Split (1)

train · 901k rows

article_title stringlengths 1 1.44k	abstract stringlengths 10 22.8k	contributor stringlengths 29 38.3k	journal stringlengths 3 246	full_text stringlengths 22 2.53M
Recreational and occupational field exposure to freshwater cyanobacteria – a review of anecdotal and case reports, epidemiological studies and the challenges for epidemiologic assessment	<p>Cyanobacteria are common inhabitants of freshwater lakes and reservoirs throughout the world. Under favourable conditions, certain cyanobacteria can dominate the phytoplankton within a waterbody and form nuisance blooms. Case reports and anecdotal references dating from 1949 describe a range of illnesses associated with recreational exposure to cyanobacteria: hay fever-like symptoms, pruritic skin rashes and gastro-intestinal symptoms are most frequently reported. Some papers give convincing descriptions of allergic reactions while others describe more serious acute illnesses, with symptoms such as severe headache, pneumonia, fever, myalgia, vertigo and blistering in the mouth. A coroner in the United States found that a teenage boy died as a result of accidentally ingesting a neurotoxic cyanotoxin from a golf course pond. This death is the first recorded human fatality attributed to recreational exposure to cyanobacteria, although uncertainties surround the forensic identification of the suspected cyanotoxin in this case.</p><p>We systematically reviewed the literature on recreational exposure to freshwater cyanobacteria. Epidemiological data are limited, with six studies conducted since 1990. Statistically significant increases in symptoms were reported in individuals exposed to cyanobacteria compared to unexposed counterparts in two Australian cohort studies, though minor morbidity appeared to be the main finding. The four other small studies (three from the UK, one Australian) did not report any significant association. However, the potential for serious injury or death remains, as freshwater cyanobacteria under bloom conditions are capable of producing potent toxins that cause specific and severe dysfunction to hepatic or central nervous systems. The exposure route for these toxins is oral, from ingestion of recreational water, and possibly by inhalation.</p><p>A range of freshwater microbial agents may cause acute conditions that present with features that resemble illnesses attributed to contact with cyanobacteria and, conversely, acute illness resulting from exposure to cyanobacteria or cyanotoxins in recreational waters could be misdiagnosed. Accurately assessing exposure to cyanobacteria in recreational waters is difficult and unreliable at present, as specific biomarkers are unavailable. However, diagnosis of cyanobacteria-related illness should be considered for individuals presenting with acute illness following freshwater contact if a description is given of a waterbody visibly affected by planktonic mass development.</p>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Stewart</surname><given-names>Ian</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Webb</surname><given-names>Penelope M</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Schluter</surname><given-names>Philip J</given-names></name><xref ref-type="aff" rid="I5">5</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Shaw</surname><given-names>Glen R</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I3">3</xref><xref ref-type="aff" rid="I6">6</xref><email>[email protected]</email></contrib>	Environmental Health	<sec><title>Introduction</title><p>Cyanobacteria are a diverse group of prokaryotes that occupy a broad range of ecological niches by virtue of their age, having first appeared some 2.5 billion years ago, and specialisation. All cyanobacteria are photoautotrophic organisms, yet many can grow heterotrophically, using light for energy and organic compounds as a carbon source [<xref ref-type="bibr" rid="B1">1</xref>]. The cyanobacteria are a remarkably widespread and successful group, colonising freshwater, marine and terrestrial ecosystems, including extreme habitats such as Antarctic lakes, salt works and hot springs [<xref ref-type="bibr" rid="B2">2</xref>]. Cyanobacteria are common inhabitants of freshwater lakes and reservoirs throughout the world. Under favourable conditions, certain cyanobacteria can dominate the phytoplankton within a waterbody and form nuisance blooms.</p><p>Cyanobacteria have come to the attention of public health workers because many freshwater and brackish species can produce a range of potent toxins. This observation was first reported over 120 years ago, when sheep, horses, dogs and pigs were seen to die within hours of drinking from a lake affected by a bloom of the brackish-water cyanobacterium <italic>Nodularia spumigena </italic>[<xref ref-type="bibr" rid="B3">3</xref>]. Since then, many reports of livestock and wild animal deaths have appeared in the literature. Such reports have been collated and discussed by several authors [<xref ref-type="bibr" rid="B4">4</xref>-<xref ref-type="bibr" rid="B10">10</xref>]. Some reports are dramatic in terms of the number of animals affected or the rapid progression of illness and death, with mass deaths of thousands of animals [<xref ref-type="bibr" rid="B11">11</xref>], and large animals succumbing within minutes [<xref ref-type="bibr" rid="B12">12</xref>]. Laboratory-based toxicological investigations have confirmed that freshwater and brackish cyanobacteria produce several categories of toxin that are (with one exception – the saxitoxins) unique to cyanobacteria. The topic of cyanobacterial toxins has been widely studied, and many excellent texts and reviews are available, e.g. [<xref ref-type="bibr" rid="B8">8</xref>-<xref ref-type="bibr" rid="B10">10</xref>,<xref ref-type="bibr" rid="B13">13</xref>-<xref ref-type="bibr" rid="B26">26</xref>]. Details of the principal cyanotoxin groups that are significant from the public health perspective of acute exposure and outcome are summarised in Table <xref ref-type="table" rid="T1">1</xref>. Lipopolysaccharides, which are defining structural components of the cell walls of Gram-negative bacteria, are discussed in the accompanying review by Stewart et al [<xref ref-type="bibr" rid="B27">27</xref>]. Cyanobacteria are rich sources of bioactive compounds; structurally diverse metabolites with cytotoxic, tumour-promoting and enzyme-inhibiting properties are known and presumably many more await discovery. Some of these metabolites are discussed by Bickel et al[<xref ref-type="bibr" rid="B28">28</xref>] and Forchert et al [<xref ref-type="bibr" rid="B29">29</xref>].</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Cyanotoxins with public health significance from acute exposures</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="center"><bold>Toxin or toxin group</bold></td><td align="center"><bold>Classification by principal target organ systems</bold></td><td align="center"><bold>Toxin-producing genera</bold></td><td align="center"><bold>LD<sub><bold>50</bold></sub>(i.p. mouse)</bold></td><td align="center"><bold>References</bold></td></tr></thead><tbody><tr><td align="center">Microcystins</td><td align="center">Hepatotoxins</td><td align="center"><italic>Anabaena, Anabaenopsis, Aphanocapsa, Arthrospira, Hapalosiphon, Microcystis, Nostoc, Oscillatoria, Planktothrix, Snowella, Woronichinia</italic></td><td align="center">25->1000 μg/kg</td><td align="center">[10, 19, 26, 125-128]</td></tr><tr><td align="center">Nodularins</td><td align="center">Hepatotoxins</td><td align="center"><italic>Nodularia</italic></td><td align="center">30–60 μg/kg</td><td align="center">[8, 26, 129]</td></tr><tr><td align="center">Anatoxin-a, homoanatoxin-a</td><td align="center">Neurotoxins</td><td align="center"><italic>Anabaena, Aphanizomenon, Arthrospira, Cylindrospermum, Microcystis, Oscillatoria, Phormidium, Planktothrix, Raphidiopsis</italic></td><td align="center">200–375 μg/kg</td><td align="center">[8, 10, 18, 26, 130-135]</td></tr><tr><td align="center">Anatoxin-a(s)</td><td align="center">Neurotoxin</td><td align="center"><italic>Anabaena</italic></td><td align="center">20–40 μg/kg</td><td align="center">[8, 26, 132]</td></tr><tr><td align="center">Saxitoxins</td><td align="center">Neurotoxins</td><td align="center"><italic>Anabaena, Aphanizomenon, Cylindrospermopsis, Lyngbya, Planktothrix</italic></td><td align="center">10–30 μg/kg</td><td align="center">[26, 127, 132, 136-140]</td></tr><tr><td align="center">Cylindrospermopsin</td><td align="center">General cytotoxin (multiple organ systems affected, incl. liver, kidney, gastrointestinal tract, heart, spleen, thymus, skin)</td><td align="center"><italic>Anabaena, Aphanizomenon, Cylindrospermopsis, Raphidiopsis, Umezakia</italic></td><td align="center">2.1 mg/kg (24 hours) 200 μg/kg (5–6 days)</td><td align="center">[8, 10, 17, 132, 141-145]</td></tr><tr><td align="center">Aplysiatoxin, debromoaplysiatoxin</td><td align="center">Dermal toxins; probable gastro-intestinal inflammatory toxin</td><td align="center"><italic>Lyngbya</italic></td><td align="center">107–117 μg/kg</td><td align="center">[146-152]</td></tr><tr><td align="center">Lyngbyatoxin A</td><td align="center">Possible gastro-intestinal inflammatory toxin</td><td align="center"><italic>Lyngbya</italic></td><td align="center">250 μg/kg (?LD<sub>100</sub>)</td><td align="center">[153]</td></tr></tbody></table></table-wrap><p>A recent report has shown that <italic>β-N</italic>-methylamino-L-alanine (BMAA), a neurotoxic non-protein amino acid associated with an atypical motor neurone disease/Parkinsonism/Alzheimer's-like dementia complex, is produced by a wide variety of cyanobacteria [<xref ref-type="bibr" rid="B30">30</xref>]. BMAA is thought to be capable of binding to endogenous proteins, in which form it may function as a "slow toxin", and may be implicated in the aetiology of other long-latency neurodegenerative diseases such as Alzheimer's disease [<xref ref-type="bibr" rid="B31">31</xref>]. The public health implications of this cyanobacteria-related neurotoxicity hypothesis have been further discussed [<xref ref-type="bibr" rid="B32">32</xref>].</p><p>Cyanobacterial poisoning of humans has occurred through known and suspected exposure to cyanotoxin-contaminated drinking water supplies [<xref ref-type="bibr" rid="B33">33</xref>] and reviewed in: [<xref ref-type="bibr" rid="B8">8</xref>-<xref ref-type="bibr" rid="B10">10</xref>,<xref ref-type="bibr" rid="B34">34</xref>]; confirmed and suspected exposure to contaminated dialysate in patients undergoing haemodialysis [<xref ref-type="bibr" rid="B35">35</xref>-<xref ref-type="bibr" rid="B39">39</xref>]; and through recreational and occupational contact. This review will concentrate on the latter exposures.</p><sec><title>Rationale and search criteria</title><p>All references that could be found in the medical and scientific literature, including conference proceedings, which describe specific incidents involving human illness and exposure to freshwater cyanobacteria in recreational or in-field occupational settings are summarised in <xref ref-type="supplementary-material" rid="S1">Additional File 1</xref>. The following citation sources were not examined for this exercise:</p><p>• Reports of cyanobacteria-associated illness from recreational exposures to marine or estuarine waters.</p><p>• Publications written in languages other than English – with the exception of three papers which we were opportunistically able to have translated [<xref ref-type="bibr" rid="B40">40</xref>-<xref ref-type="bibr" rid="B42">42</xref>].</p><p>• Newspaper reports – with three exceptions: two reports describe the first human fatality to be attributed to recreational contact with cyanobacteria [<xref ref-type="bibr" rid="B43">43</xref>,<xref ref-type="bibr" rid="B44">44</xref>]. At the time this review was submitted, these were apparently the only published references to describe the events surrounding this tragedy, so were included here because of their importance. The cyanobacteria research community awaits publication of a comprehensive case report in the scientific or medical literature. Another news article supplements a cursory description in an academic journal (though not a health-related journal) of cyanobacteria-associated illnesses; both the news report and the scientific publication appear to describe the same incident, with more detail provided by the journalist [<xref ref-type="bibr" rid="B45">45</xref>,<xref ref-type="bibr" rid="B46">46</xref>]. There are undoubtedly many more publications in the news media that report suspected cyanobacteria-related human and animal morbidity and mortality: for example Duggan [<xref ref-type="bibr" rid="B47">47</xref>] and Ruff [<xref ref-type="bibr" rid="B48">48</xref>] reported on cyanobacteria blooms in Nebraska lakes that were associated with two dog deaths and more than 40 complaints of acute eye, upper respiratory, gastrointestinal and skin symptoms.</p><p>Anecdotal and case reports presented in this review were identified by the following search strategy:</p><p>1. PubMed and Web of Science electronic databases were searched with the MeSH and textword string "(cyanobacter* AND disease outbreaks) OR (cyanobacter* AND environmental exposure) OR (cyanobacter* AND recreation) OR (cyanobacter AND epidemiology)".</p><p>2. Titles and abstracts (when available electronically) were perused to determine suitability for inclusion.</p><p>3. Bibliographies of identified primary papers and related review articles were reviewed to search for references not identified by electronic sources.</p><p>4. Publications and other sources identified and forwarded by experts working in this field were included.</p><p>The most recent update of the aforementioned electronic searches, conducted in June 2005, gave 257 citations, of which 244 were English-language publications and 13 were non-English-language papers. Of these 13 reports, three (two reviews and one primary article) were identified from abstracts and/or article titles as worth perusing for the presence of information about health-related events associated with recreational exposure to cyanobacteria [<xref ref-type="bibr" rid="B41">41</xref>,<xref ref-type="bibr" rid="B49">49</xref>,<xref ref-type="bibr" rid="B50">50</xref>]. One of these papers (the primary article, in Dutch) and another German review paper we found with a different search strategy were translated for us, but there were no previously unreported references in those papers to specific illness events that were attributed to contact with cyanobacteria [<xref ref-type="bibr" rid="B41">41</xref>,<xref ref-type="bibr" rid="B42">42</xref>]. Therefore it does not appear that there is a significantly large body of unexplored literature written in languages other than English that could contribute to this review. We also corresponded with an author of a publication in Finnish that we were unable to have translated; this paper discussed cyanobacteria-related illness in saunas [<xref ref-type="bibr" rid="B51">51</xref>]. The findings of that work were presented at an international conference, from which an English-language abstract was published. The authors reported that 18 subjects (38% of those questioned) were likely to have experienced cyanobacteria-related symptoms [<xref ref-type="bibr" rid="B52">52</xref>].</p></sec><sec><title>Recreational and in-field occupational exposure to cyanobacteria: anecdotal and case reports</title><p>Case reports and anecdotal references presented in <xref ref-type="supplementary-material" rid="S1">Additional File 1</xref> date from 1949 [<xref ref-type="bibr" rid="B53">53</xref>], and describe a range of illnesses associated with recreational exposure to cyanobacteria: hay fever-like symptoms, pruritic skin rashes and gastro-intestinal symptoms are most frequently reported. Some papers give convincing descriptions of allergic responses to cyanobacteria [<xref ref-type="bibr" rid="B53">53</xref>,<xref ref-type="bibr" rid="B54">54</xref>]. Others describe more serious acute illnesses, with symptoms such as severe headache, pneumonia, fever, myalgia, vertigo and blistering in the mouth [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B55">55</xref>-<xref ref-type="bibr" rid="B57">57</xref>]. The first and so far only description of a fatality attributed to recreational exposure to cyanotoxins appeared in news reports recently [<xref ref-type="bibr" rid="B43">43</xref>,<xref ref-type="bibr" rid="B44">44</xref>]. A U.S. coroner concluded that a teenage boy died as a result of ingesting anatoxin-a-producing cyanobacteria from a golf course pond, although there was an unusual sequence of events preceding the death insofar as the time period between exposure and death (some 48 hours) does not tally with the known mechanisms of toxicity of anatoxin-a, which initiates pathological signs and death in laboratory animals within minutes of dosing by either oral or parenteral routes [<xref ref-type="bibr" rid="B58">58</xref>-<xref ref-type="bibr" rid="B60">60</xref>]. Animals exposed to anatoxin-a-producing cyanobacteria in the field succumb within minutes to a few hours, depending on the species, the amount of toxin consumed, and prior food intake [<xref ref-type="bibr" rid="B61">61</xref>]. However, Rogers et al [<xref ref-type="bibr" rid="B60">60</xref>] demonstrated delayed mortality in toad embryos – over 6–13 days post-exposure – to anatoxin-a. Recent reports in the scientific literature also add to the uncertainty in the case of the teenager's death, with suggestions that the preliminary mass spectrometric identification of anatoxin-a in the forensic samples may not be reliable [<xref ref-type="bibr" rid="B62">62</xref>-<xref ref-type="bibr" rid="B64">64</xref>].</p><p>The principal public health concerns regarding recreational exposures relate to the potential, presumably a now-realised potential if the aforementioned fatality is indeed attributable to cyanotoxin poisoning, for exposure to hazardous levels of cyanotoxins in untreated waters. Routes of exposure are through direct contact with skin and mucous membranes, via inhalation, and by ingestion, either accidental or deliberate.</p></sec><sec><title>Discussion of anecdotal and case reports</title><p>Some reports listed in <xref ref-type="supplementary-material" rid="S1">Additional File 1</xref> present scant information relevant to this topic, with little or no detail beyond location and the kind of illness reported [<xref ref-type="bibr" rid="B65">65</xref>,<xref ref-type="bibr" rid="B66">66</xref>]. On the other end of the scale are examples of thorough, considered case reports, describing relevant medical history and diagnostic investigations [<xref ref-type="bibr" rid="B53">53</xref>,<xref ref-type="bibr" rid="B54">54</xref>]. One reason for the dearth of detail may be that non-specific, mild and self-limiting illnesses do not merit much discussion, however, some references to more serious illnesses leave a great deal unanswered, for example the 12 year-old boy who reportedly lapsed into unconsciousness for a six-hour period, and developed pneumonia, myalgia and arthralgia [<xref ref-type="bibr" rid="B67">67</xref>]. It would have been very interesting to know whether or not this boy had any predisposing medical conditions (e.g. diabetes, epilepsy) that might have explained the loss of consciousness, whether any medical attention was sought, and, if so, the details of his disease progression.</p><p>The observation that repeated water contact in a particular lake preceded a skin eruption on a six year-old girl, while other bathers appeared unaffected, helped support a diagnosis of hypersensitivity in that case [<xref ref-type="bibr" rid="B54">54</xref>]. One of the few reports of mass effects, with 20–30 children suffering conjunctival and upper respiratory symptoms during a school aquatic event, is tempered by the observation that that number represented about 25% of those exposed [<xref ref-type="bibr" rid="B68">68</xref>]. So hypersensitivity reactions affecting a sub-set of allergy-prone children may also be an explanation for the latter outbreak, although this speculation – in the absence of any other reported investigations – is solely based on that estimate of 25% of those exposed experiencing symptoms.</p><p>Those reports that have indicated symptom onset time suggest that responses can be rapid, with some urticarial and hay fever-like symptoms commencing while subjects are still in the water [<xref ref-type="bibr" rid="B53">53</xref>,<xref ref-type="bibr" rid="B68">68</xref>]. While a disparate range of signs and symptoms are listed, several reports describe a collective group of symptoms resembling immediate or Type-I hypersensitivity reactions. Immediate hypersensitivity reactions are commonly associated with atopy, which is the familial tendency to react to naturally occurring antigens, mostly proteins, through an IgE-mediated process. Atopy frequently manifests as a spectrum of diseases, e.g. seasonal rhinitis, conjunctivitis, asthma and urticaria. Different atopic illnesses often affect the same individual. A fundamental feature of Type-I hypersensitivity reactions is the rapid onset of symptoms – normally seconds to minutes – following exposure to antigens [<xref ref-type="bibr" rid="B69">69</xref>-<xref ref-type="bibr" rid="B73">73</xref>].</p><p>Some serious though apparently self-limiting gastro-intestinal illnesses have been reported after contact with cyanobacteria in recreational waters, presumably through ingestion of affected water. Dillenberg & Dehnel [<xref ref-type="bibr" rid="B55">55</xref>] describe how an adult male inadvertently swallowed lake water affected by a bloom of <italic>Microcystis </italic>sp. and <italic>Anabaena circinalis</italic>. After some three hours he developed cramping abdominal pain and nausea, which progressed to painful diarrhoea followed by a fever of 38.9°C, severe headache, lassitude, myalgia and arthralgia. Such illnesses are worrying, considering the two boys that were sickened – one of whom subsequently died – after possible exposure to cyanobacteria in a golf course pond suffered acute and severe gastro-intestinal illnesses [<xref ref-type="bibr" rid="B43">43</xref>].</p><p>Occupational exposures were included in this review, although some caution should be exercised when comparing occupational and recreational exposures. Waters that are obviously discoloured or visibly affected by cyanobacteria scums may be of interest to aquatic field workers who are keen and/or obliged to collect samples. The two incidents involving British soldiers and sea cadets conducting canoe capsizing activities, presumably under orders from their supervising officers, occurred in waters that were reportedly subject to a "heavy bloom of <italic>Microcystis </italic>spp" [<xref ref-type="bibr" rid="B74">74</xref>] and a "scum of <italic>Oscillatoria</italic>..." [<xref ref-type="bibr" rid="B23">23</xref>]. Waters that are obviously suffering a loss of visual amenity may be shunned by many recreational users, although avoidance behaviour in such circumstances cannot be taken for granted [<xref ref-type="bibr" rid="B75">75</xref>].</p><p>The other reports that are of particular interest are those grouped under "cold & flu-like symptoms". Several publications describe individuals presenting with a flu-like illness, with signs and symptoms including fever, headache, lassitude, arthralgia, myalgia, sore throat, cough, diarrhoea and vomiting. A proposed explanation for this constellation of symptoms is that of a coordinated, cytokine-mediated innate immune response. Fever and malaise are events that are directed by endogenous mediators; for further discussion see [<xref ref-type="bibr" rid="B75">75</xref>]. This spectrum of signs and symptoms also mimics those reported in volunteer studies of intravenous Gram-negative bacterial LPS injection [<xref ref-type="bibr" rid="B76">76</xref>-<xref ref-type="bibr" rid="B79">79</xref>]. Mammalian responses to LPS are mediated by inflammatory cytokines (see accompanying review by Stewart et al [<xref ref-type="bibr" rid="B27">27</xref>]). The early signs and symptoms of influenza infection (fever, myalgia, fatigue, drowsiness, rhinorrhoea, sore throat, headache) are mediated by pro-inflammatory cytokines, particularly IFN-α and IL-6 [<xref ref-type="bibr" rid="B80">80</xref>-<xref ref-type="bibr" rid="B83">83</xref>]. Flu-like reactions to immunostimulant drugs are sometimes referred to as "acute cytokine syndromes" [<xref ref-type="bibr" rid="B84">84</xref>], and the flu-like syndrome of fever, rigors, tachycardia, malaise, headache, arthralgia and myalgia that accompanies interferon pharmacotherapy is thought to be due to the release of eicosanoids, IL-1 and TNF-α [<xref ref-type="bibr" rid="B85">85</xref>].</p></sec><sec><title>Epidemiology of recreational exposure to cyanobacteria</title><p>Six epidemiological studies of recreational exposure to cyanobacteria were identified with the search strategy discussed previously: three analytical cross-sectional studies from the U.K. using similar survey instruments [<xref ref-type="bibr" rid="B86">86</xref>-<xref ref-type="bibr" rid="B88">88</xref>], a small case-control study from Australia [<xref ref-type="bibr" rid="B89">89</xref>], and two larger prospective cohort studies, also from Australia [<xref ref-type="bibr" rid="B90">90</xref>,<xref ref-type="bibr" rid="B91">91</xref>]. Table <xref ref-type="table" rid="T2">2</xref> lists the pertinent findings of these studies, which are discussed in detail below.</p><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Summary of epidemiological studies investigating recreational exposure to cyanobacteria</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="center"><bold>Country; year study conducted; study author/s; reference</bold></td><td align="center"><bold>Study design</bold></td><td align="center"><bold>Main outcomes reported</bold></td><td align="center"><bold>Study size (<bold><italic>n</italic></bold>)</bold></td><td align="center"><bold>Odds ratio (95% confidence interval)</bold></td></tr></thead><tbody><tr><td align="left">UK, 1990 Philipp [86]</td><td align="center">Cross-sectional</td><td align="center">No statistically significant findings</td><td align="center">246</td><td></td></tr><tr><td align="left">UK, 1990 Philipp & Bates [87]</td><td align="center">Cross-sectional</td><td align="center">No statistically significant findings</td><td align="center">363</td><td></td></tr><tr><td align="left">UK, 1990 Philipp et al [88]</td><td align="center">Cross-sectional</td><td align="center">No statistically significant findings</td><td align="center">246</td><td></td></tr><tr><td align="left">Australia, 1992 El Saadi et al [89]</td><td align="center">Case-control</td><td align="center">No statistically significant findings</td><td align="center">Approx. 48 (subjects reporting recreational exposure)</td><td></td></tr><tr><td align="left">Australia, 1995 Pilotto et al [90]</td><td align="center">Prospective cohort</td><td align="center">Increased symptoms at 7 days following exposure to more than 5,000 cyanobacterial cells/mL for >1 hour <italic>vs </italic>non-bathers</td><td align="center">852 (total)<break/>338 (no prior exposure or symptoms)</td><td align="center">1.3 (0.7–2.6)<break/>3.4 (1.1–10.8)</td></tr><tr><td align="left">Australia & USA, 1999–2002 Stewart et al [91]</td><td align="center">Prospective cohort</td><td align="center">Increased reporting of mild respiratory symptoms and any symptom at 3 days following exposure to cyanobacteria cell surface area >12 mm<sup>2</sup>/mL <italic>vs </italic><2.4 mm<sup>2</sup>/mL</td><td align="center">1,331 (total)<break/>1,137 (no prior symptoms)<break/>1,149 (no prior respiratory symptoms)</td><td align="center">1.7 (1.0–2.9) (any symptom)<break/>2.1 (1.1–4.0) (respiratory symptoms)</td></tr></tbody></table></table-wrap><p>The three cross-sectional studies were conducted by Philipp and co-workers [<xref ref-type="bibr" rid="B86">86</xref>-<xref ref-type="bibr" rid="B88">88</xref>]. Questionnaires were distributed to recreational users of six inland waterbodies, five of which experienced cyanobacteria blooms during 1990. The questionnaires elicited information on exposure to study waters and the presence of specific symptoms in a defined period prior to receiving the form. This period ranged from 14 days [<xref ref-type="bibr" rid="B87">87</xref>] to four weeks [<xref ref-type="bibr" rid="B88">88</xref>]. One questionnaire asked about exposure to the study water on a weekend when a bloom occurred some 21/2 weeks previously [<xref ref-type="bibr" rid="B86">86</xref>]. Recreational interest groups were used to target likely users of the waterbodies; questionnaires were mailed to members of sailing and angling clubs. Site authorities distributed questionnaires at one study lake [<xref ref-type="bibr" rid="B88">88</xref>]. The results of these three studies were similar: mostly minor morbidity was reported, with similar disease patterns across sites.</p><p>The theoretical advantages of this study type are that it is reasonably cost-effective, and in this context – recreational exposure to cyanobacteria – it can be conducted opportunistically to take advantage of any sudden-onset cyanobacteria blooms. Disadvantages relate to the difficulty in establishing that exposure occurred before the outcome [<xref ref-type="bibr" rid="B92">92</xref>,<xref ref-type="bibr" rid="B93">93</xref>]. The studies conducted by Philipp and his team [<xref ref-type="bibr" rid="B86">86</xref>-<xref ref-type="bibr" rid="B88">88</xref>] were examples of analytical cross-sectional studies, in that unexposed individuals served as controls for statistical comparison of illness reporting.</p><p>A case-control study of illness rates was conducted after an extensive <italic>Anabaena circinalis</italic>-dominant bloom along South Australia's Murray River in the summer of 1991–1992 [<xref ref-type="bibr" rid="B89">89</xref>]. Patients presenting with gastro-intestinal (G-I) or dermatological complaints comprised the case group; the patient presenting after each case was identified served as the control group. Exposure was determined by identifying each subject's principal source of water for drinking, domestic use (bathing, dishwashing) and recreation during the week prior to consultation. Recreational exposure was categorised as no contact, direct exposure to river water, or other exposure, e.g. farm dams or treated water in swimming pools. The study found a significantly increased risk of G-I symptoms for those drinking chlorinated river water, and an increased risk of G-I and cutaneous symptoms in those using untreated river water for domestic purposes. There was a statistically non-significant increase in the relative odds of developing G-I or skin symptoms amongst those with recreational exposure to river water, but that risk was lower than for those exposed to other sources of recreational water (tank, farm dam or another location). The number of subjects was small for the recreational exposure component of the study, with only some 50 subjects (16% of the study group) reporting any recreational exposure during the study period [<xref ref-type="bibr" rid="B89">89</xref>].</p><p>The advantages of a case-control design for investigating recreational exposure to cyanobacteria are that studies can be conducted opportunistically in response to the development of cyanobacteria blooms, and they are very useful for investigating infrequent outcomes. The study of El Saadi et al [<xref ref-type="bibr" rid="B89">89</xref>] has another advantage over other epidemiological studies into recreational exposure to cyanobacteria in that medical practitioners ascertained outcome data, as opposed to self-reporting of symptoms. General disadvantages of the case-control design principally relate to the problem of recall bias, where individuals with the disease of interest tend to overestimate relevant past exposures [<xref ref-type="bibr" rid="B92">92</xref>,<xref ref-type="bibr" rid="B93">93</xref>]. Because the outcome has already occurred when exposure is measured, people with disease may systematically overestimate (or underestimate) their exposure compared to disease-free controls, leading to falsely elevated (or reduced) measures of risk associated with exposure. Another major issue with case-control studies is the difficulty of identifying an appropriate control group – i.e. people who would have been identified as cases if they had the disease of interest.</p><p>Recall bias may not be so much of a problem for investigating acute illnesses following recreational exposure to cyanobacteria, where a fairly short time lag between exposure and symptom onset can be anticipated, especially if recreational exposure is determined by a yes/no response. The main problem with a case-control study in this context will be in actually identifying cases. A case-control design would not be suitable for investigating outcomes from exposure to a cyanobacteria bloom in a lake adjacent to a city, as most recreational users who do develop symptoms would presumably seek medical attention after they return home, i.e. from one of a large number of medical practitioners. El Saadi et al [<xref ref-type="bibr" rid="B89">89</xref>] alluded to the difficulty of gaining the cooperation of medical practitioners, as they approached practices in 11 towns along the Murray River, yet those in three towns presumably refused to participate in their study. The diffuse spread of cases from point sources of exposure (a cyanobacteria-affected waterbody) across a large town or city would make a case-control study practically unworkable. A case-control study would also be unsuitable for recruiting subjects who did not seek medical attention for symptoms occurring after exposure. However, a well-designed case-control study would be valuable if geographical location is a primary consideration. This would require enlisting the cooperation of medical practitioners in small townships near to cyanobacteria-affected recreational waters that are sufficiently remote from larger urban centres to allow recruitment of local residents and tourists who will camp nearby.</p><p>The studies by Pilotto et al [<xref ref-type="bibr" rid="B90">90</xref>] and Stewart et al [<xref ref-type="bibr" rid="B91">91</xref>] were prospective cohort studies. Pilotto et al [<xref ref-type="bibr" rid="B90">90</xref>] recruited individuals at five recreational waterbodies in three Australian states. Cyanobacteria blooms were anticipated at these sites, based on occurrences in previous years. Individuals were approached and invited to participate in the study. Participants completed a face-to-face interview to determine health status and recreational water activities; two telephone follow-up interviews were conducted at two and seven days following the day of recruitment into the study. Individuals who did not have water contact on the recruitment day served as the control group. No significant differences in symptom occurrence were reported at the 2<sup>nd </sup>day follow-up, but the authors concluded there was a significant increase in symptoms at 7 days, after excluding subjects with symptoms or previous recent recreational water exposure. The cohort size from which these significant results were drawn was rather small, with 93 exposed subjects, and 43 unexposed controls. Pilotto et al [<xref ref-type="bibr" rid="B90">90</xref>] interpreted the increased symptom reporting at 7 days but not 2 days following exposure as possibly due to delayed allergic responses, although so-called "late phase" allergic and asthmatic responses tend to occur some 4–24 hours after allergen exposure [<xref ref-type="bibr" rid="B69">69</xref>,<xref ref-type="bibr" rid="B94">94</xref>,<xref ref-type="bibr" rid="B95">95</xref>].</p><p>Stewart et al [<xref ref-type="bibr" rid="B91">91</xref>] also conducted a cohort study of recreational exposure to cyanobacteria. 1,331 subjects were recruited from 19 recreational waterbodies in eastern Australia and central and northeast Florida; subjects completed a self-administered questionnaire to determine recreational activity, recent illness and history of any relevant chronic diseases such as asthma, hay fever and eczema. A single follow-up telephone interview was conducted after three days post-exposure. Reference subjects were recruited at recreational waters unaffected by cyanobacteria; exposure categories (low, intermediate, high) were allocated to study subjects on the basis of cyanobacteria levels measured in study water samples collected on the day they were recruited into the study. Statistically significant increased reporting of respiratory symptoms and a pooled "any symptom" category occurred amongst subjects exposed to high levels of cyanobacteria, although symptoms were predominantly rated as mild by study subjects. A similar but non-significant relationship was also seen for reporting of skin, ear and fever symptom groups.</p><p>The studies of Pilotto et al [<xref ref-type="bibr" rid="B90">90</xref>] and Stewart et al [<xref ref-type="bibr" rid="B91">91</xref>] are both examples of a prospective cohort design, where study subjects have their exposure status determined, and are then followed forward in time to observe the development of disease. For these investigations into recreational exposure to cyanobacteria, exposure status was determined by collecting water samples on the day subjects were recruited into the study; cyanobacteria were identified and enumerated and the resultant cell counts or biomass estimates formed the basis of exposure at any given site on a particular day. One of the problems with this approach is that cyanobacteria blooms are dynamic and can change rapidly. Unless the presence of significant cyanobacterial biomass can be predicted with some degree of certainty, a prospective cohort design can result in wasted effort if the water samples reveal lower than anticipated levels of cyanobacteria. This problem undoubtedly occurred in some instances during the study conducted by Stewart et al [<xref ref-type="bibr" rid="B91">91</xref>]. One possible approach to dealing with this would be to conduct a historical cohort study, where a cohort of subjects is identified after some have experienced the outcome of interest and relevant exposure information is obtained from historical records (i.e. as in a prospective cohort study the exposure information was recorded <italic>before </italic>any outcomes occurred).</p><p>Whether a cohort study is conducted prospectively or retrospectively, the basic study design is identical – exposed and unexposed groups are compared with respect to disease outcome [<xref ref-type="bibr" rid="B93">93</xref>]. General advantages of a cohort design are the ability to determine disease onset (the exposure precedes the disease), and the study of exposures in natural settings [<xref ref-type="bibr" rid="B92">92</xref>]. General disadvantages relate to confounding, which refers to differences in the distribution of risk factors other than the exposure of interest between exposed and unexposed groups. Cohort studies can be expensive and resource intensive [<xref ref-type="bibr" rid="B92">92</xref>].</p><p>Further discussion of some common epidemiological study designs that may be useful for investigating the topic of recreational exposure to aquatic cyanobacteria, with particular emphasis on the relative advantages and disadvantages of experimental epidemiology (randomised trials) is presented by Stewart [<xref ref-type="bibr" rid="B75">75</xref>].</p></sec><sec><title>Cyanobacteria and water-related disease: some complicating factors</title><p>Other explanations for disease need to be considered by both clinicians and epidemiologists in their respective endeavours. Epidemiological studies usually aim to identify and adjust for confounding variables such as smoking and age of study participants. The following sections will discuss some freshwater-related risk factors, mostly microbial, that may confound epidemiological studies and complicate clinical diagnoses of cyanobacteria-related illness linked to recreational exposures. The final section of this review will discuss the possibility of misdiagnosis from the opposite direction: a water-borne disease outbreak in Finland that was subject to epidemiological scrutiny, but cyanobacterial exotoxin contamination of reticulated supplies was apparently not considered at the time.</p></sec><sec><title>Freshwater-related dermatoses</title><p>• <bold>Avian cercariae: </bold>avian cercariae are schistosome larvae for which humans are an accidental host. Pruritus and macules are the initial signs and symptoms; sometimes a diffuse erythema and urticaria can develop and last for several hours [<xref ref-type="bibr" rid="B96">96</xref>-<xref ref-type="bibr" rid="B99">99</xref>]. Fever, nausea and vomiting can also accompany severely affected cases [<xref ref-type="bibr" rid="B97">97</xref>,<xref ref-type="bibr" rid="B100">100</xref>]. The clinical presentation of cercarial dermatitis can be difficult to delineate from the picture of cyanobacterial dermatitis.</p><p>• <bold>Gram-negative bacteria: </bold><italic>Aeromonas hydrophila </italic>and <italic>Chromobacterium violaceum </italic>are abundant in freshwater habitats. Both usually cause infection through a pre-existing skin wound, though the clinical presentations in each case do not match of any of the reports listed in <xref ref-type="supplementary-material" rid="S1">Additional File 1</xref>. <italic>A. hydrophila </italic>causes cellulitis and a purulent discharge; aspiration of water can cause pneumonia and septicaemia. <italic>C. violaceum </italic>infections present with various cutaneous signs that are secondary to systemic disease, including sepsis [<xref ref-type="bibr" rid="B101">101</xref>]. <italic>Vibrio vulnificus </italic>has reportedly caused soft tissue infection after contact in brackish inland waters, though most cases are associated with estuarine contact [<xref ref-type="bibr" rid="B102">102</xref>]. <italic>Pseudomonas aeruginosa </italic>is widely-distributed in natural and artificial aquatic environments. Cutaneous infection presents as an erythematous or urticarial rash some 18–24 hours after water contact and progresses to a follicular dermatitis. Fever and pruritus are uncommon. Most reports of pseudomonal dermatitis are related to spa pool or hot-tub exposures [<xref ref-type="bibr" rid="B102">102</xref>,<xref ref-type="bibr" rid="B103">103</xref>]. <italic>P. aeruginosa </italic>in recreational waters is a common cause of otitis externa, presenting as a purulent discharge [<xref ref-type="bibr" rid="B102">102</xref>]. Diagnostic criteria include culturing the organism from skin or ear swabs; the incubation period would also help to distinguish <italic>P. aeruginosa </italic>infection from cyanobacteria-related dermatoses.</p><p>• <bold>Non-allergic urticaria: </bold>physical stimuli such as heat, cold and exercise can induce itching and hives in susceptible individuals [<xref ref-type="bibr" rid="B99">99</xref>,<xref ref-type="bibr" rid="B104">104</xref>].</p></sec><sec><title>Gastro-intestinal illness</title><p>• <bold>Shigellosis: </bold>Shigella outbreaks are the most commonly reported cause of disease associated with untreated inland recreational water in the USA, with 16 events affecting almost 1,300 people between 1985 and 1994 [<xref ref-type="bibr" rid="B102">102</xref>]. The incubation period is typically 2–3 days, with an upper limit of about 7 days. Illness severity is strain-dependent, with most <italic>S. sonnei </italic>infections being mild and self-limiting, and <italic>S. dysenteriae </italic>type 1 associated with severe diarrhoea which may progress to a life-threatening illness [<xref ref-type="bibr" rid="B102">102</xref>].</p><p>• <bold><italic>Escherichia coli</italic>: </bold><italic>E. coli </italic>are markers of faecal pollution in recreational waters. Disease outbreaks traced to enterohaemorrhagic <italic>E. coli </italic>0157 have been reported from recreational water exposures [<xref ref-type="bibr" rid="B102">102</xref>,<xref ref-type="bibr" rid="B105">105</xref>].</p><p>• <bold>Norwalk-like viruses: </bold>Various transmission routes, including recreational water outbreaks have been documented [<xref ref-type="bibr" rid="B105">105</xref>].</p></sec><sec><title>Other microbial pathogens</title><p>• <bold><italic>Naegleria fowleri</italic>: </bold><italic>N. fowleri </italic>is a free-living thermotolerant amoeba found in warm or thermally polluted waters. It is the causative organism of primary amoebic meningoencephalitis, a fulminating, typically fatal illness. The entry route is via the nasal mucosa; fit, immunocompetent children and young adults with a recent history of freshwater recreational activity are those most commonly affected. The causative organism and diagnosis are usually confirmed at autopsy. Several reviews are available, e.g. [<xref ref-type="bibr" rid="B106">106</xref>-<xref ref-type="bibr" rid="B113">113</xref>].</p><p>• <bold>Viruses: </bold>Pharyngo-conjunctival fever outbreaks associated with non-enteric adenoviruses in recreational waters have been reported [<xref ref-type="bibr" rid="B105">105</xref>].</p><p>• <bold>Legionella: </bold><italic>Legionella </italic>infections have been associated with recreational water contact [<xref ref-type="bibr" rid="B105">105</xref>].</p></sec><sec><title>Possible under-diagnosis of cyanobacteria-related illness</title><p>The examples given above highlight some of the differential diagnoses that need to be worked through when considering possible cases of cyanobacteria-related illness from recreational exposures. Competent history-taking and diagnostic microbiology support will correctly diagnose many such cases. Competent history-taking and clinical diagnostic support also operated in several of the case reports listed in <xref ref-type="supplementary-material" rid="S1">Additional File 1</xref>, with the early dermatological testing and microscopic examination of stool and vomitus samples lending strong support to the suspicion of cyanobacteria-related morbidity.</p><p>Misdiagnosis of cyanobacteria-related disease may occur in both directions. In 1978, nearly half the population of an industrial town in Finland were affected by a flu-like illness, with symptoms of fever, fatigue, cough, dyspnoea and myalgia. Symptoms occurred some 3–6 hours after taking a bath, shower or sauna and persisted for 8–16 hours. The outbreak lasted for some four months. This epidemic was investigated on several fronts, and provocation testing demonstrated an obvious link to the reticulated water supply. Tap water was cultured in a range of organic media for fungal and bacterial pathogens. No definitive pathogen was identified to explain the epidemic, yet in three published reports the authors describe how the shallow lake that was the town water source had taken on a distinct opaque blue-green appearance, had a musty smell, and the sand filtration system was covered by a mat of cyanobacteria. This change occurred in the same month (August, i.e. late summer) that the epidemic began. Analysis of crude lake water in the third month after the onset of the epidemic showed high coliform counts, <italic>Aspergillus fumigatus </italic>and unspecified blue-green algae. Investigations centred on identifying antibodies to mesophilic actinomycetes, which the authors [<xref ref-type="bibr" rid="B114">114</xref>] note were not pathogenic, whereas aquatic cyanobacteria were known at the time to be toxic. The health workers investigating the outbreak apparently did not consider the possibility of a cyanobacterial exotoxin breakthrough into the reticulated supply [<xref ref-type="bibr" rid="B114">114</xref>-<xref ref-type="bibr" rid="B118">118</xref>]. The epidemiological report of Aro et al [<xref ref-type="bibr" rid="B115">115</xref>] came closest to suggesting that cyanobacteria may have been involved, suggesting that "towards the end of summer....the microorganisms in the lake multiply rapidly and produce some toxic substance or allergen", and reported that cyanobacterial endotoxin concentration in lake and tap water was high. This incident appears to have been retrospectively attributed to the presence of cyanobacterial endotoxins (i.e. LPS) in the reticulated supply [<xref ref-type="bibr" rid="B119">119</xref>]. A similar outbreak occurred almost three years earlier in a Swedish town, though with a much smaller proportion of cases identified. Cyanobacteria were known to affect the town's raw water supply, and the investigators did consider the possibility that cyanotoxins may have been responsible for the outbreak [<xref ref-type="bibr" rid="B120">120</xref>], though the analytical technique used by investigators at the time – gas chromatography – would have failed to detect the presence of cyanobacterial exotoxins in the post-treatment water supply. While no conclusions can be made about events that occurred over 25 years ago, from the descriptions of the outbreaks and the raw water supplies, most cyanobacterial toxicologists would rate cyanotoxin exposure with a high index of suspicion.</p><p>A similar outbreak occurred more recently in Homa Bay, Kenya, in 1998. Apparently associated with a mass development of cyanobacteria in Lake Victoria, an epidemic of fever, malaise, dizziness and upper respiratory symptoms was related to hot water bathing. Symptoms lasted 12–24 hours, and returned when a shower or bath was taken again. This outbreak was reported in a conference abstract; the authors suggested cyanobacterial endotoxins were responsible, though it is not stated whether any investigation of cyanobacterial exotoxins was conducted [<xref ref-type="bibr" rid="B121">121</xref>].</p></sec></sec><sec><title>Conclusion</title><p>The true incidence of acute cyanobacteria-associated illness from recreational exposure is unknown, as many outcomes are likely to be mild and self-limiting, so medical attention is not sought. With a long-standing knowledge gap amongst primary healthcare providers, non-specific signs and symptoms caused by cyanobacterial products are likely to be under-diagnosed [<xref ref-type="bibr" rid="B8">8</xref>]. Codd [<xref ref-type="bibr" rid="B122">122</xref>] stated:</p><p>"Evidence linking human illnesses with cyanobacterial cells and toxins is open to criticism because of shortfalls in early detailed case definitions, because diagnoses were made by exclusion, and because identification and quantification of cyanobacterial toxins in health incidents have, until recently, been lacking."</p><p>The collation of anecdotal and case reports of illness associated with recreational exposure to cyanobacteria in <xref ref-type="supplementary-material" rid="S1">Additional File 1</xref> will hopefully highlight some of the knowledge gaps. Particular attention should be given to determining the onset and duration of individual symptoms in future case reporting, as well as detailing the presence or absence of any predisposing medical conditions.</p><p>A recent initiative of UNESCO's International Hydrology Programme has been to establish CyanoNet, which is a "Global network for the hazard management of cyanobacterial blooms and toxins in water resources". The CyanoNet website will carry information on various associated topics, including "Reported incidents of adverse health effects including case studies" and "Surveys and epidemiological studies investigating associations between cyanobacterial populations, cyanotoxins and health" [<xref ref-type="bibr" rid="B123">123</xref>].</p><p>The most important advances in understanding the health impacts of cyanobacteria have come from the discipline of toxicology. The major toxins have been extensively studied and characterised, and while there is still much to be discovered in the field of cyanobacterial toxicology, significant advances in the future will be made at the interface of toxicology and epidemiology. Molecular epidemiology techniques using yet-to-be discovered biomarkers of exposure, susceptibility and outcome will refine knowledge of the risks associated with various acute and chronic exposures to cyanotoxins. The collaborative skills that epidemiologists and toxicologists can bring to this endeavour were viewed with a mildly jaundiced eye by Paddle [<xref ref-type="bibr" rid="B124">124</xref>], whose chapter on epidemiology for toxicologists is an excellent general primer:</p><p>"The total evidence about the risk to humans...will consist of the toxicologist's precise, experimental data about the wrong species at the wrong exposure, and the epidemiologist's imprecise, observational data about the right species at the right exposure."</p><p>In conclusion, anecdotal and case reports of variable reliability have suggested a range of symptoms are associated with exposure to cyanobacteria in recreational or occupational settings. Some reports of cutaneous reactions are strongly suggestive of allergic reactions, and symptoms such as rhinitis, conjunctivitis, asthma and urticaria also hint at immediate hypersensitivity responses. Flu-like illnesses involving a constellation of symptoms including fever, malaise, myalgia, arthralgia, severe headache, cough and sore throat are, in our opinion, explained by a cascade action of pro-inflammatory cytokines. If correct, this implies that some cyanobacterial products possess ligands that induce innate immune responses, and such responses may need to be considered in terms of their potential to direct pathological changes in the liver and other organ systems.</p><p>The epidemiology of recreational exposure to cyanobacteria is incomplete at present. All common epidemiological approaches have their own inherent advantages and disadvantages; identification of biomarkers for exposure, susceptibility and outcome in the future should lead to a significantly improved perception of the risks of bathing in cyanobacteria-affected waters.</p></sec><sec><title>Abbreviations</title><p>BMAA <italic>β-N</italic>-methylamino-L-alanine</p><p>G-I gastro-intestinal</p><p>GP General Practitioner (aka Family Physician)</p><p>IFN interferon</p><p>IgE immunoglobulin-E</p><p>IL interleukin</p><p>i.p. intra-peritoneal</p><p>LD<sub>50 </sub>lethal dose for 50% of test animals</p><p>LPS lipopolysaccharide(s)</p><p>TNF-α tumour necrosis factor-alpha</p><p>UNESCO United Nations Educational, Scientific and Cultural Organization</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>IS conducted the review; PMW, PJS and GRS supervised the work and contributed to redrafting the paper. All authors read and endorsed the final manuscript.</p></sec><sec sec-type="supplementary-material"><title>Supplementary Material</title><supplementary-material content-type="local-data" id="S1"><caption><title>Additional File 1</title><p>Anecdotal and case reports of human morbidity and mortality attributed to recreational or occupational field exposure to freshwater cyanobacteria.</p></caption><media xlink:href="1476-069X-5-6-S1.pdf" mimetype="application" mime-subtype="pdf"><caption><p>Click here for file</p></caption></media></supplementary-material></sec>
Toward a treaty on safety and cost-effectiveness of pharmaceuticals and medical devices: enhancing an endangered global public good	<p>• Expert evaluations of the safety, efficacy and cost-effectiveness of pharmaceutical and medical devices, prior to marketing approval or reimbursement listing, collectively represent a globally important public good. The scientific processes involved play a major role in protecting the public from product risks such as unintended or adverse events, sub-standard production and unnecessary burdens on individual and governmental healthcare budgets.</p><p>• Most States now have an increasing policy interest in this area, though institutional arrangements, particularly in the area of cost-effectiveness analysis of medical devices, are not uniformly advanced and are fragile in the face of opposing multinational industry pressure to recoup investment and maintain profit margins.</p><p>• This paper examines the possibility, in this context, of States commencing negotiations toward bilateral trade agreement provisions, and ultimately perhaps a multilateral Treaty, on safety, efficacy and cost-effectiveness analysis of pharmaceuticals and medical devices. Such obligations may robustly facilitate a conceptually interlinked, but endangered, global public good, without compromising the capacity of intellectual property laws to facilitate local product innovations.</p>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Faunce</surname><given-names>Thomas Alured</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	Globalization and Health	<sec><title>Background: regulating the global medicines and medical devices industries</title><p>The global market for "innovative" pharmaceuticals and medical devices has become one of the most significant sectors for government healthcare spending, particularly as higher corporate rents are leveraged from elevated intellectual property standards[<xref ref-type="bibr" rid="B1">1</xref>]. Its influence on public policy is set to expand exponentially, as the products involved are innovatively re-shaped by nano and gene technology and priced accordingly[<xref ref-type="bibr" rid="B2">2</xref>]. Aging populations and normal profit-seeking behaviour by multinational corporate manufacturers and private insurers, in a regulatory environment with diminished government controls, are also likely to be major factors[<xref ref-type="bibr" rid="B3">3</xref>].</p><p>"Medicines" may be divided into subcategories depending on whether they are available to the public by physician prescription or over-the-counter pharmacy sales, have synthetic or biologic components, are patented or generic, or are complementary (outside the traditional medical evidence base) in nature[<xref ref-type="bibr" rid="B4">4</xref>]. The term "medical device" has been defined in various terms by regulatory agencies, but generally refers to any instrument, apparatus, appliance, or related article that is intended for use in the diagnosis, prevention, monitoring, treatment, or alleviation of disease, or is intended to affect the structure or function of the human anatomy[<xref ref-type="bibr" rid="B5">5</xref>].</p><p>Efficacy and safety evaluation are now routine initial regulatory hurdles in most nations for any newly created prescription medicine and medical device. Animal studies (particularly for teratogenicity, carcinogenicity and mutagenicity) and then three phase human clinical trial data, are widely used for institutional approval (licensing or registration) of pharmaceuticals and a variety of other sources for post-approval surveillance[<xref ref-type="bibr" rid="B6">6</xref>].</p><p>As shall be discussed in more detail, nations such as Canada, Australia, New Zealand and the UK, possess institutions that have achieved international recognition for excellence in cost-effectiveness analysis of pharmaceuticals ("CEAP") as a final component of safety and efficacy evaluation ("SE/CEAP")[<xref ref-type="bibr" rid="B7">7</xref>]. The literature and institutional arrangements for cost-effectiveness analysis of medical devices ("CEAMD") after safety and efficacy approval ("SE/CEAMD"), is much less developed[<xref ref-type="bibr" rid="B8">8</xref>]. This article will discuss some significant recent industry challenges to such processes.</p><p>International benchmark organizations for medicines and medical devices safety and efficacy evaluation, such as the US Food and Drug Administration ("FDA") have also recently come under intense public and governmental scrutiny for perceived inadequacies and conflicts of interest[<xref ref-type="bibr" rid="B9">9</xref>]. Additional concerns in this area are corporate-lead international harmonisation processes in safety and efficacy evaluation of medical devices, that appear to undermine the precautionary principle by shifting the burden of proof to public authorities post marketing approval[<xref ref-type="bibr" rid="B10">10</xref>]. Given that such regulatory processes are under pressure from multinational industry interests, this article explores whether the most efficient public or State response may be to work toward a multilateral treaty in this area.</p></sec><sec><title>The global spread of medical safety, efficacy and cost-effectiveness analysis</title><p>Increasing international interest exists in CEAP prior to government reimbursement as a necessary value approval stage after safety and efficacy evaluation[<xref ref-type="bibr" rid="B11">11</xref>]. Australia was one of the first nations to embrace this concept, through Pharmaceutical Benefits Scheme ("PBS") guidelines, in the early 1990s[<xref ref-type="bibr" rid="B12">12</xref>]. The resultant processes, operating under the aegis of Australia's Pharmaceutical Benefits Advisory Committee ("PBAC"), are now widely regarded as giving Australia world class expertise in the area[<xref ref-type="bibr" rid="B13">13</xref>]. They have a major role in implementing the National Medicines Policy ("NMP") 2000, the four central objectives of which are: timely access to the medicines that Australians need, at a cost individuals and the community can afford; medicines meeting appropriate standards of quality, safety and efficacy; quality use of medicines; and maintaining a responsible and viable medicines industry[<xref ref-type="bibr" rid="B14">14</xref>]. A major advantage of the Australian system, in that the monopsony buying power of the Federal government can build on CEAP prior to Federal formulary listing to achieve major price reductions from industry[<xref ref-type="bibr" rid="B15">15</xref>].</p><p>The New Zealand Pharmaceutical Management Agency ("PHARMAC") was originally established under the Health <italic>and Disabilities Services Act </italic>(1993) (NZ) (now the <italic>Public Health and Disability Act 2000 </italic>(NZ)) with the specific purpose of improving the management of Government expenditure on pharmaceuticals already approved on safety and efficacy grounds. PHARMAC, with the assistance of independent medical experts on the Pharmacology and Therapeutics Advisory Committee ("PTAC") and its specialist sub-committees, manages, on cost-effectiveness grounds set out in guidelines, a Federal formulary, known as the Pharmaceutical Schedule. Patients and their advocacy groups have input into PHARMAC's listing decisions through a Consumer Advisory Committee. One of its major advances involves the use of tendering for low cost generic medicines[<xref ref-type="bibr" rid="B16">16</xref>].</p><p>Cost-effectiveness evaluation was introduced as a interrelated evaluation with safety and efficacy approval, by the Canadian provinces of Ontario[<xref ref-type="bibr" rid="B17">17</xref>]. and British Columbia in the early 1990's[<xref ref-type="bibr" rid="B18">18</xref>]. The Canadian Expert Drug Advisory Committee ("CEDAC") now operates under the Coordinating Office for Health Technology Assessment ("CCOHTA") to create cost-effectiveness recommendations for ten provincial and three territory governments, as well as specific Federal programs (for example, veterans and also indigenous people)[<xref ref-type="bibr" rid="B19">19</xref>]. The Canadian Patented Medicines Prices Review Board ("PMPRB") sets a maximum "factory gate" price for new, patented "breakthrough" drugs, based on the median price in seven OECD nations specified in regulations (France, Germany, Italy, Sweden, Switzerland, U.K. and the U.S.), most of which (apart from the US) rely on some form of CEAP to guide government reimbursement decisions. The PMPRB attempts to also ensure that most new patented drug prices are limited to those of comparable pharmaceuticals sold in Canada and that existing patented drug prices in that nation cannot increase by more than the Consumer Price Index (CPI), or become the highest in the world[<xref ref-type="bibr" rid="B20">20</xref>]. Although it does not advertise the fact, the PMPRB appears to utilise a form of CEAP[<xref ref-type="bibr" rid="B21">21</xref>].</p><p>In Europe safety and efficacy considerations fall within the European Medicines Agency Guidelines on Risk Management Systems for Medicinal Products for Human Use[<xref ref-type="bibr" rid="B22">22</xref>]. Governments in most OECD countries (as well as those mentioned above) utilise forms of CEAP in conjunction with safety and efficacy evaluations[<xref ref-type="bibr" rid="B23">23</xref>]. Belgium, Finland, Norway, Portugal and Sweden have introduced formal cost-effectiveness as a routine "fourth hurdle" after quality, safety and efficacy determination[<xref ref-type="bibr" rid="B24">24</xref>]. The Hungarian Office of Health Technology Assessment of the National Institute for Strategic Health Research has a mandatory role in granting social insurance subsidies related to medicines and medical devices[<xref ref-type="bibr" rid="B25">25</xref>]. The resultant expert recommendation may allow the creation of formularies for either positive or negative government reimbursement of pharmaceutical prices[<xref ref-type="bibr" rid="B26">26</xref>]. As well as cost-effectiveness, cost-utility, cost-benefit and cost-minimisation approaches are utilised[<xref ref-type="bibr" rid="B27">27</xref>]. CEAP is often linked with reference pricing, which may involve a government reimbursing the average or lowest price in a therapeutic grouping of prescription medicines[<xref ref-type="bibr" rid="B28">28</xref>]. The UK Pharmaceutical Price Regulation Scheme ("PPRS")[<xref ref-type="bibr" rid="B29">29</xref>]. links government control over manufacturer profits with a negative (non-reimbursed) list and cost-effectiveness guidance from the National Institute of Clinical Excellence ("NICE")[<xref ref-type="bibr" rid="B30">30</xref>]. Though also utilising expert analysis of systematic reviews and modelling, unlike Australia's PBAC, NICE commissions evaluations on classes of drugs, rather than having them performed by submitting corporations for specific products[<xref ref-type="bibr" rid="B31">31</xref>].</p><p>In the United States, safety and efficacy evaluations follow the FDA pharmacovigilance and risk management action plans[<xref ref-type="bibr" rid="B32">32</xref>]. CEAP is less widely utilised in conjunction with safety and efficacy analysis at the Federal level[<xref ref-type="bibr" rid="B33">33</xref>]. The same true in Japan[<xref ref-type="bibr" rid="B34">34</xref>].Industry critics have pointed to methodological flaws such as vague definitions of therapeutic class and the difficulty of obtaining good measures for societal preferences[<xref ref-type="bibr" rid="B35">35</xref>]. Politically dominant private insurance and pharmaceutical corporations have also linked CEAP with claims that indiscriminate, non-evidence-based government charges could impede patient choice concerning "innovative" medicines[<xref ref-type="bibr" rid="B36">36</xref>]. Individual healthcare facilities (with limited bargaining power) in the US are encouraged by industry to develop formularies useful to patient care using managed care guidelines[<xref ref-type="bibr" rid="B37">37</xref>]. A group of States have organised a Drug Evaluation Review Process ("DERP") to assist their managed care plans[<xref ref-type="bibr" rid="B38">38</xref>]. Health Management Organisations ("HMO's") have begun to require pharmaceutical manufacturers to make formulary submissions according to guidelines prepared by the Academy of Managed Care Pharmacy ("AMCP") and increased prominence has been given to the work of the Agency for Health Research and Quality ("AHRQ")[<xref ref-type="bibr" rid="B39">39</xref>]. Increasing prominence has also been given to CEAP performed by the Veterans Health Administration ("VHA") and the Pharmacoeconomics Evaluation Center ("PEC") of the Department of Defence[<xref ref-type="bibr" rid="B40">40</xref>].</p><p>CEAP and CEAMD are emerging fields of academic and health policy interest for China, with the particular aim of reducing the high proportion (44%) of pharmaceutical expenditure in total healthcare expenditure[<xref ref-type="bibr" rid="B41">41</xref>]. The South Korean government has been developing pharmaco-economic guidelines after consultations with experts in Canada and Australia[<xref ref-type="bibr" rid="B42">42</xref>]. In 2001 the Singapore Ministry of Health appointed a Drug Cost Review Task Force to revise cost-effectiveness processes in connection with a Standard Drug List[<xref ref-type="bibr" rid="B43">43</xref>]. In Thailand, three taxation funded public insurance schemes provide a minimum pharmaceutical package through a cost-effectiveness evaluated National List of Essential Drugs[<xref ref-type="bibr" rid="B44">44</xref>]. Malaysia and Pakistan have governments very interested in cost-effectiveness analysis of pharmaceuticals, but evaluations are limited by lack of funding, lack of trained personnel, lack of protected research time, limited access to data and information, poor dissemination and official uptake of research outcomes[<xref ref-type="bibr" rid="B45">45</xref>].</p><p>Developing countries in general frequently lack the resources to train and support officials with the requisite pharmaco-economic expertise to permit interlinked safety, efficacy and CEAP/CEAMD evaluations on an effective, national scale[<xref ref-type="bibr" rid="B46">46</xref>]. To respond to community (and their own employees') social justice concerns about public health problems arising from high intellectual property rents, the multinational pharmaceutical industry has proposed self-regulatory alternatives emphasising pharmaco-philanthropy, public-private partnership initiatives and covert differential pricing[<xref ref-type="bibr" rid="B47">47</xref>]. Many developing nations, such as India, rely upon the World Health Organisation's ("WHO") Essential Medicines List[<xref ref-type="bibr" rid="B48">48</xref>]. This assesses cost of such pharmaceuticals per case, per cure, per month of treatment, per case prevented, per clinical event prevented, or, if possible and relevant, cost per quality-adjusted life year gained[<xref ref-type="bibr" rid="B49">49</xref>].</p><p>The intense recent interest focused on the global problems with safety and cost-effectiveness of pharmaceuticals, has lead to medical devices becoming somewhat of a silent partner in such regulatory discussions. The International Society for Pharmacoeconomics and Outcomes Research ("ISPOR") is attempting to redress this imbalance[<xref ref-type="bibr" rid="B50">50</xref>]. Devices do create unique difficulties, particularly through difficulties obtaining blinded trial data, the skill involvement with diagnosis (they are not therefore fully embodied technologies and have cost-effectiveness learning curves), the frequency of product modifications and poor development of regulatory theory in this area[<xref ref-type="bibr" rid="B51">51</xref>]. The Global Harmonization Task Force (GHTF) comprises representatives from national medical device regulatory authorities and industry from European Union, the United States of America, Canada and Japan was established ostensibly to encourage convergence in safety, efficacy and cost-effectiveness evaluations, whilst also promoting technological innovation and facilitating international trade[<xref ref-type="bibr" rid="B52">52</xref>].</p><p>An important point to note from the above survey is that established and effective forms of CEAP and CEAMD work in close conceptual association with safety and efficacy evaluations. We can now examine whether it may make better regulatory sense to consider these as integrally linked processes.</p></sec><sec><title>Advantages and disadvantages of SE/CEAP and SE/CEAMD</title><p>Affordable access to essential medicines is increasingly recognised as a global public good, providing an essential precondition to a reasonable quality of life for a significant proportion of every human population, being systematically underprovided by private market forces and imposing burdensome international externality costs on third parties[<xref ref-type="bibr" rid="B53">53</xref>]. Further, affordable access to essential medicines appears to be emerging, both academically and in practise, as a core part of the international right to health in article 12 of the <italic>International Covenant on Economic, Cultural and Social Rights </italic>(article 25 of the <italic>Universal Declaration of Human Rights)</italic>[<xref ref-type="bibr" rid="B54">54</xref>]. One recent manifestation was the <italic>Doha Declaration</italic>, which affirmed the capacity of WTO members to use to the full exceptions in the Trade Related Intellectual Property Rights agreement ("TRIPS") to promote public health by facilitating access to affordable medicines[<xref ref-type="bibr" rid="B55">55</xref>]. It is also specifically referred to in article 14 of the UNESCO <italic>Universal Declaration on Bioethics and Human Rights</italic>[<xref ref-type="bibr" rid="B56">56</xref>]. There seems to be little reason why in theory or practice, affordable access to essential medical devices should not to subject to the same considerations.</p><p>SE/CEAP and SE/CEAMD processes, however, despite their value to contemporary health technology assessment and their capacity to facilitate access to medicines, have not themselves been widely discussed as a global public good, or as in any obvious way connected with normative systems of distributive justice and the international human right to health. Neither is primarily regarded as a cost-containment strategy, chiefly because their related formularies generally lack a capped budget and their fiscal effects are predicated on prescribers adhering to recommended indications[<xref ref-type="bibr" rid="B57">57</xref>]. SE/CEAP and SE/CEAMD, create no barriers to market access, or infringements of intellectual property rights. They merely attempt to rationalise, according to scientific evaluation of a hierarchy of clinical trial evidence, government or other third party (private health insurer) reimbursement expenditure [<xref ref-type="bibr" rid="B58">58</xref>].</p><p>SE/CEAP and SE/CEAMD have three key advantages, which may allow them to evolve into an important global public good. The first involves an emphasis on scientific evidence, the second a commitment to equity, to ensuring value for a whole community and the third, the capacity of SE/CEAP and SE/CEAMD to act as fiscal brakes on rent flowing to prior intellectual property owners without inhibiting encouragement of local innovation through high intellectual property protection.</p><p>One of the major disadvantages of SE/CEAP and SE/CEAMD, is the common presence of methodological flaws either in the evaluations by regulators, or in economic submissions made by industry[<xref ref-type="bibr" rid="B59">59</xref>]. SE/CEAMD faces comparative difficulties with "blinding," variable physician technique and a shorter product life cycle. Yet, they may benefit from easier <italic>in vitro </italic>assessment and a greater capacity to characterise incremental design changes by laboratory bench testing.</p><p>Another disadvantage, from the regulators' point of view, is the lack of "hard" outcome data such as Quality Adjusted Life Years ("QALYs"), particularly at initial evaluation of an innovative product. Manufacturers often claim it is too early to produce such published trial data and prefer to rely on surrogate outcomes, such as readily measured changes in biochemical markers of disease. Another disadvantage is that CEAP and CEAMD analysis is often (unless it is linked to Federal monopsony buying power) unable to question the initial price given by industry. Direct, rather than inferred, evidence of marginal cost of production is denied to evaluators, often on "commercial-in-confidence" grounds. This means that CEAP and CEAMD however excellently performed, often metaphorically take place on an uncertain foundation[<xref ref-type="bibr" rid="B60">60</xref>]. There is also issue of nations training enough pharmaco-economic experts to facilitate CEAP and CEAMD for, for example, both pre and post reimbursement listing.</p><p>SE/CEAP and SE/CEAMD also commonly be "gamed" by industry. If, for example, in a system such as that of Australia, if a safety regulator approves 5 clinical indications, this could lead to submissions to a cost-effectiveness evaluator on only one indication with the industry expectation of prescription "leakage" outside recommendations, compromising fiscal savings for the taxpayer. Similarly, expert evaluations considering a medicine's toxicity may play an important CEAP role by factoring disutility into modelled analysis, calculating compliance, or altering indications.</p><p>Hasty safety approvals could endanger public health, yet heightened industry pressure for "fast-tracking" may arise from diverse sources: prior notification of submission schemes, differing standards of proof, industry applications "salami slicing" indications to fit "orphan" drug categories, by inadequate conflict of interest protections given full cost recovery from industry and pressure for development collaborations with regulators. Over-cautious rejections could delay patient benefits, reduce export earnings and stifle investor confidence; yet safety classifications of innovative nanotechnology products at the device/medicine 'boundary' will be distinctly complex. The public may react adversely to new internationally harmonised medical devices safety regulations that shift burdens of proof to safety regulators after approval, possibly in anticipation of the difficulty in obtaining credible published trial data in this area (recruitment of subjects to nanomedicine safety and cost-effectiveness trials will be unusually difficult). The limited published systematic reviews, may unduly restrict SE/CEAP and SE/CEAMD for nanotechnology products to surrogate outcome measures, rather than quality-adjusted life years.</p></sec><sec><title>Threats from global industry interests</title><p>Though well entrenched in the policies of most States, evolution and enhancement of SE/CEAP and SE/CEAMD as a global public good should not be taken for granted. Brand name pharmaceutical multinationals, in particular, are currently involved in a global strategy, using international trade arrangements, carefully funded and seeded academic articles, strategic surveys of relevant processes in Europe and Asia (and how well they respond to the corporate lobbying principle of innovation), to separate cost-effectiveness analysis from safety and efficacy evaluations and central government monopsony buying power and replace it with medicines provision models emphasising privatised insurance,[<xref ref-type="bibr" rid="B61">61</xref>]. medicines savings accounts[<xref ref-type="bibr" rid="B62">62</xref>]. and direct-to-consumer advertising[<xref ref-type="bibr" rid="B63">63</xref>]. This process has already produced large scale adverse public health consequences in China[<xref ref-type="bibr" rid="B64">64</xref>]. and Singapore[<xref ref-type="bibr" rid="B65">65</xref>]. Nevertheless it is still being promoted by industry as a credible policy alternative to universal taxpayer-funded access schemes in developed nations such as Australia, usually in the guise of enhancing "consumer" choice and responsibility[<xref ref-type="bibr" rid="B66">66</xref>]. Critics point to the lack of logic or compassion in industry emphasising the decision-making capacity of sick people, particularly the disabled and poor patients, concerning their health and therapies, as if what they were purchasing was a new car, house, or suit of clothes.</p><p>The United Nations Human Development Report 2005 has emphasised, for example, that the World Trade Organisation's ("WTO's") corporate-sponsored Trade-Related Intellectual Property Rights (TRIPS) agreement, along with so-called "TRIPS-Plus" intellectual property protections in subsequent bilateral trade agreements, pose a "pronounced" threat to global public health, particularly through their expansive effect on prices for so-called "innovative" medicines[<xref ref-type="bibr" rid="B67">67</xref>]. The US pharmaceutical industry also has a powerful influence on the globally influential US legislature[<xref ref-type="bibr" rid="B68">68</xref>]. The <italic>Medicare Prescription Drug Improvement and Modernization Act </italic>2003 (US), as one instance, thwarted attempts to introduce a Federal PBAC-type process in the US, specifically prohibiting the US government from using its bulk buying power for Medicare beneficiaries from negotiating medicines price discounts in a PBAC-style approach[<xref ref-type="bibr" rid="B69">69</xref>]. A Congressional Conference Agreement on this legislation obligated US negotiators on the AUSFTA to report on whether that deal offered opportunities to achieve the objectives of the <italic>Bipartisan Trade Authority Act 2002 </italic>(US) including the "elimination of government measures such as price controls and reference pricing which deny full market access" for US pharmaceuticals[<xref ref-type="bibr" rid="B70">70</xref>].</p><p>Section 1123 of the <italic>Medicare Prescription Drug Improvement and Modernization Act </italic>2003 (US), commissioned a study by the US Department of Commerce, on so-called pharmaceutical "price controls" implemented by SE/CEAP systems in thirteen OECD countries. It claimed that these cost US drug purchasers from $5–$6 billion per year. It argued that US drug prices should serve as a benchmark for deregulated prices, despite the fact that they are 18–67% higher than those in the relevant OECD countries[<xref ref-type="bibr" rid="B71">71</xref>].</p><p>An important issue here may be the role of Article 64 of the <italic>Agreement on Trade-Related Aspects of Intellectual Property Rights </italic>("TRIPS")[<xref ref-type="bibr" rid="B72">72</xref>]. The United States, for example, subsequently has argued that the initial and subsequent moratoria is over and the Non-Violation-Nullification of Benefits ("NVNB") remedy must now be accepted, by all WTO Members, as applying to the TRIPS Agreement[<xref ref-type="bibr" rid="B73">73</xref>]. At the WTO meeting in Hong Kong in December 2005, United States negotiators attempted to obtain concessions in return for their support for the continuance of the NVNB moratorium[<xref ref-type="bibr" rid="B74">74</xref>]. NVNB claims, permitting dispute resolution proceedings for breaching the "spirit" of a trade agreement could both support and undermine CEAP, depending on the undertakings made about it at the time such agreements were entered. The Australian government, for example, quite explicitly gave undertakings that the fundamental architecture of Australia's CEAP system would not be altered by the AUSFTA[<xref ref-type="bibr" rid="B75">75</xref>] and backed this up by passing implementing legislation against the process of patent "evergreening" predicated on such an assumption. Crucially important in this context could be Annex 2C (1) of the Australia-United States Free Trade Agreement ("AUSFTA,") where "innovation" is uniquely linked with the socially-oriented concepts of 'high quality health care', 'affordability', 'accountability' and "objectively demonstrated therapeutic significance'. Whether "innovation" should sit within CEAP, or the patent system, or both, is a major conceptual conundrum that probably goes to the heart of the industry agenda in this area.</p><p>On 1–2 December 2005, a meeting took place in Paris under the auspices of the OECD "Project on Pharmaceutical Pricing Policies and Innovation." Inclusion of the term "innovation" in the title discloses what was probably the chief purpose of this Project (though attempts were made by the US delegation to obfuscate this agenda, particularly by initial statements ostensibly withdrawing support and ensuring a significant role for nations such as Canada and Mexico). This was to broach the first stages of implementation of the US Department of Commerce report mentioned previously. Its stated terms of reference appear to confirm this. They are:</p><p>1) to add to the base of information about pharmaceutical pricing policy in OECD countries and develop a taxonomy and framework for making international comparisons of policies [the European Union was running a similar investigation already]</p><p>2) to analyze cross-national impacts and implications of policies, particularly with respect to impact on pharmaceutical prices paid in other countries and impact on pharmaceutical research and development[<xref ref-type="bibr" rid="B76">76</xref>].</p></sec><sec><title>Toward a multilateral treaty</title><p>It seems remarkable, in an age of corporate globalisation, that medicines and medical devices national safety regulators and cost-effectiveness evaluators continue to work largely in formal isolation to assess the same products. Given the importance of SE/CEAP and SE/CEAMD to sustainability and legitimacy of public health systems, it is also peculiar that governments have not already perceived the advantages of creating a multilateral treaty in this area.</p><p>One intermediate suggestion is to include provisions establishing SE/CEAP and SE/CEAMD committees or working groups in bilateral trade agreements. The aims of such arrangements would include fostering relevant international regulatory collaborations, capacity building expertise (by facilitating the relevant trade in services) and overcoming regulatory safety concerns that might provide barriers to the entry of cheap generic medicines (for example from China to Australia). Such provisions would not impact adversely on intellectual property rights. Consequently, they would not infringe any prohibitions on restricting intellectual property rights or discriminating against fields of technology emerging from the TRIPS Agreement.</p><p>For each such provision, a government department (usually the respective Ministries of Health) would need to assume responsibility for operationalising the related obligations and requirements. Qualifications and process of appointment of relevant experts would need to be resolved, as would the reporting mechanisms. Establishing such a mechanism in a trade agreement would promote SE/CEAP and SE/CEAMD expertise in relevant universities, building careers in this area, with the prospects of governments saving more money as greater numbers of relevant experts become available to preform both pre and post-listing evaluations.</p><p>Such a provision might be as brief as the following annex at the end of a trade in goods chapter:</p><p>"<bold>Medicines and Medical Devices Safety, Efficacy and Cost-Effectiveness Committee </bold>The Parties hereby establish this Committee, comprising relevant officials and expert advisors from each Party. Its primary objective shall be to promote discussion and mutual understanding, collaborations, training, education and sharing of expertise with a view to enhancing and developing techniques of, and research related to, safety, efficacy and cost-effectiveness evaluations of medicines and medical devices."</p><p>In time, the increased interest in SE/CEAP and SE/CEAMD generated by such provisions may lead to a <italic>Treaty on Safety, Efficacy and Cost-Effectiveness Evaluation of Medicines and Medical Devices</italic>. Such a Treaty could be sponsored either by UNESCO, or the World Health Organisation ("WHO") or, hopefully, both organizations in collaboration.</p><p>The relevant terms of reference could involve negotiations in the following areas:</p><p>1) the appropriate interrelationship of safety, efficacy and cost-effectiveness evaluations</p><p>2) the social theories that should underpin such evaluations including the blance between global public goods and private rights, perspectives on the relative importance and interaction in this context of bioethical equity and social justice, the international human rights to health, international trade norms preventing non-tarriff barriers and industry lobbying principles such as recognition of innovation.</p><p>3) how to improve access by regulators, health professionals, consumers and industry to public data bases of large-scale, randomised, double blind clinical trails involving head to head comparisons using therapeutically equivalent dosage forms for the most commonly prescribed pharmacological analogues or non-drug therapies for the same indication.</p><p>4) whether SE/CEAP and SE/CEAMD can progressively involve greater use of "hard" outcome measures, such as deaths prevented or quality-adjusted life years (QALYs) gained, rather than "surrogate" pharmacological outcomes (for example low density lipoprotein levels or blood pressure).</p><p>5) improving existing SE/CEAP and SE/CEAMD systems efficiencies in specifics such as reference pricing and tendering for ultra low cost generic medicines, but also whether the concept of "innovation" in relation to medicines and medical devices should be defined to include elements of safety, efficacy, affordability and objectively demonstrated therapeutic significance.</p><p>6) discussions on post marketing responsibilities which could include price-volume and binding health outcome agreements between regulators and industry, as well as the appropriate structure of vigilance trials, adverse incident reporting, impact of fraud, prescribing habits and alternative or complementary therapies.</p><p>7) discussions on how to globally capacity build SE/CEAP and SE/CEAMD as a career for health professionals and facilitate trade in services training programmes, expert exchanges and collaborations.</p><p>8) discussions on improving data in areas such as choice of comparitor, measurement of relevant costs and benefits, length of follow up, peculiarities of local setting and appropriate valuation of economic, clinical and patient-reported (or humanistic) outcomes.</p><p>9) negotiations on public interest limits about commercial-in-confidence protections and on disclosing local and international marginal costs of production for each drug. Important principles on the issue of commercial-in-confidence, for example, emerging from the parallel processes of UK NICE and Canadian CCOHTA, are that it should not so inhibit transparency as to prevent manufacturers disclosing enough information to make their submission understandable to the public or governments, or that it should not endanger public safety and should not be set unilaterally by industry[<xref ref-type="bibr" rid="B77">77</xref>].</p><p>10) horizon scanning processes to ensure all Parties are speedily appraised of recommended SE/CEAP and SE/CEAMD regulatory responses to developments in new fields such as nano and gene-based technologies.</p></sec><sec><title>Conclusion</title><p>This article has argued that despite its obvious attraction to fiscally responsible governments in a time of ageing demographics, neither the continuance, nor enhancement of science-based SE/CEAP and SE/CEAMD processes should be taken for granted. Nation states are just becoming used to the change in sovereignty associated with fully privatised healthcare sectors coexisting with international trade obligations to provide national treatment to multinational corporations. In this context, much official concern has been expressed about growing public disenchantment with the policy influence of the multinational pharmaceutical industry[<xref ref-type="bibr" rid="B78">78</xref>].</p><p>There are both responsive and pro-active reasons for seeking to include provisions facilitating SE/CEAP and SE/CEAMD in bilateral and multilateral trade agreements. The responsive reason relates to ensuring a more transparent debate about the future enhancement of these processes in relation to an industry agenda which often appears to perceive their stringent application as an impediment to their freedom to manufacture, obtain speedy safety and efficacy approval and market direct to both patients and physicians, with only limited stringent scientific scrutiny about either the marginal cost of production or overall comparative worth to the community.</p><p>The pro-active reasons for including SE/CEAP and SE/CEAMD in trade agreements relate to the possibility of creating an important, transparent playing field where the next generation of great debates between public goods and private rights in this sector can take place. They also concern the facilitation of trade-in-services, capacity building relevant expertise, improving relevant processes (including the efficiency of sharing data and reviews), as well as the need to commence negotiations with pharmaceutical multinationals on a more rational approach to important issues such as commercial-in-confidence and marginal cost of production.</p><p>Possible disadvantages in proceeding this way include the possibility of such a treaty becoming a lightning rod for a contrary agenda by the pharmaceutical and medical device industries. The aims of such a treaty, for example, could be altered to provide a vehicle for corporate strategies such as "linkage" of regulatory evaluation of a generic pharmaceuticals patent status with quality and safety evaluation prior to marketing approval, or reversal of the precautionary principle with regard to regulatory approval of new medical device technologies.</p><p>At this point in the age of corporate globalisation, perhaps it is time to start respecting scientific cost-effectiveness evaluation of medicines and medical devices as a potentially endangered global public good, which should not be conceptually or operationally separated from safety and efficacy evaluations. Governments wishing to take a popular strategy to elections with an ageing population could promote the type of multilateral treaty discussed here (or provisions facilitating SE/CEAP and SE/CEAMD in bilateral trade deals) as a rational and scientific way of restraining medicines prices and ensuring value for public expenditure in this area of the health sector.</p></sec>
The time-course of a scrapie outbreak	<sec><title>Background</title><p>Because the incubation period of scrapie has a strong host genetic component and a dose-response relationship, it is possible that changes will occur during an outbreak, especially in the genotypes of cases, age-at-onset of disease and, perhaps, the clinical signs displayed. We investigated these factors for a large outbreak of natural scrapie, which yielded sufficient data to detect temporal trends.</p></sec><sec><title>Results</title><p>Cases occurred mostly in two genotypes, VRQ/VRQ and VRQ/ARQ, with those early in the outbreak more likely to be of the VRQ/VRQ genotype. As the epidemic progressed, the age-at-onset of disease increased, which reflected changes in the genotypes of cases rather than changes in the age-at-onset within genotypes. Clinical signs of cases changed over the course of the outbreak. As the epidemic progressed VRQ/VRQ and VRQ/ARQ sheep were more likely to be reported with behavioural changes, while VRQ/VRQ sheep only were less likely to be reported with loss of condition.</p></sec><sec><title>Conclusion</title><p>This study of one of the largest scrapie outbreaks in the UK allowed investigation of the effect of PrP genotype on other epidemiological parameters. Our analysis indicated that, although age-at-onset and clinical signs changed over time, the observed changes were largely, but not exclusively, driven by the time course of the PrP genotypes of cases.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>McIntyre</surname><given-names>K Marie</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Gubbins</surname><given-names>Simon</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Goldmann</surname><given-names>Wilfred</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Stevenson</surname><given-names>Emily</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A5" corresp="yes" contrib-type="author"><name><surname>Baylis</surname><given-names>Matthew</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib>	BMC Veterinary Research	<sec><title>Background</title><p>Scrapie is an infectious disease, but one that is unusual in that the incubation period leading to clinical disease is governed by a strong host genetic component [<xref ref-type="bibr" rid="B1">1</xref>] and a strong dose-response relationship [<xref ref-type="bibr" rid="B2">2</xref>]. These factors can interact to yield intriguing epidemiological patterns.</p><p>During an outbreak, the PrP genotypes of affected animals are expected to change over time. Those encoding PrP genotypes with the shortest incubation periods should be the first to reach clinical onset, followed by their co-occurrence with those encoding PrP genotypes with longer incubation periods, then those with the longest incubation periods. If most sheep are infected at a similar age, this could manifest as a change in the age-at-onset of scrapie over time. The level of infectivity to which susceptible animals are exposed could also increase through the epidemic as the cumulative total of infected animals rises. A higher dose may shorten the incubation period and lead to a reduction in the age-at-onset over time. If PrP genotype, incubation period or age-at-onset change with time during an outbreak, then any linkage between any of these factors and the clinical signs of scrapie could also lead to changes in the clinical manifestation of the disease over time. A time course of clinical signs may also arise from the farmer, who may be able to recognise scrapie on the basis of fewer or different clinical signs as more cases are encountered.</p><p>These ideas are supported by the reports of farmers running commercial flocks affected by natural scrapie. In a field study undertaken by the Institute for Animal Health (IAH), twenty-four percent of farmers reported a change in the clinical signs being reported through the course of a scrapie epidemic, while thirty-five percent reported a change in the ages of cases.</p><p>Detection of a time trend, and teasing apart the causal factors behind it, is made difficult by the small number of cases confirmed in most outbreaks and the range of PrP genotypes involved. Furthermore, variability between flocks in the characteristics of outbreaks precludes, in most cases, the combining of datasets.</p><p>Here, we have investigated in detail the scrapie outbreak of one of very few flocks for which sufficient data are available for time course analysis. Between 1997 and 2004 there were over 130 confirmed cases of scrapie in a single flock of Welsh Mountain sheep. Almost all cases were in just two genotypes. This outbreak is possibly the largest reported for any flock in Great Britain over the same time period. The scale of the outbreak gave us an opportunity to explore in detail the time course of the scrapie epidemic at the level of PrP genotype and with extensive detail of the clinical presentation of disease.</p></sec><sec><title>Results</title><sec><title>The flock</title><p>Ten (out of 15) PrP genotypes were found in this flock upon blood sampling in 2001 (Table <xref ref-type="table" rid="T1">1</xref>). A quarter (25.2%) carried the VRQ allele (associated with the highest risk of scrapie), but only 0.5% were of the most susceptible genotype, VRQ/VRQ. The genotype frequencies correspond to allele frequencies of: ARR, 39.9%; AHQ, 14.7%; ARQ, 32.2%; and VRQ, 13.2%. The ARH allele was not detected in this flock. Animals also had PrP haplotypes ARQ-F<sup>141 </sup>and ARQ-S<sup>241 </sup>which occurred at frequencies of 21.5% and 14.2%, respectively. Neither of these haplotypes was associated with the occurrence of scrapie in this flock.</p></sec><sec><title>The outbreak</title><p>The farmer did not know at what point his flock acquired infection. Although he thought that there had been one case of scrapie (unconfirmed) in 1982, he did not recognise any more animals with scrapie-like signs until the start of the current epidemic in September 1997. The affected animals exhibited clinical signs of incoordination (ataxia), nervousness/excitability (nervous) and rubbing/scratching (pruritis); he thought that both the clinical signs and the age of the animals succumbing to disease changed during the course of the epidemic.</p><p>Between September 1997 and February 2004 there were 131 confirmed cases in the flock, in animals of 3 genotypes (Table <xref ref-type="table" rid="T1">1</xref>). The frequency of cases increased to a peak in 2000 and subsequently decreased (Figure <xref ref-type="fig" rid="F1">1</xref>). The number of cases in 2001 appears unusually low; this may be related to difficulties in reporting scrapie during the UK's foot-and-mouth disease epidemic. More cases were reported in the first than the second half of each year, but seasonal trends were not significant (<italic>t</italic><sub>13 </sub>= 0.54, <italic>P </italic>= 0.59).</p><p>Most cases were initially in animals of the VRQ/VRQ genotype, which peaked in 1999, after which cases became more prevalent in animals of the VRQ/ARQ genotype, which peaked in 2000 (Figure <xref ref-type="fig" rid="F1">1</xref>). Three cases also occurred in animals of the less-susceptible VRQ/ARR genotype during the decline phase of the epidemic (Figure <xref ref-type="fig" rid="F1">1</xref>). Considering only the time from when the flock was blood sampled, the proportion of animals which subsequently died from scrapie were: 2.1% (2/96) in VRQ/ARR; 27.0% (17/63) in VRQ/ARQ; and 25.0% (1/4) in VRQ/VRQ.</p></sec><sec><title>The time-course</title><p>As predicted, the genotype of cases changed during the time-course of the outbreak (Figure <xref ref-type="fig" rid="F2">2A</xref>). Earlier in the epidemic cases were more likely to be of the VRQ/VRQ than the VRQ/ARQ genotype (AOR = 0.93; 95% CI: 0.90–0.98). Exploratory analysis suggested that animals of the VRQ/ARR genotype were likely to exhibit clinical signs later in the epidemic than either VRQ/ARQ or VRQ/VRQ animals (Figure <xref ref-type="fig" rid="F2">2A</xref>), but their small sample size precluded more detailed analyses.</p><p>As observed by the farmer, cases which occurred later in the epidemic tended to be older than those seen earlier (Figure <xref ref-type="fig" rid="F2">2A</xref>; <italic>F</italic><sub>1,129 </sub>= 28.951, <italic>P </italic>< 0.001). Cases of the VRQ/VRQ genotype developed clinical signs at an earlier age than VRQ/ARQ animals (Figure <xref ref-type="fig" rid="F2">2B</xref>; log-rank test: χ<sup>2 </sup>= 16.90, df = 1, <italic>P </italic>< 0.001), but age within each genotype did not change significantly during the outbreak (<italic>P </italic>= 0.329). This implies that the increase in age-at-onset was due to changes in the PrP genotype of cases rather than changes in age within each genotype.</p></sec><sec><title>Clinical signs</title><p>Analysis was undertaken for those clinical signs with sufficient observations (n>25): ataxia, change of behaviour, fleece loss, loss of condition, pruritus and trembling. As the epidemic progressed, animals were more likely to be reported showing a change in behaviour, independent of PrP genotype (AOR = 1.09; 95% CI: 1.05–1.13). VRQ/VRQ animals were less likely to show loss of condition (AOR = 0.77; 95% CI: 0.62–0.97) or fleece loss (AOR = 0.83; 95% CI: 0.72–0.94) as the outbreak progressed. This pattern was not seen in VRQ/ARQ cases. Throughout the outbreak, older cases of any genotype were more likely to show ataxia (AOR = 1.03; 95% CI: 1.00–1.06) and trembling (AOR = 1.06; 95% CI: 1.02–1.09) than younger ones. Finally, animals exhibiting fleece loss were more likely to be of the VRQ/VRQ rather than the VRQ/ARQ genotype (AOR = 3.29; 95% CI: 1.21–8.95).</p></sec></sec><sec><title>Discussion</title><p>Although there have been several studies describing within-flock outbreaks of scrapie [<xref ref-type="bibr" rid="B3">3</xref>-<xref ref-type="bibr" rid="B7">7</xref>] the results of this paper represent the first analysis addressing confounding between the effects of time since the beginning of the epidemic, age at onset and PrP genotype. The major limitations of this study are that we do not know when the flock first became infected with scrapie; and genotypes were unavailable for 16 cases, including the first 14 during the epidemic.</p><p>An interesting feature of this outbreak is that cases occurred in the VRQ/VRQ, VRQ/ARQ and VRQ/ARR genotypes only, despite the high frequency of sheep of the ARQ/ARQ and VRQ/AHQ genotypes in the flock (Table <xref ref-type="table" rid="T1">1</xref>). These findings confirm the resistance of the VRQ/AHQ genotype to classical scrapie [<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B9">9</xref>] and the resistance of the ARQ/ARQ genotype to certain scrapie strains in some sheep breeds [<xref ref-type="bibr" rid="B10">10</xref>,<xref ref-type="bibr" rid="B11">11</xref>].</p><p>A key finding of this study is that the cases which occurred later in the outbreak tended to be older. Our analysis has shown that this is not related to changes in the age-at-onset within each genotype during the outbreak. Rather, it is an effect of the PrP genotype of cases, which influences the age at which animals succumb to disease. Our results differ from an earlier study, which reported a decline in the age-at-onset for four outbreaks [<xref ref-type="bibr" rid="B7">7</xref>], though the decline was significant (P < 0.001) in only one outbreak in a flock of Suffolk sheep [<xref ref-type="bibr" rid="B4">4</xref>]. This Suffolk flock was bred to maximise the incidence of disease; hence, susceptible animals were constantly being introduced to the flock and exposed to an increasing load of infection. By contrast, most, though not all, susceptible animals in our study flock succumbed to scrapie, but their susceptible alleles were not replaced. The resulting pattern of the outbreak means that animals of the same genotype were, on average, exposed to similar loads of infection.</p><p>Animals of the VRQ/VRQ genotype had a younger age-at-onset than VRQ/ARQ animals (Figure <xref ref-type="fig" rid="F2">2</xref>), in common with other studies [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B10">10</xref>]. The age at onset of clinical signs depends on a number of factors, including the infectious load, age-dependent exposure, age-dependent susceptibility and incubation period. Modelling analysis of an outbreak in a Cheviot flock suggested that the incubation period for VRQ/VRQ animals was shorter than that for VRQ/ARQ and that there was evidence for age-dependent susceptibility [<xref ref-type="bibr" rid="B12">12</xref>]. This suggests that the differences in the age-at-onset are likely to reflect differences in incubation period rather than age-dependent exposure. More detailed analyses are needed, however, to tease apart the interactions of various age and genotype effects in the age-at-onset of clinical signs.</p><p>Our study is the first to report changes in clinical signs during a scrapie outbreak. Animals were more likely to be reported showing a change of behaviour as the epidemic progressed, irrespective of PrP genotype. One hypothesis would be that at the start of the epidemic the farmer required physical signs to identify scrapie, but as the epidemic progressed he became better able to detect disease on the basis of more subtle behavioural changes. Indeed, behavioural changes are reported prior to the onset of other clinical signs in both goats [<xref ref-type="bibr" rid="B13">13</xref>] and mice [<xref ref-type="bibr" rid="B14">14</xref>].</p><p>Temporal changes in clinical signs also differed between PrP genotypes. Early in the epidemic, VRQ/VRQ animals were more likely to show loss of condition and fleece loss/change, while those reported later in the epidemic were less likely to do so. This could be related to the aforementioned greater reliance of the farmer on physical (as opposed to behavioural) signs earlier in the outbreak; however, this seems unlikely, as the pattern did not occur in animals of the VRQ/ARQ genotype. Alternatively the effect may be real, or it may reflect variation in the clinical signs recorded by the seven reporting officers. Exploratory analysis provided some support for this latter hypothesis, but there were too few cases to be able to test this rigorously.</p></sec><sec><title>Conclusion</title><p>In this paper we have described one of the largest scrapie outbreaks in the UK. As the cases were almost exclusively in two genotypes, this has facilitated the investigation of the effect of PrP genotype on other epidemiological parameters. Although age-at-onset and clinical signs changed over time, our analysis indicated that the effect was largely, but not exclusively, driven by the time course in the PrP genotype of cases.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Flock details</title><p>A flock comprising sheep predominantly of the Welsh Mountain breed and numbering about 850 breeding animals was recruited into the aforementioned IAH scrapie field study. The first cases of scrapie within this flock were confirmed in September 1997. As of February 2004, there were 131 confirmed cases of scrapie. The entire breeding flock (n = 829) was blood sampled for PrP genotyping in August 2001.</p></sec><sec><title>Identification of confirmed cases</title><p>The data for scrapie cases within the study flock were retrieved from the Scrapie Notification Database (SND) held at the Veterinary Laboratories Agency (VLA). This includes the breed, sex, date of birth, date of death, PrP genotype and clinical signs in animals submitted as suspect scrapie cases. Tissue samples are also obtained from cases, which are subject to routine analysis for evidence of scrapie.</p></sec><sec><title>Breed of cases</title><p>The confirmed cases in the flock were a mixture of pure-bred (n = 117) and cross-bred Welsh Mountain (n = 4), Welsh Half-bred (n = 3) and Beulah Speckled Face (n = 7). Because of the small number of cases in most breeds and crossbreeds within the flock, breed was not considered further in the analyses.</p></sec><sec><title>PrP genotype analysis</title><p>Susceptibility to scrapie is strongly associated with polymorphisms in the prion protein (PrP) gene [<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B9">9</xref>,<xref ref-type="bibr" rid="B15">15</xref>], of which five alleles are commonly found (defined by amino acids at codons 136, 154 and 171): VRQ, ARQ, ARH, AHQ and ARR [<xref ref-type="bibr" rid="B10">10</xref>,<xref ref-type="bibr" rid="B16">16</xref>]. PrP genotype analysis was undertaken using approximately 5 ml of blood collected into an EDTA-vacutainer from each sheep, as described previously [<xref ref-type="bibr" rid="B17">17</xref>].</p></sec><sec><title>Clinical signs</title><p>The clinical signs used within analyses were those observed and recorded by the reporting officers who visited the farm when notified by the farmer of a suspect case. Data on clinical signs were available for 120 out of 131 confirmed cases; recorded signs included abnormal head position, ataxia, change of behaviour (or temperament), biting, dribbling (of saliva or cud), fleece loss (or change), loss of condition, nervousness, nibbling reflex, pruritus, teeth grinding, trembling and visual impairment.</p></sec><sec><title>Statistical methods</title><p>'Age' in months relates to when clinical signs were reported; 'time' in months was from the first confirmed scrapie case (i.e. the start of the reported epidemic). Analyses were restricted to confirmed clinical cases of scrapie.</p><p>A binary logistic regression model was used to examine the effects of time and age upon the PrP genotype of cases. Cox proportional hazard models, with age as the survival measure, were used to analyse the effect of time and PrP genotype upon age [<xref ref-type="bibr" rid="B18">18</xref>]. Binary logistic regression models were used to investigate the effect of time, age and PrP genotype on the presence or absence of clinical signs. Statistical significance was determined by a <italic>P</italic>-value of less than 0.05. Models were derived by step-wise deletion of insignificant main effects, excluding those within significant interaction terms. Within logistic regression models, the adjusted odds ratios (AOR) and 95% confidence intervals (CI) were calculated for significant independent variables using standard methods.</p></sec></sec><sec><title>Authors' contributions</title><p>KMM carried out the statistical analyses and drafted the manuscript. SG carried out the statistical analyses and critically revised the manuscript. WG and ES undertook the PrP genotyping. MB conceived the study and critically revised the manuscript. All authors read and approved the final manuscript.</p></sec>
Evaluation of the Edinburgh Post Natal Depression Scale using Rasch analysis	<sec><title>Background</title><p>The Edinburgh Postnatal Depression Scale (EPDS) is a 10 item self-rating post-natal depression scale which has seen widespread use in epidemiological and clinical studies. Concern has been raised over the validity of the EPDS as a single summed scale, with suggestions that it measures two separate aspects, one of depressive feelings, the other of anxiety.</p></sec><sec sec-type="methods"><title>Methods</title><p>As part of a larger cross-sectional study conducted in Melbourne, Australia, a community sample (324 women, ranging in age from 18 to 44 years: mean = 32 yrs, SD = 4.6), was obtained by inviting primiparous women to participate voluntarily in this study. Data from the EPDS were fitted to the Rasch measurement model and tested for appropriate category ordering, for item bias through Differential Item Functioning (DIF) analysis, and for unidimensionality through tests of the assumption of local independence.</p></sec><sec><title>Results</title><p>Rasch analysis of the data from the ten item scale initially demonstrated a lack of fit to the model with a significant Item-Trait Interaction total chi-square (chi Square = 82.8, df = 40; p < .001). Removal of two items (items 7 and 8) resulted in a non-significant Item-Trait Interaction total chi-square with a residual mean value for items of -0.467 with a standard deviation of 0.850, showing fit to the model. No DIF existed in the final 8-item scale (EPDS-8) and all items showed fit to model expectations. Principal Components Analysis of the residuals supported the local independence assumption, and unidimensionality of the revised EPDS-8 scale. Revised cut points were identified for EPDS-8 to maintain the case identification of the original scale.</p></sec><sec><title>Conclusion</title><p>The results of this study suggest that EPDS, in its original 10 item form, is not a viable scale for the unidimensional measurement of depression. Rasch analysis suggests that a revised eight item version (EPDS-8) would provide a more psychometrically robust scale. The revised cut points of 7/8 and 9/10 for the EPDS-8 show high levels of agreement with the original case identification for the EPDS-10.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Pallant</surname><given-names>Julie F</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Miller</surname><given-names>Renée L</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Tennant</surname><given-names>Alan</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib>	BMC Psychiatry	<sec><title>Background</title><p>The Edinburgh Postnatal Depression Scale (EPDS) [<xref ref-type="bibr" rid="B1">1</xref>] is a 10 item self-rating post-natal depression scale which was developed almost twenty years ago. Its 10 polytomous items are summated to an overall score ranging from 0–30, with cut points to identify the likely presence of depression. It has seen widespread use in epidemiological and clinical studies [<xref ref-type="bibr" rid="B2">2</xref>-<xref ref-type="bibr" rid="B4">4</xref>]. Although originally intended as a measure of depression, a number of authors have speculated that it may be measuring something more general. Green [<xref ref-type="bibr" rid="B5">5</xref>] suggests that, given its high correlation with a variety of other measures, the EPDS may be measuring what she refers to as postnatal 'dysphoria' (p.153).</p><p>Concern has also been raised over the validity of the EPDS as a single unidimensional summed scale, with suggestions that it measures two separate aspects, one of depressive feelings, the other of anxiety [<xref ref-type="bibr" rid="B6">6</xref>-<xref ref-type="bibr" rid="B9">9</xref>]. Brouwers et al. [<xref ref-type="bibr" rid="B7">7</xref>], for example, identified two subscales using exploratory factor analysis (EFA), representing anxiety (items 3,4,5) and depression (items 1,2,8), with the two subscales showing only a moderate correlation of .37. This result was confirmed by Jomeen and Martin [<xref ref-type="bibr" rid="B10">10</xref>] who also found a separation of anxiety and depression items in both EFA and confirmatory factor analysis (CFA). CFA assessment of the unidimensional model as proposed by Cox recorded the worst model fit statistics, when compared with the alternative multidimensional models proposed by Brouwers et al. [<xref ref-type="bibr" rid="B7">7</xref>] and Ross et al. [<xref ref-type="bibr" rid="B8">8</xref>]. In each of these studies item 10 (<italic>The thought of harming myself has occurred to me</italic>) loaded on a third and separate factor and was not included in subsequent analyses. A three factor solution was identified in a study by Chabrol and Teissedre[<xref ref-type="bibr" rid="B11">11</xref>], distinguishing anxiety, depressive mood and anhedonia.</p><p>The majority of scales in the health and social sciences have been developed using traditional psychometric approaches involving the assessment of validity and reliability [<xref ref-type="bibr" rid="B12">12</xref>]. Construct validity has often been supported through factor analytic techniques which confirm the presence of one or more valid unidimensional scales. Unfortunately, rating scales give ordinal data which fail to meet the assumptions of parametric factor analysis, and it is known that the misuse of the technique can lead to incorrect interpretations [<xref ref-type="bibr" rid="B13">13</xref>]. However these traditional techniques are now being complemented and, in some cases replaced, by Item Response Theory approaches and particularly by the application of the Rasch measurement model [<xref ref-type="bibr" rid="B14">14</xref>-<xref ref-type="bibr" rid="B19">19</xref>].</p><p>This paper examines the potential contribution of Rasch analysis in understanding measurement issues associated with the EPDS. In particular it addresses the question: does the scale provide a psychometrically valid single unidimensional measure of post natal depression, or are there two subscales within its ten items? [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B9">9</xref>]. In addition it explores the appropriateness of the response format used, and assesses the potential bias of items by age and educational level of the mother, and age of the child.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Participants</title><p>A total of 324 women, ranging in age from 18 to 44 years (mean = 32 yrs, SD = 4.6), participated in this study. The age of women's babies at the time of completing the questionnaire ranged from 6 weeks to 6 months, with a mean age of 13 weeks (SD = 5.0). The majority of women (94%) were married (n = 248) or in a defacto relationship (n = 59), with 9 women (2.8%) in a non-cohabiting relationship, 5 women (1.5%) were single, 2 women (0.6%) were divorced, and 1 woman (0.3%) was widowed. 103 women (31.9%), reported having had no tertiary education, 107 women (33.1%) had completed undergraduate university degrees, and 113 women (34.8%) had completed postgraduate university degrees.</p></sec><sec><title>Procedure</title><p>As part of a larger cross-sectional study[<xref ref-type="bibr" rid="B20">20</xref>] conducted in Melbourne, Australia, a community sample was obtained by inviting primiparous women (recruited primarily through mothers' groups) to participate voluntarily in this study. The study was approved by the Swinburne University of Technology Ethics Committee. Women were asked to complete a questionnaire and return it via post, with no identifying information included. In order to reduce the potential confounds of additional children, criteria for inclusion limited participants to first time mothers with no step or foster children. Participants were required to be between 6 weeks and 6 months postnatal.</p></sec><sec><title>Rasch analysis</title><p>Data are fitted to the Rasch model using the RUMM2020 software [<xref ref-type="bibr" rid="B21">21</xref>]. According to Linacre [<xref ref-type="bibr" rid="B22">22</xref>] if a scale is well targeted (i.e. 40–60% endorsement rates on dichotomous test items) then a sample size of 108 will give 99% confidence of the person estimate being within ± 0.5 logits. If the scale is not well targeted (i.e. < 15% or > 85% endorsement rate), then the sample size required for accurate estimation increases to 243. Consequently the sample size of 324 women in the current study is large enough to give good precision, regardless of the targeting of the sample (the relationship between the distribution of persons and the distribution of items on the same metric scale).</p><p>The Rasch methodology adopted in this study is described in detail elsewhere [<xref ref-type="bibr" rid="B23">23</xref>]. Briefly, the Rasch model [<xref ref-type="bibr" rid="B24">24</xref>] is seen as a template which puts into operation the formal axioms which underpin additive conjoint measurement [<xref ref-type="bibr" rid="B25">25</xref>]. Dichotomous [<xref ref-type="bibr" rid="B24">24</xref>] and polytomous [<xref ref-type="bibr" rid="B26">26</xref>] versions of the model are available and a further variant of the latter, which is used in this paper, is known as the partial credit model [<xref ref-type="bibr" rid="B27">27</xref>]:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M1" name="1471-244X-6-28-i1" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:msub><mml:mi>P</mml:mi><mml:mrow><mml:mi>n</mml:mi><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow></mml:msub></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:msub><mml:mi>P</mml:mi><mml:mrow><mml:mi>n</mml:mi><mml:mi>i</mml:mi><mml:mi>j</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:msub></mml:mrow></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>=</mml:mo><mml:msub><mml:mi>θ</mml:mi><mml:mi>n</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:msub><mml:mi>b</mml:mi><mml:mrow><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow></mml:msub></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacyGGSbaBcqGGUbGBdaqadaqaamaalaaabaGaemiuaa1aaSbaaSqaaiabd6gaUjabdMgaPjabdQgaQbqabaaakeaacqaIXaqmcqGHsislcqWGqbaudaWgaaWcbaGaemOBa4MaemyAaKMaemOAaOMaeyOeI0IaeGymaedabeaaaaaakiaawIcacaGLPaaacqGH9aqpiiGacqWF4oqCdaWgaaWcbaGaemOBa4gabeaakiabgkHiTiabdkgaInaaBaaaleaacqWGPbqAcqWGQbGAaeqaaaaa@4939@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>where P<sub><italic>nij </italic></sub>is the probability that person <italic>n </italic>will answer affirm category <italic>j </italic>of item <italic>i </italic>[or be able to do the level of a task specified by that category within the item], <italic>θ </italic>is person ability, and <italic>b </italic>is the item difficulty parameter. From this, the expected pattern of responses to an item set is determined given the estimated <italic>θ </italic>and b. The expected pattern is a probabilistic form of Guttman scaling [<xref ref-type="bibr" rid="B28">28</xref>], and a variety of fit statistics determine if this is the case [<xref ref-type="bibr" rid="B29">29</xref>]. Three overall fit statistics are considered. Two are item-person interaction statistics transformed to approximate a z-score, representing a standardized normal distribution where perfect fit to the model would have a mean of approximately zero and a standard deviation of 1. A third is an item-trait interaction statistic reported as a Chi-Square, reflecting the property of invariance across the trait. A significant Chi-Square indicates that the hierarchical ordering of the items varies across the trait, so compromising the required property of invariance.</p><p>In addition to these overall summary fit statistics, individual person- and item fit statistics are presented, both as residuals (a summation of individual person and item deviations) and as a chi-square statistic. In the former case residuals between ± 2.5 are deemed to indicate adequate fit to the model. In the latter case a chi-square fit statistic is available for each item, and the overall chi-square for items is summed to give the item trait-interaction statistic. An estimate of the internal consistency reliability of the scale is also available, based on the Person Separation Index (PSI) where the estimates on the logit scale for each person are used to calculate reliability.</p><p>Sources of deviation from model expectation are examined to see if the scale construct can be improved. For a good fitting model we would expect that, for each of the items, respondents with high levels of the attribute being measured would endorse high scoring responses, while individuals with low levels of the attribute would consistently endorse low scoring responses. In Rasch analysis terms this would be indicated by an ordered set of response thresholds for each of the items. The term <italic>threshold </italic>refers to the point between two adjacent response categories where either response is equally probable. For a given item the number of thresholds is always one less than the number of response options. Disordered thresholds occur when respondents have difficulty consistently discriminating between response options. This can occur when there are too many response options, or when the labelling of options is confusing. Usually, although not always, collapsing of categories where disordered thresholds occur improves overall fit to the model.</p><p>Another issue that can affect model fit is differential item functioning (DIF). This occurs when different groups within the sample (e.g. males and females), despite equal levels of the underlying characteristic being measured, respond in a different manner to an individual item. For example men and women with equal levels of depression may respond systematically differently to an item in a depression inventory [<xref ref-type="bibr" rid="B30">30</xref>]. Two types of DIF may be identified. Uniform DIF is where the group shows a consistent systematic difference in their responses to an item, across the whole range of the attribute being measured, and non-Uniform DIF is where differences vary across levels of the attribute. Analysis of variance is conducted for each item comparing scores across each level of the 'person factor' (eg. gender) and across different levels of trait (referred to as class intervals). Uniform DIF is indicated by a significant main effect for the person factor (gender), while the presence of non-uniform DIF is indicated by a significant interaction effect (person factor X class interval).</p><p>Finally, when issues of threshold disordering, DIF and fit have been resolved a Principal Components Analysis (PCA) of the residuals detects any signs of multidimensionality. After the 'Rasch' factor has been extracted there should be no associations left in the data. There are several ways to detect this, including the proportion of variance attributable to the first residual factor compared with that attributable to the first (Rasch) factor, and whether or not estimates derived from subsets of items are invariant (specific objectivity). This latter is formally tested by allowing the factor loadings on the first residual to determine 'subsets' of items and then testing, by a paired <italic>t</italic>-test, to see if the person estimate (the logit of person 'ability' or, in this case 'depression') derived from these subsets significantly differs between subsets [<xref ref-type="bibr" rid="B31">31</xref>]. If the person estimate is found to differ between the subsets of items this would indicate the presence of multidimensionality. An effect size for the difference can also be calculated to determine the substantive nature of such a difference.</p><p>Where the data fit the model, and the assumptions of local independence are met, a unidimensional linear scale is derived from the ordinal raw score, thus opening up the opportunity to validly apply parametric approaches [<xref ref-type="bibr" rid="B32">32</xref>,<xref ref-type="bibr" rid="B33">33</xref>]. Thus, fitting data to the Rasch model offers a useful approach to addressing key methodological aspects of scale development, including dimensionality, category ordering and item bias.</p></sec></sec><sec><title>Results</title><p>Rasch analysis of the data from the ten item scale using RUMM2020 showed a lack of fit to the Rasch model with a significant Item-Trait Interaction total chi-square (chi-square = 82.8, df = 40; p < .001). The mean residual for items was -0.50 with a standard deviation (SD) of 1.575, whereas the latter would be expected to be much closer to 1, given adequate fit to the model. The mean residual for persons was -0.287 with a SD of 0.855, indicating no serious misfit among the respondents in the sample.</p><p>Initially, the pattern of thresholds was examined to see if disordering may be affecting fit. In the current example all thresholds were ordered (Figure <xref ref-type="fig" rid="F1">1</xref>). The threshold distances vary across items (see varying lengths of category one across items), supporting the use of the partial credit model for the analysis of this scale. A log likelihood ratio test statistic confirmed that this was the case (p < 0.001).</p><fig position="float" id="F1"><label>Figure 1</label><caption><p>Threshold Map for 10-item EPDS.</p></caption><graphic xlink:href="1471-244X-6-28-1"/></fig><p>Two items initially showed misfit to model expectations, Item 8 (<italic>I have felt sad or miserable</italic>) and Item 5 (<italic>I have felt scared or panicky for no very good reason</italic>) (see Table <xref ref-type="table" rid="T1">1</xref>). Item 8 showed a Fit Residual value of -3.275 and a chi-square probability value of 0.002, less than the Bonferroni adjusted alpha value of .005, indicating significant deviation from the model expectation. The negative Fit Residual value obtained suggests a high level of discrimination, shown by the ICC for the item where observed responses are steeper than the expected curve (Figure <xref ref-type="fig" rid="F2">2</xref>). Thus responses from the lowest group (low levels of depression) are below what is expected by the model and those for the highest group (high levels of depression), are above model expectation. This high negative residual is usually associated with dependency, and a high item-total correlation, signifying redundancy of the item.</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Item fit statistics</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left">Item</td><td align="center">Location</td><td align="center">SE</td><td align="center">FitResid</td><td align="center">DF</td><td align="center">ChiSq</td><td align="center">DF</td><td align="center">Prob</td></tr></thead><tbody><tr><td align="left">1. I have been able to laugh and see the funny side of things</td><td align="center">1.067</td><td align="center">0.125</td><td align="center">-0.993</td><td align="center">281.46</td><td align="center">3.26</td><td align="center">4</td><td align="center">0.515</td></tr><tr><td align="left">2. I have looked forward with enjoyment to things</td><td align="center">2.429</td><td align="center">0.125</td><td align="center">-0.315</td><td align="center">280.57</td><td align="center">2.62</td><td align="center">4</td><td align="center">0.623</td></tr><tr><td align="left">3. I have blamed myself unnecessarily when things went wrong</td><td align="center">-1.958</td><td align="center">0.094</td><td align="center">1.958</td><td align="center">280.57</td><td align="center">13.52</td><td align="center">4</td><td align="center">0.009</td></tr><tr><td align="left">4. I have been anxious or worried for no good reason</td><td align="center">-1.456</td><td align="center">0.089</td><td align="center">0.843</td><td align="center">281.46</td><td align="center">4.62</td><td align="center">4</td><td align="center">0.329</td></tr><tr><td align="left"><bold>5. I have felt scared or panicky for no very good reason</bold></td><td align="center"><bold>-0.541</bold></td><td align="center"><bold>0.094</bold></td><td align="center"><bold>0.794</bold></td><td align="center"><bold>281.46</bold></td><td align="center"><bold>15.84</bold></td><td align="center"><bold>4</bold></td><td align="center"><bold>0.003</bold></td></tr><tr><td align="left">6. Things have been getting on top of me</td><td align="center">-1.069</td><td align="center">0.109</td><td align="center">-1.162</td><td align="center">279.67</td><td align="center">8.10</td><td align="center">4</td><td align="center">0.088</td></tr><tr><td align="left">7. I have been so unhappy that I have had difficulty sleeping</td><td align="center">0.118</td><td align="center">0.104</td><td align="center">-0.935</td><td align="center">281.46</td><td align="center">5.34</td><td align="center">4</td><td align="center">0.255</td></tr><tr><td align="left"><bold>8. I have felt sad or miserable</bold></td><td align="center"><bold>-0.407</bold></td><td align="center"><bold>0.103</bold></td><td align="center"><bold>-3.275</bold></td><td align="center"><bold>281.46</bold></td><td align="center"><bold>17.55</bold></td><td align="center"><bold>4</bold></td><td align="center"><bold>0.002</bold></td></tr><tr><td align="left">9. I have been so unhappy that I have been crying</td><td align="center">0.133</td><td align="center">0.112</td><td align="center">-2.364</td><td align="center">281.46</td><td align="center">7.86</td><td align="center">4</td><td align="center">0.097</td></tr><tr><td align="left">10. The thought of harming myself has occurred to me</td><td align="center">1.684</td><td align="center">0.174</td><td align="center">0.349</td><td align="center">281.46</td><td align="center">4.15</td><td align="center">4</td><td align="center">0.386</td></tr></tbody></table><table-wrap-foot><p>Misfitting items are shown in bold.</p></table-wrap-foot></table-wrap><fig position="float" id="F2"><label>Figure 2</label><caption><p>Fit of item 8 <italic>I have felt sad or miserable.</italic></p></caption><graphic xlink:href="1471-244X-6-28-2"/></fig><p>Removal of Item 8 led to an improvement in fit to the model with a non-significant (Bonferroni adjusted) Item-Trait Interaction total chi-square (chi-square = 60.2, df = 36, p = 0.007). The Residual mean value for items became -0.47 with a standard deviation (SD) of 0.909, showing much better fit to the model. Individual person fit statistics showed that no respondents had residuals outside the acceptable range for the 9-item solution. Following the removal of item 8, individual item fit statistics were again reviewed, and item 5, which initially showed misfit to the model, now showed a response pattern consistent with model expectation, and was therefore retained.</p><p>In the 9-item solution the possibility of item bias was explored for the age of the mother, educational level of the mother, and the age of the child, using a Bonferroni adjusted p value of 0.003 (0.05/18). Just one of the items Item 7 (<italic>I have been so unhappy that I have had difficulty sleeping</italic>) recorded a probability value exceeding the adjusted alpha value, showing some degree of uniform DIF for age of child (see Figure <xref ref-type="fig" rid="F3">3</xref>). Inspection of the DIF graph suggests that, at equal levels of depression, mothers with very young babies (6 to 12 weeks) are less likely to endorse this item. As DIF is a breach of unidimensionality, this item was also deleted. This gave a non-significant (Bonferroni adjusted) Item-Trait Interaction total chi-square (chi-square = 53.8, df = 32, p = 0.009). The Residual mean value for items was -0.467 with a standard deviation (SD) of 0.850, showing fit to the model. No DIF now existed in this 8-item scale (EPDS-8) and all items showed fit to model expectations.</p><fig position="float" id="F3"><label>Figure 3</label><caption><p>Differential item functioning for age of baby for Item 7.</p></caption><graphic xlink:href="1471-244X-6-28-3"/></fig><p>Figure <xref ref-type="fig" rid="F4">4</xref> shows the distributions of persons and item thresholds of the revised scale, with persons on the upper part of the graph, and the item thresholds on the lower part. The average mean person location value of -2.465 suggests that the respondents were well below the average of the scale. However, for a screening instrument this is not necessarily of great concern, as the cut point for a clinical case is the key issue. The PSI Statistic was 0.804, which indicates that the scale has adequate person separation reliability.</p><fig position="float" id="F4"><label>Figure 4</label><caption><p>Targeting map for 8-item EPDS.</p></caption><graphic xlink:href="1471-244X-6-28-4"/></fig><p>A principal component analysis of the residuals revealed a first residual factor accounting for 1.8% of the total variance in the data, or 22% of the variance in the residuals. Two sets of items were found to load positively and negatively on the first residual component. A paired t-test indicated that neither of these two sets gave a person estimate significantly different to the other (p = 0.14) and the effect size of the difference was 0.08. Consequently the assumption of local independence is upheld, and the EPDS-8 can be considered to be a unidimensional scale.</p><p>To determine cut points on this revised 8-item scale individuals were first classified according to the original 10-item EPDS cut points [<xref ref-type="bibr" rid="B1">1</xref>]. This allowed each person to be identified as not depressed (range 0–9); minor depression (range 10–12) or more major depression (range 13 or more). For minor depression a cut point of 8 or more on the EPDS-8 maximised the kappa (0.9), identifying 95% of those classified as such by the original 10-item scale. This cut point also identified 96.7% of those identified as not depressed by the original scale. For major depression a cut point of 9 or more on the EPDS-8 identified all those so classified by the original and 91.9% of those without major depression, but the kappa was lower (0.71) than a cut point of 10+ (0.86) which identified 97.2% of those classified as having major depression on the original, and 96.8% of those without major depression.</p><p>Figure <xref ref-type="fig" rid="F5">5</xref> shows the distribution of scores on the EPDS-8 for each group classified using the original EPDS. The cut point of 8 or more for minor depression, and 10 or more for major depression (shown as the horizontal lines on the graph) clearly separates cases with no evidence of depression, as defined by the original scale, from those with minor and major depression (Kruskal-Wallis: chi-square = 179.1; df = 2; p < .001).</p><fig position="float" id="F5"><label>Figure 5</label><caption><p>Boxplot showing 8-item EPDS scores for women classified into groups using the original EPDS cutpoints (scores 0 to 9, 10 to 12, 13+).</p></caption><graphic xlink:href="1471-244X-6-28-5"/></fig><p>The results of the above analysis suggest that an eight item version of the scale would be more psychometrically robust, in that it would be free of item bias caused by the influence of baby age on Item 7, and also removes Item 8 which showed misfit to the model. It also has high levels of agreement with the original case identification. The scale has an approximate linear range only for the raw score range of 4 to 20 (from a range of 0–24 on the EPDS-8).</p></sec><sec><title>Discussion</title><p>Despite its widespread use, the viability of the original 10-item EPDS has been found to fall short of the rigorous standards of measurement defined by the Rasch model. The use of Rasch analysis in this study has enabled a detailed examination of the structure and operation of the scale. The ordering of response categories (threshold ordering) has not been examined previously, and the evidence from this current study supports the response format used, but not the validity of the full original 10-item scale. It was necessary to remove two items from the scale, in order to achieve model fit. Item 7 (<italic>I have been so unhappy that I have had difficulty sleeping</italic>) was removed because it showed differential item functioning for the age of the baby. Although there are techniques to accommodate uniform DIF by allowing the item difficulty to vary by group [<xref ref-type="bibr" rid="B34">34</xref>], we thought this inappropriate as this option is not practical in an everyday screening environment. Consequently we chose to delete the biased item (Item 7). Inspection of the item wording revealed that this item is potentially confusing as it mixes the concepts of unhappiness with sleep, which may be confounded by the mother's expectations, and/or the child's lack of sleep. This could be one reason why the item works differently according to the age of the child. Removal of this item from the scale improved the overall model fit, supporting this decision.</p><p>Item 8 (<italic>I have felt sad or miserable</italic>) was also removed from the scale due to misfit to the model. At first sight this may seem a strange omission, in that the item appears to have face validity. The high negative residual misfit indicates that it adds nothing to the information gained by other items, and its removal significantly improved the fit of those remaining items. In some respects it is more like a summary item with a high item-total correlation (0.8). Further research is needed to assess the replicability of this finding in other samples. The results of this study support the retention of Item 10 <italic>The thought of harming myself has occurred to me</italic>, despite previous factor analytic studies which led to its exclusion. Its low endorsement rate and item-total correlation may have contributed to this, as it is known that factor analysis can misidentify factors by frequency (92% scored a value of zero on this item).</p><p>The results of this study do not support the structure of the original 10-item scale as proposed by the scale developers [<xref ref-type="bibr" rid="B1">1</xref>]. However there is also no evidence supporting the alternative structure identified by Brouwers et al. [<xref ref-type="bibr" rid="B7">7</xref>], Ross et al. [<xref ref-type="bibr" rid="B8">8</xref>] and Pop et al. [<xref ref-type="bibr" rid="B9">9</xref>] separating the anxiety (items 3,4,5) and depression items (items 1,2,8). Although two sets of items were identified in PCA of the residuals, the person estimates (Rasch logit-based estimate) derived from the subsets were not significantly different from one another, thereby supporting its unidimensionality.</p><p>Finally, the fit of the EPDS to the Rasch model has shown that the scale in its raw form is ordinal. This is not necessarily a problem when the scale is used with the cut points to identify those with depression, as an ordinal scale does just as well under these circumstances. However, depending on distribution of patients, there would be a problem if change scores needed to be calculated [<xref ref-type="bibr" rid="B32">32</xref>] and this would consequently need a Rasch transformed score.</p><p>The focus of this study has been the use of Rasch analysis to assess the measurement properties of the EPDS in terms of its structure, item fit and freedom from bias. This does not however provide a test of the clinical validity of the scale. Further studies are required to formally assess the revised format of the scale (EPDS-8) in clinical settings and to explore the appropriateness of the recommended cut-points using alternative assessment tools, such as standardized diagnostic interviews. The screening capacity of the shortened version of the EPDS identified in this study will need to be clinically assessed against the original 10-item EPDS.</p></sec><sec><title>Conclusion</title><p>In summary, it would appear that the total EPDS, in its original 10-item form, is not a valid scale for the measurement of depression. The results of this study suggest that a revised eight item version, the EPDS-8, would provide a more psychometrically robust scale. The revised cut points of 7/8 and 9/10 for the EPDS-8 show high levels of agreement with the original case identification for the EPDS-10.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>JFP supervised the design of the study and the statistical analyses undertaken. RLM collected the data and helped to design the study. AT assisted with the analysis of the data and interpretation of the results. All authors contributed to the preparation of the article. All authors read and approved the final manuscript.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1471-244X/6/28/prepub"/></p></sec>
Transcriptional profiling of mesenchymal stromal cells from young and old rats in response to Dexamethasone	<sec><title>Background</title><p>Marrow-derived stromal cells (MSCs) maintain the capability of self-renewal and differentiation into multiple lineages in adult life. Age-related changes are recognized by a decline in the stemness potential that result in reduced regeneration potential of the skeleton. To explore the molecular events that underline skeletal physiology during aging we catalogued the profile of gene expression in <italic>ex vivo </italic>cultured MSCs derived from 3 and 15 month old rats. The <italic>ex vivo </italic>cultured cells were analyzed following challenge with or without Dexamethasone (Dex). RNA retrieved from these cells was analyzed using Affymetrix Gene Chips to compare the effect of Dex on gene expression in both age groups.</p></sec><sec><title>Results</title><p>The molecular mechanisms that underline skeletal senescence were studied by gene expression analysis of RNA harvested from MSCs. The analysis resulted in complex profiles of gene expression of various differentiation pathways. We revealed changes of lineage-specific gene expression; in general the pattern of expression included repression of proliferation and induction of differentiation. The functional analysis of genes clustered were related to major pathways; an increase in bone remodeling, osteogenesis and muscle formation, coupled with a decrease in adipogenesis. We demonstrated a Dex-related decrease in immune response and in genes that regulate bone resorption and an increase in osteoblastic differentiation. Myogenic-related genes and genes that regulate cell cycle were induced by Dex. While Dex repressed genes related to adipogenesis and catabolism, this decrease was complementary to an increase in expression of genes related to osteogenesis.</p></sec><sec><title>Conclusion</title><p>This study summarizes the genes expressed in the <italic>ex vivo </italic>cultured mesenchymal cells and their response to Dex. Functional clustering highlights the complexity of gene expression in MSCs and will advance the understanding of major pathways that trigger the natural changes underlining physiological aging. The high throughput analysis shed light on the anabolic effect of Dex and the relationship between osteogenesis, myogenesis and adipogenesis in the bone marrow cells.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Akavia</surname><given-names>Uri David</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Shur</surname><given-names>Irena</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Rechavi</surname><given-names>Gideon</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" corresp="yes" contrib-type="author"><name><surname>Benayahu</surname><given-names>Dafna</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Genomics	<sec><title>Background</title><p>The stromal compartment of the bone marrow contains mesenchymal stem and progenitor cells with high proliferating capacity in addition to cells at different stages of maturation. The mesenchymal cells are at the front of cell research today which differentiate <italic>in vitro </italic>and <italic>in vivo </italic>to multiple lineages including fibroblasts, adipocytes, cartilage, myogenic, and osteogenic cells [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B7">7</xref>]. MSCs also harbor the potential of trans-differentiation to many different lineages, thus providing a possible source of progenitors for cell therapy and tissue repair including bone, cartilage, cardiac, pancreas regeneration and neural injury repair [<xref ref-type="bibr" rid="B8">8</xref>]. Cell differentiation through distinct maturational stages involves coordination and activation of different sets of genes. Progenitor cells derived from the stroma compartment of the bone marrow differentiate under the control of transcription factors, which serve as lineage specific master genes for discrete differentiation steps. Definition of the key differentiation signals is important in order to induce the desired <italic>ex vivo </italic>lineage-specific maturation pathways.</p><p>Age related hormonal changes, for example a decline in sex hormones levels, are associated with a decrease in the number and activity of osteogenic cells and an increase in numbers of adipocytes [<xref ref-type="bibr" rid="B9">9</xref>-<xref ref-type="bibr" rid="B15">15</xref>]. It is generally accepted that these changes arise from a decrease in the stemness potential accompanied by a decrease in the proliferative ability and osteogenic capacity of the bone marrow cells [<xref ref-type="bibr" rid="B11">11</xref>,<xref ref-type="bibr" rid="B15">15</xref>-<xref ref-type="bibr" rid="B18">18</xref>]. The changes in stemness with age result in reduced osteogenesis and increased adipogenesis, affecting the skeletal structure and the immune system. Age-related changes associated with osteoporosis were previously studied by us in animal models [<xref ref-type="bibr" rid="B15">15</xref>,<xref ref-type="bibr" rid="B19">19</xref>] and in <italic>ex vivo </italic>cultures of stromal cells [<xref ref-type="bibr" rid="B20">20</xref>]. It is clear that the physiological status of the body affects the skeleton at the cellular level, but the underlying molecular mechanism remains unresolved.</p><p>The action of native or pharmacological glucocorticoid hormones, such as Dexamethasone (Dex), is mediated via glucocorticoid receptors (GRs). Dex is recognized by multiple effects on a wide range of tissues and physiological conditions in the body [<xref ref-type="bibr" rid="B21">21</xref>]. Dexamethasone promotes osteogenesis <italic>in vitro </italic>[<xref ref-type="bibr" rid="B22">22</xref>], and induces the expression of osteogenic markers in MSCs [<xref ref-type="bibr" rid="B20">20</xref>,<xref ref-type="bibr" rid="B23">23</xref>-<xref ref-type="bibr" rid="B25">25</xref>].</p><p>In this study, we analyze the molecular changes in aged rats that influence the cellular potential and the response to Dex. We used GeneChip technology to explore the molecular changes regulating the processes that govern the commitment and differentiation of the MSCs in young and aged animals. This approach enables us to analyze genome-wide patterns of mRNA expression and to provide an efficient access to genetic information. We compared gene profiling between MSCs cultured <italic>ex vivo </italic>from young and old rats (3 and 15 month old). The cells were maintained <italic>in vitro </italic>and maintained in presence or absence of Dex. The RNA extracted was analyzed to assess the transcriptome profile of MSCs. From the microarrays we further analyzed the lineage-specific gene expression in the MSCs enabling us to reveal genome-wide patterns of mRNA expression and to sort the gene profiles that govern various cell activities.</p></sec><sec><title>Results</title><sec><title>The GeneChip analysis</title><p>Primary marrow stromal cells (MSCs) derived from the bone marrow include stem and progenitor cells with high proliferating capacity. We have earlier studied MSC in rat [<xref ref-type="bibr" rid="B19">19</xref>] and mouse [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B15">15</xref>] models and demonstrated that the decline in stemness with aging is associated with bone atrophy, increase of adipocytes and augmentation of T-lymphopoiesis.</p><p>The present study aimed to compare the profile of genes expressed by cultured MSCs derived from 3 and 15 month old rats, expanded <italic>in vitro </italic>and treated or untreated with Dex. We used microarray technology to compare the transcriptome profile between cultured MSCs from 4 experimental groups: young Y (3 month) and old O (15 month) rats that were treated (YT, OT) or untreated (YU, OU) with Dex. The molecular analysis represents snapshot of cellular states from these rats and reflects their potential for tissue differentiation. The analysis resulted with 10290 Probe Sets (PS) that changed with fold change of two or higher. The PS clustering attempts to elucidate the pattern of gene expression which is unique to MSCs.</p></sec><sec><title>Genes differentially expressed between experimental groups</title><p>The gene expression was studied for RNA from each age group, and the overlap between them was calculated. RNA was retrieved from four groups of cells according to treatment and animal age: untreated cells from 3 month old rats (young untreated – YU – 1), 15 month old rats (old untreated – OU – 2) and treated cells from young and old rats (YT – 3, OT – 4). Differentially expressed PS were marked either as increasing or decreasing according to expression ratios between Dex-treated and untreated cells. In cells derived from young rats, the analysis resulted in 3318 differentially expressed PS, 1604 increasing (YI) and 1714 decreasing (YD). In cells derived from old rats, 2725 PS were differentially expressed – 1314 increasing (OI), 1411 decreasing (OD) following treatment.</p><p>A total of 888 PS increased and 1077 PS decreased in cells derived from both young and old rats due to the Dex treatment. The analysis of differentially expressed PS was repeated for fold change of 3, 4 and 5 and resulted in smaller numbers of differentially expressed PS. The number of PS differentially expressed at fold change higher then 2 decreased, but the ratio of expressed PS in both ages versus PS expressed in cells derived from young or old rats did not change much, indicating a common response to Dex in both ages (Figure <xref ref-type="fig" rid="F1">1</xref>).</p><p>The ratio of PS changed in both groups relative to PS changed in cells from old rats remained constant. The ratio of PS increased in both groups (BI) relative to PS increased in old (OI) and the ratio of decreasing PS (BD/OD) remained at about 60% and 70%, respectively. The PS increased in both groups (BI) relative to PS increased in young (YI) lowered from 55% to 32% as the number of differentially expressed PS decreased (due to increasing fold change). The ratio of PS decreased in both groups (BD) relative to PS that decreased in young (YD) lowered from 63% to 48% as the number of PS in general decreased.</p></sec><sec><title>Clustering analysis</title><p>PS were clustered to profile distinct Dex effect for both age groups; 4060 PS were differentially expressed, up or down regulated more than 2-fold, due to the Dex treatment in cells derived from young and old rats. The analysis resulted in six clusters that were divided into PS that increased (A-C) or decreased (D-F) (Figure <xref ref-type="fig" rid="F2">2</xref>). Cluster A (704 PS) represents PS that increased in cells derived from young rats, Cluster B, represents 420 PS increased in old rats, and PS that increased in cells from both age groups are included in Cluster C (888 PS). PS that decreased in cells derived from young rats are put in Cluster D (631 PS), PS that decreased in cells from old rats – in Cluster E (322 PS), and Cluster F (1071 PS) presents PS that decreased in cells from both age groups. A small number of PS increased in cells from young rats and decreased in cells from old rats (12 PS) or the opposite pattern of expression (6 PS) (data not shown). PS who had a known Gene ID were analyzed for functional properties of the genes.</p></sec><sec><title>Functional pathways involved in mesenchymal cells differentiation</title><p>Comparing gene expression profiles between Dex-treated or untreated cultured cells obtained from young and old rats results with clustering of 6 groups (Figure <xref ref-type="fig" rid="F2">2</xref>) analyzed for statistically significant functions, using the GOTM. This analysis shows functions affecting response to Dex and differentiation of skeletal and bone marrow cells. Tables <xref ref-type="table" rid="T1">1</xref> and <xref ref-type="table" rid="T2">2</xref> describe the major functions that have a high fold change and highlight various differentiation pathways of mesenchymal cells affected by Dex treatment.</p><p>Series of genes increased following Dex treatment of MSCs are represented in clusters A-C (Figure <xref ref-type="fig" rid="F2">2</xref>): genes increased in cells derived from young rats (A), in cells derived from old rats (B) and genes increased in cells derived from both age groups (C) (Table <xref ref-type="table" rid="T1">1</xref>). The overall gene expression reflects the Dex-related shift of cellular metabolism, which results in a decrease in proliferation, coupled with differentiation of MSCs to osteogenic and myogenic directions and a repression of the adipogenic pathway.</p><p>Cluster A (genes that increased only in cells derived from young rats) includes <italic>VEGF</italic>, <italic>Id1</italic>, <italic>Madh5 </italic>and <italic>beta-catenin </italic>which are related to angiogenesis (p = 0.00125), implying that neovascularization is linked to osteoblast maturation and bone remodeling (p = 0.00448), which included the genes <italic>PTHr1</italic>, <italic>IBSP</italic>, <italic>Madh5 </italic>and bone ECM protein <italic>Mepe</italic>. These genes are related to osteogenesis and indicate an osteogenic differentiation of these cells. Human and mouse orthologs disclosed components of ECM structural constituent such as <italic>collagens IV, V, VI </italic>and <italic>laminin </italic>(p = 0.00287). Dex has an anabolic effect on cells from young rats that increases osteogenesis-related markers, while such an effect was undetected in cells derived from old rats and is possibly related to the decrease in bone formation associated with aging.</p><p>Cluster C presents genes that increased due to Dex treatment in cells derived from both young and old rats (Table <xref ref-type="table" rid="T1">1</xref>) which are related to cell growth (p = 2.36E-06), regulation of cell cycle (p = 0.00418), muscle differentiation and contraction (p = 0.00555, p = 0.00157), cytoskeletal components (p = 0.00868) including structural constituents of the cytoskeleton and actins (p = 0.00533, p = 0.00803). A reciprocal relationship of proliferation and differentiation is recognized by a decrease in genes that activate proliferation as cells differentiate. Human and mouse orthologs of the rat genes are related to the dystrophin complex (p = 0.00311) which plays a role in development of skeletal muscle. An increase in muscle related genes due to Dex indicates the potential of mesenchymal stem cells to differentiate into myotubes.</p><p>Genes decreased following Dex treatment are clustered in D-F (Figure <xref ref-type="fig" rid="F2">2</xref>) for cells derived from young rats (D), old rats (E) and from both age groups (F) (Table <xref ref-type="table" rid="T2">2</xref>). Genes repressed following Dex treatment in cells derived from young rats are related to adipocyte differentiation, regulation and function (Table <xref ref-type="table" rid="T2">2</xref>, Cluster D). <italic>PPARγ </italic>and <italic>CEBPα </italic>are TFs essential for adipocyte differentiation (p = 0.0056). <italic>PPARγ </italic>maintains adipogenesis affecting early differentiation and survival of mature adipocytes and <italic>CEBPα </italic>is activated later than <italic>PPARγ</italic>. An inhibition of <italic>PPARγ </italic>and <italic>CEBPα </italic>in response to Dex results with a decrease of adipogenesis. The effect of Dex in lowering the adipogenesis (Table <xref ref-type="table" rid="T2">2</xref>, cluster D) is complementary to increasing the osteogenesis (Table <xref ref-type="table" rid="T1">1</xref>, cluster A).</p><p>Cluster F (Table <xref ref-type="table" rid="T2">2</xref>) summarizes genes repressed by Dex treatment in cells derived from both age groups and includes genes related to defense response (p = 0.0006). We noticed a decrease in RAS signaling genes (p = 0.00016), such as <italic>Nras</italic>, <italic>Kras2</italic>, <italic>Grb2 </italic>and <italic>Aps</italic>, which is complimentary to the proliferation and differentiation genes induced by Dex (Table <xref ref-type="table" rid="T1">1</xref>, Cluster C). Dex repressed genes related to catabolism (p = 0.00027) and proteolysis (p = 0.00312), including mainly lysosomal (p = 1.4E-07) hydrolases (p = 0.00019). The suppression of adipogenic differentiation is monitored by the decrease of genes related to lipid binding (p = 0.00024), lipid transport (p = 0.0096) and diacylglycerol binding (p = 0.00474) including genes such as <italic>FABP-4 </italic>(<italic>aP-2</italic>) and <italic>lipoprotein lipase </italic>(<italic>Lpl</italic>). Protein Kinase C isoforms comprised some of the genes identified as lipid binding cluster. The variants of this gene have a role in muscle differentiation and mediate PTH receptor action in repressing adipogenesis and inducing osteogenesis.</p><p>The functional analysis of the clusters resulted in major pathways – an increase in osteogenesis and muscle formation, coupled with a decrease in adipogenesis. We selected the genes representing these differentiation pathways (Table <xref ref-type="table" rid="T3">3</xref>) and divided them into clusters and functional groups. Genes which increased only in cells from young rats belong to the functional group of bone remodeling and angiogenesis. Genes that increased in both cells derived from young and old rats are related to muscle development and cell cycle. Other genes that decreased only in cells derived from young rats are related to adipocyte differentiation and genes that decreased in both groups are related to lipid binding, catabolic peptidases, immune response and RAS signaling. To validate the results from the gene array we employed RT-PCR analysis summarized in Figure <xref ref-type="fig" rid="F3">3</xref> (p values are specified in the legend). Analyses were performed for the expression of <italic>BSP</italic>, <italic>Biglycan </italic>and <italic>BMP4 </italic>to follow osteogenic pathway; <italic>desmin </italic>and <italic>actin γ2 </italic>– to follow the effect of age and Dex on the cell potential to differentiate to myoblasts. <italic>AP-2</italic>, <italic>LPL </italic>and <italic>Adipsin </italic>were used to follow adipogenesis. Analyzed markers showed age and Dex related changes detected with a similar fold change levels for both RT-PCR and array analysis. The expression of <italic>GR </italic>was analyzed to verify that the response to Dex resulted with down regulation of the receptor. The expression level of <italic>GR </italic>decreased in cells from both age groups – the ratios were 0.4 and 0.54 on the chip, and 0.44 and 0.42 by RT-PCR. The fold change for desmin on the chip array was 2.4 for cells derived from young rats and 2.36 for cells derived from old rats and 1.9 and 2.05 by RT-PCR, respectively. <italic>AP-2 </italic>expression decreased on the chip (0.04 and 0.01), and by RT-PCR (0.2 and 0.12). Dex triggered an increase in the osteogenic related genes that was more prominent in young animals. An increase was noted in muscle-related genes in cells from both age groups; and a decrease in genes related to the lipid metabolism. The expression level of these genes behaved in a similar manner on the chip and when checked by RT-PCR. The comparison between the two methods confirmed the biological effect of Dex on cells from young and old rats.</p></sec></sec><sec><title>Discussion</title><p>Mesenchymal stem cells have an ability to differentiate into multiple lineages including adipocytes, myocytes and bone forming cells [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B7">7</xref>]. Aging affects the bone marrow microenvironment in multiple ways, repressing the "stemness" of cells which results in a decrease in bone formation. The number of precursor cells of the hematopoietic and osteogenic cells as well as their proliferative ability decrease with aging [<xref ref-type="bibr" rid="B10">10</xref>,<xref ref-type="bibr" rid="B11">11</xref>]. Although there are some contradictory findings in the literature related to proliferative and osteogenic potential, some of them rely on studies performed with total bone marrow population [<xref ref-type="bibr" rid="B17">17</xref>,<xref ref-type="bibr" rid="B26">26</xref>]. Other experiments [<xref ref-type="bibr" rid="B9">9</xref>-<xref ref-type="bibr" rid="B15">15</xref>], including ones in our laboratory [<xref ref-type="bibr" rid="B15">15</xref>,<xref ref-type="bibr" rid="B19">19</xref>], have focused on the mesenchymal stem cells and have shown a decrease in the stemness potential followed by a decline in the proliferative and osteogenic capacities. Consequently the molecular mechanism of the decrease in stemness associated with a decrease in bone formation with aging is yet unclear.</p><p>To better understand the molecular mechanisms governing the cellular changes we profiled the response of <italic>ex vivo </italic>cultured stromal cells challenged by Dex and compared cells derived from young and old rats. The gene expression profile revealed overlap of Dex action between the MSC derived from the two age groups. The functional analysis of the differentially expressed genes highlighted a general shift in the metabolism of the MSCs which included repression of proliferation and was complemented by induction of differentiation. Specifically, we monitored genes related to induction of myogenic and osteogenic differentiation coupled with a decrease in adipogenesis.</p><p>The results presented show a decrease in proliferation coupled with an increase in muscle related genes, observed in both age groups (Cluster C). The expression of muscle related genes indicates that the potential of mesenchymal stem cells to differentiate to myotubes is enhanced by Dex, which was reported earlier [<xref ref-type="bibr" rid="B27">27</xref>]. A reciprocal relationship of proliferation and differentiation is known – acquisition of differentiated phenotype is accompanied by a decrease in cells' proliferation potential, a relationship which is well documented for osteoblasts [<xref ref-type="bibr" rid="B28">28</xref>] and for myoblasts [<xref ref-type="bibr" rid="B29">29</xref>]. Dysfunction of tumor suppressor proteins results in attenuation of osteogenesis and myogenesis, <italic>in vivo </italic>[<xref ref-type="bibr" rid="B30">30</xref>,<xref ref-type="bibr" rid="B31">31</xref>] and <italic>in vitro </italic>[<xref ref-type="bibr" rid="B32">32</xref>]. It was shown that myoblasts lacking <italic>Rb </italic>failed to differentiate into myotubes even when transfected with <italic>MyoD</italic>, an essential transcription factor for myogenesis [<xref ref-type="bibr" rid="B33">33</xref>]. In the MSCs analyzed in the present study, Dex decreased proliferation and triggered myogenesis. In addition, the genes induced are known to play a role in osteogenesis, which was also induced. Few examples are shown by the expression of <italic>PDGFa </italic>[<xref ref-type="bibr" rid="B34">34</xref>], <italic>TGFb3 </italic>and <italic>p53 </italic>[<xref ref-type="bibr" rid="B32">32</xref>], genes known to play a role in osteogenesis [<xref ref-type="bibr" rid="B35">35</xref>] in addition to myogenic differentiation.</p><p>Aging is accompanied by a reduced stemness potential of MSCs resulting in a decrease of osteogenesis and replacement of bone marrow with fat cells. This phenomena is associated with osteoporosis [<xref ref-type="bibr" rid="B36">36</xref>] and overall reduction of osteogenic stem cells with aging [<xref ref-type="bibr" rid="B12">12</xref>]. In agreement with this, in our study the cells derived from young rats, but not the ones derived from old rats, responded to Dex with an increase in genes involved in bone remodeling, angiogenesis and a decrease in genes related to adipocyte differentiation.</p><p>Genes increased in cells derived from young rats (Cluster A) were related to bone remodeling and included <italic>PTHr1</italic>, <italic>BSP</italic>, <italic>Madh5</italic>, <italic>Mepe</italic>, <italic>byglican </italic>and <italic>BMP4 </italic>which are expressed in osteoblasts during differentiation and bone regeneration [<xref ref-type="bibr" rid="B35">35</xref>,<xref ref-type="bibr" rid="B37">37</xref>,<xref ref-type="bibr" rid="B38">38</xref>], indicating an anabolic effect of Dex on the cells derived from young rats. The anabolic effect of Dex is known, as demonstrated by increases in the expression of bone related genes including <italic>BSP </italic>in rat osteoblastic cells [<xref ref-type="bibr" rid="B39">39</xref>,<xref ref-type="bibr" rid="B40">40</xref>]. Such an effect was not observed on cells derived from old rats, which might be related to the decrease in bone formation potential recognized with aging. Neovascularization is linked to osteoblast maturation and bone deposition [<xref ref-type="bibr" rid="B41">41</xref>]. The angiogenesis functional group included <italic>VEGF</italic>, <italic>Id1</italic>, <italic>Madh5</italic>, <italic>beta-catenin </italic>and other genes. These genes are recognized in the literature for their role in osteoblastic differentiation in addition to their role in angiogenesis [<xref ref-type="bibr" rid="B42">42</xref>-<xref ref-type="bibr" rid="B45">45</xref>]. The expressed genes' function in angiogenesis implies that MSCs induce or attract formation of blood vessels. In addition, examining the human and mouse orthologs revealed ECM structural constituents including proteins that build up the basement membrane, like <italic>collagens IV, V, VI </italic>and <italic>laminin</italic>. Expression of these genes can indicate a potential for an endothelial differentiation.</p><p>Genes repressed following Dex treatment in cells derived from young rats (Cluster D) are related to adipocyte differentiation, regulation and function. This function included the two transcription factors <italic>PPARγ </italic>and <italic>CEBPα</italic>, which are essential for adipocyte differentiation. <italic>PPARγ </italic>maintains adipogenesis and is necessary for early differentiation and survival of mature adipocytes [<xref ref-type="bibr" rid="B46">46</xref>]. <italic>CEBPα </italic>is activated later than <italic>PPARγ </italic>during adipogenesis [<xref ref-type="bibr" rid="B47">47</xref>] and lack of this TF results in non-functional adipocytes [<xref ref-type="bibr" rid="B48">48</xref>]. These factors are sufficient for adipogenic differentiation in non-adipogenic mesenchymal cells [<xref ref-type="bibr" rid="B49">49</xref>] including myoblasts [<xref ref-type="bibr" rid="B50">50</xref>]. Adipogenesis and osteogenesis represent two divergent pathways of differentiation from the mesenchymal stem cells. Genes up-regulated in osteogenesis, such as <italic>Msx2</italic>, repress adipogenesis by inhibition of <italic>PPARγ </italic>and <italic>CEBPα </italic>[<xref ref-type="bibr" rid="B49">49</xref>]. Alternatively, PPARγ inhibits osteogenesis suppressing the osteoblastic phenotype [<xref ref-type="bibr" rid="B51">51</xref>]. We have demonstrated an interchanging relationship between osteogenesis and adipogenesis affected by Dex. Dex increased <italic>beta-catenin </italic>in the cells derived from young rats that is concurrent with an increase in osteoblastic potential of these cells. <italic>Beta-catenin </italic>represses adipogenesis [<xref ref-type="bibr" rid="B52">52</xref>] and induces osteogenesis [<xref ref-type="bibr" rid="B45">45</xref>]. Thus, the Dex effect of suppressing the adipogenesis (Table <xref ref-type="table" rid="T2">2</xref>, cluster D) and increasing the osteogenesis (Table <xref ref-type="table" rid="T1">1</xref>, cluster A) suggests that <italic>beta-catenin </italic>plays a role in the Dex dependent modulation of the two lineages.</p><p>The genes repressed in both ages (Cluster F, Table <xref ref-type="table" rid="T2">2</xref>) include proliferation related genes, which complement the genes induced in cells from both age groups. We also observed a decrease in adipogenesis related genes, and in catabolic peptidases, indicating an anabolic effect of Dex. In addition, Dex decreased the expression of genes related to defense and immune response (Table <xref ref-type="table" rid="T2">2</xref>). Dex is known to decrease the expression of immune genes, mostly tested on lymphocytes [<xref ref-type="bibr" rid="B21">21</xref>]. Mesenchymal cells have been shown to express immune cell markers, known to produce supportive stroma that regulates the HIM in the differentiation of the immune cells [<xref ref-type="bibr" rid="B53">53</xref>,<xref ref-type="bibr" rid="B54">54</xref>]. Thus, the decrease in immune-related genes may be related to the stromal function. This is strengthened by the fact that repressed genes also included sub-sets of membranous proteins and genes related to antigen processing and presentation (human and mouse orthologs). Hormonal modulation affects cells signaling. We identified repression of RAS pathway, which included the genes <italic>Nras</italic>, <italic>Kras2</italic>, <italic>Grb2 </italic>and <italic>Aps </italic>and also play a role in control of cell proliferation and can also repress differentiation in the myogenic [<xref ref-type="bibr" rid="B55">55</xref>] and osteogenic [<xref ref-type="bibr" rid="B56">56</xref>] pathways. <italic>Grb2 </italic>adaptor protein mediates the proliferative anti-myogenic differentiation action of <italic>Met </italic>(HGF Recptor) [<xref ref-type="bibr" rid="B57">57</xref>], <italic>FGFR </italic>[<xref ref-type="bibr" rid="B58">58</xref>] and is involved in signaling of Dystrophin complex [<xref ref-type="bibr" rid="B59">59</xref>]. The decrease in Ras signaling genes is complimentary to the proliferation and differentiation genes induced by Dex (Table <xref ref-type="table" rid="T1">1</xref>, Cluster C) and indicates a decrease in cell proliferation, which is coupled with myoblastic or osteoblastic differentiation.</p><p>Suppression of adipogenic differentiation was noted by decrease of genes related to lipid binding, lipid transport and diacylglycerol binding including representative genes, such as <italic>FABP-4 </italic>(<italic>aP-2</italic>) [<xref ref-type="bibr" rid="B60">60</xref>,<xref ref-type="bibr" rid="B61">61</xref>] and <italic>lipoprotein lipase </italic>(<italic>Lpl</italic>) [<xref ref-type="bibr" rid="B62">62</xref>]. <italic>Adipsin </italic>repression marked the catabolism function of adipocytes [<xref ref-type="bibr" rid="B63">63</xref>]. Protein Kinase C isoforms are associated with various differentiation pathways. PKC repression increases myogenesis [<xref ref-type="bibr" rid="B64">64</xref>] and as a mediator of PTH receptor actions this enzyme is known to increase osteogenesis and repress adipogenesis in mesenchymal cell lines [<xref ref-type="bibr" rid="B65">65</xref>]. PKC is also associated with inducing osteoblast production of IL-6 [<xref ref-type="bibr" rid="B66">66</xref>] which promotes bone resorption. Thus, the decrease in PKC induces mesenchymal differentiation to the myogenic pathway.</p><p>Catabolic peptidases and lysosomal proteins decreased following Dex treatment. <italic>Matrix metalloproteinases 7,9,12 & 13</italic>, genes which play a role in collagen catabolism and bone remodeling [<xref ref-type="bibr" rid="B67">67</xref>] were included in this group. Additional genes were lysosomal proteases important in lympho-myeloid cells and cathepsins. <italic>Cathepsin K </italic>is the main protease in osteoclastic function that plays a role in bone resorption [<xref ref-type="bibr" rid="B68">68</xref>]. The decrease in expression of these genes indicates a lower rate of bone resorption, which consequently results in an increase in gene related to bone formation mediating the effects of Dex.</p></sec><sec><title>Conclusion</title><p>Aging affects the bone marrow microenvironment in multiple ways, including a decrease in the differentiation potential of cells associated with decrease in bone formation. To better understand molecular mechanisms governing the cellular changes we profiled the response of <italic>ex vivo </italic>cultured stromal cells to Dex and compared cells derived from young and old rats. The gene expression profiling highlights age-independent modes of Dex action which overlapped between the two age groups. While this study demonstrates the effect of Dex on in <italic>ex vivo </italic>cultured cells, it would be interesting to expand the scope to include the <italic>in vivo </italic>effects of Dexamethasone. In general, pattern of expression included repression of proliferation and induction of differentiation. The RNA analyzed was extracted from pooled samples from multiple animals, a method which has been proven to reduce variability, recommended when using a small number of microarrays [<xref ref-type="bibr" rid="B69">69</xref>]. It is possible that additional replicates would further refine these gene lists and may alter some of our conclusions. Continuing research and verification of this interesting subject are required in the future. Specifically, we catalogued genes related to induction of myogenic differentiation coupled with a decrease in adipogenesis. We demonstrated Dex-related decrease in immune response genes; Dex decreased genes that regulate bone resorption and induced osteoblastic differentiation. The high throughput analysis enlightens the effect of Dex and the relationship between osteogenesis and adipogenesis in the bone marrow due to aging. We have also demonstrated the plasticity of cells and the reciprocal relationship of the osteogenic, myogenic and adipogenic lineages in the bone marrow.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title><italic>Ex vivo </italic>cultures of mesenchymal cells</title><p>Mesenchymal stromal cells (MSCs) were cultured from female Wistar rats. Cells were cultured in Dulbecco's modified essential Medium (DMEM) with 10% heat-inactivated fetal calf serum (FCS) (Gibco, USA) [<xref ref-type="bibr" rid="B4">4</xref>]. The bone marrow cells were harvested from young rats (3 months old) and old rats (15 months old). Cells were collected as previously described [<xref ref-type="bibr" rid="B15">15</xref>]. In brief, the bone marrow cells were flushed from the long bone collected and prepared for single cells suspension. Cells pooled from six rats in each group were cultured in 75 cm<sup>2 </sup>flasks (Falcon, USA) containing 20 ml of growth medium, Dulbecco's Modified Essential Medium (DMEM) containing 10% heat-inactivated fetal calf serum (FCS). Under these conditions, the hematopoietic cells died and the cultures finally remained only with cells forming the adherent stromal fibroblast-like layer. Cells were plated and after one week of culturing were trypsinized, single cell suspensions were re-cultured for 7 days and were grown in absence or presence of 10<sup>-8 </sup>M Dexamethasone (Dex). On day 14 cells were harvested for RNA extraction and created four samples – young untreated (1 – YU), old untreated (2 – OU), young treated (3 – YT) and old treated cells (4 – OT).</p></sec><sec><title>RNA analysis and microarrays</title><p>Total RNA were extracted from the design four experimental groups. Each group is composed from multiple animals, a method which has been proven to reduce variability, and is recommended when using a small number of microarrays [<xref ref-type="bibr" rid="B69">69</xref>]. RNA was extracted using the TRIzol and cRNA prepared according to Affymetrix protocols. Gene expression was measured by hybridization to Affymetrix RAE230A Gene Chip DNA microarrays (Affymetrix, Santa Clara, CA, USA), containing 15,766 probe sets (PS) (excluding controls), comprising 14,280 Unigene clusters (11,497 Gene IDs, 4,699 full-length transcripts). The data is available as accession GSE3339 of the Gene Expression Omnibus (GEO).</p></sec><sec><title>Gene expression analysis</title><p>Total RNA extracted from cultured cells following ex vivo expansion was reverse transcribed using avian myeloblastosis virus reverse transcriptase (AMV-RT) and oligo-dT to generate cDNA that served as a template for the polymerase chain reaction (PCR) (Takara Shuzo Co. Ltd., Japan) with gene specific primers (Table <xref ref-type="table" rid="T4">4</xref>). The integrity of the RNA, the efficiency of the RT reaction and the quality of cDNA subjected to the RT-PCR was controlled by amplification of Gluteraldehyde-3-Phosphate Dehydrogenase (G3PDH) (Clontech, Palo Alto, CA). The reaction products were separated by electrophoresis in 1% agarose gels (SeaKem GTG, FMC, USA) in Tris Borate EDTA (TBE) buffer. The amplified DNA fragments were stained by ethidium bromide, and their optical density was measured using Bio Imaging System, BIS 202D and analyzed using "TINA" software. PCR amplification was performed at least twice and subjected to semi-quantitative analyses by comparison of OD of gene-specific PCR products normalized to the OD of co-amplified G3PDH-PCR product in four groups YO, OU, YT and OT.</p></sec><sec><title>The computerized data analysis</title><p>We used the MAS 5.0 algorithm to provide a baseline expression level and detection for each PS. PS were filtered according to the following criteria: (a) at least one sample was detected as Present (P) when calculating MAS 5.0 detection call; and (b) at least one sample had an expression level higher than 20. The expression levels of 10290 retained PS normalized using the Quantile Normalization method, provided by the software package 'affy' (version 1.5.8) [<xref ref-type="bibr" rid="B70">70</xref>], available as part of the Bioconductor project. For every PS, the ratio between expression level of the treated and untreated samples was calculated. PS were analyzed when the expression ratio at least 2 fold, in either age group. Gene expression of cells cultured from young rats (young treated versus young untreated) was compared to cells from old rats (old treated versus old untreated). After filtering the PS were standardized by transforming each expression pattern to have a mean of 0 and a variance of 1.</p></sec><sec><title>Clustering analysis</title><p>We used a rule-based clustering method on the PS that were differentially expressed when the fold change was at least two and were analyzed for either increasing or decreasing expression between Dex-treated and untreated cells. For every PS, the ratio between expression level of the treated and untreated samples was calculated. Every PS was graded as increased, decreased or unchanged. The young grade (YT/YU) was compared to the old grade (OT/OU), and this determined the cluster. This method resulted in 4060 PS, clustered in 8 distinct clusters. This clustering method separated increased and decreased PS due to Dex, and highlighted difference between old and young rats. Partial data is discussed in this paper and complete lists of functions and genes are available upon request.</p></sec><sec><title>Gene Ontology analysis</title><p>The clusters were analyzed for Gene Ontology (GO) annotations appearing in a statistically significant manner in the cluster, when compared to a background. The background was comprised of all expressed genes – a list of all genes expressed in at least one of the four samples. For each GO term the number of genes with this term in a cluster was compared to the number of genes with this term in the background. Additionally the ratio between number of genes in the cluster and number of gene in the background was calculated. The two ratios served to calculate a p-value, and terms with a p-value lower than 0.01 were listed. The analysis was done using GO Tree Machine (GOTM) [<xref ref-type="bibr" rid="B71">71</xref>], the 08/09/04 version. For verification, the same functional analysis was done on Human and Mouse orthologs (provided by Affymetrix 29/07/04), using the EXPANDER software suite [<xref ref-type="bibr" rid="B72">72</xref>].</p></sec></sec><sec><title>Authors' contributions</title><p>UDA carried out the computational analysis, IS carried out the molecular studies. DB conceived of the study, designed and coordinated it. The GeneChip hybridization was performed at the Functional Genomics unit, The Chaim Sheba Medical Center and Sackler School of Medicine headed by GR. All authors helped to draft the manuscript, read and approved the final manuscript.</p></sec>
Direct and heterologous approaches to identify the LET-756/FGF interactome	<sec><title>Background</title><p>Fibroblast growth factors (FGFs) are multifunctional proteins that play important roles in cell communication, proliferation and differentiation. However, many aspects of their activities are not well defined. LET-756, one of the two <italic>C. elegans </italic>FGFs, is expressed throughout development and is essential for worm development. It is both expressed in the nucleus and secreted.</p></sec><sec><title>Results</title><p>To identify nuclear factors associated with LET-756, we used three approaches. First, we screened a two-hybrid cDNA library derived from mixed stages worms and from a normalized library, using LET-756 as bait. This direct approach allowed the identification of several binding partners that play various roles in the nucleus/nucleolus, such as PAL-1, a transcription regulator, or RPS-16, a component of the small ribosomal subunit. The interactions were validated by co-immunoprecipitation and determination of their site of occurrence in mammalian cells. Second, because patterns of protein interactions may be conserved throughout species, we searched for orthologs of known mammalian interactors and measured binary interaction with these predicted candidates. We found KIN-3 and KIN-10, the orthologs of CK2α and CK2β, as new partners of LET-756. Third, following the assumption that recognition motifs mediating protein interaction may be conserved between species, we screened a two-hybrid cDNA human library using LET-756 as bait. Among the few FGF partners detected was 14-3-3β. In support of this interaction we showed that the two 14-3-3β orthologous proteins, FTT-1 and FTT-2/PAR-5, interacted with LET-756.</p></sec><sec><title>Conclusion</title><p>We have conducted the first extensive search for LET-756 interactors using a multi-directional approach and established the first interaction map of LET-756/FGF with other FGF binding proteins from other species. The interactors identified play various roles in developmental process or basic biochemical events such as ribosome biogenesis.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Popovici</surname><given-names>Cornel</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Berda</surname><given-names>Yael</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Conchonaud</surname><given-names>Fabien</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Harbis</surname><given-names>Aurélie</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Birnbaum</surname><given-names>Daniel</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A6" corresp="yes" contrib-type="author"><name><surname>Roubin</surname><given-names>Régine</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Genomics	<sec><title>Background</title><p>FGFs constitute a superfamily of pleiotropic growth factors involved in multiple cellular processes such as mitogenesis, angiogenesis and mesoderm induction [<xref ref-type="bibr" rid="B1">1</xref>]. There are 22 FGFs in humans. Except FGF11-14, they exert their biological activities by acting as extracellular growth factors binding to receptors (FGFR1-4) of the tyrosine kinase receptor superfamily [<xref ref-type="bibr" rid="B2">2</xref>]. In addition, FGF1-3 and FGF11-14 are localized in the nucleus and function intracellularly [<xref ref-type="bibr" rid="B3">3</xref>]. Intracellular FGFs bind to several proteins that play a role in FGF trafficking: FIBP, which allows FGF1 to shuttle between the cytosol and the nucleus [<xref ref-type="bibr" rid="B4">4</xref>], synaptotagmin-1, which allows FGF1 exocytosis [<xref ref-type="bibr" rid="B5">5</xref>], Cystein Rich FGF receptor (CFR), which forms complexes with various FGFs and allows their secretion [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B7">7</xref>], and LRP-1 and 2 (lipoprotein receptor-related proteins), which in conjunction with DAB-1 (Disabled) regulate EGL-17/FGF export in <italic>C. elegans </italic>[<xref ref-type="bibr" rid="B8">8</xref>]. FGF interactors may also regulate FGF nuclear activity; this is the case of Casein Kinase II regulatory subunits [<xref ref-type="bibr" rid="B9">9</xref>] and splicing factor SF3a66 [<xref ref-type="bibr" rid="B10">10</xref>], which both interact with FGF2. Finally, proteins of the extracellular matrix such as fibstatin and fibrinogen interact with FGF2 [<xref ref-type="bibr" rid="B11">11</xref>,<xref ref-type="bibr" rid="B12">12</xref>].</p><p>LET-756 is one of the two FGFs of <italic>C. elegans </italic>[13, 14 for reviews]. It is essential for worm development [<xref ref-type="bibr" rid="B15">15</xref>]. Like some mammalian FGFs, it acts both intra and extracellularly. The molecular motif allowing secretion [<xref ref-type="bibr" rid="B16">16</xref>] and some of LET-756 extracellular functions have been described [<xref ref-type="bibr" rid="B17">17</xref>,<xref ref-type="bibr" rid="B18">18</xref>] but the intracellular functions remain poorly defined, although nuclear localization is probably of importance [<xref ref-type="bibr" rid="B19">19</xref>]. To further characterize the functions of LET-756, we used yeast two-hybrid screens to identify proteins that interact with this FGF. We identified several interacting proteins involved in various developmental processes or in basic biochemical events such as ribosome biogenesis, and validated some of the interactions by co-immunoprecipitation and/or colocalization.</p></sec><sec><title>Results</title><sec><title>Identification of nematode LET-756 binding proteins by yeast two-hybrid library screens</title><p>To identify worm proteins that interact with LET-756, we used the two-hybrid system in the MAV103 yeast with LET-756 fused to the Gal-4 DNA binding domain (pDB) as bait and two <italic>C. elegans </italic>libraries. The latter were either normalized [<xref ref-type="bibr" rid="B20">20</xref>] to contain one representative of each expressed gene of the whole genome (ORFeome) or derived from a mixed stage worm population. Library clones were coupled to the Gal-4 activating domain (pAD). The bait did not show any intrinsic transcriptional activation of the three yeast reporter genes. In a screen of approximately 4 × 10<sup>6 </sup>transformants, 41 clones were positive for the two reporters tested, or for only one reporter but with great intensity. The gap repair technique confirmed 9 clones (Table <xref ref-type="table" rid="T1">1</xref>). Sequencing of these clones and blastn or tblastx interrogation of databases revealed unidentified sequences (UIS) and sequences coding for proteins with known functions: UMP synthase, an enzyme involved in de novo nucleic acid synthesis, cathepsin (aspartic peptidase A1, pepsinogen family member), RPS-16 (small ribosomal subunit), transcription factor PAL-1 involved in the anterior-posterior development of the male [<xref ref-type="bibr" rid="B21">21</xref>], DAF-21, a chaperone of the HSP-90 family, involved in chemosensory transduction and insulin signalization [<xref ref-type="bibr" rid="B22">22</xref>,<xref ref-type="bibr" rid="B23">23</xref>], COL-129, an isoform of collagen, and SKR-2 (homolog of skp1 of <italic>S. cerevisiae</italic>), a component of the skp1p/cullin/F-box SCF complex with ubiquitin ligase activity [<xref ref-type="bibr" rid="B24">24</xref>,<xref ref-type="bibr" rid="B25">25</xref>]. The latter also shows similarities with P19, which is associated with cyclinA/CDK2 complex in humans.</p><p>To confirm the results on some on the interactors we judged potentially relevant, pAD interactor clones were picked individually from the ORFeome library and tested directly in yeast two-hybrid system against pDB LET-756. PAL-1 interacted strongly with LET-756 (two positive tests). For the other interactors, only one test was clearly positive (Table <xref ref-type="table" rid="T2">2A</xref>).</p></sec><sec><title>Identification of LET-756-binding proteins among orthologs of known mammalian FGF interactors</title><p>The conservation of signaling pathways between worms and mammals, as well as the conserved FGF structures in different species [<xref ref-type="bibr" rid="B26">26</xref>,<xref ref-type="bibr" rid="B27">27</xref>], suggested that orthologs of human FGF-binding proteins could interact with LET-756. We used blast interrogation to determine the most conserved orthologs supposed to retain the ancestral function of the known human FGF-interactors. These orthologs were recovered from the ORFeome library and tested in binary interactions (Table <xref ref-type="table" rid="T2">2B</xref>). We found <italic>kin-10 </italic>(CK2β) and <italic>rpl-6 </italic>positive for two reporter genes, and <italic>kin-3 </italic>(CK2α), <italic>hsp-6 </italic>(mortalin), F14E5.2 (cystein rich FGF receptor, CFR) and C18A3.3 (NoBP) positive for one reporter gene.</p></sec><sec><title>Identification of human LET-756-binding proteins</title><p>We based our third approach on the assumption that protein interaction motifs may be conserved between species. We made a heterologous screen using a cDNA human library as prey and LET-756 as bait. The yeast two-hybrid system was used with LET-756 fused to LexA DNA binding domain as bait (pDB) for the screening of a human placenta cDNA library containing the Gal-4 activating domain (pAD). Table <xref ref-type="table" rid="T3">3</xref> indicates the number of times the interactors were isolated and the strength of the interaction. Identified partners were different from those unveiled by the <italic>C. elegans </italic>screens but showed similar biologic activities (Table <xref ref-type="table" rid="T4">4</xref>). MBD1 and ZN420 are transcription factors, 14-3-3β (YWHAB, tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein, β polypeptide) has chaperone activity, and FBLI1 (Filamin-binding LIM protein 1) is a protein of the Zyxin family.</p><p>Finally, the orthologs of human 14-3-3β <italic>, ftt-1/par-5 </italic>and <italic>ftt-2</italic>, and the orthologs of FBL1, <italic>zyx-1a </italic>and <italic>zyx-1b</italic>, were obtained from the nematode ORFeome library and tested for direct interaction in yeast two-hybrid system. Table <xref ref-type="table" rid="T2">2C</xref> indicates that FTT-1 and FTT-2, but not ZYX-1a or ZYX-1b, reacted with the LET-756 bait.</p></sec><sec><title>Confirmation of interactions by co- immunoprecipitation experiments</title><p>A number of interactors we identified have been described as false positives in various studies. It is the case for HSP family members and ribosomal proteins, and to some extent for collagen-related proteins, Zn finger proteins and proteasome subunits [<xref ref-type="bibr" rid="B28">28</xref>]. To validate the interaction of LET-756 with the candidate partners, co-immunoprecipitation experiments were done in Cos-1 cells. Cos-1 cells were transiently cotransfected with HA-tagged partner constructs and LET-756::GFP, and the lysates were immunoprecipitated with anti-GFP. Fig. <xref ref-type="fig" rid="F1">1</xref> shows the result of a western blot probed first with anti-HA to reveal the co-immunoprecipitated proteins, and second with anti-GFP to normalize the transfection with LET-756. All tested partners immunoprecipitated with LET-756, although with different strength, unrelated to their level of expression (not shown). The 14-3-3β and KIN-10 proteins were reproducibly the less efficient. To make sure that overexpression of the two tagged proteins was not responsible for the immunoprecipitation, we used TACC1 as an unrelated HA-tagged control. In similar condition, TACC1 was unable to immunoprecipitate LET-756.</p></sec><sec><title>Confirmation of interactions by subcellular colocalization experiments</title><p>To further confirm FGF/partner interaction in mammalian cells, HA-tagged partners were co-expressed with LET-756::GFP in Cos-1 cells and their respective subcellular localization was examined. As already described [<xref ref-type="bibr" rid="B16">16</xref>,<xref ref-type="bibr" rid="B19">19</xref>], LET-756::GFP localized in specific regions of the nucleus where splicing factors are concentrated. Immunofluorescence microscopy using anti-HA antibodies revealed PAL-1 in foci in the nucleus that colocalized with LET-756 (Fig. <xref ref-type="fig" rid="F2">2</xref>). We have previously established [<xref ref-type="bibr" rid="B19">19</xref>] that treatment of LET-756 expressing cells with actinomycin D (a drug inhibiting Pol I and Pol II activities) displaces LET-756 to the perinucleolar and nucleolar compartments. Addition of actinomycin D did not move the PAL-1 protein to the nucleolus as it did for LET-756 but kept both partners in close association in nucleoplasmic foci. Other partners, such as RPL-6, were delocalized by transfection of LET-756. RPL-6 was localized in large foci mainly in the nucleoplasm when transfected alone (column I, Fig. <xref ref-type="fig" rid="F2">2</xref>) but was most often dispersed through the nucleoplasm and associated with LET-756 when cotransfected with the FGF (Fig. <xref ref-type="fig" rid="F2">2</xref>, column II to IV). FTT-1/PAR-5 and FTT-2 localized preferentially in the cytoplasm when transfected alone, but observed also in the nucleus when transfected together with LET-756. In addition, vesicles containing both LET-756 and FTT-1 or FTT-2 were visible. KIN-3 colocalized with LET-756 in nuclei and exhibited a strong expression in cytoplasm whether LET-756 was present or not whereas KIN-10 present in nucleoplasm of untransfected LET-756 cells moved with LET-756 in the speckles when co-transfected. Upon actinomycin D treatment both LET-756 and KIN-10 formed enlarged speckles and moved to the nucleolus (Fig. <xref ref-type="fig" rid="F2">2</xref>). This stricking delocalization observed upon actinomycin D treated KIN-10 co-transfected cells did not occurred with KIN-3. In other instances, the partner modified LET-756 localization: RPS-16 concentrated LET-756 in large foci when both proteins were present in the nucleus (Fig. <xref ref-type="fig" rid="F2">2</xref>). The protein encoded by C15C6.2 did not show any gross colocalization (Fig. <xref ref-type="fig" rid="F2">2</xref>). Finally, COL-129 was localized only at the Golgi apparatus, whether LET-756 was present or not, and CFR was localized only in the cytoplasm.</p></sec></sec><sec><title>Discussion</title><p>Several growth factors are found in the nucleus in addition to their other localizations. This is the case for LET-756 but not for EGL-17, the other <italic>C. elegans </italic>FGF [<xref ref-type="bibr" rid="B29">29</xref>]. The role of LET-756 in the nucleus is not known. To help characterize this role we searched to identify intracellular binding partners of LET-756. By using different two-hybrid screens in yeast, we identified several proteins involved in various aspects of protein synthesis or degradation. Further analysis by co-immunoprecipitation and colocalization confirmed the interactions identified. We demonstrated that not only LET-756 could interact with mammalian partners as well as their orthologs (e.g 14-3-3β vs FTT-1/PAR-5 and FTT-2) but also that orthologs of mammalian FGF partners could interact with LET-756 (CK2β <italic>vs </italic>KIN-10). Analyses of these partners could be of interest in the study of mammalian FGFs.</p><p>The majority of the proteins we identified are nuclear, which was expected since the two-hybrid system needs the fusion to be targeted to the nucleus. However, some interactions have been identified as false positive in other screens [<xref ref-type="bibr" rid="B28">28</xref>]. By performing co-immunoprecipitations and studying subcellular localization, we validated the interaction of LET-756 with RPS-16, FTT-1/PAR-5, FTT-2, KIN-10 and RPL-6 with high score, and with PAL-1, DAF-21, SKR-2, KIN-3 and C18A3.3 with lower strength (see Table <xref ref-type="table" rid="T6">6</xref>). The function of some interacting partners is relevant to FGF biology. The 14-3-3/FTT-1/FTT-2 proteins, which belong to the highly conserved family of chaperone molecules transit to the nucleus and participate in nucleo-cytoplasmic transport, regulating intracellular transduction [<xref ref-type="bibr" rid="B30">30</xref>]. We did not find a 14-3-3 conventional phosphorylated binding site on LET-756. However, other domains have been involved in 14-3-3 binding, such as nuclear localization signals (31 for review). No role for FTT-1 or FTT-2 in modulating secretion has been assigned in <italic>C. elegans</italic>. It will be interesting to analyze whether interaction of these proteins with the EFVSVA motif of secretion described in LET-756 [<xref ref-type="bibr" rid="B16">16</xref>] causes its secretion. The interaction of LET-756 with PAL-1 is of interest because PAL-1 is also highly conserved during evolution. It is the ortholog of caudal (<italic>Drosophila</italic>), CDX1, 2 and 4 (mammals) and Xcad3 (Xenopus) paraHOX proteins. Caudal proteins are involved in the transcriptional regulation of multiple genes that are involved in posterior patterning [<xref ref-type="bibr" rid="B32">32</xref>]. The interaction of LET-756 with PAL-1 could activate the expression of various genes involved in nematode anterior-posterior development as it is the case for the interaction of Xenopus e-FGF with Xcad3 and the resulting activation of <italic>HOX </italic>genes [<xref ref-type="bibr" rid="B33">33</xref>,<xref ref-type="bibr" rid="B34">34</xref>]. In addition, <italic>pal-1 </italic>mutant exhibits aberrant cell position in posterior muscle cells [<xref ref-type="bibr" rid="B35">35</xref>], a site of LET-756 expression [<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B19">19</xref>] as well as in posterior hypodermis, a site of LET-756 action [<xref ref-type="bibr" rid="B17">17</xref>]. Both muscle and epidermis evolve from the C lineage. In the absence of PAL-1, the C blastomeres fail to develop. Protein phosphorylation by the coordinated activities of protein kinases and phosphatases is central to many signal transduction pathways. The combined action of LET-756/FGF, EGL-15 receptor, CLR-1 phosphatase (for a review see [<xref ref-type="bibr" rid="B14">14</xref>]) and KIN-3 and KIN-10, the respective catalytic and regulatory subunits of CK2, might regulate various processes involved in proliferation -as it is described for FGF1 and 2 [<xref ref-type="bibr" rid="B43">43</xref>,<xref ref-type="bibr" rid="B9">9</xref>] – or in other functions. KIN-3 and KIN-10 have been recently implicated in primary cilia biology [<xref ref-type="bibr" rid="B55">55</xref>]. Finally, some ribosomal proteins interact with mammalian FGF [<xref ref-type="bibr" rid="B36">36</xref>-<xref ref-type="bibr" rid="B38">38</xref>] to regulate their signaling and trafficking to the nucleus; reciprocally, FGFs may regulate ribosome biogenesis and protein synthesis during the G1 phase of the cell cycle. In contrast to these relevant interactors, others appear irrelevant, such as MBD1 since no methylation occurs in <italic>C. elegans</italic>.</p><p>The interactions revealed by the two-hybrid screens are rather weak. This could be due to 1) the high stringency associated with the system using the MAV 103/203 yeast strains; it is worth noting that in large-scale screenings of interactions no partner for LET-756 was found [<xref ref-type="bibr" rid="B20">20</xref>]; 2) a bad exposure of the binding site in the fusion proteins; 3) the existence of ternary interactions as seen in the ligand – tyrosine kinase receptor – heparan sulphate complex and 4) the need for post-translational modifications of the proteins that occur in mammalian cells and not in yeast, explaining why better interactions between glycosylated LET-756 [<xref ref-type="bibr" rid="B16">16</xref>] and various partners were obtained in immunoprecipitation and immunofluorescence experiments than in yeast two-hybrid screens.</p><p>Finally, it will be interesting to know whether the expression pattern of LET-756, which is mainly muscular and neuronal in the worm, overlap with that of the various partners. Search in the literature (56, 57) was not conclusive since the majority of the partners were found in eggs (FTT-1/PAR-5, FTT-2, DAF-21, PAL-1), intestine (HSP-6, SKR-2) or cuticle (COL-129).</p></sec><sec><title>Conclusion</title><p>We have conducted the first extensive search for LET-756 interactors and established the first interaction map of LET-756/FGF with FGF binding proteins (Fig. <xref ref-type="fig" rid="F3">3</xref> and Table <xref ref-type="table" rid="T6">6</xref>). This could help understand FGF functions. Proteins of interest were involved in developmental processes or in basic biochemical events such as ribosome biogenesis and protein synthesis. In addition, to get insight in the evolution of the FGF interactome network, which we have illustrated in Fig. <xref ref-type="fig" rid="F3">3</xref>, we tested 6 of 20 orthologs of human FGF interactors (Table <xref ref-type="table" rid="T6">6</xref>), and found KIN-10 and RPL-6 as new potential interactors. Looking for physical interactions in a physiological system will determine which of these interactions are essential.</p><p>In conclusion, 1) combining the yeast two-hybrid screen with bioinformatics and computational biology, we have delineated potential interactors of LET-756, and possibly of the entire FGF family; 2) comparative genomics analysis yielded valuable insights into conserved and divergent aspects of function, regulation, and evolution since not all pathways are conserved as demonstrated by the ortholog analysis; 3) the information given herein, although not complete, might be useful for people working in the field.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Yeast two-hybrid assays</title><p>A full-length <italic>let-756 </italic>transcript was fused in-frame with the coding sequence of the DNA binding domain (DB) of Gal4, and was used in a Y2H screen system as described in [<xref ref-type="bibr" rid="B20">20</xref>] for the <italic>C elegans </italic>libraries. Two worm libraries fused to the activation domain (AD) of Gal4, a cDNA and the AD-ORFeome libraries [<xref ref-type="bibr" rid="B53">53</xref>] were screened. The MAV103 yeast strain based Y2H assay contains three reporter genes (<italic>HIS3</italic>, <italic>lacZ </italic>and <italic>URA3</italic>) [<xref ref-type="bibr" rid="B54">54</xref>]. A cDNA human placenta library fused to the activation domain of LexA was also screened. The L40 yeast strain based Y2H assay contains only two reporters (<italic>HIS3 </italic>and <italic>lacZ</italic>). In this case, the full length <italic>let-756 </italic>transcript was cloned into the LexA DNA binding domain bait expression vector pBTM116B Kana. Yeast assays were done using conventional lithium acetate-based method. Clones were assigned scores for LacZ expression, growth on plates lacking histidine but containing 20 or 40 mM 3-amino triazol and in addition, growth on plates lacking uracil for the <italic>C. elegans </italic>libraries. To ascertain interactions, the gap repair technique was performed as in [<xref ref-type="bibr" rid="B54">54</xref>].</p></sec><sec><title>Plasmid construction</title><p>To generate prey-tagged expression vectors used in co-immunoprecipitation assay or immunofluorescence, the coding regions of various genes were amplified by PCR using as template the corresponding EST clones obtained from RZPD (Berlin, Germany) and then inserted in the expression vectors using Gateway technology (Invitrogen, Carlsbad, CA).</p><p>LET-756::GFP was obtained as previously described [<xref ref-type="bibr" rid="B16">16</xref>].</p></sec><sec><title>Cell culture and in vivo interaction assay</title><p>Cos-1 cells grown in DMEM supplemented with 10% fetal calf serum were plated in 60-mm dishes at a concentration of 2 × 10<sup>6 </sup>cells/dish and immediately transfected with 1μg DNA in Fugene, according to the manufacturer instructions. Twenty four hours after transfection, cells were lysed in 1 ml triton buffer (10 mM Tris, pH7.4, 100 mM NaCl, 2.5 mM MgCl2, 1% triton, 1 mM EDTA, 10 mM DTT). Detergent insoluble materials were removed by 30 min centrifugation at 13000 rpm at 4°C. Whole cell lysates were first incubated with protein G-sepharose beads and then with the relevant antibody for at least 2 hr. Protein G-sepharose beads were then added for another additional 2 hr and washed 3 times with lysis buffer. Bound proteins were eluted by boiling in SDS sample buffer and resolved on a 10% SDS-PAGE gel and analyzed by Western blots. For immunofluorescence analysis, cells grown on glass coverslips were fixed and permeabilized in 3.7% PAF and 0.1%Triton or in methanol for 6 min at -20°C. Similar results were obtained using these different modes of fixation. Cells were incubated with primary antibody for 1 hr and then incubated with Texas Red-conjugated secondary antibody for another hr. Plasmid LET-756::GFP was visualized by autofluorescence. Coverslips were examined using a Leica TCS NT confocal microscope.</p><p>The following antibodies were used: rat monoclonal anti-HA (12CA5) antibody was from Roche (Indianopolis, IN, USA), rabbit polyclonal anti-GFP from Abcam (Cambridge, UK), Texas Red anti-rat antibody from Molecular Probes (Eugene, OR, USA), peroxydase anti-mouse from Santa Cruz (Santa Cruz, CA, USA)</p></sec></sec><sec><title>Abbreviations</title><p>DMEM, Dulbecco's mofied Eagle's medium; FBS, fetal bovine serum; DTT dithiotreitol; Y2H, yeast two hybrid; GFP, green fluorescent protein</p></sec><sec><title>Authors' contributions</title><p>CP, YB and AH performed the two-hybrid screenings and the analyses of the data, CP and FC the gap-repair confirmation of clones, FC and RR the co-immunoprecipitation and immunofluorescence experiments, CP the art work. CP played the major role in the bioinformatics analysis. DB initiated the <italic>C. elegans </italic>project and helped draft the manuscript. RR conceived and coordinated the study and wrote the manuscript. All authors read and approved the final manuscript.</p></sec>
In vivo – in vitro toxicogenomic comparison of TCDD-elicited gene expression in Hepa1c1c7 mouse hepatoma cells and C57BL/6 hepatic tissue	<sec><title>Background</title><p><italic>In vitro </italic>systems have inherent limitations in their ability to model whole organism gene responses, which must be identified and appropriately considered when developing predictive biomarkers of <italic>in vivo </italic>toxicity. Systematic comparison of <italic>in vitro </italic>and <italic>in vivo </italic>temporal gene expression profiles were conducted to assess the ability of Hepa1c1c7 mouse hepatoma cells to model hepatic responses in C57BL/6 mice following treatment with 2,3,7,8-tetrachlorodibenzo-<italic>p</italic>-dioxin (TCDD).</p></sec><sec><title>Results</title><p>Gene expression analysis and functional gene annotation indicate that Hepa1c1c7 cells appropriately modeled the induction of xenobiotic metabolism genes <italic>in vivo</italic>. However, responses associated with cell cycle progression and proliferation were unique to Hepa1c1c7 cells, consistent with the cell cycle arrest effects of TCDD on rapidly dividing cells. In contrast, lipid metabolism and immune responses, representative of whole organism effects <italic>in vivo</italic>, were not replicated in Hepa1c1c7 cells.</p></sec><sec><title>Conclusion</title><p>These results identified inherent differences in TCDD-mediated gene expression responses between these models and highlighted the limitations of <italic>in vitro </italic>systems in modeling whole organism responses, and additionally identified potential predictive biomarkers of toxicity.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Dere</surname><given-names>Edward</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Boverhof</surname><given-names>Darrell R</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Burgoon</surname><given-names>Lyle D</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A4" corresp="yes" contrib-type="author"><name><surname>Zacharewski</surname><given-names>Timothy R</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib>	BMC Genomics	<sec><title>Background</title><p>Advances in microarray and related technologies continue to revolutionize biomedical research and are being incorporated into toxicology and risk assessment. These technologies not only facilitate a more comprehensive elucidation of the mechanisms of toxicity, but also support mechanistically-based quantitative risk assessment [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B5">5</xref>]. In addition, these technologies are being used to develop predictive toxicity screening assays to screen drug candidates with adverse characteristics earlier in the development pipeline in order to prioritize resources and maximize successes in clinical trials [<xref ref-type="bibr" rid="B6">6</xref>-<xref ref-type="bibr" rid="B8">8</xref>]. Comparable screening strategies are also being proposed to rank and prioritize commercial chemicals, natural products, and environmental contaminants that warrant further toxicological investigation. Traditionally, rodent models or surrogates for ecologically-relevant species are typically used in regulatory testing. However, public and regulatory pressure, especially in Europe, seek to minimize the use of animals in testing [<xref ref-type="bibr" rid="B9">9</xref>]. Similar policies in the US, such as the ICCVAM Authorization Act of 2000, provide guidelines to facilitate the regulatory acceptance of alternative testing methods. These initiatives combined with the need to assess an expanding list of drug candidates and commercial chemicals for toxicity, have increased demand for the development and implementation of high-throughput <italic>in vitro </italic>screening assays that are predictive of toxicity in humans and ecologically-relevant species.</p><p>Various <italic>in vitro </italic>hepatic models including the isolated perfused liver, precision cut liver slices, isolated primary liver cells and a number of immortalized liver cell lines, have been used as animal alternatives [<xref ref-type="bibr" rid="B10">10</xref>]. In addition to providing a renewable model, <italic>in vitro </italic>systems are a cost-effective alternative and are amenable to high-throughput screening. These models, particularly immortalized cell lines, also allow for more in-depth biochemical and molecular investigations, such as over-expression, knock-down, activation or inhibition strategies, thus further elucidating mechanisms of action. However, inherent limitations in the ability of cell cultures to model whole organism responses must also be considered when identifying putative biomarkers for high-throughput toxicity screening assays, and elucidating relevant mechanisms of toxicity that support quantitative risk assessment. Despite several <italic>in vitro </italic>toxicogenomic reports [<xref ref-type="bibr" rid="B11">11</xref>-<xref ref-type="bibr" rid="B13">13</xref>], few have systematically examined the ability of <italic>in vitro </italic>systems to predict <italic>in vivo </italic>gene expression profiles in response to chemical treatment [<xref ref-type="bibr" rid="B10">10</xref>,<xref ref-type="bibr" rid="B14">14</xref>].</p><p>2,3,7,8-Tetrachlorodibenzo-<italic>p</italic>-dioxin (TCDD) is a widespread environmental contaminant that elicits a number of adverse effects including tumor promotion, teratogenesis, hepatotoxicity, and immunotoxicity as well as the induction of several metabolizing enzymes [<xref ref-type="bibr" rid="B15">15</xref>]. Many, if not all of these effects, are due to alterations in gene expression mediated by the aryl hydrocarbon receptor (AhR), a basic-helix-loop-helix-PAS (bHLH-PAS) transcription factor [<xref ref-type="bibr" rid="B15">15</xref>,<xref ref-type="bibr" rid="B16">16</xref>]. Ligand binding to the cytoplasmic AhR complex triggers the dissociation of interacting proteins and results in the translocation of the ligand-bound AhR to the nucleus where it heterodimerizes with the aryl hydrocarbon receptor nuclear translocator (ARNT), another member of the bHLH-PAS family. The heterodimer then binds specific DNA elements, termed dioxin response elements (DREs), within the regulatory regions of target genes leading to changes in expression that ultimately result in the observed responses [<xref ref-type="bibr" rid="B17">17</xref>]. Although the role of AhR is well established, the gene regulatory pathways responsible for toxicity are poorly understood and warrant further investigation to assess the potential risks to humans and ecologically relevant species.</p><p>Hepa1c1c7 cells and C57BL/6 mice are well-established models routinely used to examine the mechanisms of action of TCDD and related compounds. In this study, TCDD-elicited temporal gene expression effects were systematically compared in order to assess the ability of Hepa1c1c7 cells to replicate C57BL/6 hepatic tissue responses. Our results indicate that several phase I and II metabolizing enzyme responses are aptly reproduced. However, many responses were model-specific and reflect inherent <italic>in vitro </italic>and <italic>in vivo </italic>differences that must be considered in mechanistic studies and during the selection of biomarkers for developing toxicity screening assays.</p></sec><sec><title>Results</title><sec><title>In vitro microarray data analysis</title><p>Temporal gene expression profiles were assessed in Hepa1c1c7 wild type cells following treatment with 10 nM TCDD using cDNA microarrays with 13,362 spotted features. Empirical Bayes analysis of the <italic>in vitro </italic>time course data identified 331 features representing 285 unique genes with a P1(<italic>t</italic>) value greater than 0.9999 at one or more time points, and differential expression greater than ± 1.5 fold relative to time-matched vehicle controls. The number of differentially regulated genes gradually increased from 1 to 24 hrs, followed by a slight decrease at 48 hrs (Figure <xref ref-type="fig" rid="F1">1A</xref>). <italic>In vitro </italic>dose-response data performed at 12 hrs with TCDD covering 6 different concentrations (0.001, 0.01, 0.1, 1.0, 10 and 100 nM), identified 181 features representing 155 unique genes (P1(<italic>t</italic>) > 0.9999 and an absolute fold change > 1.5 at one or more doses; Figure <xref ref-type="fig" rid="F1">1B</xref>). Complete <italic>in vitro </italic>time course and dose-response data are available in Additional file <xref ref-type="supplementary-material" rid="S1">1</xref> and <xref ref-type="supplementary-material" rid="S2">2</xref>, respectively.</p><fig position="float" id="F1"><label>Figure 1</label><caption><p>Number of genes differentially regulated (P 1(<italic>t</italic>) > 0.9999 and Ifold changel > 1.5-fold) as measured by microarray analysis for the (A) time course and (B) and dose-response studies in mouse hepatoma Hepa1c1c7 cells. For the time course study, cells were treated with 10 nM TCDD and harvested at 1, 2, 4, 8, 12, 24 or 48 hrs after treatment. Cells for the 12 hr dose-response study were treated with 0.001, 0.01, 0.1, 1.0, 10 and 100 nM of TCDD</p></caption><graphic xlink:href="1471-2164-7-80-1"/></fig><p>As a control, the gene expression effects elicited by 10 nM TCDD in ARNT-deficient c4 Hepa1c1c7 mutants [<xref ref-type="bibr" rid="B18">18</xref>] were examined at 1 and 24 hrs (data not shown). Only ATPase, H<sup>+ </sup>transporting, V1 subunit E-like 2 isoform 2 (Atp6v1e2) and SUMO/sentrin specific peptidase 6 (Senp6) exhibited a significant change in expression using the same criteria (P1(<italic>t</italic>) > 0.9999 and an absolute fold change > 1.5). Neither Atp6v1e2 nor Senp6 were among the active genes in wild-type Hepa1c1c7 cells or in C57BL/6 liver samples [<xref ref-type="bibr" rid="B19">19</xref>]. These results provide further evidence that the AhR/ARNT signaling pathway mediates TCDD-elicited gene expression responses, which are consistent with <italic>in vivo </italic>microarray results with AhR knockout mice [<xref ref-type="bibr" rid="B20">20</xref>].</p><p>Hierarchical clustering of the genes expressed in Hepa1c1c7 time course indicate that 2 and 4 hrs were most similar, as were 8 and 12 hrs, and 24 and 48 hrs, while the 1 hr time point was segregated (Figure <xref ref-type="fig" rid="F2">2A</xref>). A strong dose-response relationship was also evident with clusters sequentially branching out with increasing concentration (Figure <xref ref-type="fig" rid="F2">2B</xref>). At 12 hrs, 117 genes were differentially expressed with 112 exhibiting a dose-dependent response. Moreover, the fold changes measured in both the time course and dose-response studies using 10 nM TCDD were comparable. For example, xanthine dehydrogenase (Xdh) and NAD(P)H dehydrogenase, quinone 1 (Nqo1) were induced 2.39- and 4.89-fold respectively in the time course and 2.93- and 4.71-fold in the dose-response study. There is a strong correlation (R = 0.97) between the differentially expressed genes at 12 hrs in the time course with the differentially regulated genes in the dose-response study at 10 nM, demonstrating the reproducibility between independent studies and providing further evidence that these genes are regulated by TCDD.</p><fig position="float" id="F2"><label>Figure 2</label><caption><p>Hierarchical clustering of the differentially regulated gene lists for A) temporal and B) dose-response microarray studies in mouse hepatoma Hepa1c1c7 cells. The results illustrate time and dose-dependent clustering patterns. From the A) temporal results, the early (2 hr and 4 hr), intermediate (8 hr and 12 hr) and late (24 hr and 48 hr) time points cluster separately while the 1 hr time point clusters alone. Results from the B) dose-response show that highest doses clustered together, while the remaining doses branched out in a dose-dependent manner</p></caption><graphic xlink:href="1471-2164-7-80-2"/></fig><p>The list of temporally regulated genes was subjected to <italic>k</italic>-means clustering using the standard correlation distance metrics. Five <italic>k</italic>-means clusters best characterized the dataset and identified clusters representing A) up-regulated early and sustained, B) up-regulated intermediate and sustained, C) up-regulated intermediate, D) up-regulated immediate and E) down-regulated late (Figure <xref ref-type="fig" rid="F3">3</xref>). These were comparable to the <italic>k</italic>-means clusters identified in hepatic tissue of C57BL/6 mice following treatment with 30 μg/kg TCDD [<xref ref-type="bibr" rid="B19">19</xref>]. Although, no discernable functional category is over-represented in any one cluster, the sustained up-regulation of early (Cluster A) and intermediate (Cluster B) responding genes include classic TCDD-responsive genes such as cytochrome P450, family 1, subfamily a, polypeptide 1 (Cyp1a1), Xdh and Nqo1. Many down-regulated late genes were associated with cell cycle regulation such as myelocytomatosis oncogene (Myc). Additionally, targets of Myc, including cyclin D1 and ornithine decarboxylase (Odc1), were also down-regulated suggesting a mechanism for cell cycle arrest [<xref ref-type="bibr" rid="B21">21</xref>-<xref ref-type="bibr" rid="B23">23</xref>], a common <italic>in vitro </italic>response to TCDD.</p><fig position="float" id="F3"><label>Figure 3</label><caption><p>K-means clustering of temporally differentially regulated genes <italic>in vitro</italic>. Five k-mean clusters corresponding to (A) up-early and sustained, (B) up-intermediate and sustained, (C) up-regulated intermediate, (D) up-regulated immediate and (E) down-regulated late. Time and expression ratio are indicated on the <italic>x</italic>- and y-axis respectively. The color of individual gene expression profiles reflects the expression ratio observed at 24 hrs</p></caption><graphic xlink:href="1471-2164-7-80-3"/></fig></sec><sec><title>Classification of gene expression responses for common regulated genes</title><p>Using the same filtering criteria (P1(<italic>t</italic>) > 0.9999 and an absolute fold change > 1.5), 678 features representing 619 unique genes were differentially expressed as previously reported in a time course study conducted in hepatic tissue from C57BL/6 mice orally gavaged with 30 μg/kg TCDD [<xref ref-type="bibr" rid="B19">19</xref>]. The number of responsive <italic>in vivo </italic>genes and their temporal expression patterns closely paralleled the results from this <italic>in vitro </italic>study. The fewest number of active genes was observed at 2 hrs, followed by a large increase at 4 hrs, which was sustained to 72 hrs. However, the substantial increase in expressed <italic>in vivo </italic>genes at 168 hrs was attributed to triglyceride accumulation and immune cell infiltration, which was not observed in Hepa1c1c7 cells. This list of 619 of <italic>in vivo </italic>genes served as the basis for subsequent comparisons against TCDD-elicited <italic>in vitro </italic>responses.</p><p>Comparison of <italic>in vitro </italic>and <italic>in vivo </italic>differentially expressed gene lists identified common and model specific responses (Figure <xref ref-type="fig" rid="F4">4A</xref>). TCDD treatment resulted in a total of 838 regulated genes in either model and with 67 common to both. TCDD elicited 218 gene expression changes unique to Hepa1c1c7 cells while 552 genes were specific to C57BL/6 hepatic samples. Although 67 genes were regulated in both models, not all possessed similar temporal patterns of expression. Contingency analysis using a 2 × 2 table and the χ<sup>2 </sup>test resulted in a p-value < 0.001 (α = 0.05) illustrate a statistically significant association between the lists of differentially regulated genes <italic>in vitro </italic>and <italic>in vivo</italic>. Further stratification revealed genes that were either induced in both models (class I), repressed in both models (class II), induced <italic>in vivo </italic>while repressed <italic>in vitro </italic>(class III), or repressed <italic>in vivo </italic>while induced <italic>in vitro </italic>(class IV; Figure <xref ref-type="fig" rid="F4">4B</xref>). Genes regulated in a similar fashion in both models (classes I and II) accounted for 49 of the 67 common active genes, while the remaining genes exhibited divergent expression profiles (classes III and IV). Hierarchical clustering of the temporal expression values for the 67 overlapping genes identified the same four classes (Figure <xref ref-type="fig" rid="F4">4C</xref>). The pattern across model and time illustrates that the earliest time points (i.e. 1 hr <italic>in vitro </italic>and 2 hr <italic>in vivo </italic>time points) cluster together while the remaining clusters branch into <italic>in vitro </italic>or <italic>in vivo </italic>clusters according to time. These results suggest that potential biomarkers of acute TCDD-mediated responses may best be predicted by the immediate-early <italic>in vitro </italic>gene responses.</p><fig position="float" id="F4"><label>Figure 4</label><caption><p>Comparison of common significant <italic>in vitro </italic>and <italic>in vivo </italic>TCDD elicited time-dependent gene expression changes. A) 285 differentially regulated <italic>in vitro </italic>genes and 619 differentially regulated <italic>in vivo </italic>genes were identified, with 67 genes common to both studies. B) The temporal gene expression profiles from both studies were categorized into (I) induced in both, (II) repressed in both, (III) induced <italic>in vivo </italic>and repressed <italic>in vitro</italic>, and (IV) repressed <italic>in vitro </italic>and induced <italic>in vivo</italic>. C) Hierarchical clustering identified similar classification groups. Clustering across both time and model, separated samples from <italic>in vitro </italic>and <italic>in vivo</italic>, with the exception of the early time points from both studies (1 hr <italic>in vitro </italic>and 2 hr <italic>in vivo</italic>), which clustered together. * identifies <italic>in vitro </italic>time points</p></caption><graphic xlink:href="1471-2164-7-80-4"/></fig><p><italic>In vitro </italic>and <italic>in vivo </italic>induced genes (class I) include xenobiotic and oxidoreductase enzymes such as abhydrolase domain containing 6 (Abhd6), Cyp1a1, dehydrogenase/reductase (SDR family) member 3 (Dhrs3), Nqo1, prostaglandin-endoperoxide synthase 1 (Ptgs1), UDP-glucose dehydrogenase (Ugdh) and Xdh (Table <xref ref-type="table" rid="T1">1</xref>). These genes have previously been reported to be TCDD-responsive [<xref ref-type="bibr" rid="B19">19</xref>,<xref ref-type="bibr" rid="B24">24</xref>], with Cyp1a1 and Nqo1 being members of the "AhR gene battery" [<xref ref-type="bibr" rid="B25">25</xref>]. Glutathione S-transferase, alpha 4 (Gsta4) was also induced <italic>in vitro </italic>and <italic>in vivo</italic>, 1.7- and 2.0-fold respectively, consistent with TCDD-mediated induction of phase I and II metabolizing enzymes. Of the 35 genes responding similarly in both models, approximately 71% of were similarly up-regulated (class I) while the remaining genes were repressed across both models (class II). Repressed class II genes include minichromosome maintenance deficient 6 (Mcm6), glycerol kinase (Gyk) and ficolin A (Fcna) (repressed 1.6-, 1.6- and 1.7-fold <italic>in vitro</italic>, respectively). Overall, repressed genes did not share any common discernable biological function.</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Classification of common differentially regulated temporal gene expression responses to TCDD in both <italic>in vitro </italic>and <italic>in vivo </italic>models</p></caption><table frame="hsides" rules="groups"><tbody><tr><td align="left">Accession</td><td align="left">Gene name</td><td align="center">Gene symbol</td><td align="center">Entrez Gene ID</td><td align="center" colspan="3"><italic>In vivo</italic></td><td align="center" colspan="3"><italic>In vitro</italic></td></tr><tr><td></td><td></td><td></td><td></td><td colspan="3"><hr></hr></td><td colspan="3"><hr></hr></td></tr><tr><td></td><td></td><td></td><td></td><td align="center">Fold change<sup>a</sup></td><td align="center">Time points<sup>b</sup></td><td align="center">EC50<sup>c,d </sup>(μg/kg)</td><td align="center">Fold change<sup>a</sup></td><td align="center">Time points<sup>b</sup></td><td align="center">EC50<sup>c,d </sup>(pM)</td></tr><tr><td align="left" colspan="10">I) Induced both <italic>in vivo </italic>and <italic>in vitro</italic><sup>e</sup></td></tr><tr><td align="left">BE689910</td><td align="left">RIKEN cDNA 2310001H12 gene</td><td align="center">2310001H12Rik</td><td align="center">69504</td><td align="center">2.7</td><td align="center">2<sup>f</sup>, 168</td><td align="right">48.02</td><td align="right">3.9</td><td align="center">1<sup>f</sup></td><td align="right">ND</td></tr><tr><td align="left">BF226070</td><td align="left">RIKEN cDNA 2600005C20 gene</td><td align="center">2600005C20Rik</td><td align="center">72462</td><td align="center">2.1</td><td align="center">4, 12, 18, 24<sup>f</sup>, 72, 168</td><td align="right">2.18</td><td align="right">2.3</td><td align="center">4, 8, 12<sup>f</sup>, 24, 48</td><td align="right">265.50</td></tr><tr><td align="left">AI043124</td><td align="left">RIKEN cDNA 2810003C17 gene</td><td align="center">2810003C17Rik</td><td align="center">108897</td><td align="center">1.6</td><td align="center">12<sup>f</sup></td><td align="right">37.02</td><td align="right">1.7</td><td align="center">4<sup>f</sup></td><td align="right">ND</td></tr><tr><td align="left">AW537038</td><td align="left">expressed sequence AA959742</td><td align="center">AA959742</td><td align="center">98238</td><td align="center">7.2</td><td align="center">4, 8, 12<sup>f</sup>, 18, 24, 72, 168</td><td align="right">1.71</td><td align="right">5.2</td><td align="center">4, 8<sup>f</sup>, 12, 24, 48</td><td align="right">67.79</td></tr><tr><td align="left">W34507</td><td align="left">abhydrolase domain containing 6</td><td align="center">Abhd6</td><td align="center">66082</td><td align="center">1.7</td><td align="center">4, 8<sup>f</sup>, 12, 18, 24, 72, 168</td><td align="right">154.30</td><td align="right">1.5</td><td align="center">48<sup>f</sup></td><td align="right">138.50</td></tr><tr><td align="left">NM_026410</td><td align="left">cell division cycle associated 5</td><td align="center">Cdca5</td><td align="center">67849</td><td align="center">8.8</td><td align="center">4, 8, 12, 18, 24, 72<sup>f</sup>, 168</td><td align="right">ND</td><td align="right">1.7</td><td align="center">4<sup>f</sup></td><td align="right">ND</td></tr><tr><td align="left">BG063743</td><td align="left">craniofacial development protein 1</td><td align="center">Cfdp1</td><td align="center">23837</td><td align="center">3.6</td><td align="center">4, 8, 12<sup>f</sup>, 18, 24, 72, 168</td><td align="right">14.27</td><td align="right">2.3</td><td align="center">4<sup>f</sup>, 8, 12, 24, 48</td><td align="right">42.64</td></tr><tr><td align="left">AA073604</td><td align="left">procollagen, type I, alpha 1</td><td align="center">Col1a1</td><td align="center">12842</td><td align="center">1.7</td><td align="center">18, 24, 72<sup>f</sup></td><td align="right">0.65</td><td align="right">1.6</td><td align="center">4, 8, 12<sup>f</sup></td><td align="right">17.25</td></tr><tr><td align="left">NM_009992</td><td align="left">cytochrome P450, family 1, subfamily a, polypeptide 1</td><td align="center">Cyp1a1</td><td align="center">13076</td><td align="center">38.4</td><td align="center">2, 4, 8, 12, 18, 24<sup>f</sup>, 72, 168</td><td align="right">0.05</td><td align="right">37.7</td><td align="center">1, 2, 4, 8, 12, 24, 48<sup>f</sup></td><td align="right">14.06</td></tr><tr><td align="left">BE457542</td><td align="left">Dehydrogenase/reductase (SDR family) member 3</td><td align="center">Dhrs3</td><td align="center">20148</td><td align="center">2.0</td><td align="center">4, 8, 12<sup>f</sup>, 18, 72, 168</td><td align="right">0.67</td><td align="right">1.5</td><td align="center">8<sup>f</sup></td><td align="right">2.43</td></tr><tr><td align="left">AW552715</td><td align="left">DnaJ (Hsp40) homolog, subfamily B, member 11</td><td align="center">Dnajb11</td><td align="center">67838</td><td align="center">1.7</td><td align="center">12, 18, 24<sup>f</sup>,168</td><td align="right">3.95</td><td align="right">1.6</td><td align="center">8, 12<sup>f</sup></td><td align="right">9.85</td></tr><tr><td align="left">AK015223</td><td align="left">dermatan sulphate proteoglycan 3</td><td align="center">Dspg3</td><td align="center">13516</td><td align="center">6.2</td><td align="center">4, 8, 12, 18, 24<sup>f</sup>, 72, 168</td><td align="right">0.13</td><td align="right">8.4</td><td align="center">2, 4, 8, 12, 24, 48<sup>f</sup></td><td align="right">16.34</td></tr><tr><td align="left">NM_008655</td><td align="left">growth arrest and DNA-damage-inducible 45 beta</td><td align="center">Gadd45b</td><td align="center">17873</td><td align="center">4.6</td><td align="center">2<sup>f</sup>, 4, 72</td><td align="right">133.30</td><td align="right">3.7</td><td align="center">1<sup>f</sup>, 2</td><td align="right">1440.00</td></tr><tr><td align="left">W54349</td><td align="left">glutathione S-transferase, alpha 4</td><td align="center">Gsta4</td><td align="center">14860</td><td align="center">2.0</td><td align="center">18, 24, 72<sup>f</sup></td><td align="right">0.48</td><td align="right">1.7</td><td align="center">8<sup>f</sup>, 12</td><td align="right">56.38</td></tr><tr><td align="left">BG067127</td><td align="left">interferon regulatory factor 1</td><td align="center">Irf1</td><td align="center">16362</td><td align="center">1.5</td><td align="center">168<sup>f</sup></td><td align="right">ND</td><td align="right">1.7</td><td align="center">2<sup>f</sup>, 4</td><td align="right">ND</td></tr><tr><td align="left">AA015278</td><td align="left">integrin beta 1 (fibronectin receptor beta)</td><td align="center">Itgb1</td><td align="center">16412</td><td align="center">1.6</td><td align="center">4, 18, 24, 168<sup>f</sup></td><td align="right">97.23</td><td align="right">4.2</td><td align="center">4, 8, 12, 24, 48<sup>f</sup></td><td align="right">72.92</td></tr><tr><td align="left">AA041752</td><td align="left">Jun proto-oncogene related gene d1</td><td align="center">Jund1</td><td align="center">16478</td><td align="center">2.0</td><td align="center">12<sup>f</sup>, 18, 24</td><td align="right">0.99</td><td align="right">2.1</td><td align="center">4, 8, 12, 24, 48<sup>f</sup></td><td align="right">50.34</td></tr><tr><td align="left">BF538945</td><td align="left">lectin, mannose-binding, 1</td><td align="center">Lman1</td><td align="center">70361</td><td align="center">1.9</td><td align="center">12, 72, 168<sup>f</sup></td><td align="right">13.49</td><td align="right">2.0</td><td align="center">4<sup>f</sup>, 8, 24, 48</td><td align="right">40.72</td></tr><tr><td align="left">BG066626</td><td align="left">lipin 2</td><td align="center">Lpin2</td><td align="center">64898</td><td align="center">3.0</td><td align="center">4, 12, 24<sup>f</sup>, 72</td><td align="right">3.13</td><td align="right">2.3</td><td align="center">2, 4<sup>f</sup>, 8, 12, 24, 48</td><td align="right">23.83</td></tr><tr><td align="left">BI440950</td><td align="left">leucine rich repeat containing 39</td><td align="center">Lrrc39</td><td align="center">109245</td><td align="center">2.9</td><td align="center">2<sup>f</sup>, 4</td><td align="right">49.71</td><td align="right">3.1</td><td align="center">1<sup>f</sup>, 2</td><td align="right">68.57</td></tr><tr><td align="left">AW413953</td><td align="left">mitochondrial ribosomal protein L37</td><td align="center">Mrpl37</td><td align="center">56280</td><td align="center">8.3</td><td align="center">2, 4, 8<sup>f</sup>, 12, 18, 24, 72, 168</td><td align="right">8.77</td><td align="right">2.7</td><td align="center">2, 4<sup>f</sup>, 8, 12, 24, 48</td><td align="right">49.59</td></tr><tr><td align="left">BE623489</td><td align="left">NAD(P)H dehydrogenase, quinone 1</td><td align="center">Nqo1</td><td align="center">18104</td><td align="center">4.6</td><td align="center">4, 8, 12<sup>f</sup>, 18, 24, 72, 168</td><td align="right">1.00</td><td align="right">5.2</td><td align="center">4, 8<sup>f</sup>, 12, 24, 48</td><td align="right">33.74</td></tr><tr><td align="left">NM_026550</td><td align="left">PAK1 interacting protein 1</td><td align="center">Pak1ip1</td><td align="center">68083</td><td align="center">3.8</td><td align="center">4, 8, 12, 18, 24, 72<sup>f</sup>, 168</td><td align="right">0.26</td><td align="right">2.2</td><td align="center">4<sup>f</sup>, 8, 12, 24, 48</td><td align="right">7.00</td></tr><tr><td align="left">AA152754</td><td align="left">prostaglandin-endoperoxide synthase 1</td><td align="center">Ptgs1</td><td align="center">19224</td><td align="center">1.6</td><td align="center">168<sup>f</sup></td><td align="right">1.11</td><td align="right">2.3</td><td align="center">4, 8<sup>f</sup>, 12, 24, 48</td><td align="right">37.96</td></tr><tr><td align="left">BG063583</td><td align="left">solute carrier family 20, member 1</td><td align="center">Slc20a1</td><td align="center">20515</td><td align="center">2.2</td><td align="center">2, 4<sup>f</sup>, 8</td><td align="right">ND</td><td align="right">1.8</td><td align="center">2<sup>f</sup>, 4</td><td align="right">ND</td></tr><tr><td align="left">AJ223958</td><td align="left">solute carrier family 27 (fatty acid transporter), member 2</td><td align="center">Slc27a2</td><td align="center">26458</td><td align="center">1.9</td><td align="center">12<sup>f</sup>, 18, 24, 72, 168</td><td align="right">2.88</td><td align="right">2.1</td><td align="center">8, 12<sup>f</sup>, 24, 48</td><td align="right">17.42</td></tr><tr><td align="left">BG066820</td><td align="left">solute carrier family 6 (neurotransmitter transporter, taurine), member 6</td><td align="center">Slc6a6</td><td align="center">21366</td><td align="center">1.8</td><td align="center">4<sup>f</sup>, 12</td><td align="right">2.48</td><td align="right">1.7</td><td align="center">48<sup>f</sup></td><td align="right">3.06</td></tr><tr><td align="left">AI592773</td><td align="left">suppression of tumorigenicity 5</td><td align="center">St5</td><td align="center">76954</td><td align="center">1.6</td><td align="center">8, 12<sup>f</sup></td><td align="right">28.85</td><td align="right">1.7</td><td align="center">4<sup>f</sup>, 8, 12</td><td align="right">14.69</td></tr><tr><td align="left">BG067168</td><td align="left">TCDD-inducible poly(ADP-ribose) Polymerase</td><td align="center">Tiparp</td><td align="center">99929</td><td align="center">10.3</td><td align="center">2, 4<sup>f</sup>, 12, 18, 24, 72, 168</td><td align="right">36.49</td><td align="right">6.4</td><td align="center">1, 2<sup>f</sup>, 4, 8, 12, 24, 48</td><td align="right">18.03</td></tr><tr><td align="left">BG065761</td><td align="left">tumor necrosis factor, alpha-induced protein 2</td><td align="center">Tnfaip2</td><td align="center">21928</td><td align="center">5.5</td><td align="center">2, 4<sup>f</sup>, 12, 18, 72</td><td align="right">36.41</td><td align="right">6.3</td><td align="center">2, 4<sup>f</sup>, 8, 12, 24, 48</td><td align="right">41.15</td></tr><tr><td align="left">AA067191</td><td align="left">UDP-glucose dehydrogenase</td><td align="center">Ugdh</td><td align="center">22235</td><td align="center">3.1</td><td align="center">4, 8, 12 <sup>f</sup>, 18, 24, 72, 168</td><td align="right">0.79</td><td align="right">1.5</td><td align="center">2, 4, 8, 12<sup>f</sup>, 48</td><td align="right">4.33</td></tr><tr><td align="left">NM_011709</td><td align="left">whey acidic protein</td><td align="center">Wap</td><td align="center">22373</td><td align="center">5.9</td><td align="center">2, 4, 8, 12, 18, 24, 72, 168<sup>f</sup></td><td align="right">0.12</td><td align="right">4.2</td><td align="center">2, 4, 8, 12, 24, 48<sup>f</sup></td><td align="right">17.44</td></tr><tr><td align="left">BG075778</td><td align="left">Xanthine dehydrogenase</td><td align="center">Xdh</td><td align="center">22436</td><td align="center">2.7</td><td align="center">4, 8, 12<sup>f</sup>, 18, 24, 72, 168</td><td align="right">1.24</td><td align="right">2.6</td><td align="center">4, 8, 12, 24, 48<sup>f</sup></td><td align="right">34.92</td></tr><tr><td align="left">BG073881</td><td align="left">zinc finger protein 36, C3H type-like 1</td><td align="center">Zfp36l1</td><td align="center">12192</td><td align="center">2.2</td><td align="center">2<sup>f</sup></td><td align="right">ND</td><td align="right">1.7</td><td align="center">1, 2<sup>f</sup></td><td align="right">2427.00</td></tr><tr><td align="left">AA031146</td><td align="left">zinc finger protein 672</td><td align="center">Zfp672</td><td align="center">319475</td><td align="center">1.6</td><td align="center">4<sup>f</sup></td><td align="right">3.09</td><td align="right">1.5</td><td align="center">2<sup>f</sup></td><td align="right">ND</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left" colspan="10">II) Repressed both <italic>in vivo </italic>and <italic>in vitro</italic><sup>e</sup></td></tr><tr><td align="left">BG146493</td><td align="left">RIKEN cDNA 6330406L22 gene</td><td align="center">6330406L22Rik</td><td align="center">70719</td><td align="center">-1.5</td><td align="center">18<sup>f</sup></td><td align="right">0.51</td><td align="right">-1.8</td><td align="center">8, 12<sup>f</sup></td><td align="right">25.67</td></tr><tr><td align="left">AA140059</td><td align="left">DNA methyltransferase (cytosine-5) 1</td><td align="center">Dnmt1</td><td align="center">13433</td><td align="center">-1.9</td><td align="center">168<sup>f</sup></td><td align="right">ND</td><td align="right">-1.6</td><td align="center">8<sup>f</sup>, 12</td><td align="right">ND</td></tr><tr><td align="left">AI327022</td><td align="left">ficolin A</td><td align="center">Fcna</td><td align="center">14133</td><td align="center">-1.6</td><td align="center">18, 24<sup>f</sup></td><td align="right">ND</td><td align="right">-1.7</td><td align="center">12, 24<sup>f</sup></td><td align="right">ND</td></tr><tr><td align="left">AA288963</td><td align="left">fibrinogen-like protein 1</td><td align="center">Fgl1</td><td align="center">234199</td><td align="center">-1.9</td><td align="center">24<sup>f</sup></td><td align="right">ND</td><td align="right">-1.5</td><td align="center">24<sup>f</sup>, 48</td><td align="right">ND</td></tr><tr><td align="left">BE626913</td><td align="left">GTP binding protein 6 (putative)</td><td align="center">Gtpbp6</td><td align="center">107999</td><td align="center">-3.4</td><td align="center">24, 72<sup>f</sup></td><td align="right">ND</td><td align="right">-1.7</td><td align="center">24, 48<sup>f</sup></td><td align="right">116.40</td></tr><tr><td align="left">AA275564</td><td align="left">glycerol kinase</td><td align="center">Gyk</td><td align="center">14933</td><td align="center">-1.5</td><td align="center">12<sup>f</sup></td><td align="right">10.2</td><td align="right">-1.6</td><td align="center">24<sup>f</sup></td><td align="right">ND</td></tr><tr><td align="left">BG070106</td><td align="left">lipocalin 2</td><td align="center">Lcn2</td><td align="center">16819</td><td align="center">-2.8</td><td align="center">24<sup>f</sup></td><td align="right">ND</td><td align="right">-1.5</td><td align="center">24, 48<sup>f</sup></td><td align="right">ND</td></tr><tr><td align="left">AW049427</td><td align="left">leucine zipper domain protein</td><td align="center">Lzf</td><td align="center">66049</td><td align="center">-1.6</td><td align="center">24<sup>f</sup></td><td align="right">ND</td><td align="right">-1.6</td><td align="center">48<sup>f</sup></td><td align="right">78.29</td></tr><tr><td align="left">AA016759</td><td align="left">minichromosome maintenance deficient 6</td><td align="center">Mcm6</td><td align="center">17219</td><td align="center">-1.6</td><td align="center">18<sup>f</sup></td><td align="right">3.34</td><td align="right">-1.6</td><td align="center">8<sup>f</sup></td><td align="right">58.04</td></tr><tr><td align="left">BF011268</td><td align="left">mitochondrial methionyl-tRNA formyltransferase</td><td align="center">Mtfmt</td><td align="center">69606</td><td align="center">-1.8</td><td align="center">24, 72<sup>f</sup>, 168</td><td align="right">ND</td><td align="right">-1.6</td><td align="center">24, 48<sup>f</sup></td><td align="right">ND</td></tr><tr><td align="left">AA683699</td><td align="left">RNA (guanine-7-) methyltransferase</td><td align="center">Rnmt</td><td align="center">67897</td><td align="center">-2.0</td><td align="center">12<sup>f</sup></td><td align="right">ND</td><td align="right">-1.6</td><td align="center">8<sup>f</sup></td><td align="right">ND</td></tr><tr><td></td><td align="left">syntrophin, gamma 1</td><td align="center">Sntg1</td><td align="center">71096</td><td align="center">-1.6</td><td align="center">24<sup>f</sup></td><td align="right">15.27</td><td align="right">-1.7</td><td align="center">4<sup>f</sup></td><td align="right">66.61</td></tr><tr><td align="left">AA199550</td><td align="left">syntaxin 12</td><td align="center">Stx12</td><td align="center">100226</td><td align="center">-1.5</td><td align="center">18<sup>f</sup></td><td align="right">ND</td><td align="right">-1.6</td><td align="center">48<sup>f</sup></td><td align="right">ND</td></tr><tr><td align="left">AA047942</td><td align="left">thymidine kinase 1</td><td align="center">Tk1</td><td align="center">21877</td><td align="center">-1.7</td><td align="center">18<sup>f</sup>, 24, 72</td><td align="right">0.34</td><td align="right">-2.0</td><td align="center">8, 12<sup>f</sup></td><td align="right">153.90</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left" colspan="10">III) Induced <italic>in vivo </italic>and repressed <italic>in vitro</italic><sup>e</sup></td></tr><tr><td align="left">AA122925</td><td align="left">carbonic anhydrase 2</td><td align="center">Car2</td><td align="center">12349</td><td align="center">2.4</td><td align="center">12, 72, 168<sup>f</sup></td><td align="right">2.00</td><td align="right">-1.8</td><td align="center">24<sup>f</sup>, 48</td><td align="right">55.96</td></tr><tr><td align="left">AI327078</td><td align="left">coactosin-like 1</td><td align="center">Cotl1</td><td align="center">72042</td><td align="center">1.6</td><td align="center">168<sup>f</sup></td><td align="right">ND</td><td align="right">-1.7</td><td align="center">24, 48<sup>f</sup></td><td align="right">25.19</td></tr><tr><td align="left">NM_007935</td><td align="left">enhancer of polycomb homolog 1</td><td align="center">Epc1</td><td align="center">13831</td><td align="center">1.6</td><td align="center">168<sup>f</sup></td><td align="right">1.16</td><td align="right">-2.7</td><td align="center">12, 24<sup>f</sup></td><td align="right">75.21</td></tr><tr><td align="left">BC002008</td><td align="left">fatty acid binding protein 5, epidermal</td><td align="center">Fabp5</td><td align="center">16592</td><td align="center">3.9</td><td align="center">8, 12<sup>f</sup></td><td align="right">2.43</td><td align="right">-1.9</td><td align="center">8, 12<sup>f</sup>, 24</td><td align="right">54.14</td></tr><tr><td align="left">NM_026320</td><td align="left">growth arrest and DNA-damage- inducible, gamma interacting protein 1</td><td align="center">Gadd45gip1</td><td align="center">102060</td><td align="center">1.8</td><td align="center">168<sup>f</sup></td><td align="right">4.67</td><td align="right">-1.5</td><td align="center">8<sup>f</sup></td><td align="right">40.49</td></tr><tr><td align="left">W11419</td><td align="left">inhibitor of DNA binding 3</td><td align="center">Id3</td><td align="center">15903</td><td align="center">1.8</td><td align="center">168<sup>f</sup></td><td align="right">0.34</td><td align="right">-1.5</td><td align="center">24, 48<sup>f</sup></td><td align="right">88.83</td></tr><tr><td align="left">AA009268</td><td align="left">myelocytomatosis oncogene</td><td align="center">Myc</td><td align="center">17869</td><td align="center">3.7</td><td align="center">4, 12<sup>f</sup>, 168</td><td align="right">5.59</td><td align="right">-2.2</td><td align="center">2<sup>f</sup></td><td align="right">148.40</td></tr><tr><td align="left">NM_011033</td><td align="left">poly A binding protein, cytoplasmic 2</td><td align="center">Pabpc2</td><td align="center">18459</td><td align="center">7.0</td><td align="center">2<sup>f</sup></td><td align="right">ND</td><td align="right">-1.6</td><td align="center">12<sup>f</sup></td><td align="right">ND</td></tr><tr><td></td><td align="left">REST corepressor 1</td><td align="center">Rcor1</td><td align="center">217864</td><td align="center">1.9</td><td align="center">4, 8, 18, 72, 168</td><td align="right">3.70</td><td align="right">-1.6</td><td align="center">24<sup>f</sup></td><td align="right">116.50</td></tr><tr><td align="left">BE980584</td><td align="left">secretory granule neuroendocrine protein 1, 7B2 protein</td><td align="center">Sgne1</td><td align="center">20394</td><td align="center">3.3</td><td align="center">168<sup>f</sup></td><td align="right">0.74</td><td align="right">-1.5</td><td align="center">48<sup>f</sup></td><td align="right">175.00</td></tr><tr><td align="left">AA462951</td><td align="left">transcription factor 4</td><td align="center">Tcf4</td><td align="center">21413</td><td align="center">1.6</td><td align="center">12<sup>f</sup>, 168</td><td align="right">5.77</td><td align="right">-1.5</td><td align="center">24<sup>f</sup></td><td align="right">74.44</td></tr><tr><td align="left">AA003942</td><td align="left">tenascin C</td><td align="center">Tnc</td><td align="center">21923</td><td align="center">1.6</td><td align="center">168<sup>f</sup></td><td align="right">0.37</td><td align="right">-1.8</td><td align="center">24<sup>f</sup>, 48</td><td align="right">59.34</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left" colspan="10">IV) Repressed <italic>in vivo </italic>and induced <italic>in vitro</italic><sup>e</sup></td></tr><tr><td align="left">W36712</td><td align="left">B-cell translocation gene 2, anti- proliferative</td><td align="center">Btg2</td><td align="center">12227</td><td align="center">-1.8</td><td align="center">18<sup>f</sup>, 24</td><td align="right">ND</td><td align="right">1.5</td><td align="center">4<sup>f</sup></td><td align="right">ND</td></tr><tr><td align="left">AA174215</td><td align="left">cathepsin L</td><td align="center">Ctsl</td><td align="center">13039</td><td align="center">-1.6</td><td align="center">24<sup>f</sup>, 72, 168</td><td align="right">ND</td><td align="right">1.6</td><td align="center">8, 48<sup>f</sup></td><td align="right">ND</td></tr><tr><td align="left">AA419858</td><td align="left">cysteine rich protein 61</td><td align="center">Cyr61</td><td align="center">16007</td><td align="center">-1.6</td><td align="center">2<sup>f</sup></td><td align="right">0.07</td><td align="right">1.6</td><td align="center">8, 48<sup>f</sup></td><td align="right">ND</td></tr><tr><td align="left">AW488956</td><td align="left">polo-like kinase 3</td><td align="center">Plk3</td><td align="center">12795</td><td align="center">-1.6</td><td align="center">4<sup>f</sup></td><td align="right">ND</td><td align="right">1.6</td><td align="center">4, 48<sup>f</sup></td><td align="right">ND</td></tr><tr><td align="left">BG068288</td><td align="left">solute carrier organic anion transporter family, member 1b2</td><td align="center">Slco1b2</td><td align="center">28253</td><td align="center">-1.7</td><td align="center">8<sup>f</sup>, 12, 18, 24, 72, 168</td><td align="right">ND</td><td align="right">1.6</td><td align="center">4<sup>f</sup></td><td align="right">1.18</td></tr><tr><td align="left">NM_011470</td><td align="left">small proline-rich protein 2D</td><td align="center">Sprr2d</td><td align="center">20758</td><td align="center">-1.6</td><td align="center">18<sup>f</sup>, 72</td><td align="right">1.97</td><td align="right">1.6</td><td align="center">4<sup>f</sup></td><td align="right">ND</td></tr></tbody></table><table-wrap-foot><p><sup>a</sup>Maximum absolute fold change determined by microarray analysis</p><p><sup>b</sup>Time point where genes are differentially regulated with P1(t) > 0.9999 and \|fold change\| > 1.5</p><p><sup>c</sup>EC50 valued determined from microarray results</p><p><sup>d</sup>ND = not determined from microarray results</p><p><sup>e</sup>Classification groups as defined in Figure 4B</p><p><sup>f</sup>Time point representing the maximum \|fold change\|</p></table-wrap-foot></table-wrap><p>Forty-two of the 67 common differentially expressed genes were dose responsive at 12 and 24 hrs <italic>in vitro </italic>and <italic>in vivo</italic>, respectively, further suggesting the role of the AhR in mediating these responses. Microarray-based EC<sub>50 </sub>values spanned at least 3 orders of magnitude ranging from 0.05 μg/kg to >150 μg/kg <italic>in vivo</italic>, and 0.00118 nM to 2.4 nM <italic>in vitro </italic>(Table <xref ref-type="table" rid="T1">1</xref>). Cyp1a1, the prototypical marker of TCDD exposure, had EC<sub>50 </sub>values of 0.05 μg/kg and 0.014 nM, <italic>in vivo </italic>and <italic>in vitro </italic>respectively, and was induced 38-fold in both time course studies. Complete data sets for the <italic>in vivo </italic>time course and dose-responses experiments are available in Additional file <xref ref-type="supplementary-material" rid="S3">3</xref> and <xref ref-type="supplementary-material" rid="S4">4</xref>.</p><p>Of the 67 overlapping genes, 18 exhibited divergent temporal profiles (classes III and IV). Class III contains 12 genes induced <italic>in vivo </italic>but repressed <italic>in vitro</italic>, while 6 were repressed <italic>in vivo </italic>and induced <italic>in vitro </italic>(class IV). Example genes include Myc (class III) and B-cell translocation gene 2 (Btg2, class IV) which are both involved in regulating cell cycle progression [<xref ref-type="bibr" rid="B23">23</xref>,<xref ref-type="bibr" rid="B26">26</xref>-<xref ref-type="bibr" rid="B30">30</xref>]. Myc was induced 3.7-fold <italic>in vivo </italic>and repressed 2.2-fold <italic>in vitro</italic>, while Btg2 was repressed 1.8-fold <italic>in vivo </italic>and induced 1.5-fold <italic>in vitro</italic>.</p><p>In addition to the regulated genes common to both models, 218 <italic>in vitro</italic>- and 559 <italic>in </italic><italic>vivo</italic>-specific genes were identified. Many of the unique <italic>in vitro </italic>responses are involved in cell cycle regulation, including cyclins D1 and B2 (Table <xref ref-type="table" rid="T2">2</xref>). Cyclin D1, which complexes with cyclin-dependent kinase 4 (Cdk4) to regulate the progression from G<sub>1 </sub>to S phase [<xref ref-type="bibr" rid="B31">31</xref>,<xref ref-type="bibr" rid="B32">32</xref>], was down-regulated early and repressed 1.7-fold to 48 hrs. Furthermore, cyclin B2 and cell division cycle 2 homolog A (Cdc2a) which interact to form an active kinase required for G<sub>2 </sub>promotion, were down-regulated, 1.8-fold and 1.5-fold, respectively. In addition to cell cycle related genes, UDP glucuronosyltransferase 1 family, polypeptide A2 (Ugt1a2), a phase II metabolizing enzyme, was induced 2.8-fold <italic>in vitro</italic>, but not significantly regulated <italic>in vivo</italic>.</p><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Examples of TCDD-elicited gene expression responses unique to Hepa1c1c7 cells</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left">Accession</td><td align="left">Gene name</td><td align="center">Gene Symbol</td><td align="center">Entrez Gene ID</td><td align="center">Fold change<sup>a</sup></td><td align="center">Time points<sup>b </sup>(hrs)</td></tr></thead><tbody><tr><td align="left">AA111722</td><td align="left">cyclin D1</td><td align="center">Ccnd1</td><td align="center">12443</td><td align="center">-1.7</td><td align="center">4, 8, 12, 24<sup>c</sup>, 48</td></tr><tr><td align="left">AA914666</td><td align="left">cyclin-dependent kinase inhibitor 2B (p15, inhibits CDK4)</td><td align="center">Cdkn2b</td><td align="center">12579</td><td align="center">2.4</td><td align="center">4<sup>c</sup>, 8, 48</td></tr><tr><td align="left">BC008247</td><td align="left">cyclin B2</td><td align="center">Ccnb2</td><td align="center">12442</td><td align="center">-1.8</td><td align="center">24<sup>c</sup></td></tr><tr><td align="left">BG064846</td><td align="left">cell division cycle 2 homolog A (S. pombe)</td><td align="center">Cdc2a</td><td align="center">12534</td><td align="center">-1.5</td><td align="center">12, 24<sup>c</sup></td></tr><tr><td align="left">AA011839</td><td align="left">minichromosome maintenance deficient 2 mitotin (S. cerevisiae)</td><td align="center">Mcm2</td><td align="center">17216</td><td align="center">-1.8</td><td align="center">8<sup>c</sup>, 12</td></tr><tr><td align="left">BG074721</td><td align="left">minichromosome maintenance deficient 7 (S. cerevisiae)</td><td align="center">Mcm7</td><td align="center">17220</td><td align="center">-1.7</td><td align="center">8, 12<sup>c</sup>, 24</td></tr><tr><td align="left">AA003042</td><td align="left">myeloblastosis oncogene-like 2</td><td align="center">Mybl2</td><td align="center">17865</td><td align="center">-2.2</td><td align="center">8<sup>c</sup>, 12, 24</td></tr><tr><td align="left">L27122</td><td align="left">UDP glucuronosyltransferase 1 family, polypeptide A2</td><td align="center">Ugt1a2</td><td align="center">22236</td><td align="center">2.8</td><td align="center">4, 8, 12<sup>c</sup>, 24, 48</td></tr></tbody></table><table-wrap-foot><p><sup>a</sup>Maximum absolute fold change determined by microarray analysis</p><p><sup>b</sup>Differentially regulated genes with P1(t) > 0.9999 and \|fold change\| > 1.5</p><p><sup>c</sup>Time point representing the maximum \|fold change\|</p></table-wrap-foot></table-wrap><p>Analysis of the C57BL/6 hepatic time course identified 552 unique genes that were solely regulated <italic>in vivo</italic>. This included TCDD induced transcripts for microsomal epoxide hydrolase 1 (Ephx1) and carbonyl reductase 3 (Cbr3) which both function as xenobiotic metabolizing enzymes. Notch gene homolog 1 (Notch1) and growth arrest specific 1 (Gas1) which are both associated with development and differentiation but serve undetermined roles in the liver, were also induced by TCDD (Table <xref ref-type="table" rid="T3">3</xref>). Genes related to immune cell accumulation were also specific to the <italic>in vivo </italic>study, coincident with immune cell accumulation at 168 hr as determined by histopathological examination [<xref ref-type="bibr" rid="B19">19</xref>].</p><table-wrap position="float" id="T3"><label>Table 3</label><caption><p>Examples of TCDD-elicited gene expression responses unique to C57BL/6 hepatic tissue</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left">Accession</td><td align="left">Gene name</td><td align="center">Gene Symbol</td><td align="center">Entrez Gene ID</td><td align="center">Fold change<sup>a</sup></td><td align="center">Time points<sup>b </sup>(hrs)</td></tr></thead><tbody><tr><td align="left">AA170585</td><td align="left">carbonic anhydrase 3</td><td align="center">Car3</td><td align="center">12350</td><td align="center">-3.5</td><td align="center">12<sup>c</sup>, 18, 24, 168</td></tr><tr><td align="left">AK003232</td><td align="left">carbonyl reductase 3</td><td align="center">Cbr3</td><td align="center">109857</td><td align="center">2.2</td><td align="center">12, 18<sup>c</sup></td></tr><tr><td align="left">AA571998</td><td align="left">CD3 antigen, delta polypeptide</td><td align="center">Cd3d</td><td align="center">12500</td><td align="center">-2.4</td><td align="center">12, 18<sup>c</sup>, 24, 72, 168</td></tr><tr><td align="left">BG072496</td><td align="left">ELOVL family member 5, elongation of long chain fatty acids</td><td align="center">Elovl5</td><td align="center">68801</td><td align="center">2.0</td><td align="center">8, 12<sup>c</sup>, 18, 24, 72, 168</td></tr><tr><td align="left">BG072453</td><td align="left">epoxide hydrolase 1, microsomal</td><td align="center">Ephx1</td><td align="center">13222</td><td align="center">1.9</td><td align="center">8, 12 18, 24<sup>c</sup></td></tr><tr><td align="left">W84211</td><td align="left">growth arrest specific 1</td><td align="center">Gas1</td><td align="center">14451</td><td align="center">-1.9</td><td align="center">4, 8, 18, 24, 72, 168<sup>c</sup></td></tr><tr><td align="left">W41175</td><td align="left">glycerol phosphate dehydrogenase 2, mitochondrial</td><td align="center">Gpd2</td><td align="center">14571</td><td align="center">-2.3</td><td align="center">8, 12, 18, 24, 72<sup>c</sup>, 168</td></tr><tr><td align="left">W29265</td><td align="left">glutathione S-transferase, alpha 2 (Yc2)</td><td align="center">Gsta2</td><td align="center">14858</td><td align="center">7.2</td><td align="center">12, 18, 24, 72<sup>c</sup>, 168</td></tr><tr><td align="left">AA145865</td><td align="left">lymphocyte antigen 6 complex, locus A</td><td align="center">Ly6a</td><td align="center">110454</td><td align="center">2.5</td><td align="center">72, 168<sup>c</sup></td></tr><tr><td align="left">W98998</td><td align="left">Notch gene homolog 1 (Drosophila)</td><td align="center">Notch1</td><td align="center">18128</td><td align="center">3.3</td><td align="center">2, 4<sup>c</sup>, 8, 12, 18, 24, 72, 168</td></tr></tbody></table><table-wrap-foot><p><sup>a</sup>Maximum absolute fold change determined by microarray analysis</p><p><sup>b</sup>Differentially regulated genes with P1(t) > 0.9999 and \|fold change\| > 1.5</p><p><sup>c</sup>Time point representing the maximum \|fold change\|</p></table-wrap-foot></table-wrap></sec><sec><title>Comparison of basal gene expression levels in Hepa1c1c7 cells and hepatic tissue</title><p>In order to further investigate differences in gene expression levels, Hepa1c1c7 cells and C57BL/6 liver samples were directly compared by competitive hybridization on the same array, to identify basal gene expression level differences. Subsequent linear regression analysis of the mean normalized signal intensities from the untreated samples resulted in a correlation value of R = 0.75 (Figure <xref ref-type="fig" rid="F5">5</xref>), which is consistent with basal gene expression comparisons of various <italic>in vitro </italic>rat hepatic systems against whole livers, where correlation values decreased between liver slices (R = 0.97), primary cells (R = 0.85), BRL3A (R = 0.3) and NRL clone 9 (R = 0.32) rat liver cell lines [<xref ref-type="bibr" rid="B10">10</xref>]. Overall, the correlation illustrates reasonable concordance in basal gene expression levels between the two models. However, data points which deviate from the fitted line indicate differences in the basal expression of individual genes between the Hepa1c1c7 cells and hepatic tissue from C57BL/6 mice. Although there are differences, they may be negligible if the TCDD-elicited responses are conserved <italic>in vitro </italic>and <italic>in vivo</italic>. Complete microarray data for the untreated comparisons are available in <xref ref-type="supplementary-material" rid="S5">Additional file 5</xref>.</p><fig position="float" id="F5"><label>Figure 5</label><caption><p>Comparison of Hepalclc7 cell and C57BL/6 hepatic tissue basal gene expression. Untreated samples from Hepalclc7 cells and hepatic tissue from immature ovariectomized C57BL/6 mice taken at 0 hrs were competitively hybridized to the 13,362 feature cDNA microarray. Log2 normalized signal intensities were plotted for <italic>in vitro </italic>versus <italic>in vivo </italic>data to generate the correlation coefficient. The linear correlation coefficient R was 0.75 between <italic>in vitro </italic>and <italic>in vivo </italic>models</p></caption><graphic xlink:href="1471-2164-7-80-5"/></fig><p>The relative basal expression of the 67 common active features was further investigated (Figure <xref ref-type="fig" rid="F5">5</xref>). In general, class I (i.e. induced in both models) genes fell close to the regression line, indicating that the basal expression of induced genes were comparable as were their <italic>in vitro </italic>and <italic>in vivo </italic>responses to TCDD. In contrast, basal expression levels of class III genes (i.e. induced <italic>in vivo </italic>while repressed <italic>in vitro</italic>) were generally higher in the Hepa1c1c7 cells, while levels in class II and IV (i.e. repressed in both models and repressed <italic>in vivo </italic>while induced <italic>in vitro</italic>, respectively) genes were scattered around the fitted linear line in Figure <xref ref-type="fig" rid="F5">5</xref>.</p></sec><sec><title>Quantitative real-time PCR verification of microarray responses</title><p>In total, 14 <italic>in vitro </italic>and 24 <italic>in vivo </italic>responsive genes representing common and model-specific genes were verified by quantitative real-time PCR (QRTPCR) (see <xref ref-type="supplementary-material" rid="S6">Additional file 6</xref>). Of the selected genes regulated in both models, all displayed temporal patterns comparable to the microarray data (Figure <xref ref-type="fig" rid="F6">6</xref>). For example, Xdh, Myc and fatty acid binding protein (Fabp5) exhibited good agreement in fold change and temporal expression pattern when comparing microarray and QRTPCR data. However, significant data compression was evident when comparing <italic>in vitro </italic>and <italic>in vivo </italic>Cyp1a1 induction by QRTPCR, although <italic>in vitro </italic>and <italic>in vivo </italic>microarray induction levels were comparable. Previous studies suggest this is likely due to the limited fluorescence intensity range (0 – 65,535) of microarrays resulting in signal saturation and compression of the true magnitude of induction of transcript levels [<xref ref-type="bibr" rid="B33">33</xref>,<xref ref-type="bibr" rid="B34">34</xref>]. Cross hybridization of homologous probes to a given target sequence on the microarray may also be a contributing factor, especially in comparison to other, more gene-specific measurement techniques [<xref ref-type="bibr" rid="B35">35</xref>].</p><fig position="float" id="F6"><label>Figure 6</label><caption><p>Quantitative real-time PCR verification of <italic>in vitro </italic>and <italic>in vivo </italic>microarray results. The same RNA used for cDNA microarray analysis was examined by QRTPCR. All fold changes were calculated relative to time-matched vehicle controls. Bars (left axis) and line (right axis) represent data obtained by QRTPCR and cDNA microarrays, respectively. Genes are indicated by official gene symbols, and results are the average of four biological replicates. Classes refer to the respective classification categories as illustrated in Figure 4B. Error bars represent the standard error of measurement for the average fold change. p < 0.05 for QRTPCR</p></caption><graphic xlink:href="1471-2164-7-80-6"/></fig></sec></sec><sec><title>Discussion</title><p>Microarrays have become an invaluable tool in toxicogenomics for comprehensively characterizing gene expression responses following treatment with an environmental contaminant, commercial chemical, natural product or drug as well as for investigating complex mixtures relevant to human and wildlife exposures. An emerging consensus suggests that toxicogenomics will accelerate drug development and significantly improve quantitative risk assessments [<xref ref-type="bibr" rid="B36">36</xref>,<xref ref-type="bibr" rid="B37">37</xref>]. In addition, toxicogenomics supports the development and refinement of predictive <italic>in vitro </italic>high-throughput toxicity screening assays that can be used as alternatives to traditional <italic>in vivo </italic>testing. Ideally, <italic>in vitro </italic>high-throughput toxicity screens can be used to rank and prioritize drug candidates, environmental contaminants, and commercial chemicals, which warrant further development or testing. Although <italic>in vitro </italic>responses are assumed to reflect a subset of comparable <italic>in vivo </italic>responses, few studies have completed a comprehensive and systematic comparison. This study closely examined two well-established models, and comprehensively compared the TCDD-elicited gene expression to assess the predictive value of <italic>in vitro </italic>systems.</p><p>Comparative analysis of Hepa1c1c7 cell and hepatic C57BL/6 microarray data identified 67 differentially expressed genes co-regulated by TCDD. Four classes based on their temporal expression patterns were identified (Figure <xref ref-type="fig" rid="F4">4B</xref> and <xref ref-type="fig" rid="F4">4C</xref>), with 42 of the 67 common regulated genes exhibiting dose-response characteristics in both models. <italic>In vitro </italic>EC<sub>50 </sub>values ranged from 0.001182 nM to 2.4 nM, while <italic>in vivo </italic>the values ranged from 0.05 μg/kg to >150 μg/kg. The wide range of EC<sub>50 </sub>values illustrate the varying sensitivity of regulated genes to TCDD in both models.</p><p>Hepa1c1c7 cells and hepatic tissue from C57BL/6 mice are the prototypical models used to investigate the mechanisms of action of TCDD and other related compounds and both exhibited the classic induction of phase I and II metabolizing enzymes including Cyp1a1 and Nqo1 [<xref ref-type="bibr" rid="B38">38</xref>,<xref ref-type="bibr" rid="B39">39</xref>]. Gsta4 and Xdh were also up-regulated in both models further demonstrating Hepa1c1c7 cells as a suitable model for investigating TCDD-regulated induction of xenobiotic metabolizing genes. In addition to these genes, the responses of Nqo1, Ugdh and Tnfaip2 were also conserved across models and were categorized as class I genes (similarly induced in both models; Figure <xref ref-type="fig" rid="F4">4B</xref> and <xref ref-type="fig" rid="F4">4C</xref>). However, Gsta2 was induced <italic>in vivo </italic>while no significant effect was detected in Hepa1c1c7 cells, and Ugt1a2 was induced <italic>in vitro </italic>but not differentially expressed in C57BL/6 hepatic tissue. Although many phase I and II metabolizing enzyme responses were conserved, differences exist that may limit Hepa1c1c7 cells from accurately modeling the full spectrum of <italic>in vivo </italic>hepatic responses elicited by TCDD.</p><p>A direct comparison of untreated Hepa1c1c7 cells and C57BL/6 hepatic tissue was performed to further investigate innate differences between the two models. Comparison of the normalized signal intensities revealed a good correlation (R = 0.75) between <italic>in vitro </italic>and <italic>in vivo </italic>basal expression levels (Figure <xref ref-type="fig" rid="F5">5</xref>). This illustrates that many genes are basally expressed to similar levels in both models as illustrated by the cluster of class I (similarly induced genes) closely surrounding the fitted line. Although a correlation exists, there are still differences in basal expression which may be associated with the origins of the models (i.e. normal hepatic tissue versus hepatoma derived Hepa1c1c7 cells), as well as the inability of <italic>in vitro </italic>systems to effectively model complex interactions between different cell types (e.g. Kupffer and stellate cells). For example, Myc, a G<sub>1 </sub>to S phase cell cycle regulator [<xref ref-type="bibr" rid="B23">23</xref>,<xref ref-type="bibr" rid="B26">26</xref>-<xref ref-type="bibr" rid="B29">29</xref>], was repressed <italic>in vitro </italic>while being induced <italic>in vivo </italic>and the model-specific responses may be related to difference in basal expression levels between the two models (Table <xref ref-type="table" rid="T1">1</xref>). The levels of Myc transcripts in untreated Hepa1c1c7 cells were higher relative to untreated C57BL/6 hepatic tissue, consistent with the proliferative state of the <italic>in vitro </italic>system (data not shown). Examination of other class III genes suggests that they are more highly expressed <italic>in vitro </italic>when compared to <italic>in vivo </italic>(Figure <xref ref-type="fig" rid="F5">5</xref>). Consequently, differences in basal expression may be a factor contributing to divergent <italic>in vitro – in vivo </italic>responses. Another possible source for the model-specific responses may be related to DNA methylation status of the promoter region of TCDD-responsive genes in either model. DNA methylation results in gene silencing [<xref ref-type="bibr" rid="B40">40</xref>,<xref ref-type="bibr" rid="B41">41</xref>] and a previous study with Hepa1c1c7 has shown that TCDD-elicited gene expression responses are influenced by DNA methylation status [<xref ref-type="bibr" rid="B42">42</xref>]. The differing methylation states between the <italic>in vitro </italic>and <italic>in vivo </italic>systems may further contribute to the model-specific gene expression responses.</p><p>Many <italic>in vitro </italic>specific gene expression responses elicited by TCDD were associated with cell cycle progression and cell cycle arrest. Myc and its downstream target, cyclin D1, which forms a kinase complex with Cdk4 [<xref ref-type="bibr" rid="B43">43</xref>,<xref ref-type="bibr" rid="B44">44</xref>] were both repressed by TCDD. In contrast, Cdkn1a, an inhibitor of cyclin-dependent kinase 2 (Cdk2)-cyclin E complex kinase activity [<xref ref-type="bibr" rid="B43">43</xref>], was induced. Inactivation of the Cdk2-cyclin E complex prevents the phosphorylation of pRb resulting in cell cycle arrest during G<sub>1</sub>. Additionally, the <italic>in vitro </italic>induction of Btg2 suggests an alternative mechanism for cell cycle arrest during the G<sub>2 </sub>phase. Constitutively active BTG2 in human leukemia U937 cells, induces G<sub>2</sub>/M cell cycle arrest by inhibiting the formation of the cyclin B1 and Cdc2 complex, thereby inhibiting the active kinase function of the complex [<xref ref-type="bibr" rid="B30">30</xref>]. Collectively, these results corroborate and extend previous <italic>in vitro </italic>TCDD-mediated cell cycle arrest studies [<xref ref-type="bibr" rid="B45">45</xref>-<xref ref-type="bibr" rid="B48">48</xref>].</p><p>TCDD treatment resulted in a number of divergent gene responses across both models as represented by classes III and IV (Figures <xref ref-type="fig" rid="F4">4B</xref> and <xref ref-type="fig" rid="F4">4C</xref>). Genes related to immune cell accumulation, including major histocompatibility complex (MHC) molecules were only observed <italic>in vivo</italic>, and are likely a response to hepatic damage mediated by ROS or fatty accumulation and therefore independent of direct AhR action [<xref ref-type="bibr" rid="B19">19</xref>]. This is characteristic of the complex interaction between different cell types responding to liver injury that cannot be modeled in homogenous cultures of cells.</p><p>Pharmacokinetics may also contribute to response differences between the two models. Hepa1c1c7 cells were directly treated, whereas <italic>in vivo</italic>, TCDD must first be delivered to the liver and targeted cells prior to eliciting its effects. Additionally, C57BL/6 studies were able to be carried out to 168 hrs following TCDD treatment, while <italic>in vitro </italic>studies were limited to 48 hrs to minimize potentially confounding effects due to cell confluency. However, early responses associated with classes I and II (induced or repressed in both models; Figure <xref ref-type="fig" rid="F4">4B</xref> and <xref ref-type="fig" rid="F4">4C</xref>) are well conserved and exhibit comparable levels of induction or repression in both models. Hierarchical clustering of the common active genes (Figure <xref ref-type="fig" rid="F4">4C</xref>) illustrates gene induction occurs early while gene repression occurs later in both models. Clustering across both time and model revealed that gene expression profiles at 1 hr <italic>in vitro </italic>and 2 hr <italic>in vivo </italic>were most similar. This clustering pattern implies that early <italic>in vitro </italic>responses may accurately model early <italic>in vivo </italic>gene expression effects.</p></sec><sec><title>Conclusion</title><p>Comparative analysis of global gene expression from Hepa1c1c7 cells and hepatic tissue from C57BL/6 mice identified several model-specific responses to TCDD that should be considered when extrapolating <italic>in vitro </italic>results to potential <italic>in vivo </italic>effects. Despite these differences, immortalized cells as well as other emerging <italic>in vitro </italic>systems (e.g., primary cells, stem cells and 3-D culture systems) provide valuable mechanistic information that supports the further development of high-throughput toxicity screening assays. However, the relevance of <italic>in vitro </italic>responses requires complementary <italic>in vivo </italic>verification. Furthermore, comparative studies exploiting other <italic>in vitro </italic>and <italic>in vivo </italic>systems, different structurally diverse ligands and other relevant model species will not only corroborate the relevance of the mechanisms, but will also support more appropriate extrapolations between rodent studies and potential effects in humans and ecologically-relevant species.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Culture and treatment of cell lines</title><p>Hepa1c1c7 wild-type and c4 ARNT-deficient cell lines (gifts from O. Hankinson, University of California, Los Angeles, CA) were maintained in phenol-red free DMEM/F12 media (Invitrogen, Carlsbad, CA) supplemented with 5% fetal bovine serum (FBS) (Hyclone, Logan, UT), 2.5 μg/mL amphotericin B (Invitrogen), 2.5 μg/mL amphotericin B (Invitrogen), 50 μg/mL gentamycin (Invitrogen), 100 U/mL penicillin and 100 μg/mL streptomycin (Invitrogen). 1 × 10<sup>6</sup> cells were seeded into T175 culture flasks (Sarstedt, Newton, NC) and incubated under standard conditions (5% CO<sub>2</sub>, 37°C). Time course studies were performed with wild-type and c4 mutant cells where both were dosed with either 10 nM TCDD (provided by S. Safe, Texas A&M University, College Station, TX) or DMSO (Sigma, St. Louis, MO) vehicle and harvested at 1, 2, 4, 8, 12, 24 or 48 hrs. Additional untreated control cells were harvested at the time of dosing (i.e. 0 hrs). For the dose-response study, wild-type cells were treated with DMSO vehicle or 0.001, 0.01, 0.1, 1.0, 10 or 100 nM TCDD and harvested at 12 hrs. The treatment and harvesting regimen for cell culture studies are illustrated in <xref ref-type="supplementary-material" rid="S7">Additional file 7</xref>.</p></sec><sec><title>Animal treatment</title><p>The handling and treatment of female C57BL/6 mice has been previously described [<xref ref-type="bibr" rid="B19">19</xref>]. Briefly, immature ovariectomized mice were orally gavaged with 30 μg/kg TCDD for the time course study and sacrificed at 2, 4, 8, 12, 18, 24 72 or 168 hrs after treatment. For the dose-response study, mice were treated with 0.001, 0.01, 0.1, 1, 10, 100 or 300 μg/kg TCDD and sacrificed 24 hrs after dosing. Animals were sacrificed by cervical dislocation and tissue samples were removed, weighed, flash frozen in liquid nitrogen and stored at -80°C until further use.</p></sec><sec><title>RNA isolation</title><p>Cells were harvested by scraping in 2.0 mL of Trizol Reagent (Invitrogen). Frozen liver samples (approximately 70 mg) were transferred to 1.0 mL of Trizol Reagent and homogenized in a Mixer Mill 300 tissue homogenizer (Retsch, Germany). Total RNA from each study was isolated according to the manufacturer's protocol with an additional acid phenol:chloroform extraction. Isolated RNA was resuspended in The RNA Storage Solution (Ambion Inc., Austin, TX), quantified (A<sub>260</sub>), and assessed for purity by determining the A<sub>260</sub>/A<sub>280 </sub>ratio and by visual inspection of 1.0 μg on a denaturing gel.</p></sec><sec><title>Microarray experimental design</title><p>Changes in gene expression were assessed using customized cDNA microarrays containing 13,362 features representing 8,284 unique genes. For the time course study, TCDD-treated samples were compared to time-matched vehicle controls using an independent reference design [<xref ref-type="bibr" rid="B49">49</xref>]. In this design, treated Hepa1c1c7 cell or hepatic tissue samples were compared to the corresponding time-matched vehicle control with two independent labelings (dye swaps; <xref ref-type="supplementary-material" rid="S8">Additional file 8</xref>). Four replicates of this design were performed, each using independent cell culture samples or different animals. Dose-response changes in gene expression were analyzed using a common reference design in which samples from TCDD-treated cells or mice were co-hybridized with a common vehicle reference (i.e. independent DMSO treated Hepa1c1c7 cell samples, hepatic samples from independent sesame oil treated C57BL/6 mice) using two independent labelings (<xref ref-type="supplementary-material" rid="S8">Additional file 8</xref>). Four replicates with two independent labelings were performed for both <italic>in vitro </italic>and <italic>in vivo </italic>samples. Co-hybridizations of untreated Hepa1c1c7 cells and hepatic tissue from C57BL/6 mice were performed to investigate differences in basal gene expression levels between models (<xref ref-type="supplementary-material" rid="S8">Additional file 8</xref>). Four replicates were performed with two independent labelings per sample (dye swap).</p><p>More detailed protocols regarding the microarray assay, including microarray preparation, labeling of the cDNA probe, sample hybridization and washing can be obtained from the dbZach website [<xref ref-type="bibr" rid="B50">50</xref>]. Briefly, polymerase chain reaction (PCR) amplified cDNAs were robotically arrayed onto epoxy-coated glass slides (Schott-Nexterion, Duryea, PA) using an Omnigrid arrayer (GeneMachines, San Carlos, CA) equipped with 48 (4 × 12) Chipmaker 2 pins (Telechem) at Michigan State University's Research Technology Support Facility [<xref ref-type="bibr" rid="B51">51</xref>]. Total RNA (30 μg) was reverse transcribed in the presence of Cy3- or Cy5-deoxyuridine triphosphate (dUTP) to create fluorescence-labeled cDNA, which was purified using a Qiagen PCR kit (Qiagen, Valencia, CA). Cy3 and Cy5 samples were mixed, vacuum dried and resuspended in 48 μL of hybridization buffer (40% formamide, 4× SSC, 1% sodium dodecyl sulfate [SDS]) with 20 μg polydA and 20 μg of mouse COT-1 DNA (Invitrogen) as competitor. This probe mixture was heated at 95°C for 3 min and hybridized on the array under a 22 × 60 mm LifterSlip (Erie Scientific Company, Portsmouth, NH) in a light-protected and humidified hybridization chamber (Corning Inc., Corning, NY) for 18–24 hrs in a 42°C water bath. Slides were then washed, dried by centrifugation and scanned at 635 nm (Cy5) and 532 nm (Cy3) on an Affymetrix 428 Array Scanner (Santa Clara, CA). Images were analyzed for feature and background intensities using GenePix Pro 5.0 (Molecular Devices, Union City, CA).</p></sec><sec><title>Microarray data quality assurance, normalization and analysis</title><p>Microarray data were first passed through a quality assurance protocol prior to further analysis to ensure consistently high quality data throughout the dose-response and time course studies prior to normalization and further analysis [<xref ref-type="bibr" rid="B52">52</xref>]. All the collected data were then normalized using a semi-parametric approach [<xref ref-type="bibr" rid="B53">53</xref>]. Empirical Bayes analysis was used to calculate posterior probabilities (P1(<italic>t</italic>) value) of activity on a per gene and time point or dose group basis using the model-based <italic>t</italic>-value [<xref ref-type="bibr" rid="B54">54</xref>]. The data were filtered using a P1(<italic>t</italic>) cutoff of 0.9999 and ± 1.5 fold change to identify the most robust changes in gene expression and to obtain an initial subset of differentially regulated genes for further investigation and data interpretation. Subsequent analysis included agglomerative hierarchical and <italic>k</italic>-means clustering using the standard correlation distance metric implemented in GeneSpring 6.0 (Silicon Genetics, Redwood City, CA). Functional categorization of differentially regulated genes were mined and statistically analyzed from Gene Ontology [<xref ref-type="bibr" rid="B55">55</xref>] using GOMiner [<xref ref-type="bibr" rid="B56">56</xref>].</p></sec><sec><title>Quantitative real-time PCR analysis</title><p>For each sample, 1.0 μg of total RNA was reverse transcribed by Superscript II using an anchored oligo-dT primer as described by the manufacturer (Invitrogen). The cDNA (1.0 μL) was used as a template in a 30 μL PCR reaction containing 0.1 μM of forward and reverse gene-specific primers designed using Primer 3 [<xref ref-type="bibr" rid="B57">57</xref>], 3 mM MgCl<sub>2</sub>, 1.0 mM dNTPs, 0.025 IU AmpliTaq Gold, and 1× SYBR Green PCR buffer (Applied Biosystems, Foster City, CA). PCR amplification was conducted in MicroAmp Optical 96-well reaction plates (Applied Biosystems) on an Applied Biosystems PRISM 7000 Sequence Detection System under the following conditions: initial denaturation and enzyme activation for 10 min at 95°C, followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. A dissociation protocol was performed to assess the specificity of the primers and the uniformity of the PCR-generated products. Each plate contained duplicate standards of purified PCR products of known template concentration covering 7 orders of magnitude to interpolate relative template concentrations of the samples from the standard curves of log copy number versus threshold cycle (Ct). No template controls (NTC) were also included on each plate. Samples with a Ct value within 2 standard deviations of the mean Ct values for the NTCs were considered below the limits of detection. The copy number of each unknown sample for each gene was standardized to the geometric mean of three house-keeping genes (β-actin, Gapd and Hprt) to control for differences in RNA loading, quality, and cDNA synthesis. For graphing purposes, the relative expression levels were scaled such that the expression level of the time-matched control group was equal to 1. Statistical analysis was performed with SAS 8.02 (SAS Institute, Cary, NC). Data were analyzed by analysis of variance (ANOVA) followed by Tukey's <italic>post hoc </italic>test. Differences between treatment groups were considered significant when <italic>p </italic>< 0.05. Official gene names and symbols, RefSeq and Entrez Gene IDs, forward and reverse primer sequences, and amplicon sizes are listed in Table <xref ref-type="table" rid="T4">4</xref>.</p><table-wrap position="float" id="T4"><label>Table 4</label><caption><p>Gene names and primer sequences for QRTPCR</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left">RefSeq</td><td align="left">Gene name</td><td align="center">Gene Symbol</td><td align="center">Entrez Gene ID</td><td align="center">Forward Primer</td><td align="center">Reverse Primer</td><td align="center">Product Size (bp)</td></tr></thead><tbody><tr><td align="left">NM_007393</td><td align="left">actin, beta, cytoplasmic</td><td align="center">Actb</td><td align="center">11461</td><td align="left">GCTACAGCTTCACCACCACA</td><td align="left">TCTCCAGGGAGGAAGAGGAT</td><td align="center">123</td></tr><tr><td align="left">NM_009992</td><td align="left">cytochrome P450, family 1, subfamily a, polypeptide 1</td><td align="center">Cyp1a1</td><td align="center">13076</td><td align="left">AAGTGCAGATGCGGTCTTCT</td><td align="left">AAAGTAGGAGGCAGGCACAA</td><td align="center">140</td></tr><tr><td align="left">NM_010634</td><td align="left">fatty acid binding protein 5, epidermal</td><td align="center">Fabp5</td><td align="center">16592</td><td align="left">TGTCATGAACAATGCCACCT</td><td align="left">CTGGCAGCTAACTCCTGTCC</td><td align="center">87</td></tr><tr><td align="left">NM_008084</td><td align="left">glyceraldehyde-3- phosphate dehydrogenase</td><td align="center">Gapd</td><td align="center">2597</td><td align="left">GTGGACCTCATGGCCTACAT</td><td align="left">TGTGAGGGAGATGCTCAGTG</td><td align="center">125</td></tr><tr><td align="left">NM_013556</td><td align="left">hypoxanthine phosphoribosyl transferase</td><td align="center">Hprt</td><td align="center">24465</td><td align="left">AAGCCTAAGATGAGCGCAAG</td><td align="left">TTACTAGGCAGATGGCCACA</td><td align="center">104</td></tr><tr><td align="left">NM_010849</td><td align="left">myelocytomatosis oncogene</td><td align="center">Myc</td><td align="center">17869</td><td align="left">CTGTGGAGAAGAGGCAAACC</td><td align="left">TTGTGCTGGTGAGTGGAGAC</td><td align="center">127</td></tr><tr><td align="left">NM_011723</td><td align="left">xanthine dehydrogenase</td><td align="center">Xdh</td><td align="center">22436</td><td align="left">GTCGAGGAGATCGAGAATGC</td><td align="left">GGTTGTTTCCACTTCCTCCA</td><td align="center">124</td></tr></tbody></table></table-wrap></sec></sec><sec><title>Authors' contributions</title><p><italic>In vitro </italic>work and associated microarrays and QRTPCR were conducted by ED, similarly, <italic>in vivo </italic>studies and associated microarrays and QRTPCR were performed by DRB. LDB provided the normalization, statistical analysis and database support of microarray data. Comparison of <italic>in vitro </italic>and <italic>in vivo </italic>data was primarily carried out by ED with support by DRB. ED produced the initial draft of the manuscript. TRZ was responsible for the design and oversaw the completion of the study.</p></sec><sec sec-type="supplementary-material"><title>Supplementary Material</title><supplementary-material content-type="local-data" id="S1"><caption><title>Additional File 1</title><p><bold>Hepalclc7 TCDD time course microarray data</bold>. Ratios represent expression relative to the time matched vehicle control. P1(<italic>t</italic>)-values represent posterior probabilities of activity on a per gene and time-point basis using the model-based t-value.</p></caption><media xlink:href="1471-2164-7-80-S1.xls" mimetype="application" mime-subtype="vnd.ms-excel"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S2"><caption><title>Additional File 2</title><p><bold>Hepalclc7 TCDD dose-response microarray data</bold>. Ratios represent expression relative to the time matched vehicle control. P1(<italic>t</italic>)-values represent posterior probabilities of activity on a per gene and dose basis using the model-based t-value.</p></caption><media xlink:href="1471-2164-7-80-S2.xls" mimetype="application" mime-subtype="vnd.ms-excel"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S3"><caption><title>Additional File 3</title><p><bold>C57BL/6 mice hepatic tissue TCDD time course microarray data</bold>. Ratios represent expression relative to the time matched vehicle control. P1(<italic>t</italic>)-values represent posterior probabilities of activity on a per gene and time-point basis using the model-based t-value.</p></caption><media xlink:href="1471-2164-7-80-S3.xls" mimetype="application" mime-subtype="vnd.ms-excel"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S4"><caption><title>Additional File 4</title><p><bold>C57BL/6 mice hepatic tissue TCDD dose-response microarray data</bold>. Ratios represent expression relative to the time matched vehicle control. P1(<italic>t</italic>)-values represent posterior probabilities of activity on a per gene and dose basis using the model-based t-value.</p></caption><media xlink:href="1471-2164-7-80-S4.xls" mimetype="application" mime-subtype="vnd.ms-excel"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S5"><caption><title>Additional File 5</title><p><bold>Untreated Hepalclc7 and C57BL/6 sample microarray data</bold>. Ratios represent basal expression of Hepalclc7 cells relative to hepatic tissue from C5VBL/6 mice. P1(<italic>t</italic>)-values represent posterior probabilities of activity on a per gene and dose basis using the model-based t-value.</p></caption><media xlink:href="1471-2164-7-80-S5.xls" mimetype="application" mime-subtype="vnd.ms-excel"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S6"><caption><title>Additional File 6</title><p><bold>Gene names and primer sequences (5'-3') for transcripts verified by QRTPCR</bold>. Primer pair sequences used to verify <italic>in vitro </italic>and <italic>in vivo </italic>microarray results using QRTPCR. </p></caption><media xlink:href="1471-2164-7-80-S6.xls" mimetype="application" mime-subtype="vnd.ms-excel"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S7"><caption><title>Additional File 7</title><p><bold>Hepa1c1c7 TCDD treatment and harvesting regimen</bold>. For the time course study, wild-type and ARNT-deficient c4 mutant cells were treated with 10 nM TCDD or 0.1% DMSO vehicle and harvested at 1, 2, 4, 8, 12, 24, or 48 hrs post-treatment. Untreated controls were harvested at 0 hrs (as indicated by ). The dose-response study was done performed with Hepa1c1c7 wild-type cells and treated with 0.001, 0.01, 0.1, 1.0, 10, 100 nM TCDD or 0.1% DMSO vehicle and harvested 12 hrs post-treatment (as indicated by ‡).</p></caption><media xlink:href="1471-2164-7-80-S7.png" mimetype="image" mime-subtype="png"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S8"><caption><title>Additional File 8</title><p><bold>Microarray experimental designs for A) temporal, B) dose-response and C) basal expression studies</bold>. A) Temporal gene expression patterns were analyzed by an independent reference design in which cells treated with TCDD (T) were co-hybridized to time-matched vehicle controls (V). This design involves two independent labelings per sample for a total of 14 arrays per replicate. Four biological replicates were conducted for a total of 56 microarrays. Numbers indicate time points for comparison. B) Dose-dependent changes in gene expression were analyzed 12 hrs after treatment using a common reference design in which cells treated with TCDD were co-hybridized with a common vehicle control. This design involves two independent labelings per sample for a total of 12 arrays per replicate. Four biological replicates were conducted for a total of 48 microarrays. Numbers indicate TCDD concentration in nM units. C) Comparative basal gene expression levels between untreated <italic>in vitro </italic>and <italic>in vivo </italic>samples were analyzed by an independent reference design. Four biological replicates of untreated Hepa1c1c7 cells and hepatic tissue from C57BL/6 mice harvested at 0 hrs were co-hybridized and two independent labelings were performed per sample for a total of 8 arrays. Double-headed arrows indicate dye swaps (each sample labeled with Cy3 and Cy5 on different microarrays).</p></caption><media xlink:href="1471-2164-7-80-S8.png" mimetype="image" mime-subtype="png"><caption><p>Click here for file</p></caption></media></supplementary-material></sec>
The complete chloroplast genome sequence of <italic>Gossypium hirsutum</italic>: organization and phylogenetic relationships to other angiosperms	<sec><title>Background</title><p>Cotton (<italic>Gossypium hirsutum</italic>) is the most important fiber crop grown in 90 countries. In 2004–2005, US farmers planted 79% of the 5.7-million hectares of nuclear transgenic cotton. Unfortunately, genetically modified cotton has the potential to hybridize with other cultivated and wild relatives, resulting in geographical restrictions to cultivation. However, chloroplast genetic engineering offers the possibility of containment because of maternal inheritance of transgenes. The complete chloroplast genome of cotton provides essential information required for genetic engineering. In addition, the sequence data were used to assess phylogenetic relationships among the major clades of rosids using cotton and 25 other completely sequenced angiosperm chloroplast genomes.</p></sec><sec><title>Results</title><p>The complete cotton chloroplast genome is 160,301 bp in length, with 112 unique genes and 19 duplicated genes within the IR, containing a total of 131 genes. There are four ribosomal RNAs, 30 distinct tRNA genes and 17 intron-containing genes. The gene order in cotton is identical to that of tobacco but lacks <italic>rpl22 </italic>and <italic>infA</italic>. There are 30 direct and 24 inverted repeats 30 bp or longer with a sequence identity ≥ 90%. Most of the direct repeats are within intergenic spacer regions, introns and a 72 bp-long direct repeat is within the <italic>psaA </italic>and <italic>psaB </italic>genes. Comparison of protein coding sequences with expressed sequence tags (ESTs) revealed nucleotide substitutions resulting in amino acid changes in <italic>ndhC, rpl23, rpl20, rps3 </italic>and <italic>clpP</italic>. Phylogenetic analysis of a data set including 61 protein-coding genes using both maximum likelihood and maximum parsimony were performed for 28 taxa, including cotton and five other angiosperm chloroplast genomes that were not included in any previous phylogenies.</p></sec><sec><title>Conclusion</title><p>Cotton chloroplast genome lacks <italic>rpl22 </italic>and <italic>infA </italic>and contains a number of dispersed direct and inverted repeats. RNA editing resulted in amino acid changes with significant impact on their hydropathy. Phylogenetic analysis provides strong support for the position of cotton in the Malvales in the eurosids II clade sister to <italic>Arabidopsis </italic>in the Brassicales. Furthermore, there is strong support for the placement of the Myrtales sister to the eurosid I clade, although expanded taxon sampling is needed to further test this relationship.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Lee</surname><given-names>Seung-Bum</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Kaittanis</surname><given-names>Charalambos</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Jansen</surname><given-names>Robert K</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Hostetler</surname><given-names>Jessica B</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Tallon</surname><given-names>Luke J</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Town</surname><given-names>Christopher D</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A7" corresp="yes" contrib-type="author"><name><surname>Daniell</surname><given-names>Henry</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Genomics	<sec><title>Background</title><p>The chloroplast is the site of photosynthesis, where light energy in photons is converted into chemical bond energy, via redox reactions, including inorganic carbon fixation at Calvin's cycle, finally yielding energy-rich carbohydrate molecules. Therefore, apart from the antennae, photosystem I and II complexes, which are found in the thylakoid membrane, the chloroplast contains the entire enzymatic machinery for carbohydrate biosynthesis in the stroma. Anabolic pathways such as protein, fatty acid, vitamin, and pigment biosynthesis take place in the chloroplast as well, indicating the organelle's ability to synthesize complex molecules. The chloroplast genome maintains a highly conserved organization [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B2">2</xref>] with most land plant genomes composed of a single circular chromosome with a quadripartite structure that includes two copies of an inverted repeat (IR) that separate the large and small single copy regions (LSC and SSC) [<xref ref-type="bibr" rid="B3">3</xref>]. The recent surge of interest in sequencing chloroplast genomes has provided a plethora of information on the organization and evolution of these genomes and new data for reconstructing phylogenetic relationships [<xref ref-type="bibr" rid="B2">2</xref>].</p><p>Chloroplast genetic engineering offers numerous advantages, including a high-level of transgene expression [<xref ref-type="bibr" rid="B4">4</xref>], multi-gene engineering in a single transformation event [<xref ref-type="bibr" rid="B4">4</xref>-<xref ref-type="bibr" rid="B7">7</xref>], transgene containment via maternal inheritance [<xref ref-type="bibr" rid="B8">8</xref>-<xref ref-type="bibr" rid="B11">11</xref>] or cytoplasmic male sterility [<xref ref-type="bibr" rid="B12">12</xref>], lack of gene silencing [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B13">13</xref>], position effect due to site specific transgene integration [<xref ref-type="bibr" rid="B14">14</xref>], and pleiotropic effects due to sub-cellular compartmentalization of transgene products [<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B15">15</xref>,<xref ref-type="bibr" rid="B16">16</xref>]. Apart from expressing therapeutic agents, biopolymers, or transgenes to confer agronomic traits, plastid genetic engineering has been used to study plastid biogenesis and function, revealing mechanisms of plastid DNA replication origins, intron maturases, translation elements and proteolysis, import of proteins and several other processes [<xref ref-type="bibr" rid="B18">18</xref>]. Despite the potential of chloroplast genetic engineering, this technology has only recently been extended to the major crops, including soybean [<xref ref-type="bibr" rid="B19">19</xref>], carrot [<xref ref-type="bibr" rid="B20">20</xref>] and cotton [<xref ref-type="bibr" rid="B21">21</xref>], via somatic embryogenesis, achieving transgene expression in non-green plastids [<xref ref-type="bibr" rid="B22">22</xref>]. All other previous studies focused on direct organogenesis by bombardment of leaves containing mature green chloroplasts [<xref ref-type="bibr" rid="B22">22</xref>]. Lack of complete chloroplast genome sequences to provide 100% homologous species-specific chloroplast transformation vectors, containing suitable selectable markers and endogenous regulatory elements, is one of the major limitations to extend this concept to other useful crops [<xref ref-type="bibr" rid="B22">22</xref>,<xref ref-type="bibr" rid="B23">23</xref>].</p><p>The need for sequencing the cotton plastome is obvious, when considering its annual retail value of about $120 billion, making it America's most value-added crop. This is justified by the fact that cotton is the single most important textile fiber grown in 90 countries; the US accounts for 21% of the total world fiber production. In 2004–2005, US farmers planted 79% of the 5.7-million hectares of nuclear transgenic cotton. Upland cotton, <italic>Gossypium hirsutum</italic>, has the potential to hybridize with <italic>G. tomentosum</italic>, feral populations of <italic>G. hirsutum</italic>, and <italic>G. hirsutum/G. barbadense </italic>[<xref ref-type="bibr" rid="B21">21</xref>]. Therefore, geographical restrictions in planting genetically modified cotton are in place because of reports of pollen dispersal from transgenic cotton plants [<xref ref-type="bibr" rid="B25">25</xref>]. Chloroplast genetic engineering could minimize transgene escape because of maternal inheritance of transgenes [<xref ref-type="bibr" rid="B8">8</xref>-<xref ref-type="bibr" rid="B11">11</xref>]. In addition, other failsafe mechanisms, including cytoplasmic male sterility could be employed to contain transgenes [<xref ref-type="bibr" rid="B12">12</xref>].</p><p>The examination of phylogenetic relationships among angiosperms has received considerable attention during the past decade [reviewed in [<xref ref-type="bibr" rid="B26">26</xref>]]. Although there is considerable consensus about the circumscription and relationships among many of the major clades, most molecular phylogenetic analyses have examined numerous taxa but have relied on only a few gene sequences. Completely sequenced chloroplast genomes provide a rich source of nucleotide sequence data that can be used to address phylogenetic questions. Several recent studies have attempted to use completely sequenced genomes to resolve the identification of the basal lineages of flowering plants [<xref ref-type="bibr" rid="B27">27</xref>-<xref ref-type="bibr" rid="B29">29</xref>]. Use of many or all of the genes from the chloroplast genome provides many more characters for phylogeny reconstruction in comparison with previous studies that have relied on only a few genes. However, the limited number of available whole chloroplast genome sequences can result in misleading estimates of relationship [<xref ref-type="bibr" rid="B27">27</xref>,<xref ref-type="bibr" rid="B30">30</xref>]. This problem can be overcome as more complete chloroplast genome sequences become available.</p><p>In this article, we present the complete sequence of the chloroplast genome of upland cotton, <italic>Gossypium hirsutum</italic>. One goal of this paper is to examine gene content and gene order, and determine the distribution and location of repeated sequences. Secondly, the RNA editing sites in the cotton chloroplast genome are identified and examined, by comparing the DNA sequences with available expressed sequence tag (EST) sequences, because RNA editing plays a major role in several lineages of plants [<xref ref-type="bibr" rid="B31">31</xref>,<xref ref-type="bibr" rid="B32">32</xref>]. Lastly, protein-coding sequences from 61 genes are used to estimate phylogenetic relationships of cotton with 25 other angiosperms.</p></sec><sec><title>Results</title><sec><title>Size, gene content, order and organization of the cotton chloroplast genome</title><p>Cotton complete chloroplast genome is 160,301 bp in length (Fig. <xref ref-type="fig" rid="F1">1</xref>), and includes a pair of inverted repeats 25,608 bp long, separated by a small and a large single copy region of 20,269 bp and 88,816 bp, respectively. There are 112 unique genes within the cotton chloroplast genome and 19 of these are duplicated in the IR, giving a total of 131 genes (Fig. <xref ref-type="fig" rid="F1">1</xref>). Furthermore, there are four ribosomal and 30 distinct tRNA genes; seven of the tRNA genes and all rRNA genes are duplicated within the IR. There are 17 intron-containing genes, 15 of which contain one intron, whereas the remaining two have two introns. The gene order in the cotton plastid genome is identical to that of tobacco, but cotton lacks the <italic>rpl22 </italic>and <italic>infA </italic>genes. Overall, genomic content is 37.25% GC and 62.75% AT, where 56.46% of the genome corresponds to protein coding genes and 43.54% to non-coding regions, including introns and intergenic spacers.</p></sec><sec><title>Repeat structure</title><p>Repeat analysis identified 30 direct and 24 inverted repeats 30 bp or longer with a sequence identity of at least 90% (Fig. <xref ref-type="fig" rid="F2">2</xref> and Table <xref ref-type="table" rid="T1">1</xref>). Twenty-three direct and 15 inverted repeats are 30 to 40 bp long, and the longest direct repeat is 72 bp. Most of the direct repeats are within intergenic spacer regions, intron sequences and <italic>ycf2</italic>, an essential hypothetical chloroplast gene [<xref ref-type="bibr" rid="B33">33</xref>]. Interestingly, a 72 bp-long direct repeat was found in the <italic>psaA </italic>and <italic>psaB </italic>genes, whereas a 34-bp forward repeat was within the <italic>rrn23 </italic>gene, and a shorter, 32 bp-long direct repeat was identified in two serine transfer-RNA(<italic>trnS</italic>) genes that recognize different codons; <italic>trnS-GCU </italic>and <italic>trnS-UGA</italic>.</p></sec><sec><title>RNA editing</title><p>Comparison of the nucleotide sequences of protein coding genes and EST sequences retrieved from GenBank revealed that <italic>rps16, rpl2, rpoC2, rps4 </italic>and <italic>ycf1 </italic>have 100% sequence identity with their respective ESTs (data not shown). Eleven non-synonymous nucleotide substitutions, resulting in a total of nine amino acid changes, were identified within <italic>ndhC, rpl23, rpl20, rps3 </italic>and <italic>clpP </italic>compared to respective ESTs, although their sequence identity was above 98% (Table <xref ref-type="table" rid="T2">2</xref>). Surprisingly, there were no synonymous substitutions. All of the five aforementioned genes experienced one or two nucleotide substitutions, apart from the protease-encoding <italic>clpP</italic>, which had five variable sites. Lastly, in all but <italic>rpl23</italic>, the nucleotide substitutions had an impact on the hydropathy of the amino acid because they changed the amino acids from aliphatic to hydrophilic, and vice versa.</p></sec><sec><title>Phylogenetic analysis</title><p>The data matrix for phylogenetic analyses included 61 protein-coding genes for 28 taxa (Table <xref ref-type="table" rid="T3">3</xref>), including 26 angiosperms and two gymnosperm outgroups (<italic>Pinus </italic>and <italic>Ginkgo</italic>). The data set comprised 45,573 nucleotide positions but when the gaps were excluded there were 39,624 characters. Maximum Parsimony (MP) analyses resulted in a single, fully resolved tree with a length of 49,957, a consistency index of 0.46 (excluding uninformative characters) and a retention index of 0.62 (Fig. <xref ref-type="fig" rid="F3">3</xref>). Bootstrap analyses indicated that 24 of the 26 nodes were supported by values ≥ 95% with 19 of these with bootstrap values of 100%. Maximum Likelihood (ML) analysis resulted in a tree with a –lnL = 311251.33. The ML and MP trees had identical topologies so only the MP tree is shown in Figure <xref ref-type="fig" rid="F3">3</xref>.</p><p>Several major groups were supported within angiosperms and these groups are generally in agreement with recent classifications [<xref ref-type="bibr" rid="B26">26</xref>]. The most basal lineage was <italic>Amborella </italic>followed by the Nymphaeales. The next branch included <italic>Calycanthus</italic>, the sole representative of magnoliids in the data set. This was followed by a strongly supported clade of monocots, represented by members of three different orders (Acorales, Asparagales, and Poales). The monocots were then sister to the eudicots with the Ranunculales forming the earliest diverging eudicot clade. Within the core eudicots there were two major clades, one including the rosids and the second including the Caryophyllales sister to asterids. Within the rosid clade there were two major groups, the eurosids II and a group that included the Myrtales sister to the eurosids I. <italic>Gossypium </italic>in the Malvales was sister to <italic>Arabidopsis </italic>in the Brassicales.</p></sec></sec><sec><title>Discussion</title><sec><title>Implications for integration of transgenes</title><p>We have recently demonstrated stable transformation of the cotton plastid genome and maternal inheritance of transgenes via somatic embryogenesis [<xref ref-type="bibr" rid="B21">21</xref>]. In contrast to previous reports on integrating foreign genes in tomato and potato chloroplast genomes using tobacco flanking sequences that do not have 100% sequence identity [<xref ref-type="bibr" rid="B24">24</xref>,<xref ref-type="bibr" rid="B34">34</xref>,<xref ref-type="bibr" rid="B35">35</xref>], the cotton plastid transformation vector was constructed using the PCR-amplified native cotton 16S/<italic>trnI</italic>-<italic>trnA/</italic>23S sequence. However, regulatory sequences used in the cotton plastid transformation were derived from tobacco or other heterologous sequences. With the availability of the entire cotton chloroplast genome sequence, it should now be possible to utilize endogenous regulatory sequences. Species-specific vectors should be effective for plastid transformation, especially in recalcitrant plants, because of transgene integration using flanking sequences with 100% sequence identity and endogenous promoters, 5' & 3'untranslated regions, thereby enhancing transcription and translation of transgenes. Also, the complete chloroplast genome provides the option of transgene integration into transcriptionally silent, active or read-through spacer regions for optimal transgene integration.</p><p>Thus far, transgenes conferring several useful agronomic traits, including insect [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B36">36</xref>,<xref ref-type="bibr" rid="B37">37</xref>], herbicide [<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B38">38</xref>], and disease resistance [<xref ref-type="bibr" rid="B39">39</xref>], drought [<xref ref-type="bibr" rid="B13">13</xref>] and salt tolerance [<xref ref-type="bibr" rid="B20">20</xref>], phytoremediation [<xref ref-type="bibr" rid="B5">5</xref>], as well as cytoplasmic male sterility [<xref ref-type="bibr" rid="B12">12</xref>], have been stably integrated and expressed, via the tobacco chloroplast genome. Using the chloroplast as a bioreactor, vaccine antigens [<xref ref-type="bibr" rid="B15">15</xref>,<xref ref-type="bibr" rid="B40">40</xref>-<xref ref-type="bibr" rid="B42">42</xref>], human therapeutic proteins [<xref ref-type="bibr" rid="B17">17</xref>,<xref ref-type="bibr" rid="B43">43</xref>-<xref ref-type="bibr" rid="B45">45</xref>], industrial enzymes [<xref ref-type="bibr" rid="B46">46</xref>] and biomaterials [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B47">47</xref>,<xref ref-type="bibr" rid="B48">48</xref>] have been produced successfully in an environmental friendly way. Although many successful examples of plastid engineering in tobacco have set a solid foundation for various future applications, this technology has not been extended to many of the major crops, primarily due to the lack of complete chloroplast genome sequences and challenges in achieving homoplasmy in recalcitrant crops.</p></sec><sec><title>Evolutionary implications</title><p>Other than the IR, repeated sequences are generally considered to be uncommon in chloroplast genomes [<xref ref-type="bibr" rid="B1">1</xref>]. Furthermore, previous studies based on both filter hybridization and DNA sequencing have indicated that dispersed repeats are found more commonly in genomes that have experienced changes in genome organization [<xref ref-type="bibr" rid="B49">49</xref>,<xref ref-type="bibr" rid="B56">56</xref>], especially in highly rearranged algal genomes [<xref ref-type="bibr" rid="B51">51</xref>,<xref ref-type="bibr" rid="B52">52</xref>]. The most extensive examination of repeat structure in angiosperms was performed in legumes [<xref ref-type="bibr" rid="B3">3</xref>], which do have a single inversion and in some taxa a loss of one copy of the IR. These repeat analyses identified a substantial number highly conserved repeats ≥ 30 bp with a sequence identity of ≥ 90%. Many of these repeats were located in intergenic spacer regions and introns, with several located in the coding regions of <italic>psaA</italic>, <italic>psaB</italic>, and <italic>ycf2</italic>. Our examination of repeats in the cotton chloroplast genome (Table <xref ref-type="table" rid="T1">1</xref>, Fig. <xref ref-type="fig" rid="F2">2</xref>) identified similar numbers of repeats as in legumes [<xref ref-type="bibr" rid="B3">3</xref>], and these are also located mostly in intergenic spacer regions and introns. Repeats in coding regions of cotton are located in the same genes as in legumes. Overall, it appears that dispersed repeats are very common in angiosperm chloroplast genomes, even in genomes that have not experienced rearrangements. Future comparative studies are needed to determine the functional and evolutionary role these repeats may play in chloroplast genomes.</p><p>DNA and EST sequence comparisons identified many nucleotide substitutions resulting in amino acid changes. Based on previous studies of <italic>Atropa </italic>[<xref ref-type="bibr" rid="B53">53</xref>] and tobacco [<xref ref-type="bibr" rid="B54">54</xref>], posttranscriptional RNA editing events result predominantly in C-to-U edits. However, analysis of the cotton genome and EST sequences indicates that only two of the eleven differences were C-to-U changes, suggesting that most of these changes are not mRNA edits but may simply represent intra-species polymorphisms. Evolutionary loss of RNA editing sites has been previously observed and could possibly be due to a decrease in the effect of RNA-editing enzymes [<xref ref-type="bibr" rid="B31">31</xref>]. Additionally, conversions other than C-to-U in cotton, as well as other crops, suggest that chloroplast genomes may be accumulating considerable amounts of nucleotide substitutions, where some genes might accrue more alterations than others, such as the <italic>petL </italic>and <italic>ndh </italic>genes that have a high frequency of RNA editing [<xref ref-type="bibr" rid="B55">55</xref>]. Therefore, despite the plastome's high conservation, variations occur post-transcriptionally, promoting translational efficiency due to transcript-protein complex binding and/or changes in the chloroplast microenvironment, like redox potential or light intensity [<xref ref-type="bibr" rid="B56">56</xref>,<xref ref-type="bibr" rid="B57">57</xref>].</p><p>The phylogeny based on 61 protein-coding genes for 28 angiosperms is congruent with relationships suggested in previous studies [summarized in [<xref ref-type="bibr" rid="B26">26</xref>]]. There is strong support for the monophyly all of the major clades of angiosperms, including monocots, eudicots, rosids, asterids, eurosids I, eurosids II, asterids I and asterid II. Our phylogenetic analyses have greatly expanded the taxon sampling of entire genomes because we included six genomes (in bold in Table <xref ref-type="table" rid="T1">1</xref> and Fig. <xref ref-type="fig" rid="F3">3</xref>) that have not been included in recently published phylogenies based on complete chloroplast genomes [<xref ref-type="bibr" rid="B27">27</xref>-<xref ref-type="bibr" rid="B29">29</xref>,<xref ref-type="bibr" rid="B58">58</xref>]. The sampling is particularly expanded in the rosids with four of the six genomes from this clade. Thus, we will focus our discussion of the phylogenetic implications of this expanded analysis on this group.</p><p>The rosid clade is very large and includes nearly 140 families representing almost one third of all angiosperms. The most recent phylogenies of this group [summarized in chapter 8 in [<xref ref-type="bibr" rid="B26">26</xref>]] indicate that there are seven major clades whose relationships still remain unresolved. Representatives of three of these major clades are included in our analyses, eurosids I, eurosids II, and Myrtales. The position of the Myrtales has been especially controversial with no clear resolution of the relationship of this order to other members of the rosids. Our 61 gene chloroplast phylogeny (Fig. <xref ref-type="fig" rid="F3">3</xref>) provides strong support for a sister relationship of the Myrtales with the eurosid I clade. A three-gene phylogeny of 560 angiosperms is congruent with our results [<xref ref-type="bibr" rid="B59">59</xref>], although support was very weak. However, a sister relationship between eurosids I and Myrtales is in conflict with two other recent phylogenies based on two chloroplast genes (<italic>atpB</italic>, <italic>rbcL</italic>), which placed the Myrtales sister to the eurosid II clade with weak support [<xref ref-type="bibr" rid="B60">60</xref>,<xref ref-type="bibr" rid="B61">61</xref>]. Although our results clearly favor a closer relationship of Myrtales to the eurosid I clade, expanded sampling of complete chloroplast genome sequences of rosids is needed to resolve this issue, especially since limited taxon sampling can lead to erroneous tree topologies [<xref ref-type="bibr" rid="B27">27</xref>,<xref ref-type="bibr" rid="B30">30</xref>].</p><p>Our chloroplast phylogeny (Fig. <xref ref-type="fig" rid="F3">3</xref>) also supports the sister relationship between the orders Cucurbitales and Fabales, two of the four nitrogen fixing clades of eurosids I. Furthermore, the position of cotton, a member of the order Malvales, as sister to <italic>Arabidopsis </italic>in the Brassicales, is in agreement with recently phylogenies of the eurosid II clade [<xref ref-type="bibr" rid="B26">26</xref>].</p></sec></sec><sec><title>Conclusion</title><p>Our complete sequence of the cotton chloroplast genome provides the needed information for expanding chloroplast genetic engineering to this important crop plant. Although genome organization of cotton is very similar to other unrearranged angiosperm chloroplast genomes, identification of disperse repeats and potential RNA editing sites provides new insights into the evolution of this genome. Finally, phylogenetic analyses of sequences of 61 protein-coding genes for 26 angiosperms suggests that the order Myrtales is sister to the eurosid I clade but denser sampling is needed to test this result rigorously.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>DNA isolation and amplification</title><p><italic>Gossypium hirsutum </italic>plants cv. Coker310FR were grown from seedlings in soil pots, until they were 1 m tall. Prior to DNA extraction, the plants were placed in the dark for two days to reduce the chloroplast starch levels. After that, 10 g of young leaf tissue was collected for cpDNA isolation based on the sucrose step gradient centrifugation method by Sandbrink <italic>et al </italic>[<xref ref-type="bibr" rid="B62">62</xref>]. Isolation was followed by whole chloroplast genome Rolling Circle Amplification (RCA), using the Repli-g RCA kit (Qiagen, Inc.) following the methods outlined in [<xref ref-type="bibr" rid="B63">63</xref>]. After incubation at 30°C for 16 hr, the reaction was terminated with 10-minute incubation at 65°C. Digestion of the RCA product with <italic>BstXI, EcoRI </italic>and <italic>HindIII </italic>allowed verification of successful RCA plastome amplification, as well as assessment of its quality, prior to DNA sequencing.</p></sec><sec><title>DNA sequencing and genome assembly</title><p>DNA was sheared by nebulization, size fractionated to 4–6 kb, linker ligated and cloned into pHOS2, a TIGR medium copy vector. A total of 1619 good reads with an average length of 812 bases was generated during the random (1396 reads) and closure (223 reads) phases of sequencing. Sequences were assembled using TIGR assembler [<xref ref-type="bibr" rid="B64">64</xref>] and scaffolded using Bambus [<xref ref-type="bibr" rid="B65">65</xref>]. Sequence finishing included directed PCR to span gaps, directed primer walking on clones and transposon mediated sequencing of full clones to cover the entire genome and complete regions of low coverage and manual editing of sequences to resolve inconsistencies.</p></sec><sec><title>Gene annotation</title><p>The cotton genome was annotated using DOGMA [Dual Organellar GenoMe Annotator, [<xref ref-type="bibr" rid="B66">66</xref>]], after uploading a FASTA-formatted file of the complete plastid genome to the program's server. BLASTX and BLASTN searches, against a custom database of previously published plastid genomes, identified cotton's putative protein-coding genes, and tRNAs or rRNAs. For genes with low sequence identity, manual annotation was performed, after identifying the position of the start and stop codons, as well as the translated amino acid sequence, using the plastid/bacterial genetic code.</p></sec><sec><title>Examination of repeat structure</title><p>REPuter [<xref ref-type="bibr" rid="B67">67</xref>] was used to locate and count the direct (forward) and inverted (palindromic) repeats within the cotton chloroplast genome. For repeat identification, the following constraints were used: (i) minimum repeat size of 30 bp, and (ii) 90% or greater sequence identity, based on Hamming distance equal to 3 bp [<xref ref-type="bibr" rid="B3">3</xref>]. Manual verification of the identified repeats was performed in EditSeq, while performing intragenomic blast search of the identified repeat sequence.</p></sec><sec><title>Variation between coding sequences and cDNAs</title><p>Each of the gene sequences from the cotton chloroplast genome was used to perform a BLAST search of expressed sequence tags (ESTs) from GenBank. The retrieved <italic>Gossypium hirsutum </italic>ESTs were aligned with the corresponding annotated gene using ClustalX [<xref ref-type="bibr" rid="B68">68</xref>], followed by screening for nucleotide and amino acid changes using Megalign and its' plastid/bacterial genetic code. Because of variation in the length between an EST and the related gene, the length of the analyzed sequence was recorded.</p></sec><sec><title>Phylogenetic analysis</title><p>The 61 genes included in the analyses of Goremykin et al. [<xref ref-type="bibr" rid="B28">28</xref>,<xref ref-type="bibr" rid="B29">29</xref>] and Leebens-Mack et al. [<xref ref-type="bibr" rid="B27">27</xref>] were extracted from our new chloroplast genome sequences of cotton using the organellar genome annotation program DOGMA. [<xref ref-type="bibr" rid="B66">66</xref>]. The same set of 61 genes was extracted from chloroplast genome sequences of five other recently sequenced angiosperm chloroplast genomes, including tomato, potato, soybean, cucumber, and <italic>Eucalyptus </italic>(see Table <xref ref-type="table" rid="T3">3</xref> for complete list of genomes examined). In general, alignment of the DNA sequences was straightforward and simply involved removing gaps included in the data set because of the elimination of non-seed plants and adding the 61 genes for the new angiosperms to the aligned data matrix from Leebens-Mack et al. [<xref ref-type="bibr" rid="B27">27</xref>]. In some cases, small in frame insertions or deletions were required for correct alignment. For two genes, <italic>ccsA </italic>and <italic>matK</italic>, the DNA sequences were more divergent, requiring alignment using ClustalX [<xref ref-type="bibr" rid="B68">68</xref>] followed by manual adjustments.</p><p>Phylogenetic analyses using maximum parsimony (MP) and maximum likelihood (ML) were performed using PAUP* version 4.10 [<xref ref-type="bibr" rid="B69">69</xref>]. All phylogenetic analyses excluded gap regions. All MP searches were heuristic with 100 random addition replicates and TBR branch swapping with the Multrees option. The Hasegawa-Kishino-Yano (HKY; [<xref ref-type="bibr" rid="B70">70</xref>]) model of molecular evolution was used in ML analyses of the nucleotide sequences. ML analyses used TBR branch swapping with the Multrees option and one random addition replicate. Non-parametric bootstrap analyses [<xref ref-type="bibr" rid="B71">71</xref>] were performed for MP analyses with 1000 replicates with TBR branch swapping, one random addition replicate, and the Multrees option and for ML analyses with 100 replicates with NNI branch swapping, one random addition replicate, and the Multrees option.</p></sec></sec><sec><title>Abbreviations</title><p>cpDNA, chloroplast DNA; IR inverted repeat; SSC, small single copy; LSC, large single copy, bp, base pair; ycf, hypothetical chloroplast reading frame; rrn, ribosomal RNA; MP, maximum parsimony; ML, maximum likelihood; EST, expressed sequence tag; cDNA, complementary DNA.</p></sec><sec><title>Authors' contributions</title><p>SBL isolated chloroplasts, performed RCA amplification of cpDNA, genome annotation, analysis and submission of data to the GenBank; CK performed the repeat analyses, comparisons of DNA and EST sequences, assisted with extraction & alignment of DNA sequences for phylogenetic analyses and wrote a few sections of the first draft; JBH, LJT and CDT performed DNA sequencing and genome assembly; RKJ assisted with extracting and aligning DNA sequences, performed phylogenetic analyses, and wrote the phylogenetic portions of the manuscript; HD conceived and designed this study, interpreted data, wrote and revised several versions of this manuscript. All authors read and approved the final manuscript.</p></sec>
Marked stem cell factor expression in the airways of lung transplant recipients	<sec><title>Background</title><p>Airways repair is critical to lung function following transplantation. We hypothesised that the stem cell factor (SCF) could play a role in this setting.</p></sec><sec sec-type="methods"><title>Methods</title><p>We studied 9 lung transplant recipients (LTx recipients) during their first year postgraft, and evaluated SCF mRNA expression in bronchial biopsy specimens using on-line fluorescent PCR and SCF protein levels in bronchoalveolar lavage (BAL) and serum using ELISA. The expression of SCF receptor Kit was assessed using immunostaining of paraffin-embedded bronchial sections.</p></sec><sec><title>Results</title><p>SCF mRNA was highly expressed during the early postgraft period [Month (M)1-M3] (300% increase vs controls: 356 vs 1.2 pg SCF/μg GAPDH cDNA, <italic>p </italic>< 0.001) and decreased thereafter (M4-M12: 187 pg/μg), although remaining at all times 10–100 times higher than in controls. While SCF protein levels in BAL were similar in LTx recipients and in controls, the SCF serum levels were at all times higher in LTx recipients than in controls (<italic>p </italic>< 0.05), with no relationship between these levels and the acute complications of the graft. Finally, Kit was strongly expressed by the mast cells as well as by the bronchial epithelium of LTx recipients.</p></sec><sec><title>Conclusion</title><p>SCF and Kit are expressed in bronchial biopsies from lung transplant recipients irrespective of the clinical status of the graft. A role for these factors in tissue repair following lung transplantation is hypothesised.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Da Silva</surname><given-names>Carla A</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Adda</surname><given-names>Mélanie</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Stern</surname><given-names>Marc</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>de Blay</surname><given-names>Frédéric</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Frossard</surname><given-names>Nelly</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A6" corresp="yes" contrib-type="author"><name><surname>Israel-Biet</surname><given-names>Dominique</given-names></name><xref ref-type="aff" rid="I2">2</xref><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib>	Respiratory Research	<sec><title>Background</title><p>Injury processes of various types can damage a grafted organ. Some are due to the surgical procedure itself-the section of vessels and nerves and, for lung transplants, conducting airways. Others are inflammatory in nature, due to reperfusion of the graft or early allogeneic reactions. The lack of efficient tissue repair mechanisms would severely impair graft functioning. The restoration of transplanted airways involves a variety of cell types: local progenitors, found mainly among basal and Clara cells [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B3">3</xref>], and host stem cells from the bone marrow, a rich reservoir of progenitors for different cell types including mast cells [<xref ref-type="bibr" rid="B4">4</xref>-<xref ref-type="bibr" rid="B6">6</xref>]. Experimental studies in murine models provide evidence that bone marrow stem cells can differentiate into type I and type II pneumocytes and bronchial epithelial cells [<xref ref-type="bibr" rid="B5">5</xref>,<xref ref-type="bibr" rid="B7">7</xref>]. The ability of such cells to engraft into the human lung is still controversial although studies after hematopoietic stem cell transplantation in humans show evidence of donor-derived cells (chimerism) in the lung [<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B9">9</xref>]. In this context, SCF is a key factor in mobilisation of stem cells from bone marrow where it facilitates egress from the marrow to the circulation [<xref ref-type="bibr" rid="B10">10</xref>]. On the other hand, there is some evidence for the involvement of local progenitor cells in repair processes following tissue injury in murine models [<xref ref-type="bibr" rid="B11">11</xref>-<xref ref-type="bibr" rid="B13">13</xref>] as well as in human situations [<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B15">15</xref>]. The SCF/Kit pathway has also been shown to play a key role in these latter.</p><p>To our knowledge, SCF production has never been evaluated in the lung in settings other than asthma, where it is markedly upregulated [<xref ref-type="bibr" rid="B16">16</xref>]. More specifically, it has never been evaluated in the context of solid organ transplantation. We hypothesised that it might be produced soon after transplantation to participate in airway tissue repair. We therefore evaluated SCF mRNA expression in bronchial biopsy specimens and SCF protein expression in bronchoalveolar lavage (BAL) and serum during a one-year follow-up of human LTx recipients. We also assessed SCF receptor Kit expression in bronchial epithelium using immunodetection. Finally, because SCF is a well-known mast cell growth factor (for review see [<xref ref-type="bibr" rid="B17">17</xref>,<xref ref-type="bibr" rid="B18">18</xref>]), we also quantified the number of mast cells infiltrating the airways.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Study population</title><p>Nine (3 single and 6 double) lung transplant recipients (3 males, 6 females), aged 41 (median; range: 22–60 years), were included into this study. Their initial diagnoses were cystic fibrosis (n = 3), emphysema (n = 2), diffuse bronchiectasis (n = 2), pulmonary artery hypertension associated with systemic sclerosis (n = 1), and lymphangioleiomyomatosis (n = 1). All patients were initially placed on an immunosuppressive regimen including cyclosporin A, azathioprine and prednisone, until the third episode of acute lung rejection (ALR) after which cyclosporin A was switched to tacrolimus and azathioprine to mycophenolate mofetil. In addition to clinical and functional evaluation, fiberoptic bronchoscopy with bronchoalveolar lavage (BAL), transbronchial biopsies and proximal bronchial biopsies were routinely performed in each patient either to detect asymptomatic complications at day 30 after surgery (then every month during the first 6 months and every 6 months thereafter) or as part of the routine procedure for the diagnosis of complications suspected on clinical and/or functional grounds. One patient died 8 months after the transplant of infectious (Aspergillus) complications. None of the patient was diagnosed with bronchiolitis obliterans syndrome (BOS) during the study period but 4 of them subsequently developed this complication between 15 and 33 months post-transplantation, leading to death in two.</p><p>Control subjects (4 males and 5 females, aged 25 (median; range: 21–38 years) were all healthy volunteers. The study was approved by the local Ethics committee, and all patients and control subjects provided written informed consent.</p></sec><sec><title>Tissue specimens</title><p>Bronchial biopsy specimens were fixed in 10% buffered formalin and paraffin-embedded. They were cut into four- μm sections for morphological and immunohistochemical evaluation. An additional biopsy sample was snap-frozen in liquid nitrogen for subsequent RNA extraction. Four to six serial samples per patient, recovered during the first year post-transplantation, were evaluated.</p></sec><sec><title>RNA extraction and reverse transcription</title><p>Total RNA was extracted from biopsy specimens with TriReagent<sup>® </sup>(Molecular Research Center Inc., Cincinnati, OH, USA). Isolated RNA was diluted in RNase-free water and quantified by absorbance measurement at 260 nm. Two μg of total RNA were incubated with 0.5 μg of random primers for 5 min at 70°C, and allowed to cool down at room temperature. RNA was subsequently reverse-transcribed in 1× reverse transcription (RT) buffer (75 mM KCl, 3 mM MgCl2, 10 mM dithiothreitol, and 50 mM Tris-HCl, pH8.3), containing 1 unit/μl RNasin ribonuclease inhibitor, 1 mM of each dNTP, and 10 units/μl RNase H(-)-Moloney leukemia virus reverse transcriptase (all reagents from Promega, Madison, WI, USA). The reaction was conducted for 1 h at 37°C, and then the reverse transcriptase was heat-inactivated at 99°C for 5 min.</p></sec><sec><title>Quantification of SCF cDNA</title><p>Reverse transcribed cDNAs were amplified by on-line fluorescent PCR (LightCycler™-SYBR GreenI, Roche Diagnostics), with primers leading to single 149-bp and 240-bp PCR products for quantification of SCF and GAPDH (used as a housekeeping gene) cDNAs, respectively. PCR reactions were performed in 1× PCR reaction buffer [2 μl of the reaction mix, containing FastStart Taq DNA polymerase, dNTP mix, SYBR Green I, 3 mM MgCl<sub>2 </sub>(Roche Diagnostics)], and 10 pmol of each primer:</p><p>SCF amplification: sense primer 5'-TGGATAAGCGAGATGGTAGT-3'antisense primer 5'-TTTTCTTTCACGCACTCCAC-3', GAPDH amplification:sense primer 5'-GGTGAAGGTCGGAGTCAACGGA-3' antisense primer5'-GAGGGATCTCGCTCCTGGAAGA-3', in a 20 μl final volume for 35 cycles. Each cycle consisted of 15 sec denaturation at 95°C, 10 sec annealing (at 53°C for SCF cDNA amplification and 60°C for GAPDH) and 10 sec extension at 72°C. Amplified SCF and GAPDH cDNAs and standard SCF and GAPDH cDNAs were analysed on-line by fluorescence (LightCycler™, Roche Diagnostics).</p><p>The standard SCF and GAPDH cDNA came from pulmonary fibroblast total cDNA. After amplification by PCR, with the primers described above, the 149 bp- and 240 bp-PCR products were electrophoresed on a 2% agarose gel, stained with ethidium bromide, purified on QIAEX II (QIAEX II gel extraction kit, QIAGEN, Courtaboeuf, France), and quantified by fluorescence (PicoGreen<sup>®</sup>, Molecular Probes Inc.) according to a standard curve obtained with a double-stranded phage λ DNA (0.005–1 μg/ml). The purified SCF cDNA and GAPDH cDNA were used to establish a standard curve from 1 to 300 fg/ml.</p><p>Quantification of SCF cDNA (pg) was normalised to the GAPDH cDNA measurement. Results were expressed as pg SCF cDNA/μg GAPDH cDNA.</p></sec><sec><title>Soluble SCF protein measurement in BAL and serum</title><p>A sensitive ELISA procedure quantified immunoreactive SCF released into patients' BAL fluid and serum. It used a capture anti-human SCF monoclonal antibody (R&D Systems Europe, Abingdon, UK, clone 13302) and an anti-human SCF biotinylated polyclonal antibody (R&D Systems Europe), revealed by extravidin-horseradish peroxidase and a 3,3',5,5'-tetramethylbenzidine liquid substrate system (Sigma Chemicals, St Louis, MO, USA). Standard curves were generated with recombinant human SCF (R&D Systems Europe) diluted in fœtal calf serum. They were linear from 16 to 500 pg/ml. Eight normal blood donors provided serum for control purposes. Soluble SCF concentrations were expressed as pg/ml of BAL fluid or serum.</p></sec><sec><title>Kit immunodetection</title><p>Four- μm sections of paraffin-embedded biopsies were cut and used for immunohistochemistry. Briefly, tissue sections previously deparaffinised in toluene and rehydrated through graded concentrations of ethanol, were incubated for 60 min at room temperature with a primary polyclonal rabbit anti-human CD117 (Kit) (1/100) (Santa Cruz Biotechnology, Inc; CA, USA). After two washings with Tris buffer at pH 7.6, staining was revealed according to the avidin-biotin method using a LSAB AP kit (Dako Corp; Carpinteria, CA, USA) and FastRed as a substrate. Tissue sections were finally counterstained with hematoxylin and mounted with Ultramount medium (Dako Corp; CA, USA). To ensure the anti-Kit binding specificity, two types of negative controls were used: 1) by omitting the primary antibody and 2) by preincubating the primary antibody with a specific blocking peptide (sc-168 P, Santa Cruz Biotechnology, Inc; CA, USA).</p></sec><sec><title>Mast cell counts</title><p>The immunodetection of mast cells was similar to that described above with the primary antibody being a monoclonal mouse anti-human tryptase (0.25 μg/ml, Dako Corp; CA, USA). Results are expressed as the number of tryptase-positive cells per mm<sup>2 </sup>of bronchial structure.</p></sec><sec><title>Statistical analysis</title><p>All results are expressed as median values (SCF mRNA, SCF protein levels, mast cell numbers) collected during two different periods of time: M1-M3 (months 1 to 3 postgraft, during which the immunosuppressive regimens are usually the most intensive) and M4-M12 (months 4 to 12 post-graft, during which corticosteroids doses are usually tapered in case of an uneventful evolution of the graft). Comparisons were made using the non-parametric Mann Whitney U-test for unpaired series. A <italic>p </italic>value <0.05 was considered significant.</p></sec></sec><sec><title>Results</title><sec><title>SCF mRNA</title><p>High levels of SCF mRNA expression were observed in the airways of every lung transplant recipient. This expression was particularly marked during the early post-transplant period (M1-M3), when it was 300% of the level in controls (356 pg SCF/μg GAPDH (range: 59–1826) in patients and 1.2 pg SCF/μg GAPDH mRNA (range: 0.4–5.8) in controls, <italic>p </italic>< 0.001) (Figure <xref ref-type="fig" rid="F1">1</xref>). SCF expression thereafter decreased in lung transplant recipients (M4-M12: 187 pg/μg GAPDH mRNA (range: 10–987), <italic>p </italic>< 0.05 <italic>vs </italic>M1-M3) but remained 10–100 times higher in patients than controls throughout this period (Figure <xref ref-type="fig" rid="F1">1</xref>). We found no statistical relation between clinical status and SCF mRNA expression, which remained at comparable levels when the patient's condition was stable and during acute complications.</p><fig position="float" id="F1"><label>Figure 1</label><caption><p>SCF mRNA expression in lung transplant recipients. SCF mRNA was quantified after total RNA reverse transcription by on-line fluorescent PCR in biopsy specimen from control subjects (Control) and lung tranplant patients at Months (Mo) postgraft 1 to 3 (M1-M3), and 4 to 12 (M4-M12). Results were normalised to GAPDH mRNA expression, and expressed in pg/μg as the ratio [SCF mRNA]/[GAPDH mRNA]. Symbols denote individual episodes (◆ : stable condition, ○ : acute lung rejection, × : infection). Median values are also shown (bold line) (NS; non-significative, ; <italic>p </italic>< 0.05, ; <italic>p </italic>< 0.01, *; <italic>p </italic>< 0.001).</p></caption><graphic xlink:href="1465-9921-7-90-1"/></fig></sec><sec><title>SCF protein</title><p>It was detected in every BAL sample taken from LTx recipients. Its concentration remained comparable to controls throughout the study period (70 pg/ml (range: 34–182) during M1-M3, 67 pg/ml (range: 35–144) during M4-M12 and 58 pg/ml (range: 1–90) in controls, NS for all comparisons). In contrast, serum SCF levels remained at all times higher in LTx recipients than in controls (142 pg/ml (range: 104–265) during M1-M3, 193 pg/ml (range: 67–338) during M4-M12 and 53 pg/ml (range: 28–174) in controls, <italic>p </italic>< 0.05 controls vs both post-transplant periods) (Figure <xref ref-type="fig" rid="F2">2</xref>).</p><fig position="float" id="F2"><label>Figure 2</label><caption><p>SCF protein levels in serum. SCF protein levels were assessed by ELISA in serum from control subjects (Control) and lung transplant recipients at Months (Mo) postgraft 1 to 3 (M1-M3), and 4 to 12 (M4-M12). Results are expressed as pg/ml, and represented as individual (symbols) and median values (bold line) (NS; non-significative, ; <italic>p </italic>< 0.05, ; <italic>p </italic>< 0.01, *; <italic>p </italic>< 0.001).</p></caption><graphic xlink:href="1465-9921-7-90-2"/></fig></sec><sec><title>Kit expression</title><p>SCF receptor Kit could be evaluated in 5 of 9 LTx recipients. Figure <xref ref-type="fig" rid="F3">3</xref> shows a marked staining of the mast cells as well as of the bronchial epithelium with a clear immunolocalisation on some basal and some ciliated epithelial cells (Figure <xref ref-type="fig" rid="F3">3</xref>).</p><fig position="float" id="F3"><label>Figure 3</label><caption><p>Photomicrograph of Kit expression in LTx recipients biopsy sections. Kit was immunodetected using a polyclonal rabbit anti-human CD117 (Kit) in bronchial biopsy specimen from LTx recipients in mast cells (arrow head) and epithelial cells (magnitude ×40).</p></caption><graphic xlink:href="1465-9921-7-90-3"/></fig></sec><sec><title>Tryptase-positive cells</title><p>Non degranulated tryptase-positive cells were observed in almost all biopsy specimens from LTx recipients. They were mainly located in the bronchial submucosa but also infiltrated the bronchial epithelium (Figure <xref ref-type="fig" rid="F4">4</xref>). In the early post-transplant period, mast cell counts were not different in patients and controls (36 (range: 0–215) <italic>vs </italic>30 (range: 7–71)/mm<sup>2 </sup>in the M1-M3 period and in controls, respectively; NS). They subsequently increased significantly in LTx recipients: 71 (range 2–224)/mm<sup>2 </sup>during M4-M12 (<italic>p </italic>< 0.05 vs M1-M3 and vs controls). This increase appeared independent of patients' clinical condition (stable condition or acute complications).</p><fig position="float" id="F4"><label>Figure 4</label><caption><p>Tryptase-immunoreactive mast cells in the bronchial biopsy sections from Ltx patients. Mast cell immunolabelling was performed using a mouse monoclonal antibody raised against human tryptase and revealed by FastRed<sup>® </sup>in lung biopsy specimen from lung transplant recipients (magnitude ×40).</p></caption><graphic xlink:href="1465-9921-7-90-4"/></fig></sec></sec><sec><title>Discussion</title><p>Our study shows that SCF was highly expressed in 9 of 9 lung transplant recipients irrespective of their clinical status (infection, rejection). The level of this expression is remarkable, being 300-times greater than in controls in the early post-graft period and albeit decreasing thereafter, remaining at least 10-fold over the control values until the end of the first year post-transplant. To our knowledge, such high levels of SCF transcripts have never been reported so far in human airways. They are striking given the high corticosteroid doses administered to the patients during the early post-transplant period, knowing the usual inhibition of SCF expression by this drug as reported in asthma patients treated with glucocorticoids [<xref ref-type="bibr" rid="B16">16</xref>] as well as in pulmonary fibroblasts <italic>in vitro </italic>[<xref ref-type="bibr" rid="B19">19</xref>]. The only inflammatory airway condition in which SCF expression been reported before is asthma [<xref ref-type="bibr" rid="B16">16</xref>,<xref ref-type="bibr" rid="B20">20</xref>], albeit with levels largely lower than those reported in this study. None of the underlying diseases which indicated pulmonary transplantation in our patients has ever been reported to be associated with SCF airway expression.</p><p>With respect to SCF protein, although it was detected in the BAL fluid of every LTx recipients, it remained at levels comparable with those of controls. In contrast, SCF serum levels were significantly higher in LTx recipients than in controls. One possible explanation for the apparent discrepancy between SCF BAL and serum levels is that inflammatory alveolar cells, the main cell compartment sampled by BAL, are not its main source in the transplanted lung. Although our study does not allow us to draw conclusions about the nature of producing cells, we can hypothesise that bronchial cells are also at least partly involved in this production since all resident lung cells have this ability i.e. normal epithelial cells, smooth muscle cells and fibroblasts [<xref ref-type="bibr" rid="B20">20</xref>-<xref ref-type="bibr" rid="B22">22</xref>].</p><p>The role of SCF in this context can only be speculative. In our small cohort of patients, we found no relationship between mRNA levels and the current clinical situation, particularly acute complications of the graft. It was highly and early expressed even in subjects in stable condition. One hypothesis is that the SCF produced locally might play a role in inflammatory/repair processes following transplantation. Along with this hypothesis, it is of interest to note the expression of the SCF receptor Kit, which we have shown in the bronchial epithelium of LTx recipients. The SCF/Kit pathway, functioning in an autocrine and/or paracrine manner, has clearly been implicated in repair processes taking place after different types of tissue injury in animal models [<xref ref-type="bibr" rid="B11">11</xref>-<xref ref-type="bibr" rid="B13">13</xref>] as well as in humans [<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B15">15</xref>]. Alternatively, SCF having the ability to mobilise stem cells of hematopoietic origin [<xref ref-type="bibr" rid="B10">10</xref>,<xref ref-type="bibr" rid="B23">23</xref>], it could participate in their recruitment to the lung, in keeping with a recent experimental model of lung injury showing a partial pulmonary reconstitution by originally hematopoietic cells [<xref ref-type="bibr" rid="B6">6</xref>]. Indeed, some allografts exhibit a low but consistent chimerism [<xref ref-type="bibr" rid="B24">24</xref>-<xref ref-type="bibr" rid="B26">26</xref>]. This phenomenon has been recently confirmed in human lung allografts [<xref ref-type="bibr" rid="B27">27</xref>,<xref ref-type="bibr" rid="B28">28</xref>]. SCF might be one of the factors involved in the mobilisation and homing of hematopoietic lung progenitors to the injured/repairing zone. On the other hand, SCF is the main chemotactic, activation and growth factor for mast cells [<xref ref-type="bibr" rid="B17">17</xref>,<xref ref-type="bibr" rid="B18">18</xref>]. Of note is the fact that we have shown a certain degree of mast cell infiltration of transplanted airways particularly during the second post-transplant period (M4-M12). This delay might be partly explained by the high doses of corticosteroids included in the early postgraft immunosuppressive regimens, to which mast cells are known to be highly sensitive. In any case, mast cells can exert strong beneficial effects in inflamed organs through the release of a wide range of mediators involved in different aspects of wound healing [<xref ref-type="bibr" rid="B29">29</xref>-<xref ref-type="bibr" rid="B31">31</xref>]. Conversely, they have also been implicated in various fibrotic processes [<xref ref-type="bibr" rid="B32">32</xref>-<xref ref-type="bibr" rid="B35">35</xref>] as well as in chronically rejected lungs [<xref ref-type="bibr" rid="B36">36</xref>,<xref ref-type="bibr" rid="B37">37</xref>].</p></sec><sec><title>Conclusion</title><p>Whether mast cells, which infiltrate the transplanted lung, are innocent bystanders attracted by potent chemotactic factors or are actively recruited for specific purposes is presently unknown. Only the continued follow-up of our patients and the repeated evaluation of these factors over time will tell us about the potential relationship between SCF expression, mast cell presence in the airways and lung transplant outcome.</p></sec><sec><title>Abbreviations</title><p><underline>ALR</underline>: Acute Lung Rejection</p><p><underline>BAL</underline>: BronchoAlveolar Lavage</p><p><underline>BOS</underline>: Bronchiolitis Obliterans Syndrome</p><p><underline>LTx</underline>: Lung Transplant</p><p><underline>SCF</underline>: Stem Cell Factor</p></sec><sec><title>Authors' contributions</title><p><bold>CADS </bold>performed SCF mRNA and protein measurements. Contributed to the writing of the paper. <bold>MA </bold>performed most of the mast cell immunolabeling and counting. Contributed to the writing of the paper. <bold>MS </bold>Head of the department of Pulmonary Medicine of the CMC Foch, where lung transplantation is performed. In charge of the entire clinical management of the patients, including the performance of fiberoptic bronchoscopies and bronchial biopsies. <bold>FdB </bold>Performed the bronchial biopsies. <bold>NF </bold>Head of the research team who performed mRNA and ELISA studies. Co-directed the study. Co-wrote the paper. <bold>DIB </bold>Head of the research team who performed immunohistochemistry. In charge of the immunological monitoring of lung transplant recipients referred above. Co-directed the study. Co-wrote the paper.</p></sec>
RhoA signaling modulates cyclin D1 expression in human lung fibroblasts; implications for idiopathic pulmonary fibrosis	<sec><title>Background</title><p>Idiopathic Pulmonary Fibrosis (IPF) is a debilitating disease characterized by exaggerated extracellular matrix deposition and aggressive lung structural remodeling. Disease pathogenesis is driven by fibroblastic foci formation, consequent on growth factor overexpression and myofibroblast proliferation. We have previously shown that both CTGF overexpression and myofibroblast formation in IPF cell lines are dependent on RhoA signaling. As RhoA-mediated regulation is also involved in cell cycle progression, we hypothesise that this pathway is key to lung fibroblast turnover through modulation of cyclin D1 kinetic expression.</p></sec><sec sec-type="methods"><title>Methods</title><p>Cyclin D1 expression was compared in primary IPF patient-derived fibroblasts and equivalent normal control cells. Quantitative real time PCR was employed to examine relative expression levels of cyclin D1 mRNA; protein expression was confirmed by western blotting. Effects of Rho signaling were investigated using transient transfection of constitutively active and dominant negative RhoA constructs as well as pharmacological inhibitors. Cellular proliferation of lung fibroblasts was determined by BrdU incorporation ELISA. To further explore RhoA regulation of cyclin D1 in lung fibroblasts and associated cell cycle progression, an established Rho inhibitor, Simvastatin, was incorporated in our studies.</p></sec><sec><title>Results</title><p>Cyclin D1 expression was upregulated in IPF compared to normal lung fibroblasts under exponential growth conditions (p < 0.05). Serum deprivation inhibited cyclin D1 expression, which was restored following treatment with fibrogenic growth factors (TGF-β1 and CTGF). RhoA inhibition, using a dominant negative mutant and a pharmacological inhibitor (C3 exotoxin), suppressed levels of cyclin D1 mRNA and protein in IPF fibroblasts, with significant abrogation of cell turnover (p < 0.05). Furthermore, Simvastatin dose-dependently inhibited fibroblast cyclin D1 gene and protein expression, inducing G1 cell cycle arrest. Similar trends were observed in control experiments using normal lung fibroblasts, though exhibited responses were lower in magnitude.</p></sec><sec><title>Conclusion</title><p>These findings report for the first time that cyclin D1 expression is deregulated in IPF through a RhoA dependent mechanism that influences lung fibroblast proliferation. This potentially unravels new molecular targets for future anti-IPF strategies; accordingly, Simvastatin inhibition of Rho-mediated cyclin D1 expression in IPF fibroblasts merits further exploitation.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Watts</surname><given-names>KL</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Cottrell</surname><given-names>E</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Hoban</surname><given-names>PR</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Spiteri</surname><given-names>MA</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	Respiratory Research	<sec><title>Background</title><p>Idiopathic pulmonary fibrosis (IPF) is an insidious fibroproliferative disorder, characterised by interstitial alveolar fibrosis thought to be consequent on aberrant responses to undefined microinsults. Lung injury maybe exacerbated by concurrent failure of re-epithelialisation and excessive fibroblast differentiation [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B2">2</xref>], underpinned by erratic deposition of extracellular matrix (ECM) proteins and progressive lung tissue remodelling. Although a number of scientific advances have been made in understanding disease pathogenesis, no efficacious therapy is available to halt or alter these exaggerated pro-fibrotic processes.</p><p>It follows that IPF pathogenesis must involve aberrations within regulatory pathways critical to the described cellular – biomolecular events. Under such conditions, fibroblasts acquire an aggressive, contractile myofibroblast phenotype, with potent capability for ECM protein production [<xref ref-type="bibr" rid="B3">3</xref>]. Fibroblast-myofibroblast differentiation, is driven by an upregulated pool of growth factors, of which connective tissue growth factor (CTGF) is a key player [<xref ref-type="bibr" rid="B4">4</xref>]. CTGF induction primarily, but not exclusively, is mediated by TGF-β1 through a TGF-β response element in the CTGF promoter [<xref ref-type="bibr" rid="B5">5</xref>]. CTGF modulates IPF fibroblast differentiation through a signalling pathway involving RhoA [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B7">7</xref>]. Interestingly, RhoA is also known to be instrumental in the kinetics of cyclin D1 expression, specifically in G1 phase of the cell cycle [<xref ref-type="bibr" rid="B8">8</xref>]. It follows that as relentless proliferation and differentiation of fibroblasts are crucial to IPF progression, deregulated expression of key cell cycle genes and transcription factors may be pivotal to disease pathogenesis.</p><p>The cell cycle regulator cyclin D1 is a critical factor in the development of proliferative disease [<xref ref-type="bibr" rid="B9">9</xref>], including specific organ oncogenesis [<xref ref-type="bibr" rid="B10">10</xref>-<xref ref-type="bibr" rid="B12">12</xref>]. This 36-kDa protein has a widely accepted role in positive regulation of G1-S progression [<xref ref-type="bibr" rid="B13">13</xref>]. Functioning as a 'mitogenic sensor', in the presence of growth factors, cyclin D1 gene (<italic>CCND1</italic>) drives target cells through the restriction point in the G1 phase of their cycle (thus committing them to cell division). This function is facilitated through binding and activation of cyclin-dependent kinases (CDK) 4 and 6, with phosphorylation of the retinoblastoma protein (Rb), and release of sequestered transcription factors such as E2F [<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B15">15</xref>]. Furthermore, <italic>in vitro </italic>induction of <italic>CCND1 </italic>augments cellular proliferation and transformation of mammalian cells [<xref ref-type="bibr" rid="B16">16</xref>]; which in rodent cells is characterised by a shortened G1 phase with reduced dependence on mitogens [<xref ref-type="bibr" rid="B17">17</xref>].</p><p>A key histological feature of IPF lungs is presence of fibroblast proliferation, with fibroblastic foci formation. We hypothesise that cyclin D1 plays an instrumental role in these pro-fibrogenic processes, augmented by <italic>in situ </italic>growth factor overproduction and exaggerated extracellular matrix deposition [<xref ref-type="bibr" rid="B18">18</xref>]. We contend that cyclin D1 influence in fibroblasts is mediated via a RhoA signalling pathway, especially as RhoA is known to regulate G1 progression of cells [<xref ref-type="bibr" rid="B19">19</xref>]. Accordingly, our study explores for the first time expression levels of cyclin D1 in IPF patient-derived fibroblasts (and equivalent controls) and identifies the influence of Rho, using constitutively active and dominant negative RhoA constructs as well as pharmacological inhibitors, including the agent Simvastatin. This agent selectively blocks a key cascade enzyme, 3-hydroxy-3-methylglutaryl coenzyme A reductase (HMG CoA), inhibiting essential post-translational modification of RhoA, thus inactivating its signalling function.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Human lung fibroblast cell culture</title><p>Three separate human lung fibroblast cell lines isolated from IPF patients (LL29 and LL97a both ATCC, Manassas, USA; and HIPF – a generous gift from R.J. McAnulty, UCL London,) and normal control equivalents (CCD8LU, ATCC, Manassas, USA). The control cell line (CCD8LU) is an adult lung fibroblast cell line, derived from a 48 year old male with cerebral thrombosis, which are a good representative control cell line for analysis of IPF specific effects. All cells were cultured in Dulbecco's modified Eagles medium (DMEM, Sigma Aldrich, Dorset, UK). Media was supplemented with penicillin/streptomycin (100 U/ml) and L-glutamine (2 mM) (both Gibco BRL, Paisley, Scotland) with 10% fetal calf serum (FCS, Labtech, Sussex, UK). All cell lines were cultured and utilized at passages 5–8 to limit passage dependent effects on the observed effects. For experiments, medium was replaced with serum free DMEM (SF-DMEM), for 48 hours to induce quiescence before treatment.</p></sec><sec><title>Treatment with fibrogenic growth factors</title><p>Following serum depravation for 48 hours the fibroblasts were stimulated with fibrogenic growth factors; human recombinant TGF-β1 (R&D systems, Oxford, UK) dose of 1 ng/ml and 5 ng/ml; and human recombinant CTGF (Fibrogen, CA, USA) doses of 10 ng/ml and 100 ng/ml. Fibroblasts were treated with the above-mentioned growth factors for 8 hours for gene expression analysis and 24 hours for protein expression studies. The chosen time points and concentrations of growth factors were determined and established in previous and ongoing studies within our laboratories [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B7">7</xref>].</p></sec><sec><title>C3 exotoxin treatment of lung fibroblasts</title><p>Quiescent lung fibroblasts were incubated overnight (16 hours) with <italic>Clostridium botulinum </italic>C3 exotoxin (Upstate cell signalling solutions, NY, USA) in SF-DMEM. C3 exotoxin was used at concentrations of 0.5 μg/ml, 1 μg/ml and 5 μg/ml; these doses have been previously shown to inhibit Rho signalling pathways in similar fibroblast lines [<xref ref-type="bibr" rid="B6">6</xref>].</p></sec><sec><title>Simvastatin treatment</title><p>Simvastatin is used clinically for the treatment of hypercholesterolaemia due its ability to abrogate the cholesterol synthesis pathway via HMG CoA inhibition. The statins also possess a range of secondary effects arising from disruption of guanosine triphosphatase (GTPase) signalling, including members of the Rho and Ras family. Simvastatin (Merck Sharp and Dohme, Hertfordshire, UK) was dissolved and filter sterilised before use in cell culture studies [<xref ref-type="bibr" rid="B20">20</xref>]. Quiescent lung fibroblasts were then incubated with physiological concentrations of Simvastatin (0.1 μM, 1 μM 10 μM) for 16 hours in serum free cell culture media. Following Simvastatin pre-conditioning, cells were stimulated with human recombinant TGF-β1 (R&D systems, Oxford, UK) at a dose of 5 ng/ml, cells were harvested at 8 hours for mRNA studies and 24 hours for protein analysis.</p></sec><sec><title>Transient transfection of dominant negative/constitutively active RhoA constructs</title><p>Transfection of dominant-negative and constitutively active RhoA (accession number L25080) constructs into human lung fibroblasts (IPF-derived and CCD8LU cells) were performed using Transfast mammalian transfection system (Promega, Southampton, UK). Transfection was performed in lung fibroblasts at 90% confluency following the manufacturer's recommendations. 0.75 μg of DNA was transfected per well (18 mm diameter) using a 1:1 ratio of DNA/Transfast reagent in serum-negative cultures. 90% confluent cells were incubated in the transfection mix containing the RhoA plasmid for 1 hour; DMEM containing 10% FCS was added up to a volume of 1 ml, and cultures were left for 4 hours. Following this, the transfected cells were serum deprived for 48 hours before treatment with TGF-β1 (5 ng/ml) for 8 hours. RhoA G14V (a construct containing a mutation at G14V to render it constitutively active) and RhoA T19N (a construct containing a mutation at T19N, giving it a dominant negative phenotype) constructs were utilized in a cDNA3.1+ vector and were obtained from the Guthrie research institute <ext-link ext-link-type="uri" xlink:href="http://www.cdna.org"/>.</p></sec><sec><title>Real time PCR</title><p>Stored cDNA samples isolated from normal and IPF isolated lung fibroblasts were used to assess CTGF and α-SMA gene expression. 2 μl of undiluted cDNA was used per 25 μl reaction; the primer and probe sets were 'pre-designed assay on demand' probes (Applied Biosystems, Foster City, CA); these pre-designed primers are tested and standardised for reproducible expression analysis. Primer and cDNA were added to the TaqMan universal PCR master mix (Applied Biosystems, Foster City, CA), containing all the reagents for PCR. Absolute quantification of the PCR products was carried out with an ABI prism 7000 (Applied Biosystems, Foster City, CA) utilising the relative standard curve method. cDNA that positively expresses the target gene is used to create a dilution series with arbitrary units. To ensure reproducibility, quantitative data were taken at a point in which each sample was in the exponential phase of amplification. The mean quantity of target gene expression was determined from the generated standard curve; then all samples were normalised against an internal standard β actin or 18s in all quantitative PCR reactions. All data are presented as the fold-change over control in cyclin-D1 gene expression.</p></sec><sec><title>Western blotting</title><p>Total cell proteins were extracted in lysis buffer comprising 1% (v/v) Triton X-100, 20 mM Tris HCL (pH 8.0), 10% (v/v) glycerol, 1 mM sodium orthovanadate, 2 mM EDTA, 1 mM phenylmethylsulfonyl fluoride (PMSF), 20 μM leupeptin and 0.15 U/ml aprotinin. Recovered cells were lysed in above lysis buffer and placed on ice for 20 minutes. The lysates were then centrifuged at 10 000 g, 4°C to pellet cell debris. The supernatant containing the protein was recovered and assayed for total protein using a commercial microplate assay (Bio-Rad, Hemel Hempsted, UK). 25 μg of total protein was combined with sample buffer and boiled prior to gel loading. In addition full-length, recombinant human cyclin D1 protein a 61 Kda tagged fusion protein corresponding to amino acids 1–295 (Santa Cruz Biotechnology, CA, USA) was also loaded onto the gels to ensure detection of the protein of interest. Proteins were resolved on a 12.5% polyacrylamide gel by electrophoresis at 120 V in reducing buffer and transfer was carried out at 100 V. Membranes were blocked with 5% (w/v) BSA in TBS-T buffer overnight. For detection of the cyclin D1 protein DCS-6 (Santa Cruz Biotechnology, CA, USA) antibody was used at 1:100 dilution in TBS-T and 1% BSA. Secondary detection was carried out with horseradish peroxidase-conjugated (HRP) Affinipure goat anti-mouse IgG antibody (Jackson Immunoresearch) at 1:25,000 in TBS-T containing 1% BSA. The cyclin D1 band was visualised by enhanced chemiluminescence (ECL; Amersham Pharmacia Biotech, Buckinghamshire, UK) according to the manufacturer's recommendations and blots were quantified by densitometrical analysis, which involved correcting each blot for background density on each gel. Ponceau S staining of blots after transfer revealed equal loading of total protein; additionally the membranes were reprobed for GAPDH using rabbit polyclonal antibody to GAPDH (1:1000 dilution, Abcam, UK) to ensure equal loading.</p></sec><sec><title>DNA synthesis of proliferating cells</title><p>DNA synthesis was assessed by colorimetric cell proliferation Biotrak ELISA method according to the manufacturer's recommendations (Amersham Biosciences, UK) based on the measurement of 5-bromo-2'-deoxyuridine (BrdU) incorporation during DNA synthesis of proliferating cells. Briefly 30,000 cells were seeded per well of a 96 well plate and left for 24 hours. Cells were then synchronised <italic>in situ </italic>by incubation with serum-depleted media for 48 hours. Cells were then treated with the recognised Rho inhibitor <italic>Clostridium botulinum </italic>C3 exotoxin, (0.5–5 μg/ml) (Upstate cell signalling solutions, Lake Placid, NY) overnight prior to treatment with recombinant human TGF-β1 (5 ng/ml) for up to 5 days. BrdU incorporation was measured daily, during which cells were subjected to BrdU incorporation for 4 hours. The colorimetric change was measured at 450 nm on a Dynatech MR50000 microplate reader (Dynex Laboratories, UK).</p></sec><sec><title>FACS analysis</title><p>LL97a lung fibroblasts were grown to approximately 60% confluency prior to serum deprivation for 48 hours (this ensures the cells become quiescent and are synchronised in the cell cycle). The lung fibroblasts were then treated, accordingly with Simvastatin (0.1 μg/ml or 10 μg/ml) with or without TGF-β1 (5 ng/ml) for 24 hours. The cells were then harvested and the cell suspension fixed in 70% ice-cold ethanol. The cell suspension was centrifuged at 200 rpm and the cell pellet resuspended in PBS. RNase (1 mg/ml) and propidium iodide (0.5 mg/ml) were added and incubated for 30 minutes at 37°C. To ensure no clumping of the cells the suspension was passed through a 25 g needle. The cells were analysed on a MOFLO cell sorter (Dakocytomation, Glostrup, Denmark) at a wavelength of 488 nm and speed of 100 events per second (eps). A minimum of 20,000 events per data profile was collected.</p></sec><sec><title>Statistical analysis</title><p>Data are shown as a mean ± SEM. An unpaired student's t test was employed for comparing 2 groups of data. Multiple comparisons were made using analysis of variance (ANOVA) followed by Tukeys pairwise comparison. All p values < 0.05 were considered significant.</p></sec></sec><sec><title>Results</title><sec><title>Cyclin D1 gene expression is upregulated in IPF fibroblasts</title><p>The expression of the cyclin D1 gene was quantified in 3 IPF-derived lung fibroblast cell lines (HIPF, LL29, LL97a) and the adult normal lung fibroblast cell line CCD8LU using a real time PCR approach (Fig <xref ref-type="fig" rid="F1">1</xref>). Under exponential growth conditions (cells grown in 10% FCS, i.e. actively dividing cells), IPF-derived lung fibroblasts demonstrated a 4.72 to 11.29 fold elevation of cyclin D1 mRNA expression (average of 10.10 fold increase) compared to the CCD8LU normal lung fibroblast cell line (p < 0.05). We compared these data to A431 cells, a human epithelial squamous carcinoma cell line with a known 5 fold amplification of cyclin D1 [<xref ref-type="bibr" rid="B21">21</xref>]; the IPF fibroblast cell lines studies significantly exceeded the amplified cyclin D1 mRNA expression of A431 by an average of 2.45 fold.</p><fig position="float" id="F1"><label>Figure 1</label><caption><p><bold>Expression levels of cyclin D1 mRNA in human lung fibroblasts during exponential growth</bold>. Quantitative real-time RT-PCR was performed on three separate human lung fibroblast cell lines from IPF patients (HIPF, LL29, and LL97a) and normal control equivalents CCD8LU. Quantification of mRNA was performed by determining the threshold cycle; and standard curves were constructed using the values obtained from serially diluted positively expressing human cDNA. All cells were under conditions of exponential growth (10% FCS supplemented media). 3 PCR reactions were performed from 3 independent cell culture experiments, graph represents mean cyclin D1 expression ± S.E.M; * = p < 0.05.</p></caption><graphic xlink:href="1465-9921-7-88-1"/></fig></sec><sec><title>Cyclin D1 gene and protein levels are augmented by growth factors</title><p>Cyclin D1 gene expression was measured in the normal lung fibroblasts and the 3 IPF-derived lung fibroblast cell lines following growth factor treatment (Fig <xref ref-type="fig" rid="F2">2a</xref>), Cells were serum deprived for 48 hours to ensure quiescence and to synchronise cell proliferation; cells were then exposed to physiologically relevant concentrations of growth factors (CTGF and TGF-β1) known to be implicated in IPF pathogenesis [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B5">5</xref>]. Serum deprivation inhibited cyclin D1 expression (as expected); however expression was restored upon treatment with recombinant growth factors. Cyclin D1 augmentation was more pronounced in the IPF-derived lung fibroblasts, especially in the presence of TGF-β1 (1 ng/ml and 5 ng/ml) and CTGF (10 ng/ml) (p < 0.05). Interestingly, in cultures containing the higher concentration of CTGF (100 ng/ml), we observed fibroblast apoptosis especially in IPF-related cell line (data not shown); and no further increase in cyclin D1 expression. It is also of interest to note that in the absence of mitogens the levels of cyclin D1 mRNA are not significantly different between the cell lines studied and expression lies are within are narrow range (2.10–6.65 × 10<sup>-4</sup>).</p><fig position="float" id="F2"><label>Figure 2</label><caption><p><bold>Expression of cyclin D1 mRNA and protein in human lung fibroblasts; response to fibrogenic growth factors</bold>. Figure 2a: Cyclin D1 mRNA levels were determined by quantitative real-time PCR on three separate human lung fibroblast cell lines from IPF patients (HIPF, LL29, and LL97a) and normal control equivalents CCD8LU exposed to fibrogenic mediators TGF-β1 (5 ng/ml and 10 ng/ml) and CTGF (10 ng/ml and 100 ng/ml) for 8 hrs. Data shown demonstrates analysis from LL97a and CCD8LU fibroblasts, no significant difference was observed in baseline cyclin D1 expression between the cell lines. Data are representative of 3 independent experiments, within each of which PCRs were performed in triplicate. Data represents mean cyclin D1 expression ± S.E.M; * = p < 0.05 compared to serum free, † = p < 0.05 compared to normal lung fibroblasts. Figure 2b: Quantification of cyclin D1 protein expression was performed by western blotting in all three IPF fibroblast lines and normal equivalents. Quiescent serum deprived lung fibroblasts were stimulated with fibrogenic growth factors for 24 hours. Cyclin D1 protein levels found in 25 μg of total protein from normal and IPF derived lung fibroblasts was determined by western blotting. Data shown demonstrates analysis from LL97a and CCD8LU fibroblasts. Data are representative of 3 independent westerns and represented as mean density ± SEM. * = p < 0.05 compared to normal lung fibroblasts. Figure 2c: Representative western blots for cyclin D1 and GAPDH protein expression in a representative IPF lung fibroblast cell line (LL97a). (i) cyclin D1 blot-lane 1 = marker lane 2 = control (serum deprived); lane 3 = 10% FCS; lane 4 = TGF-β1 1 ng/ml; lane 5 = TGF-β1 5 ng/ml; lane 6 = CTGF 10 ng/ml; lane 6 = CTGF 100 ng/ml. (ii) GAPDH blot-lane 1 = control (serum deprived); lane 2 = 10% FCS; lane 3 = TGF-β1 1 ng/ml; lane 4 = TGF-β1 5 ng/ml; lane 5 = marker; lane 6 = CTGF 10 ng/ml; lane 6 = CTGF 100 ng/ml. Ponceau S staining of blots after transfer revealed equivalent loading of total protein.</p></caption><graphic xlink:href="1465-9921-7-88-2"/></fig><p>To further confirm above findings, cyclin D1 protein expression was determined in the same cell lines under the same experimental conditions using western blot analysis (Fig <xref ref-type="fig" rid="F2">2b</xref>). We observed the same patterns of expression and induction in cyclin D1 protein, reflecting results of cyclin mRNA expression. A representative blot from one of the IPF derived cell lines is shown in Fig <xref ref-type="fig" rid="F2">2c</xref>. Data shown in fig <xref ref-type="fig" rid="F2">2a</xref>, <xref ref-type="fig" rid="F2">2b</xref> and <xref ref-type="fig" rid="F2">2c</xref> is taken from the patient cell line LL97a; these mRNA and protein data reflect results obtained with the other 2 IPF fibroblast cell lines studied.</p></sec><sec><title>RhoA modulates cyclin D1 gene and protein levels in lung fibroblasts</title><p>Transient transfection of dominant-negative RhoA (RhoA T19N) and constitutively active (RhoA G14V) constructs were utilised to confirm the regulatory role of Rho in cyclin D1 induction (Fig <xref ref-type="fig" rid="F3">3a</xref>). These data revealed that cyclin D1 mRNA expression levels are of comparable magnitude between cells stimulated with TGF-β1 (5 ng/ml) alone compared to those expressing constitutively active RhoA (G14V RhoA transfected). When G14V active, RhoA cultures were subsequently conditioned with TGF-β1, significant upregulation (p < 0.05) in cyclin D1 gene expression was observed in LL97a; this trend was replicated in the other 2 IPF derived cell lines studied. These data support a role for Rho in cyclin D1 induction; which is further confirmed by the use of a dominant-negative RhoA construct. Transfection with Rho T19N induced significant reduction (p < 0.05) in cyclin D1 gene expression producing a 25.5% and a 33% reduction in normal and IPF derived lung fibroblasts respectively compared to cells treated with 5 ng/ml TGF-β1 alone, further supporting involvement of RhoA in cyclin D1 expression. In our experiments, we did not observe complete inhibition of cyclin D1 gene, as expected of the transient transfection method used. As the average transfection efficiency achieved was about 40%, thus a proportion of the cells in our culture will not have inhibited RhoA signalling. The above result trends were consistent throughout the 3 IPF cell lines analysed.</p><fig position="float" id="F3"><label>Figure 3</label><caption><p><bold>RhoA signalling directly influences cyclin D1 mRNA and protein expression in lung fibroblasts</bold>. Fig 3a. Human lung fibroblasts (3 IPF fibroblast lines and normal equivalents) were transfected with RhoA T19N (dominant negative) and RhoA G14V (constitutively active) constructs. Cells were serum deprived for 48 hours post transfection before incubation with TGF-β1 (5 ng/ml) for 8 hours. Data shown demonstrates analysis from LL97a and CCD8LU fibroblasts. Data are representative of transfection performed in triplicate from three independent experiments. Data are expressed as mean fold change in cyclin D1 transcript ± SEM. * = p < 0.05 relative to control, † = p < 0.05 relative to normal lung fibroblasts, †† = p < 0.05 relative to TGF-β1 treated fibroblasts. Fig 3b. Human lung fibroblasts (normal CCD8LU and IPF-derived HIPF, LL29 and LL97a) were treated with 0.5 μg/ml and 5 μg/ml of C3 exotoxin with or without subsequent TGF-β1 (5 ng/ml) stimulation. The control shown represents fibroblasts not exposed to C3 exotoxin and/or TGF-β1. Cyclin D1 protein levels found in 25 μg of total protein was determined by western blotting. Data shown demonstrates analysis from LL97a and CCD8LU fibroblasts. Data is representative of westerns performed in triplicate and shown as mean density ± SEM, * = p < 0.05.</p></caption><graphic xlink:href="1465-9921-7-88-3"/></fig><p>To confirm above findings, cyclin D1 protein expression was analysed following pharmacological inhibition of RhoA utilising C3 exotoxin (a recognised inhibitor of RhoA) (Fig <xref ref-type="fig" rid="F3">3b</xref>). Compared to TGF-β1 treatment alone (5 ng/ml), both test concentrations of C3 exotoxin significantly (p < 0.05) abrogated cyclin D1 protein expression in both normal and IPF lung fibroblasts, irrespective of subsequent TGF-β1 exposure.</p></sec><sec><title>DNA synthesis is suppressed by RhoA inhibition</title><p>We used a sensitive BrdU incorporation ELISA that measures DNA synthesis to determine if Rho inhibition would alter cell proliferative capability. Firstly DNA synthesis in response to growth factor treatment was determined (Fig <xref ref-type="fig" rid="F4">4a</xref>). Actively dividing cells (cultured in media supplemented with 10% FCS) had the fastest DNA synthesis rate. As expected, which was almost halted in serum-deprived (quiescent) cells; but exhibited some restoration over the 120-hour time course on exposure to fibrogenic factors TGF-β1 (5 ng/ml) and CTGF (10 ng/ml). At the end time point (120 hr), TGF-β1 and CTGF induced a 44.85% and 36.88% increase respectively (p < 0.05) in BrdU incorporation IPF fibroblasts compared to serum-starved equivalent controls. This data is representative of all 3 IPF cell lines studied. Similar trends are replicated in the normal fibroblast equivalents although the magnitude of BrdU incorporation was approximately 3-fold lower in the controls (data not shown).</p><fig position="float" id="F4"><label>Figure 4</label><caption><p><bold>Cell proliferation is enhanced in IPF lung fibroblasts and can be abrogated by RhoA inhibition</bold>. 4a. Cell proliferation was determined by incorporation of BrdU in three separate human lung fibroblast cell lines from IPF patients (HIPF, LL29, and LL97a) and normal control equivalents CCD8LU data represents analysis in the IPF lung fibroblast cell line LL97a. Cells were subjected to BrdU incorporation at 24-hour intervals as described. Data are representative of the mean of 3 independent experiments (standard error bars have been omitted to simplify the figure). * = p < 0.05 significant elevation relative to serum free controls. 4b. Cell proliferation in response to the recognised Rho inhibitor C3 exotoxin (0.5–5 μg/ml) with or without subsequent TGF-β1 (5 ng/ml) stimulation was measured by BrdU incorporation in three separate human lung fibroblast cell lines from IPF patients (HIPF, LL29, and LL97a) and normal control equivalents CCD8LU. Data shows analysis in the IPF cell line LL97a. Data are representative of the mean of 3 independent experiments (standard error bars have been omitted to simplify the figure). * = p < 0.05 relative to TGF-β1 stimulated cells.</p></caption><graphic xlink:href="1465-9921-7-88-4"/></fig><p>Involvement of RhoA in above DNA synthesis was determined using C3 exotoxin, which specifically ADP ribosylates and inactivates Rho. The inhibitor was used at optimal concentrations of 0.5 μg/ml and 5 μg/ml with or without additional TGF-β1 (5 ng/ml) stimulation (Fig <xref ref-type="fig" rid="F4">4b</xref>). Exposure to C3 exotoxin inhibited BrdU incorporation, even in the presence of TGF-β1 at both 0.5 and 5 μg/ml concentrations. Both C3 exotoxin treatments suppressed DNA synthesis over the 120 hour time course compared to control IPF lung fibroblasts treated with 5 ng/ml TGF-β1 alone; becoming significant at p < 0.05 over time from 72 hours onwards.</p></sec><sec><title>Simvastatin inhibits fibroblast cyclin D1 expression via a Rho signalling pathway</title><p>The effect of cell pre-incubation with varying concentrations of Simvastatin (0.1 μM, 1 μM and 10 μM) on cyclin D1 gene expression in the IPF derived lung fibroblast cell lines and equivalent normal controls was analysed by real time PCR (Fig <xref ref-type="fig" rid="F5">5</xref>). The physiological concentrations of Simvastatin used abrogated cyclin D1 gene expression, irrespective of TGF-β1 presence (p < 0.05). Although 0.1 μM Simvastatin had little effect on cyclin D1 expression, 10 μM Simvastatin was efficacious enough to reduce even basal levels of cyclin D1 mRNA in test fibroblasts, inducing a 1.66 fold and 2.1 fold respective inhibition of the gene compared to untreated cells and TGF-β1-lone treated cells respectively. Furthermore the inhibition of cyclin D1 was further confirmed at the protein level by western blotting (data not shown). The data in fig <xref ref-type="fig" rid="F5">5</xref> is from LL97a IPF derived lung fibroblasts; these data are also representative of the other 2 IPF lung fibroblast cell lines studied. Again trends were replicated within the normal fibroblast equivalents but with a lower magnitude of cyclin D1 expression compared to patient samples.</p><fig position="float" id="F5"><label>Figure 5</label><caption><p><bold>Simvastatin abrogates cyclin D1 gene expression levels in a dose dependent fashion in human lung fibroblasts</bold>. Serum deprived cells were incubated with Simvastatin (Sim 0.1–10 μM) for 16 hours. Subsequent TGF-β1 stimulation (5 ng/ml) was carried out for 8 hours. Experiments were performed in three separate human lung fibroblast cell lines from IPF patients (HIPF, LL29, and LL97a) and normal control equivalents CCD8LU. The control shown represents fibroblasts not exposed to Simvastatin and/or TGF-β1. The gene expression of cyclin D1 was then determined by real time PCR. Data shown demonstrates analysis from LL97a and CCD8LU fibroblasts. Data is representative of the mean of triplicate PCRs obtained from 3 independent experiments. Data are expressed as the mean fold change in cyclin D1 expression ± SEM. * = p < 0.05 compared to control untreated, † = p < 0.05 compared to TGF-β1 treatment.</p></caption><graphic xlink:href="1465-9921-7-88-5"/></fig></sec><sec><title>Simvastatin induces G1 arrest in IPF lung fibroblasts</title><p>To explore the influence, as yet unrecognised, of Simvastatin on IPF lung fibroblast proliferation, we analysed DNA content in Simvastatin-treated patient-derived lung fibroblasts (LL97a) using FACS analysis of propidium iodide stained of cells (Fig <xref ref-type="fig" rid="F6">6</xref>). Fibroblasts grown in DMEM containing 10% FCS (6a) showed their progression through the cell cycle; whereas serum deprivation limited G1 progression and entry in S phase by 51% (6b). Compared to serum-depleted samples, cells incubated in 5 ng/ml TGF-β1 (6c) presented a profile similar to that of cells grown in 10% FCS; with 5.04% of cells entering G1 phase and 9.73% of cells in S phase transition. Further analysis revealed that fibroblasts were G1 arrested following treatment with Simvastatin; small responses were observed at a dose of 0.1 μM (6d) and a more pronounced response is seen at the higher concentration of 10 μm (6e), irrespective of TGF-β1 treatment (5 ng/ml). Such cells were prevented from entering S phase of the cell cycle, thus reducing the percentage of cells in G2 phase of the cell cycle by 40.8% and 76.2% respectively. These findings are summarised in Fig <xref ref-type="fig" rid="F7">7</xref> where Simvastatin is observed to induce a decrease in the percentage number of fibroblasts in G2 phase of the cell cycle with concurrent increase in cells remaining within G1 phase of the cell cycle.</p><fig position="float" id="F6"><label>Figure 6</label><caption><p><bold>Simvastatin influences cell cycle progression in human lung fibroblasts by inducing G1 arrest</bold>. LL97a lung fibroblasts at 60% confluency were serum deprived for 48 hours; cells were then harvested following treatment (6a) DMEM containing 10% FCS (exponential growth) (6b) serum free (quiescent cells) (6c) TGF-β1 alone (5 ng/ml) (6d) Simvastatin (10 μM) alone (6e) Simvastatin (10 μM) and TGF-β1 (5 ng/ml) stimulation. FACS sorting was used to assess cell cycle progression using propidium iodide staining of cellular DNA content. Data are representative of FACS analysis performed in triplicate.</p></caption><graphic xlink:href="1465-9921-7-88-6"/></fig><fig position="float" id="F7"><label>Figure 7</label><caption><p><bold>Simvastatin modulation of cell cycle progression as determined by FACS analysis</bold>. Data is summarised from the same treatments and FACS data from Fig 6 in LL97a lung fibroblasts. The % number of cells present in G1, S and G2 phase of the cell cycle are presented ± SEM and is representative of 3 independent experiments. * = p < 0.05 compared to serum free control, † = p < 0.05 compared to 10% FCS treatment.</p></caption><graphic xlink:href="1465-9921-7-88-7"/></fig></sec></sec><sec><title>Discussion</title><p>Cyclin D1 is a critical regulator in progression of the cell cycle, specifically passage through the G1 phase and entry into S phase, beyond which cells are committed to mitosis. <italic>CCND1 </italic>is a recognised oncogene; thus, when <italic>CCND1 </italic>is over-expressed pathologically such as in oncogenesis, affected cells enter S phase more rapidly resulting in accelerated speed and frequency of proliferation [<xref ref-type="bibr" rid="B22">22</xref>]. There is increasing evidence that Rho family members promote cell cycle progression by regulating cyclin D1 and associated genes such as p21cip1, p27kip1 [<xref ref-type="bibr" rid="B23">23</xref>]. We have previously demonstrated that Rho is a key driver in fibroblast-mediated growth factor expression and myofibroblast formation [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B7">7</xref>]. In this study we have explored the role of cyclin D1 and interaction with RhoA signalling to determine key influences in observed fibroblast over-proliferation in IPF.</p><p>Our study data demonstrate for the first time that cyclin D1 gene and protein are upregulated in IPF-derived lung fibroblasts under basal proliferating conditions (media supplemented with 10% FCS). Indeed, levels of cyclin D1 mRNA expression greatly exceed those of the control cell line A431 that has a known 5-fold amplification of the gene [<xref ref-type="bibr" rid="B21">21</xref>]. The reason for the observed elevated levels of cyclin D1 in IPF cells lines is as yet unknown and will be addressed in separate lung tissue studies; however possibilities include amplification of gene copy number, hyper-stimulation of the RhoA pathway through an aberrant disease-associated mutation (<italic>or </italic>pathogenic mutation causing abrogation of pathway inhibitors) or simply, factor/s within the profibrogenic milieu. Nonetheless, the findings to date support our hypothesis that cyclin D1 deregulation could explain exaggerated fibroblast proliferation observed in IPF lungs, and possibly propagate, albeit partly, associated formation of fibroblastic foci. Interestingly, we observed that specific pro-fibrogenic growth factors, known to be associated with IPF pathogenesis [<xref ref-type="bibr" rid="B5">5</xref>], can induce cyclin D1 expression in serum-deprived fibroblasts. Cells treated with TGF-β1 show gene upregulation at both 1 ng/ml and 5 ng/ml, with the greatest response seen at the higher dose. CTGF at 10 ng/ml also induced cyclin D1 mRNA; however this trend was not replicated at the higher dose of 100 ng/ml in IPF fibroblasts. This result could be explained by CTGF-induced cell apoptosis in these cells at high concentrations [<xref ref-type="bibr" rid="B24">24</xref>].</p><p>We also believe that the growth factor effect on cyclin D1 expression in fibroblasts is not only dependent on the concentration of the particular mediator, but may also be factor-specific. Preliminary data in our laboratory reveals that another known pro-fibrogenic mediator, thrombin (1 ng/ml and 2.5 ng/ml) only induces small, insignificant responses in same fibroblast cyclin D1 expression. Thus not all fibrogenic growth factors have similar effects on <italic>CCND1 </italic>expression profiles; known differential effects of the test growth factors on the Rho signalling pathway may explain such discrepancy. Specifically, TGF-β1 and CTGF act via a Rho signalling pathway to induce changes in cyclin D1. However, thrombin has recently been shown to suppress RhoA activity by inducing tyrosine phosphorylation coinciding with a decrease in Rho activity [<xref ref-type="bibr" rid="B25">25</xref>]; accounting for its limited observed response on fibroblast cyclin D1 expression (in-house data).</p><p>Taken together, these observations support a crucial function for RhoA signalling in cyclin D1 expression in IPF lung fibroblasts, with consequence on their proliferative activity. We have demonstrated that inhibition of RhoA signalling (using both dominant negative transfection and pharmacological inhibitors) downregulates cyclin D1 expression in lung fibroblasts, reflected functionally, albeit indirectly, by altered cell turnover. There is evidence that there are 2 opposing mechanisms for Rho mediated control of cyclin D1; a stimulatory axis mediated through ERK signaling and a concurrent inhibitory axis acting through Rac/cdc42 [<xref ref-type="bibr" rid="B8">8</xref>]. These 2 mechanisms may account for some of the findings in this manuscript. We observe that constitutively active RhoA (G14V) augments cyclin D1 expression, however in separate experiments we also show that C3 exotoxin a Rho inhibitor is also able to increases cyclin D1 expression; thus suggesting that these 2 pathways may be active in the lung fibroblasts studied. Further experiments are needed to further identify the presence and role of ERK and Rac/cdc42 dependent pathways in relation to lung fibroblasts and IPF mechanisms. Also of interest is that the constitutively active RhoA construct (G14V) in the presence of TGFβ1 (5 ng/ml) is able to further elevate cyclin D1 mRNA expression in the IPF cell line with only little or no further effect in the control fibroblasts. Thus this may highlight a deregulated mechanism specific to the IPF cohort and thus present a suitable target for therapeutic intervention. We feel that this observation may be related to deregulation of pathways involved in suppression of cytokine signalling (SOCS) genes, which may increase IPF fibroblasts susceptibility to growth factors such as TGFβ1. This is a potential mechanism that has be highlighted in liver fibrosis [<xref ref-type="bibr" rid="B29">29</xref>] and emerging findings from our own experiments support the concept of deregulated SOCS 3 expression in IPF lung fibroblasts (in house data).</p><p>Experiments using the specific HMG CoA inhibitor agent, Simvastatin also support the concept that RhoA modulates cyclinD1 expression. Interestingly such statin agents possess increasingly recognised pleiotropic effects beyond that of cholesterol lowering, including CTGF inhibition, preventing myofibroblast formation and anti-fibrotic effects in kidney disease and heart disease [<xref ref-type="bibr" rid="B26">26</xref>,<xref ref-type="bibr" rid="B27">27</xref>]. These additional effects are due to Simvastatin's ability to modulate RhoA signalling; occurring as a result of inhibited post-translational modification of the RhoA molecule (a pre-requisite for its activation). Using Simvastatin we achieved abrogation of cyclin D1 mRNA and protein expression in a concentration dependent manner, irrespective of TGF-β1 conditioning. Simvastatin treatment was able to lower IPF fibroblast cyclin D1 levels to basal expression of normal cells. Functionally, Simvastatin also induced G1 arrest in the IPF fibroblasts, again overriding inductive effects of TGF-β1, resulting in suppressed cell proliferation. An alternative mechanism for the observed changes in cell cycle progression and cyclin D1 expression is Simvastatin-mediated disruption of lipid raft localisation. The lipid rafts are essential for efficient signal transduction by a number of cell types including B and T cells [<xref ref-type="bibr" rid="B28">28</xref>] resulting in altered growth factor and GTPase signalling such as Ras. However our data is consistent with Rho being the central mechanism for CCND1 disruption as the specific Rho inhibitor C3 exotoxin is able to influence expression, in addition we have preliminary data (in house data) in which we have utilised Simvastatin to inhibit GTPase activity, Rho activity can be restored by introducing geranylgeranylpyrophosphate (GGPP) with associated augmented cyclin D1 and growth factor expression. However restoring Ras activity by the incorporation of farnesylpyrophospahe (FPP) is unable to have the same effects and expression of cyclinD1 and other key growth factors is not returned. These observations may suggest that selective inhibitory manipulation of Rho signalling pathway components could be exploited to attempt therapeutic reversal of the fibroproliferative processes associated IPF.</p></sec><sec><title>Conclusion</title><p>Our studies further enhance understanding of the pathogenic events within IPF lungs, highlighting fibroblast cell cycle deregulation via a cyclin D1 mechanism as a key factor in disease progression. Tentatively, we provide evidence to support future exploitation of direct RhoA inhibition (using HMG CoA inhibitor agents) as a novel strategic option for fibroproliferative abrogation in lung fibrosis.</p></sec><sec><title>Abbreviations</title><p>α-Smooth Muscle Actin α-SMA</p><p>5-bromo-2'-deoxyuridine BrdU</p><p>cyclin D1 gene CCND1</p><p>Connective Tissue Growth Factor CTGF</p><p>Extracellular Matrix ECM</p><p>Fetal Calf Serum FCS</p><p>Fluorescence Activated Cell Sorting FACS</p><p>Farnesylpyrophosphate FPP</p><p>Geranylgeranylpyrophosphate GGPP</p><p>Glyceraldehyde-3-phosphate dehydrogenase GAPDH</p><p>Guanine nucleotide-binding regulatory protein G protein</p><p>Guanosine triphosphatase GTPase</p><p>3 hydroxy3methylglutaryl Coenzyme A HMG CoA</p><p>Idiopathic Pulmonary Fibrosis IPF</p><p>Phosphate buffered saline PBS</p><p>Reverse Transcription Polymerase Chain Reaction RT-PCR</p><p>Serum-free DMEM media SF-DMEM</p><p>Suppressor of cytokine Signalling SOCS</p><p>Transforming Growth Factor-β1 TGF-β1</p></sec><sec><title>Competing interests</title><p>None of the authors are aware of any competing interests regarding submission/publication of this manuscript.</p></sec><sec><title>Authors' contributions</title><p><italic>KW </italic>has worked full time as a post-doctoral researcher on this project (funded by the British Lung Foundation) including its design, experimental work and data analysis; she has led production of this manuscript.</p><p><italic>EC </italic>worked as a project student on the study under the guidance of KW and PH. EW helped perform the Simvastatin experiments and subsequent analysis that appears in Fig <xref ref-type="fig" rid="F5">5</xref>.</p><p><italic>PH </italic>has given guidance to KW on experimental design and has helped in manuscript preparation.</p><p><italic>MS </italic>is director of the lung fibrosis programme, closely supervising and advising KW; and has extensively revised manuscript drafts.</p></sec>
Cationic polyamines inhibit anthrax lethal factor protease	<sec><title>Background</title><p>Anthrax is a human disease that results from infection by the bacteria, <italic>Bacillus anthracis </italic>and has recently been used as a bioterrorist agent. Historically, this disease was associated with <italic>Bacillus </italic>spore exposure from wool or animal carcasses. While current vaccine approaches (targeted against the protective antigen) are effective for prophylaxis, multiple doses must be injected. Common antibiotics that block the germination process are effective but must be administered early in the infection cycle. In addition, new therapeutics are needed to specifically target the proteolytic activity of lethal factor (LF) associated with this bacterial infection.</p></sec><sec><title>Results</title><p>Using a fluorescence-based assay to identify and characterize inhibitors of anthrax lethal factor protease activity, we identified several chemically-distinct classes of inhibitory molecules including polyamines, aminoglycosides and cationic peptides. In these studies, spermine was demonstrated for the first time to inhibit anthrax LF with a K<sub>i </sub>value of 0.9 ± 0.09 μM (mean ± SEM; n = 3). Additional linear polyamines were also active as LF inhibitors with lower potencies.</p></sec><sec><title>Conclusion</title><p>Based upon the studies reported herein, we chose linear polyamines related to spermine as potential lead optimization candidates and additional testing in cell-based models where cell penetration could be studied. During our screening process, we reproducibly demonstrated that the potencies of certain compounds, including neomycin but not neamine or spermine, were different depending upon the presence or absence of nucleic acids. Differential sensitivity to the presence/absence of nucleic acids may be an additional point to consider when comparing various classes of active compounds for lead optimization.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Goldman</surname><given-names>Mark Evan</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Cregar</surname><given-names>Lynne</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Nguyen</surname><given-names>Dominique</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Simo</surname><given-names>Ondrej</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>O'Malley</surname><given-names>Sean</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Humphreys</surname><given-names>Tom</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Pharmacology	<sec><title>Background</title><p>Anthrax is a disease of animals including humans and results from infection by <italic>Bacillus (B.) anthracis </italic>[<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B3">3</xref>]. Anthrax spores reside in soil samples worldwide and are resistant to environmental insults such as temperature, moisture and UV irradiation. Spores enter host animals via inhalation, epidermal or gastrointestinal routes with respiratory route being the most fatal. Once inside hosts, the spores germinate and secrete three toxin components, called lethal factor (LF), edema factor (EF) and protective antigen (PA), encoded by the pXO1 plasmid. LF, a metalloprotease, plus PA is termed lethal toxin and EF plus PA is termed edema toxin [<xref ref-type="bibr" rid="B3">3</xref>]. The combination of bacteremia and release of the protein toxins leads to sepsis, pulmonary edema and other fatal effects [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B5">5</xref>].</p><p>PA is responsible for translocating the two other gene products, LF and EF, into the cytosol of susceptible cells [<xref ref-type="bibr" rid="B6">6</xref>-<xref ref-type="bibr" rid="B9">9</xref>]. The precursor form of PA (PA83) binds to ubiquitous cell surface receptors including von Willebrand factor, tumor endothelial marker 8 (TEM8) and capillary morphogenesis protein 2 [<xref ref-type="bibr" rid="B10">10</xref>,<xref ref-type="bibr" rid="B11">11</xref>]. PA83 is cleaved by furin as well as by serum proteases [<xref ref-type="bibr" rid="B12">12</xref>-<xref ref-type="bibr" rid="B14">14</xref>]. The active form of PA (PA63) then heptamerizes and binds with a high affinity to LF or EF [<xref ref-type="bibr" rid="B3">3</xref>]. The complex of PA with LF or EF forms a channel to allow LF/EF to translocate from the endosome to the cytosol where the toxic effects associated with LF are manifest [<xref ref-type="bibr" rid="B3">3</xref>]. Cationic peptides that inhibit furin-mediated activation of PA83 to PA63 are also effective in blocking lethal toxin cytotoxicity [<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B16">16</xref>,<xref ref-type="bibr" rid="B17">17</xref>].</p><p>EF is a calmodulin-dependent adenylate cyclase and thus elevates intracellular cAMP levels of intoxicated cells [<xref ref-type="bibr" rid="B18">18</xref>]. As a result of this mechanism, EF causes the additional pathological effects in the host although it is less virulent than LF. Recent studies have demonstrated that adefovir diphosphate is a potent inhibitor of anthrax EF, <italic>in vitro </italic>[<xref ref-type="bibr" rid="B19">19</xref>].</p><p>Anthrax LF is a representative member of the zinc-dependent endopeptidases family as demonstrated by the presence of the HEXXH zinc-binding consensus sequence [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B15">15</xref>]. LF, an 89 kD protein, is one of the main virulence factors of anthrax [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B15">15</xref>]. LF contains numerous anionic sites both within the active site and at distant sites [<xref ref-type="bibr" rid="B20">20</xref>-<xref ref-type="bibr" rid="B22">22</xref>].</p><p>Macrophages are target cells of LF toxicity in animal model systems. Exposure of murine macrophages to lethal toxin resulted in rapid loss of cell viability [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B23">23</xref>,<xref ref-type="bibr" rid="B24">24</xref>]. Conversely, mice depleted of macrophages were not sensitive to lethal toxin [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B3">3</xref>]. The mechanism of lethality has been attributed to release of cytokines or apoptosis as well as other mechanisms but is highly dependent upon the LF concentration [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B5">5</xref>,<xref ref-type="bibr" rid="B26">26</xref>-<xref ref-type="bibr" rid="B29">29</xref>].</p><p>The mitogen activated protein kinase/extracellular signal-regulated protein kinase (MAPK/ERK) pathway is a major regulator for communication of extracellular signals to the nucleus and is involved in cellular adaptations to the environment [<xref ref-type="bibr" rid="B30">30</xref>,<xref ref-type="bibr" rid="B31">31</xref>]. In the cytosol, LF cleaves members of the mitogen activated protein kinase kinase (MAPKK) family in the N-terminal region including MAPKK family members 1–3 [<xref ref-type="bibr" rid="B31">31</xref>,<xref ref-type="bibr" rid="B32">32</xref>]. The reduced levels of MAPKKs then prevent p38 kinase-mediated activation of immune mechanisms <italic>B. anthracis </italic>to evade host immunological mechanisms. Recently published studies have demonstrated that small molecules can inhibit LF activity and subsequently block LF-mediated cytotoxicity [<xref ref-type="bibr" rid="B21">21</xref>,<xref ref-type="bibr" rid="B29">29</xref>,<xref ref-type="bibr" rid="B33">33</xref>-<xref ref-type="bibr" rid="B36">36</xref>].</p><p>At present, the only mechanism to fatally "intoxicate" cells with lethal factor is via host infection with <italic>B. anthracis </italic>spores that germinate in lymphatic tissues and secrete their toxin components. None of the three individual gene products of pXO1 are toxic <italic>in vivo </italic>[<xref ref-type="bibr" rid="B37">37</xref>]. Inhibitors of other proteases such as angiotensin converting enzyme and HIV-1/HIV-2 proteases are effective and highly specific drugs for the treatment of chronic diseases [<xref ref-type="bibr" rid="B38">38</xref>] and therefore suggest a logical strategy for identifying anthrax lethal factor inhibitors. Based upon the demonstration of the anionic rich regions of LF [<xref ref-type="bibr" rid="B21">21</xref>,<xref ref-type="bibr" rid="B22">22</xref>,<xref ref-type="bibr" rid="B29">29</xref>], we chose chemical libraries that included cationic compounds to test for LF inhibition. These studies were directed at identification of compounds that selectively inhibited LF both at the enzyme level then evaluation of their effects in cell culture based assays. An ideal therapeutic would penetrate susceptible cells and be effective protease inhibitors in "post-exposure" models of treatment.</p></sec><sec><title>Results</title><sec><title>Substrate kinetics</title><p>Using MAPKKide™ as substrate, velocity vs. substrate curves were analyzed by DYNAFIT, a nonlinear fitting program [<xref ref-type="bibr" rid="B39">39</xref>]. The results of the forward progress curves demonstrated a clear substrate inhibition process as previously shown with a different substrate [<xref ref-type="bibr" rid="B20">20</xref>]. The K<sub>m </sub>and K<sub>i </sub>values for MAPKKide™ were calculated to be 8.6 ± 1.5 μM and 85 ± 17 μM, respectively. These data suggest that multiple inhibitory mechanisms may be available as sites for binding of LF inhibitors (Kuzmic <italic>et al</italic>., submitted).</p></sec><sec><title>Endogenous polyamines inhibit Lethal Factor enzyme activity</title><p>Based upon the presence of anionic sites on LF, we hypothesized that cationic compounds, including members of known drug-like chemical families, might inhibit LF enzyme activity. An initial focused library of commercially available cationic compounds (n~100 compounds) from numerous chemical classes was assembled and tested at a concentration of approximately 10 μg/ml in the LF enzyme inhibition assay. One of these compoundsspermine, was found to inhibit anthrax protease activity in a concentration-dependent manner with a K<sub>i </sub>value of 0.9 ± 0.09 μM (mean ± SEM; Figure <xref ref-type="fig" rid="F1">1</xref>; Figure <xref ref-type="fig" rid="F2">2</xref>; Table <xref ref-type="table" rid="T1">1</xref>). In contrast to the potent inhibition of anthrax LF protease enzyme activity, this compound was >40-fold weaker as a botulinum protease inhibitor (K<sub>i </sub>value = 46 ± 6 μM) and much less active on mammalian proteases including trypsin, cathepsin B and cathepsin D (IC<sub>50 </sub>values > 500 μM). Several other endogenous polyamines, including spermidine, and ornithine, were evaluated for activity as LF inhibitors; these endogenous compounds were weaker than spermine but were still concentration-dependent inhibitors (Table <xref ref-type="table" rid="T1">1</xref>).</p><fig position="float" id="F1"><label>Figure 1</label><caption><p>Chemical structures of compounds used in this study.</p></caption><graphic xlink:href="1471-2210-6-8-1"/></fig><fig position="float" id="F2"><label>Figure 2</label><caption><p>Spermine inhibits anthrax lethal factor protease in a concentration-dependent manner. These results (mean ± SEM) are averaged from 3 separate experiments.</p></caption><graphic xlink:href="1471-2210-6-8-2"/></fig><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Endogenous polyamine and aminoglycoside-mediated inhibition of anthrax lethal factor activity-comparison with other proteases-(Mean ± SEM <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M1" name="1471-2210-6-8-i1" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mtext>K</mml:mtext><mml:mi>i</mml:mi><mml:mrow><mml:mi>a</mml:mi><mml:mi>p</mml:mi><mml:mi>p</mml:mi></mml:mrow></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqqGlbWsdaqhaaWcbaGaemyAaKgabaGaemyyaeMaemiCaaNaemiCaahaaaaa@336E@</mml:annotation></mml:semantics></mml:math></inline-formula> or IC<sub>50 </sub>values)</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="center">LF</td><td align="left">Bot</td><td align="left">MMP-9</td><td align="left">Furin</td></tr></thead><tbody><tr><td align="left">Name</td><td align="center"><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M2" name="1471-2210-6-8-i1" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mtext>K</mml:mtext><mml:mi>i</mml:mi><mml:mrow><mml:mi>a</mml:mi><mml:mi>p</mml:mi><mml:mi>p</mml:mi></mml:mrow></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqqGlbWsdaqhaaWcbaGaemyAaKgabaGaemyyaeMaemiCaaNaemiCaahaaaaa@336E@</mml:annotation></mml:semantics></mml:math></inline-formula> μM</td><td align="left">IC<sub>50 </sub>μM</td><td align="left"><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M3" name="1471-2210-6-8-i1" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mtext>K</mml:mtext><mml:mi>i</mml:mi><mml:mrow><mml:mi>a</mml:mi><mml:mi>p</mml:mi><mml:mi>p</mml:mi></mml:mrow></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqqGlbWsdaqhaaWcbaGaemyAaKgabaGaemyyaeMaemiCaaNaemiCaahaaaaa@336E@</mml:annotation></mml:semantics></mml:math></inline-formula> μM</td><td align="left"><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M4" name="1471-2210-6-8-i1" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mtext>K</mml:mtext><mml:mi>i</mml:mi><mml:mrow><mml:mi>a</mml:mi><mml:mi>p</mml:mi><mml:mi>p</mml:mi></mml:mrow></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqqGlbWsdaqhaaWcbaGaemyAaKgabaGaemyyaeMaemiCaaNaemiCaahaaaaa@336E@</mml:annotation></mml:semantics></mml:math></inline-formula> μM</td></tr><tr><td align="left">Neomycin B tris-sulfate</td><td align="center">0.71 ± 0.04</td><td align="left">92.5</td><td align="left">>300</td><td align="left">N/T</td></tr><tr><td align="left">Sisomicin Sulfate</td><td align="center">1.8</td><td align="left">>300</td><td align="left">>300</td><td align="left">N/T</td></tr><tr><td align="left">Spermine, diphosphate salt</td><td align="center">0.9 ± 0.09</td><td align="left">57</td><td align="left">>300</td><td align="left">N/T</td></tr><tr><td align="left">Amikacin</td><td align="center">23.3</td><td align="left">>300</td><td align="left">>300</td><td align="left">N/T</td></tr><tr><td align="left">Neamine (free base)</td><td align="center">31.1 ± 5.6</td><td align="left">175</td><td align="left">>300</td><td align="left">N/T</td></tr><tr><td align="left">Spermidine</td><td align="center">>100</td><td align="left">76</td><td align="left">N/T</td><td align="left">N/T</td></tr><tr><td align="left">Apramycin</td><td align="center">111</td><td align="left">>300</td><td align="left">>300</td><td align="left">N/T</td></tr><tr><td align="left">Putrescine (1,4-diaminobutane)</td><td align="center">>300</td><td align="left">>300</td><td align="left">>300</td><td align="left">N/T</td></tr><tr><td align="left">Ac-CRATKML-N</td><td align="center">>300</td><td align="left">3.5</td><td align="left">N/T</td><td align="left">N/T</td></tr><tr><td align="left">GM 6001</td><td align="center">7.2 ± 1.76</td><td align="left">>300</td><td align="left">0.002 ± 0.001</td><td align="left">N/T</td></tr><tr><td align="left">H-RRRRRR-OH</td><td align="center">0.24</td><td align="left">N/T</td><td align="left">N/T</td><td align="left">0.06</td></tr><tr><td align="left">Ac-RRRRRR-OH</td><td align="center">0.29 ± 0.04</td><td align="left">N/T</td><td align="left">N/T</td><td align="left">0.05</td></tr><tr><td align="left">Ac-RRRRRR-NH2</td><td align="center">0.12</td><td align="left">N/T</td><td align="left">N/T</td><td align="left">N/T</td></tr><tr><td align="left">H-(D-Arg)-(D-Arg)-(D-Arg)-(D-Arg)-(D-Arg)-(D-Arg)-NH2</td><td align="center">0.04 ± 0.02</td><td align="left">N/T</td><td align="left">N/T</td><td align="left">0.06</td></tr><tr><td align="left">H-R(NO2)R(NO2)</td><td align="center">>300</td><td align="left">N/T</td><td align="left">N/T</td><td align="left">N/T</td></tr><tr><td align="left">Neomycin B hexaguanyl hexatrifluoroacetate salt</td><td align="center">0.03</td><td align="left">14</td><td align="left">>170</td><td align="left">1.5</td></tr><tr><td align="left">Tetraguanyl neamine, free base</td><td align="center">0.3</td><td align="left">N/T</td><td align="left">N/T</td><td align="left">N/T</td></tr></tbody></table><table-wrap-foot><p>N/T = not tested</p></table-wrap-foot></table-wrap></sec><sec><title>Aminoglycosides inhibit LF enzyme activity</title><p>Aminoglycoside antibiotics bind to the polyamine class of glutamate receptors by a mechanism unrelated to antibiotic activity [<xref ref-type="bibr" rid="B40">40</xref>-<xref ref-type="bibr" rid="B42">42</xref>]. We therefore evaluated a series of commercially-available aminoglycosides (n = 31) to determine their potencies as LF inhibitors. The results demonstrated that some members of this chemical class were potent inhibitors of anthrax LF cleavage of the substrate (Table <xref ref-type="table" rid="T1">1</xref>). Both natural aminoglycosides and synthetic aminoglycosides (Table <xref ref-type="table" rid="T1">1</xref>) were active. Of these, neomycin was the most potent aminoglycoside with a K<sub>i </sub>value of 0.3 ± 0.1 μM. Based upon chemical size, we chose to focus on neamine and related compounds (n = 20 neamine derivatives).</p></sec><sec><title>Exogenous nucleic acids alter LF inhibition by neomycin</title><p>Since aminoglycosides are known to bind to nucleic acids [<xref ref-type="bibr" rid="B43">43</xref>-<xref ref-type="bibr" rid="B45">45</xref>] we evaluated key compounds in the absence of DNA (standard assay) and in the presence of a variety of nucleic acids. At concentrations <10 μg/ml, nucleic acids did not affect LF enzyme activity. As shown in Fig <xref ref-type="fig" rid="F3">3</xref>, the potency of neomycin was greater in the absence of DNA compared to in the presence of salmon testes DNA (4 and 8 μg/ml). The higher concentration of DNA caused a ~10-fold right shift in potency of neomycin. In contrast, the concentration-dependent inhibitory activities of neamine or spermine were unaffected by DNA or RNA (specifically human placental DNA, type III RNA, polyA-polyU; results not shown).</p><fig position="float" id="F3"><label>Figure 3</label><caption><p>Influence of nucleic acids on concentration-dependent LF inhibition. Anthrax lethal factor activity was measured in the absence of DNA (■) as well as in the presence of salmon sperm DNA at 4 μg/ml (▲) and 8 μg/ml (▼), performed in triplicate.</p></caption><graphic xlink:href="1471-2210-6-8-3"/></fig></sec><sec><title>Cationic peptides inhibit furin enzyme activity</title><p>Cross-inhibition of LF and furin has been demonstrated for polyarginine based inhibitors [<xref ref-type="bibr" rid="B46">46</xref>]. We therefore examined the ability of our panel of LF inhibitors to inhibit furin in an <italic>in vitro </italic>substrate cleavage assay. As expected, several polyarginine derivatives inhibited furin activity (Table <xref ref-type="table" rid="T1">1</xref>). None of the remaining compounds interfered with furin activity at concentrations up to 100 μM.</p></sec></sec><sec><title>Discussion</title><p>In the initial phase of this study, we sought to identify compounds that selectively inhibited anthrax lethal factor enzyme activity. Such compounds were hypothesized to be potential lead molecules for optimization as drugs to treat <italic>B. anthracis </italic>infection. Since inhibitors of this protease were not known at the time, we chose to screen structurally diverse collections of individual compounds (a "library" consisting of ~500 compounds in several chemical classes) as one approach towards lead identification. We included simple linear cationic polyamines (n = 17) in the screening library with the hypothesis that they might bind to anionic sites on LF and thus block substrate cleavage. The data presented in this study show that spermine (a simple linear polyamine) is a concentration-dependent, sub-micromolar inhibitor of LF with reduced inhibitory potencies (termed selectivity) versus other bacterial and mammalian proteases. Polyamine analogs of spermine, including spermidine and ornithine were less active than spermine but still displayed concentration-dependent inhibitory effects as LF inhibitors.</p><p>Based upon literature demonstrating that both polyamines and aminoglycoside antibiotics bind to the N-methyl-D-aspartate receptor [<xref ref-type="bibr" rid="B47">47</xref>,<xref ref-type="bibr" rid="B48">48</xref>], we also evaluated aminoglycoside antibiotics for LF inhibition. In our independent studies reported here and identified by other laboratories [<xref ref-type="bibr" rid="B49">49</xref>-<xref ref-type="bibr" rid="B51">51</xref>] we found that gentamicin inhibited LF enzyme activity without inhibiting other proteases from bacterial and mammalian sources. We then showed that other compounds were more potent LF inhibitors than gentamicin. To further validate the mechanism, we tested cationic peptides (n~5) such as D- and L-hexaarginine as well as non-peptidyl cationic polymers including poly-L-arginine and poly-L-lysine (molecular weight ranges = 5,000–15,000); the larger cationic polymers (both peptidyl and non-peptidyl) were more potent inhibitors. While these large molecules will not be drug leads, they validated the mechanistic hypotheses of LF inhibition. Based upon these data, we concluded that neamine possessed the most relevant combination of drug-like properties and it was used as a scaffold for designing more potent and cell permeable analogs [<xref ref-type="bibr" rid="B52">52</xref>].</p><p>Aminoglycosides are effective antibiotics for the treatment of Gram-positive and Gram-negative infections as well as certain mycobacterial infections [<xref ref-type="bibr" rid="B53">53</xref>,<xref ref-type="bibr" rid="B54">54</xref>]. Their use, however, is limited by lack of oral absorption and toxicity at high doses including both ototoxicity and nephrotoxicity [<xref ref-type="bibr" rid="B55">55</xref>]. Because of such toxicities, intravenous use of aminoglycosides in large and diverse age/health populations would pose a significant risk if used as prophylactic agents. Orally active/non-toxic compounds are still needed for protection against bioterrorist threats based on <italic>B. anthracis </italic>and its toxins.</p><p>High affinity polyamine interactions with nucleic acids are well known in both cell-free and cellular systems [<xref ref-type="bibr" rid="B56">56</xref>-<xref ref-type="bibr" rid="B60">60</xref>]. In anticancer studies, for example, exogenous polyamines are cytotoxic by depleting endogenous polyamine levels through feedback inhibition mechanisms [<xref ref-type="bibr" rid="B59">59</xref>]. The cellular uptake of linear polyamines is well-recognized and numerous transporters have been shown to modulate polyamine levels within cells and organelles [60 and references therein].</p><p>We also sought to determine if the presence of DNA or RNA in the MAPKK cleavage assays would affect polyamine inhibition of LF activity. First, we demonstrated that nucleic acids did not inhibit LF activity at concentrations below those known to be present in human blood (< 80 μg/ml; Promega tech bulletin). Subsequently, we have also demonstrated that the presence of these nucleic acids prevent certain compounds such as neomycin but not neamine from inhibiting LF activity. One hypothesis we have considered is the role of size of the inhibitory molecules; smaller compounds such as spermine and neamine did not bind nucleic acids whereas larger molecules that inhibited LF (neomycin) were rendered less potent in the presence of 4–8 μg/ml of DNA or RNA. This is a unique pharmacological discovery first demonstrated in this effort.</p><p>All three chemical classes (linear polyamines, aminoglycosides and peptides) are highly charged molecules; they were not expected to be active in cell models of anthrax lethal factor cytotoxicity. This result was confirmed in our initial LF cytotoxicity studies with RAW 264.7 macrophage cells as all compounds were not active up to the highest concentration tested (500 μM). Recent studies [<xref ref-type="bibr" rid="B50">50</xref>], however, have demonstrated that aminoglycosides at "seemingly physiological conditions" inhibit LF and exhibit antibiotic activity against <italic>B. anthracis</italic>. In addition, linear polyamines enter cells by multiple mechanisms including active transport [<xref ref-type="bibr" rid="B48">48</xref>,<xref ref-type="bibr" rid="B59">59</xref>,<xref ref-type="bibr" rid="B60">60</xref>]. These different results highlight the need for continued research in this area including longer compound exposure periods.</p><p>Inhibition of the proprotein convertase, furin, by cationic peptides was expected since cationic hexapeptides and nonapeptides have been demonstrated to inhibit furin activity [<xref ref-type="bibr" rid="B12">12</xref>,<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B46">46</xref>]. In contrast, however, the charged, nonpeptidyl compounds were inactive as furin inhibitors at concentrations up to 10 μM. It is likely that other small molecule cations will inhibit furin processing. Such compounds may be effective in a broad spectrum of diseases where furin cleavage of proteins play a role, such as Alzheimer's disease, viral infections and bacterial infections [<xref ref-type="bibr" rid="B13">13</xref>].</p></sec><sec><title>Conclusion</title><p>Taken together, the enzymologic and pharmacologic results presented in this study demonstrate the role of anionic sites on anthrax lethal factor that specifically bind cationic compounds from different chemical classes. Ultimately multi-drug and multi-dose combinations of antibiotics that suppress production of anthrax spores plus edema factor and LF inhibitors that target intracellular toxins will be employed to treat people exposed to anthrax gene products. Recently, an antiviral agent called adefovir was found to inhibit edema factor adenylate cyclase activity [<xref ref-type="bibr" rid="B19">19</xref>]. Combinations of edema factor and lethal factor inhibitors along with Gram-positive antibiotics could be utilized to treat people exposed to anthrax or its gene products. As a spin-off of these efforts, researchers may want to determine if the presence/absence of nucleic acids affects the potency of selected compounds.</p></sec><sec sec-type="methods"><title>Methods</title><sec sec-type="materials"><title>Materials</title><p>Aminoglycosides, nucleic acids and endogenous polyamines used in this study were purchased from Sigma (St. Louis MO), and Calbiochem (San Diego, CA). Synthetic polyamines were purchased from Mixture Sciences Inc. (La Jolla, CA). Tetrapeptides were synthesized by Synpep, Inc. (Dublin, CA).</p></sec><sec><title>Lethal Factor protease assay</title><p>Lethal Factor (20 nM final concentration) and MAPKK substrate (MAPKKide<sup>® </sup>12.5 μM, final concentration) were purchased from List Biological Laboratories, Campbell, CA and used according to the fluorescence resonance energy transfer (FRET) method. The assay final volume was 50 μl (Fisher #3694 96-well half area plates) consisting of 5 μl inhibitor/test sample/buffer, 25 μl buffer (20 mM Hepes, pH 7.4), 10 μl enzyme and 10 μl substrate. Test sample, buffer and enzyme were incubated briefly at room temperature. Upon addition of substrate, the reaction was linear for 15 min at room temperature. Fluorescence intensity was determined in the kinetic mode (Ex: 320 nm, Em: 420 nm; 6-minute read time; Molecular Devices Gemini fluorescence plate reader) and data was captured by SoftMax Pro (Molecular Devices, Sunnyvale, CA). Analysis of resulting kinetic data was carried out using DYNAFIT and Batch Ki (Biokin, Ltd., Pullman, WA) then plotted with Prism (Graphpad, San Diego, CA).</p></sec><sec><title>Other protease assays</title><p>Additional FRET-based substrate cleavage assays were established to monitor specificity of LF inhibitors. The botulinum neurotoxin/A (BoNT/A) assay was carried out in a similar manner as the lethal factor assay (above) except the buffer was composed of 30 mM Hepes, pH 7.3, 5 mM dithiothreitol, 0.25 mM ZnCl<sub>2</sub>, and 1 mg/mL bovine serum albumin (BSA). The substrate for the BoNT/A was SNAPtide<sup>® </sup>(12.5 μM, final concentration) purchased from List Biological Laboratories (Campbell, CA). BoNT/A enzyme was obtained from the University of Wisconsin.</p><p>Furin inhibition was quantified [<xref ref-type="bibr" rid="B12">12</xref>] by measuring the hydrolysis of the fluorogenic furin substrate Pyr-RTKR-CMK (Peptide Institute, Osaka, Japan). Assays were performed in 96-well plates using 100 μM substrate, and a serial dilution of inhibitors. The initial velocity (V<sub>o</sub>) of the 200 μl reactions was quantified using a Spectramax Gemini XS microplate reader. The IC<sub>50 </sub>of each inhibitor was calculated by plotting V<sub>0 </sub>versus log [I] and performing nonlinear regression. K<sub>i(app) </sub>was calculated from the IC<sub>50 </sub>values using the equation K<sub>i </sub>= IC<sub>50</sub>/1+([S]/K<sub>m</sub>.</p></sec><sec><title>Cell-based cytotoxicity assay</title><p>RAW 264.7 murine macrophage cells (ATCC, Manassas, VA) were grown in the presence of Dulbecco's modified Eagles medium containing 10% FBS to 70% confluency (approximately 50,000 cells/well for 24 hours) in standard Corning 96-well cell-culture grade polystyrene plates (Corning, NY). Test compounds at various concentrations (concentration-response) and appropriate vehicles were added to the medium and preincubation was continued for 60 min. At the end of this period, PA (250 ng/ml) and LF (250 ng/ml) were added sequentially to each well. After 2 hours, 3-[4,5-dimethylthiazol-2-yl]-2,5-diphenytetrazolium bromide (MTT; 0.5 mg/ml final) was added and the incubation was continued for an additional 2 hours. The supernatant fluid was removed from each well and the remaining pigment was dissolved in 100 μl of 0.5% (w/v) sodium dodecyl sulfate, 40 mM HCl in 90% (v/v) 2-propanol. Absorbance was read at 570 nm and % viability was determined as a function of control wells.</p></sec></sec><sec><title>Authors' contributions</title><p><bold>M.E.G. </bold>(principal investigator, chemical library composition, data evaluation, manuscript preparation), <bold>L.C. </bold>(execution of biochemical experiments and data analyses), <bold>D.N. </bold>(establishment of cell assays and execution of cell assays), <bold>O.S. </bold>(synthesis of compound in Table <xref ref-type="table" rid="T1">1</xref>), <bold>S.O. </bold>(synthesis of compound in Table <xref ref-type="table" rid="T1">1</xref>) and <bold>T.H. </bold>(target conception; critical intellectual discussion with principal investigator and manuscript evaluation/critique)</p></sec>
Randomized pilot study to disseminate caries-control services in dentist offices	<sec><title>Background</title><p>To determine whether education and financial incentives increased dentists' delivery of fluoride varnish and sealants to at risk children covered by capitation dental insurance in Washington state (U.S.).</p></sec><sec sec-type="methods"><title>Methods</title><p>In 1999, 53 dental offices in Washington Dental Service's capitation dental plan were invited to participate in the study, and consenting offices were randomized to intervention (n = 9) and control (n = 10) groups. Offices recruited 689 capitation children aged 6–14 and at risk for caries, who were followed for 2 years. Intervention offices received provider education and fee-for-service reimbursement for delivering fluoride varnish and sealants. Insurance records were used to calculate office service rates for fluoride, sealants, and restorations. Parents completed mail surveys after follow-up to measure their children's dental utilization, dental satisfaction, dental fear and oral health status. Regression models estimated differences in service rates between intervention and control offices, and compared survey measures between groups.</p></sec><sec><title>Results</title><p>Nineteen offices (34%) consented to participate in the study. Fluoride and sealant rates were greater in the intervention offices than the control offices, but the differences were not statistically significant. Restoration rates were lower in the intervention offices than the control offices. Parents in the intervention group reported their children had less dental fear than control group parents.</p></sec><sec><title>Conclusion</title><p>Due to low dentist participation the study lacked power to detect an intervention effect on dentists' delivery of caries-control services. The intervention may have reduced children's dental fear.</p></sec>	<contrib id="A1" equal-contrib="yes" corresp="yes" contrib-type="author"><name><surname>Grembowski</surname><given-names>David</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A2" equal-contrib="yes" contrib-type="author"><name><surname>Spiekerman</surname><given-names>Charles</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" equal-contrib="yes" contrib-type="author"><name><surname>del Aguila</surname><given-names>Michael A</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Anderson</surname><given-names>Maxwell</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Reynolds</surname><given-names>Debra</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Ellersick</surname><given-names>Allison</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A7" contrib-type="author"><name><surname>Foster</surname><given-names>James</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A8" contrib-type="author"><name><surname>Choate</surname><given-names>Leslie</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib>	BMC Oral Health	<sec><title>Background</title><p>Scientific advances and new, effective caries-control services have emerged for preventing caries, [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B2">2</xref>] yet most general dentists have not adopted them[<xref ref-type="bibr" rid="B3">3</xref>]. Dissemination, or efforts to persuade dentists to adopt effective innovations, is important for improving public health, particularly among children. Although caries has declined in the U.S. over the past three decades, 51% of children aged 5–9 and 78% of 17 year-olds have at least one carious lesion or filling [<xref ref-type="bibr" rid="B4">4</xref>-<xref ref-type="bibr" rid="B6">6</xref>]. Reliance on diffusion, or the passive spread of new technology, will not solve the problem, given evidence that application of medical technology lags an average of 17 years[<xref ref-type="bibr" rid="B7">7</xref>].</p><p>One approach to increase dentist adoption of new, caries-control services is to pay dentists for providing them. However, when fluoride varnish became a covered benefit in a fee-for-service dental plan in 1996, most dentists did not adopt the technology, [<xref ref-type="bibr" rid="B8">8</xref>] suggesting that stronger interventions are necessary to increase dentist adoption of caries-control services.</p><p>The purpose of this study is to determine whether a more intensive intervention composed of provider education and reimbursement for a package of caries-control services increases the delivery of caries-control services and reduces restorations among at risk children with capitation dental coverage.</p><sec><title>Caries Prevention Study</title><p>The Caries Prevention Study was conducted in the capitation dental plan offered by Washington Dental Service (WDS), the Delta Dental Plan in Washington state. As in most capitation plans, dentists receive a fixed, monthly payment for each capitation enrollee in the practice, and dentists provide all the care that may be required, within contract limitations, without additional payment. Thus, the dentist has a financial incentive – that is, earns more income – to prevent disease and avoid treatment costs. The plan fully covers fluoride varnish and sealants for children aged 14 and under. About 52% of U.S. children aged 6–18 with private dental insurance see a dentist each year[<xref ref-type="bibr" rid="B9">9</xref>].</p><p>The intervention has two parts: 1) fee-for-service (FFS) reimbursement for providing fluoride varnish and sealants to at risk children with capitation dental benefits; and 2) provider education about caries-control technologies and how to incorporate them into daily practice. The intervention promotes dentists' adoption by supplementing the dentist's capitation payment with fee-for-service reimbursement for sealants and fluoride varnish delivered to capitation children who are at risk for caries. Provider education included didactic instruction on caries-control services and the "business case" for dental practice based on prevention, plus a video demonstration of fluoride varnish application.</p><p>Economic theory and diffusion theory suggest the intervention will increase dentist adoption of sealants and fluoride varnish. Dentists have two financial incentives to adopt the technologies. Because financial reimbursement for caries-control services increases dentist income and dentists usually want to increase practice revenue, the financial incentive may increase dentist adoption of the technologies[<xref ref-type="bibr" rid="B10">10</xref>]. In addition, by delivering caries-control services and preventing decay, [<xref ref-type="bibr" rid="B11">11</xref>-<xref ref-type="bibr" rid="B17">17</xref>] restorations may be reduced, and therefore, dentists retain more profits from their capitation payments, an added incentive to adopt the technology.</p><p>Diffusion theory posits that innovations do not sell themselves but are adopted over time through the predictable patterns of communication in a profession[<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B19">19</xref>]. Compared to reparative treatments, preventive innovations tend to have a slower rate of adoption because clinicians have difficulty observing their relative advantages[<xref ref-type="bibr" rid="B18">18</xref>]. The combination of provider education and financial reimbursement can speed-up the diffusion process because they increase provider awareness and knowledge of the innovation and its relative advantages[<xref ref-type="bibr" rid="B18">18</xref>]. In one study, dentists who knew more about fluoride varnish were more likely to adopt the technology than those who knew less[<xref ref-type="bibr" rid="B3">3</xref>]. Well-designed classes also create communication channels for sharing information that can promote adoption[<xref ref-type="bibr" rid="B20">20</xref>].</p><p>Adoption also is more likely to occur when interventions target the entire dental office rather than just the dentist, mainly because innovations almost always require changes in office structure and ways of working together[<xref ref-type="bibr" rid="B21">21</xref>]. Consequently, provider education about caries-control services included dental office staff as well as dentists. Finally, when outside professional dental organizations, such as WDS, sponsor the intervention and thereby set practice norms for caries-control services, organizations may increase the spread of innovations, although few studies exist on this topic[<xref ref-type="bibr" rid="B21">21</xref>].</p><p>In contrast, Kuhn's model of paradigm shift in scientific disciplines suggests the intervention will not increase dentist adoption of caries control services[<xref ref-type="bibr" rid="B22">22</xref>]. The intervention is more than the adoption of a new technology; it is a paradigm shift from the traditional surgical approach to a disease-based approach, or "medical model," of dental practice. For dentists, this requires a philosophical switch that can impose a significant change in the way clinicians provide care and generate income[<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B23">23</xref>].</p><p>According to Kuhn, paradigm shift for a provider is not gradual; it is all-or-nothing. For the profession as a whole, paradigm shift can be a long-term process that begins when dentists realize that the old paradigm (the surgical model) no longer adequately addresses the problems facing the profession, and it progresses as more and more evidence supporting the new paradigm (the medical model) appears in the literature. Sufficient evidence exists presently for resin based sealants, but insufficient evidence exists for ionomor sealants[<xref ref-type="bibr" rid="B24">24</xref>]. Without an evidence base for most caries-control services, paradigm shift to the medical model is premature, and the intervention may not be strong enough to speed up this process.</p><p>In particular, dental offices may not adopt fluoride varnish because the service is not integrated fully into U.S. dental institutions. After the Federal Drug Administration approved fluoride varnish as a devise for use in 1994, it can be used "off-label," which means the agent is being used for another purpose for which FDA approval is lacking. Off-label use of fluoride varnish for caries prevention is occurring because of substantial evidence that fluoride varnish reduces caries[<xref ref-type="bibr" rid="B14">14</xref>]. The American Dental Association does not have a preventive procedure code for fluoride varnish, and all dental insurance plans do not reimburse for fluoride varnish, which may have slowed adoption of the service.</p><p>The evidence in medical and dental care is equivocal about whether financial incentives increase the delivery of preventive services [<xref ref-type="bibr" rid="B25">25</xref>-<xref ref-type="bibr" rid="B31">31</xref>]. Some studies report that incentives motivated medical providers to improve chart documentation but did not increase preventive services[<xref ref-type="bibr" rid="B25">25</xref>]. Even if financial incentives are effective, greater use of preventive services may not reduce restorations[<xref ref-type="bibr" rid="B32">32</xref>,<xref ref-type="bibr" rid="B33">33</xref>]. In contrast, small group interactive education, along with educational outreach by experts, are effective in changing provider behavior, particularly for preventive care[<xref ref-type="bibr" rid="B34">34</xref>].</p><p>Our purpose is to conduct a randomized study testing whether financial incentives and provider education increase dentists' delivery of caries-control services to children at risk of caries.</p></sec></sec><sec sec-type="methods"><title>Methods</title><sec><title>Study design, populations and intervention</title><p>The impact of the Caries Prevention Study on the delivery of dental services was evaluated using a randomized posttest-only control group study design[<xref ref-type="bibr" rid="B35">35</xref>]. In 1999 we invited 53 dentists who owned Seattle-area dental offices and were network providers in WDS's capitation dental plan, and who had 30 or more children patients aged 6–14 covered by the plan. In 1997 about 65% of capitation dentists provided dental sealants to at least 1 child, but among those dentists, only about 6% of their children patients received sealants. Of these children, an average of 3.1 sealants were provided per child. Accurate records of dentists' delivery of fluoride varnish did not exist before the study, and we assumed dentist delivery of fluoride varnish was similar to other dentists in Washington state[<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B8">8</xref>]. In short, dentists had not fully adopted these two services at baseline. The study was reviewed and approved by the Institutional Review Board of the University of Washington.</p><p>About 36% of the dentist-owners (n = 19) consented to participate. The offices were randomized in a stratified manner to ensure the groups were reasonably balanced in terms of office size. The offices of the consenting dentist owners were ordered by number of children aged 6–14 in the office. In each successive pair of offices, one office was randomly chosen for the intervention group, and the other office was assigned to the control group, yielding 9 dentist-owners in the intervention group and 10 dentist-owners in the control group.</p><p>Intervention dentists and their office staff attended an educational session about caries-control services. Dentists received continuing dental education credits, and staff attended because Washington state law allows delegation of caries-control services to auxiliaries. The 4-hour didactic sessions provided information about the clinical benefits of fluoride varnish and sealants, how to incorporate them into day-to-day practice, and how the delivery of preventive services might contribute to the financial health of the practice. A video was shown demonstrating the application of fluoride varnish, and protocols for enrolling and following children were reviewed. Control dentists and staff attended separate training sessions covering study data collection protocols.</p><p>Between August 1999 and July 2000, intervention and control dental offices invited eligible children to participate in the study at regular office visits. Eligibility criteria were: 1) coverage by WDS capitation plan; 2) aged 6–14; 3) parental consent to participate; and 4) at risk for caries (defined as ≥ 1 restoration or carious lesion)[<xref ref-type="bibr" rid="B36">36</xref>]. WDS monitored utilization records to verify that all children with restorations seen at intervention and control officers were invited to participate in the study. A total of 391 intervention and 298 control children were enrolled.</p><p>Children were followed for 2 years through July 2002. Intervention dentists received their contractual capitation payments and monthly fee-for-service reimbursement for providing fluoride varnish ($20 U.S.) and sealants ($20 U.S.) to eligible children. To preserve the value of the incentive, intervention offices received free supplies of Duraphat<sup>® </sup>fluoride varnish throughout the follow-up period. Intervention and control offices were required contractually to submit service records to WDS, and intervention offices received fee-for-service reimbursement when service records contained procedure codes for fluoride varnish (WDS code 1206), topical fluoride application (1201, 1203, 1204) or sealants (1351). Fluoride codes in WDS service records were compared with fluoride codes in office dental charts for sampled intervention and control children (n = 289). At the end of follow-up, intervention and control offices reported on the percentage of eligible children receiving fluoride varnish.</p><p>WDS mailed parents 6-month recall letters to promote regular visits. When children exited the study, WDS mailed parents a questionnaire about their children's oral health status, dental utilization and dental fear, and satisfaction with their child's dental care.</p></sec><sec><title>Measures</title><sec><title>Dental service rates</title><p>WDS service records were used to calculate dentist service rates, or the average number of times that a dental service was provided to enrolled children in the office of each dentist-owner in the 2-year follow-up period. Dental services rates were calculated for fluoride, sealants, and restorations.</p></sec><sec><title>Office characteristics</title><p>The characteristics of intervention and control offices included the number of capitation patients covered by WDS in the office, number of capitation children aged 6 to 14, and number of hygienists employed by the office.</p></sec><sec><title>Baseline caries risk</title><p>Dental offices recorded the number of decayed, missing, and filled teeth (dft and DFT) and the number of sealants for each child at enrollment. A child's caries risk at baseline was measured by summing the dft and DFT scores. Offices also reported their perception of a child's caries risk (low, moderate, or high), and whether the child was taking fluoride supplements, whether orthodontic treatment was underway, and brushing frequency.</p></sec><sec><title>Oral health status</title><p>Survey measures included the parent's self-rating of the child's oral health on a 5-point scale (poor (1), fair, good, very good, excellent (5). Parents also rated, compared to one year ago, the condition of the child's teeth on a 5-point scale (much better (1) to much worse (5).</p></sec><sec><title>Dental satisfaction and utilization</title><p>Parents rated their satisfaction with their children's dental care at the follow-up survey through two items: 1) parent rating of the dental care from the child's dentist; and 2) parent rating of the preventive services from the child's dentist. Each item was rated on a 5-point scale (poor (1), fair, good, very good, excellent (5)). Survey measures included the parent's self-report of whether the child received any sealants, fluoride varnish, or any restorations in the past year.</p></sec><sec><title>Dental fear</title><p>Parents rated their child's dental fear at the follow-up survey using a modified item from the Corah Dental Anxiety Scale[<xref ref-type="bibr" rid="B37">37</xref>]. Parents were asked how their child would feel if the child had to go to the dentist tomorrow. Children were categorized as being fearful if the parent responded the child would be afraid that it would be unpleasant or painful, or the child would be very frightened of what the dentist might do.</p></sec><sec><title>Child and household characteristics</title><p>Though the dental office, rather than the child, was the unit of randomization, child and parent characteristics might also influence dental utilization, dental satisfaction, and dental fear[<xref ref-type="bibr" rid="B38">38</xref>]. Child characteristics measured from WDS records included age, gender and years enrolled in the study, which may be less than two years for children of parents losing dental capitation benefits (range: 0–2 years). Child characteristics measured from the parent survey included use of fluoride drops or tablets and frequency of snacks, pop or juice between meals. Parent and household characteristics included age, gender, the parent's race/ethnicity, years of education, the parent's marital status, and the number of people in the household.</p></sec></sec><sec><title>Data collection</title><p>The office measures of a child's caries risk were collected using a "Tooth Chart," a version of the form previously field-tested and used in a statewide oral health survey of Washington children[<xref ref-type="bibr" rid="B39">39</xref>]. Protocols for completing the Tooth Charts were explained at training sessions. Offices completed Tooth Charts for each child at enrollment and each dental visit in the follow-up period. Intervention and control offices were reimbursed $10 (U.S.) by WDS for each completed Tooth Chart to increase compliance with study protocols and offset their data collection costs. Tooth Charts were reviewed for completeness, and offices were contacted to supply any missing data from office records when available.</p><p>The parent survey was performed by WDS and followed procedures recommended by Dillman[<xref ref-type="bibr" rid="B40">40</xref>,<xref ref-type="bibr" rid="B41">41</xref>]. When children exited the study, parents were mailed a questionnaire, cover letter, prepaid return envelope, and a $15 (U.S.) gift certificate. The initial mailing was followed by: (1) a reminder postcard; (2) a second mailing of the questionnaire and revised cover letter to nonrespondents; and (3) a third mailing of the questionnaire and revised cover letter to nonrespondents.</p></sec><sec><title>Data analysis</title><p>Bivariate statistical tests compared the characteristics of participating and nonparticipating dentists and offices. Bivariate statistical tests were performed to determine whether the characteristics of children and dental offices in the intervention group were significantly different than the characteristics in the control group.</p><p>Treatment and control group differences in amounts of fluorides, sealants, and restorations given per child over the study period were evaluated using permutation tests for group-randomized data[<xref ref-type="bibr" rid="B42">42</xref>,<xref ref-type="bibr" rid="B43">43</xref>]. The permutation test makes no distributional assumptions, and as applied here, takes into account the stratified group randomization by using all randomizations possible via this scheme to create the permutation comparison distribution. The group differences were adjusted using linear modeling, taking into account the following covariates: time in study with dental capitation coverage, child characteristics (age, gender, current orthodontia, sum of dft and DFT, dentist-evaluated caries risk, and number of sealed teeth at study entry, fluoride supplement use in restoration model), size of the office's capitation plan (number of children and number of total patients in capitation plan), and the number of hygienists employed by the office. All <italic>p</italic>-values are two-sided. Computations were performed using <italic>S-Plus</italic><sup>© </sup>2000 statistical software[<xref ref-type="bibr" rid="B44">44</xref>].</p><p>Logistic regression was performed to determine whether the age and gender of children with completed questionnaires were different from those without, controlling for group. For those with questionnaires, bivariate statistical tests were performed to determine whether children, parents and households in the intervention group were significantly different from the control group. Regression models were estimated to determine intervention effects on parent-reported dental utilization, dental satisfaction, and dental fear, and all models employed the permutation tests to determine intervention effects. Children control variables (female, age, brushing, fluoride supplements, snacks) and adult/household control variables (adult education, nonwhite race, age, female, marital status, and household size) were included in the models.</p></sec></sec><sec><title>Results</title><sec><title>Participating and non-participating dentists</title><p>Dentists who chose to participate had significantly more children aged 6–14 years old in the capitation dental plan in the previous year than dentists who did not participate (avg 220 vs. 81 children, <italic>p</italic>=.004), and participating dentists had more total patients in the capitation plan (avg 1171 vs. 479, <italic>p</italic>=.007). Otherwise no statistically significant differences existed between participating and nonparticipating dentists at baseline for the following characteristics: percentage solo practitioners (75% vs. 72%, respectively); average number of hygienists (1.06 vs. .69); average number of assistants (3.04 vs. 2.67); average number of operatories (4.92 vs. 4.31); electronic submission of dental claims (46% vs. 34%); computer in office (79% vs. 86%); accepting new patients (92% vs. 90%); average percent of children with sealants based on WDS records (10% vs. 15%). The main reasons for nonparticipation were office disruption (31% of nonparticipating dentists), lack of interest (28%), and uncertain future membership in the capitation dental plan (21%); other concerns were office staff reluctance (10%) and the poor timing of the study for the office (10%).</p></sec><sec><title>Children and dental office characteristics</title><p>Table <xref ref-type="table" rid="T1">1</xref> presents the characteristics of children and dental offices in the intervention and control groups. Although no statistically significant differences were found between groups, children in the intervention group tended to have more sealed teeth, a greater percentage of dentists rating their caries risk as high, and less likely to have orthodontia. Dental offices in the intervention group also tended to have more capitation patients and employ more hygienists.</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Baseline characteristics of dental offices</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="center"><bold>Control Group</bold></td><td align="center"><bold>Intervention Group</bold></td><td align="center"><bold><italic>p</italic>-value</bold></td></tr></thead><tbody><tr><td align="left"><underline>Dental Office Characteristics</underline></td><td align="center">(n = 10)</td><td align="center">(n = 9)</td><td></td></tr><tr><td align="left">Average number of capitation adults and children aged 6–14 and adults</td><td align="center">1042 (960)</td><td align="center">1408 (1758)</td><td align="center">.57</td></tr><tr><td align="left">Average number of capitation children aged 6–14</td><td align="center">196 (191)</td><td align="center">270 (324)</td><td align="center">.44</td></tr><tr><td align="left">Average number of hygienists</td><td align="center">0.8 (1.1)</td><td align="center">1.7 (1.3)</td><td align="center">.27</td></tr></tbody></table><table-wrap-foot><p>Standard deviations are in parentheses.</p></table-wrap-foot></table-wrap><p>Of all the intervention and control offices, only two intervention offices submitted service records with the WDS procedure code (1206) specific for fluoride varnish, as opposed to codes for other fluoride applications. About 4% of all fluoride codes in chart records were for fluoride varnish. In interviews with office personnel after the follow-up period, about half of the offices in each group reported consistent use of fluoride varnish for most eligible children.</p><p>Table <xref ref-type="table" rid="T3">3</xref> presents unadjusted dental service rates for dental offices in the control and intervention groups. Overall, the average fluoride (either fluoride varnish or topical fluoride application) and sealant rates were greater in the intervention offices than in the control offices. Average restoration rates were similar in the intervention and control offices.</p><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Baseline characteristics of enrolled children</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="center"><bold>Control Group</bold></td><td align="center"><bold>Intervention Group</bold></td><td align="center"><bold><italic>p</italic>-value</bold></td></tr></thead><tbody><tr><td align="left"><underline>Child Characteristics</underline></td><td align="center">(n = 298)</td><td align="center">(n = 391)</td><td></td></tr><tr><td align="left">Average length of follow-up</td><td align="center">1.8 (0.5)</td><td align="center">1.9 (0.4)</td><td align="center">.83</td></tr><tr><td align="left">Gender (% male)</td><td align="center">55</td><td align="center">53</td><td align="center">.72</td></tr><tr><td align="left">Average age (yrs)</td><td align="center">10.0 (2.2)</td><td align="center">9.9 (2.3)</td><td align="center">.73</td></tr><tr><td align="left">Average number of sealed teeth</td><td align="center">1.0 (1.6)</td><td align="center">1.9 (2.2)</td><td align="center">.67</td></tr><tr><td align="left">Dentist-evaluated caries risk (%)</td><td></td><td></td><td align="center">.31</td></tr><tr><td align="left"> Low</td><td align="center">13</td><td align="center">12</td><td></td></tr><tr><td align="left"> Medium</td><td align="center">32</td><td align="center">11</td><td></td></tr><tr><td align="left"> High</td><td align="center">9</td><td align="center">30</td><td></td></tr><tr><td align="left"> No evaluation</td><td align="center">45</td><td align="center">46</td><td></td></tr><tr><td align="left">Children with orthodontic treatment (%)</td><td align="center">15</td><td align="center">8</td><td align="center">.61</td></tr><tr><td align="left">Average number of decayed and filled teeth (sum of dft, DFT)</td><td align="center">4.2 (2.7)</td><td align="center">4.2 (2.7)</td><td align="center">.99</td></tr></tbody></table><table-wrap-foot><p>Standard deviations are in parentheses.</p></table-wrap-foot></table-wrap><table-wrap position="float" id="T3"><label>Table 3</label><caption><p>Unadjusted dental service rates for dental offices in the control and intervention groups</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="center"><bold>Control Group</bold></td><td align="center"><bold>Number of Children</bold></td><td align="center" colspan="3"><bold>Average Number Services per Child in 2-Year Follow-up Period</bold></td></tr></thead><tbody><tr><td align="center"><bold>Offices</bold></td><td align="center"><bold>Followed</bold></td><td align="center"><bold>Fluoride</bold></td><td align="center"><bold>Sealants</bold></td><td align="center"><bold>Restorations</bold></td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="center">1</td><td align="center">2</td><td align="center">0.5 (0.7)</td><td align="center">0.0 (0.0)</td><td align="center">0.0 (0.0)</td></tr><tr><td align="center">2</td><td align="center">5</td><td align="center">2.6 (0.9)</td><td align="center">0.0 (0.0)</td><td align="center">0.6 (0.9)</td></tr><tr><td align="center">3</td><td align="center">7</td><td align="center">2.3 (1.7)</td><td align="center">0.1 (0.4)</td><td align="center">0.4 (0.5)</td></tr><tr><td align="center">4</td><td align="center">10</td><td align="center">1.7 (1.1)</td><td align="center">0.7 (1.9)</td><td align="center">0.6 (1.1)</td></tr><tr><td align="center">5</td><td align="center">21</td><td align="center">2.8 (0.9)</td><td align="center">1.2 (1.9)</td><td align="center">1.8 (2.5)</td></tr><tr><td align="center">6</td><td align="center">21</td><td align="center">0.9 (1.1)</td><td align="center">0.9 (2.1)</td><td align="center">1.7 (2.0)</td></tr><tr><td align="center">7</td><td align="center">27</td><td align="center">1.1 (1.0)</td><td align="center">0.9 (1.8)</td><td align="center">2.9 (2.4)</td></tr><tr><td align="center">8</td><td align="center">28</td><td align="center">2.4 (1.4)</td><td align="center">1.3 (1.8)</td><td align="center">1.1 (1.4)</td></tr><tr><td align="center">9</td><td align="center">74</td><td align="center">3.0 (1.3)</td><td align="center">0.9 (1.8)</td><td align="center">1.9 (2.2)</td></tr><tr><td align="center">10</td><td align="center">103</td><td align="center">1.8 (1.0)</td><td align="center">0.6 (1.7)</td><td align="center">2.5 (3.3)</td></tr><tr><td align="center"><bold>Total Children and Average Rates for All Control Offices</bold></td><td align="center">298</td><td align="center">2.1 (1.3)</td><td align="center">0.8 (1.8)</td><td align="center">2.0 (2.6)</td></tr><tr><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>Intervention Group Offices</bold></td><td></td><td></td><td></td><td></td></tr><tr><td align="center">1</td><td align="center">1</td><td align="center">4.0 (-)</td><td align="center">0.0 (-)</td><td align="center">2.0 (-)</td></tr><tr><td align="center">2</td><td align="center">8</td><td align="center">2.3 (1.2)</td><td align="center">1.4 (1.7)</td><td align="center">1.5 (1.1)</td></tr><tr><td align="center">3</td><td align="center">9</td><td align="center">3.3 (1.2)</td><td align="center">0.4 (1.3)</td><td align="center">1.7 (2.1)</td></tr><tr><td align="center">4</td><td align="center">14</td><td align="center">2.0 (1.2)</td><td align="center">1.3 (1.4)</td><td align="center">0.6 (1.2)</td></tr><tr><td align="center">5</td><td align="center">18</td><td align="center">1.1 (1.0)</td><td align="center">0.8 (1.6)</td><td align="center">1.9 (2.2)</td></tr><tr><td align="center">6</td><td align="center">20</td><td align="center">2.9 (1.2)</td><td align="center">0.2 (0.7)</td><td align="center">1.6 (1.6)</td></tr><tr><td align="center">7</td><td align="center">33</td><td align="center">1.8 (0.9)</td><td align="center">0.8 (1.6)</td><td align="center">1.8 (2.4)</td></tr><tr><td align="center">8</td><td align="center">82</td><td align="center">2.4 (1.2)</td><td align="center">1.1 (1.7)</td><td align="center">1.5 (2.2)</td></tr><tr><td align="center">9</td><td align="center">206</td><td align="center">2.8 (1.3)</td><td align="center">1.9 (2.3)</td><td align="center">2.3 (2.6)</td></tr><tr><td align="center"><bold>Total Children and Average Rates for All Intervention Offices</bold></td><td align="center">391</td><td align="center">2.5 (1.3)</td><td align="center">1.4 (2.1)</td><td align="center">1.9 (2.4)</td></tr></tbody></table><table-wrap-foot><p>Standard deviations are in parentheses.</p></table-wrap-foot></table-wrap></sec><sec><title>Intervention effects on dental service rates</title><p>Table <xref ref-type="table" rid="T4">4</xref> presents adjusted differences in the dental service rates between the intervention and control offices. In the fluoride regression model, the fluoride rate was 0.19 greater in intervention than control offices, ninety-five percent confidence interval (-.30, 0.79). The estimated difference in the sealant rate between the intervention and control offices was 0.10 per child, 95% confidence interval (-.29, .41). Restoration rates were significantly lower, -0.46, in the intervention offices than in the control offices, 95% confidence interval (-.88, .00), <italic>p</italic>=.05.</p><table-wrap position="float" id="T4"><label>Table 4</label><caption><p>Adjusted differences in dental service rates between intervention and control dental offices</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="center"><bold>Dental Service</bold></td><td align="center"><bold>Adjusted Difference in Rates (Intervention – Control)</bold></td><td align="center"><bold>95% confidence interval</bold></td><td align="center"><bold><italic>p</italic>-value</bold></td></tr></thead><tbody><tr><td align="left">Number of fluoride applications per child</td><td align="center">0.19</td><td align="center">(-.30, 0.79)</td><td align="center">0.46</td></tr><tr><td align="left">Number of sealants applied per child</td><td align="center">0.10</td><td align="center">(-.29, .41)</td><td align="center">0.50</td></tr><tr><td align="left">Number of restorations performed per child</td><td align="center">-0.46</td><td align="center">(-.88, .00)</td><td align="center">0.05</td></tr></tbody></table><table-wrap-foot><p> Control variables include time in study with dental capitation coverage, child characteristics (age, gender, current orthodontia, sum of dft and DFT, dentist-evaluated risk, number of sealed teeth at study entry, fluoride supplement use in restoration model), size of the office's capitation plan (number of children and number of total patients in capitation plan), and the number of hygienists employed by the office.</p></table-wrap-foot></table-wrap><p>About 45% of the children were enrolled in two offices in the intervention and control groups, which might be contributing more to Table <xref ref-type="table" rid="T4">4</xref> results than the other offices. As a sensitivity analysis, when the two practices were excluded from regression models, similar results were obtained, but the restoration rates were no longer significant. When dentist-evaluated risk was excluded from the models because of a high percentage of missing values, the intervention effect on restorations was no longer significant.</p></sec><sec><title>Parent survey</title><p>The parent follow-up survey had a 70% (n= 492) response rate. The children of parents who did or did not respond had similar age and gender (p > .05). Among survey respondents, the characteristics of children and parents were generally similar in the intervention and control groups (see Table <xref ref-type="table" rid="T5">5</xref>). However, the intervention group had a higher percentage of female parents and smaller households than the control group.</p><table-wrap position="float" id="T5"><label>Table 5</label><caption><p>Characteristics of children, parents and households inthe follow-up parent survey</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left"><bold>Variable</bold></td><td align="center"><bold>Intervention Group (n = 282)</bold></td><td align="center"><bold>Control Group (n = 205)</bold></td><td align="center"><bold><italic>p</italic>-value</bold></td></tr></thead><tbody><tr><td align="left"><underline>Children Characteristics</underline></td><td></td><td></td><td></td></tr><tr><td align="left">Percent female</td><td align="center">47</td><td align="center">45</td><td align="center">.30</td></tr><tr><td align="left">Average age (years)</td><td align="center">10 (2.4)</td><td align="center">10 (2.3)</td><td align="center">.50</td></tr><tr><td align="left">Percent brushing 2+ times daily</td><td align="center">53</td><td align="center">57</td><td align="center">.48</td></tr><tr><td align="left">Percent taking fluoride tablets or drops at home</td><td align="center">9</td><td align="center">11</td><td align="center">.34</td></tr><tr><td align="left">Percent eating snacks or drinking pop/juice 2+ times daily</td><td align="center">64</td><td align="center">59</td><td align="center">.28</td></tr><tr><td></td><td></td><td></td><td></td></tr><tr><td align="left"><underline>Respondent And Household Characteristics</underline></td><td></td><td></td><td></td></tr><tr><td align="left">Percent female</td><td align="center">80</td><td align="center">70</td><td align="center">.02</td></tr><tr><td align="left">Average age (years)</td><td align="center">40 (6.4)</td><td align="center">40 (6.7)</td><td align="center">.61</td></tr><tr><td align="left">Average education (years)</td><td align="center">14 (1.9)</td><td align="center">14 (2.3)</td><td align="center">.79</td></tr><tr><td align="left">Percent nonwhite</td><td align="center">9</td><td align="center">15</td><td align="center">.27</td></tr><tr><td align="left">Percent single</td><td align="center">15</td><td align="center">14</td><td align="center">.86</td></tr><tr><td align="left">Household size</td><td align="center">4.5 (1.5)</td><td align="center">4.7 (1.4)</td><td align="center">.04</td></tr></tbody></table><table-wrap-foot><p>Standard deviations are in parentheses.</p></table-wrap-foot></table-wrap><p>Based on parents' reports, Table <xref ref-type="table" rid="T6">6</xref> presents group differences in children's dental utilization in the past 12 months. A greater percentage of intervention children received fluoride varnish and sealants, but these differences were not significant in regression models. No significant differences in restorations were reported. Parents in the intervention group also reported their children had less dental fear than control children.</p><table-wrap position="float" id="T6"><label>Table 6</label><caption><p>Group differences in dental utilization, satisfaction with child's dental care, child's dental fear and oral health status in the follow-up parent survey</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left"><bold>Variable</bold></td><td align="center"><bold>Intervention Group (n = 282)</bold></td><td align="center"><bold>Control Group (n = 205)</bold></td><td align="center"><bold><italic>p</italic>-value</bold></td></tr></thead><tbody><tr><td align="left"><underline>Dental Utilization</underline></td><td></td><td></td><td></td></tr><tr><td align="left">Percentage children receiving:</td><td></td><td></td><td></td></tr><tr><td align="left"> Fluoride varnish</td><td align="center">65</td><td align="center">53</td><td align="center">.98</td></tr><tr><td align="left"> Sealants</td><td align="center">33</td><td align="center">27</td><td align="center">.63</td></tr><tr><td align="left"> Any restorations</td><td align="center">41</td><td align="center">46</td><td align="center">.69</td></tr><tr><td></td><td></td><td></td><td></td></tr><tr><td align="left"><underline>Satisfaction With Dental Care</underline></td><td></td><td></td><td></td></tr><tr><td align="left">Average satisfaction with dental care (5=excellent)</td><td align="center">4.1 (0.9)</td><td align="center">3.7 (1.0)</td><td align="center">.17</td></tr><tr><td align="left">Average satisfaction with preventive care (5=excellent)</td><td align="center">4.2 (0.9)</td><td align="center">3.7 (1.0)</td><td align="center">.10</td></tr><tr><td></td><td></td><td></td><td></td></tr><tr><td align="left"><underline>Dental Fear</underline></td><td></td><td></td><td></td></tr><tr><td align="left">Average dental fear score (5=very fearful)</td><td align="center">2.1 (0.7)</td><td align="center">2.3 (0.9)</td><td align="center">.04</td></tr><tr><td></td><td></td><td></td><td></td></tr><tr><td align="left"><underline>Oral Health Status</underline></td><td></td><td></td><td></td></tr><tr><td align="left">Average rating of child dental health now (5=excellent)</td><td align="center">3.8 (0.9)</td><td align="center">3.7 (0.9)</td><td align="center">.37</td></tr><tr><td align="left">Average rating of condition of teeth now compared to 1 year ago (1=much better)</td><td align="center">2.3 (0.8)</td><td align="center">2.5 (0.8)</td><td align="center">.08</td></tr></tbody></table><table-wrap-foot><p>Standard deviations are in parentheses.</p><p> Statistical tests adjust for children characteristics (gender, age, brushing, fluoride supplements, snacks) and adult/household characteristics (adult education, nonwhite race, age, gender, marital status, and household size).</p></table-wrap-foot></table-wrap></sec></sec><sec><title>Discussion</title><p>We invited 53 dentists who owned their offices and were in Washington Dental Service's capitation dental plan to participate in the study. Less than half consented. Among consenting dentists, fluoride and sealant rates were greater in the intervention dental offices than the control offices, but the differences between the two groups were not statistically significant, likely because of small sample sizes from low dentist participation and therefore, lower numbers of participating children. Similar findings were obtained in the parent survey. In addition, intervention (and control) dental offices rarely used the WDS procedure code for fluoride varnish in WDS service records and office charts, indicating incomplete adoption of the service (offices typically used the American Dental Association's Current Dental Terminology code 1203 for child topical application of fluoride)[<xref ref-type="bibr" rid="B45">45</xref>]. Only half of the intervention offices self-reported using fluoride varnish consistently for eligible children.</p><p>We estimated intervention effects on fluoride rates, rather than fluoride varnish, because we lacked information distinguishing whether children received fluoride varnishes or topical fluoride applications. This occurred partly because incentive payments were not contingent on accurate coding, although incentives did improve chart documentation in some medical studies[<xref ref-type="bibr" rid="B25">25</xref>]. Offices may not have documented fluoride varnish because no preventive procedure code exists for the service in the American Dental Association's Current Dental Terminology – an indicator that fluoride varnish is not integrated fully into the profession[<xref ref-type="bibr" rid="B45">45</xref>]. If the intervention actually increased fluoride varnish, detecting this effect is harder when fluoride is the outcome measure, and, therefore, our results are conservative. In future studies, monitoring office coding might increase use of the procedure code for fluoride varnish, but monitoring also may increase the delivery of fluoride varnish in control offices.</p><p>We examined dentist and office characteristics associated with participation in the study. The number of capitation children in dental offices was the strongest predictor of office participation[<xref ref-type="bibr" rid="B46">46</xref>]. Dental offices with a smaller number of capitation children were less likely to participate, probably because participation promised small financial rewards. This result suggests that if future studies recruit offices with large numbers of at risk children, office participation rates may be high.</p><p>We may have found small differences between groups because the financial incentive lacked sufficient "economic clout" to cause increased delivery of caries-control services in intervention offices. Most intervention offices enrolled less than 50 at risk children, and total financial incentives paid to those offices were likely a very small percentage of annual revenues. This argument also suggests that financial incentives might be more effective in offices with a substantial number of at risk children.</p><p>A related factor is that dental offices had patients with both fee-for-service and capitation dental plans. When the majority of patients in an office has fee-for-service dental plans, the financial incentives of the fee-for-service plan may "spillover" and may affect how a dentist treats patients in capitation plans [<xref ref-type="bibr" rid="B47">47</xref>-<xref ref-type="bibr" rid="B49">49</xref>]. Potential spillover effects from fee-for-service dental plans may have diluted the effects of an intervention targeting capitation patients, providing additional justification for recruiting offices with large majorities of capitation patients in future studies.</p><p>The small differences between the intervention and control groups also may be caused by the Hawthorne effect. Medical studies indicate that physician behavior can be altered simply by the awareness their behavior is being monitored[<xref ref-type="bibr" rid="B50">50</xref>,<xref ref-type="bibr" rid="B51">51</xref>]. Control offices were monitored closely by WDS in the follow-up period to track recall patterns of enrolled children, and control offices performed oral health assessments at the office visits of enrolled children and submitted this information to WDS. Control offices may have knowingly or unknowingly increased the provision of caries control services as a result of observation (WDS monitoring) and adherence to data collection protocols, which may have reduced the differences between groups.</p><p>Controlling for child and office characteristics, restoration rates were lower in the intervention group than the control group. However, when dentist-evaluated risk was excluded from the models because of a high percentage of missing values, the intervention effect on restorations was no longer significant. For the offices reporting caries risk, control and intervention offices classified a similar percentage of children at low caries risk (13% vs. 12%, respectively). Control offices classified a higher percentage of children at medium risk than high risk (32% vs 9%), while intervention offices had an opposite pattern (11% medium risk vs. 30% high risk). Thus, a difference exists only at the medium-to-high threshold, which may reflect differences in how the dentists evaluated caries risk as well as differences in the oral health status of the children. Percentages might change if the caries risk data were complete. In short, restoration rates also may depend on dentists' clinical decisions and other factors, and the intervention effect may be contingent on controlling statistically for dentists' assessment of children's caries risk in the restoration regression model[<xref ref-type="bibr" rid="B32">32</xref>,<xref ref-type="bibr" rid="B52">52</xref>].</p><p>Overall, the pattern of significant and nonsignificant results in Table <xref ref-type="table" rid="T4">4</xref> indicates that intervention offices provided more caries control services and less restorative services than control offices, which may justify replication of the study with a larger sample of dental offices and children. In a prior study, Lennon et al, report that compared to fee-for-service payment, capitation payment was associated with greater preventive services and less restorations for children[<xref ref-type="bibr" rid="B53">53</xref>]. Thus, a central question of future studies is whether payment schemes that blend fee-for-service reimbursement with capitation coverage produce a similar, beneficial mix of preventive and restorative services among children patients.</p><p>Parents in the intervention group reported their children had less dental fear, were more satisfied with their children's preventive care, and were more likely to report improvements in their children's oral health, although the satisfaction and oral health effects were weak. Dental fear may be less in the intervention group because provider education may have improved skills in delivering noninvasive preventive services, which may have reduced fear among intervention children. Because higher dental prices are associated with higher quality of dental care, [<xref ref-type="bibr" rid="B54">54</xref>] the financial incentives may have increased the quality of preventive services in intervention offices, which may have increased parent satisfaction with preventive care. Parent perceptions of improved oral health, along with the other beneficial intervention effects, provide support for replicating the study in larger children populations.</p><p>If a future study with larger sample sizes and adequate power does not find statistically significant intervention effects, the results would be consistent with Kuhn's model explaining paradigm shifts in scientific disciplines[<xref ref-type="bibr" rid="B22">22</xref>]. His model predicts that paradigm shift from the traditional surgical approach to a disease-based approach of dental practice is a long-term process. Stronger and more comprehensive, multi-pronged interventions appear essential for overcoming the inertia of paradigm shift and speeding up dentist adoption of caries-control technologies[<xref ref-type="bibr" rid="B34">34</xref>,<xref ref-type="bibr" rid="B55">55</xref>,<xref ref-type="bibr" rid="B56">56</xref>]. Dissemination efforts targeting a specific caries-control service may be ineffective when the real issue is increasing paradigm shift.</p><p>Our findings are limited to Seattle-area dentists in the provider network of Washington Dental Service's capitation dental plan and who consented to participate in the study. In addition, findings are limited to Seattle-area children aged 6–14 who were at risk for caries, covered by the capitation dental plan, and seen by participating dentists. A limitation of the study is inadequate sample sizes to detect small intervention effects. Greater financial incentives or different provider education may have different results.</p></sec><sec><title>Conclusion</title><p>Because dentists with greater numbers of capitation children were more likely to participate, we recommend that future studies increase sample sizes by recruiting dental offices which have a majority of patients in capitation dental plans, which also may reduce potential spillover effects. Findings suggest the intervention may have reduced children's dental fear.</p></sec><sec><title>Competing interests</title><p>The Caries Prevention Study was a collaboration between the University of Washington and Washington Dental Service (the Delta Dental insurance plan in Washington state), and the majority of funds for the study, including the financial incentives paid to dentists, were provided by Washington Dental Service. Given findings that the intervention did not increase the delivery of caries-control services to at risk children, any potential conflict of interest for the co-authors at Washington Dental Service may not be an issue.</p></sec><sec><title>Authors' contributions</title><p>All authors read and approved the final manuscript.</p><p>David Grembowski: principal investigator responsible for the conduct of the overall study and first author</p><p>Charles Spiekerman: biostatistician responsible for randomization of dentist offices, performance of data analyses, preparing data analysis and results sections of manuscript</p><p>Michael A. del Aguila: lead investigator at Washington Dental Service, with overall responsibility for overseeing the implementation of the study in the dental offices</p><p>Maxwell Anderson: vice president of Washington Dental Service who influenced the design of the study, secured WDS resources for the study, designed and participated in the provider education component of the intervention, and assisted in the interpretation of results</p><p>Debra Reynolds: lead field representative responsible for the day-to-day conduct of the study in the dental offices</p><p>Allison Ellersick: lead data manager responsible for managing day-to-day collection of study data, including the conduct of the mail survey of parents</p><p>James Foster: responsible for constructing the study's data base at Washington Dental Service</p><p>Leslie Choate: responsible for supporting all phases of data collection at Washington Dental Service</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1472-6831/6/7/prepub"/></p></sec>
Health costs in anthroposophic therapy users: a two-year prospective cohort study	<sec><title>Background</title><p>Anthroposophic therapies (counselling, special medication, art, eurythmy movement, and rhythmical massage) aim to stimulate long-term self-healing processes, which theoretically could lead to a reduction of healthcare use. In a prospective two-year cohort study, anthroposophic therapies were followed by a reduction of chronic disease symptoms and improvement of quality of life. The purpose of this analysis was to describe health costs in users of anthroposophic therapies.</p></sec><sec sec-type="methods"><title>Methods</title><p>717 consecutive outpatients from 134 medical practices in Germany, starting anthroposophic therapies for chronic diseases, participated in a prospective cohort study. We analysed direct health costs (anthroposophic therapies, physician and dentist consultations, psychotherapy, medication, physiotherapy, ergotherapy, hospital treatment, rehabilitation) and indirect costs (sick leave compensation) in the pre-study year and the first two study years. Costs were calculated from resource utilisation, documented by patient self-reporting. Data were collected from January 1999 to April 2003.</p></sec><sec><title>Results</title><p>Total health costs in the first study year (bootstrap mean 3,297 Euro; 95% confidence interval 95%-CI 3,157 Euro to 3,923 Euro) did not differ significantly from the pre-study year (3,186 Euro; 95%-CI 3,037 Euro to 3,711 Euro), whereas in the second year, costs (2,771 Euro; 95%-CI 2,647 Euro to 3,256 Euro) were significantly reduced by 416 Euro (95%-CI 264 Euro to 960 Euro) compared to the pre-study year. In each period hospitalisation and sick-leave together amounted to more than half of the total health costs. Anthroposophic therapies and medication amounted to 3%, 15%, and 8% of total health costs in the pre-study year, first year, and second study year, respectively. The cost reduction in the second year was largely accounted for by a decrease of inpatient hospitalisation, leading to a hospital cost reduction of 519 Euro (95%-CI 377 Euro to 904 Euro) compared to the pre-study year.</p></sec><sec><title>Conclusion</title><p>In patients starting anthroposophic therapies for chronic disease, total health costs did not increase in the first year, and were reduced in the second year. This reduction was largely explained by a decrease of inpatient hospitalisation. Within the limits of a pre-post design, study findings suggest that anthroposophic therapies are not associated with a relevant increase in total health costs.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Hamre</surname><given-names>Harald J</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Witt</surname><given-names>Claudia M</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Glockmann</surname><given-names>Anja</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Ziegler</surname><given-names>Renatus</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Willich</surname><given-names>Stefan N</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Kiene</surname><given-names>Helmut</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Health Services Research	<sec><title>Background</title><p>Complementary therapies are popular and extensively used. In Germany and Switzerland some complementary therapies are reimbursed within health care budgets. In these countries there has been a debate as to whether reimbursement of complementary therapies may lead to increased overall health expenditures [<xref ref-type="bibr" rid="B1">1</xref>].</p><p>Anthroposophic medicine (AM) was founded in the 1920s by Rudolf Steiner and Ita Wegman [<xref ref-type="bibr" rid="B2">2</xref>]. AM therapies are provided by physicians (counselling, AM medication) and non-medical therapists (AM art, eurythmy movement, and massage therapy) in inpatient and outpatient settings. AM aims to stimulate long-term self-healing processes in patients [<xref ref-type="bibr" rid="B3">3</xref>], which theoretically could lead to a reduction of healthcare use. Observational data suggest AM can be associated with cost savings [<xref ref-type="bibr" rid="B4">4</xref>], but no studies of total health costs have been undertaken.</p><p>The Anthroposophic Medicine Outcomes Study (AMOS) [<xref ref-type="bibr" rid="B5">5</xref>] provided an opportunity to investigate total health costs in AM users. AMOS is a prospective cohort study of outpatients starting AM therapies for chronic disease. The study was initiated by a health insurance company in conjunction with a health benefit project including reimbursement of AM therapies [<xref ref-type="bibr" rid="B5">5</xref>]. AM therapies were implemented during the first 3–6 months after study enrolment and were followed by a substantial reduction of disease severity and an improvement of quality of life [<xref ref-type="bibr" rid="B5">5</xref>]. In a first cost analysis, the pre-study year was compared to the first study year [<xref ref-type="bibr" rid="B5">5</xref>]. Here we present a cost analysis with a larger patient sample, including costs in the second study year.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Objective and design</title><p>The objective was to study health costs in AM therapy users from the societal perspective. For this purpose, we analysed health service use in a prospective cohort study of patients starting AM therapies for chronic disease, and calculated health costs. We calculated total (direct and indirect) costs in the first and second years after study entry and compared them to costs in the pre-study year. For each year we determined the relative size of AM therapy costs vs. total health costs. Exploratory subgroup analyses were performed for different age and therapy groups.</p></sec><sec><title>Setting, participants, and therapy</title><p>Participating physicians were certified by the Physicians' Association for Anthroposophical Medicine in Germany and had an office-based practice or worked in outpatient clinics in Germany. The physicians recruited consecutive patients starting AM therapy. Patients enrolled in the period 1 Jan 1999 to 31 March 2001 were included in the present analysis (18- and 24-month follow-ups were not performed for patients enrolled before 1 Jan 1999; n = 87) if they fulfilled eligibility criteria:</p><p>Inclusion criteria: (a) Outpatients aged 1–75 years, (b) referral to AM therapy (art, eurythmy or rhythmical massage), or initial AM-related consultation ≥ 30 min for any indication (main diagnosis), (c) at least three out of five follow-up questionnaires returned within the first two study years.</p><p>Exclusion criteria: previous AM therapy (art/eurythmy/rhythmical massage/AM-related consultation ≥ 30 min) for main diagnosis, respectively.</p><p>Therapy: Patients were treated according to the physician's discretion.</p></sec><sec><title>Outcome measures</title><p>Health costs, regardless of diagnosis, in the pre-study year and in the first and second study years: direct health costs (AM therapies, physician and dentist consultations, psychotherapy, medication, physiotherapy, ergotherapy, inpatient hospital and rehabilitation treatment), indirect costs (sick leave compensation).</p></sec><sec><title>Data collection</title><p>All data were documented with questionnaires sent in sealed envelopes to the study office. Physicians documented eligibility criteria and baseline health status; all other items were documented by patients. Patient responses were not made available to physicians. Physicians were compensated 40 Euro per included and fully documented patient, while patients received no compensation.</p><p>Data were entered twice by two different persons into Microsoft<sup>® </sup>Access 97. The two datasets were compared and discrepancies resolved by checking with the original data.</p></sec><sec><title>Quality assurance, adherence to regulations</title><p>The study was approved by the Ethics Committee of the Faculty of Medicine Charité, Humboldt University Berlin, and was conducted according to the Helsinki Declaration and ICH-GCP guidelines. Written informed consent was obtained from all patients before enrolment.</p></sec><sec><title>Data analysis</title><p>Data analysis (SPSS<sup>® </sup>13.0.1, StatXact<sup>® </sup>5.0.3, S-PLUS<sup>® </sup>7.0) was performed on all patients fulfilling eligibility criteria. For total and hospital costs, bootstrap means with bias-corrected and accelerated (BCa) bootstrap 95% confidence intervals (95%-CI) were calculated, using 2000 replications per analysis [<xref ref-type="bibr" rid="B6">6</xref>]. For other continuous data Wilcoxon Signed-Rank test was used for paired samples, Mann-Whitney U-test for independent samples; median differences with 95%-CI were estimated according to Hodges and Lehmann [<xref ref-type="bibr" rid="B7">7</xref>]. For binominal data McNemar test and Fisher's exact test were used. All tests were two-tailed. Significance criteria were p < 0.05 and 95%-CI not including 0.</p><p>Resource utilisation (therapies and health services) were analysed replacing missing data for each item and follow-up period by the group mean value. Costs were analysed from the perspective of the payer (employer: sick-leave costs for first six weeks; statutory health insurance: direct costs and sick-leave costs beyond first six weeks). Patient co-payment was not subtracted from direct costs.</p><p>Unit costs (Table <xref ref-type="table" rid="T1">1</xref>) were calculated from average costs per item in Germany, year 2000 value (physicians' and dentists' fees, medication, hospital, rehabilitation, sick-leave costs [<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B9">9</xref>]) or from reimbursement fees regulated in health care benefit catalogues (AM therapies, paraclinical investigations, psychotherapy, physiotherapy, ergotherapy [<xref ref-type="bibr" rid="B10">10</xref>-<xref ref-type="bibr" rid="B13">13</xref>]).</p><p>Hospital costs were calculated from average costs in each German federal state [<xref ref-type="bibr" rid="B14">14</xref>]. Physicians' fees were calculated from average fees of general practitioners + 12 specialist categories in the Accounting Data Record Panel of the Central Research Institute of Ambulatory Health Care in Germany [<xref ref-type="bibr" rid="B15">15</xref>]. Costs for paraclinical investigations (x-rays, computer tomography scans, nuclear magnetic resonance imaging and scintigrams) were calculated separately [<xref ref-type="bibr" rid="B11">11</xref>]. Costs of AM medication (any medication produced by the pharmaceutical companies Abnoba Arzneimittel GmbH, Helixor Heilmittel GmbH & Co, WALA Heilmittel GmbH, and Weleda AG) were calculated from average costs in 51 different price groups. Costs of other medications were calculated from national average costs in 86 Anatomical Therapeutic Chemical subgroups [<xref ref-type="bibr" rid="B16">16</xref>]. Sick-leave costs were calculated from national average gender-specific earnings for civil servants, salaried employees, and wage earners (100% compensation for sick-leave days 1–42, 70% compensation thereafter) [<xref ref-type="bibr" rid="B9">9</xref>]. Costs were not discounted.</p></sec></sec><sec><title>Results</title><sec><title>Participating physicians</title><p>153 physicians screened patients. 134 physicians had evaluable patients; these physicians did not differ significantly from all AM-certified physicians in Germany (n = 362) regarding gender (56.7% vs. 62.2% males), age (mean 46.3 ± 7.2 vs. 47.5 ± 7.9 years), number of years in practice (18.5 ± 7.5 vs. 18.9 ± 7.3 years), or the proportion of primary care physicians (86.6% vs. 85.0%).</p></sec><sec><title>Patient recruitment and follow-up</title><p>From 1 Jan 1999 to 31 March 2001, a total of 999 patients were screened for inclusion. 717 patients fulfilled all eligibility criteria and were included in the analysis (Figure <xref ref-type="fig" rid="F1">1</xref>). Included and not included patients did not differ significantly regarding age, gender, diagnosis, disease duration, or baseline symptom severity. The last patient follow-up ensued on 30 April 2003. 74.3% (533/717) of patients were enrolled by general practitioners, 10.2% by internists, 5.7% by paediatricians, and 9.8% by other specialist physicians. Physicians' setting was primary care practice (87.2% of patients, n = 625/717), referral practice (7.9%), and outpatient clinic (4.9%). Each physician enrolled median 3.0 patients (interquartile range IQR 2.0–7.0 patients).</p></sec><sec><title>Baseline characteristics</title><p>Most frequent main diagnoses, classified by ICD-10 (International Classification of Diseases, Tenth Edition), were F00-F99 Mental Disorders (31.8%, 228/717 patients), M00-M99 Musculoskeletal Diseases (19.0%), J00-J99 Respiratory Diseases (8.8%), and G00-G99 Nervous System Diseases (7.0%). Median disease duration was 3.0 (IQR 0.9–8.0) years. 78.8% (565/717) of patients had at least one comorbid disease, median 1.0 (IQR 1.0–3.0) comorbid diseases per patient. Most common comorbid diseases, classified by ICD-10, were M00-M99 Musculoskeletal Diseases (15.3%, 184/1206 diagnoses), F00-F99 Mental Disorders (14.3%), I00-I99 Circulatory System Diseases (8.3%), E00-E90 Endocrine, Nutritional and Metabolic Diseases, and J00-J99 Respiratory Diseases (8.0%). Patients were recruited from 15 of 16 German federal states. Median age was 39.0 (IQR 22.0–48.0) years. Compared to the German population, patients had higher educational and occupational levels, had less daily alcohol consumers and regular smokers, and were less overweight; patients' socio-demographic status was similar to the population regarding low-income, living alone, severe disability status, and sport; and less favourable for work disability pension and sick-leave (Table <xref ref-type="table" rid="T2">2</xref>).</p></sec><sec><title>Resource utilisation</title><p>Compared to the pre-study year, the use of AM therapies was increased during both the first and the second study year, medication use and psychotherapy increased in the first year but not in the second year, whereas the number of hospital and rehabilitation days decreased progressively and were significantly decreased in the second year (Table <xref ref-type="table" rid="T3">3</xref>).</p></sec><sec><title>Costs</title><p>Total health costs averaged 3,186 Euro (bootstrap mean 3,186 Euro; 95%-CI 3,037 Euro to 3,711 Euro) per patient in the pre-study year and 3,302 Euro (bootstrap mean 3,297 Euro; 95%-CI 3,157 Euro to 3,923 Euro) in the first study year, an increase of 123 Euro (95%-CI -391 Euro to +320 Euro) from the pre-study year. In the second study year, total costs were 2,768 Euro (bootstrap mean 2,771 Euro; 95%-CI 2,647 Euro to 3,256 Euro) a decrease of 416 Euro (95%-CI 264 Euro to 960 Euro) from the pre-study year (Table <xref ref-type="table" rid="T4">4</xref>). Cost distribution was highly skewed in all periods; in the first study year, 5% of patients caused 38% of all costs. In each year hospital costs and sick-leave compensation together amounted to more than half of the costs. Costs of AM therapies and medication amounted to 3%, 15%, and 8% of total health costs in the pre-study year, first year, and second study year, respectively.</p><p>In the first study year the largest cost differences from the pre-study year were observed for AM therapies (nominal increase of 347 Euro per patient) and inpatient hospital costs (nominal decrease of 310 Euro, estimated decrease of 314 Euro, 95%-CI 130 Euro to 753 Euro); in the second year the largest differences from the pre-study year were again for AM therapies (nominal increase of 108 Euro) and hospital costs (nominal decrease of 513 Euro, estimated decrease of 519 Euro; 95%-CI 377 Euro to 904 Euro). Other costs differed little (differences < 50 Euro per patient and year).</p><p>Total health costs were analysed in age and therapy groups (Table <xref ref-type="table" rid="T5">5</xref>). Average costs in the first study year varied by a factor of 3.3 between age groups (1–19 years: 1416 Euro, 40–59 years: 4646 Euro) and by a factor of 1.5 between AM therapy groups (medical: 2614 Euro, art therapy: 3706 Euro).</p></sec></sec><sec><title>Discussion</title><p>We analysed direct and indirect health costs in German outpatients starting AM therapies for chronic disease under routine conditions. Compared to the pre-study year, costs did not differ significantly in the first year after enrolment, whereas in the second year costs were significantly reduced by 13% (416 Euro per patient).</p><p>Strengths of this study include a large patient sample, a long follow-up period, high follow-up rates, and the participation of 37% of all AM-certified physicians in Germany. Participants resembled all eligible physicians with respect to socio-demographic characteristics, and included patients resembled not included, screened patients regarding baseline characteristics. These features suggest that the study to a high degree mirrors contemporary AM use in outpatient settings. Moreover, since patients with all diagnoses were included, our study offers a comprehensive picture of AM practice. Therefore, in the present early phase of economic AM evaluation, the inclusion of all diagnoses is an advantage. On the other hand, we did not attempt to separate disease-specific costs from overall health costs. Our analysis is comprehensive, including cost domains (physician and dentist services, psychotherapy, physiotherapy, ergotherapy, medication, inpatient treatment, sick-leave compensation) amounting to 87% of healthcare expenditures of the German Statutory Health System [<xref ref-type="bibr" rid="B8">8</xref>] (13% not analysed: dentures, medical appliances, nursing, patient transport, and health prevention programs).</p><p>A limitation of the study is the absence of a comparison group. We do not know if in similar patients in similar settings receiving conventional or no treatment, costs would have increased, been stable, or been reduced.</p><p>Another limitation is that cost analysis was not based on direct cost measurement but on patient self-reporting of resource utilisation, which can be affected by recall bias. In this pre-post analysis, however, any systematic recall bias would probably have been conservative, making results appear less favourable. The reason is: While at study entry patients were asked about therapies and health services during the preceding 12 months, these items were thereafter asked every six months (medicine use also after three months). Since patients' recall of resource utilisation declines over time with a net tendency towards under-reporting [<xref ref-type="bibr" rid="B17">17</xref>], under-reporting is more likely for the 12-month pre-study period than for the shorter periods after study entry.</p><p>Dropout is unlikely to have biased the analysis of resource utilisation: For this analysis, 88% of patients were evaluable. Moreover, there is no a priori reason to assume that therapies and health services are used differently by dropouts and respondents.</p><p>Since patients were treated by AM physicians who could possibly have an interest in AM therapies having favourable outcomes, study data were largely collected by patients and not physicians. For this analysis, any bias affecting physician's documentation would not affect outcomes (resource utilisation), since these outcomes were documented by patients. Also, physicians' documentation of baseline health status (main and comorbid diagnoses) did not affect patient recruitment, since patients were enrolled regardless of diagnoses.</p><p>Major determinants of cost changes were an increased use of AM therapies (corresponding to a cost increase from the pre-study year of 377 Euro and 116 Euro per patient in the first and second year, respectively) and a reduction of hospitalisation (corresponding to a cost reduction of 310 Euro and 513 Euro, respectively), whereas other costs differed by less than 50 Euro per year. The increase of AM therapies is a consequence of the study inclusion criterion of patients starting new AM therapies. The reduction in hospitalisation was paralleled by a reduction of disease severity and improvement in quality of life [<xref ref-type="bibr" rid="B5">5</xref>] and may thus be related to successful therapies or spontaneous improvement. Another possible cause is frequent or long hospitalisation early in the course of disease (diagnosis, therapy initiation) followed by a normalisation of hospitalisation rates. Sensitivity analysis, however, suggests that this factor could at maximum explain 37% of the hospitalisation reduction in the second year (primary analysis: decrease by average 1.78 days = 100%, compared to the pre-study year; patients with disease duration of at least one year: decrease by 1.13 days = 63%).</p><p>Moreover, changes in health-care implementation may affect the frequency and duration of hospital treatment. However, during this study, the average number of hospital days per person-year in Germany decreased by only 0.21 days (1999–2003: 2.07→1.86 days) [<xref ref-type="bibr" rid="B18">18</xref>]. This reduction of 0.11 days per two years corresponds to only 6% of the observed reduction of 1.78 days per two years in our study patients. Therefore, the reduced hospitalisation in our study cannot be explained by changes in health-care implementation. A possible setting-related cause of reduced hospitalisation is the policy of AM general practitioners to provide more comprehensive patient care and avoid unnecessary referrals to secondary care [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B19">19</xref>]. Study implications: The reduction of hospital treatment in this cohort following AM therapies is in accordance with other findings: In two Dutch studies [<xref ref-type="bibr" rid="B20">20</xref>,<xref ref-type="bibr" rid="B21">21</xref>] and a British NHS audit [<xref ref-type="bibr" rid="B22">22</xref>] patients of AM physicians had 10%–35% less hospital days than local or national averages.</p><p>In Germany, patients may use specialist health services without referral from a primary care physician, generating additional costs. Our study is the first economic analysis of AM therapies taking into account such direct health costs generated outside the AM setting, as well as indirect costs (sick-leave compensation). In the first study year, costs of AM therapies amounted to 15% of total health costs and were largely outweighed by the reduced hospital costs; therefore, total costs were not significantly increased from the pre-study-year (as found in our previous analysis of a smaller patient sample of this study [<xref ref-type="bibr" rid="B5">5</xref>]), and in the second year a cost decrease of 416 Euro per patient (bootstrap 95%-CI indicating a decrease of at least 264 Euro) was found.</p></sec><sec><title>Conclusion</title><p>In patients starting anthroposophic therapies for chronic disease, costs did not increase in the first year, in spite of the intensified therapy. In the second year, a reduction of costs was observed. This reduction was largely explained by a decrease of inpatient hospitalisation. Within the limits of a pre-post design, our findings suggest that anthroposophic healthcare in Germany is not associated with a relevant increase in total health costs.</p></sec><sec><title>Abbreviations</title><p>AM: anthroposophic medicine, AMOS: Anthroposophic Medicine Outcomes Study</p></sec><sec><title>Competing interests</title><p>HJH has received funding from WALA Heilmittel GmbH and Weleda AG, who produce anthroposophic medications. These companies did not finance this manuscript and had no influence on design, planning, conduct, analysis, interpretation, or publication of this study. Otherwise all authors declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>HJH, CMW, SNW, and HK contributed to study design. HJH, AG, and HK contributed to data collection. HJH, RZ, and HK wrote analysis plan, HJH, AG, and RZ analysed data. HJH was principal author of the paper, had full access to all data, and is guarantor. All authors contributed to manuscript drafting and revision and approved the final manuscript.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1472-6963/6/65/prepub"/></p></sec>
Identifying priorities in methodological research using ICD-9-CM and ICD-10 administrative data: report from an international consortium	<sec><title>Background</title><p>Health administrative data are frequently used for health services and population health research. Comparative research using these data has been facilitated by the use of a standard system for coding diagnoses, the International Classification of Diseases (ICD). Research using the data must deal with data quality and validity limitations which arise because the data are not created for research purposes. This paper presents a list of high-priority methodological areas for researchers using health administrative data.</p></sec><sec sec-type="methods"><title>Methods</title><p>A group of researchers and users of health administrative data from Canada, the United States, Switzerland, Australia, China and the United Kingdom came together in June 2005 in Banff, Canada to discuss and identify high-priority methodological research areas. The generation of ideas for research focussed not only on matters relating to the use of administrative data in health services and population health research, but also on the challenges created in transitioning from ICD-9 to ICD-10. After the brain-storming session, voting took place to rank-order the suggested projects. Participants were asked to rate the importance of each project from 1 (low priority) to 10 (high priority). Average ranks were computed to prioritise the projects.</p></sec><sec><title>Results</title><p>Thirteen potential areas of research were identified, some of which represented preparatory work rather than research <italic>per se</italic>. The three most highly ranked priorities were the documentation of data fields in each country's hospital administrative data (average score 8.4), the translation of patient safety indicators from ICD-9 to ICD-10 (average score 8.0), and the development and validation of algorithms to verify the logic and internal consistency of coding in hospital abstract data (average score 7.0).</p></sec><sec><title>Conclusion</title><p>The group discussions resulted in a list of expert views on critical international priorities for future methodological research relating to health administrative data. The consortium's members welcome contacts from investigators involved in research using health administrative data, especially in cross-jurisdictional collaborative studies or in studies that illustrate the application of ICD-10.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>De Coster</surname><given-names>Carolyn</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I8">8</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Quan</surname><given-names>Hude</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Finlayson</surname><given-names>Alan</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Gao</surname><given-names>Min</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Halfon</surname><given-names>Patricia</given-names></name><xref ref-type="aff" rid="I5">5</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Humphries</surname><given-names>Karin H</given-names></name><xref ref-type="aff" rid="I6">6</xref><email>[email protected]</email></contrib><contrib id="A7" contrib-type="author"><name><surname>Johansen</surname><given-names>Helen</given-names></name><xref ref-type="aff" rid="I7">7</xref><email>[email protected]</email></contrib><contrib id="A8" contrib-type="author"><name><surname>Lix</surname><given-names>Lisa M</given-names></name><xref ref-type="aff" rid="I8">8</xref><email>[email protected]</email></contrib><contrib id="A9" contrib-type="author"><name><surname>Luthi</surname><given-names>Jean-Christophe</given-names></name><xref ref-type="aff" rid="I5">5</xref><email>[email protected]</email></contrib><contrib id="A10" contrib-type="author"><name><surname>Ma</surname><given-names>Jin</given-names></name><xref ref-type="aff" rid="I9">9</xref><email>[email protected]</email></contrib><contrib id="A11" contrib-type="author"><name><surname>Romano</surname><given-names>Patrick S</given-names></name><xref ref-type="aff" rid="I10">10</xref><email>[email protected]</email></contrib><contrib id="A12" contrib-type="author"><name><surname>Roos</surname><given-names>Leslie</given-names></name><xref ref-type="aff" rid="I8">8</xref><email>[email protected]</email></contrib><contrib id="A13" contrib-type="author"><name><surname>Sundararajan</surname><given-names>Vijaya</given-names></name><xref ref-type="aff" rid="I11">11</xref><email>[email protected]</email></contrib><contrib id="A14" contrib-type="author"><name><surname>Tu</surname><given-names>Jack V</given-names></name><xref ref-type="aff" rid="I12">12</xref><email>[email protected]</email></contrib><contrib id="A15" contrib-type="author"><name><surname>Webster</surname><given-names>Greg</given-names></name><xref ref-type="aff" rid="I13">13</xref><email>[email protected]</email></contrib><contrib id="A16" contrib-type="author"><name><surname>Ghali</surname><given-names>William A</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><xref ref-type="aff" rid="I14">14</xref><email>[email protected]</email></contrib>	BMC Health Services Research	<sec><title>Background</title><p>Health administrative data are frequently used for health research in Canada and abroad. In the past two decades, such data have been widely employed by health services and population health researchers to study healthcare outcomes, effectiveness, appropriateness and utilization of healthcare services, and to investigate or monitor population health status and its determinants [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B11">11</xref>]. The varied and broad use of administrative data has been facilitated by important advantages of the data, including their accessibility, their wide geographic coverage and their relatively complete capture of contacts with the health system for a defined population [<xref ref-type="bibr" rid="B12">12</xref>,<xref ref-type="bibr" rid="B13">13</xref>].</p><p>The use of health administrative data in health services research has been enabled by some key characteristics, notably the use of a standard system for coding diagnoses, the International Classification of Diseases (ICD). Established by the World Health Organization in 1893 to categorise causes of death, this system adopts a standardised format to code diagnoses, thereby enabling longitudinal and comparative studies [<xref ref-type="bibr" rid="B14">14</xref>]. The ninth revision, ICD-9, was expanded in 1977 to ICD-9-CM (Clinical Modification) to enable more precision in diagnostic codes, together with the addition of surgical intervention codes. In 1992, the 10<sup>th </sup>Revision of ICD (ICD-10) was introduced. ICD-10 has been used by many countries throughout the world for coding cause of death and for hospital diagnoses since 1994 [<xref ref-type="bibr" rid="B15">15</xref>-<xref ref-type="bibr" rid="B17">17</xref>]. It has been used for mortality data since 2000 in Canada, and provinces have adopted ICD-10 for coding hospital diagnoses in a phased approach, beginning in 2001.</p><p>One of the major advantages of ICD-10 is that it is far more detailed (there are a total of 12,420 codes in ICD-10 compared to 6,969 in ICD-9), permitting richer capture of clinical information. However, its implementation means that a number of established methodological tools applicable to ICD-9 or ICD-9-CM need to be redesigned for application in ICD-10. Another issue is that the structure of ICD-10 differs substantially from ICD-9. Furthermore, since each country licences the coding system individually from WHO and can create its own modifications, there may be more opportunity for discrepancies between countries. Finally, ICD-10 does not include procedure codes and so each country has developed its own coding system. The system used by Canada is the International Classification of Diseases, 10<sup>th </sup>revision, Canadian version, Canadian Classification of Health Interventions (ICD-10-CA/CCI).</p><p>Clearly the implementation of ICD-10 offers many benefits while also raising significant challenges for the international health services and population health research communities. In addition, research using ICD administrative data must address other limitations, largely stemming from the fact that the data were created not for research but for other purposes. Data quality is a concern; errors in the data can stem from inaccurate or missing information in the patient record, from the failure to abstract relevant data, or from incorrect coding of the abstracted data. Another concern is that administrative data lack clinical details. Even when data quality is good, the diagnoses that are coded do not reflect the severity of disease, diagnostic findings are not coded, and clinical sequence is not available.</p><p>This paper describes the origins and first symposium of a new international group that has come together to discuss how to take advantage of these potential benefits, and to address the new and ongoing challenges associated with using administrative data in health services and population health research. International collaborative research on health services has many advantages. From the methodological perspective, such research allows investigators to develop analytic tools that are more robust and more generalisable. It also allows those tools to be adopted in a systematic and uniform manner across countries, thereby fostering international exchange of research data and findings. From the policy perspective, it helps us to understand the strengths and weaknesses of various healthcare systems, and identifies opportunities for improvement in those systems.</p><sec><title>The consortium</title><p>The consortium came together through a fortuitous set of circumstances. Australian researcher Vijaya Sundararajan contacted Canadian researchers William Ghali and Hude Quan because they were all doing similar work. While on sabbatical, William Ghali met Swiss researchers with similar interests: Patricia Halfon, Jean-Christophe Luthi and Bernard Burnand. These links led to two initial collaborative projects: new ICD-10 coding algorithms for two widely-used comorbidity measures, the Charlson index and the Elixhauser comorbidity categories [<xref ref-type="bibr" rid="B18">18</xref>].</p><p>Meanwhile the Canadian Institutes of Health Research (CIHR) announced a funding opportunity for workshops. A successful proposal by Ghali and Quan to the Institute for Health Services and Policy Research permitted a seminar and workshop held June 17 and 18, 2005 in Calgary and Banff, Alberta. The objectives of the workshop were to:</p><p>1) solidify collaborative relationships through a face-to-face meeting of researchers;</p><p>2) initiate dialogue around launching a set of collaborative research projects on methodological issues surrounding the use of administrative data; and</p><p>3) stage a symposium in parallel to the workshop meetings at which the invited researchers would present their work to interested attendees.</p><p>Additional invitees to the seminar and workshop included representatives from two stakeholder organizations (Canadian Institute for Health Information (CIHI), and Statistics Canada), five Canadian collaborators, and investigators from the United States, the United Kingdom, Australia, Switzerland and China. The list of invited participants was a convenience sample whose selection was based on two criteria: they were bona fide experts in this area and/or they were known to the organisers.</p></sec></sec><sec sec-type="methods"><title>Methods</title><p>The Seminar was held on the morning of June 17 at the Faculty of Medicine, University of Calgary. Members of the international consortium gave 11 presentations to an audience of approximately 100 people, with participants from not only Calgary, but also Edmonton, Vancouver and Ontario. The workshop presentations included descriptions of administrative data systems in Switzerland, Scotland and China, and the use of administrative data to measure comorbidities, chronic disease prevalence, quality of care and waiting times.</p><p>The research planning workshop followed on Saturday, June 18 in Banff. The atmosphere was informal and collaborative. The morning sessions covered such topics as the validity of administrative data, analysis of administrative data by Statistics Canada, premature mortality in Scotland and Europe, and opportunities for using CIHI data for research. The group then engaged in a focussed discussion around ideas for future collaborative research projects necessary to advance this field. The emphasis in this research planning discussion was on high-priority methodological areas in need of research that the consortium could undertake collectively in future work. Some of the areas identified represent preparatory work rather than research <italic>per se</italic>.</p></sec><sec><title>Results</title><p>Thirteen potential areas of research were identified.</p><p>1. 'Meta-data' documentation of international administrative data: Every field in each country's hospital administrative data system would be defined and described. While not as exciting as more applied projects, a compilation of this nature would be necessary for international comparative studies, and would also serve to highlight identified problems or issues with the data from specific countries.</p><p>2. International cross-validation of new ICD-10 coding algorithms. ICD-10 versions of the Charlson and Elixhauser comorbidity indices have been developed, as mentioned previously. There has been some initial work comparing the results of the new Charlson coding algorithms across countries, but more work is necessary. ICD-10 coding algorithms need to be developed in other areas, for example chronic diseases, along with additional international comparisons</p><p>3. Patient safety indicators (PSI) translation: PSIs have been developed using ICD9-CM coding, under the auspices of the U.S. Agency for Healthcare Research and Quality, but corresponding ICD-10 codes for these indicators have not yet been developed. The PSIs are designed to screen for potentially preventable adverse effects of hospitals care. By translating the PSIs into ICD-10 and then validating this translation using data that have been independently coded according to both ICD-9-CM and ICD-10, researchers will be able to compare inpatient safety across national boundaries.</p><p>4. Learning curves: This effort would focus on the timing of uptake of ICD-10, and whether data validity assessments indicate the presence of a learning curve for coding. Canada, with its phased implementation in multiple provinces over several years, would be an ideal setting for this type of work.</p><p>5. Training standards for health record coders: It was discovered at the workshop that hospital abstract coders receive very different training from country to country. This project would explore those issues further with formal documentation of training requirements and practice guidelines for health record coders in various countries.</p><p>6. Chart-Database comparison studies: This would involve medical record reviews to determine the validity of hospital abstract data compared with the patient record across multiple countries. These are very expensive studies, especially if international comparisons are involved, but they would help researchers to characterise the importance of reporting and coding bias in international studies using administrative healthcare data.</p><p>7. Internal consistency algorithms: Algorithms can be developed to verify the logic of codes. For example, diabetic retinopathy should not occur in a patient who has never had a diagnosis of diabetes; prostatectomies cannot occur in females. Some work of this type has already been done in Switzerland and California. Different algorithms could be tested, refined, validated and then made available to others.</p><p>8. "True" gold standard: The purpose of this research would be to verify whether the trusted gold standard in observational health research, the patient's medical record, is in fact valid when compared to a 'truer' gold standard of information collected prospectively from patients and providers during a medical encounter. This research would require real-time patient assessments by independent clinicians who would observe all of the patient's interactions with physicians, as well as all of the discussions among the physicians involved in establishing and treating the patient's diagnosis. Comparisons would then be made between the independent assessment, the patient record, a nurse reviewer, and administrative data.</p><p>9. Travelling coders for comparative recoding: This research would require travelling coders who would recode previously coded records across countries to assess uniformity. By using a single team of travelling coders, researchers could estimate the nature and magnitude of international differences in coding practices.</p><p>10. Interventional studies to enhance coding quality: This research might include, for example, randomised controlled trials or pre-post studies to determine the effectiveness of educational or system interventions aimed at improving coding quality.</p><p>11. Value of diagnosis type coding: Some countries (or individual states or provinces) include a diagnosis-type code indicating whether each diagnosis is a comorbidity or a complication. Research in this area would focus on demonstrating the value of diagnosis-type codes, their validity, and the economic and human resources impact of implementation.</p><p>12. International comparisons of predictive model performance, as measured by the C (concordance)-statistic: It was determined from the group's presentations that C-statistic values differ across countries in comorbidity-based mortality predictions, but it is not understood why. The C-statistic is a measure of the discriminative accuracy of a logistic regression model [<xref ref-type="bibr" rid="B19">19</xref>,<xref ref-type="bibr" rid="B20">20</xref>]. The difference in C-statistic values may depend on the number of diagnosis fields available in abstracts, as well as the underlying coding validity and the epidemiology of disease in the population. Research in this area would aim to uncover the factors that contribute to the observed differences in model performance.</p><p>13. International scan of privacy considerations across countries and implications regarding permissible linkage activities: Discussions at the workshop revealed that there are considerable differences between countries in permissible data linkage activities, which have a great impact on the types of health services research that is possible.</p><p>After the brain-storming session, voting took place to rank-order the suggested projects. Participants were asked to rate the importance of each project from 1 (low importance), to 10 (high importance). Average ranks were computed to prioritise the projects (Table <xref ref-type="table" rid="T1">1</xref>). While all projects were considered to be of at least moderate importance, several priorities emerged, in particular, research into international meta-data documentation and translation of patient safety indicators.</p></sec><sec><title>Discussion</title><p>Objectives were achieved; the workshop was considered by all to be a big success and a memorable event. Valuable face-to-face contacts were made and the addition of outdoor activities on Sunday June 19 helped to solidify linkages between participants. The group discussions resulted in a list of expert views on critical international priorities for future methodological research relating to health administrative data. It must be acknowledged, however, that the list was limited by the experience and knowledge of the experts who attended the meeting and as such, it is certainly possible that the list omits key issues that others would consider to be important.</p><p>Since the symposium, work has continued. A paper is in preparation comparing three ICD-10 translations of the Charlson comorbidity index that were developed in Switzerland, Australia, and Canada. Within Canada, trends in the coding of Charlson comorbidities are being analyzed, assessing the impact and learning curve associated with the phased introduction of ICD-10. Preparatory dialogue is underway to plan the implementation of additional projects in the research areas outlined in the table.</p><sec><title>Knowledge exchange</title><p>The consortium is committed to the dissemination and sharing of knowledge with the broader health services and population health research communities. The PowerPoint presentations from the seminar are available on the website of the Centre for Health and Policy Studies, University of Calgary [<xref ref-type="bibr" rid="B21">21</xref>]. Useful websites which describe methodological tools, key concepts and operational definitions emanating in part from the work of consortium members include the Manitoba Centre for Health Policy's concept index [<xref ref-type="bibr" rid="B22">22</xref>], the Centre for Health and Policy Studies [<xref ref-type="bibr" rid="B21">21</xref>], the Institut Universitaire de Médecine Sociale et Préventive [<xref ref-type="bibr" rid="B23">23</xref>], AHRQ's quality indicators [<xref ref-type="bibr" rid="B24">24</xref>], and the Canadian Institute for Health Information [<xref ref-type="bibr" rid="B25">25</xref>].</p><p>The consortium's members welcome contacts from investigators involved in research using health administrative data, especially in cross-jurisdictional collaborative studies and/or in studies that illustrate the application of ICD-10. All attendees indicated commitment to carry forward the enthusiasm evident at this inaugural workshop, and hoped to hold future consortium meetings to advance the exciting and important work of this international group.</p></sec></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>HQ and WAG obtained the funding for the workshop, organised and facilitated it, and assisted in drafting the manuscript; CD drafted the manuscript; all participants were actively engaged in the workshop, contributed to the discussions and ratings about priority areas for future collaborative research, and participated in revising the paper and reviewing the final version for submission.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1472-6963/6/77/prepub"/></p></sec>
Predictors of opioid misuse in patients with chronic pain: a prospective cohort study	<sec><title>Background</title><p>Opioid misuse can complicate chronic pain management, and the non-medical use of opioids is a growing public health problem. The incidence and risk factors for opioid misuse in patients with chronic pain, however, have not been well characterized. We conducted a prospective cohort study to determine the one-year incidence and predictors of opioid misuse among patients enrolled in a chronic pain disease management program within an academic internal medicine practice.</p></sec><sec sec-type="methods"><title>Methods</title><p>One-hundred and ninety-six opioid-treated patients with chronic, non-cancer pain of at least three months duration were monitored for opioid misuse at pre-defined intervals. Opioid misuse was defined as: 1. Negative urine toxicological screen (UTS) for prescribed opioids; 2. UTS positive for opioids or controlled substances not prescribed by our practice; 3. Evidence of procurement of opioids from multiple providers; 4. Diversion of opioids; 5. Prescription forgery; or 6. Stimulants (cocaine or amphetamines) on UTS.</p></sec><sec><title>Results</title><p>The mean patient age was 52 years, 55% were male, and 75% were white. Sixty-two of 196 (32%) patients committed opioid misuse. Detection of cocaine or amphetamines on UTS was the most common form of misuse (40.3% of misusers). In bivariate analysis, misusers were more likely than non-misusers to be younger (48 years vs 54 years, p < 0.001), male (59.6% vs. 38%; p = 0.023), have past alcohol abuse (44% vs 23%; p = 0.004), past cocaine abuse (68% vs 21%; p < 0.001), or have a previous drug or DUI conviction (40% vs 11%; p < 0.001%). In multivariate analyses, age, past cocaine abuse (OR, 4.3), drug or DUI conviction (OR, 2.6), and a past alcohol abuse (OR, 2.6) persisted as predictors of misuse. Race, income, education, depression score, disability score, pain score, and literacy were not associated with misuse. No relationship between pain scores and misuse emerged.</p></sec><sec><title>Conclusion</title><p>Opioid misuse occurred frequently in chronic pain patients in a pain management program within an academic primary care practice. Patients with a history of alcohol or cocaine abuse and alcohol or drug related convictions should be carefully evaluated and followed for signs of misuse if opioids are prescribed. Structured monitoring for opioid misuse can potentially ensure the appropriate use of opioids in chronic pain management and mitigate adverse public health effects of diversion.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Ives</surname><given-names>Timothy J</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A2" corresp="yes" contrib-type="author"><name><surname>Chelminski</surname><given-names>Paul R</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Hammett-Stabler</surname><given-names>Catherine A</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Malone</surname><given-names>Robert M</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Perhac</surname><given-names>J Stephen</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Potisek</surname><given-names>Nicholas M</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A7" contrib-type="author"><name><surname>Shilliday</surname><given-names>Betsy Bryant</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A8" contrib-type="author"><name><surname>DeWalt</surname><given-names>Darren A</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A9" contrib-type="author"><name><surname>Pignone</surname><given-names>Michael P</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib>	BMC Health Services Research	<sec><title>Background</title><p>The past decade and a half has witnessed an expansion of opioid analgesic use for patients who have chronic non-cancer pain [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B5">5</xref>]. The misuse of opioid analgesics, however, is a growing public health problem [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B7">7</xref>]. National surveys show that opioid misuse has increased dramatically over the past decade and that opioid medications have surpassed cocaine and heroin use as the leading drugs of abuse [<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B9">9</xref>]. Utah and North Carolina have documented dramatic increases in unintentional overdose deaths from opioid analgesics diverted from their intended medical use [<xref ref-type="bibr" rid="B10">10</xref>,<xref ref-type="bibr" rid="B11">11</xref>]. The increased misuse is also reflected in the trauma literature which reports increases in opioid use among patients admitted to trauma centers [<xref ref-type="bibr" rid="B12">12</xref>]. As an ongoing response to the long-standing public health problem of prescription drug diversion, (as of May 2005), at least 28 states have established or are in the process of enacting legislation to establish prescription monitoring systems for controlled substances, and the medical literature is beginning to examine their effectiveness [<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B14">14</xref>].</p><p>Chronic pain is recognized as another important public health problem that is often undertreated [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B15">15</xref>,<xref ref-type="bibr" rid="B16">16</xref>]. Experts advocate the use of opioids in a carefully selected "subset" of patients with chronic non-cancer pain, but few data are available to guide selection of patients for whom opioids are likely to have net benefit [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B17">17</xref>]. The limited clinical trial data on opioid use in chronic pain derives mainly from small trials in highly selected patients seen in specialty settings [<xref ref-type="bibr" rid="B18">18</xref>-<xref ref-type="bibr" rid="B22">22</xref>]. The decision of whether and how providers should use these agents in a primary care setting, however, falls largely on expert opinion and clinical judgment. Generalists are faced with the dilemma of balancing the pain-relieving properties of opioids in selected patients with chronic pain against the reality that some patients may misuse and divert these medications. In effect, they are balancing one public health priority – the relief of suffering from pain – against another, the mitigation of substance misuse.</p><p>The incidence and prevalence of opioid misuse in patients treated for chronic pain is unclear and remains a topic of debate. Little is known about the factors predisposing patients to opioid misuse in the outpatient setting. Although histories of drug or alcohol abuse are commonly accepted proxies for patients at risk for opioid abuse [<xref ref-type="bibr" rid="B23">23</xref>], few epidemiologic data are available that clearly define risk factors for opioid misuse by chronic pain patients [<xref ref-type="bibr" rid="B24">24</xref>]. Most studies have been small (less than 50 patients) or were conducted with patients who were receiving substance abuse treatment, such as patients enrolled in methadone treatment clinics [<xref ref-type="bibr" rid="B23">23</xref>-<xref ref-type="bibr" rid="B26">26</xref>]. A case-control study of 533 hospitalized patients identified previous substance abuse, ongoing alcohol abuse, and urine toxicological screens positive for opiates as risk factors for misuse, but this study focused on inpatients hospitalized in a drug addiction unit and did not address the question of substance misuse in pain patients [<xref ref-type="bibr" rid="B27">27</xref>]. Other studies of misuse conducted in pain specialty clinics have relied on surveys and retrospective chart reviews, but did not monitor patients prospectively for predefined clinical outcomes [<xref ref-type="bibr" rid="B28">28</xref>-<xref ref-type="bibr" rid="B30">30</xref>]. Generalization of their findings to a primary care setting is limited.</p><p>We sought to determine the one-year incidence and predictors of opioid misuse in a cohort of patients enrolled in a chronic pain disease management program within an academic general internal medicine practice.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Patient recruitment</title><p>This study was conducted in patients with chronic pain who were referred to a chronic pain disease management program within an academic internal medicine practice [<xref ref-type="bibr" rid="B31">31</xref>]. Patients were eligible if they had non-cancer pain of greater than three months duration, and we encouraged referral of patients whose pain was considered difficult to manage and in whom opioid misuse was suspected. Patients were managed by a multidisciplinary team in consultation with the patient's primary care physician. The team was composed of a clinical pharmacist practitioner, an internist, a psychiatrist with sub-specialization in pain medicine, a nurse, and a program assistant. Patients were seen initially at monthly intervals during the medication titration phase. In addition to standard non-pharmacotherapeutic modalities, the use of anti-inflammatory agents, adjunctive analgesics, or long-acting (e.g., methadone) or sustained-release opioid agents (e.g., morphine ER) were preferred. Once patients achieved adequate, stable pain control with a proportionate improvement in function, they were scheduled to return every three months for monitoring of pain, depression, functional status, and misuse.</p></sec><sec><title>Defining and identifying opioid misuse</title><p>At enrollment, patients signed a medication agreement at enrollment [<xref ref-type="bibr" rid="B32">32</xref>], specifying the conditions under which opioids or controlled substances (O/CS) would be prescribed. Patients agreed to the following:</p><p>• To receive O/CS only from this practice.</p><p>• To use a single pharmacy.</p><p>• Not to sell or share medication.</p><p>• Not to abuse alcohol or illicit drugs (e.g. cocaine).</p><p>• That lost, stolen, or misplaced medication would generally not be replaced and that consideration of replacement would only occur at a clinic visit.</p><p>• That requests for medication renewals would occur only during regular clinic business hours, and not by telephone request.</p><p>• That regular urine toxicological screens would be performed, and</p><p>• That background checks for criminal drug and alcohol convictions would be performed.</p><p>As stipulated in the medication agreement, we prospectively monitored for misuse through clinical history, review of medications, review of outside medical records, communication with pharmacies and providers, and urine toxicological screening (UTS) [<xref ref-type="bibr" rid="B33">33</xref>,<xref ref-type="bibr" rid="B34">34</xref>]. Prescriptions for O/CS were documented both in the institutional electronic medical record and our disease management program database. Discrepancies and inconsistencies in opioid medication use were discussed with the patient's primary provider. Pharmacies were contacted to verify procurement of O/CS medications, and, if misuse was suspected, additional pharmacies were contacted to ascertain whether or not a patient was receiving opioids from multiple sources.</p><p>We defined opioid misuse prospectively as any of the following:</p><p>1. Negative UTS: Defined as UTS negative on at least two occasions for prescribed O/CS in the context of a reported history that the patient was taking the medication as prescribed (Repeatedly "negative" urines were considered an indicator of possible diversion)</p><p>2. Inconsistent UTS: Defined as UTS positive on at least two occasions for O/CS medications not prescribed by our practice</p><p>3. Doctor collecting: Evidence of concurrent procurement of O/CS from multiple providers</p><p>4. Diversion of O/CS</p><p>5. Prescription forgery</p><p>6. Stimulant positive (cocaine or amphetamine) UTS: Evidence of cocaine or amphetamines in the urine while being prescribed opioids was considered opioid misuse because it was in violation of the patient's medication agreement and because concurrent use of cocaine and amphetamines was felt to increase the risk of diversion in order to procure additional stimulants.</p><p>Urine toxicology screening included immunoassays for opiates, amphetamines, cannabinoids, benzodiazepines, methadone, propoxyphene, cocaine metabolite, and barbiturates. Testing was conducted at each visit and was correlated with the patient's reported history of O/CS use. In collaboration with our institution's toxicologist, results of the UTS were verified using gas chromatography/mass spectrophotometry (GC/MS) confirmatory assays. Because the UTS <italic>opiate </italic>assay has greater sensitivity and specificity for morphine and codeine, the presence or absence (i.e., inappropriately negative when it should have been positive) of other opiates (i.e., hydromorphone, oxycodone, hydrocodone, oxymorphone) were also confirmed by GC/MS. All positive results for amphetamines were confirmed with GC/MS to exclude the possibility of assay interference from other medications [<xref ref-type="bibr" rid="B34">34</xref>]. Urine samples were tested for low urine creatinine levels (i.e., < 20 ng/mL) to detect inappropriately diluted samples.</p><p>A single positive cannabinoid finding on UTS was not defined as misuse, but patients with multiple chronic positive results were strongly counseled to refrain from use of marijuana. Continued positive UTS for cannabinoids were tracked, however, to examine this variable as a potential predictor of opioid misuse as defined above. Neither past drug or alcohol abuse, nor past drug or alcohol criminal convictions disqualified patients from participating in our program, or receiving opioids within it.</p><p>Patients were advised at entry into the program that that the aforementioned violations would result in discontinuation of O/CS. A formal committee was constituted to evaluate and respond to instances of opioid misuse. It consisted of the practice medical director, two other attending physicians, the program clinical pharmacist practitioner, two resident physicians, and a nurse. The committee deliberated through secure e-mail. Patients committing opioid misuse were offered referral to substance abuse experts at our institution, or in their respective communities. Practice policy stipulated that the reinstitution of O/CS therapy could occur if the patient completed 6 months of substance abuse counseling. Patients who forged prescriptions were subject to dismissal from the practice.</p></sec><sec><title>Predictors of opioid misuse</title><p>Patients provided informed consent and underwent a comprehensive baseline medical assessment that included collection of socio-demographic data, assessment of pain, disability, mood, and literacy, using validated scales. Using the 11-point Brief Pain Inventory (BPI), patients rated their current pain and their pain at its worst, least, and average over the past month [<xref ref-type="bibr" rid="B35">35</xref>,<xref ref-type="bibr" rid="B36">36</xref>]. The seven-item Pain Disability Index (PDI), a measure of pain-related disability, asked patients to rate the degree of disability on a 10-point scale [<xref ref-type="bibr" rid="B37">37</xref>-<xref ref-type="bibr" rid="B39">39</xref>]. To assess depression, the Center for Epidemiological Studies-Depression Scale (CES-D) was used [<xref ref-type="bibr" rid="B40">40</xref>]. This twenty-item tool rates affective symptoms on a scale of 0 to 3. Literacy was measured using Rapid Estimate of Adult Literacy in Medicine (REALM) instrument [<xref ref-type="bibr" rid="B41">41</xref>], a word recognition test that assesses reading ability and uses health care terms. Previous history of cocaine or alcohol abuse was assessed by self report. Past criminal convictions for drug, or driving while impaired violations were researched using the publicly accessible database of the North Carolina Department of Correction Public Access Information System [<xref ref-type="bibr" rid="B42">42</xref>].</p></sec><sec><title>Analysis</title><p>Opioid misuse, as defined above, was the primary outcome of interest. The misuse categories described are presented individually as counts and proportions and are combined as a composite outcome in logistic regression analysis. Predictors of both opioid and other drug misuse were examined in bivariate and multivariate analyses. Bivariate analyses are reported as proportions and relative risks, with p-values and 95% confidence intervals (CI) for dichotomous variables and t-tests for continuous variables. All exposure variables with a p-value of <0.1 were analyzed in multivariate modeling. Models were reduced using the Maximum Likelihood Ratio Test. Statistical analyses were performed using Stata 7.0 (Stata Corporation, College Station, TX). The research protocol was approved by the Committee on the Protection of the Rights of Human Subjects, School of Medicine, University of North Carolina at Chapel Hill and patients provided informed consent.</p></sec></sec><sec><title>Results</title><p>Between December 2002 and December 2003, 199 consecutive patients were referred. Of that number, 196 agreed to participate, and were enrolled (Table <xref ref-type="table" rid="T1">1</xref>). The mean age was 52 years, 55% were male, 75% were white, 96% were taking opioids, 28% had a history of alcohol abuse, and 28% had a history of cocaine abuse. Eighty-five percent reported an income of less than $20,000 per year. Depression was common: the average CES-D score was 23.6, 74% of patients scored in the depressed range, and 54% scored in the severe depression range. The average literacy score using the REALM was 51.2, and 54% of patients scored below the 9th grade reading level (REALM < 60). The mean PDI score was 45.2, suggesting substantial functional impairment. Twelve percent of patients had North Carolina drug convictions; eleven percent had driving while impaired convictions (DUI); and 20% had either drug or alcohol convictions. Back pain was the most common cause of chronic pain, and the distribution of primary pain types was consistent with other reports of pain types reported in the general medicine literature, with the exception of the under-representation of headache (Data not shown) [<xref ref-type="bibr" rid="B28">28</xref>,<xref ref-type="bibr" rid="B43">43</xref>,<xref ref-type="bibr" rid="B44">44</xref>]. At 12 months, four patients were lost to follow-up, and three changed their venue of primary care.</p><sec><title>Incidence of opioid misuse</title><p>Over the one-year study period, opioid misuse occurred in sixty-two (32%) patients (Table <xref ref-type="table" rid="T2">2</xref>). Twenty-five patients were found to have positive and confirmed urine drug screens for stimulants; twenty-four were positive for cocaine metabolite, and one for amphetamines. Fifteen patients were found to have repeatedly negative urine drug screens for prescribed opioids despite being counseled on at least one occasion about the proper scheduling of their medication. The absence of the prescribed opiate was confirmed with GC/MS. Nine patients had repeatedly positive UTS for non-prescribed opioids, despite being counseled on at least one occasion that this was a violation of the medication agreement. Ten patients habitually obtained opioids from multiple providers (We did not consider the occasional use of the emergency department as a violation, but counseled patients against this practice unless clinically necessary). Two patients were found to have forged prescriptions, and one patient was diverting medications. All patients who violated the clinic opioid misuse policy were offered referral for counseling, but only two followed through, to our knowledge. Although not considered opioid misuse, eighteen percent had a UTS positive for cannabinoids at least once during the study period.</p></sec><sec><title>Predictors of opioid misuse</title><p>Predictors of opioid misuse were examined in bivariate and multivariate analyses. In bivariate analyses (Table <xref ref-type="table" rid="T3">3</xref>), misusers were more likely than non-misusers to have past cocaine abuse (68% vs 21%; p < 0.001), have a previous drug or DUI conviction (40% vs 11%; p < 0.001), be younger (48 years vs 54 years, p < 0.001), have past alcohol abuse (44% vs 23%; p = 0.004), or be male (59.7% vs. 38%; p = 0.005). Similar to cocaine abuse, the presence of cannabinoids on UTS obtained at any time during the 12 month follow-up period (33% vs 12%; p = 0.001) was a predictor of misuse. A previous drug or DUI conviction or multiple drug convictions were more strongly associated with misuse, with relative risks of 3.6, and 15.1, respectively. Race, income, education, depression score (CES-D), disability (PDI), and literacy score (REALM) were not associated with opioid misuse. There was no consistent correlation between pain scores and the risk of misuse, although misusers reported a higher intensity of <italic>current </italic>pain at baseline (Table <xref ref-type="table" rid="T4">4</xref>).</p><p>In multivariate analyses (Table <xref ref-type="table" rid="T5">5</xref>), age, self-reported histories of cocaine or alcohol abuse, drug or DUI convictions were shown to be the most powerful predictors of misuse (AUC, 0.827). The effect of a history of cocaine abuse was moderately strong (OR, 4.3; CI, 1.76 – 10.4). The odds ratios (OR) for drug or DUI convictions and a history of alcohol abuse were both 2.6. Age, though statistically significant in the model, did not clinically discriminate well between misusers and non-misusers. In the adjusted analyses, the average age was 53 years for misusers and 49 years for non-misusers. We performed analyses of the subset of opioid misusers who were not abusing stimulants (N = 37). The bivariate sub-analysis demonstrated general persistence of the statistical relationships seen in the entire sample (Table <xref ref-type="table" rid="T6">6</xref>).</p></sec></sec><sec><title>Discussion</title><p>We identified predictors of opioid misuse in a cohort of opioid-treated patients with chronic pain who were enrolled in a primary care-based disease management program. Our program and study was not designed to make systematic substance abuse, dependence, or addiction diagnoses but rather to apply a working diagnosis of misuse that defined conditions under which opioids would be prescribed. The strongest predictors of misuse in the study population were self-reported histories of previous alcohol or cocaine abuse, or previous criminal drug or alcohol-related convictions. Age was also predictive, but the effect was not large. Gender, race, literacy, disability, and measures of socioeconomic status were not associated with misuse. The most frequent type of misuse involved the concurrent use of stimulants, usually cocaine. In a separate bivariate sub-analysis of patients with opiate misuse other than cocaine or amphetamines on UTS, the relationships between predictors and outcomes were similar, as the magnitudes of the odds ratios shown in Table <xref ref-type="table" rid="T6">6</xref> suggest. Our findings stand in contradistinction to other research that has found no predictive relationship between past alcohol and substance abuse and future opioid abuse in patients with chronic pain [<xref ref-type="bibr" rid="B45">45</xref>]. The pattern of drug misuse in the study population suggested the potential for multiple co-morbid diagnoses of substance abuse or dependence, placing these individuals at especially high risk of morbidity and mortality [<xref ref-type="bibr" rid="B46">46</xref>].</p><p>The limited clinical trials in the literature examining the use of opioids in the treatment of chronic pain do not identify factors that put chronic pain patients at risk for opioid misuse. They do not provide concrete guidance about how to select appropriate candidates for opioid therapy in a primary care setting. Although the incidence of misuse that we report is higher than that reported in other studies, many studies have not clearly defined their monitoring procedures to detect opioid misuse, have excluded patients <italic>a priori </italic>with significant mental illness (even major depression) or history of drug misuse [<xref ref-type="bibr" rid="B16">16</xref>], and have been conducted in specialty settings [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B47">47</xref>-<xref ref-type="bibr" rid="B49">49</xref>].</p><p>Some authorities have asserted that substance abuse and dependence are uncommon or rare consequences of opioid use for pain; however, the heterogeneity of the available evidence does permit accurate estimates of the prevalence or incidence of abuse in opioid-treated patients. One widely cited reference estimates opioid <italic>addiction </italic>at approximately 4 in 10,000 treated patients [<xref ref-type="bibr" rid="B50">50</xref>]. Such a low prevalence of misuse in opioid-treated patients, moreover, is inconsistent with epidemiological data that conservatively estimate the 12-month prevalence of drug misuse at 80 in 10,000 [<xref ref-type="bibr" rid="B51">51</xref>]. Pain specialty clinics have reported prevalences of dependence ranging from 3% to 17% [<xref ref-type="bibr" rid="B52">52</xref>]. In primary care, a retrospective study of two clinics documented <italic>misuse </italic>of opioid medications at 24% and 31%, respectively [<xref ref-type="bibr" rid="B53">53</xref>]. A study from Sweden suggests that abuse is common is patients with chronic pain. In that study, 414 hospitalized patients with chronic pain were systematically evaluated for substance abuse using the Substance Use Disorder Diagnostic Schedule based on the <italic>Diagnostic and Statistical Manual of Mental Disorders, Third Edition</italic>. Twenty-three percent of patients were found to have active drug abuse disorders [<xref ref-type="bibr" rid="B54">54</xref>]. In general, it is difficult to apply DSM-IV criteria for substance abuse or dependence in the context of prescription opioid use.</p><p>We chose the term <italic>misuse </italic>in our study because misuse encompasses behaviors with both medical and non-medical dimensions, whereas <italic>abuse </italic>more properly denotes the medical substance abuse or dependence disorders. The standard psychiatric definitions of abuse and dependence focus on tolerance and withdrawal which cannot be used to identify aberrant behavior in patients who are prescribed and regularly taking the medication that they may or may not be abusing as well. We adhered to published guidelines and literature that discourage opioid prescribing to patients with a history of previous or ongoing substance abuse. Stimulant-positive urines were considered evidence of, or proxy for, ongoing substance abuse and hence a contraindication to prescribing opioids. Evidence of stimulant abuse thus constituted opioid misuse as defined by our medication agreement but not opioid abuse or dependence per se. Also, we suspected that another subset of patients was procuring and diverting opioids for monetary gain as evidenced by the frequent finding of negative UTS in patients who reported they were using their medication as directed. These misusers might not receive substance abuse diagnoses. Based on consistently negative UTS, diversion of O/CS medications appears to be a common form of misuse encountered in our study. While the reasons for different forms of misuse were not qualitatively examined, the high street value of prescription opioids may have led to a temptation to sell them [<xref ref-type="bibr" rid="B55">55</xref>]. Alternatively, patients with negative UTS may have "used up" their prescriptions by taking their medication at a greater than agreed upon rate, although all patients found to have negative UTS asserted that they were taking their medications correctly, and UTS confirmation should have revealed their presence. In addition, we did not often witness the physiologic opioid withdrawal one would have expected in these patients.</p><p>We chose not to define a single positive cannabinoid test on UTS while receiving O/CS pharmacotherapy as an act of misuse that would result in sanction. We did, however, advise patients against marijuana use. Research in twins suggests that marijuana use is a risk factor for developing more severe and pervasive drug misuse disorders [<xref ref-type="bibr" rid="B56">56</xref>]. Our data suggest that marijuana users may be at higher risk of misuse and might require more vigilant monitoring.</p><p>Currently, most primary care settings have not organized care in a way that allows systematic evaluation of patients with chronic pain for either response to pharmacotherapy or misuse [<xref ref-type="bibr" rid="B23">23</xref>,<xref ref-type="bibr" rid="B30">30</xref>,<xref ref-type="bibr" rid="B57">57</xref>]. We believe that our pragmatic approach to monitoring opioid misuse based on the specific elements of the medication agreement can be replicated in primary care settings that do not have the resources to systematically evaluate patients for substance abuse or dependence. It provides a rational template for treating pain effectively and compassionately with opioids [<xref ref-type="bibr" rid="B31">31</xref>], while also offering providers reassurance that their actions are not contributing to the growing public health problem of prescription drug diversion and misuse.</p><p>Striking a balance between appropriate use of opioids and prevention of misuse is important for successful management of chronic pain. This study and others have found that the multidisciplinary disease management for chronic pain, can produce significant reductions in pain, improvements in depression and health-related quality of life through the establishment of a pain diagnosis and management plan [<xref ref-type="bibr" rid="B58">58</xref>,<xref ref-type="bibr" rid="B59">59</xref>]. Recent restrictions in the Drug Enforcement Administration regulations with regard to the provision of Schedule II controlled substances [<xref ref-type="bibr" rid="B60">60</xref>,<xref ref-type="bibr" rid="B61">61</xref>], along with rare but high-profile prosecutions of pain-treating physicians [<xref ref-type="bibr" rid="B62">62</xref>,<xref ref-type="bibr" rid="B63">63</xref>], have highlighted the need for continued care in prescribing these agents. Systematic approaches to pain management and detecting opioid misuse can reassure physicians that they can alleviate suffering with opioids without inviting criminal sanction or negatively impacting public health.</p><p>This study has several limitations. As noted above, the study population was drawn from referrals within a single, academic general internal medicine practice. As such, the sample may not be representative of all opioid-treated patients in primary care settings. Because we sought referrals of patients that were difficult to manage, the incidence of opioid or other drug misuse in this investigation may be higher than in other primary care or community-based populations of opioid-treated patients. Public information on drug offenses and DUI, while easily obtained online in North Carolina, is less accessible in other states. The initial assessment of prior or current drug misuse was based on self-report and clinical assessment rather than a structured diagnostic interview; better measurement may have allowed more accurate classification and assessment of risk. In addition, we did not inquire about histories of substance abuse other than alcohol and cocaine. Finally, we have limited data about the patients' outcomes after completing the study. Patients who were identified as committing misuse usually dropped out of the program, and we were unable to assess outcomes of pain, functional status, and mental health status once contact was lost.</p></sec><sec><title>Conclusion</title><p>Identifying chronic pain patients at risk for opioid misuse remains a challenge. This study and other studies of chronic pain patients [<xref ref-type="bibr" rid="B52">52</xref>-<xref ref-type="bibr" rid="B54">54</xref>], suggest that the prevalence of any substance misuse may approach one-quarter of chronic pain patients receiving opioids. Opioid misuse was more common in patients with a self-reported history of alcohol or cocaine abuse. Previous criminal convictions for DUI or drug offenses predicted opioid misuse. Based upon these data, patients with a history of alcohol or cocaine abuse and alcohol or drug related convictions should be carefully evaluated and followed for signs of misuse if opioids are to be prescribed.</p><p>Additional prospective studies in primary care settings are needed to confirm these findings and to examine other potential predictors of opioid misuse. Also, better studies of interventions to reduce misuse of opioids and programs designed to effectively treat pain in patients with active substance abuse disorders are needed [<xref ref-type="bibr" rid="B64">64</xref>]. At the public health level, several states are considering legislation to allow better monitoring of prescriptions of controlled substances, such as state-wide registries, that may reduce some types of misuse, particularly the procurement of medication from multiple sources.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no financial or non-financial competing interests.</p></sec><sec><title>Authors' contributions</title><p>TJI developed the study design and intervention, administered surveys, directed pain management and drug misuse monitoring, assisted in the drafting, editing, and revision of the manuscript. PRC developed the study design and intervention, performed statistical analyses and drafted, edited, and revised the manuscript. CAH-S developed drug misuse monitoring protocols, provided expert consultation on clinical protocols using urine toxicological testing and edited the manuscript. RMM developed the study design, oversaw data management, and edited and revised the manuscript. JSP developed opioid misuse monitoring protocols, administered surveys, performed data management, edited the manuscript. NMP developed opioid misuse monitoring protocols, administered surveys, performed data management, edited the manuscript. BBS developed the study design, participated in data management, edited the manuscript. DAD provided statistical analytical support, and assisted in the drafting, editing, and revising the manuscript. MPP developed the study design, supervised overall conduction of the study, participated in data analysis, and assisted in the drafting, editing, and revision of the manuscript.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1472-6963/6/46/prepub"/></p></sec>
Protocol for stage 1 of the GaP study (Genetic testing acceptability for Paget's disease of bone): an interview study about genetic testing and preventive treatment: would relatives of people with Paget's disease want testing and treatment if they were available?	<sec><title>Background</title><p>Paget's disease of bone (PDB) is characterised by focal increases in bone turnover, affecting one or more bones throughout the skeleton. This disrupts normal bone architecture and causes pain, deformity, deafness, osteoarthritis, and fractures.</p><p>Genetic factors are recognised to play a role in PDB and it is now possible to carry out genetic tests for research. In view of this, it is timely to investigate the clinical potential for a programme of genetic testing and preventative treatment for people who have a family history of PDB, to prevent or delay the development of PDB.</p><p>Evidence from non-genetic conditions, that have effective treatments, demonstrates that patients' beliefs may affect the acceptability and uptake of treatment. Two groups of beliefs (illness and treatment representations) are likely to be influential.</p><p>Illness representations describe how people see their illness, as outlined in Leventhal's Self-Regulation Model. Treatment representations describe how people perceive potential treatment for their disease. People offered a programme of genetic testing and treatment will develop their own treatment representations based on what is offered, but the beliefs rather than the objective programme of treatment are likely to determine their willingness to participate. The Theory of Planned Behaviour is a theoretical model that predicts behaviours from people's beliefs about the consequences, social pressures and perceived control over the behaviour, including uptake of treatment.</p></sec><sec><title>Methods/design</title><p>This study aims to examine the acceptability of genetic testing, followed by preventative treatment, to relatives of people with PDB. We aim to interview people with Paget's disease, and their families, from the UK. Our research questions are:</p><p>1. What do <bold><italic>individuals with Paget's disease </italic></bold>think would influence the involvement of their relatives in a programme of genetic testing and preventative treatment?</p><p>2. What do <bold><italic>relatives of Paget's disease sufferers </italic></bold>think would influence them in accepting an offer of a programme of genetic testing and preventative treatment?</p></sec><sec><title>Discussion</title><p>Our research will be informed by relevant psychological theory: primarily the Self-Regulation Model and the Theory of Planned Behaviour. The results of these interviews will inform the development of a separate questionnaire-based study to explore these research questions in greater detail.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Langston</surname><given-names>Anne L</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Johnston</surname><given-names>Marie</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Robertson</surname><given-names>Clare</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Campbell</surname><given-names>Marion K</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Entwistle</surname><given-names>Vikki A</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Marteau</surname><given-names>Theresa M</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A7" contrib-type="author"><name><surname>McCallum</surname><given-names>Marilyn</given-names></name><xref ref-type="aff" rid="I5">5</xref><email>[email protected]</email></contrib><contrib id="A8" contrib-type="author"><name><surname>Ralston</surname><given-names>Stuart H</given-names></name><xref ref-type="aff" rid="I6">6</xref><email>[email protected]</email></contrib>	BMC Health Services Research	<sec><title>Background</title><sec><title>The need for this research</title><p>Over recent years, advances in human molecular genetics have resulted in the identification of polymorphisms and mutations in several genes that cause or predispose to diseases such as cancer, neurodegenerative conditions and inborn errors of metabolism. For some of these conditions, such as Huntington's disease and muscular dystrophy, it is possible to offer a genetic test that will give information on the probability of the disease occurring but this is somewhat unattractive to patients without the prospect of an effective treatment. For other diseases, such as hypercholesterolaemia and haemochromatosis, identification of genetic susceptibility can be used as a risk factor to inform management strategies and treatment decisions in a similar way to other clinical risk factors.</p><p>This application aims to test the feasibility of translating genetic knowledge into practice by examining the acceptability of offering genetic testing, followed by therapeutic intervention, to people with Paget's disease of bone.</p></sec><sec><title>Paget's disease of bone</title><p>Paget's disease of bone (PDB) affects about 3% of individuals over the age of 55 years in the UK [<xref ref-type="bibr" rid="B1">1</xref>]. It is characterised by focal increases in bone turnover, affecting one or more bones throughout the skeleton. The abnormal bone turnover disrupts normal bone architecture and causes bone pain, bone deformity, deafness, osteoarthritis, and pathological fractures [<xref ref-type="bibr" rid="B2">2</xref>]. Genetic factors have long been recognised to play an important role in Paget's disease [<xref ref-type="bibr" rid="B3">3</xref>-<xref ref-type="bibr" rid="B6">6</xref>] and recent studies by our own group and others indicate that between 40–50% of instances of familial Paget's disease are caused by mutations affecting the ubiquitin-associated (UBA) domain of the <italic>SQSTM1 </italic>gene [<xref ref-type="bibr" rid="B7">7</xref>-<xref ref-type="bibr" rid="B11">11</xref>]. Genotype-phenotype analysis has shown that <italic>SQSTM1 </italic>mutations are highly penetrant, such that between 90–100% of individuals within families who carry mutations will have developed the disease by the age of 65 years [<xref ref-type="bibr" rid="B7">7</xref>-<xref ref-type="bibr" rid="B11">11</xref>].</p><p>In view of this, it is timely to investigate the clinical potential for a programme of genetic testing and preventative treatment for patients who carry <italic>SQSTM1 </italic>mutations, in an effort to prevent or delay the development of Paget's disease or its complications.</p></sec><sec><title>Prospects for preventing Paget's disease</title><p>Bisphosphonates such as Risedronate, Pamidronate and Zoledronate have emerged as highly effective agents for reducing bone turnover and treating bone pain in patients with Paget's disease [<xref ref-type="bibr" rid="B12">12</xref>]. Clinical studies have shown that these drugs can restore elevated levels of bone turnover to normal in a high proportion of cases, can improve the appearance of Pagetic lesions on isotope bone scans [<xref ref-type="bibr" rid="B13">13</xref>], and can restore bone architecture to normal, as assessed histologically [<xref ref-type="bibr" rid="B14">14</xref>].</p><p>This raises the possibility that patients who carry <italic>SQSTM1 </italic>mutations and who are at high risk of developing Paget's disease could be offered prophylactic therapy, in an attempt to prevent the disease occurring, or to prevent complications developing. A particularly promising mode of treatment in this respect is Zoledronate, which is a highly effective treatment for Paget's disease even in microgram doses [<xref ref-type="bibr" rid="B15">15</xref>], with inhibitory effects on bone turnover that can extend for at least 12 months after a single injection [<xref ref-type="bibr" rid="B16">16</xref>]. There is also evidence to suggest that treatment with intravenous Pamidronate and oral Risedronate can suppress bone resorption in active Paget's disease for up to two years [<xref ref-type="bibr" rid="B17">17</xref>,<xref ref-type="bibr" rid="B18">18</xref>].</p><p>With the existence of effective treatment for Paget's disease that could prevent or delay the onset of the disease, we need to explore the acceptability of genetic testing followed by preventative treatment in patients with Paget's disease of bone.</p></sec><sec><title>Possible factors that could affect acceptability of genetic testing and preventative treatment</title><p>Evidence from other areas of genetic testing suggests that patients may not wish to take genetic tests. For example, in Huntington's disease, prior to the development of a genetic test, individuals at risk declared a much higher intention of taking a test than the actual numbers taking the test when it became available [<xref ref-type="bibr" rid="B19">19</xref>]. Studies have shown that there is a widespread belief that genetic conditions are not treatable. Such beliefs might affect the acceptability of diagnosis and treatment of genetic conditions. However, the additional offer of effective preventative treatment can make genetic testing more acceptable [<xref ref-type="bibr" rid="B20">20</xref>].</p></sec><sec><title>Beliefs</title><p>Evidence from other, non-genetic, conditions, for which there are effective treatments, demonstrates that patients' beliefs may affect the acceptability and uptake of treatment. Two groups of beliefs (illness representations and treatment representations) are most likely to be influential.</p><sec><title>Illness representations</title><p>Illness representations describe how people see their illness. This is most clearly outlined in the Self-regulation Model (SRM) [<xref ref-type="bibr" rid="B21">21</xref>]. The five key representations are:</p><p>• <italic>identity </italic>(e.g. the 'label', identifying symptoms);</p><p>• <italic>cause </italic>(e.g. stress, genetics);</p><p>• <italic>consequences </italic>(e.g. activity limitations, loss of wages);</p><p>• <italic>timeline </italic>(e.g. acute, fluctuating, progressing);</p><p>• <italic>cure/control </italic>(e.g. medication, diet, exercises).</p><p>In addition, the SRM identifies the <italic>emotional representation </italic>(e.g. frustrating, frightening) as a separate dimension of representation. Based on this model, Weinman and colleagues [<xref ref-type="bibr" rid="B22">22</xref>] have developed a measure, the Illness Perceptions Questionnaire (IPQ), to assess these representations and have found that they are predictive of uptake and response to a wide variety of treatments, including the bone disease Rheumatoid Arthritis [<xref ref-type="bibr" rid="B23">23</xref>]. Clearly, such beliefs may influence uptake of a new treatment offered to families with Paget's disease, and Marteau has shown that beliefs affect the decision to accept a genetic test [<xref ref-type="bibr" rid="B24">24</xref>].</p><p>Paget's disease has a highly variable presentation: age at onset, severity of symptoms and impact on sufferer's lives; and individuals might vary in their response to an offer of testing and treatment depending on the disease history within their family. Using the World Health Organisation International Classification of Functioning, Disability and Health model of health components [<xref ref-type="bibr" rid="B25">25</xref>], one might expect that all three components (impairment, activity limitations, and participation restrictions) might affect beliefs about the identity, consequences and timeline of the condition.</p></sec><sec><title>Treatment representations</title><p>Treatment representations describe how people perceive potential treatments for their disease. There is considerable evidence that patients' beliefs about treatment (e.g. accessibility, burden, perceived likelihood of success) influence uptake and adherence [<xref ref-type="bibr" rid="B26">26</xref>]. People offered a programme of genetic testing and treatment will develop their own treatment representations based on what is offered, but the beliefs rather than the objective programme of treatment, are likely to determine their willingness to participate. The Theory Of Planned Behaviour (TPB) [<xref ref-type="bibr" rid="B26">26</xref>] (is a theoretical model, which predicts behaviours from people's beliefs about the consequences, social pressures and perceived control over the behaviour, including uptake of treatment (Figure <xref ref-type="fig" rid="F1">1</xref>).</p></sec></sec><sec><title>The nature of the treatment and treatment offer</title><p>The nature of the treatment, and the way in which the treatment is offered, may affect acceptability. One needs to consider not only how to invite the affected individual to 'involve' their family, but also how to introduce the programme to invited family members. Hardeman and colleagues [<xref ref-type="bibr" rid="B27">27</xref>] in the <italic>Pro-Active </italic>Programme, have developed an interview for families of people with Type 2 diabetes that incorporates the offer of the intervention for offspring. This interview was based on the interviews used to develop questionnaires for the Theory of Planned Behaviour (TPB) addressing beliefs about involving the offspring in the programme. It has proved very effective in rendering the programme acceptable [<xref ref-type="bibr" rid="B27">27</xref>].</p><p>In addition, the treatment offered for Paget's disease could be in tablet or infusion form, and could be administered in a range of environments. All of these factors may influence the acceptability of a programme of genetic testing and preventative treatment.</p></sec><sec><title>Demographic and health characteristics</title><p>Demographic and health characteristics may also influence the acceptability of treatment. For example, someone who is older may think they are less at risk if they have passed the age of onset of Paget's disease in the affected person; or they may think they are nearing the time when the disease might affect them. Closeness to the affected person may make them view the condition as more or less serious, and they may be influenced by how the affected person copes with the illness. Furthermore, one might expect there to be more agreement within members of one family than between families due to their shared exposure to the condition as well as their opportunity to discuss and influence each other.</p></sec><sec><title>Scientific value of this study</title><p>This project takes advantage of recent advances in knowledge about the molecular-genetic basis of PDB, with advances in therapeutics to address an important clinical question that is of relevance to patients with Paget's disease and their families. The project provides added value in that modern techniques in behavioural science will be applied to a specific issue in the treatment of Paget's disease, while at the same time investigating more general theories of human behaviour. Using the Self-Regulation Model, we will investigate how genetic information and preventive opportunities affect self-regulation via illness representations and, using the Theory of Planned Behaviour, how uptake of the offer of genetic testing and treatment is predicted by the constructs found to predict behaviour in other settings. Thus the findings will be relevant to these theories. They will also contribute to ongoing work to integrate these two models in understanding behaviour related to illness.</p><p>In addition, the knowledge gained from this investigation would be relevant to other musculoskeletal diseases such as osteoporosis, which also have a strong genetic component, and which can be prevented by bisphosphonates and hormone replacement therapy. With completion of the human genome project, and advances in human molecular genetics, it is probable that DNA testing with single genetic markers or combinations of markers will be offered for a wide range of other chronic diseases for which effective treatments are available. The results of the present research project will also be of direct relevance to the development of programmes of genetic testing and preventative intervention for these conditions.</p></sec><sec><title>Study aim</title><p>The aims of the GaP study are to understand factors that might influence the acceptability of an offer of a programme of genetic screening and preventative treatment to families of Paget's disease sufferers.</p><p>The specific research questions are:</p><p><bold>1. </bold>a) What do individuals with Paget's disease think would influence the involvement of their relatives in a programme of genetic testing and preventative treatment? and;</p><p>b) What do relatives of Paget's disease sufferers think would influence them in accepting an offer of a programme of genetic testing and preventative treatment?</p><p><bold>2. </bold>Do the following factors affect acceptability of a programme of genetic testing and preventative treatment in relatives of Paget's disease sufferers?</p><p>• Illness and emotional representations of Paget's disease;</p><p>• Treatment representations;</p><p>• Presentation of the disease in affected relative(s) (age at onset, impairment, activity limitations, participation restrictions; family history as a function of the number of affected relatives and the relationship to the Subject);</p><p>• Respondent characteristics (age, gender and health);</p><p>• Beliefs of other family members; and</p><p>• The nature of the treatment offered.</p><p>By answering these research questions, this study will cast light on the feasibility of developing a programme of genetic testing and preventative treatment for individuals who carry <italic>SQSTM1 </italic>mutations that are at high risk of developing Paget's disease. Since we do not yet know whether prophylactic bisphosphonate therapy would be effective in preventing Paget's disease or its complications, we envisage that the results of the present study would be used to inform the design of a randomised controlled trial to investigate the efficacy of prophylactic bisphosphonate therapy in people with <italic>SQSTM1 </italic>mutations. Such a study would form the topic of a future research project.</p></sec><sec><title>An overview of the GaP study</title><p>The GaP study has two stages:</p><p>Stage 1: This part of the study will address the first research question and aims to identify factors that would influence involvement of relatives not known to have Paget's disease in a programme of genetic screening and preventative treatment.</p><p>Stage 2: This part of the study is <underline>not</underline> part of the current protocol, and will comprise a postal questionnaire study of people without Paget's disease but who are relatives of people with Paget's disease.</p></sec></sec><sec><title>Methods/design for stage 1</title><p>Stage 1 will employ semi-structured interviews with individuals suffering from Paget's disease, and their relatives who have not been diagnosed as suffering from Paget's disease. The results of Stage 1 will inform the development of a TPB-based questionnaire for Stage 2.</p><p>The interviews with individuals suffering from Paget's disease will use the Theory of Planned Behaviour (TPB)-based [<xref ref-type="bibr" rid="B26">26</xref>].</p><sec><title>Subjects and recruitment</title><p>Individuals with Paget's disease will be identified from an established cohort: The PRISM trial.</p><p>The PRISM cohort involves 1331 patients with Paget's disease. The PRISM trial involves 39 collaborating centres ranging in size from 3 trial participants to approx 250 trial participants.</p><p>For this part of the study we will identify potential participants from 7 centres collaborating in the PRISM trial.</p></sec><sec><title>Identifying subjects</title><p>A purposive sample of individuals with Paget's disease identified from the PRISM study will be invited to participate. Participating individuals will be asked to identify first and second-degree relatives without Paget's disease who they think might also be interested in taking part in the study. These individuals will be invited to participate in separate interviews. Relatives identified by this procedure will be selected for the study to represent a range of family relationships (e.g. siblings, children, grandchildren) and living conditions (e.g. living in same house, nearby, or far away). A sequential approach to sampling will therefore be used.</p><p>For this stage of the study, individuals will be approached from a restricted selection of PRISM centres (Table <xref ref-type="table" rid="T1">1</xref>.).</p><p>These 3 groups of centres represent regions of the UK with varying incidences of Paget's disease. A low incidence of Paget's disease occurs in Scotland, whereas Manchester and Liverpool are well recognised as areas of high incidence of Paget's disease. Lancashire is considered to be the global 'hotspot' for Paget's. The southern England region represents an area with an incidence rate between those of the northern England and Scottish regions.</p></sec><sec><title>Contacting subjects</title><p>A letter will be sent to probands inviting them to participate in an interview together with study information leaflets. The letter will include a 'tear off' section to return (in a reply paid envelope) if they are interested in discussing the study further. Probands will also be asked to provide contact details of one or two relatives who may also like to participate in an interview.</p><p>A letter will be sent to relatives identified by probands inviting them to participate in an interview together with study information leaflets. The letter will include a 'tear off' section to return (in a reply paid envelope) if they are interested in discussing the study further.</p><p>A researcher will telephone those who return the slip, and who are interested in taking part, to discuss the study further. If the person is still interested, an appropriate time and place for interview, and arrangements for formal consent will be discussed. There will be no follow-up of those probands or relatives that decline to participate or who do not reply.</p><p>A Thank You letter will be sent to probands and relatives taking part in Stage 1, once each individual interview is completed.</p></sec><sec><title>Sample size</title><p>Sampling will continue until saturation, i.e. no 'new' ideas are being introduced within the three TPB beliefs (behavioural, normative and control) and 5 SRM domains (identity, cause, consequences, cure/control, timeline) plus emotional representations). It is expected that this will involve 10–20 individuals from the PRISM cohort together with 10–20 relatives of these affected individuals.</p><p>Purposive sampling will be used to identify probands and relatives to be approached for participation in this stage. Subjects are not chosen at random but instead are chosen because they are expected to facilitate investigation of the range of views and opinions relevant to the research. The aim of sampling is therefore to maximise variety for analysis. Subjects will be purposively selected to include a good spread across the following factors:</p></sec><sec><title>Probands</title><p>• Gender (male, female);</p><p>• Age (<40 years, 41–60 years, >61 years);</p><p>• Severity of disease (monostotic, polyostotic);</p><p>• Time elapsed since diagnosis (<5 years, > 5 years); and</p><p>• Family history (Yes, No).</p></sec><sec><title>Relatives</title><p>• Gender (male, female);</p><p>• Age (<40 years, 41–60 years, >61 years);</p><p>• Severity of disease in proband (monostotic, polyostotic);</p><p>• Relationship to proband (parent, sibling, grandparent, aunt/uncle); and</p><p>• Living proximity to proband (within 50 miles, further than 50 miles).</p><p>Anne Langston will undertake selection of interviewees, and the interviewer (Clare Robertson) will be blinded to these details.</p></sec><sec><title>Exclusion criteria</title><p>Exclusion criteria include:</p><p>• Known* limited life expectancy (under one year); or</p><p>• Aged less than 18 years.</p><p>* <italic>The caveat of 'known' is used, as it may not be possible to determine limited life expectancy of relatives</italic>.</p><p>A lower age limit has been imposed (minimum age of 18 years). Clinical studies suggest that people with <italic>SQSTM1 </italic>mutations develop Paget's disease from approximately 45 years of age onwards [<xref ref-type="bibr" rid="B9">9</xref>]. Clinical intuition would suggest that starting a programme of preventative treatment at age 40 years would be therefore be appropriate. However, the clinical evidence to support this is still lacking. In addition, it is likely that a psychologically 'acceptable' age for beginning preventative treatment will differ from that indicated by existing evidence. As such, we propose to include all adult relatives of probands to determine the effect of age of the acceptability of genetic testing, and preventative treatment.</p></sec><sec><title>Special arrangements for including people with specific communication needs</title><p>We have not made special arrangements for non-English speaking subjects for two reasons:</p><p>1. Providing multi-lingual facilities for this small number of interviews will be logistically and financially prohibitive;</p><p>2. Paget's disease is rare in non-Caucasians and therefore we anticipate that English will be the first language of the majority of eligible individuals within the UK.</p><p>3. It has not proved necessary to provide special arrangements for participants of the PRISM trial, and therefore it is not anticipated that the requirement will exist for this study.</p><p>Participants whose first language is not English will not automatically be excluded from the study if they are either able to communicate in English, or an interpreter is available to assist with communication. Such arrangements will be accommodated, if it is practicable to do so, on an individual basis.</p><p>Individuals with limited life expectancy are excluded from Stage 1 of the study as the interview process may be cognitively and emotionally taxing for them.</p><p>Individuals with known limited hearing ability may be excluded from Stage 1 due to practical difficulties of carrying out an interview and the distress and frustration this may cause the participant. If a participant is deaf but can lip-read, or has a friend or family member who can use Sign Language for them, the individual will not be excluded. Since deafness caused by Paget's disease of the skull is an important complication of Paget's disease, every attempt will be made to include these individuals in Stage 1 of the study, where practicable and where it will not cause distress to the participant.</p><p>Information leaflets and consent forms will be made available in large print upon request.</p></sec><sec><title>Setting</title><p>For Stage 1, identified PRISM probands and relatives will be invited to interview at home or in a convenient clinical or research environment. The place of the interview will be according to the subject's preference. Travel expenses incurred by the subject will be reimbursed. Interviews will be tape-recorded and will be conducted face-to-face to allow the interviewer to counsel for consent.</p></sec><sec><title>Consent process</title><p>Consent will be sought at interview and participating probands and relatives will be asked to sign a consent form. One copy will be kept by the participants and one returned to the researcher.</p><p>The consent form has been designed in accordance with current ethical good practice [<xref ref-type="bibr" rid="B28">28</xref>]. The consent form explicitly states that genetic testing is not available as part of this study.</p></sec><sec><title>Withdrawal of subjects</title><p>Individuals will be able to withdraw from the process at any time. If individuals wish to withdraw their consent after the interview has taken place, the audio recording and any associated transcript will be destroyed if specifically requested. They will not be contacted for participation in Stage 2 or future research.</p></sec><sec><title>Measures</title><p>Semi-structured interview schedules will be developed to address Behavioural, Normative and Control beliefs using the TPB preliminary interview format [<xref ref-type="bibr" rid="B26">26</xref>,<xref ref-type="bibr" rid="B29">29</xref>], and for probands will also address illness perceptions using an interview schedule based on the IPQ format [<xref ref-type="bibr" rid="B22">22</xref>].</p><p>For probands, the interview will also address barriers and facilitators to inviting their non-Pagetic relatives to participate in interviews about genetic testing and preventative treatment (a topic guide will be used). For non-Pagetic relatives, the interview will address barriers and facilitators to participating in a programme of genetic testing and preventative treatment (a topic guide will be used).</p></sec><sec><title>Data recording</title><p>With the permission of the subject, the entire interview will be digitally audio recorded. The subject will be assigned an anonymous code (study number). Audio recordings will be labelled with the date of interview and study number.</p><p>Not all of the interviews will be transcribed. Transcription will occur on a limited selection of interviews (maximum of 5) to allow more detailed analysis. When transcription does take place it will be carried out by a trained data transcriber (not the interviewer). Pauses, including 'hmmms' and 'ahs', and interruptions of one speaker by another will also be indicated in the transcription. Transcribed interviews will be reviewed by the interviewer responsible for data collection (with reference to the original recording) and transcription errors will be rectified. Transcriptions of interviews will follow recognised guidelines and will not include any identifying information.</p></sec><sec><title>Analysis</title><p>Data will be content-analysed using simple TPB & IPQ coding frameworks. A sequential approach to sampling will be used whereby data will be analysed throughout recruitment on an ongoing basis until either the point of data saturation is reached (where no new responses emerge) or a maximum of 20 probands and 20 relatives have been interviewed.</p></sec><sec><title>End of study procedures</title><p>During the consent process for the study we will ascertain whether participants would like to be sent a report of the study results. For those participants indicating that they would like study results, a short report will be prepared and circulated at the end of the study.</p><p>Once the interview is completed the participant will be sent a Thank You letter. This will include a tear-off reply slip that participants will be asked to return indicating whether or not they would like to participate in Stage 2 of the GaP study.</p></sec><sec><title>Confidentiality</title><p>Transcripts and audio recordings will be kept in separate locked filing cabinets, and destroyed after 20 years in accordance with current MRC guidelines.</p><p>The Study Office, based in the Health Services Research Unit (HSRU), is responsible for the confidentiality of all study records. In accordance with the HSRU code of conduct, all data will be password protected against unauthorised access and stored in accordance with the Data Protection Act 1998. All stored data will be anonymised.</p></sec><sec><title>Finance and indemnity</title><p>The study is supported by a grant from the Medical Research Council (MRC). The University of Aberdeen holds the grant, and has accepted Sponsorship responsibilities for the study, including indemnity for negligent and non-negligent harm.</p></sec><sec><title>Reporting and dissemination</title><p>A summary of the results of the study will be prepared and distributed, not only to the appropriate funding body, but also to all subjects who take part (if they so wish). Results will be published in peer-reviewed academic journals.</p></sec></sec><sec><title>Abbreviations</title><p><bold>DNA </bold>Deoxyribonucleic acid. The substance that makes up genes.</p><p><bold>Familial Paget's disease </bold>Paget's disease that affects more than one member of a family</p><p><bold>IPQ-R </bold>The Illness Perceptions Questionnaire (R stands for revised version). This measures what people think about an illness in themselves or others.</p><p><bold>MRC </bold>Medical Research Council. A UK government funded organisation that supports medical research [<xref ref-type="bibr" rid="B30">30</xref>].</p><p><bold>NARPD </bold>National Association for the Relief of Paget's Disease. A support group for people with Paget's disease, their family and carers [<xref ref-type="bibr" rid="B31">31</xref>].</p><p><bold>Non-familial Paget's disease </bold>Paget's disease that only affects one person in a family.</p><p><bold>PDFR </bold>Paget's Disease Family Register. A study led by Prof Ralston, investigating the genetics of Paget's disease.</p><p><bold>PRISM </bold>A trial led by Prof Ralston and co-ordinated by Dr Langston, investigating the treatment of Paget's disease.</p><p><bold>Proband </bold>Initial person of contact within a family who has Paget's disease</p><p><bold>Relative </bold>A relative of a proband (i.e. brother, sister, mother, father, grandmother, grandfather, son, daughter, grandson, grand daughter, nephew, niece)</p><p><bold>SF-36 </bold>A questionnaire that measures general health, and quality of life.</p><p><bold>SQSTM1 </bold>Sequestosome 1 [RNA NM_003900]. One of the genes identified that, when 'faulty', causes Paget's disease.</p><p><bold>SRM </bold>The Self Regulation Model. A psychological theory that aims to identify how people perceive their illness.</p><p><bold>TPB </bold>The Theory of Planned Behaviour. A psychological theory that aims to identify predictors of behaviour using a defined framework.</p></sec><sec><title>Competing interests</title><p>Anne Langston received a small travel bursary in 2003 from The Alliance for Better Bone Health, an alliance between Proctor & Gamble, and Sanofi Aventis, which are pharmaceutical companies who manufacture drugs used in the treatment of Paget's disease. She is also a Board Member and Trustee for the National Association for the Relief of Paget's Disease. Stuart Ralston acts as a consultant for Proctor & Gamble, Sanofi Aventis and Novartis, which are pharmaceutical companies who manufacture drugs used in the treatment of Paget's disease. Prof Ralston is also a Board Member and Trustee for the National Association for the Relief of Paget's Disease. Marilyn McCallum is the Chief Executive for the National Association for the Relief of Paget's Disease. There are no competing interests for the other authors.</p></sec><sec><title>Authors' contributions</title><p>Anne Langston prepared and wrote the first draft of this protocol. Clare Robertson contributed to the amendments and re-drafting of the protocol to achieve ethical approval. Anne Langston, Marie Johnston, Marion Campbell, Vikki Entwistle, Theresa Marteau, Marilyn McCallum, and Stuart Ralston designed the study and are grant-holders. All authors have read and approved the final manuscript.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1472-6963/6/71/prepub"/></p></sec>
Racial and ethnic disparities in the control of cardiovascular disease risk factors in Southwest American veterans with type 2 diabetes: the Diabetes Outcomes in Veterans Study	<sec><title>Background</title><p>Racial/ethnic disparities in cardiovascular disease complications have been observed in diabetic patients. We examined the association between race/ethnicity and cardiovascular disease risk factor control in a large cohort of insulin-treated veterans with type 2 diabetes.</p></sec><sec sec-type="methods"><title>Methods</title><p>We conducted a cross-sectional observational study at 3 Veterans Affairs Medical Centers in the American Southwest. Using electronic pharmacy databases, we randomly selected 338 veterans with insulin-treated type 2 diabetes. We collected medical record and patient survey data on diabetes control and management, cardiovascular disease risk factors, comorbidity, demographics, socioeconomic factors, psychological status, and health behaviors. We used analysis of variance and multivariate linear regression to determine the effect of race/ethnicity on glycemic control, insulin treatment intensity, lipid levels, and blood pressure control.</p></sec><sec><title>Results</title><p>The study cohort was comprised of 72 (21.3%) Hispanic subjects (H), 35 (10.4%) African Americans (AA), and 226 (67%) non-Hispanic whites (NHW). The mean (SD) hemoglobin A1c differed significantly by race/ethnicity: NHW 7.86 (1.4)%, H 8.16 (1.6)%, AA 8.84 (2.9)%, p = 0.05. The multivariate-adjusted A1c was significantly higher for AA (+0.93%, p = 0.002) compared to NHW. Insulin doses (unit/day) also differed significantly: NHW 70.6 (48.8), H 58.4 (32.6), and AA 53.1 (36.2), p < 0.01. Multivariate-adjusted insulin doses were significantly lower for AA (-17.8 units/day, p = 0.01) and H (-10.5 units/day, p = 0.04) compared to NHW. Decrements in insulin doses were even greater among minority patients with poorly controlled diabetes (A1c ≥ 8%). The disparities in glycemic control and insulin treatment intensity could not be explained by differences in age, body mass index, oral hypoglycemic medications, socioeconomic barriers, attitudes about diabetes care, diabetes knowledge, depression, cognitive dysfunction, or social support. We found no significant racial/ethnic differences in lipid or blood pressure control.</p></sec><sec><title>Conclusion</title><p>In our cohort, insulin-treated minority veterans, particularly AA, had poorer glycemic control and received lower doses of insulin than NHW. However, we found no differences for control of other cardiovascular disease risk factors. The diabetes treatment disparity could be due to provider behaviors and/or patient behaviors or preferences. Further research with larger sample sizes and more geographically diverse populations are needed to confirm our findings.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Wendel</surname><given-names>Christopher S</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Shah</surname><given-names>Jayendra H</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Duckworth</surname><given-names>William C</given-names></name><xref ref-type="aff" rid="I3">3</xref><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A4" corresp="yes" contrib-type="author"><name><surname>Hoffman</surname><given-names>Richard M</given-names></name><xref ref-type="aff" rid="I4">4</xref><xref ref-type="aff" rid="I5">5</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Mohler</surname><given-names>M Jane</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Murata</surname><given-names>Glen H</given-names></name><xref ref-type="aff" rid="I4">4</xref><xref ref-type="aff" rid="I5">5</xref><email>[email protected]</email></contrib>	BMC Health Services Research	<sec><title>Background</title><p>Type 2 diabetes causes a substantial burden of suffering for minorities. Compared to non-Hispanic whites, all minorities except Alaskan natives have a 2- to 6-fold increased risk of acquiring the disease [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B3">3</xref>], and the prevalence is rising in some groups, including African Americans, Hispanic Americans, and American Indians [<xref ref-type="bibr" rid="B4">4</xref>]. Many population studies have shown that minorities do not achieve the same level of glycemic control as non-Hispanic whites [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B5">5</xref>-<xref ref-type="bibr" rid="B8">8</xref>]. Cultural and socioeconomic differences may create barriers to health by making it difficult to adhere to customary self-care recommendations for diet, weight loss, exercise, or blood glucose monitoring [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B9">9</xref>-<xref ref-type="bibr" rid="B11">11</xref>]. Finally, minorities have been shown to have an increased risk of developing micro- or macrovascular disease complications [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B12">12</xref>,<xref ref-type="bibr" rid="B13">13</xref>]. An excess risk has been described for nephropathy [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B14">14</xref>-<xref ref-type="bibr" rid="B16">16</xref>], retinopathy [<xref ref-type="bibr" rid="B17">17</xref>-<xref ref-type="bibr" rid="B19">19</xref>], amputations and foot problems [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B20">20</xref>], coronary artery disease [<xref ref-type="bibr" rid="B3">3</xref>], and stroke [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B21">21</xref>] compared to non-Hispanic whites. This health care burden makes it imperative to develop appropriate interventions for minority patients.</p><p>Evaluating racial and ethnic variations in the intermediate outcomes of diabetes care may be the best way to assess diabetes management. A comprehensive evaluation is essential because socio-cultural barriers may differentially influence various clinical outcomes. Evaluations should encompass weight control, diet, exercise, glycemic control, smoking status, lipid management, and blood pressure control. The American Southwest is an appropriate region for evaluating racial and ethnic differences in diabetes care because of the high prevalence of disease and the large minority populations. The purpose of this study was to examine the association between race/ethnicity and control of cardiovascular disease risk factors, adjusted for socioeconomic, clinical, and behavioral factors, in a large cohort of insulin-treated veterans with type 2 diabetes.</p></sec><sec sec-type="methods"><title>Methods</title><p>The Diabetes Outcomes in Veterans Study (DOVES) was a prospective, observational study of insulin-treated veterans with type 2 diabetes mellitus, designed to examine the association between clinical, demographic, lifestyle, socioeconomic, and psychological variables and the clinical outcomes of glycemic control and disease management. DOVES was conducted under the auspices of the Southwestern Group for Outcomes Research in Diabetes (SWORD), a consortium of the largest VA facilities in Veterans Integrated Service Network 18. The protocol for this study was described in detail elsewhere [<xref ref-type="bibr" rid="B22">22</xref>]. The institutional review board of each study site approved the protocol. Briefly, computerized pharmacy records were used to identify insulin-treated patients receiving care at the New Mexico VA Health Care System (Albuquerque, NM), the Carl T. Hayden VA Medical Center (Phoenix, AZ), and the Southern Arizona VA Health Care System (Tucson, AZ). Patients were eligible for this study if they had type 2 diabetes diagnosed after age 35; took at least daily one injection of a long-acting insulin preparation, did not self-titrate their insulin doses; and their diabetes medication regimen had been relatively unchanged in the preceding 2 months (the dose of all oral hypoglycemic medications remained unchanged, no oral hypoglycemic medications were added to the treatment regimen, and the total daily insulin dose was changed by no more than 10 units or 15%, whichever was less), thus ensuring that their A1c had equilibrated at the current insulin dose.</p><p>We excluded patients with less than a one-year expected survival; alcoholism or substance abuse listed as an active problem in the electronic medical record; a history of diabetic ketoacidosis or type I diabetes; or co-morbidities affecting glucose homeostasis: diabetes resulting from pancreatitis or pancreatic resection; cirrhosis, chronic active hepatitis, hemochromatosis, Wilson's disease or other liver disease; endocrinopathies such as pituitary adenoma, Cushing's or Addison's disease; hereditary or acquired forms of insulin resistance; glucocorticoid treatment; immunosuppression or treatment with immunosuppressant drugs; or chronic infectious diseases (e.g. osteomyelitis or refractory skin ulcers). We also excluded homeless patients because they would have difficulty with the intensive self-monitoring of blood glucose required for the DOVES prospective study.</p><p>All measures in this report were collected at the two baseline visits, scheduled two weeks apart, with the exception of insulin doses measured at a 26-week follow-up visit. At the entry visit, research coordinators explained the project, answered questions, and obtained informed consent. All subjects underwent an evaluation of their psychological status, socio-cultural barriers to diabetes management, disabilities, dietary habits, exercise patterns, and micro- and macrovascular disease risk factors. Race/ethnicity was determined from a structured category question that asked the patient to describe his or her background. The listed responses included non-Hispanic white, Hispanic, Native American, African-American, and Asian. Subjects also had the option to check an "other" category and write in a response. Subjects then answered questions on their family obligations, living arrangements, means of transportation, occupation, and financial status. Subjects rated their physical ability to do the following activities: work, yard work, household projects, shopping, exercise, cooking or light housekeeping, and personal care. Research coordinators conducted structured interviews with study subjects to collect data about medical treatment, including insulin dose, number of injections, types of preparations, and the dose, type, and frequency of use of oral hypoglycemic medications. Research coordinators then ascertained the number of units of each insulin type at each daily dosing time and summed them to determine total insulin units per day.</p><p>Psychosocial testing was performed in private sessions during the baseline visit and the second visit two weeks later. A research coordinator was present to provide instructions, answer questions, clarify items, or in some cases, read the questions. Psychological instruments were administered in random order and included the University of Michigan Diabetes Knowledge Test [<xref ref-type="bibr" rid="B23">23</xref>], the Mini-Mental State Examination (MMSE) [<xref ref-type="bibr" rid="B24">24</xref>], the Geriatric Depression Scale [<xref ref-type="bibr" rid="B25">25</xref>], the Diabetes Family Behavior Check List [<xref ref-type="bibr" rid="B26">26</xref>], and the Diabetes Care Profile [<xref ref-type="bibr" rid="B27">27</xref>]. We used the Compendium of Physical Activities to rate physical activities [<xref ref-type="bibr" rid="B28">28</xref>] and the Fred Hutchinson Cancer Research Center Food Frequency Analysis to assess dietary habits [<xref ref-type="bibr" rid="B29">29</xref>]. Baseline physiologic measurements made upon entry to the study included hemoglobin A1c (A1c), blood pressure, height, weight, smoking status, and blood lipids.</p><sec><title>Statistical analyses</title><p>We analyzed group differences in continuous variables by the unpaired Student's t-test and one-way analysis of variance. For the latter procedure, homogeneity of variances was examined by Levene's test, and the Brown-Forsythe test was used in place of the standard ANOVA F-test if the variances were significantly different (performed on BMDP software). The Mann-Whitney U-test and the Kruskal-Wallis one-way analysis of variance by ranks were used for variables with highly skewed distributions. Group differences in nominal variables were tested by chi-square analysis. The relationship between continuous variables was examined by simple regression. We used stepwise multiple linear regression analyses to identify factors affecting the baseline A1c and daily insulin doses. Predictors associated with the dependent variable in univariate analysis (p < 0.10) were entered into multivariate model using a forward- and backward-stepping procedure with an α ≤ 0.05 to enter and an α > 0.10 to remove. Continuous variables were expressed as mean ± standard deviation (SD).</p></sec></sec><sec><title>Results</title><p>We identified over 10,000 insulin-treated patients at the three participating medical centers. We randomly selected approximately 3000 subjects from this list and invited 589 eligible subjects to participate. We enrolled 359 (61%) subjects, but subsequently excluded 21 with incomplete data, leaving a cohort of 338 for the analysis. Their mean age was 65.1 ± 9.7 years, 96% were men, and 59% were married. Based on self-description, the cohort was comprised of 226 (67%) non-Hispanic white, 72 (21%) Hispanic, and 35 (10%) African-American subjects. Over two-thirds of the subjects had at least one microvascular disease complication and an equivalent number had a macrovascular disease complication. Although average daily insulin doses were substantial (66 units), only one-third of the subjects were concurrently treated with an oral hypoglycemic medication. Most subjects had an elevated baseline A1c value suggesting poor glycemic control, including 99 (29%) with a value between 7.0% and 8.0%, and 148 (44%) with a value ≥ 8.0%. The median level of activity was 7 met-hours of activity per day (equivalent to 2.5 hours of light home activities plus 0.5 hours of moderate walking). Sixty-two percent had a BMI ≥ 30 and 22.2% were current smokers. However, average lipid and blood pressure measurements were close to the target values recommended by the American Diabetes Association (ADA) [<xref ref-type="bibr" rid="B30">30</xref>,<xref ref-type="bibr" rid="B31">31</xref>].</p><p>Clinical and socioeconomic characteristics stratified by race/ethnicity are presented in Table <xref ref-type="table" rid="T1">1</xref>. African-American subjects had the highest A1c and were most likely to have poor glycemic control. Fifty-one percent of African Americans had an A1c ≥ 8% compared to 49% for Hispanics and 40% for non-Hispanic whites (p = 0.27). Mean ± SD A1c levels were 7.9 ± 1.4% for non-Hispanic whites, 8.2 ± 1.6% for Hispanics, and 8.8 ± 2.9% for African Americans (P = 0.05). We found significant differences in the daily units of insulin, with non-Hispanic whites receiving 70.6 ± 48.8 units compared to 58.4 ± 32.6 units for Hispanics and 53.1 ± 36.2 units for African Americans (p < 0.01). However, we found no differences between minorities and non-Hispanic whites in the daily number of insulin injections, the number of different insulin preparations used, or the use of oral hypoglycemic medications. Body mass index, amount of exercise, smoking status, lipid levels, and blood pressures were also similar between groups.</p><p>We also evaluated potential barriers to care. African-American subjects considered themselves less disabled for work (Table <xref ref-type="table" rid="T1">1</xref>). Hispanic subjects had the most dependents, and reported the highest psychosocial barriers with respect to language preference, education, depression, diabetes knowledge, and performance on the MMSE (Table <xref ref-type="table" rid="T2">2</xref>). African-Americans perceived the fewest problems with glycemic control, had the fewest negative attitudes about diabetes, and had the highest self-ratings for self-care abilities and dietary adherence. They also tended to have the strongest convictions about the importance of self-care and the fewest perceived barriers to exercise.</p><p>We found that higher depression scores (r = 0.11, p = 0.049), greater work hours per week (r = 0.17, p = 0.002), greater number of household dependents (r = 0.13, p = 0.023), being employed (p = 0.004), and age (r = -0.21, p = 0.0001) were significantly associated with a higher baseline A1c. After adjusting for these covariates with a multivariate linear regression analysis, baseline A1c was significantly higher for African-American subjects (+0.93%, p = 0.002), though not for Hispanics (+0.25%, p = 0.29), compared to non-Hispanic whites.</p><p>We stratified daily insulin dose by race and ethnicity and level of glycemic control (Table <xref ref-type="table" rid="T3">3</xref>). We found that unadjusted daily insulin units increased monotonically with A1c in non-Hispanic whites, but not in Hispanics or African Americans. We found the most significant racial/ethnic differences in insulin doses among subjects with poorly controlled diabetes (A1c ≥ 8.0%), with non-Hispanic whites receiving approximately 22 daily units more than Hispanics and 26 daily units more than African Americans (p < 0.01). At follow-up, the treatment patterns remained essentially the same. Overall, 63% of each minority group and 70% of the non-Hispanic whites returned for the 26-week follow-up visit. For non-Hispanic whites, Hispanics, and African Americans, the mean ± SD daily insulin units were 71.1 ± 51.2, 63.1 ± 36.9, and 51.4 ± 26.6, respectively (p = 0.14). Among subjects with a baseline A1c ≥ 8.0%, the mean ± SD daily insulin units at follow-up also differed significantly (p = 0.04): non-Hispanic whites = 80.3 ± 51.3, Hispanics = 63.0 ± 34.0, and African Americans = 47.8 ± 23.1.</p><p>We found that race/ethnicity, BMI, baseline A1c, and use of any other oral hypoglycemic medication were significant predictors of daily insulin units on multivariate linear regression analyses (Table <xref ref-type="table" rid="T4">4</xref>). African-Americans received an insulin dose that was an average of 17.8 units less (P = 0.01) and Hispanics an average of 10.5 units less (P = 0.04) than non-Hispanic whites. No significant interactions were observed. When the adjusted model was limited to those with baseline A1c ≥ 8%, African-Americans received an insulin dose that was an average of 26.5 units less (P = 0.01) and Hispanics an average of 15.9 units less (P = 0.03) than non-Hispanic whites. These differences persisted after adjusting for practice site.</p></sec><sec><title>Discussion</title><p>We evaluated racial/ethnic differences in glycemic control, medical therapy, and control of cardiovascular disease risk factors in insulin-treated Southwest American veterans with type 2 diabetes. We found that African Americans and Hispanics had poorer glycemic control and received less intensive insulin treatment, particularly African-Americans with A1c ≥ 8.0% who received over 25 units of insulin less per day less than non-Hispanic whites. We found that blood lipids and blood pressure control were close to ADA target values for all racial/ethnic groups. This finding is consistent with previous findings that failure to intensify insulin treatment contributed to poor glycemic control in urban African-Americans [<xref ref-type="bibr" rid="B32">32</xref>]. This under-treatment is an important health problem, as evidence suggests that African Americans respond better to insulin treatment [<xref ref-type="bibr" rid="B33">33</xref>].</p><p>Several psychosocial factors and barriers to care or self-care varied among the racial/ethnic groups in this study, and could be related to the observed disparities in glycemic control. Hispanics were the most disadvantaged group in terms of language preference and education and scored higher on the depression inventory. These factors may explain their performance on the MMSE and Diabetes Knowledge tests – tasks that required comprehension of complex instructions. On the other hand, African-Americans had more favorable responses in several areas rated by the Diabetes Care Profile. We could not readily attribute poorer glycemic control to a negative outlook in this group.</p><p>Other investigators have found that racial/ethnic differences in attitudes towards type 2 diabetes might contribute to poor outcomes [<xref ref-type="bibr" rid="B34">34</xref>,<xref ref-type="bibr" rid="B35">35</xref>]. Possibly, instruments specifically developed for that purpose might have explained some of the effects of race/ethnicity on glycemic control and treatment intensity. For example, the Veterans Ecocultural Self Report [<xref ref-type="bibr" rid="B36">36</xref>] adaptation measure better explains the effect of minority status on glycemic control than does Hispanic ethnicity. Our inability to identify these factors may also be due to the fact that our instrument did not specifically target insulin therapy. Hunt and associates [<xref ref-type="bibr" rid="B37">37</xref>] used an open-ended interviewing technique to examine the attitudes of 44 low-income Mexican Americans towards insulin therapy. Negative aspects were much more frequently discussed than positive aspects and focused on anxiety about pain, proper techniques, disrupting daily activities, low blood sugars, and other complications of therapy. The subjects also expressed concern that previous treatment efforts had failed and that the disease had progressed into a more serious phase. Our study suggests that integrating an abbreviated interview technique could be considered for future studies of minority subjects with an inadequate response to insulin therapy.</p><p>Potential cultural barriers to a more intensive insulin regimen in minorities also include: 1) a greater aversion to parenteral injections or to multiple injections; 2) greater fear of hypoglycemic events; 3) greater aversion to glucose monitoring; and 4) cultural barriers relating to images of wellness, such as a cultural aversion to public display of illness.</p><p>This study has some potential limitations. We cannot explain why minorities with poor glycemic control did not receive higher insulin doses. One possibility is a selection bias that precluded the enrollment of minorities with more aggressive insulin regimens and better glycemic control. We randomly selected subjects from a sample frame generated from administrative pharmacy files; however, a number of eligible patients refused to participate. Because participation in this study required time, travel, disruption of daily routines, and some discomfort, even minor cultural barriers may have played a role in the loss of more intensively-treated minorities. We were unable to ascertain the race/ethnicity of non-participants. Since provider adjustments to insulin dose are not captured in the administrative pharmacy database, we relied on self-report from patients for current insulin units per day via structured interviews with research coordinators. While this was the best practical source of information, it is possible that minority patients systematically under-reported their insulin dose. Another possibility for the difference in treatment intensity is that minority patients had fewer clinic visits. We were unable to assess this because we did not have data on the number of clinic visits since being started on insulin. However, it is unlikely that an effect from fewer visits would persist over an average of 8 years of insulin treatment. Additionally, given the equal access to VA health care shared by all veterans, the frequency of clinic visits would not likely be an explanation for differential treatment, but rather a marker for confounding socioeconomic or psychological factors. Finally, because of the population studied, our results may be less generalizable to younger patients, non-veterans, or women.</p><p>Insulin treatment may have been less intensive in minorities for clinical reasons. Hypoglycemia is a major deterrent to tight glycemic control. Although age has been identified as a risk factor for drug-induced hypoglycemia [<xref ref-type="bibr" rid="B38">38</xref>], the roles of race and ethnicity have not been established. Possibly, some providers perceive a higher hypoglycemia risk for minority patients that may affect treatment intensity. This hypothesis requires further evaluation. The risk of hypoglycemia could be affected by dietary habits that are influenced by race and culture. Another possibility is racial/ethnic differences in the intensity of glucose monitoring. Because monitoring is used to titrate insulin doses, targets may not have been reached in those who tested less frequently. In 1993, Harris and co-workers [<xref ref-type="bibr" rid="B39">39</xref>] reported data from the 1989 National Health Interview Survey on 2,405 diabetic patients ≥ 18 years of age. They found that African-Americans were 60% less likely to test their blood glucose at least once daily compared to non-Hispanic whites and Hispanics. The effect of race/ethnicity was independent of age, insulin use, education, intensity of physician visits, or diabetes education. Unfortunately, we did not obtain information on monitoring practices or the rate of hypoglycemia when patients entered this study.</p><p>Finally, it should be noted that the differences among the racial/ethnic groups were limited to insulin use. We found no racial/ethnic differences in the use of oral hypoglycemic medications or in the control level of 5 other risk factors for cardiovascular disease (weight exercise, smoking status, lipids, and blood pressure). In contrast, Heisler and co-workers [<xref ref-type="bibr" rid="B40">40</xref>] found that African American veterans with type 2 diabetes were more likely to have poor lipid and blood pressure control compared to white veterans, although there was no difference in intensity of treatment for those in poor control (the intensity of the glycemic regimen was not measured). Our study may have been underpowered to detect small differences in lipids, blood pressure, and other risk factors. However, the finding of comparable control across racial/ethnic groups suggests that subjects did not face substantial access barriers and that providers were addressing these cardiovascular risk factors. However, glycemic control and insulin treatment require more motivation and patient education than other aspects of cardiovascular disease risk factor control. Failure to achieve treatment goals may have been due to specific problems with the patient-provider interaction. One possibility is that the patient was not able to establish an effective therapeutic relationship with a provider of another race. A second possibility is that the providers were not aware of the specific needs of minorities with diabetes [<xref ref-type="bibr" rid="B41">41</xref>]. Providers often address multiple acute and chronic conditions during medical encounters; this may present a barrier for consistently providing preventive services and optimal disease management for diabetes, particularly if minority patients have more comorbidity than non-Hispanic white patients [<xref ref-type="bibr" rid="B42">42</xref>,<xref ref-type="bibr" rid="B43">43</xref>]. Although provider effects are an important determinant of diabetes control, we did not have sufficient power to model this factor in our analyses.</p></sec><sec><title>Conclusion</title><p>In summary, insulin-treated veterans who are minorities may have an increased risk of poor glycemic control and receiving lower doses of insulin. African-Americans in this sample were the most likely patients to experience this problem, while Hispanics had an intermediate risk. No racial or ethnic differences were found for the control of other cardiovascular disease risk factors. The disparities in glycemic control and treatment intensity could not be explained by the socioeconomic barriers, attitudes, level of knowledge, depression, cognitive dysfunction, or social support rated by the instruments in this study. The treatment disparity could be due to provider behaviors and/or patient behaviors or preferences. Further research with larger sample sizes and more geographically diverse populations are needed to confirm our findings and to elucidate the reasons for any observed disparities.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>CSW served as data manager and analyst and drafted the manuscript. JHS participated in study conception, design, management, and interpretation. WCD participated in study conception, design, management, and interpretation. RMH participated in study conception, design, and interpretation, and helped to refine the manuscript. MJM participated in study conception and design, and helped to refine the manuscript. GHM conceived of and designed the study, managed it as PI, lead the statistical analysis, and helped to draft the manuscript. All authors read and approved the final manuscript.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1472-6963/6/58/prepub"/></p></sec>
The basic principles of migration health: Population mobility and gaps in disease prevalence	Could not extract abstract	<contrib contrib-type="author"><name><surname>Gushulak</surname><given-names>Brian D</given-names></name><address><email>[email protected]</email></address><xref ref-type="aff" rid="Aff1">1</xref><xref ref-type="aff" rid="Aff3">3</xref></contrib><contrib contrib-type="author" corresp="yes"><name><surname>MacPherson</surname><given-names>Douglas W</given-names></name><address><email>[email protected]</email></address><xref ref-type="aff" rid="Aff1">1</xref><xref ref-type="aff" rid="Aff2">2</xref><xref ref-type="aff" rid="Aff3">3</xref></contrib><aff id="Aff1"><label>1</label>Migration Health Consultants, Inc., Vienna, Austria </aff><aff id="Aff2"><label>2</label><institution-wrap><institution-id institution-id-type="GRID">grid.25073.33</institution-id><institution-id institution-id-type="ISNI">0000000419368227</institution-id><institution>Faculty of Health Sciences, </institution><institution>McMaster University, </institution></institution-wrap>Hamilton, Ontario Canada </aff><aff id="Aff3"><label>3</label>Migration Health Consultants, Inc., Cheltenham, Ontario Canada </aff>	Emerging Themes in Epidemiology	<sec id="Sec1"><title>Introduction</title><p>The relationships between disease, travel and migration have historical roots that continue to influence modern medical activities [<xref ref-type="bibr" rid="CR1">1</xref>]. Traditional medical approaches dealing with migrant health have focused on the recognition, identification and management of specific diseases, illnesses or health concerns in mobile populations at the time and place of their arrival [<xref ref-type="bibr" rid="CR2">2</xref>]. These activities have often been based on the principles of protecting the recipient population through policies of exclusion directed at the migrant or arriving traveller. Derived from the historical practices of quarantine, similar processes continue in a modern context through immigration medical screening [<xref ref-type="bibr" rid="CR3">3</xref>] and border control practices intended to reduce threats to public health or to mitigate potential impacts on healthcare services.</p><p>The epidemiological analysis of illnesses and disease in migrants is most commonly approached in one of two ways in receiving countries. The first is to consider the health issue of concern in terms of the status at the time of migration, while the second is to study the evolution of the health characteristic over time [<xref ref-type="bibr" rid="CR4">4</xref>]. The reference population for the first analytical approach is normally the host or receiving population, while the reference group for the second approach can be either the host population or a comparison cohort at the migrants' place of origin (see table <xref rid="Tab1" ref-type="table">1</xref>).<table-wrap id="Tab1"><label>Table 1</label><caption><p>Epidemiological approaches to health and migration</p></caption><table frame="hsides" rules="groups"><thead><tr><th align="left">Time Period</th><th align="left">Epidemiological Characteristic</th><th align="left">Reference Population</th><th/></tr></thead><tbody><tr><td align="left"> <italic>Arrival Phase (Point Prevalence)</italic> </td><td align="left">Prevalence</td><td/><td align="left">Host/Receiving population</td></tr><tr><td align="left"> <italic>Post Arrival Phase (Point Prevalence or Longitudinal Studies)</italic> </td><td align="left">Incidence/Prevalence</td><td align="left">Similar non-migrating cohort at migrants' place of origin</td><td align="left">Host/Receiving population</td></tr></tbody></table></table-wrap></p><p>The quarantine-associated historical basis of migration health practices has ensured that much of the interest in health and migration has been directed towards communicable diseases [<xref ref-type="bibr" rid="CR5">5</xref>]. Commonly, migrant medical screening focuses on conditions differentially prevalent between the migrant and host population, such as tuberculosis [<xref ref-type="bibr" rid="CR6">6</xref>], leprosy [<xref ref-type="bibr" rid="CR7">7</xref>], or syphilis [<xref ref-type="bibr" rid="CR8">8</xref>]. Medical screening has been used to quantify and document aspects of health and disease in migrant cohorts, most often in relation to national public health statistics. Over time, these studies have described some of the immediate and long-term impacts of population movement in individual migrant receiving nations.</p><p>Recently, the growing international importance of migration has stimulated new interest in other aspects of migrant health. In addition to communicable diseases, attention is now focused on pre-existing non-infectious diseases [<xref ref-type="bibr" rid="CR9">9</xref>] and other health domains, including behaviour [<xref ref-type="bibr" rid="CR10">10</xref>], morality [<xref ref-type="bibr" rid="CR11">11</xref>] and genetic or ethnic profiles [<xref ref-type="bibr" rid="CR12">12</xref>] in migrant populations. Epidemiological studies now involve chronic illnesses [<xref ref-type="bibr" rid="CR13">13</xref>] such as malignancies [<xref ref-type="bibr" rid="CR14">14</xref>], renal failure [<xref ref-type="bibr" rid="CR15">15</xref>] and severe cardiac disease [<xref ref-type="bibr" rid="CR16">16</xref>], as well as mental and psychosocial health [<xref ref-type="bibr" rid="CR17">17</xref>] and maternal and child health [<xref ref-type="bibr" rid="CR18">18</xref>]. Lifestyle-associated health issues, including tobacco use, alcohol consumption and substance abuse, are also being examined in relation to the process of migration in some migrant receiving countries [<xref ref-type="bibr" rid="CR19">19</xref>].</p><p>As migrant demographics often vary between receiving nations, international comparisons involving the pooled analysis of several host nations would be challenging in interpretation and of questionable utility. However, interest in the global implications of the epidemiological aspects of migration is growing [<xref ref-type="bibr" rid="CR20">20</xref>]. The increasing desire for improved information is, however, complicated by the greater diversity now manifest in modern migrant populations. In additional to traditional immigrants, current mobile populations are often comprised of several other cohorts that are not similarly distributed between migrant receiving nations. Those other groups include refugees and asylum seekers, temporary migrants such as international students and migrant workers, and complex groups of irregular or illegal migrants, including those who have arrived through smuggling or trafficking.</p><p>The current volume and diversity in migration often exceeds the scope and intent of the traditional methods used to assess and manage health issues in immigrants. As a consequence of this more diverse demographic and fluid environment, the perspectives derived solely from traditional immigrant medical screening practices may be limited. These limitations may be overcome by applying population health principles to the study of migration, a process that may result in observations that are more applicable to immigration health policy and programme development [<xref ref-type="bibr" rid="CR21">21</xref>].</p><p>A population health-based approach considers the relationship between migration and health as a progressive, interactive process influenced by temporal and local variables. The observations are less related to the administrative mechanics of migration [<xref ref-type="bibr" rid="CR22">22</xref>] and more sensitive to the driving forces that cause people to migrate. A population-based approach also facilitates the consideration and study of the long-term consequences of movement between locations with different health determinants and health outcomes. Use of this approach supports the examination of the issues from a global perspective. Such methodologies are already in use in the health sector. The Global Fund approach to tuberculosis, malaria and HIV similarly represents an integrated, inter-regional action plan in the face of persistent global health challenges [<xref ref-type="bibr" rid="CR23">23</xref>].</p><p>As presented below, considering migration-relevant diseases in terms of population-based risk may be more relevant to disease control programs than time-of-entry screening for individual conditions. Evaluating the migration and mobility history of populations has the additional advantage of supporting longitudinal analysis of health characteristics. This context is important in the consideration of illnesses and diseases with long latency or delayed diagnostic time [<xref ref-type="bibr" rid="CR24">24</xref>], an area beyond the scope of traditional time-of-entry immigration screening. It also facilitates investigation of the acquisition of positive and negative health attributes by migrants following arrival [<xref ref-type="bibr" rid="CR25">25</xref>], which is also of increasing interest in migration health. Finally, population-based methods reflect the nature and role of globalisation and can be used to support and rationalise strategies to deal with health challenges at their source.</p><p>The population health approach to migration health is based on the standardised examination of two factors: (1) sustained disparate health environments and (2) the movement of populations between regions of differential prevalence of health indicators and outcomes.</p><sec id="Sec2"><title>The dynamics of health disparities</title><p>Some diseases or illnesses are sustained by differences that are purely geographic or environmental in origin. In other situations, differences in health outcomes, and the factors that determine or influence health outcomes, result from more complex interactions. The environment [<xref ref-type="bibr" rid="CR26">26</xref>], socio-economics, genetics and biology, and behavioural factors influence population measurements of disease prevalence individually and in combination.</p><p>Examples of environmentally-limited diseases include vector-borne conditions, for which environmental factors determine the distribution of disease transmission, as observed in the global epidemiology of malaria, Chagas' disease, yellow fever and West Nile Virus. Environmentally-related non-communicable disease epidemiological disparities include micronutrient deficiencies [<xref ref-type="bibr" rid="CR27">27</xref>] and geographically-defined exposure risks, such as health outcomes related to extreme weather or altitude [<xref ref-type="bibr" rid="CR28">28</xref>]. Movement of the population out of the risk environment (i.e. African refugees in Europe and North America and malaria) or establishment of disease transmission outside of the usual environmental constraints (i.e. West Nile Virus in North America) will impact on the epidemiology of the condition in the receiving region and on the local population health outcomes.</p><p>Social and economic influences can be significant factors in the creation and maintenance of differences in health and disease outcomes between populations. Poverty, education, housing and nutrition are directly related to disease prevalence and illness outcomes [<xref ref-type="bibr" rid="CR29">29</xref>]. The capacities and capabilities of medical and health sectors can affect health through the availability, accessibility [<xref ref-type="bibr" rid="CR30">30</xref>] and affordability of health promotion, disease prevention and treatment services [<xref ref-type="bibr" rid="CR31">31</xref>, <xref ref-type="bibr" rid="CR32">32</xref>]. Additional factors that influence health risks and outcomes include language skills [<xref ref-type="bibr" rid="CR33">33</xref>], behavioural and cultural practices [<xref ref-type="bibr" rid="CR34">34</xref>, <xref ref-type="bibr" rid="CR35">35</xref>], such as the use of tobacco, dietary practices and population norms for body mass and physical exercise [<xref ref-type="bibr" rid="CR36">36</xref>].</p><p>Migrants and other mobile populations reflect the health characteristics of their place and environment of origin and carry several of these characteristics with them when they move [<xref ref-type="bibr" rid="CR37">37</xref>, <xref ref-type="bibr" rid="CR38">38</xref>]. In addition, migrants are also subject to other specific influences that may affect their health. These factors result from the process of migration itself, for example, during the travel phase between origin and destination. This is frequently observed in refugees, displaced persons and disadvantaged migrant populations such as trafficked or smuggled persons [<xref ref-type="bibr" rid="CR39">39</xref>–<xref ref-type="bibr" rid="CR41">41</xref>], and includes events such as trauma and torture [<xref ref-type="bibr" rid="CR42">42</xref>, <xref ref-type="bibr" rid="CR43">43</xref>]. Other migration-specific health influences are observed in migrant worker populations [<xref ref-type="bibr" rid="CR44">44</xref>, <xref ref-type="bibr" rid="CR45">45</xref>], the children of migrants [<xref ref-type="bibr" rid="CR46">46</xref>] and returning travellers who have been visiting family and friends [<xref ref-type="bibr" rid="CR47">47</xref>]. A demographic and mobility process approach to these considerations is presented in table <xref rid="Tab2" ref-type="table">2</xref>.<table-wrap id="Tab2"><label>Table 2</label><caption><p>The impact of different health environments and the phases of population mobility</p></caption><table frame="hsides" rules="groups"><thead><tr><th align="left">Occurrence</th><th align="left">Examples</th><th align="left">Consequence at Destination</th></tr></thead><tbody><tr><td align="left"> <italic>Pre-departure existing medical condition</italic> </td><td align="left"><p>- prevalence of endemic disease</p><p>- level of development</p><p>- access to care</p><p>- availability of care</p></td><td align="left"><p>Arriving population displays health indicators of origin:</p><p>• Differing incidence and prevalence of illness</p><p>• Differences in awareness of and use of healthcare services:</p><p>• preventive</p><p>• promotional</p><p>• diagnostic</p><p>• therapeutic</p></td></tr><tr><td align="left"> <italic>Health impacts during migration</italic> </td><td align="left"><p>- trauma (physical-psychosocial)</p><p>- deprivation</p><p>- violence</p><p>- exposure</p><p>- injury</p></td><td align="left"><p>Some populations display greater prevalence of illness resulting from torture, trauma, abuse and exposure</p><p>• Refugees</p><p>• Refugee claimants or asylum seekers</p><p>• Trafficked/smuggled migrants</p></td></tr><tr><td align="left"> <italic>Health impacts arising after arrival</italic> </td><td align="left"><p>administrative/legal limits</p><p>- poverty</p><p>- language culture</p><p>- occupational risks</p></td><td align="left"><p>Awareness of and use of healthcare services in migrant populations may be limited by immigration status, poverty, language and culture</p><p>Working conditions may be associated with health risks:</p><p>• Migrant agricultural labor</p><p>• Commercial sex workers</p><p>• Illegal workers</p><p>• Trafficked migrants</p></td></tr><tr><td align="left"> <italic>Health consequences of return travel</italic> </td><td align="left"><p>Health environment at origin may have changed</p><p>- health systems improvements or declines</p><p>Children born to foreign-born parents have no exposure to risks present at origin</p></td><td align="left"><p>Populations making return journeys to place of origin (particularly children born at new destination) may be at increased risk of disease or illness:</p><p>"Visiting friends and relative" travellers</p><p>- Locally born children of foreign-born parents</p></td></tr></tbody></table></table-wrap></p><p>The impact of genetic and biological determinants of health and disease may be intuitively self-evident. However, in non-endemic regions these influences, and their linkages to population mobility, may be poorly appreciated in the early phases of migration due to lack of awareness, knowledge or experience in the healthcare delivery sector [<xref ref-type="bibr" rid="CR48">48</xref>].</p><p>Disparities in health determinants and disease outcomes are not absolute, but change over time. This temporal variability adds an important dimension of complexity to the analysis and investigation of migrant health concerns, which can affect cohort comparability. Economic and social environments can change rapidly in the modern world. If those changes influence health determinants, consequential changes in health outcomes can be observed over relatively short periods of time. For example, in the thirty years following 1965, the difference between life expectancy for males in the United Kingdom and Russia increased by more than ten years (range from 3.6 to 15.1 years) [<xref ref-type="bibr" rid="CR49">49</xref>]. Basic public health improvements such as the provision of adequate, safe drinking water, improved sewerage and housing can significantly reduce the incidence and prevalence of diseases of major public health importance in the space of less than a generation [<xref ref-type="bibr" rid="CR50">50</xref>]. Similarly, conflict, environmental change, natural disasters and population growth can result in new risk exposures and acquisition of adverse health outcomes over short time periods [<xref ref-type="bibr" rid="CR51">51</xref>]. Genetic admixture and behavioural characteristics of individuals and populations that impact on health outcomes can also occur singly and in combination with other determinants vary over time.</p><p>These rapid temporal changes influence the epidemiology of health disparities. If they take place against a background of sustained or growing migration, they may also influence the interpretation of longitudinal and comparative studies involving migrants. Health outcomes for current departing migrants may differ significantly from previous cohorts originating at the same location, as health indicators change over time at the migrants' place of origin. Depending on the nature of the local changes, either improved or worsening health characteristics in current migrant cohorts may be observed. Examples of this ''time phase'' phenomenon can be seen in the improved population health indicators of several Asian nations that have occurred following recent trends in large-scale migration from Asia to Europe and North America. Some of these changes are significant, as indicated by the changing mortality data due to hypertensive cardiac disease in Korea. During the interval between 1984 and 1999, the age-adjusted mortality for hypertensive cardiac disease in Korean men decreased by 92% (from 51.6 to 4.1/100,000) and 84% for women (from 34.1 to 5.3/100,000) [<xref ref-type="bibr" rid="CR52">52</xref>]. By contrast, examples of diminishing levels of good health indicators can be observed in some Central and Eastern European nations in the period following the dissolution of the Soviet Union [<xref ref-type="bibr" rid="CR53">53</xref>].</p></sec><sec id="Sec3"><title>Modern migration and population mobility</title><p>While migration has always been a fluid process subject to change, these changes must now be assessed in terms of the rate of change and global magnitude of population movement. During the past 50 years, the process of migration and concomitant movement of other mobile populations has been markedly influenced by:<list list-type="order"><list-item><p>The decolonialisation of many nations, including those in Africa, the Middle East, Asia, Latin America and the Caribbean;</p></list-item><list-item><p>Large refugee movements following conflicts and civil disturbances, including South East Asia, the Balkans, Central America and Central Africa; <italic>and</italic></p></list-item><list-item><p>The political, social and economic consequences of the collapse of the former Soviet Union.</p></list-item></list></p><p>As a result of these events, between 1960 and 2000 the legal and administrative restrictions on the ability to travel, work and move internationally have changed for hundreds of millions of individuals. This has been associated with a profound shift in the demography of people on the move and the nature of migration itself [<xref ref-type="bibr" rid="CR54">54</xref>]. In traditional migrant-receiving regions such as Australia and North America, patterns of migrant origin have shifted from Europe to source countries in Asia, Africa, Central and South America and the Middle East [<xref ref-type="bibr" rid="CR55">55</xref>] in the timespan of little more than one generation.</p><p>This evolution has not been limited to regulated, traditional immigration and emigration. It has also involved refugee and humanitarian movements and an increase in irregular arrivals (refugee claimants, asylum seeking, smuggling and trafficking in humans). Complex humanitarian emergencies are often associated with large population displacements of refugees, humanitarian evacuees and other displaced populations. Unlike refugee movements before the Second World War that were often passive, modern international attention and efforts are frequently directed at assisting the international relocation of vulnerable migrant populations. International organisations such as the United Nations High Commission for Refugees [<xref ref-type="bibr" rid="CR56">56</xref>] and the International Organization for Migration [<xref ref-type="bibr" rid="CR57">57</xref>], as well as other infrastructures, now support and facilitate the selection, movement and resettlement of these populations. These activities are now global in scope and planning, involve an increasing number of nations and, when triggered, take place more rapidly than historical refugee resettlements [<xref ref-type="bibr" rid="CR58">58</xref>].</p><p>Against this backdrop of political, social and civil societal change, the nature, speed and access to international travel has also undergone marked evolution. Travel patterns have been affected by changes in transportation technology, accessibility, and affordability. Growth in air travel has functionally reduced previous limits on the rapid international movement of large numbers of individuals. In 1960, there were approximately 70 million international journeys globally. The number of similar international journeys in 2004 was in excess of 760 million [<xref ref-type="bibr" rid="CR59">59</xref>]. The high volume of international travel supports greater population exchange and return flows between migrant origin and destination locations.</p><p>Increased international travel has also been an integral component of the growing process of globalisation. The progressive integration of global economic and communications sectors has been accompanied, if not preceded by, a corresponding growth in the international demand and flow of labour and manpower. The International Labour Organization estimates the foreign-born migrant labour force to be nearly 90 million persons worldwide [<xref ref-type="bibr" rid="CR60">60</xref>]. In several locations, there is a repetitive flow of workers between regions of origin and regions of employment. Some of these movements are regular and organised. However, modern population pressures and economic push-pull factors related to marked global differences in opportunity are increasingly associated with irregular population flows facilitated by either smuggling or trafficking of those seeking a better life [<xref ref-type="bibr" rid="CR61">61</xref>].</p><p>Another recent series of factors affecting population mobility and migration has resulted from geo-political changes such as the collapse of the former Soviet Union [<xref ref-type="bibr" rid="CR62">62</xref>]. The resulting economic, social and political consequences have both direct and indirect impacts on the demography and movement patterns of migrants to Western Europe, parts of Middle-East Asia and the Americas [<xref ref-type="bibr" rid="CR63">63</xref>]. Current immigration, family reunification and return dynamics in Europe and Central Asia are profoundly different from those of only two decades previously. These changes have been manifested in both voluntary migration, such as that observed in migrants from the former Soviet Union to Israel [<xref ref-type="bibr" rid="CR64">64</xref>], as well as the flow of refugees and those displaced by regional conflict, as exemplified in the Balkans in the mid to late 1990s [<xref ref-type="bibr" rid="CR65">65</xref>].</p></sec><sec id="Sec4"><title>The epidemiological consequences of mobility across differentials in disease prevalence</title><p>In much of the developed world, infections that were historically significant causes of illness and death have decreased in incidence and prevalence or been eliminated. This was accomplished through sanitation, immunisation, antibiotic therapy, and improved healthcare and public health practices. By the end of the last millennium, several globally important infections had reached the point where they were no longer of public health significance in economically advantaged areas of the world. In some areas of the developed world, domestic transmission of serious infectious diseases, such as measles [<xref ref-type="bibr" rid="CR66">66</xref>] and polio [<xref ref-type="bibr" rid="CR67">67</xref>], has been eliminated. This is far in advance of what was observed or possible in the developing world. This can create enormous differences in the prevalence of certain conditions between locations. In a mobile world, travellers and migrants crossing these prevalence gaps can become the source for outbreaks of these diseases [<xref ref-type="bibr" rid="CR68">68</xref>].</p><p>The creation of such prevalence differences is not limited to communicable diseases. The ability to treat and manage non-infectious diseases in highly developed regions of the world likewise differs from the situation in many developing regions. Access to and use of complex and costly interventions, such as cancer treatment, organ system support, transplantation and extensive pharmacotherapy vary according to levels of national economic development. These disparities in social investments in health services availability and population access to healthcare are associated with several differential inter-regional health outcomes, including premature death and increased morbidity in the developing world [<xref ref-type="bibr" rid="CR69">69</xref>].</p><p>In the absence of extensive international travel, population mobility and migration, the effects of differences in disease prevalence would have limited global significance. Nations and regions would strive to improve their domestic health capacities and reduce the domestic burden of disease and illness within their population much as they have done throughout history. However, expanding travel and migration across these prevalence differentials now function as an increasing, population-based bridge between the disparities. The net result is the global extension of what was predominantly a local risk.</p><p>The global extension of regional epidemiological differences can have marked impact on the local epidemiology of disease. Examples include tuberculosis, sexually-acquired infections, Chagas' disease and strongyloidiasis, for which foreign-born migrants from hyper-endemic areas represent the majority of national case burden in low incidence or non-endemic nations [<xref ref-type="bibr" rid="CR70">70</xref>–<xref ref-type="bibr" rid="CR73">73</xref>]. Similar patterns of epidemiological evolution exist for other long-term infections such as hepatitis B and C, as well as HIV/AIDS in some European nations, where these diseases are reported to be more common in foreign-born migrants than in native-born residents [<xref ref-type="bibr" rid="CR74">74</xref>–<xref ref-type="bibr" rid="CR76">76</xref>].</p><p>Migration-related epidemiological influences on domestic disease patterns for migrant-receiving nations are also observed for non-infectious diseases [<xref ref-type="bibr" rid="CR77">77</xref>]. Migrants arriving from less developed regions of the world may have had less access to preventive care, health promotion programmes, and diagnostic or therapeutic interventions for illness and disease. Cancer detection programs [<xref ref-type="bibr" rid="CR78">78</xref>] and periodic health examinations may not be commonly accessible in many populations. Access to healthcare providers and basic services are unequally distributed or subject to limited availability in many places. Similar differences in availability can be observed for smoking and substance abuse prevention programmes, programmes to detect and manage vitamin or micronutrient deficiencies, promotion of dental health and programmes to manage genetic or biological conditions [<xref ref-type="bibr" rid="CR79">79</xref>]. As a consequence, migrants may present with disease in more advanced stages than normally observed by providers in the destination country [<xref ref-type="bibr" rid="CR80">80</xref>].</p><p>In considering migration in a population health context, it is important to note that not all of the migration-associated epidemiological changes relate to situations where the migrants are less advantaged than the host population. In terms of lifestyle-related non-infectious illnesses, many migrants and new arrivals display health parameters that are better than those of the receiving population [<xref ref-type="bibr" rid="CR81">81</xref>]. Over time and resulting from a variety of factors including acculturation, diet and behavioural changes, immigrant populations may acquire and display common adverse health indicators more similar to those of the receiving population [<xref ref-type="bibr" rid="CR82">82</xref>]. Ironically, some of the beneficial factors arriving with immigrants may be adapted locally to the general benefit of the host population, but may be lost over time in immigrant population and their locally-born children.</p></sec><sec id="Sec5"><title>Temporal effects of migration on local health and disease epidemiology</title><p>Migration-associated influences on the epidemiology of disease have both immediate and long-term effects on host country health indicators due to differentials in disease prevalence, as well as magnitude factors associated with population census shifts (because of the number of migrants and births to the foreign-born cohort). For diseases of rare or limited occurrence, particularly where national incidence has been reduced to very low levels, the presentation of even a single case can have important implications locally and internationally [<xref ref-type="bibr" rid="CR83">83</xref>]. This can result in a heightened perception of threat to the public health of the local population and increase concerns regarding capacity and response of healthcare service delivery. Recent examples of this include the global public health control efforts resulting from the 2003 SARS events [<xref ref-type="bibr" rid="CR84">84</xref>, <xref ref-type="bibr" rid="CR85">85</xref>], avian-to-human influenza transmission [<xref ref-type="bibr" rid="CR86">86</xref>, <xref ref-type="bibr" rid="CR87">87</xref>], periodic outbreaks of viral haemorrhagic fevers [<xref ref-type="bibr" rid="CR88">88</xref>] and the impact of HIV/AIDS cases in Europe and North America [<xref ref-type="bibr" rid="CR89">89</xref>, <xref ref-type="bibr" rid="CR90">90</xref>] acquired abroad.</p><p>The continued arrival of new residents from high prevalence areas will contribute to the existing disease base of low incidence migrant-receiving locations. Over time, as observed with regard to globally prevalent but non-uniformly distributed diseases such as tuberculosis [<xref ref-type="bibr" rid="CR91">91</xref>], imported cases among migrants and other mobile populations can come to represent the majority of the case load in the recipient nation. In these situations, where domestic epidemiology comes to reflect global disease distribution through the process of migration, health policy implications become apparent. Long-term healthcare policy and planning in migrant-receiving nations will have to encompass an international and more global focus to be effective. Reliance on historical, domestic epidemiology for policy implementation in these nations will have limited relevance when disease volumes and case-burdens originate beyond the mandate and jurisdiction of national prevention and control efforts [<xref ref-type="bibr" rid="CR92">92</xref>]. For example, hepatitis A infection associated with population mobility in some western nations has raised concerns for the need of national domestic immunisation programs to deal with what is, in fact, the international persistence of the disease [<xref ref-type="bibr" rid="CR93">93</xref>].</p><p>In terms of non-infectious diseases, migration-associated pressures result from the need to provide service delivery in culturally or linguistically sensitive programmes for the prevention or treatment of illness in migrant communities. While uncomplicated in principle, the provision of health promotion or prevention advice, such as that recently recommended for asthma and atopy in migrant populations [<xref ref-type="bibr" rid="CR94">94</xref>], can be both a logistical and resource challenge [<xref ref-type="bibr" rid="CR95">95</xref>].</p><p>The reintroduction of diseases into low incidence locations through migration coupled with the growth of new populations with wide linguistic and cultural diversity can create difficulties in recognition, diagnosis and treatment [<xref ref-type="bibr" rid="CR96">96</xref>]. In some situations, delayed treatment can have important consequences. These consequences may directly affect the patient [<xref ref-type="bibr" rid="CR97">97</xref>], influence demands on programs and, for some contagious conditions, challenge the public health management of the exposed population. The growth and ease of travel, in conjunction with the increased numbers of those who are mobile, continues to raise the likelihood of the global extension and presentation of many diseases and illnesses. Maintaining sufficient degrees of clinical suspicion as well as laboratory diagnostic expertise, capacity and access can be costly and complicated. Growing needs for cultural and linguistic competencies in the health sector increases this complexity [<xref ref-type="bibr" rid="CR98">98</xref>]. Such preparedness requires the education, training and maintenance of competence of healthcare providers and the associated infrastructures for diagnosis and care [<xref ref-type="bibr" rid="CR99">99</xref>] (see table <xref rid="Tab3" ref-type="table">3</xref>).<table-wrap id="Tab3"><label>Table 3</label><caption><p>Health service issues resulting from international population mobility</p></caption><table frame="hsides" rules="groups"><thead><tr><th align="left">Immediate Term: Response to Imported/Introduced Illness</th></tr></thead><tbody><tr><td align="left"><p>Continued/Enhanced need for Clinical/Laboratory Capacities for imported diseases</p><p>• Provision of services and facilities</p><p>• Sustained laboratory capacity for rare or exotic diseases</p><p>• Continuing education of healthcare providers in global health issues</p><p>• Post-graduate training in low probability but high impact diseases</p><p>• Maintenance of competency in international health issues and response</p><p>• Contingency planning and exercise testing of plans</p><p>• Global health programs in universities and medical/nursing schools</p><p>• Development of specialized reference centers and international networks</p></td></tr><tr><td align="left"> <bold><italic>Long Term: Response to Growing Foreign Born Population Component</italic></bold> </td></tr><tr><td align="left"><p>Increasing Demands for Service/Access</p><p>Provision of appropriate diagnostic and treatment services</p><p>• Modification of training programs for health providers</p><p>• Translation and interpretation services</p><p>• Increased migration of health professionals from migrant source regions</p><p>• Training/certification of migrants who have linguistic skills</p><p>• Cultural awareness and sensitivity programs and training</p></td></tr></tbody></table></table-wrap></p></sec><sec id="Sec6"><title>Future impact of population mobility on global health</title><p>Negative health outcomes resulting from migration and population mobility can be expected to increasingly exert major influences on both national and global health planning. Mobility is a basic and fundamental component of the rapidly expanding globalisation process. Analysis suggests that the volume of immigration, travel and the migration of labour are expected to remain at current levels or increase for the foreseeable future. At the same time, current regional and global health disparities are anticipated to remain or increase, despite the international desire and efforts to reduce them. Global efforts aimed at reducing global disparities and impacts of disease and ill health, such as attempts to achieve the Millennium Development Goals [<xref ref-type="bibr" rid="CR100">100</xref>], are currently underway. These are long-term initiatives that will take time, resources and extensive effort to achieve and maintain. The longer inter-regional disparities in health and health outcomes persist, the longer they will continue to influence the health of migrants, as well as mobile and non-mobile populations, and the greater the challenge and cost will be to effect control of these conditions.</p><p>Historically, the majority of the health issues associated with migration, or occurring as a result of migration, have been managed at the national level. This has been accomplished through either immigration health activities or exclusion, or as a component of other domestic health programs. Demand for these services will remain in some locations due to specific situational issues or particular migratory movements. Nationally mandated immigration health screening [<xref ref-type="bibr" rid="CR101">101</xref>] and the management of health issues associated with the Hajj [<xref ref-type="bibr" rid="CR102">102</xref>] are two examples of those situations. In other areas, the evolution of travel and migration has reduced the effectiveness of many national, point-of-arrival activities. New patterns of population mobility require reconsideration of the practicality and viability of border-health inspection for exclusion or containment strategies. Some nations with large immigration medical programs maintain specific screening or intervention programs for targeted diseases such as tuberculosis, syphilis and HIV/AIDS. Accumulating evidence suggests that more effective outcomes could be obtained through interventions focused on disease control efforts in source nations rather than reliance on arrival screening alone [<xref ref-type="bibr" rid="CR103">103</xref>]. The globalisation of risks – manifest by the mobility of large numbers of individuals flowing across and between disparities in health environments and disease prevalence – will require increased investment in globally-focused resources and management commitments, as opposed to inward-looking national management strategies and programs.</p><p>The increased recognition of the importance of these issues has already resulted in reconsideration of international policy and program activity [<xref ref-type="bibr" rid="CR104">104</xref>]. Global response initiatives have been designed, and in some cases implemented, primarily as a result of concerns regarding the potential adverse population health outcomes posed by certain infectious disease threats. International collaboration is underway to track and monitor disease in the global context. Some of these activities are represented by the internationalisation of epidemiological surveillance, reporting and response reflected in the recent revisions to the International Health Regulations [<xref ref-type="bibr" rid="CR105">105</xref>].</p><p>It is evident that health program and policy planning processes need to anticipate and manage the future impacts of population mobility [<xref ref-type="bibr" rid="CR106">106</xref>]. In several western nations, migrants and other mobile populations represent sizeable and growing population sectors. Over time, the specific health needs and characteristics of these mobile population cohorts will exert greater influence on the health sector in the receiving host nations [<xref ref-type="bibr" rid="CR107">107</xref>]. The influence of genetic and biological factors in health and disease will increase in parallel with the size of diverse migrant populations. At the same time, the impact and influence of the health environment and epidemiology of disease at the migrants' place of origin will continue to be reflected in both migrants and the children of migrant parents [<xref ref-type="bibr" rid="CR108">108</xref>] long after the period of immigration.</p><p>Addressing these challenges at the national level will require some changes in the epidemiological context that is applied to prospective policy development and programme management (see table <xref rid="Tab4" ref-type="table">4</xref>). In a globalised world in which travel and migration represent the experience of a large and growing population cohort, they can be expected to assume a standard role as a determination factor influencing many health outcomes [<xref ref-type="bibr" rid="CR109">109</xref>]. In a manner similar to age, sex, genetics, biology, behaviour, and educational and wealth attainment, the mobility history at both the individual and population levels will need to become a routine consideration in healthcare policy, planning, education, training and service delivery [<xref ref-type="bibr" rid="CR110">110</xref>].<table-wrap id="Tab4"><label>Table 4</label><caption><p>Health policy issues resulting from international population mobility</p></caption><table frame="hsides" rules="groups"><tbody><tr><td align="left"><p>• National point-of-arrival activities – for example, immigration medical screening programs for specific targeted disease at the airport – will become less effective, more costly, and increasingly irrelevant</p><p>• International intervention programs for specific diseases at the migrants' country of origin may be more effective than national intervention programs dealing with low incidence diseases</p><p>• Mobility will become a more important determination factor influencing many health outcomes, along with age, sex, genetics, biology, behaviour and educational and wealth attainment</p><p>• Mobile population health policy frameworks will increasingly require integration and harmonisation at all jurisdictional levels with international economic, trade and security approaches</p></td></tr></tbody></table></table-wrap></p><p>Similar population-based approaches are already in use in some of the national and international infectious disease surveillance and monitoring systems described above [<xref ref-type="bibr" rid="CR111">111</xref>]. The analysis and interpretation of the information collected by those systems has demonstrated the importance of migration-related travel in terms of imported tropical infections [<xref ref-type="bibr" rid="CR112">112</xref>]. The extension of the collection of the travel and mobility history to non-infectious disease surveillance is both logical and supported by preliminary study.</p></sec><sec id="Sec7"><title>Summary</title><p>At the national and international level, the epidemiological outcomes and issues related to migration can be seen to result from the predictable effects of population flows between and across regional disparities and disease prevalence differentials. The growing number of migrants of diverse nature is bridging existing and developing gaps in health outcome indicators. The dynamics of migration and population mobility are evolving at a rate that creates health challenges for existing policy and programme frameworks that differ from those observed in historic migratory movements [<xref ref-type="bibr" rid="CR113">113</xref>].</p><p>The net result is an ongoing globalisation of health influences and indicators currently relevant at both national and global level. The epidemiological impact of population mobility is now evident in a considerable amount of infectious disease surveillance information [<xref ref-type="bibr" rid="CR114">114</xref>] and similar impacts can be anticipated for non-infectious illnesses in immigration receiving nations [<xref ref-type="bibr" rid="CR115">115</xref>].</p><p>As long as global health disparities and prevalence differentials exist, national health programs and policies in migrant receiving nations will continue to be challenged by illness and disease arising beyond their jurisdiction. National control and regulatory systems alone will be unable to extend their immediate mandate or authority to the source of the problem. To be effective, the management of health issues resulting from population mobility will require an integration of national and global health initiatives for both infectious [<xref ref-type="bibr" rid="CR116">116</xref>] and non-infectious [<xref ref-type="bibr" rid="CR117">117</xref>] disease conditions.</p></sec></sec>
Concomitant infections of <italic>Plasmodium falciparum </italic>and <italic>Wuchereria bancrofti </italic>on the Kenyan coast	<sec><title>Background</title><p><italic>Anopheles gambiae s.l</italic>. and <italic>An. funestus </italic>are important vectors of malaria and bancroftian filariasis, which occur as co-endemic infections along the Kenyan Coast. However, little is known about the occurrence and prevalence of concomitant infections of the two diseases in mosquito and human populations in these areas. This study reports the prevalence of concomitant infections of <italic>Plasmodium falciparum </italic>and <italic>Wuchereria bancrofti </italic>in mosquito and human populations in Jilore and Shakahola villages in Malindi, Kenya.</p></sec><sec sec-type="methods"><title>Methods</title><p>Mosquitoes were sampled inside houses by pyrethrum spray sheet collection (PSC) while blood samples were collected by finger prick technique at the end of entomological survey.</p></sec><sec><title>Results</title><p>A total of 1,979 female <italic>Anopheles </italic>mosquitoes comprising of 1,919 <italic>Anopheles gambiae s.l </italic>and 60 <italic>An. funestus </italic>were collected. Concomitant infections of <italic>P. falciparum </italic>sporozoites and filarial worms occurred in 1.1% and 1.6% of <italic>An. gambiae s.l </italic>collected in Jilore and Shakahola villages respectively. <italic>Wuchereria</italic>-infected mosquitoes had higher sporozoite rates compared to non-infected mosquitoes, but multiple infections appeared to reduce mosquito survivorship making transmission of such infections rare. None of the persons examined in Shakahola (n = 107) had coinfections of the two parasites, whereas in Jilore (n = 94), out of the 4.3% of individuals harbouring both parasites, 1.2% had <italic>P. falciparum </italic>gametocytes and microfilariae and could potentially infect the mosquito with both parasites simultaneously.</p></sec><sec><title>Conclusion</title><p>Concerted efforts should be made to integrate the control of malaria and bancroftian filariasis in areas where they co-exist.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Muturi</surname><given-names>Ephantus J</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Mbogo</surname><given-names>Charles M</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Mwangangi</surname><given-names>Joseph M</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Ng'ang'a</surname><given-names>Zipporah W</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>zipng'ang'[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Kabiru</surname><given-names>Ephantus W</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Mwandawiro</surname><given-names>Charles</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A7" contrib-type="author"><name><surname>Beier</surname><given-names>John C</given-names></name><xref ref-type="aff" rid="I5">5</xref><email>[email protected]</email></contrib>	Filaria Journal	<sec><title>Background</title><p>Malaria and lymphatic filariasis are the world's most important parasitic infections with an estimated loss of 4.5 × 10<sup>7 </sup>and 4.9 × 10<sup>6 </sup>disability adjusted life years (DALYs), respectively [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B2">2</xref>]. The two diseases occur as co-endemic infections in many tropical developing countries affecting the same human hosts and sharing common vectors [<xref ref-type="bibr" rid="B3">3</xref>]. In Papua New Guinea, Burkot <italic>et al</italic>. [<xref ref-type="bibr" rid="B4">4</xref>] reported the occurrence of concomitant infections of <italic>P. falciparum </italic>and <italic>W. bancrofti </italic>in <italic>An. punctulatus </italic>mosquitoes, whereas in India and South America, the two parasites have been reported to occur simultaneously in humans [<xref ref-type="bibr" rid="B5">5</xref>-<xref ref-type="bibr" rid="B7">7</xref>].</p><p>Several reports indicate that a high prevalence of bancroftian filariasis and falciparum malaria occurs along the Kenyan Coast [<xref ref-type="bibr" rid="B8">8</xref>-<xref ref-type="bibr" rid="B12">12</xref>]. Moreover, <italic>An. gambiae s.l </italic>and <italic>An. funestus </italic>combine the dual role in their transmission [<xref ref-type="bibr" rid="B13">13</xref>-<xref ref-type="bibr" rid="B16">16</xref>]. Concomitant infections of the two diseases is therefore expected to be a common feature in both humans and mosquito vectors in these areas but little information exists about the occurrence of this phenomenon. In Tanzania, multiple infections of malarial and filarial parasites were documented five decades ago in humans and <italic>An. gambiae </italic>[<xref ref-type="bibr" rid="B17">17</xref>] but little is known about how the two parasites interact during concomitant parasitism. Studies in India have demonstrated that the intensity of <italic>P. falciparum </italic>is generally lower in microfilaraemic individuals than in amicrofilaraemic ones [<xref ref-type="bibr" rid="B6">6</xref>]. Moreover, laboratory studies using vertebrate hosts have revealed that filarial infections may have either benign or suppressive effect on malaria development [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B19">19</xref>]. For instance, microfilaraemic infections in owl monkeys, <italic>Aotus trivirgatus griseimembra </italic>resulted in more benign <italic>P. falciparum </italic>infections than in amicrofilaraemic monkeys [<xref ref-type="bibr" rid="B18">18</xref>]. This indicates that interactions between malaria and filarial parasites may influence the clinical presentation, pathogenicity, and even epidemiology of the diseases they cause [<xref ref-type="bibr" rid="B6">6</xref>], making them important to both the clinicians and epidemiologists.</p><p>In mosquito vectors, interaction between pathogens may affect susceptibility of the vector to infection. Under laboratory conditions, Turrell <italic>et al</italic>. [<xref ref-type="bibr" rid="B20">20</xref>] demonstrated that <italic>Aedes taerniorhynchus </italic>mosquitoes infected with <italic>Brugia pahangi </italic>were more susceptible to Rift Valley Fever (RVF) Virus. This was attributed to the physical disruption of midgut by migrating microfilariae, allowing penetration of virus into the haemocoel. A similar mechanism was suggested to be responsible for the high number of <italic>W. bancrofti </italic>larvae observed in <italic>Plasmodium</italic>-infected <italic>An. punctulatus </italic>[<xref ref-type="bibr" rid="B4">4</xref>]. In contrast however, heavier and/or mixed malaria and filarial infections affect vector survival and flight behaviour [<xref ref-type="bibr" rid="B21">21</xref>-<xref ref-type="bibr" rid="B24">24</xref>] resulting to reduced transmission of both parasites simultaneously [<xref ref-type="bibr" rid="B17">17</xref>]. It is therefore of epidemiological significance to study how malaria and filarial parasites interact in <italic>An. gambiae s.l </italic>and <italic>An. funestus</italic>, the Africa's most important vectors of malaria and LF. The present study was conducted to investigate the occurrence and prevalence of concomitant infections of <italic>P. falciparum </italic>and <italic>W. bancrofti </italic>in mosquito and human populations along the Kenyan Coast. The results of this study provide important information on the need for integrating the control of malaria and lymphatic filariasis.</p></sec><sec sec-type="materials\|methods"><title>Materials and methods</title><sec><title>Study area</title><p>The study was carried out in Shakahola and Jilore villages in Malindi District in Coastal Kenya. The District lies between latitude 2.20 degrees East and 4 degrees South and longitude 3 degrees and 4.14 degrees East. The study area is hot and humid all year round with the annual mean temperatures ranging between 22.5°C and 34°C and the average relative humidity ranging between 60% and 80%. There are two main rainfall seasons in a year. The long rains start from April to June with a peak in May while the short rains fall from October to November. The annual average rainfall ranges from 400 mm in the hinterland to 1200 mm in the Coastal belt. The soils are mainly sandy and infertile supporting small patches of natural forest interpassed with bushes and grasslands.</p><p>Jilore Village is located approximately 30 km west of Malindi town and has a population of 543 people. The village borders the extensive Sokoke forest to the south. The site has a hilly terrain with few plateaus and sandy soils. The population is composed of subsistence farmers, growing mainly maize and cassava and keeping chicken, ducks, and goats as domestic animals. Coconuts and cashew nuts are produced for commercial purposes. Small-scale fishing is also practiced in Lake Jilore. The inhabitants are mainly the Giriama, one of the groups making up the nine-mijikenda people of the East Africa coastal region. Majority of them live in mud-walled houses thatched with coconut leaves. The houses have unscreened widows, holes in the walls, and large open eaves that provide easy entry for mosquitoes. Homesteads are scattered and separated from one another either by agricultural land or small patches of natural vegetation. Each homestead has a number of houses, which are as a result of extended families sharing one compound. Domestic water is collected from the permanent Sabaki River, which together with Lake Jilore provide suitable larval habitats for mosquitoes. The population seeks treatment from Jilore Dispensary. In the year 2002 the prevalence of malaria in this dispensary was 40.5% all due to <italic>P. falciparum </italic>[<xref ref-type="bibr" rid="B25">25</xref>]. Lymphatic filariasis is also common in the village as depicted by the high number of people with overt symptoms of the disease (Mwandawiro, personal communication). However, it is rarely diagnosed in the dispensary. Historically, there has never been a filariasis control programme in the village, but it is currently underway.</p><p>Shakahola village is located approximately 90 km southwest of Malindi town and 45 km from Jilore. The village has a population of 1,009 people. The site is located on a fairly flat terrain with few hilly terrains. The population is composed of subsistence farmers, and essentially they grow the same crops and keep the same domestic animals as those in Jilore. Palm-thatched-mud walled houses, scattered homesteads, and extended families also characterize the village. River Sabaki is the main source of domestic water for the inhabitant. The population is composed of the Kauma and Giriama sub-tribes, which are among the groups making up the nine-mijikenda people of the East Africa coastal region with the Giriama constituting over 75% of the population. The inhabitants seek treatment from the neighboring Chakama Dispensary. Malaria is an important health problem in the area and had a prevalence of 29.5% in the year 2002 all due to <italic>Plasmodium falciparum </italic>[<xref ref-type="bibr" rid="B25">25</xref>]. Lymphatic filariasis is also highly endemic although it is rarely diagnosed. Currently there is an ongoing filariasis control programme in the village and by the beginning of this study, all the inhabitants had already taken the first annual single dose combination of diethycarbamazine and albendazole drugs five months earlier. The prevalence of microfilaraemia in humans before treatment was 17.7% while the filarial infection and infectivity rates were 9.4% and 3.0% in <italic>Anopheles gambiae s.l</italic>, and 3.3% and 1.0% in <italic>An. funestus </italic>(Mwandawiro, unpublished report).</p></sec><sec><title>Registration of inhabitants</title><p>Before fieldwork commenced, permission to conduct the study was obtained from Kenya Medical Research Institute (KEMRI) ethical review committee. Meetings were held in the villages to explain the purpose of the study to the inhabitants. It was made clear that participation in the study was voluntary and that all members of the household aged four years and above were eligible for enrolment in the study. Before commencing the collection of blood, census of all people was done by house-to-house visits, during which personal information was taken for each individual. Individual consent was obtained from each participant or (if < 16 years) from one of their parents or guardian. Individuals found positive for malaria, filarial or both parasites were advised to seek treatment.</p></sec><sec><title>Mosquito collection</title><p>In each village, ten randomly selected houses were sampled for mosquitoes between 0700 and 1000 hours. In Shakahola, mosquitoes were collected in each of the 10 houses once in a month over a three-month period namely; September 2002 and January and February 2003. Due to logistical difficulties sampling was not done between October and December 2002. In Jilore, the collections were done over a 6-month period, sampling each of the ten houses every alternate day for five days in September, three days in October, November and December 2002 and once in January and February 2003. The monthly differences in mosquito sampling times were also as a result of logistical difficulties.</p><p>During sampling, pyrethrum spray sheet collection [<xref ref-type="bibr" rid="B26">26</xref>] was conducted in five of the ten selected houses per village while the remaining five houses were sprayed the following day. All the knocked down female anopheline mosquitoes from different households were picked up from among other insects on the sheets, placed into labeled petri dishes lined with moist cotton wool and transported to the laboratory in a cool box for identification and dissection.</p></sec><sec><title>Mosquito identification and processing</title><p>Mosquitoes were identified morphologically to species using taxonomic keys [<xref ref-type="bibr" rid="B27">27</xref>]. The head, thorax and abdomen of each <italic>An gambiae s.l </italic>and <italic>An. funestus </italic>were dissected separately from each other in a drop of phosphate buffered saline (PBS) on a slide and examined for filarial worms [<xref ref-type="bibr" rid="B28">28</xref>]. The worms were classified as L<sub>1 </sub>(sausage stage), L<sub>2</sub>(motile short) and L<sub>3 </sub>(motile, infective and with caudal papillae) larvae [<xref ref-type="bibr" rid="B29">29</xref>]. The number of larvae was counted to determine the infection load per mosquito.</p><p>The debris of the heads and thoraces of all <italic>Anopheles </italic>mosquitoes examined for filarial worms were removed from the slides and placed singly into labeled plastic vials paying much attention to avoid contamination. Eventually, 50 μl of boiled casein blocking buffer with Nonidet 40 were added into each vial, and the samples ground using sterile pestles. Subsequently, 200 μl of blocking buffer were added bringing the final volume to 250 μl. The samples were stored at -20°C until time of testing. Fifty microlitres aliquots were tested by an enzyme-linked immunosorbent assay (ELISA) using monoclonal antibodies to detect circumsporozoite (CS) proteins of <italic>P. falciparum </italic>[<xref ref-type="bibr" rid="B30">30</xref>]. Samples were assessed visually for positivity [<xref ref-type="bibr" rid="B31">31</xref>].</p></sec><sec><title>Parasitological survey</title><p>Blood samples were taken once, from volunteers aged four years and above and living in the two villages. Briefly, 130 μl of finger-prick blood was collected in a heparinized capillary tube between 2100 and 2400 hrs [<xref ref-type="bibr" rid="B32">32</xref>]. One hundred microliters of this blood was immediately transferred into a plastic vial containing 0.9 ml of 3% acetic acid. In the laboratory, each specimen was transferred to a clean counting chamber and examined under a microscope for enumeration of <italic>Wuchereria bancrofti </italic>microfilariae [<xref ref-type="bibr" rid="B33">33</xref>].</p><p>The remaining 30 μl of blood was used to prepare thick and thin smears for examination of malaria parasites. The smears were stained with 10% Giemsa for 10 minutes, and then examined microscopically under oil immersion for <italic>Plasmodium </italic>species [<xref ref-type="bibr" rid="B32">32</xref>]. A person was considered to be malaria positive if malaria parasites were detected in the thick blood smear and negative if no parasites were found in 200 fields of the thick smear. The thin smears were used for <italic>Plasmodium </italic>species identification. When malaria parasites were demonstrable in the thick blood film, the parasitaemia (parasite density) was determined by counting the number of parasites per 200 leucocytes and total number of parasites obtained multiplied by 40 based on a mean leukocyte count of 8,000 per microliter of blood [<xref ref-type="bibr" rid="B34">34</xref>]. A person was considered to have concomitant infections of malaria and bancroftian filariasis if positive for both malaria and microfilariae parasites.</p></sec><sec><title>Statistical analyses</title><p>Data from precoded forms was checked for accuracy and entered into the computer using FoxPro programme. The contents of the computer files were then checked against the original precoded data sheets for errors or omissions. All statistical analyses were performed using SPSS software (Version 11.5 for widows, SPSS Inc., Chicago, IL). The differences in sporozoite rates, filarial infection, and infectivity rates between species and villages as well as the differences in prevalence of microfilaraemia and malaria parasites between the two villages were compared by chi-square or Fisher's exact test (as appropriate). Interaction between malaria and filarial parasites was determined using the student t-test. The geometric mean density of malaria parasites in humans was calculated after logarithm transformation to normalize the distribution and minimize the standard error.</p></sec></sec><sec><title>Results</title><sec><title>Prevalence of <italic>P. falciparum </italic>and <italic>W. bancrofti </italic>in <italic>An. gambiae s.l </italic>and <italic>An. funestus</italic></title><p>Table <xref ref-type="table" rid="T1">1</xref> shows the prevalence of <italic>P. falciparum </italic>and <italic>W. bancrofti </italic>in <italic>An. gambiae s.l </italic>and <italic>An. funestus</italic>. In Jilore, <italic>P. falciparum </italic>sporozoite rates and <italic>W. bancrofti </italic>infection and infectivity rates were 7.7%, 5.9% and 1.1%, respectively in <italic>An. gambiae s.l </italic>and 1.7%, 6.9% and 1.7% in <italic>An. funestus</italic>. These differences in sporozoite rates and filarial infection and infectivity rates between the two species were not significant (Fisher's exact test, p = 0.123, p = 0.775 and p = 0.484). The corresponding sporozoite rates, filarial infection and infectivity rates for <italic>An. gambiae s.l</italic>. in Shakahola were 5.9%, 13.0% and 0.5%, respectively. None of the two <italic>An. funestus </italic>caught was found positive for any of the two parasites. Filarial infection rates in <italic>An. gambiae s.l </italic>were significantly higher in Shakahola than in Jilore (Fisher's exact test, p = 0.01).</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p><italic>Plasmodium falciparum </italic>sporozoites rates and filarial infection rates in <italic>An. gambiae s.l </italic>and <italic>An. funestus </italic>in Jilore and Shakahola villages</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left">villages</td><td align="center">Mosquito species</td><td align="center">Number examined</td><td align="center">Sporozoite rates</td><td align="center">Filarial infection rates (L<sub>1</sub>-L<sub>3</sub>)</td><td align="center">Infectivity rate (L<sub>3</sub>)</td></tr></thead><tbody><tr><td align="left">Jilore</td><td align="center"><italic>An. gambiae s.l</italic></td><td align="center">1734</td><td align="center">7.7</td><td align="center">5.9</td><td align="center">1.1</td></tr><tr><td></td><td align="center"><italic>An. funestus</italic></td><td align="center">58</td><td align="center">1.7</td><td align="center">6.9</td><td align="center">1.7</td></tr><tr><td align="left">Shakahola</td><td align="center"><italic>An. gambiae s.l</italic></td><td align="center">185</td><td align="center">5.9</td><td align="center">13.0</td><td align="center">0.5</td></tr><tr><td></td><td align="center"><italic>An. funestus</italic></td><td align="center">2</td><td align="center">0.0</td><td align="center">0.0</td><td align="center">0.0</td></tr><tr><td align="left">Total</td><td align="center"><italic>An. gambiae s.l</italic></td><td align="center">1919</td><td align="center">7.6</td><td align="center">6.6</td><td align="center">1.0</td></tr><tr><td></td><td align="center"><italic>An. funestus</italic></td><td align="center">60</td><td align="center">1.7</td><td align="center">6.7</td><td align="center">1.7</td></tr></tbody></table></table-wrap></sec><sec><title>Concomitant infections of <italic>P. falciparum </italic>and <italic>W. bancrofti </italic>in <italic>An. gambiae s.l</italic></title><p>Concomitant infections of <italic>P. falciparum </italic>sporozoites and <italic>W. bancrofti </italic>larvae were recorded in 1.1% (n = 1,734) and 1.6% (n = 185) of <italic>An. gambiae s.l </italic>from Jilore and Shakahola, respectively (Table <xref ref-type="table" rid="T2">2</xref>). Only 10.5% (n = 19) of mosquitoes infected with both parasites were observed harbouring the infective larvae (L<sub>3</sub>) of <italic>W. bancrofti </italic>together with <italic>P. falciparum </italic>sporozoites compared to 89.5% of mosquitoes that harboured <italic>P. falciparum </italic>sporozoites together with immature stages (L<sub>1 </sub>and L<sub>2</sub>) of <italic>W. bancrofti</italic>. None of the 60 <italic>An. funestus </italic>harboured both parasites simultaneously.</p><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Mixed infections of <italic>P. falciparum sporozoites </italic>and <italic>W. bancrofti </italic>larvae in <italic>An. gambiae s.l</italic>. in Jilore and Shakahola villages</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="center" colspan="2">Number of mosquitoes infected (%)</td></tr></thead><tbody><tr><td align="left">Mixed infections</td><td align="center">Jilore (n= 1734)</td><td align="center">Shakahola (n= 185)</td></tr><tr><td colspan="3"><hr></hr></td></tr><tr><td align="left">L1 and Sporozoites</td><td align="center">11 (0.6)</td><td align="center">2 (1.1)</td></tr><tr><td align="left">L2 and sporozoites</td><td align="center">3 (0.17)</td><td align="center">1 (0.54)</td></tr><tr><td align="left">L3 and sporozoites</td><td align="center">1 (0.06)</td><td align="center">0 (0.0)</td></tr><tr><td align="left">L1, L2 and sporozoites</td><td align="center">3 (0.17)</td><td align="center">0 (0.0)</td></tr><tr><td align="left">L1, L3 and sporozoites</td><td align="center">1 (0.07)</td><td align="center">0 (0.0)</td></tr><tr><td align="left">L2, L3 and sporozoites</td><td align="center">0 (0.0)</td><td align="center">0 (0.0)</td></tr><tr><td align="left">L1, L2, L3 and sporozoites</td><td align="center">0 (0.0)</td><td align="center">0 (0.0)</td></tr><tr><td colspan="3"><hr></hr></td></tr><tr><td align="left">Total</td><td align="center">19 (1.1)</td><td align="center">3 (1.62)</td></tr></tbody></table></table-wrap><p>The <italic>P. falciparum </italic>sporozoite rate in mosquitoes with and without filarial infections is shown in Table <xref ref-type="table" rid="T3">3</xref>. <italic>An. gambiae s.l </italic>infected with filarial worms in both villages had higher sporozoites rates than those without filarial infection, although significant difference was only observed in Jilore (χ<sup>2</sup>, p < 0.001).</p><table-wrap position="float" id="T3"><label>Table 3</label><caption><p><italic>Plasmodium falciparum </italic>infections in mosquitoes with and without filarial worms (all stages) in Jilore and Shakahola villages</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left">Site</td><td align="center">Species</td><td align="center">Filarial infection</td><td align="center">No. of mosquitoes</td><td align="center">No. with sporozoite (%)</td><td align="center">P value</td></tr></thead><tbody><tr><td align="left">Jilore</td><td align="center">An. gambiae s.l</td><td align="center">Non-infected</td><td align="center">1631</td><td align="center">115 (7.1)</td><td></td></tr><tr><td></td><td></td><td align="center">Infected</td><td align="center">103</td><td align="center">19 (18.4)</td><td></td></tr><tr><td></td><td align="center">Overall</td><td></td><td align="center">1734</td><td align="center">134 (7.7)</td><td align="center">χ<sup>2</sup>, p < 0.001</td></tr><tr><td></td><td colspan="5"><hr></hr></td></tr><tr><td></td><td align="center">An. funestus</td><td align="center">Non-infected</td><td align="center">54</td><td align="center">1 (1.9)</td><td></td></tr><tr><td></td><td></td><td align="center">Infected</td><td align="center">4</td><td align="center">0 (0.0)</td><td></td></tr><tr><td></td><td align="center">Overall</td><td></td><td align="center">58</td><td align="center">1 (1.7)</td><td align="center">FET, p = 1.000</td></tr><tr><td colspan="6"><hr></hr></td></tr><tr><td align="left">Shakahola</td><td align="center">An. gambiae s.l</td><td align="center">Non-infected</td><td align="center">161</td><td align="center">8 (5.0)</td><td></td></tr><tr><td></td><td></td><td align="center">Infected</td><td align="center">24</td><td align="center">3 (12.5)</td><td></td></tr><tr><td></td><td align="center">Overall</td><td></td><td align="center">185</td><td align="center">11 (5.9)</td><td align="center">FET, p = 0.157</td></tr><tr><td></td><td colspan="5"><hr></hr></td></tr><tr><td></td><td align="center">An. funestus</td><td align="center">Non-infected</td><td align="center">2</td><td align="center">0 (0.0)</td><td></td></tr><tr><td></td><td></td><td align="center">Infected</td><td align="center">0</td><td align="center">0 (0.0)</td><td></td></tr><tr><td></td><td align="center">Overall</td><td></td><td align="center">2</td><td align="center">0 (0.0)</td><td></td></tr></tbody></table><table-wrap-foot><p>FET; Fisher's exact test</p></table-wrap-foot></table-wrap></sec><sec><title>Prevalence of malaria parasites</title><p>Table <xref ref-type="table" rid="T4">4</xref> shows the prevalence of malaria parasites in humans. A total of 208 individuals were examined for malaria parasites of which 47.6% (n = 99) were from Jilore and 52.4% (n = 109) from Shakahola. The prevalence of <italic>P. falciparum </italic>was significantly higher in Jilore (36.4%) than in Shakahola (17.4%) (χ<sup>2</sup>, p = 0.002), whereas <italic>P. malariae </italic>was only found in Jilore (1.0%) but not in Shakahola. The geometric mean density of malaria parasites was 726.9 parasites/μl of blood in Shakahola and did not differ significantly from 505.0 parasites/μl of blood in Jilore (t, p = 0.227). The prevalence of <italic>P. falciparum </italic>gametocytes was 4.0% in Jilore whereas in Shakahola none of the persons examined had gametocytes. The geometric mean density of gametocytes in Jilore was 115.7 gametocytes/μl.</p><table-wrap position="float" id="T4"><label>Table 4</label><caption><p>Malaria and microfilariae prevalence among individuals in Jilore and Shakahola villages.</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="center" colspan="3"><italic>P. falciparum </italic>prevalence</td><td align="center"><italic>P. malariae </italic>prevalence</td><td align="center">Gametocyte prevalence</td><td align="center">Micofilaraemia</td><td></td></tr></thead><tbody><tr><td align="left">Village</td><td align="center">No. examined</td><td align="center">No. positive (%)</td><td align="center">No. positive (%)</td><td align="center">No. positive (%)</td><td align="center">No. examined</td><td align="center">No. positive (%)</td></tr><tr><td colspan="7"><hr></hr></td></tr><tr><td align="left">Jilore</td><td align="center">99</td><td align="center">36 (36.4)</td><td align="center">1 (1.0)</td><td align="center">4 (4.0)</td><td align="center">94</td><td align="center">15 (16.0)</td></tr><tr><td align="left">Shakahola</td><td align="center">109</td><td align="center">19 (17.4)</td><td align="center">0 (0)</td><td align="center">0 (0)</td><td align="center">107</td><td align="center">3 (2.8)</td></tr><tr><td colspan="7"><hr></hr></td></tr><tr><td align="left">Overall</td><td align="center">208</td><td align="center">55 (26.4)</td><td align="center">1 (0.5)</td><td align="center">4 (1.9)</td><td align="center">201</td><td align="center">18 (9.0)</td></tr></tbody></table></table-wrap></sec><sec><title>Prevalence of microfilaraemia</title><p>The prevalence of microfilaraemia in Jilore and Shakahola villages is shown in Table <xref ref-type="table" rid="T4">4</xref>. Of the 208 individuals examined for malaria parasites, 201 of them were further examined for microfilariae. Overall, Jilore (16.0%) had significantly higher microfilaraemia prevalence compared to Shakahola (2.8%) (χ<sup>2</sup>, p < 0.001).</p><p>Microfilaraemia prevalence in both villages increased with age, peaking in the over 50 year's age group where the overall prevalence was 42.9% and 16.7% in Jilore and Shakahola, respectively (data not shown).</p></sec><sec><title>Prevalence of concomitant infections of malaria and bancroftian filariasis in humans</title><p>Two hundred and one people from both villages were examined for malaria and microfilariae parasites, out of which 2.0% had both <italic>P. falciparum </italic>and microfilariae parasites. None of the persons examined in Shakahola (n = 107) had mixed malaria and filarial infections whereas in Jilore (n = 94), 4.3% of the individuals examined harboured microfilariae and <italic>P. falciparum</italic>, 1.2% of these carrying both microfilariae and <italic>P. falciparum </italic>gametocytes. All the persons harbouring both parasites were from 10–29 years age group. When the difference in the geometric mean density of <italic>P. falciparum </italic>was compared between microfilaraemic and amicrofilaraemic persons, no significant difference was observed (t, p > 0.05).</p></sec></sec><sec><title>Discussion</title><p>Malaria and LF co-exist in human populations along the Kenyan coast and transmitted by common vectors. However, there has been inadequate information on the occurrence and prevalence of concomitant infections of the two diseases in both the vector and human populations in these areas. The World Health Organization is currently implementing a new framework for vector control based on a strategy of integrated vector-management targeting both diseases simultaneously. It is thus deemed essential to obtain local information on the occurrence, distribution and prevalence of co-infections of the two diseases as a first step towards this goal. The main aim of the present study was to collect baseline data on which efforts towards designation and implementation of an integrated control strategy may be based. We therefore considered two sets of villages, one in which filariasis control programme was already in place and another one where filariasis control programme is under way. Due to logistical difficulties, we were unable to replicate each set of village but we felt that the two villages would provide essential ideas on the actual situation on the ground.</p><p>In Jilore where no LF control programme was in place, we observed significantly higher sporozoite rates in <italic>Wuchereria</italic>-infected <italic>An. gambiae s.l </italic>than in non-infected mosquitoes. This indicates that infection with <italic>W. bancrofti </italic>may increase mosquito susceptibility to <italic>P. falciparum</italic>. In Papua New Guinea, <italic>Plasmodium</italic>-infected <italic>An. punctulatus </italic>had higher number of W. <italic>bancrofti </italic>larvae compared to uninfected mosquitoes [<xref ref-type="bibr" rid="B4">4</xref>]. Mosquito gut is known to be a significant barrier to malaria infection [<xref ref-type="bibr" rid="B35">35</xref>], but its physical disruption by migrating microfilariae removes this barrier. This may account for the higher sporozoite rates in <italic>Wuchereria</italic>-infected mosquitoes compared to non-infected mosquitoes. Surprisingly, despite the high sporozoite rates in <italic>Wuchereria</italic>-infected mosquitoes, the results demonstrated that simultaneous transmission of concomitant infections of the two parasites is rare. This is reflected by the absence of concomitant infections in <italic>An. funestus </italic>and the low number of <italic>An. gambiae s.l </italic>that were found harbouring both <italic>P. falciparum </italic>sporozoites and infective larvae of <italic>W. bancrofti</italic>. A similar study along the Kenyan coast reported that only 0.4% of <italic>An. gambiae s.l </italic>harboured the infective stages of malaria and filarial parasites [<xref ref-type="bibr" rid="B36">36</xref>]. In Tanzania, only a single mosquito was found harbouring infective stages of malaria and filarial parasites out of 15 mosquitoes that were found infected with malaria or filaria [<xref ref-type="bibr" rid="B17">17</xref>]. This suggests that majority of mosquitoes that pick up mixed malarial and filarial infections do not live long enough for the two parasites to reach the infective stage. Under laboratory conditions, a greater proportion of <italic>B. pahangi </italic>failed to complete development to the second stage in <italic>Aedes aegypti </italic>mosquitoes harbouring <italic>Plasmodium gallinaceum </italic>[<xref ref-type="bibr" rid="B37">37</xref>]. This confirms previous findings that enhanced susceptibility to multiple infections in mosquitoes is of no advantage to parasite transmission [<xref ref-type="bibr" rid="B23">23</xref>]. The findings of this study have important implications towards control of malaria and bancroftian filariasis. Based on the results from Jilore, one may be misled to conclude that elimination of bancroftian filariasis would result to a reduction in sporozoite rates in mosquitoes. However, although further studies are needed to assess the effect of LF control on malaria transmission where both diseases are co-endemic, the findings of this study did not establish any significant difference in <italic>P. falciparum </italic>sporozoite rates between <italic>Wuchereria</italic>-infected and non-infected mosquitoes in Shakahola where there was an ongoing LF control programme. This indicates that differential elimination of LF in malaria endemic areas may result to an increase in malaria transmission since many mosquitoes that die before <italic>W. bancrofti </italic>larvae become infective [<xref ref-type="bibr" rid="B38">38</xref>] or due to multiple infections [<xref ref-type="bibr" rid="B23">23</xref>] will be spared. This clearly supports the need for integrated control of the two diseases. Currently, the Global Programme for Elimination of Lymphatic Filariasis (GPELF) and the Roll Back Malaria Partnership (RBM) carry out most of vector control. RBM is targeting the <italic>Anopheles </italic>vectors of malaria and its activities also cut filarial transmission, particularly where <italic>Anopheles </italic>are also the main vectors of <italic>W. bancrofti</italic>. Co-ordination of the activities of RBM and GPELF could ensure the reduction of both diseases, thereby offsetting the increase in malaria transmission that could arise from differential elimination of bancroftian filariasis.</p><p>Our results indicate that co-infections of malaria and LF in the population are more likely to occur when the prevalence of both diseases is high. This is clearly demonstrated by the presence of co-infections of the two diseases in Jilore and their absence in Shakahola. In coastal areas of Georgetown, Guyana low incidence of malaria was used to explain the low transmission of concomitant infections of malaria and filariasis [<xref ref-type="bibr" rid="B7">7</xref>]. Reducing the prevalence of both diseases would thus reduce the prevalence of concomitant infections of the two diseases. Such a goal can be achieved through early diagnosis and treatment of malaria and filariasis. Unlike studies in India and South America [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B7">7</xref>], the present study has demonstrated that some individuals harbour both malaria gametocytes and microfilariae parasites. A mosquito feeding on such individuals may therefore pick both parasites simultaneously. On the other hand, this study has only demonstrated the occurrence of <italic>P. falciparum </italic>together with <italic>W. bancrofti </italic>but not with other <italic>Plasmodium </italic>species like <italic>P. vivax </italic>as has been observed in India [<xref ref-type="bibr" rid="B5">5</xref>]. This was expected since <italic>P. falciparum </italic>constituted 99.5% of malaria prevalence while <italic>P. malariae </italic>constituted the remaining 0.5%. Since concurrent transmission of the two parasites by anophelines on the Kenyan Coast is rare [<xref ref-type="bibr" rid="B36">36</xref>], and mass treatment of people with DEC and albendazole interrupts their occurrence in humans, as depicted by their absence in Shakahola, promotion of insecticide-treated bed nets together with mass administration of antimalarial and filaricidal drugs can be utilized in control of malaria, bancroftian filariasis and co-infection with the two diseases.</p><p>Although this study was limited by the small sample size, concomitant infections of malaria and bancroftian filariasis were only observed in persons aged between 10 and 29 years. Chadee <italic>et al </italic>[<xref ref-type="bibr" rid="B7">7</xref>] has also reported the absence of such infections in children below 10 years in South America. The differences in behaviour and occupation of people in different age groups may account for the distribution of concomitant infections. The 10 to 29 years age group is composed of young school going children and energetic working individuals. These children are active and may engage in outdoor activities until late evening exposing themselves to mosquito bites, while the working individuals may be exposed to infective bites while working. The fishing in Lake Jilore, and tapping of palm wine (<italic>Mnazi</italic>) both of which are conducted early in the morning and late in the evening may expose people to mosquitoes. Young children and older adults retire to their homes earlier, but are also at risk since these vectors are known to be highly endophilic and anthropophilic [<xref ref-type="bibr" rid="B39">39</xref>].</p><p>The present study did not reveal any significant difference between the geometric mean density of <italic>P. falciparum </italic>in microfilaraemic and amicrofilaraemic individuals, an indication that filarial infections do not affect <italic>P. falciparum </italic>parasitaemia. This was not expected since helminthes are known to induce a TH2-dominant response, which may alter cell mediated immune function to other microbial agents [<xref ref-type="bibr" rid="B40">40</xref>]. Moreover, studies in India [<xref ref-type="bibr" rid="B6">6</xref>] and in animal models [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B19">19</xref>] have demonstrated that filarial infection result in either benign or suppressive effect on <italic>P. falciparum </italic>infections.</p></sec><sec><title>Conclusion</title><p><italic>Wuchereria-</italic>infected mosquitoes seem to be more susceptible to <italic>P. falciparum </italic>infections, but this does not offer any advantage to the transmission of both parasites due to increased mortality of the affected mosquitoes. Thus differential elimination of bancroftian filariasis could potentially increase mosquitoes' life-span resulting to increase in malaria transmission. Furthermore, concomitant infections of malaria and filariasis are likely to occur when the prevalence of both parasites is high. As such, an integrated control strategy targeting the two diseases in areas where they are co-endemic is recommended. Further studies should be conducted under field and laboratory conditions, to ascertain the hypothesis that <italic>Wuchereria </italic>infection increases mosquito susceptibility to <italic>P. falciparum </italic>and appropriate control strategies designed and implemented.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>Ephantus J. Muturi conducted the survey and drafted the manuscript. Charles M. Mbogo provided scientific guidance in data collection, analysis and manuscript preparation and planning, and implementation of day-to-day field activities. Joseph M. Mwangangi, Zipporah W. Ng'ang'a and Ephantus W. Kabiru offered scientific guidance in data analysis and manuscript preparation. Charles Mwandawiro assisted in data collection, provision of ethical clearance and guided in data collection, analysis and manuscript preparation. John C. Beier provided overall supervision of the study and preparation of manuscript. All authors read and approved the final manuscript.</p></sec>
Quality of care associated with number of cases seen and self-reports of clinical competence for Japanese physicians-in-training in internal medicine	<sec><title>Background</title><p>The extent of clinical exposure needed to ensure quality care has not been well determined during internal medicine training. We aimed to determine the association between clinical exposure (number of cases seen), self- reports of clinical competence, and type of institution (predictor variables) and quality of care (outcome variable) as measured by clinical vignettes.</p></sec><sec sec-type="methods"><title>Methods</title><p>Cross-sectional study using univariate and multivariate linear analyses in 11 teaching hospitals in Japan. Participants were physicians-in-training in internal medicine departments. Main outcome measure was standardized t-scores (quality of care) derived from responses to five clinical vignettes.</p></sec><sec><title>Results</title><p>Of the 375 eligible participants, 263 (70.1%) completed the vignettes. Most were in their first (57.8%) and second year (28.5%) of training; on average, the participants were 1.8 years (range = 1–8) after graduation. Two thirds of the participants (68.8%) worked in university-affiliated teaching hospitals. The median number of cases seen was 210 (range = 10–11400). Greater exposure to cases (p = 0.0005), higher self-reports of clinical competence (p = 0.0095), and type of institution (p < 0.0001) were significantly associated with higher quality of care, using a multivariate linear model and adjusting for the remaining factors. Quality of care rapidly increased for the first 100 to 200 cases seen and tapered thereafter.</p></sec><sec><title>Conclusion</title><p>The amount of clinical exposure and levels of self-reports of clinical competence, not years after graduation, were positively associated with quality of care, adjusting for the remaining factors. The learning curve tapered after about 200 cases.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Hayashino</surname><given-names>Yasuaki</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Fukuhara</surname><given-names>Shunich</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Matsui</surname><given-names>Kunihiko</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Noguchi</surname><given-names>Yoshinori</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Minami</surname><given-names>Taro</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Bertenthal</surname><given-names>Dan</given-names></name><xref ref-type="aff" rid="I5">5</xref><email>[email protected]</email></contrib><contrib id="A7" contrib-type="author"><name><surname>Peabody</surname><given-names>John W</given-names></name><xref ref-type="aff" rid="I5">5</xref><email>[email protected]</email></contrib><contrib id="A8" contrib-type="author"><name><surname>Mutoh</surname><given-names>Yoshitomo</given-names></name><xref ref-type="aff" rid="I6">6</xref><email>[email protected]</email></contrib><contrib id="A9" contrib-type="author"><name><surname>Hirao</surname><given-names>Yoshihiko</given-names></name><xref ref-type="aff" rid="I7">7</xref><email>[email protected]</email></contrib><contrib id="A10" contrib-type="author"><name><surname>Kikawa</surname><given-names>Kazuhiko</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A11" contrib-type="author"><name><surname>Fukumoto</surname><given-names>Yohei</given-names></name><xref ref-type="aff" rid="I8">8</xref><email>[email protected]</email></contrib><contrib id="A12" contrib-type="author"><name><surname>Hayano</surname><given-names>Junichiro</given-names></name><xref ref-type="aff" rid="I9">9</xref><email>[email protected]</email></contrib><contrib id="A13" contrib-type="author"><name><surname>Ino</surname><given-names>Teruo</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A14" contrib-type="author"><name><surname>Sawada</surname><given-names>Umihiko</given-names></name><xref ref-type="aff" rid="I10">10</xref><email>[email protected]</email></contrib><contrib id="A15" contrib-type="author"><name><surname>Seino</surname><given-names>Jin</given-names></name><xref ref-type="aff" rid="I11">11</xref><email>[email protected]</email></contrib><contrib id="A16" contrib-type="author"><name><surname>Higuma</surname><given-names>Norio</given-names></name><xref ref-type="aff" rid="I12">12</xref><email>[email protected]</email></contrib><contrib id="A17" contrib-type="author"><name><surname>Ishimaru</surname><given-names>Hiroyasu</given-names></name><xref ref-type="aff" rid="I13">13</xref><email>[email protected]</email></contrib>	BMC Medical Education	<sec><title>Background</title><p>Healthcare systems throughout the world are searching for better ways of delivering high quality care. Attention to quality of patient care has become an important healthcare issue during the last decade, not only for health authorities, policymakers, and managers, but also for physicians and patients. Improving the quality of healthcare involves a broad range of discrete activities such as rigorous evaluation of conventional treatments, incorporating patients' views in healthcare decisions, and audit and feedback of healthcare practices. Physicians are one of the main healthcare providers and are confronted with increasing pressure to provide and improve care. The skills and knowledge of physicians improve through a combination of didactic and experiential learning that can in turn contribute to improving patient care [<xref ref-type="bibr" rid="B1">1</xref>]. Learning occurs through repeated experience with many clinical cases. The number of clinical cases seen might be an important factor linked to quality of care.</p><p>The number of clinical cases needed to meet optimal levels of proficiency in surgical procedures [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B3">3</xref>], such as colonoscopy [<xref ref-type="bibr" rid="B4">4</xref>], has been evaluated often. However, there is very little literature that evaluates the impact of the amount of clinical exposure on quality of care in internal medicine, that is, the learning curve, especially for residents [<xref ref-type="bibr" rid="B5">5</xref>]. Bugelski suggested in the 1970's that the major increment in learning occurs in the early stages of exposure and that less is learned during later stages [<xref ref-type="bibr" rid="B6">6</xref>]. This theory has been tested especially for various surgical procedures and invasive diagnostic tests [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B7">7</xref>], but there was only one study (Medline search) that examined the learning curve for internal medicine training. Day et al. concluded that the rate of increase in self-reports of clinical competence in specific skills was influenced by the number of post-graduation years [<xref ref-type="bibr" rid="B5">5</xref>]. Residents in the first post-graduation year (PGY1) reported that their skills improved by an average of 196% during that year compared to less than 50% for residents in the third year. The goal of this study was to examine the relationship between self-estimates of clinical exposure (number of cases seen) and self-reports of clinical competence and quality of care using specially designed clinical vignettes.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Sample</title><p>An anonymous survey was administered in departments of internal medicine at 11 teaching hospitals in Japan, a convenience sample from 6 university-affiliated and 5 non-affiliated teaching hospitals (with oversampling of university settings). All physicians-in-training within 10 years of graduation from medical school were eligible for the study (n = 375). To avoid contamination, the survey was administered on the same day (03/14/2003) in all institutions.</p></sec><sec><title>Measurement method</title><p>Quality of care was defined as the delivery of patient care in a manner that leads to better outcomes for individuals and populations [<xref ref-type="bibr" rid="B8">8</xref>]. Clinical vignettes have been used to measure variations in quality of care [<xref ref-type="bibr" rid="B9">9</xref>]. The scores derived from the vignettes reliably reflected actual levels of physician practice, resulting in higher criterion validity compared to scores derived from chart abstractions. Based on disease prevalence in Japan, we began by selecting six clinical vignettes to measure quality of care: four common outpatient chronic conditions (diabetes mellitus, chronic obstructive pulmonary disease, vascular disease, and depression) and two acute emergency room conditions (subarachnoid hemorrhage and gastrointestinal bleeding).</p><p>Two detailed clinical vignettes were developed for each chronic condition, for a total of 8 vignettes. These vignettes were originally developed to measure quality of care in the United States. The vignettes were translated in Japanese and partly revised to match clinical practice in Japan, for example, using equivalent drugs and screening procedures. In addition, we developed two original Japanese vignettes for the two acute conditions. From the 10 vignettes available, each participant received five randomly selected vignettes, one from each condition (4 chronic and 1 acute). The vignettes required open-ended responses to questions that were presented in sections characteristic of a typical patient encounter: presenting complaint, history, physical examination, radiological or laboratory tests, diagnosis, and treatment and management plans. Each section began with the presentation of new information. After answering a given section, participants could not return to previous sections to revise (possibly improve) their answers. Participants were given 85 minutes to complete all 5 vignettes.</p><p>Clinical exposure was measured using participant self-estimates of the number of patients seen in in-patient wards, outpatient clinics, and emergency rooms. Data was also collected on the number of years after graduation, type of institution (university-affiliated teaching hospitals or non-affiliated), self-reports of clinical competence (i.e., problem-solving ability, basic procedural skills [e.g., venipuncture, bone marrow aspiration], and basic medical knowledge), and communication ability (i.e., attitude toward patients and their family and cooperativeness with other medical staff). Self-reports were rated using a five-point ordinal rating scale (i.e., unsatisfactory, satisfactory, good, excellent, or outstanding). The overall model consisted of one quality-of-care outcome variable, portrayed by the vignettes, and four predictor variables, that is, self-estimates of total number of patients seen, type of institution, and self-reports of clinical competence and communication ability. The latter two variables were also summed to create a global self-reported competence variable.</p></sec><sec><title>Scoring</title><p>The responses to the vignettes were scored by the authors. To ensure consistency in scoring, given conditions were scored by the same author. With regard to chronic conditions, we used the scoring criteria developed by the original American authors who based their criteria on national guidelines [<xref ref-type="bibr" rid="B9">9</xref>]. These criteria were then reviewed and ratified by expert panels of academic and community physicians in Japan, in fields relevant to each condition; in the end the original criteria were adopted. Scoring criteria for the acute conditions were developed de novo, using expert panels of Japanese physicians. To verify the equivalence of the Japanese version with the original English version, the 10 vignettes were back-translated into English and verified by the original American authors. Based on their recommendations and consensus among the authors, the vignettes and scoring criteria were finalized. Each vignette contained an average of 37 criteria (range = 26–50). Each criterion was rated according to a three-level quality-of-care scale: adequate, unnecessary, and inappropriate care. A one-point credit was assigned for each criterion when adequate care was proposed. An overall vignette score was assigned by summing the scores from the individual criteria.</p><p>First, we used the general linear model to test for a vignette (disease condition) effect; there was no such effect (p = 0.239). Thus the scores from all five vignettes were added for each participant and then converted to a standardized t-score with a mean of 50 and a standard deviation of 10; t-scores were used as the outcome (criterion) variable for quality of care in the analyses. This transformation facilitated the interpretation of the relative importance of the predictor variables by comparing the corresponding t-scores to means of 50.</p></sec><sec><title>Predictor variables</title><p>The amount of clinical exposure was computed by adding the number of cases that each subject had seen in each setting (inpatient, outpatient and ER) and then scored according to five ordinal categories: 0–100, 201–300, 301–400, 401–600, 601–800, >800 cases. The data distribution was skewed to the left and consequently ordinal categories were used because they fit the model better than log transformations (based on Akaike's information criteria – AIC [<xref ref-type="bibr" rid="B10">10</xref>]). We also used a broader range (200 vs. 100) for numbers above 401 because the data were skewed and sparse in those categories. The proportion of cases in the in-patient setting was also calculated and incorporated into the analyses because training occurs mostly in in-patient settings in Japan. The number of years after graduation could influence the amount of clinical exposure and was thus incorporated into the analysis using three ordinal categories (because again they fit the model better than log transformations): one year after graduation (PGY1); two years after graduation (PGY2); and more than 3 years after graduation (≥PGY3).</p><p>In addition to examining the relationship between overall clinical exposure and t-scores, we also looked specifically at the exposure to disorders similar to the ones in the vignettes. This was measured using a common disease index (CDI), defined as (numerator:) the number of cases seen that were similar to the diseases in the vignettes (i.e., stroke, gastrointestinal bleeding, COPD, heart failure, ischemic heart disease, depression, and diabetes mellitus), divided by (denominator:) the total number of cases seen. The t-scores were plotted against CDI to interpret graphically the relationship between the CDI and t-scores. (a measure of quality of care). We also examined the relationship between CDI and quality of care, adjusting for overall clinical exposure in order to verify whether simply increasing the proportion of clinical exposure to similar disorders would lead to higher quality for a fixed amount of exposure.</p><p>Clinical competence was defined as the sum of the self-assessed ratings for the three elements of competence (i.e., problem-solving ability, basic procedural skills, and basic medical knowledge). CDI and clinical competence scores were each further divided into three-level variables: low (up to 33<sup>rd </sup>percentile), middle (up to 67<sup>th </sup>percentile), and high (greater than 67<sup>th </sup>percentile). The maximum score within each level of competence was 15, that is, the sum of 5 points maximum for each element of competence (e.g., problem solving, procedural skills, and knowledge).</p><p>Quality of care can vary depending on the type of institution [<xref ref-type="bibr" rid="B9">9</xref>], and thus it was also incorporated into the analyses. Two types of teaching hospitals were included: university-affiliated and non-university-affiliated (community) teaching hospitals. While this variable was included in the analyses, specific nominal results about this factor are not reported because some hospitals did not consent to revealing type of institution.</p></sec><sec><title>Analyses</title><p>Descriptive statistics included rates and proportions for categorical data and means and standard deviations (SD) for continuous data. We first performed univariate analyses to evaluate the relationship between predictor variables and quality of care. Analysis of covariance or pooled t-test was used for categorical data. Pearson or Spearman correlation coefficients were used for continuous data.</p><p>Multivariate linear regression models were then constructed to examine the association between clinical exposure and quality of care. We incorporated all predictor variables into the model because all of the variables were thought to be important factors that could potentially be associated with levels of quality of care. We tested the interaction between the amount of clinical exposure and type of institution and self-reports of clinical competence as well as for a case (vignette type) main effect. For all analyses, alpha was set at 0.05. Analyses were done using commercially available software (Intercooled STATA 8.0; STATA Corporation, TX, USA). Ethics approval was granted for this study by the Kyoto University Faculty of Medicine Institutional Review Board.</p></sec></sec><sec><title>Results</title><p>Of the 375 eligible physicians-in-training, 263 (70.1%) consented to participate, the majority of whom were first (57.8%) and second-year residents (28.5%). The mean number of years after graduation was 1.8 years (range: 1–8). Two thirds of the participants (68.8%) worked in university-affiliated hospitals. The median number of cases reported seen was 210 (range: 10–11400). The proportion of cases seen in inpatient settings was 48.6%, but the variance was large (SD = 27.4%). A third of the participants (34.7%) had seen more than 400 cases overall. See Table <xref ref-type="table" rid="T1">1</xref> for details.</p><p>Univariate analyses revealed that t-scores were significantly associated with the amount of clinical exposure, type of institution, number of years after graduation, and self-reports of clinical competence, but not with CDI, the proportion of inpatient cases, and self-reports of communication. The mean t-scores between types of institution were significantly different (p < .0001): 55.5 (SD = 2.1) compared to 47.5 (SD = 4.5) (the type of institution is concealed because certain institutions did not consent to revealing their identity). The mean t-scores increased with the number of post-graduation years, amount of exposure, and self-reports of clinical competence. Results of univariate and multivariate analyses are shown in Table <xref ref-type="table" rid="T2">2</xref>. CDI was not associated with t-scores (Figure <xref ref-type="fig" rid="F1">1</xref>).</p><p>In the multivariate model, the number of post-graduation years was not statistically significant (p = 0.6942) while the proportion of inpatient cases seen was a statistically significant predictor (p = 0.0095). There was no significant interaction between the amount of clinical exposure and type of institution or levels of competence; consequently the interaction term was not included in the final model. CDI was not associated with t-scores, even after adjusting for the other factors.</p><p>The slope of t-scores went up sharply for the first 100 to 200 cases and tapered thereafter. The t-scores were higher for one type of institution and for high competence levels, but the slope of the curves was the same among stratified groups, for given levels of amount of clinical exposure. The t-scores and amount of clinical exposure, stratified by type of institution and levels of self-reports of clinical competence, are graphically shown in Figure <xref ref-type="fig" rid="F2">2</xref>.</p></sec><sec><title>Discussion</title><p>This study shows that quality of care for physicians-in-training in internal medicine in Japan increased as physicians saw more cases, especially during the initial stages, and tapered off thereafter. These results are consistent with those of Day et al. for PGY1 and PGY3 residents [<xref ref-type="bibr" rid="B5">5</xref>]. However, the number of years since graduation, which Day et al. suggested was an important predictor, was not significantly associated with quality of care when clinical exposure was included as a variable.</p><p>Although we have found that the overall amount of clinical exposure is an important determinant of the quality of care for physicians-in-training, a related problem is still unsolved: which of the amount of clinical exposure and quality of education is more strongly related to quality of care? To answer this question, we determine the association between the proportion having clinical exposure to certain diseases (CDI) and the quality of care (t-score) for those conditions, instead of evaluating the overall conditions. In the present study, the CDI was not associated with quality of care when adjusting for the overall amount of clinical exposure. Figure <xref ref-type="fig" rid="F1">1</xref> illustrates this result. The CDI for institution B was among the lowest, indicating that physicians in that institution saw fewer similar diseases to the vignettes than physicians in other institutions. However, the average t-score for institution B was the second highest among all institutions. Although there is no accepted indicator for teaching, this institution is well known for its excellent teaching, and is the first educational hospital to have started residency training system in Japan; all authors agree that this hospital provides an excellent education. This suggests that physicians could have high quality of care in specific fields even if they had limited clinical exposure, if the quality of their education was excellent. The same point is made with Institution J, also known for its good teaching, which has a high t-score and high CDI.</p><p>There was a discrepancy between the effect on quality of care of the overall amount of exposure and the case-specific clinical exposure (CDI). The overall amount of clinical exposure was associated with better quality of care, whereas case-specific clinical exposure was not. A possible reason is that the most important skills to be acquired by internists in their training include basic skills and knowledge: history taking, physical examination, interpersonal skills, competence in continuing care, competence in diagnosis, selection of appropriate diagnostic studies, skills in searching evidence, clinical reasoning skills, and problem-solving skills [<xref ref-type="bibr" rid="B11">11</xref>,<xref ref-type="bibr" rid="B12">12</xref>]. These skills and knowledge could be acquired by experience in seeing various diseases, and could be applicable to any type of case seen subsequently, so that the amount of case-specific clinical experience is less important given the same amount of clinical exposure.</p><p>The present results indicate that self-reports of clinical competence are significantly associated with quality of care. However, some studies suggest that self-reports of competence are not in agreement with objective measures of clinical skills [<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B14">14</xref>]. A possible explanation is misclassification of self-reports due to the anonymous (blinded) nature of the study. The participants might have considered that their self-assessments could influence their future career, so that they rated themselves higher than their actual performance. This hypothesis is supported by the observation of Woolliscroft et al. that the bottom quartile of medical students, according to objective evaluations, rate themselves higher than others [<xref ref-type="bibr" rid="B15">15</xref>]. Some other studies [<xref ref-type="bibr" rid="B14">14</xref>-<xref ref-type="bibr" rid="B16">16</xref>] did not state explicitly whether the self-reported ratings were blinded to evaluators. The anonymous self-reports were a strength of the present study, and this might have helped the participants make more accurate self-assessments with no resulting misclassification bias.</p><p>A limitation of the present study is that we did not adjust for medical school provenance or academic performance. Some studies suggest that selection of a medical school may influence practice outcomes [<xref ref-type="bibr" rid="B17">17</xref>,<xref ref-type="bibr" rid="B18">18</xref>]. This type of adjustment is important because of variations in the quality of education among schools and also because high achievers prefer to be educated in an institution with acknowledged excellent teaching. This could confuse the relation between type of institution and quality of care. Instead we adjusted for self-reported levels of competence, because we believe that the effect of current (self-assessed) competence was greater than that of past educational experiences.</p></sec><sec><title>Conclusion</title><p>In summary, the overall amount of clinical exposure (number of cases seen) and levels of self-reports of clinical competence, but not the number of years after graduation, were significantly associated with quality of care, after adjusting for the remaining factors. Quality of education (e.g., the number and quality of the faculty) should be taken into account in future studies. It is possible that a selection bias exists whereby better quality students apply to more highly rated institutions.</p></sec><sec><title>Competing interests</title><p>Fukuhara Shunichi was partly supported by a grant-in-aid from the scientific fund of the Ministry of Health, Labor and Welfare in Japan.</p></sec><sec><title>Authors' contributions</title><p>YH developed the Japanese version of clinical vignettes and coordinated the study with SF, KM, YN and TM. YH, SF, KM, YH, TM and JWP conceived the study, and participated in the study design. YH and DB performed statistical analysis and drafted the initial manuscript for journal submission and participated in revisions. JWP developed the original version of clinical vignettes, and contributed to the evaluation of back-translated Japanese version. YM, YH, KK, YF, JH, TI, US, JS, NH, and HI coordinated data collection at each sites. All authors read and approved the final manuscript.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1472-6920/6/33/prepub"/></p></sec>
Caloric restriction in C57BL/6J mice mimics therapeutic fasting in humans	<sec><title>Background</title><p>Caloric restriction (CR) has long been recognized as a dietary therapy that improves health and increases longevity. Little is known about the persistent effects of CR on plasma biomarkers (glucose, ketone bodies, and lipids) following re-feeding in mice. It is also unclear how these biomarker changes in calorically restricted mice relate to those observed previously in calorically restricted humans.</p></sec><sec><title>Results</title><p>Three groups of individually housed adult female C57BL/6J (B6) mice (n = 4/group) were fed a standard rodent chow diet either: (1) unrestricted (UR); (2) restricted for three weeks to reduce body weight by approximately 15–20% (R); or (3) restricted for three weeks and then re-fed unrestricted (<italic>ad libitum</italic>) for an additional three weeks (R-RF). Body weight and food intake were measured throughout the study, while plasma lipids and levels of glucose and ketone bodies (β-hydroxybutyrate) were measured at the termination of the study. Plasma glucose, phosphatidylcholine, cholesterol, and triglycerides were significantly lower in the R mice than in the UR mice. In contrast, plasma fatty acids and β-hydroxybutyrate were significantly higher in the R mice than in the UR mice. CR had no effect on plasma phosphatidylinositol levels. While body weight and plasma lipids of the R-RF mice returned to unrestricted levels upon re-feeding, food intake and glucose levels remained significantly lower than those prior to the initiation of CR.</p></sec><sec><title>Conclusion</title><p>CR establishes a new homeostatic state in B6 mice that persists for at least three weeks following <italic>ad libitum </italic>re-feeding. Moreover, the plasma biomarker changes observed in B6 mice during CR mimic those reported in humans on very low calorie diets or during therapeutic fasting.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Mahoney</surname><given-names>Lisa B</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Denny</surname><given-names>Christine A</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" corresp="yes" contrib-type="author"><name><surname>Seyfried</surname><given-names>Thomas N</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	Lipids in Health and Disease	<sec><title>Background</title><p>Caloric restriction (CR) has long been recognized as a natural therapy that improves health and extends longevity in humans and rodents [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B7">7</xref>]. CR diminishes inflammation and oxidative stress that occurs from aging by decreasing the production of reactive oxygen species [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B9">9</xref>]. In rodents and primates, CR lowers plasma insulin, cholesterol, triglycerides, and insulin-like growth factor (IGF-1) levels, while elevating plasma high-density lipoprotein (HDL) levels [<xref ref-type="bibr" rid="B10">10</xref>-<xref ref-type="bibr" rid="B14">14</xref>]. These changes in plasma metabolites reduce risk for atherosclerosis, diabetes, and obesity [<xref ref-type="bibr" rid="B15">15</xref>]. Additional health benefits of CR likely result from reduced glucose levels and elevated ketone bodies (β-hydroxybutyrate), which reduce oxygen free radicals and increase the ΔG' of ATP hydrolysis [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B16">16</xref>,<xref ref-type="bibr" rid="B17">17</xref>].</p><p>Numerous studies in humans have used fasting as a treatment for obesity, diabetes, and cancer [<xref ref-type="bibr" rid="B18">18</xref>-<xref ref-type="bibr" rid="B21">21</xref>]. Therapeutic fasting differs from starvation in mobilizing fat rather than protein for energy. Very low calorie diets (approximately 300 kilocalories per day) often produce effects that are similar to those seen during therapeutic fasting [<xref ref-type="bibr" rid="B22">22</xref>-<xref ref-type="bibr" rid="B24">24</xref>]. During the initial stages of a full food fast (water only), blood glucose levels are initially maintained by the mobilization of stored glycogen (glyogenolysis). As glycogen stores become depleted, the body gradually transitions to fatty acids and ketone bodies for additional energy. Although gluconeogenesis also increases, this is insufficient alone to provide enough energy, especially for the brain [<xref ref-type="bibr" rid="B21">21</xref>,<xref ref-type="bibr" rid="B25">25</xref>-<xref ref-type="bibr" rid="B28">28</xref>]. Continued fasting decreases total plasma cholesterol, low-density lipoprotein (LDL) levels, and triglycerides, while elevating fatty acids [<xref ref-type="bibr" rid="B20">20</xref>,<xref ref-type="bibr" rid="B29">29</xref>,<xref ref-type="bibr" rid="B30">30</xref>]. Since the brain does not generally metabolize fatty acids for energy [<xref ref-type="bibr" rid="B31">31</xref>], ketone bodies provide the largest source of energy for the brain during prolonged fasting [<xref ref-type="bibr" rid="B32">32</xref>]. Ketone bodies are a more efficient energy source than either glucose or fatty acids because they are more reduced (a greater hydrogen/carbon ratio) than pyruvate and do not uncouple the mitochondrial proton gradient as occurs with fatty acid metabolism [<xref ref-type="bibr" rid="B17">17</xref>].</p><p>Few studies have examined the longer-term effects of CR or fasting on the concentration of plasma metabolites following <italic>ad libitum </italic>re-feeding. Most previous studies examined biomarker changes following brief periods of re-feeding (approximately 4 days) [<xref ref-type="bibr" rid="B33">33</xref>-<xref ref-type="bibr" rid="B35">35</xref>]. In general, re-feeding restored levels of cholesterol, triglycerides, glucose, ketone bodies, fatty acids, and body weight to the levels seen prior to the initiation of CR or fasting [<xref ref-type="bibr" rid="B15">15</xref>,<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B34">34</xref>-<xref ref-type="bibr" rid="B37">37</xref>]. No prior studies, to our knowledge, have determined to what extent CR-induced plasma biomarker changes persist in mice following <italic>ad libitum </italic>re-feeding for several weeks. It is also unclear how plasma biomarker changes in mice under CR relate to those observed in humans under food restricted diets.</p><p>In this study, we found that three weeks of moderate CR in adult female C57BL/6J (B6) mice significantly reduced plasma glucose, cholesterol, triglycerides and body weight, while elevating fatty acids and ketone bodies. Although <italic>ad libitum </italic>re-feeding for three weeks restored body weight and most CR-induced biomarker changes, food intake and glucose levels remained lower in the R-RF mice than in the UR mice. These findings suggest that the health benefits of CR persist for at least three weeks in B6 mice thus producing a physiological state more energy efficient than that prior to CR. Moreover, the plasma biomarker changes found in B6 mice during three weeks of CR mimic those reported in humans during a very low calorie diet or therapeutic fasting.</p></sec><sec><title>Results</title><p>Compared to the UR mice, the R mice were healthier and more active as assessed by ambulatory and grooming behavior. There were no signs of vitamin or mineral deficiency in the R mice according to standard criteria [<xref ref-type="bibr" rid="B38">38</xref>]. These findings are consistent with the well-established health benefits of mild to moderate CR in rodents and why it is unnecessary to supplement with vitamins and minerals during short-term (up to 12 weeks) CR studies [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B39">39</xref>-<xref ref-type="bibr" rid="B41">41</xref>].</p><sec><title>Influence of caloric restriction and re-feeding on food intake and body weight</title><p>Adult virgin female mice were used for this study because their food intake and body weights are relatively stable from about 120 to 170 days of age (Figs. <xref ref-type="fig" rid="F1">1</xref>, <xref ref-type="fig" rid="F2">2A</xref> and <xref ref-type="fig" rid="F2">2B</xref>). The average total food intake for the UR group during weeks 2–4 was 86.9 ± 2.2 g (n = 4), and over the next three weeks was 87.5 g (n = 2). The amount of food provided for the R mice was initially 60% (40% restriction) of that eaten prior to the initiation of CR (pretrial period). The amount of food given to the R mice was then adjusted each day (± 5%) to achieve a final body weight reduction of approximately 15%. The average total food intake for all restricted mice (n = 8) during the three week restriction period was 52.2 ± 1.5 g. This represents an overall average total food restriction of approximately 40% over the three week period. Body weight was chosen as an endpoint for CR rather than food intake because body weight is a more stable variable than food intake, which differs significantly among mice, even within the same strain. [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B16">16</xref>]. A 15% body weight reduction achieved by a 40% restriction in food represents a moderate caloric restriction for adult mice [<xref ref-type="bibr" rid="B4">4</xref>]. Body weight was reduced in the R group at day 15 and remained significantly lower than that of the UR group (p < 0.01) until day 30. On the day of re-feeding, the R-RF mice binge ate and consumed approximately twice as much food (8.5 g/day/mouse) as they did during the pretrial period (about 4.2 g/day/mouse). Food intake in the R-RF mice decreased rapidly, but remained greater than that of the UR mice until day 33 (Fig. <xref ref-type="fig" rid="F2">2A</xref>). The average total food intake for the R-RF group (n = 4) for the three week re-feeding period, including the three day binge period, was 76.9 ± 2.0 g. Interestingly, the food intake following the binge period (from day 38 through the end of the study) was less in the R-RF mice (3.35 g/day/mouse (n = 4)) than in the UR mice (4.17 g/day/mouse (n = 2) (Fig. <xref ref-type="fig" rid="F2">2A</xref>)). The R-RF mice also ate significantly less food per day during this period than they did during the pretrial period (4.08 g/day/mouse (n = 4) (p < 0.05)) as determined by the paired <italic>t</italic>-test. Despite reduced food intake, the body weights of the R-RF mice returned to the levels observed during the pre-trial period and were similar to the body weights of the UR mice (Fig. <xref ref-type="fig" rid="F2">2B</xref>).</p><fig position="float" id="F1"><label>Figure 1</label><caption><p>Flow chart of the study design. Body weight and food intake were measured every other day over the seven day pre-trial period. All mice received food <italic>ad libitum </italic>during the pre-trial period. After the pre-trial period, the mice were divided into three groups (n = 4 mice/group) where the average body weight of each group was similar. The mice in each group were then fed the same diet in different amounts: 1) the standard chow diet unrestricted (UR), 2) the standard chow diet restricted to achieve an approximate 15–20% body weight reduction from the pre-trial weight (R), or 3) the standard chow diet restricted to achieve an approximate 15–20% body weight reduction from the pre-trial weight for a period of three weeks, followed by unrestricted re-feeding for a period of three weeks (R-RF). Each mouse in the R and the R-RF groups served as its own control for body weight reduction as previously described [16]. Based on food intake and body weight during the pre-trial period, food in the R and the R-RF groups was reduced until each mouse achieved the target weight reduction of approximately 15–20%. The study was terminated and plasma was collected for two UR mice and four R mice on day 30, and for the remainder of the mice on day 51.</p></caption><graphic xlink:href="1476-511X-5-13-1"/></fig><fig position="float" id="F2"><label>Figure 2</label><caption><p>Influence of CR and re-feeding on food intake (A) and body weight (B). Values are expressed as means and 4–8 mice were analyzed in each group. The black arrow indicates the initiation of CR on day 8. The white arrow indicates the initiation of <italic>ad libitum </italic>re-feeding on day 30. The * indicates that the food intake average of the days 38 to 50 of re-feeding for the R-RF mice was significantly less than their food intake prior to initiation of CR, as determined by the paired <italic>t</italic>-test.</p></caption><graphic xlink:href="1476-511X-5-13-2"/></fig></sec><sec><title>Influence of CR and re-feeding on plasma glucose and β-hydroxybutyrate levels</title><p>Glucose levels were 41% less in the R mice than in the UR mice (Table <xref ref-type="table" rid="T1">1</xref>). Although the glucose levels increased following re-feeding in the R-RF mice, the levels remained significantly lower than those in the UR mice. Plasma β-hydroxybutyrate levels were 367% greater in the R mice than in the UR mice (Table <xref ref-type="table" rid="T1">1</xref>). Once re-fed, the plasma β-hydroxybutyrate levels for the R-RF mice returned to those measured in the UR mice. These findings are consistent with our previous studies that β-hydroxybutyrate levels are increased under CR and that circulating β-hydroxybutyrate levels are inversely related to circulating glucose levels [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B16">16</xref>].</p></sec><sec><title>Influence of CR and re-feeding on neutral and acidic lipids</title><p>The influence of CR and re-feeding on the qualitative and quantitative distribution of plasma neutral lipids and acidic lipids in B6 mice is shown in Figs. <xref ref-type="fig" rid="F3">3</xref> and <xref ref-type="fig" rid="F4">4</xref>, respectively, and in Table <xref ref-type="table" rid="T1">1</xref>. Triglycerides, cholesterol, and phosphatidylcholine were significantly reduced, while fatty acids were significantly elevated in the R mice when compared to the UR mice. All plasma lipids in the R-RF mice returned to the levels seen in the UR mice. In contrast to phosphatidylcholine, which was reduced in the R mice and returned to normal levels in the R-RF mice, CR and re-feeding had no effect on plasma levels of phosphatidylinositol. Although sphingomyelin and lysophosphatidylcholine were detected in the plasma of all groups (Fig. <xref ref-type="fig" rid="F3">3</xref>), no statistically significant differences were found among the groups for these lipids due to sample variability. It is important to mention that the solvent front (SF) does not include lipids, but contains slight impurities from the organic solvents used in the developing system.</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Influence of Caloric Restriction and Re-feeding on Plasma Metabolites in B6 Mice<sup>a</sup></p></caption><table frame="hsides" rules="groups"><thead><tr><td align="center">Metabolites</td><td align="center">UR</td><td align="center">R</td><td align="center">Difference(%)</td><td align="center">R-RF</td><td align="center">F<sup>d</sup>(2,9)</td></tr></thead><tbody><tr><td align="center">Glucose<sup>b</sup></td><td align="center">15.5 ± 0.86</td><td align="center">9.1 ± 1.83*</td><td align="center">41</td><td align="center">12.5 ± 1.61</td><td align="center">18.3</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center">β-hydroxybutyrate<sup>b</sup></td><td align="center">0.3 ± 0.12</td><td align="center">1.4 ± 0.13*</td><td align="center">367</td><td align="center">0.3 ± 0.05</td><td align="center">141.4</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><italic>Neutral lipids </italic><sup>c</sup></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center">Triglycerides</td><td align="center">2.4 ± 0.63</td><td align="center">1.0 ± 0.16</td><td align="center">-58</td><td align="center">2.5 ± 1.00</td><td align="center">6.1</td></tr><tr><td align="center">Cholesterol</td><td align="center">0.5 ± 0.23</td><td align="center">0.2 ± 0.06</td><td align="center">-60</td><td align="center">0.4 ± 0.14</td><td align="center">3.7</td></tr><tr><td align="center">Phosphatidylcholine</td><td align="center">2.8 ± 0.53</td><td align="center">1.7 ± 0.25</td><td align="center">-39</td><td align="center">2.8 ± 0.30</td><td align="center">10.3</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><italic>Acidic Lipids </italic><sup>c</sup></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center">Fatty acids</td><td align="center">0.3 ± 0.08</td><td align="center">0.8 ± 0.05</td><td align="center">167</td><td align="center">0.3 ± 0.05</td><td align="center">69.7</td></tr><tr><td align="center">Phosphatidylinositol</td><td align="center">0.2 ± 0.03</td><td align="center">0.2 ± 0.01</td><td align="center">0</td><td align="center">0.2 ± 0.03</td><td align="center">1.2</td></tr></tbody></table><table-wrap-foot><p><sup>a </sup>Values represent the mean ± 95% CI. Four independent samples were analyzed per group.</p><p><sup>b </sup>Values are expressed as mM.</p><p><sup>c </sup>Determined from densitometric scanning of HPTLC as shown in Figures 3 and 4. Values are expressed as mg lipid/ml plasma.</p><p><sup>d</sup> /F/ ratio and degree of freedom (df) obtained from one way analysis of variance.</p><p>Astericks indicate that the value is significantly different from that of the control at the p < 0.05 and ** p < 0.01 as determined by ANOVA followed by Fisher's PLSD</p></table-wrap-foot></table-wrap><fig position="float" id="F3"><label>Figure 3</label><caption><p>HPTLC of plasma neutral lipids in B6 mice. The amount of neutral lipids spotted per lane was equivalent to 2.5 μl of plasma. The plate was developed as described in the Methods. CE, cholesterol esters; TG, triglycerides; IS, internal standard; C, cholesterol; Cer, ceramide; CB, cerebrosides (doublet); PE, phosphatidylethanolamine; PC, phosphatidylcholine; SM, sphingomyelin; LPC, lysophosphatidylcholine; O, origin; and SF, solvent front of the first developing solvent system.</p></caption><graphic xlink:href="1476-511X-5-13-3"/></fig><fig position="float" id="F4"><label>Figure 4</label><caption><p>HPTLC of plasma acidic lipids in B6 mice. The amount of acidic lipids spotted per lane was equivalent to 15 μl of plasma. The plate was developed as described in the Methods. FA, fatty acids; IS, internal standard; CL, cardiolipin; PA, phosphatidic acid; Sulf, sulfatides (doublet); PS, phosphatidylserine; PI, phosphatidylinositol; O, origin; and SF, solvent front of the first developing solvent system.</p></caption><graphic xlink:href="1476-511X-5-13-4"/></fig></sec></sec><sec><title>Discussion</title><p>Reliable biomarkers can be useful for gauging the degree and efficacy of CR as a therapy for a variety of diseases to include: aging, neurological and neurodegenerative diseases, and cancer. Our data show that reductions in plasma glucose, cholesterol, phosphatidylcholine, and triglycerides, combined with elevations of ketone bodies (β-hydroxybutyrate), and fatty acids, are robust biomarker changes for CR in the B6 mouse. Similar changes in glucose and ketone bodies have been observed in other mouse strains and rodent models under CR [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B16">16</xref>,<xref ref-type="bibr" rid="B42">42</xref>]. Cholesterol esters, sphingomyelin, and lysophosphatidylcholine are less reliable plasma biomarkers of CR due to variability between individual mice. It is interesting to note that phosphatidylinositol levels were unchanged as a result of CR and re-feeding, suggesting that this lipid might serve as an internal control for assessing the degree of change in other plasma biomarkers of CR.</p><p>Little is known about the persistent effects of CR on plasma biomarkers (glucose, ketone bodies, and lipids) following re-feeding in mice. Previous studies showed that CR-induced biomarker changes return to levels seen prior to CR following brief periods of <italic>ad libitum </italic>re-feeding (approximately 4 days) [<xref ref-type="bibr" rid="B34">34</xref>,<xref ref-type="bibr" rid="B35">35</xref>]. Our results showed that all biomarkers in the R-RF mice returned to the levels seen prior to CR with the exception of food intake and glucose levels. Since blood glucose levels are directly related to food intake, the persistent reduction in blood glucose reflects the reduction in food intake. These findings suggest that the R-RF mice have established a new, more efficient homeostatic state, in which reduced food intake can maintain body weight similar to that seen during the pretrial period.</p><p>Our results are in agreement with those of other investigators [<xref ref-type="bibr" rid="B43">43</xref>-<xref ref-type="bibr" rid="B45">45</xref>] who observed an energy conservation mechanism due to a decrease in thermogenesis, allowing less energy to be lost as heat and more accumulated as protein, fat, and glycogen. This increase in metabolic efficiency could be the result of several factors involved in homeostasis, but most likely is the result of a decrease in total heat production of the thermoregulatory system. In addition to increasing ATP production, while reducing oxygen consumption, ketone body metabolism also reduces production of damaging free radicals [<xref ref-type="bibr" rid="B17">17</xref>,<xref ref-type="bibr" rid="B46">46</xref>]. For these and other reasons, Veech has described ketone bodies as "super fuel" [<xref ref-type="bibr" rid="B17">17</xref>]. We suggest that the health benefits of CR result in part from a bioenergetic mechanism made more efficient through an increase in ketone body metabolism coupled with a decrease in glucose metabolism. Further studies will be needed to identify those physiological changes within the mitochondria that contribute to or underlie the more efficient metabolic state.</p><p>The physiological relationship between CR in mice and humans is unclear. Although rodents and other animals can be maintained on calorie-restricted diets for prolonged periods [<xref ref-type="bibr" rid="B47">47</xref>], this draconian dietary practice is impractical in humans. Since the basal metabolic rate of mice is about seven times that of humans [<xref ref-type="bibr" rid="B48">48</xref>], it is unlikely that similar degrees of CR will have similar physiological effects in man and mouse. Indeed, a review of the literature generally shows that the plasma biomarker changes we observed in B6 mice, which received approximately 60% of the food given to the UR mice on a daily basis, are generally similar to those observed previously in humans during very low calorie diets or during "water only" therapeutic fasting (Table <xref ref-type="table" rid="T2">2</xref>). While prolonged therapeutic fasting (for one to three weeks) can be healthy for some humans [<xref ref-type="bibr" rid="B49">49</xref>], severe food deprivation beyond a few days is unhealthy in rodents due to increased oxidative stress [<xref ref-type="bibr" rid="B50">50</xref>]. Our findings indicate that moderate CR in B6 mice mimics very low calorie diets or therapeutic fasting in humans. Hence, the numerous health benefits documented in mice following CR may be experienced in humans on very low calorie diets or during periodic therapeutic fasting.</p><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Influence of Fasting/Very Low Calorie Diet on Plasma Metabolites in Humans<sup>a</sup></p></caption><table frame="hsides" rules="groups"><thead><tr><td align="center">Metabolites</td><td align="center">Length (days)</td><td align="center">Unrestricted</td><td align="center">Fasted</td><td align="center">Difference (%)</td><td align="center">References</td><td></td></tr></thead><tbody><tr><td align="center">Glucose</td><td align="center">21</td><td align="center">7.06</td><td align="center">4.39</td><td align="center">-38</td><td align="left">Owen et al. 1998</td><td align="left">[19]</td></tr><tr><td></td><td align="center">21–35</td><td align="center">5.11</td><td align="center">3.89</td><td align="center">-24</td><td align="left">Streja et al. 1977</td><td align="left">[58]</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center">β-hydroxybutyrate</td><td align="center">2</td><td align="center">0.03</td><td align="center">1.67</td><td align="center">5.E+03</td><td align="left">Pan et al. 2000</td><td align="left">[59]</td></tr><tr><td></td><td align="center">3</td><td align="center">0.03</td><td align="center">3.15</td><td align="center">1.E+04</td><td></td><td></td></tr><tr><td></td><td align="center">21</td><td align="center">0.19</td><td align="center">4.60</td><td align="center">2.E+03</td><td align="left">Owen et al. 1998</td><td align="left">[19]</td></tr><tr><td></td><td align="center">21–35</td><td align="center">0.11</td><td align="center">4.56</td><td align="center">4.E+03</td><td align="left">Streja et al. 1977</td><td align="left">[58]</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><italic>Neutral lipids</italic></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center">Triglycerides</td><td align="center">7</td><td align="center">3.46</td><td align="center">2.50</td><td align="center">-28</td><td align="left">Balazsi et al. 1983</td><td align="left">[29]</td></tr><tr><td></td><td align="center">14</td><td align="center">3.46</td><td align="center">1.77</td><td align="center">-49</td><td></td><td></td></tr><tr><td></td><td align="center">28</td><td align="center">1.13</td><td align="center">0.95</td><td align="center">-16</td><td align="left">Shoji et al. 1992</td><td align="left">[30]</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center">Cholesterol</td><td align="center">7</td><td align="center">4.90</td><td align="center">6.73</td><td align="center">37</td><td align="left">Savendahl and Underwood 1999</td><td align="left">[20]</td></tr><tr><td></td><td align="center">7</td><td align="center">5.48</td><td align="center">5.16</td><td align="center">-6</td><td align="left">Balazsi et al. 1983</td><td align="left">[29]</td></tr><tr><td></td><td align="center">14</td><td align="center">5.48</td><td align="center">4.36</td><td align="center">-20</td><td></td><td></td></tr><tr><td></td><td align="center">14</td><td align="center">5.14</td><td align="center">4.01</td><td align="center">-22</td><td align="left">Schouten et al. 1981</td><td align="left">[60]</td></tr><tr><td></td><td align="center">15</td><td align="center">5.39</td><td align="center">4.43</td><td align="center">-18</td><td align="left">Cominacini et al. 1991</td><td align="left">[61]</td></tr><tr><td></td><td align="center">28</td><td align="center">5.22</td><td align="center">4.21</td><td align="center">-19</td><td align="left">Shoji et al. 1992</td><td align="left">[30]</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center">LDL Cholesterol</td><td align="center">7</td><td align="center">2.91</td><td align="center">2.96</td><td align="center">2</td><td align="left">Balazsi et al. 1983</td><td align="left">[29]</td></tr><tr><td></td><td align="center">14</td><td align="center">2.91</td><td align="center">2.43</td><td align="center">-16</td><td></td><td></td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center">Phosphatidylcholine</td><td align="center">7</td><td align="center">2.21</td><td align="center">2.39</td><td align="center">8</td><td align="left">Savendahl et al. 1997</td><td align="left">[62]</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><italic>Acidic Lipids</italic></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center">Fatty acids</td><td align="center">21</td><td align="center">0.84</td><td align="center">1.19</td><td align="center">42</td><td align="left">Owen et al. 1998</td><td align="left">[19]</td></tr><tr><td></td><td align="center">21–35</td><td align="center">0.51</td><td align="center">0.85</td><td align="center">67</td><td align="left">Streja et al. 1977</td><td align="left">[58]</td></tr></tbody></table><table-wrap-foot><p><sup>a </sup>Values are expressed as mM.</p></table-wrap-foot></table-wrap></sec><sec><title>Conclusion</title><p>CR establishes a new homeostatic state in B6 mice that persists for at least three weeks following <italic>ad libitum </italic>re-feeding. Moreover, the plasma biomarker changes observed in B6 mice during CR mimic those reported in humans on very low calorie diets or during therapeutic fasting.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Mice</title><p>C57BL/6J (B6) mice were obtained from the Jackson Laboratory (Bar Harbor, ME, USA) and were propagated in the Boston College Animal Care Facility. Adult female mice were used and were housed individually in plastic cages with filter tops containing Sani-Chip bedding (P.J. Murphy Forest Products Corp., Montville, NJ, USA. Cotton nesting pads were provided to all mice for warmth for the duration of the experiment, and room was maintained at 22°C on a 12 h light – 12 h dark cycle. The procedures for animal use were in strict accordance with the NIH Guide for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care Committee.</p></sec><sec><title>Caloric restriction, body weight and food intake measurements</title><p>All mice received PROLAB RMH 3000 chow (LabDiet, Richmond, IN, USA). This contained a balance of mouse nutritional ingredients and delivers 4.4 kcal g<sup>-1 </sup>gross energy, where fat, carbohydrate, protein, and fiber comprised 55, 520, 225, and 45 g kg<sup>-1 </sup>of the diet, respectively. A total of 12 singly caged, adult female B6 mice were used for the study. The mice were matched for age (120 ± 8 days), sex (virgin females), and body weight (22.0 ± 1.0 g). The experimental design for implementation of CR and re-feeding is outlined in Fig. <xref ref-type="fig" rid="F1">1</xref>. Body weight and food intake measurements were taken at approximately the same time of day (11:00 AM – 1:00 PM) for all mice. Body weight was measured every two days for the UR and R mice. The R-RF mice were weighed daily during the binge period, and every two days thereafter. Food intake for the UR mice was determined daily by subtracting the weight of the food pellets remaining in the food hopper after two days from the initial amount given (approximately 80 g) and dividing the difference by two. For mice in the R and R-RF groups, weighed food pellets were dropped directly into each cage for easy access. Water was provided <italic>ad libitum </italic>for all mice.</p></sec><sec><title>Glucose and β-hydroxybutyrate measurements</title><p>Mice were sacrificed with isofluorane (Halocarbon Laboratories, River Edge, NJ, USA) and blood was collected into heparinized tubes from either the retro-orbital sinus or the heart. The blood was centrifuged at 6,000 × <italic>g </italic>for 10 min, the plasma was collected, and aliquots were stored at -80°C until analysis. Plasma glucose concentration was measured spectrophotometrically using the Trinder Assay (Sigma-Aldrich, St. Louis, MO, USA). The ketone body β-hydroxybutyrate was measured enzymatically using the Stanbio β-Hydroxybutyrate LiquiColor<sup>® </sup>assay kit (Stanbio, Boerne, TX, USA).</p></sec><sec><title>Lipid isolation and purification</title><p>Acidic and neutral lipids were isolated and purified from plasma using modifications of previously described procedures [<xref ref-type="bibr" rid="B51">51</xref>-<xref ref-type="bibr" rid="B53">53</xref>]. Briefly, total lipids were extracted by adding chloroform (CHCl<sub>3</sub>) and methanol (CH<sub>3</sub>OH) to an aliquot of plasma to produce a ratio of CHCl<sub>3 </sub>: CH<sub>3</sub>OH : aqueous plasma (30:60:8 by vol). The plasma volume was used to calculate the volume of CHCl<sub>3 </sub>and CH<sub>3</sub>OH needed to achieve the ratio. Solvent A (CHCl<sub>3 </sub>: CH<sub>3</sub>OH : dH<sub>2</sub>0; 30:60:8 by vol) was added to increase the total volume of each sample. The solution was placed on a magnetic stirrer at room temperature overnight and then centrifuged for 20 min at 1200 × <italic>g</italic>. The supernatant was collected and the pellet was washed with solvent A, placed on the stirrer for 30 min, and centrifuged as before. The supernatants were combined.</p><p>The neutral lipids and acidic lipids were purified using DEAE-Sephadex (A-25, Pharmacia Biotech, Upsala, Sweden) column chromatography as previously described [<xref ref-type="bibr" rid="B52">52</xref>,<xref ref-type="bibr" rid="B53">53</xref>]. The total lipid mixture was applied to a DEAE-Sephadex column with a bed volume of 1.2 ml that had been equilibrated prior with solvent A. Neutral lipids were eluted from the column by washing two times with 20 ml of solvent A. Acidic lipids were then eluted from the column with 30 ml of solvent B (CHCl<sub>3 </sub>: CH<sub>3</sub>OH : 0.8 M Na acetate, 30:60:8 by vol). The neutral lipid fraction was dried using rotary evaporation, washed with 1 ml dH<sub>2</sub>0 and 4 ml CHCl<sub>3</sub>:CH<sub>3</sub>OH (2:1 by vol), and centrifuged at 1200 × <italic>g </italic>to partition neutral lipids into the Folch lower phase [<xref ref-type="bibr" rid="B54">54</xref>,<xref ref-type="bibr" rid="B55">55</xref>]. The upper phase was removed and the lower phase was washed once with the Folch pure solvent upper phase [PSUP] (CHCl<sub>3</sub>:CH<sub>3</sub>OH:dH<sub>2</sub>0, 3:48:47 by vol) and centrifuged again at 1200 × g for 15 min. The upper phase was removed and the lower phase was then evaporated under a stream of nitrogen, re-suspended in 5 ml of CHCl<sub>3</sub>:CH<sub>3</sub>OH (2:1 by vol), and stored at 4°C.</p><p>The acidic lipid fraction was evaporated under vacuum and 7 ml of CHCl<sub>3</sub>:CH<sub>3</sub>OH (1:1 by vol) was added. CHCl<sub>3 </sub>(3.5 ml) and dH<sub>2</sub>0 (2.6 ml) were added, and the mixture was inverted, vortexed, and centrifuged to partition acidic lipids into the lower phase. The upper phase was removed and the lower organic phase was washed once with 4.5 ml of the Folch PSUP and centrifuged. The upper phase was removed and the lower phase was evaporated under a stream of nitrogen, re-suspended in 5 ml of CHCl<sub>3</sub>:CH<sub>3</sub>OH (2:1 by vol), and stored at 4°C.</p></sec><sec><title>Qualitative and quantitative analysis of plasma lipids</title><p>Neutral and acidic lipids were analyzed qualitatively by high-performance thin-layer chromatography (HPTLC) following modifications of previously described methods [<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B51">51</xref>,<xref ref-type="bibr" rid="B52">52</xref>,<xref ref-type="bibr" rid="B56">56</xref>]. Lipids were spotted on 10 × 20 Silica gel 60 HPTLC plates (E. Merck, Darmstadt, Germany) using a Camag Linomat III auto-TLC spotter (Camag Scientific Inc., Wilmington, NC, USA). The amount of plasma per lane was equivalent to 15 μl for acidic lipids and 2.5 μl for neutral lipids. To enhance precision, an internal standard (oleoyl alcohol) was added to the neutral and acidic lipid standards and the plasma samples as previously described [<xref ref-type="bibr" rid="B52">52</xref>]. Purified lipid standards were purchased from Matreya, Inc. (Pleasant Gap, PA, USA), Avanti Polar Lipids, Inc. (Alabaster, AL, USA), and Sigma (St. Louis, MO, USA).</p><p>For neutral and acidic lipids, the plate was developed to a height of 4.5 and 6.0 cm, respectively, with chloroform : methanol : acetic acid : formic acid : water (35:15:6:2:1 by vol), and was then developed to the top with hexanes : diisopropyl ether : acetic acid (65:35:2 by vol). Neutral and acidic lipids were visualized by charring with 3% cupric acetate in 8% phosphoric acid solution, followed by heating in an oven at 160–170°C for 7 min as previously described [<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B51">51</xref>,<xref ref-type="bibr" rid="B52">52</xref>,<xref ref-type="bibr" rid="B56">56</xref>].</p><p>The density and percentage distribution of the individual lipid bands was determined by scanning the plate on a Personal Densitometer SI with ImageQuant software (Molecular Dynamics, Sunnyvale, CA, USA) for neutral and acidic lipids. The density values for each neutral and acidic lipid were fit to a standard curve of the respective lipid and used to calculate individual concentrations as described previously [<xref ref-type="bibr" rid="B52">52</xref>]. All plasma lipid concentrations are expressed as milligram of lipid per milliliter of plasma.</p></sec><sec><title>Statistical analysis</title><p>Analysis of variance (ANOVA) followed by a Fisher's protected least significant difference (PLSD) test were used to evaluate the significance of differences between the UR, R, and R-RF groups. A paired <italic>t</italic>-test was used to analyze differences within the R-RF group (Statview, v. 5.0) [<xref ref-type="bibr" rid="B57">57</xref>]. In each figure, <italic>n </italic>designates the number of individual mice analyzed.</p></sec></sec><sec><title>Abbreviations</title><p>Adenosine triphosphate (ATP)</p><p>Caloric restriction (CR)</p><p>High-density lipoprotein (HDL)</p><p>High-performance thin-layer chromatography (HPTLC)</p><p>Low-density lipoprotein (LDL)</p><p>Restricted (R)</p><p>Restricted/Re-fed (R-RF)</p><p>Unrestricted (UR)</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>LBM participated in the design of the study, carried out the study, performed the lipid analysis, performed the statistical analysis, and helped to draft the manuscript. CAD participated in the design of the study, performed statistical analysis, and coordinated and helped to draft the manuscript. TNS designed the study and helped to draft the manuscript. All authors read and approved the final manuscript.</p></sec>
The evolution of methods for the capture of human movement leading to markerless motion capture for biomechanical applications	<p>Over the centuries the evolution of methods for the capture of human movement has been motivated by the need for new information on the characteristics of normal and pathological human movement. This study was motivated in part by the need of new clinical approaches for the treatment and prevention of diseases that are influenced by subtle changes in the patterns movement. These clinical approaches require new methods to measure accurately patterns of locomotion without the risk of artificial stimulus producing unwanted artifacts that could mask the natural patterns of motion. Most common methods for accurate capture of three-dimensional human movement require a laboratory environment and the attachment of markers or fixtures to the body's segments. These laboratory conditions can cause unknown experimental artifacts. Thus, our understanding of normal and pathological human movement would be enhanced by a method that allows the capture of human movement without the constraint of markers or fixtures placed on the body. In this paper, the need for markerless human motion capture methods is discussed and the advancement of markerless approaches is considered in view of accurate capture of three-dimensional human movement for biomechanical applications. The role of choosing appropriate technical equipment and algorithms for accurate markerless motion capture is critical. The implementation of this new methodology offers the promise for simple, time-efficient, and potentially more meaningful assessments of human movement in research and clinical practice. The feasibility of accurately and precisely measuring 3D human body kinematics for the lower limbs using a markerless motion capture system on the basis of visual hulls is demonstrated.</p>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Mündermann</surname><given-names>Lars</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Corazza</surname><given-names>Stefano</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Andriacchi</surname><given-names>Thomas P</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib>	Journal of NeuroEngineering and Rehabilitation	<sec><title>Introduction</title><p>Over the last several centuries our understanding of human locomotion has been a function of the methods to capture human movement that were available at the time. In many cases the expanded need for enhancing our understanding of normal and pathological human movement drove the introduction of new methods to capture human movement.</p><sec><title>Historical examples</title><p>A look at the history of the study of human locomotion provides some interesting examples of contemporary problems driving the development of new methods for the capture and analysis of human movement. For example, the Weber brothers (1836) reported one of the first quantitative studies of the temporal and distance parameters during human locomotion [<xref ref-type="bibr" rid="B1">1</xref>]. Their work established a model for subsequent quantitative studies of human locomotion. The works of two contemporaries, Marey (1873) and Muybridge (1878), were among the first to quantify patterns of human movement using photographic techniques [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B3">3</xref>]. Also during that time period, Wilhelm Braune (an anatomist) and Otto Fisher (a mathematician) reported measurements of body segment movements to calculate joint forces and energy expenditures using Newtonian mechanics [<xref ref-type="bibr" rid="B4">4</xref>]. Interestingly, their work was motivated by military applications related to improving the efficiency of troop movement.</p><p>During the 1950s there was a need for an improved understanding of locomotion for the treatment of World War II veterans. The classic work at the University of California [<xref ref-type="bibr" rid="B5">5</xref>,<xref ref-type="bibr" rid="B6">6</xref>] provided a tremendous resource of knowledge related to the mechanics of human movement. The work at the University of California formed the basis for many of the fundamental techniques currently used for the study of human locomotion. More recently, instrumentation and computer technologies have provided new opportunities for the advancement of the study of human locomotion. The limitations with respect to automated motion capture as well as measurement reduction no longer exist. New methodology has made it feasible to extend the application of kinetic analysis to clinical problems.</p></sec><sec><title>Current state of the art</title><p>As discussed the expanded need for improved knowledge of locomotion drove the invention of new methods of observation. At present, the most common methods for accurate capture of three-dimensional human movement require a laboratory environment and the attachment of markers, fixtures or sensors to the body segments. These laboratory conditions can cause unknown experimental artifacts.</p><p>Currently, one of the primary technical factors limiting the advancement of the study of human movement is the measurement of skeletal movement from markers or sensors placed on the skin. The movement of the markers is typically used to infer the underlying relative movement between two adjacent segments (e.g. knee joint) with the goal of precisely defining the movement of the joint. Skin movement relative to the underlying bone is a primary factor limiting the resolution of detailed joint movement using skin-based systems [<xref ref-type="bibr" rid="B7">7</xref>-<xref ref-type="bibr" rid="B11">11</xref>].</p><p>Skeletal movement can also be measured directly using alternative approaches to a skin-based marker system. These approaches include stereoradiography [<xref ref-type="bibr" rid="B12">12</xref>], bone pins [<xref ref-type="bibr" rid="B9">9</xref>,<xref ref-type="bibr" rid="B13">13</xref>], external fixation devices [<xref ref-type="bibr" rid="B10">10</xref>] or single plane fluoroscopic techniques [<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B15">15</xref>]. While these methods provide direct measurement of skeletal movement, they are invasive or expose the test subject to radiation. More recently, real-time magnetic resonance imaging (MRI) using open-access MRI provide non-invasive and harmless <italic>in vivo </italic>measurement of bones, ligaments, muscle, etc. [<xref ref-type="bibr" rid="B16">16</xref>]. However, all these methods also impede natural patterns of movements and care must be taken when attempting to extrapolate these types of measurements to natural patterns of locomotion. With skin-based marker systems, in most cases, only large motions such as flexion-extension have acceptable error limits. Cappozzo et al. [<xref ref-type="bibr" rid="B17">17</xref>] have examined five subjects with external fixator devices and compared the estimates of bone location and orientation between coordinate systems embedded in the bone and coordinate systems determined from skin-based marker systems for walking, cycling and flexion-extension activities. Comparisons of bone orientation from true bone embedded markers versus clusters of three skin-based markers indicate a worst-case root mean square artifact of 7°.</p><p>The most frequently used method for measuring human movement involves placing markers or fixtures on the skin's surface of the segment being analyzed [<xref ref-type="bibr" rid="B18">18</xref>]. The vast majority of current analysis techniques model the limb segment as a rigid body, then apply various estimation algorithms to obtain an optimal estimate of the rigid body motion. One such rigid body model formulation is given by Spoor and Veldpas [<xref ref-type="bibr" rid="B19">19</xref>]; they have described a rigid body model technique using a minimum mean square error approach that lessens the effect of deformation between any two time steps. This assumption limits the scope of application for this method, since markers placed directly on skin will experience non-rigid body movement. Lu and O'Connor [<xref ref-type="bibr" rid="B20">20</xref>] expanded the rigid body model approach; rather than seeking the optimal rigid body transformation on each segment individually, multiple, constrained rigid body transforms are sought, modeling the hip, knee, and ankle as ball and socket joints. The difficulty with this approach is modeling the joints as ball and sockets where all joint translations are treated as artifact, which is clearly a limitation for knee motion. Lucchetti et al. [<xref ref-type="bibr" rid="B21">21</xref>] presented an entirely different approach, using artifact assessment exercise to determine the correlation between flexion-extension angles and apparent skin marker artifact trajectories. A limitation of this approach is the assumption that the skin motion during the quasi-static artifact assessment movements is the same as during dynamic activities.</p><p>A recently described [<xref ref-type="bibr" rid="B22">22</xref>,<xref ref-type="bibr" rid="B23">23</xref>] point cluster technique (PCT) employs an overabundance of markers (a cluster) placed on each segment to minimize the effects of skin movement artifact. The basic PCT [<xref ref-type="bibr" rid="B24">24</xref>] can be extended to minimize skin movement artifact by optimal weighting of the markers according to their degree of deformation. Another extension of the basic PCT corrects for error induced by segment deformation associated with skin marker movement relative to the underlying bone. This is accomplished by extending the transformation equations to the general deformation case, modeling the deformation by an activity-dependent function, and smoothing the deformation over a specified interval to the functional form. A limitation of this approach is the time-consuming placement of additional markers.</p><p>In addition to skin movement artifact, many of the previously described methods can introduce an artificial stimulus to the neurosensory system while measuring human movement yielding motion patterns that do not reflect natural patterns of movement. For example, even walking on a treadmill can produce changes in the stride length-walking speed relationships [<xref ref-type="bibr" rid="B25">25</xref>]. Insertion of bone pins, the strapping of tight fixtures around limb segments or constraints to normal movement patterns (such as required for fluoroscopic or other radiographic imaging measurements) can introduce artifacts into the observation of human movement due to local anesthesia and/or interference with musculoskeletal structures. In some cases, these artifacts can lead to incorrect interpretations of movement data.</p><p>The potential for measurement-induced artifact is particularly relevant to studies where subtle gait changes are associated with pathology. For example, the success of newer methods for the treatment and prevention of diseases such as osteoarthritis [<xref ref-type="bibr" rid="B26">26</xref>] is influenced by subtle changes in the patterns of locomotion. Thus, the ability to accurately measure patterns of locomotion without the risk of an artificial stimulus producing unwanted artifacts that could mask the natural patterns of motion is an important need for emerging health care applications.</p><p>Ideally, the measurement system/protocol should be neither invasive nor harmful and only minimally encumber the subject. Furthermore, it should allow measuring subjects in their natural environment such as their work place, home, or on sport fields and be capable of measuring natural activities/motion over a sufficiently large field of view. The purpose of this paper is to examine the development of markerless methods for providing accurate representation of three-dimensional joint mechanics and addressing emerging needs for a better understanding of the biomechanics of normal and pathological motion. The terms markerless and marker-free are used interchangeable for motion capture system without markers. In this review we will use the term markerless motion capture.</p></sec><sec><title>Markerless methods for human motion capture</title><p>Motion capture is an important method for studies in biomechanics and has traditionally been used for the diagnosis of the patho-mechanics related to musculoskeletal diseases [<xref ref-type="bibr" rid="B27">27</xref>,<xref ref-type="bibr" rid="B28">28</xref>]. Recently it has also been used in the development and evaluation of rehabilitative treatments and preventive interventions for musculoskeletal diseases [<xref ref-type="bibr" rid="B29">29</xref>]. Although motion analysis has been recognized as clinically useful, the routine clinical use of gait analysis has seen very limited growth. The issue of its clinical value is related to many factors, including the applicability of existing technology to addressing clinical problems and the length of time and costs required for data collection, processing and interpretation [<xref ref-type="bibr" rid="B30">30</xref>]. A next critical advancement in human motion capture is the development of a non-invasive and markerless system. A technique for human body kinematics estimation that does not require markers or fixtures placed on the body would greatly expand the applicability of human motion capture. Eliminating the need for markers would also considerably reduce patient preparatory time and enable simple, time-efficient, and potentially more meaningful assessments of human movement in research and clinical practice. To date, markerless methods are not widely available because the accurate capture of human movement without markers is technically challenging yet recent technical developments in computer vision provide the potential for markerless human motion capture for biomechanical and clinical applications.</p><p>One of the challenges for a markerless system is the acquisition and representation of human movement. Systems are typically divided into two categories, namely active and passive vision systems. Active systems emit light-information in the visible or infrared light spectrum in the form of laser light, light patterns or modulated light pulses, while passive systems rely purely on capturing images. In general, active systems such as laser scanners, structured light systems and time-of-flight sensors provide very accurate 3D measurements, but require a controlled laboratory environment and often are limited to static measurements. For example, a full body laser scan typically takes several seconds to capture the surface of a human body. Therefore, the main focus on the development of vision systems for markerless motion capture currently lies on employing passive systems. Passive systems are advantageous as they only rely on capturing images and thus provide an ideal framework for capturing subjects in their natural environment.</p><p>The development of markerless motion capture systems originated from the fields of computer vision and machine learning, where the analysis of human actions by a computer is gaining increasing interest. Potential applications of human motion capture are the driving force of system development, and the major application areas are: smart surveillance, identification, control, perceptual interface, character animation, virtual reality, view interpolation, and motion analysis [<xref ref-type="bibr" rid="B31">31</xref>,<xref ref-type="bibr" rid="B32">32</xref>]. Over the past two decades, the field of registering human body motion using computer vision has grown substantially, and a great variety of vision-based systems have been proposed for tracking human motion. These systems vary in the number of cameras used (camera configuration), the representation of captured data, types of algorithms, use of various models, and the application to specific body regions and whole body. Employed configurations typically range from using a single camera [<xref ref-type="bibr" rid="B33">33</xref>-<xref ref-type="bibr" rid="B35">35</xref>] to multiple cameras [<xref ref-type="bibr" rid="B36">36</xref>-<xref ref-type="bibr" rid="B40">40</xref>].</p><p>An even greater variety of algorithms has been proposed for estimating human motion including constraint propagation [<xref ref-type="bibr" rid="B41">41</xref>], optical flow [<xref ref-type="bibr" rid="B42">42</xref>,<xref ref-type="bibr" rid="B43">43</xref>], medial axis transformation [<xref ref-type="bibr" rid="B44">44</xref>], stochastic propagation [<xref ref-type="bibr" rid="B45">45</xref>], search space decomposition based on cues [<xref ref-type="bibr" rid="B36">36</xref>], statistical models of background and foreground [<xref ref-type="bibr" rid="B46">46</xref>], silhouette contours [<xref ref-type="bibr" rid="B47">47</xref>], annealed particle filtering [<xref ref-type="bibr" rid="B48">48</xref>], silhouette based techniques [<xref ref-type="bibr" rid="B49">49</xref>,<xref ref-type="bibr" rid="B50">50</xref>], shape-encoded particle propagation [<xref ref-type="bibr" rid="B51">51</xref>], and fuzzy clustering process [<xref ref-type="bibr" rid="B52">52</xref>]. These algorithms typically derive features either directly in the single or multiple 2D image planes [<xref ref-type="bibr" rid="B42">42</xref>,<xref ref-type="bibr" rid="B45">45</xref>] or, in the case of multiple cameras, at times utilize a 3D representation [<xref ref-type="bibr" rid="B36">36</xref>,<xref ref-type="bibr" rid="B50">50</xref>] for estimating human body kinematics, and are often classified into model-based and model-free approaches. The majority of approaches is model-based in which an a priori model with relevant anatomic and kinematic information is tracked or matched to 2D image planes or 3D representations. Different model types have been proposed including stick-figure [<xref ref-type="bibr" rid="B35">35</xref>], cylinders [<xref ref-type="bibr" rid="B33">33</xref>], super-quadrics [<xref ref-type="bibr" rid="B36">36</xref>], and CAD model [<xref ref-type="bibr" rid="B43">43</xref>]. Model-free approaches attempt to capture skeleton features in the absence of an a priori model. These include the representation of motion in form of simple bounding boxes [<xref ref-type="bibr" rid="B53">53</xref>] or stick-figure through medial axis transformation [<xref ref-type="bibr" rid="B44">44</xref>], and the use of Isomaps [<xref ref-type="bibr" rid="B54">54</xref>] and Laplacian Eigenmaps [<xref ref-type="bibr" rid="B55">55</xref>] for transforming a 3D representation into a pose-invariant graph for extracting kinematics.</p><p>Several surveys concerned with computer-vision approaches have been published in recent years, each classifying existing methods into different categories [<xref ref-type="bibr" rid="B31">31</xref>,<xref ref-type="bibr" rid="B32">32</xref>,<xref ref-type="bibr" rid="B56">56</xref>-<xref ref-type="bibr" rid="B58">58</xref>]. For instance, Moeslund et al. [<xref ref-type="bibr" rid="B31">31</xref>] reviewed more than 130 human motion capture papers published between 1980 and 2000 and categorized motion capture approaches by the stages necessary to solve the general problem of motion capture. Wang et. al [<xref ref-type="bibr" rid="B32">32</xref>] provided a similar survey of human motion capture approaches in the field of computer vision ranging mainly from 1997 to 2001 with a greater emphasize on categorizing the framework of human motion analysis in low-level vision, intermediate-level vision, and high-level vision systems.</p><p>While many existing computer vision approaches offer a great potential for markerless motion capture for biomechanical applications, these approaches have not been developed or tested for this applications. To date, qualitative tests and visual inspections are most frequently used for assessing approaches introduced in the field of computer vision and machine learning. Evaluating existing approaches within a framework focused on addressing biomechanical applications is critical. The majority of research on human motion capture in the field of computer vision and machine learning has concentrated on tracking, estimation and recognition of human motion for surveillance purposes. Moreover, much of the work reported in the literature on the above has been developed for the use of a single camera. Single image stream based methods suffer from poor performance for accurate movement analysis due to the severe ill-posed nature of motion recovery. Furthermore, simplistic or generic models of a human body with either fewer joints or reduced number of degrees of freedom are often utilized for enhancing computational performance. For instance, existing methods for gait-based human identification in surveillance applications use mostly 2D appearance models and measurements such as height, extracted from the side view. Generic models typically lack accurate joint information and thus lack accuracy for accurate movement analysis. However, biomechanical and, in particular, clinical applications typically require knowledge of detailed and accurate representation of 3D joint mechanics. Some of the most challenging issues in whole-body movement capture are due to the complexity and variability of the appearance of the human body, the nonlinear and non-rigid nature of human motion, a lack of sufficient image cues about 3D body pose, including self-occlusion as well as the presence of other occluding objects, and exploitation of multiple image streams. Human body self-occlusion is a major cause of ambiguities in body part tracking using a single camera. The self-occlusion problem is addressed when multiple cameras are used, since the appearance of a human body from multiple viewpoints is available.</p><p>Approaches from the field of computer vision have previously been explored for biomechanical applications. These include the use of a model-based simulated annealing approach for improving posture prediction from marker positions [<xref ref-type="bibr" rid="B59">59</xref>] and marker-free systems for the estimation of joint centers [<xref ref-type="bibr" rid="B60">60</xref>], tracking of lower limb segments [<xref ref-type="bibr" rid="B61">61</xref>], analysis of movement disabilities [<xref ref-type="bibr" rid="B47">47</xref>,<xref ref-type="bibr" rid="B52">52</xref>], and estimation of working postures [<xref ref-type="bibr" rid="B62">62</xref>]. In particular, Persson [<xref ref-type="bibr" rid="B61">61</xref>] proposed a marker-free method for tracking the human lower limb segments. Only movement in the sagittal plane was considered. Pinzke and Kopp [<xref ref-type="bibr" rid="B62">62</xref>] tested the usability of different markerless approaches for automatic tracking and assessing identifying and evaluating potentially harmful working postures from video film. Legrand et al. [<xref ref-type="bibr" rid="B47">47</xref>] proposed a system composed of one camera. The human boundary was extracted in each image and a two-dimensional model of the human body, based on tapered super-quadrics, was matched. Marzani et al. [<xref ref-type="bibr" rid="B52">52</xref>] extended this approach to a system consisting of three cameras. A 3D model based on a set of articulated 2D super-quadrics, each of them describing a part of the human body, was positioned by a fuzzy clustering process.</p><p>These studies demonstrate the applicability of techniques in computer vision for automatic human movement analysis, but the approaches were not validated against marker-based data. To date, the detailed analysis of 3D joint kinematics through a markerless system is still lacking. Quantitative measurements of movement and continuous tracking of humans using multiple image streams is crucial for 3D gait studies. A markerless motion capture system based on visual hulls from multiple image streams and the use of detailed subject-specific 3D articulated models with soft joint constraints is demonstrated in the following section. To critically analyze the effectiveness of markerless motion capture in the biomechanical/clinical environment, we quantitatively compared data obtained from this new system with data obtained from marker-based motion capture.</p></sec><sec><title>Markerless human movement analysis through visual hull and articulated ICP</title><p>The overall goal of our research is to develop a markerless system using multiple optical sensors that will efficiently and accurately provide 3D measurements of human movement for application in clinical practice. Our approach employs an articulated iterative closest point (ICP) algorithm with soft joint constraints [<xref ref-type="bibr" rid="B63">63</xref>] for tracking human body segments in visual hull sequences (a standard 3D representation of dynamic sequences from multiple images). The soft joint constraints approach extends previous approaches [<xref ref-type="bibr" rid="B42">42</xref>,<xref ref-type="bibr" rid="B50">50</xref>] for tracking articulated models that enforced hard constraints on the joints of the articulated body. Small movements at the joint are allowed and penalized in least-squares terms. As a result a more anatomically correct matching suitable for biomechanical applications is obtained with an objective function that can be optimized in an efficient and straightforward manner.</p><p>The articulated ICP algorithm is a generalization of the standard ICP algorithm [<xref ref-type="bibr" rid="B64">64</xref>,<xref ref-type="bibr" rid="B65">65</xref>] to articulated models. The objective is to track an articulated model in a sequence of visual hulls. The articulated model <italic>M </italic>is represented as a discrete sampling of points <italic>p</italic><sub><italic>1</italic></sub>, ..., <italic>p</italic><sub><italic>P </italic></sub>on the surface, a set of rigid segments <italic>s</italic><sub><italic>1</italic></sub>, ..., <italic>s</italic><sub><italic>S</italic></sub>, and a set of joints <italic>q</italic><sub><italic>1</italic></sub>, ..., <italic>q</italic><sub><italic>Q </italic></sub>connecting the segments. Each visual hull is represented as a set of points <italic>V </italic>= <italic>v</italic><sub><italic>1</italic></sub>, ..., <italic>v</italic><sub><italic>N</italic></sub>, which describes the appearance of the person at that time. For each frame of the sequence, an alignment <italic>T </italic>is computed, which brings the surfaces of <italic>M </italic>and <italic>V </italic>into correspondence, while respecting the model joints <italic>q</italic>. The alignment <italic>T </italic>consists of a set of rigid transformations <italic>T</italic><sub><italic>j</italic></sub>, one for each rigid part <italic>s</italic><sub><italic>j</italic></sub>. Similar to ICP, this algorithm iterates between two steps. In the first step, each point <italic>p</italic><sub><italic>i </italic></sub>on the model is associated to its nearest neighbor <italic>v</italic><sub><italic>s</italic>(<italic>i</italic>) </sub>among the visual hull points <italic>V</italic>, where <italic>s(i) </italic>defines the mapping from the index of a surface point <italic>p</italic><sub><italic>i </italic></sub>to its rigid part index. In the second step, given a set of corresponding pairs (<italic>p</italic><sub><italic>i</italic></sub>, <italic>v</italic><sub><italic>s</italic>(<italic>i</italic>)</sub>), a set of transformations <italic>T </italic>is computed, which brings them into alignment. The second step is defined by an objective function of the transformation variables given as <italic>F(T) </italic>= <italic>H(T) </italic>+ <italic>G(T)</italic>. The term <italic>H(T) </italic>ensures that corresponding points (found in the first step) are aligned.</p><p><inline-graphic xlink:href="1743-0003-3-6-i1.gif"/></p><p>The transformation <italic>T</italic><sub><italic>j </italic></sub>of each rigid part <italic>s</italic><sub><italic>j </italic></sub>is parameterized by a 3 × 1 translation vector <italic>t</italic><sub><italic>j </italic></sub>and a 3 × 1 twist coordinates vector <italic>r</italic><sub><italic>j </italic></sub>(twists are standard representations of rotation [<xref ref-type="bibr" rid="B66">66</xref>]), and <italic>R(r</italic><sub><italic>s</italic>(<italic>i</italic>)</sub><italic>) </italic>denotes the rotation matrix induced by the twist parameters <italic>r</italic><sub><italic>s</italic>(<italic>i</italic>)</sub>. The term <italic>G(T) </italic>ensures that joints are approximately preserved, where each joint <italic>q</italic><sub><italic>i</italic>,<italic>j </italic></sub>can be viewed as a point belonging to parts <italic>s</italic><sub><italic>i </italic></sub>and <italic>s</italic><sub><italic>j </italic></sub>simultaneously. The transformations <italic>T</italic><sub><italic>i </italic></sub>and <italic>T</italic><sub><italic>j </italic></sub>are forced to predict the joint consistently.</p><p><inline-graphic xlink:href="1743-0003-3-6-i2.gif"/></p><p>Decreasing the value of <italic>w</italic><sub><italic>G </italic></sub>allows greater movement at the joint, which potentially improves the matching of body segments to the visual hull. The center of the predicted joint locations (belonging to adjacent segments) provides an accurate approximation of the functional joint center. As a result, the underlying kinematic model can be refined and a more anatomically correct matching is obtained.</p><p>The algorithm was evaluated in a theoretical and experimental environment [<xref ref-type="bibr" rid="B67">67</xref>,<xref ref-type="bibr" rid="B68">68</xref>]. The accuracy of human body kinematics was evaluated by tracking articulated models in visual hull sequences. Most favorable camera arrangements for a 3 × 1.5 × 2 m viewing volume were used [<xref ref-type="bibr" rid="B69">69</xref>]. This viewing volume is sufficiently large enough to capture an entire gait cycle. The settings w<sub>H </sub>= 1, w<sub>G </sub>= 5000 (Equations 1 and 2) were used to underscore the relative importance of the joints. The theoretical analysis was conducted in a virtual environment using a realistic human 3D model. The virtual environment permitted the evaluation of the quality of visual hulls on extracting kinematics while excluding errors due to camera calibration and fore-/background separation. To simulate a human form walking, 120 poses were created using Poser (Curious Labs, CA) mimicking one gait cycle. The poses of the human form consisted of 3D surfaces and had an average volume of 68.01 ± 0.06 liters. Visual hulls of different quality using 4, 8, 16, 32 and 64 cameras with a resolution of 640 × 480 pixels and an 80-degree horizontal view were constructed of the Poser sequence. In the experimental environment, full body movement was captured using a marker-based and a markerless motion capture system simultaneously. The marker-based system consisted of an eight-Qualisys camera optoelectronic system monitoring 3D marker positions for the hip, knees and ankles at 120 fps. The markerless motion capture system consisted of eight Basler CCD color cameras (656 × 494 pixels; 80-degree horizontal view) synchronously capturing images at 75 fps. Internal and external camera parameters and a common global frame of reference were obtained through offline calibration. Images from all cameras were streamed in their uncompressed form to several computers during acquisition.</p><p>The subject was separated from the background in the image sequence of all cameras using intensity and color thresholding [<xref ref-type="bibr" rid="B70">70</xref>] compared to background images (Figure <xref ref-type="fig" rid="F1">1</xref>). The 3D representation was achieved through visual hull construction from multiple 2D camera views [<xref ref-type="bibr" rid="B71">71</xref>-<xref ref-type="bibr" rid="B73">73</xref>]. Visual hulls were created with voxel edges of λ = 10 mm, which is sufficiently small enough for these camera configurations [<xref ref-type="bibr" rid="B74">74</xref>]. The number of cameras used for visual hull construction greatly affects the accuracy of visual hulls [<xref ref-type="bibr" rid="B69">69</xref>]. The accuracy of visual hulls also depends on the human subject's position and pose within an observed viewing volume [<xref ref-type="bibr" rid="B69">69</xref>]. Simultaneous changes in position and pose result in decreased accuracy of visual hull construction (Figure <xref ref-type="fig" rid="F2">2</xref>). Increasing the number of cameras leads to decreased variations across the viewing volume and a better approximation of the true volume value.</p><fig position="float" id="F1"><label>Figure 1</label><caption><p>(a) Selected background images (top) and separated subject data (bottom). (b) Camera configuration, video sequences with separated subject data, and selected visual hulls.</p></caption><graphic xlink:href="1743-0003-3-6-1"/></fig><fig position="float" id="F2"><label>Figure 2</label><caption><p>(a) Volume values of visual hulls as a function of position and pose in the viewing volume. (b) Average, min and max volume values across the viewing volume as a function of number of cameras. The dotted line indicates the human form's volume.</p></caption><graphic xlink:href="1743-0003-3-6-2"/></fig><p>A subject-specific 3D articulated model was tracked in the 3D representations constructed from the image sequences. An articulated model is typically derived from a morphological description of the human body's anatomy plus a set of information regarding the kinematic chain and joint centers. The morphological information of the human body can be a general approximation (cylinders, super-quadrics, etc.) or an estimation of the actual subject's outer surface. Ideally, an articulated model is subject-specific and created from a direct measurement of the subject's outer surface. The kinematic chain underneath an anatomic model can be manually set or estimated through either functional [<xref ref-type="bibr" rid="B49">49</xref>,<xref ref-type="bibr" rid="B75">75</xref>] or anthropometric methods [<xref ref-type="bibr" rid="B76">76</xref>,<xref ref-type="bibr" rid="B77">77</xref>]. The more complex the kinematic description of the body the more information can be obtained from the 3D representation matched by the model. While in marker-based systems the anatomic reference frame of a segment is acquired from anatomical landmarks tracked consistently through the motion path, in the markerless system the anatomical reference frames are defined by the model joint centers and reference pose. During the tracking process, the reference frames remain rigidly attached to their appropriate model anatomic segment, thus describing the estimated position and orientation in the subject's anatomic segments. In this study, an articulated body was created from a detailed full body laser scan with markers affixed to the subject's joints (Figure <xref ref-type="fig" rid="F3">3</xref>). The articulated body consisted at least of 15 body segments (head, trunk, pelvis, and left and right arm, forearm, hand, thigh, shank and foot) and 14 joints connecting these segments.</p><fig position="float" id="F3"><label>Figure 3</label><caption><p>(a) Laser scan. (b) Body segments. (c) Joint centers.</p></caption><graphic xlink:href="1743-0003-3-6-3"/></fig><p>The subject's pose was roughly matched to the first frame in the motion sequence and subsequently tracked automatically over the gait cycle (Figure <xref ref-type="fig" rid="F4">4</xref>). Joint center locations were extracted for all joints and joint centers of adjacent segments were used to define segment coordinate axes. Joint angles for the lower limbs for the sagittal and frontal planes were calculated as angles between corresponding axes of neighboring segments projected into the corresponding planes. Accuracy of human body kinematics was calculated as the average deviation of the deviation of joint angles derived from visual hulls compared to joint angles derived from the theoretical sequence and marker-based system over the gait cycle, respectively. The joint angles (sagittal and frontal plane) for the knee calculated as angles between corresponding axes of neighboring segments are used as preliminary basis of comparison between the marker-based and markerless systems (Figure <xref ref-type="fig" rid="F5">5</xref>). The accuracy of sagittal and frontal plane knee joint angles calculated from experiments was within the scope of the accuracy estimated from the theoretical calculations (accuracy<sub>experimental</sub>: 2.3 ± 1.0° (sagittal); 1.6 ± 0.9° (frontal); accuracy<sub>theoretical</sub>: 2.1 ± 0.9° (sagittal); 0.4 ± 0.7° (frontal); [<xref ref-type="bibr" rid="B67">67</xref>,<xref ref-type="bibr" rid="B68">68</xref>]). A similar method, with different model matching formulation and limited to hard joint constraints, was recently explored by the authors [<xref ref-type="bibr" rid="B78">78</xref>]. This method utilized simulated annealing and exponential maps to extract subject's kinematics, and resulted in comparable accuracy.</p><fig position="float" id="F4"><label>Figure 4</label><caption><p>Articulated body matched to visual hulls. (a) Human body segments. (b) Kinematic chain.</p></caption><graphic xlink:href="1743-0003-3-6-4"/></fig><fig position="float" id="F5"><label>Figure 5</label><caption><p>Motion graphs for (a) knee flexion and (b) knee abduction angles (gray = marker-based; black = markerless).</p></caption><graphic xlink:href="1743-0003-3-6-5"/></fig><p>This markerless system was recently used to investigate the role of trunk movement in reducing medial compartment load [<xref ref-type="bibr" rid="B79">79</xref>]. Conventional marker-based motion capture methods are not well suited to study whole body movement since they require a large number of markers placed all over the body. Subjects performed walking trials at a self-selected normal speed in their own low top, comfortable walking shoes with a) normal and b) increased medio-lateral trunk motion. On average, subjects increased their medio-lateral trunk sway by 7.9 ± 4.5° (P = 0.002) resulting in an average reduction of the first peak knee adduction moment of 68.1 ± 16.5% (P < 0.001). Subjects with greater increase in medio-lateral trunk sway experienced greater reductions in the first peak knee adduction moment. The magnitude of reductions in the first peak knee adduction moments were in some cases substantially greater than for conventional interventions including high tibial osteotomy or footwear interventions. The trunk movement assessed was similar to the natural gait compensation adopted by patients with knee OA such as Trendelenburg gait supporting previous findings [<xref ref-type="bibr" rid="B80">80</xref>,<xref ref-type="bibr" rid="B81">81</xref>] that the load distribution between the medial and lateral compartments at the knee during walking is critical. These results demonstrate that introducing a markerless motion capture system into clinical practice will provide meaningful assessments.</p></sec></sec><sec><title>Discussion</title><p>The development of markerless motion capture methods is motivated by the need to address contemporary needs to understand normal and pathological human movement without the encumbrance of markers or fixtures placed on the subject, while achieving the quantitative accuracy of marker based systems. Markerless motion capture has been widely used for a range of applications in the surveillance, film and game industries. However, the biomechanical, medical, and sports applications of markerless capture have been limited by the accuracy of current methods for markerless motions capture.</p><p>Previous experience has demonstrated that minor changes in patterns of locomotion can have a profound impact on the outcome of treatment or progression of musculoskeletal pathology. The ability to address emerging clinical questions on problems that influence normal patterns of locomotion requires new methods that would limit the risk of producing artifact due to markers or the constraints of the testing methods. For example, the constraints of the laboratory environment as well as the markers placed on the subjects can mask subtle but important changes to the patterns of locomotion. It has been shown that the mechanics of walking was changed in patients with anterior cruciate ligament deficiency of the knee [<xref ref-type="bibr" rid="B26">26</xref>,<xref ref-type="bibr" rid="B82">82</xref>]; functional loading influenced the outcome of high tibial osteotomy [<xref ref-type="bibr" rid="B83">83</xref>]; functional performance of patients with total knee replacement was influenced by the design of the implant [<xref ref-type="bibr" rid="B84">84</xref>], and the mechanics of walking influenced the disease severity of osteoarthritis of the knee [<xref ref-type="bibr" rid="B26">26</xref>,<xref ref-type="bibr" rid="B29">29</xref>,<xref ref-type="bibr" rid="B80">80</xref>,<xref ref-type="bibr" rid="B85">85</xref>]. It should be noted that each of the clinical examples referenced above were associated with subtle but important changes to the mechanics of walking.</p><p>The work cited above indicates several necessary requirements for the next significant advancement in our understanding of normal and pathological human movement. First, we need to capture the kinematics and kinetics of human movement without the constraints of the laboratory or the encumbrance of placing markers on the limb segments. Second, we need to relate the external features of human movement to the internal anatomical structures (e.g. muscle, bone, cartilage and ligaments) to further our knowledge of musculoskeletal function and pathology.</p><p>The results presented here demonstrate that markerless motion capture has the potential to achieve a level of accuracy that facilitates the study of the biomechanics of normal and pathological human movement. The errors affecting the accuracy of a markerless motion capture system can be classified into errors due to limitations of the technical equipment and errors due to the shape and/or size of the object or body under investigation. For instance, the accuracy of markerless methods based on visual hulls is dependent on the number of cameras. Configurations with fewer than 8 cameras resulted in volume estimations greatly deviating from original values and fluctuating enormously for different poses and positions across the viewing volume. Visual hulls were not able to capture surface depressions such as eye sockets and lacked accuracy in narrow spaces such as the arm pit and groin regions. However, a human form can be approximated accurately with the appropriate number of cameras for the specific viewing volume. Configurations with 8 and more cameras provided good volume estimations and consistent results for different poses and positions across the viewing volume. Thus, one multi-camera system can be used for both capturing human shape and human movement.</p><p>The work presented here systematically points out that choosing appropriate technical equipment and approaches for accurate markerless motion capture is critical. The processing modules used in this study including background separation, visual hull, iterative closest point methods, etc. yielded results that were comparable to a marker-based system for motion at the knee. While additional evaluation of the system is needed, the results demonstrate the feasibility of calculating meaningful joint kinematics from subjects walking without any markers attached to the limb.</p><p>The markerless framework introduced in this work can serve as a basis for developing the broader application of markerless motion capture. Each of the modules can be independently evaluated and modified as newer methods become available, thus making markerless tracking a feasible and practical alternative to marker based systems. Markerless motion capture systems offer the promise of expanding the applicability of human movement capture, minimizing patient preparation time, and reducing experimental errors caused by, for instance, inter-observer variability. In addition, gait patterns can not only be visualized using traces of joint angles but sequences of snapshots (Figure <xref ref-type="fig" rid="F4">4</xref>) can be easily obtained that allow the researcher or clinician to combine the qualitative and quantitative evaluation of a patient's gait pattern. Thus, the implementation of this new technology will allow for simple, time-efficient, and potentially more meaningful assessments of gait in research and clinical practice.</p></sec>
Sexually dimorphic gene expression that overlaps maturation of type II pneumonocytes in fetal mouse lungs	<sec><title>Background</title><p>In human, respiratory distress of the neonates, which occurs in prematurity, is prevalent in male. Late in gestation, maturation of type II pneumonocytes, and consequently the surge of surfactant synthesis are delayed in male fetuses compared with female fetuses. Although the presence of higher levels of androgens in male fetuses is thought to explain this sex difference, the identity of genes involved in lung maturation that are differentially modulated according to fetal sex is unknown. We have studied the sex difference in developing mouse lung by gene profiling during a three-day gestational window preceding and including the emergence of mature PTII cells (the surge of surfactant synthesis in the mouse occurs on GD 17.5).</p></sec><sec sec-type="methods"><title>Methods</title><p>Total RNA was extracted from lungs of male and female fetal mice (gestation days 15.5, 16.5, and 17.5), converted to cRNA, labeled with biotin, and hybridized to oligonucleotide microarrays (Affymetrix MOE430A). Analysis of data was performed using MAS5.0, LFCM and Genesis softwares.</p></sec><sec><title>Results</title><p>Many genes involved in lung maturation were expressed with no sex difference. Of the approximative 14 000 transcripts covered by the arrays, only 83 genes presented a sex difference at one or more time points between GDs 15.5 and 17.5. They include genes involved in hormone metabolism and regulation (i.e. steroidogenesis pathways), apoptosis, signal transduction, transcriptional regulation, and lipid metabolism with four apolipoprotein genes. Genes involved in immune functions and other metabolisms also displayed a sex difference.</p></sec><sec><title>Conclusion</title><p>Among these sexually dimorphic genes, some may be candidates for a role in lung maturation. Indeed, on GD 17.5, the sex difference in surfactant lipids correlates with the sex difference in pulmonary expression of apolipoprotein genes, which are involved in lipid transport. This suggests a role for these genes in the surge of surfactant synthesis. Our results would help to identify novel genes involved in the physiopathology of the respiratory distress of the neonates.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Simard</surname><given-names>Marc</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Provost</surname><given-names>Pierre R</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A3" corresp="yes" contrib-type="author"><name><surname>Tremblay</surname><given-names>Yves</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib>	Reproductive Biology and Endocrinology	<sec><title>Background</title><p>Hyaline membrane disease (respiratory distress of the neonate) occurs primarily in premature infants. A major cause of this disease is surfactant deficiency. Hyaline membrane disease and surfactant synthesis are both affected by fetal sex. Indeed, the surge of surfactant synthesis is normally delayed in the developing male lung when compared with the female, while hyaline membrane disease is prevalent in males [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B5">5</xref>].</p><p>Surfactant synthesis occurs in type II pneumonocytes (PTII) following maturation of these cells, which is promoted by fibroblast-PTII cell communication [<xref ref-type="bibr" rid="B5">5</xref>]. This maturation process is stimulated by glucocorticoids [<xref ref-type="bibr" rid="B6">6</xref>] and involves some cytokines including epidermal growth factor (EGF) [<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B8">8</xref>], neuregulin-1 [<xref ref-type="bibr" rid="B9">9</xref>], and keratinocyte growth factor (KGF) [<xref ref-type="bibr" rid="B10">10</xref>] as positive regulators and transforming growth factor-β1 (TGF-β1) [<xref ref-type="bibr" rid="B11">11</xref>-<xref ref-type="bibr" rid="B14">14</xref>] as negative regulator.</p><p>Androgens have been shown to both delay fetal lung maturation <italic>in vivo </italic>[<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B15">15</xref>] and block the stimulatory effect of corticosteroids on surfactant synthesis <italic>in vitro </italic>[<xref ref-type="bibr" rid="B16">16</xref>]. These effects are mediated through binding of androgens to androgen receptors [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B15">15</xref>,<xref ref-type="bibr" rid="B17">17</xref>], which are present in both male and female lung tissues [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B19">19</xref>]. There is an active androgen metabolism in the developing lung where androgen synthesis [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B20">20</xref>] and inactivation [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B21">21</xref>] occur. In mice, many genes involved in androgen metabolism are regulated specifically on gestation day (GD) 17.5 in parallel with the emergence of mature PTII cells ([<xref ref-type="bibr" rid="B18">18</xref>] and unpublished data).</p><p>The surge of surfactant synthesis occurs on GD 17.5 in the mouse [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B18">18</xref>] with a sex difference in pulmonary surfactant lipid levels [<xref ref-type="bibr" rid="B3">3</xref>]. Male mice carrying the <italic>Tfm </italic>gene (male with testicular feminization), which have no functional androgen receptors, have surfactant levels comparable with those of normal females at a comparable developmental time point [<xref ref-type="bibr" rid="B3">3</xref>]. Therefore, the mouse is a good model to study the effect of fetal sex on the timing of the developmental events related to the surge of surfactant synthesis.</p><p>Knowing that the surge of surfactant synthesis is delayed in male fetal mouse lungs compared with females, we were interested to identify genes that are expressed with a sex difference during the gestational period that overlaps the surge of surfactant production. We found, by microarray analysis, genes exhibiting a sex difference in expression in lung development during a three-day window preceding and including the emergence of mature PTII cells. To date, no gene profiling study of sex differences in the embryonic lung tissue exists. Using Affymetrix technology, we have studied about 14,000 transcripts and variants with more than 22,600 probe sets in male and female fetal lungs on GDs 15.5, 16.5, and 17.5.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Animals</title><p>Protocols were approved by the animal care and use committee and the institutional review board of the Centre de Recherche du Centre Hospitalier Universitaire de Québec (protocol 2002-080). Balb/C mice (<italic>Mus musculus</italic>) were mated during the night. Appearance of the copulatory plug was considered as gestation day 0.5 (GD 0.5). Pregnant females were euthanized by exposure to CO<sub>2</sub>. Fetal sex was determined by examination of the genital tract with a dissecting microscope at 15 × magnification. Fetal lungs were collected and one pool of tissues was prepared for each sex and each pregnant animal prior RNA extraction.</p></sec><sec><title>RNA extraction</title><p>Total RNA was extracted using Tri-reagent, a mixture of phenol and guanidine thiocyanate in a monophasic solution (Molecular Research Center, Cincinnati, OH) as described previously [<xref ref-type="bibr" rid="B22">22</xref>]. Each RNA sample was purified on a CsCl gradient as described [<xref ref-type="bibr" rid="B23">23</xref>], using a TLA 120.2 rotor in an Optima MAX centrifuge (Beckman, Mississauga, ON, Canada).</p></sec><sec><title>Preparation of probes</title><p>Samples were processed following the Small Sample Labeling Protocol version II from Affymetrix [<xref ref-type="bibr" rid="B24">24</xref>]. This protocol is based on the principle of performing two cycles of cDNA synthesis and <italic>in vitro </italic>transcription reactions for target amplification. Briefly, 10 μg of total RNA were converted to cDNA by incubation with 400 units of SuperScript II reverse transcriptase (Invitrogen, Carlsbad, CA), a T7 oligonucleotide-d(T)<sub>24 </sub>as a primer (5'-GGCCAGTGAATTGTAATACGACTCACTATAGGGAGGCGG(T)<sub>24</sub>-3'), combined with 1 mM dNTPs (deoxynucleotide triphosphates) in 1 × first strand buffer (50 mM Tris HCl pH 8.3, 75 mM KCl, 3 mM MgCl<sub>2</sub>, 10 mM DTT) at 42°C for 1 h. Second strand cDNA synthesis was performed using 40 units of DNA polymerase I (Invitrogen), 10 units of <italic>E. coli </italic>DNA ligase (Invitrogen), 2 units of RNase H (Invitrogen), and 0.2 mM dNTPs in 1 × reaction buffer (18.8 mM Tris-HCl pH 8.3, 90.6 mM KCl, 4.6 mM MgCl<sub>2</sub>, 3.8 mM DTT, 0.15 mM NAD, 10 mM (NH<sub>4</sub>)<sub>2</sub>SO<sub>4</sub>) at 16°C for 2 h. Each cDNA sample was blunt ended by addition of 10 units of T4 polynucleotide kinase (Invitrogen) and incubation at 16°C for 10 min. cDNA samples were purified by phenol-chloroform extraction using phase lock gels (Brinkmann Instruments Inc., Mississauga, ON, Canada), ethanol precipitated and resuspended in 10 μl of DEPC- (diethylpyrocarbonate) treated H<sub>2</sub>O. First cycle amplification was performed using a MEGAscript T7 Kit (Ambion, Austin TX). The mixture (10 μl final volume) was incubated at 37°C for 5 h. cRNA was purified using a RNeasy Mini Kit (Qiagen, Valencia, CA) according to the protocol of the manufacturer. Purified cRNA was reverse-transcribed to cDNA for a second time following the protocol used for the first cycle. For the second amplification, a T7 BioArray High Yield RNA Transcript Labeling Kit (Enzo Diagnostics, Farmingdale, NY) was used to produce biotinylated cRNA. The mixture (20 μl final volume) was incubated at 37°C for 5 h with gentle mixing every 30 min. Labelled cRNA was purified using a RNeasy Mini Kit (Qiagen) according to the protocol of the manufacturer. Purified cRNA was fragmented into segments of 20–300 nucleotide length by incubation in a fragmentation buffer (100 mM potassium acetate, 30 mM magnesium acetate, 40 mM Tris-acetate pH 8.1) for 20 min at 94°C. The quality of cRNA amplification and cRNA fragmentation was monitored by micro-capillary electrophoresis (Bioanalyser 2100, Agilent Technologies, Mississauga, ON, Canada).</p></sec><sec><title>Microarray hybridization, scanning, and analysis</title><p>Each preparation of cRNA probe was hybridized to two GeneChip Mouse Genome 430 A arrays (Affymetrix, Santa Clara, CA). Each microarray was pre-hybridized in 1 × hybridization buffer (0.1 mg/ml herring sperm DNA, 0.5 mg/ml acetylated BSA) at 45°C for 10 min under constant rotation (60 rpm). Then, the buffer was replaced by a mixture containing 15 μg of fragmented cRNA in 1 × hybridization buffer, and the following internal controls from Affymetrix: 5 nM control oligonucleotide B2 and 1 × eukaryotic hybridization control (1.5 pM <italic>BioB</italic>, 5 pM <italic>BioC</italic>, 25 pM <italic>BioD </italic>and 100 pM <italic>cre</italic>). Samples were incubated at 45°C for 16 h under constant rotation. Microarrays were processed using an Affymetrix GeneChip Fluidic Station 400 (protocol EukGE-WS2Av4). Staining was initiated with streptavidin-conjugated phycoerythrin (SAPE), followed by amplification using a biotinylated anti-streptavidin antibody and by a second round of SAPE. GeneChips were scanned using a GeneChip Scanner 3000 with autoloader (Affymetrix). Data acquisition and analysis were performed using the Microarray Suite 5.0 software (Affymetrix). Signal intensities for β-actin and GAPDH genes were used as internal quality controls; their ratio of fluorescence intensities for the 5' and 3' ends was <2. Differentially expressed genes were determined using the LFCM software [<xref ref-type="bibr" rid="B25">25</xref>]. Briefly, variable fold change limit (LFC) decreasing with gene expression value was used to select differentially expressed genes. The LFC equation is Y = a + b/X, were X is the minimum gene expression intensity from two arrays and Y is the fold change limit. The parameters a and b were estimated from the distribution of ratios calculated from replicated chips. The resulting cut-off point, Y = 1.8 + 62.0/X, gave an approximately constant rate of false positive modulated genes of 0.1%. All the genes having a fold change above this curve were considered significantly modulated. Data were also analysed with the Genesis 1.6.0 Beta1 software [<xref ref-type="bibr" rid="B26">26</xref>]. The MOE430A microarray provides coverage of over 22,600 probe sets corresponding to about 14,000 transcripts and variants. The probe sets were selected from sequences derived from GenBank, dbEST and RefSeq. The sequence clusters were created from the UniGene database (Build 107, June 2002) and then refined by analysis and comparison with the publicly available draft assembly of the mouse genome from the Whitehead Institute Center for Genome Research (MSCG, April 2002) (for more details see [<xref ref-type="bibr" rid="B24">24</xref>]).</p></sec></sec><sec><title>Results</title><p>Pregnant Balb/C mice were sacrificed on GDs 15.5, 16.5, and 17.5. Total RNA prepared from male and female fetal lungs, from each litter, was pooled in order to obtain the following six samples: GD 15.5 males; GD 15.5 females (each from 2 litters); GD 16.5 males; GD 16.5 females (each from 2 litters); GD 17.5 males; and GD 17.5 females (each from 1 litter) (number of fetuses per RNA preparation were respectively of 6, 9, 7, 9, 2, and 4). Previously, the litters corresponding to GDs 15.5 and 17.5 have shown sex differences in expression for several genes [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B27">27</xref>]. Microarray analysis was performed using six probes prepared from these RNA samples and 12 GeneChip Mouse Genome 430 A array (in duplicate). Values obtained for each gene were compared between sexes for each GD.</p><sec><title>Validation of data</title><p>Internal controls assessing the validity and reproducibility of the data were satisfied (see Methods). For each experiment, Table <xref ref-type="table" rid="T1">1</xref> shows the number of probe sets detected (<italic>p</italic>-value = 0.05 based on raw signals obtained for each probe set). Of these, the number of associated known and unknown transcripts was determined from analysis of probe set IDs using NetAffx (Affymetrix) and Excel (Microsoft) softwares. Expressed sequence tags (ESTs), Riken sequences, and all other unknown transcripts ("hypothetical","similar to"...) were sorted into the category named "Unknowns".</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Overview of the microarray data characteristics</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="center" colspan="2"><bold>GD15.5</bold></td><td align="center" colspan="2"><bold>GD16.5</bold></td><td align="center" colspan="2"><bold>GD17.5</bold></td></tr></thead><tbody><tr><td></td><td align="center"><bold>M</bold></td><td align="center"><bold>F</bold></td><td align="center"><bold>M</bold></td><td align="center"><bold>F</bold></td><td align="center"><bold>M</bold></td><td align="center"><bold>F</bold></td></tr><tr><td colspan="7"><hr></hr></td></tr><tr><td align="left"><bold>Detected probesets (% present)</bold></td><td align="center">12955 (57,3%)</td><td align="center">12988 (57,4%)</td><td align="center">12660 (56%)</td><td align="center">12455 (55%)</td><td align="center">12234 (54,1%)</td><td align="center">12808 (56,6%)</td></tr><tr><td align="left"><bold>Known genes</bold><sup>†</sup></td><td align="center">6490</td><td align="center">6464</td><td align="center">6466</td><td align="center">6424</td><td align="center">6321</td><td align="center">6561</td></tr><tr><td align="left"><bold>Unknown transcripts</bold><sup>†</sup></td><td align="center">2406</td><td align="center">2433</td><td align="center">2030</td><td align="center">2019</td><td align="center">1965</td><td align="center">2060</td></tr></tbody></table><table-wrap-foot><p>- Number of probe sets detected (p-value ≤ 0.05) relative to the total number of probe sets on the array</p><p><sup>†</sup>- When multiple probe sets are associated to the same transcript, only one is considered M, male; F, female</p></table-wrap-foot></table-wrap><p>Sex determination of fetuses was confirmed by analysis of several genes associated with the Y-chromosome. Specifically, Ddx3y (DEAD (Asp-Glu-Ala-Asp) box polypeptide 3, Y-linked) (Id. #56, Table <xref ref-type="table" rid="T2">2</xref>), Jarid1d (jumonji, AT rich interactive domain 1D (Rbp2 like)) (Id. #57), also called Smcy, Eif2s3y (Y chromosome-encoded subunit of the initiation of translation factor Eif2) (Id. #50) and Uty (ubiquitously transcribed tetratricopeptide repeat gene, Y chromosome) (Id. #65) were expressed, as expected, exclusively in male samples for each GD studied (Table <xref ref-type="table" rid="T2">2</xref>).</p><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Differentially expressed genes in the lung according to sex at GDs 15.5, 16.5 and 17.5</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left"><bold>Id</bold></td><td align="left"><bold>Accession number</bold></td><td align="center"><bold>Gene name</bold><sup>†</sup></td><td></td><td align="center"><bold>Fold changes</bold><sup>‡</sup></td><td></td><td></td><td></td></tr><tr><td></td><td></td><td></td><td colspan="5"><hr></hr></td></tr><tr><td></td><td></td><td></td><td align="center"><bold>Symbol</bold></td><td align="center"><bold>GD15.5 (62)</bold><sup>§</sup></td><td align="center"><bold>GD16.5 (33)</bold></td><td align="center"><bold>GD17.5 (38)</bold></td><td align="center"><bold>Chr.</bold></td></tr></thead><tbody><tr><td align="left" colspan="8"><bold>Apoptosis</bold></td></tr><tr><td align="left">#1</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BI662863">BI662863</ext-link></td><td align="center">Rho-associated coiled-coil forming kinase 1</td><td align="center">Rock1</td><td></td><td align="center">2.25</td><td></td><td></td></tr><tr><td align="left">#2</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_011997">NM_011997</ext-link></td><td align="center">caspase 8 associated protein 2</td><td align="center">Casp8ap2</td><td></td><td align="center">2.15</td><td></td><td align="center">4 11.4 cM</td></tr><tr><td align="left" colspan="8"><bold>Cell adhesion</bold></td></tr><tr><td align="left">#3</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_009925">NM_009925</ext-link></td><td align="center">procollagen, type X, alpha 1</td><td align="center">Col10a1</td><td align="center">M</td><td></td><td></td><td align="center">10 22.0 cM</td></tr><tr><td align="left">#4</td><td align="left"><ext-link ext-link-type="gen" xlink:href="L20232">L20232</ext-link></td><td align="center">integrin binding sialoprotein</td><td align="center">Ibsp</td><td align="center">M</td><td></td><td></td><td align="center">5 56.0 cM</td></tr><tr><td align="left">#5</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_031163">NM_031163</ext-link></td><td align="center">procollagen, type II, alpha 1</td><td align="center">Col2a1</td><td align="center">2.96</td><td align="center">F</td><td></td><td align="center">15 54.5 cM</td></tr><tr><td align="left">#6</td><td align="left">NM_007729</td><td align="center">procollagen, type XI, alpha 1</td><td align="center">Col11a1</td><td align="center">2.25</td><td></td><td></td><td align="center">3 53.1 cM</td></tr><tr><td align="left">#7</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AK004383">AK004383</ext-link></td><td align="center">procollagen, type IX, alpha 1</td><td align="center">Col9a1</td><td align="center">1.96</td><td></td><td></td><td align="center">1 15.0 cM</td></tr><tr><td align="left">#8</td><td align="left"><ext-link ext-link-type="gen" xlink:href="U08020">U08020</ext-link></td><td align="center">procollagen, type I, alpha 1</td><td align="center">Col1a1</td><td></td><td></td><td align="center">(1.77)</td><td align="center">11 56.0 cM</td></tr><tr><td align="left" colspan="8"><bold>Cell growth</bold></td></tr><tr><td align="left">#9</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_008341">NM_008341</ext-link></td><td align="center">insulin-like growth factor binding protein 1</td><td align="center">Igfbp1</td><td align="center">M</td><td></td><td></td><td align="center">11 1.3 cM</td></tr><tr><td align="left" colspan="8"><bold>Coagulation</bold></td></tr><tr><td align="left">#10</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_008877">NM_008877</ext-link></td><td align="center">plasminogen</td><td align="center">Plg</td><td align="center">F</td><td></td><td align="center">F</td><td align="center">17 7.3 cM</td></tr><tr><td align="left">#11</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AK011118">AK011118</ext-link></td><td align="center">fibrinogen, B beta polypeptide</td><td align="center">Fgb</td><td align="center">F</td><td></td><td align="center">F</td><td align="center">3 48.2 cM</td></tr><tr><td align="left">#12</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_133862">NM_133862</ext-link></td><td align="center">fibrinogen, gamma polypeptide</td><td align="center">Fgg</td><td align="center">(3.81)</td><td></td><td align="center">F</td><td align="center">3 41.3 cM</td></tr><tr><td align="left">#13</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BC005467">BC005467</ext-link></td><td align="center">fibrinogen, alpha polypeptide</td><td align="center">Fga</td><td align="center">(3.10)</td><td></td><td align="center">F</td><td align="center">3 44.8 cM</td></tr><tr><td align="left">#14</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_010168">NM_010168</ext-link></td><td align="center">coagulation factor II</td><td align="center">F2</td><td></td><td></td><td align="center">(2.67)</td><td align="center">2 47.5 cM</td></tr><tr><td align="left" colspan="8"><bold>Endopeptidase activity</bold></td></tr><tr><td align="left">#15</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BC012874">BC012874</ext-link></td><td align="center">serine (or cysteine) proteinase inhibitor, clade A, member 1a</td><td align="center">Serpina1a</td><td align="center">F</td><td></td><td align="center">F</td><td align="center">12 51.0 cM</td></tr><tr><td align="left">#16</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_007443">NM_007443</ext-link></td><td align="center">alpha 1 microglobulin/bikunin</td><td align="center">Ambp</td><td align="center">F</td><td></td><td align="center">F</td><td align="center">4 30.6 cM</td></tr><tr><td align="left">#17</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_010582">NM_010582</ext-link></td><td align="center">inter-alpha trypsin inhibitor, heavy chain 2</td><td align="center">Itih2</td><td align="center">F</td><td></td><td align="center">F</td><td align="center">2 1.0 cM</td></tr><tr><td align="left">#18</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_008407">NM_008407</ext-link></td><td align="center">inter-alpha trypsin inhibitor, heavy chain 3</td><td align="center">Itih3</td><td align="center">F</td><td></td><td align="center">F</td><td></td></tr><tr><td align="left">#29</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_007618">NM_007618</ext-link></td><td align="center">serine (or cysteine) proteinase inhibitor, clade A, member 6</td><td align="center">Serpina6</td><td align="center">(2.66)</td><td></td><td align="center">F</td><td align="center">12 51.0 cM</td></tr><tr><td align="left">#20</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_019429">NM_019429</ext-link></td><td align="center">protease, serine, 16 (thymus)</td><td align="center">Prss16</td><td></td><td align="center">(2.65)</td><td></td><td align="center">13 10.0 cM</td></tr><tr><td align="left" colspan="8"><bold>Erythroid-associated</bold></td></tr><tr><td align="left">#21</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_013848">NM_013848</ext-link></td><td align="center">erythroblast membrane-associated protein</td><td align="center">Ermap</td><td align="center">F</td><td></td><td></td><td align="center">4 57.0 cM</td></tr><tr><td align="left">#22</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AF069311">AF069311</ext-link></td><td align="center">Rhesus blood group CE and D</td><td align="center">Rhced</td><td align="center">F</td><td></td><td></td><td align="center">4 65.7 cM</td></tr><tr><td align="left">#23</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_053149">NM_053149</ext-link></td><td align="center">hemogen</td><td align="center">Hemgn</td><td align="center">(3.29)</td><td align="center">F</td><td align="center">F</td><td></td></tr><tr><td align="left">#24</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_010635">NM_010635</ext-link></td><td align="center">Kruppel-like factor 1 (erythroid)</td><td align="center">Klf1</td><td align="center">(3.01)</td><td></td><td></td><td></td></tr><tr><td align="left">#25</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_133245">NM_133245</ext-link></td><td align="center">erythroid associated factor</td><td align="center">Eraf</td><td align="center">(2.38)</td><td></td><td></td><td></td></tr><tr><td align="left">#26</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_010369">NM_010369</ext-link></td><td align="center">glycophorin A</td><td align="center">Gypa</td><td align="center">(2.27)</td><td></td><td></td><td align="center">8 36.0 cM</td></tr><tr><td align="left">#27</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AJ007909">AJ007909</ext-link></td><td align="center">erythroid differentiation regulator 1</td><td align="center">Erdr1</td><td></td><td></td><td align="center">(1.75)</td><td></td></tr><tr><td align="left" colspan="8"><bold>Hormone metabolism/regulation</bold></td></tr><tr><td align="left">#28</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_008293">NM_008293</ext-link></td><td align="center">hydroxysteroid dehydrogenase-1, delta<5>-3-beta</td><td align="center">Hsd3b1</td><td align="center">M</td><td></td><td></td><td align="center">3 49.1 cM</td></tr><tr><td align="left">#29</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AV021656">AV021656</ext-link></td><td align="center">aldo-keto reductase family 1, member B7</td><td align="center">Akr1b7</td><td align="center">M</td><td></td><td></td><td align="center">6 14.0 cM</td></tr><tr><td align="left">#30</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_009995">NM_009995</ext-link></td><td align="center">cytochrome P450, family 21, subfamily a, polypeptide 1</td><td align="center">Cyp21a1</td><td align="center">M</td><td></td><td></td><td></td></tr><tr><td align="left">#31</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AI195150">AI195150</ext-link></td><td align="center">group specific component</td><td align="center">Gc</td><td align="center">F</td><td></td><td align="center">F</td><td align="center">5 44.0 cM</td></tr><tr><td align="left">#32</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BG141874">BG141874</ext-link></td><td align="center">transthyretin</td><td align="center">Ttr</td><td align="center">(3.93)</td><td align="center">F</td><td align="center">F</td><td align="center">18 7.0 cM</td></tr><tr><td align="left">#33</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_031192">NM_031192</ext-link></td><td align="center">renin 1 structural</td><td align="center">Ren1</td><td align="center">3.67</td><td align="center">F</td><td></td><td></td></tr><tr><td align="left">#34</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_007423">NM_007423</ext-link></td><td align="center">alpha fetoprotein</td><td align="center">Afp</td><td align="center">(2.64)</td><td></td><td align="center">F</td><td align="center">5 50.0 cM</td></tr><tr><td align="left">#35</td><td align="left"><ext-link ext-link-type="gen" xlink:href="U63146">U63146</ext-link></td><td align="center">retinol binding protein 4, plasma</td><td align="center">Rbp4</td><td align="center">(1.89)</td><td></td><td align="center">(2.16)</td><td align="center">19 38.0 cM</td></tr><tr><td align="left" colspan="8"><bold>Immune functions</bold></td></tr><tr><td align="left">#36</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_011558">NM_011558</ext-link></td><td align="center">T-cell receptor gamma, variable 4</td><td align="center">Tcrg-V4</td><td align="center">F</td><td align="center">F</td><td align="center">F</td><td></td></tr><tr><td align="left">#37</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_009019">NM_009019</ext-link></td><td align="center">recombination activating gene 1</td><td align="center">Rag1</td><td></td><td align="center">F</td><td align="center">F</td><td align="center">2 56.0 cM</td></tr><tr><td align="left">#38</td><td align="left"><ext-link ext-link-type="gen" xlink:href="M58149">M58149</ext-link></td><td align="center">CD3 antigen, gamma polypeptide</td><td align="center">Cd3g</td><td align="center">F</td><td align="center">F</td><td></td><td align="center">9 26.0 cM</td></tr><tr><td align="left">#39</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AF119253">AF119253</ext-link></td><td align="center">histocompatibility 2, class II antigen A, alpha</td><td align="center">H2-Aa</td><td></td><td align="center">F</td><td></td><td></td></tr><tr><td align="left">#40</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AV058500">AV058500</ext-link></td><td align="center">P lysozyme structural</td><td align="center">Lzp-s</td><td align="center">3.42</td><td></td><td align="center">1.70</td><td></td></tr><tr><td align="left">#41</td><td align="left"><ext-link ext-link-type="gen" xlink:href="X67128">X67128</ext-link></td><td align="center">T-cell receptor beta, variable 13</td><td align="center">Tcrb-V13</td><td align="center">(1.97)</td><td align="center">F</td><td align="center">F</td><td align="center">6 20.5 cM</td></tr><tr><td align="left">#42</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BC011474">BC011474</ext-link></td><td align="center">lymphocyte protein tyrosine kinase</td><td align="center">Lck</td><td></td><td align="center">(2.87)</td><td></td><td align="center">4 59.0 cM</td></tr><tr><td align="left">#43</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BC003476">BC003476</ext-link></td><td align="center">Ia-associated invariant chain</td><td align="center">Ii</td><td></td><td align="center">(2.10)</td><td></td><td align="center">18 32.0 cM</td></tr><tr><td align="left" colspan="8"><bold>Lipid metabolism</bold></td></tr><tr><td align="left">#44</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_013475">NM_013475</ext-link></td><td align="center">apolipoprotein H</td><td align="center">Apoh</td><td align="center">F</td><td></td><td align="center">F</td><td align="center">11 63.0 cM</td></tr><tr><td align="left">#45</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_009692">NM_009692</ext-link></td><td align="center">apolipoprotein A-I</td><td align="center">Apoa1</td><td align="center">F</td><td></td><td align="center">F</td><td align="center">9 27.0 cM</td></tr><tr><td align="left">#46</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_009695">NM_009695</ext-link></td><td align="center">apolipoprotein C-II</td><td align="center">Apoc2</td><td align="center">(3.20)</td><td></td><td align="center">F</td><td align="center">7 4.0 cM</td></tr><tr><td align="left">#47</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_013474">NM_013474</ext-link></td><td align="center">apolipoprotein A-II</td><td align="center">Apoa2</td><td align="center">(2.27)</td><td></td><td align="center">F</td><td align="center">1 92.6 cM</td></tr><tr><td align="left">#48</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BC002148">BC002148</ext-link></td><td align="center">fatty acid binding protein 4, adipocyte</td><td align="center">Fabp4</td><td align="center">2.33</td><td></td><td></td><td align="center">3 13.9 cM</td></tr><tr><td align="left">#49</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BG922397">BG922397</ext-link></td><td align="center">p47 protein</td><td align="center">p47</td><td align="center">1.84</td><td></td><td></td><td></td></tr><tr><td align="left" colspan="8"><bold>Protein biosynthesis</bold></td></tr><tr><td align="left">#50</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_012011">NM_012011</ext-link></td><td align="center">eukaryotic translation initiation factor 2, subunit 3, structural gene Y-linked</td><td align="center">Eif2s3y</td><td align="center">M</td><td align="center">M</td><td align="center">M</td><td></td></tr><tr><td align="left">#51</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_009079">NM_009079</ext-link></td><td align="center">ribosomal protein L22</td><td align="center">Rpl22</td><td align="center">1.86</td><td></td><td></td><td></td></tr><tr><td align="left">#52</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AI642440">AI642440</ext-link></td><td align="center">ribosomal protein S13</td><td align="center">Rps13</td><td align="center">(1.81)</td><td></td><td></td><td></td></tr><tr><td align="left" colspan="8"><bold>Signal transduction</bold></td></tr><tr><td align="left">#53</td><td align="left"><ext-link ext-link-type="gen" xlink:href="X67702">X67702</ext-link></td><td align="center">secretoglobin, family 1A, member 1 (uteroglobin)</td><td align="center">Scgb1a1</td><td align="center">M</td><td></td><td></td><td></td></tr><tr><td align="left">#54</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BF537798">BF537798</ext-link></td><td align="center">receptor (calcitonin) activity modifying protein 2</td><td align="center">Ramp2</td><td align="center">2.07</td><td align="center">2.16</td><td></td><td align="center">11 61.5 cM</td></tr><tr><td align="left">#55</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_016721">NM_016721</ext-link></td><td align="center">IQ motif containing GTPase activating protein 1</td><td align="center">Iqgap1</td><td align="center">1.96</td><td></td><td></td><td align="center">7 39.0 cM</td></tr><tr><td align="left" colspan="8"><bold>Transcriptionnal regulation</bold></td></tr><tr><td align="left">#56</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AA210261">AA210261</ext-link></td><td align="center">DEAD (Asp-Glu-Ala-Asp) box polypeptide 3, Y-linked</td><td align="center">Ddx3y</td><td align="center">M</td><td align="center">M</td><td align="center">M</td><td align="center">Y 2.07 cM</td></tr><tr><td align="left">#57</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AF127244">AF127244</ext-link></td><td align="center">jumonji, AT rich interactive domain 1D (Rbp2 like)</td><td align="center">Jarid1d</td><td align="center">M</td><td align="center">M</td><td align="center">M</td><td align="center">Y 2.03 cM</td></tr><tr><td align="left">#58</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AA617392">AA617392</ext-link></td><td align="center">Max protein</td><td align="center">Max</td><td></td><td align="center">F</td><td></td><td align="center">12 32.0 cM</td></tr><tr><td align="left">#59</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BB393998">BB393998</ext-link></td><td align="center">flap structure specific endonuclease 1</td><td align="center">Fen1</td><td align="center">3.70</td><td align="center">(2.96)</td><td></td><td></td></tr><tr><td align="left">#60</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_007496">NM_007496</ext-link></td><td align="center">AT motif binding factor 1</td><td align="center">Atbf1</td><td></td><td></td><td align="center">(2.16)</td><td align="center">8 E1</td></tr><tr><td align="left" colspan="8"><bold>Transport</bold></td></tr><tr><td align="left">#61</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BC024643">BC024643</ext-link></td><td align="center">albumin 1</td><td align="center">Alb1</td><td align="center">(2.71)</td><td></td><td align="center">(8.68)</td><td align="center">5 50.0 cM</td></tr><tr><td align="left">#62</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AF440692">AF440692</ext-link></td><td align="center">Transferring</td><td align="center">Trf</td><td align="center">(2.02)</td><td></td><td align="center">(3.29)</td><td align="center">9 56.0 cM</td></tr><tr><td align="left">#63</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BB448377">BB448377</ext-link></td><td align="center">solute carrier family 4 (anion exchanger), member 1</td><td align="center">Slc4a1</td><td align="center">(2.07)</td><td></td><td></td><td align="center">11 62.0 cM</td></tr><tr><td align="left">#64</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_026331">NM_026331</ext-link></td><td align="center">mitochondrial solute carrier protein</td><td align="center">Mscp</td><td align="center">(1.90)</td><td></td><td></td><td></td></tr><tr><td align="left" colspan="8"><bold>Others</bold></td></tr><tr><td align="left">#65</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BB742957">BB742957</ext-link></td><td align="center">ubiquitously transcribed tetratricopeptide repeat gene, Y chr.</td><td align="center">Uty</td><td align="center">M</td><td align="center">M</td><td align="center">M</td><td align="center">Y 2.06 cM</td></tr><tr><td align="left">#66</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AI256465">AI256465</ext-link></td><td align="center">alpha-2-HS-glycoprotein</td><td align="center">Ahsg</td><td align="center">F</td><td></td><td align="center">F</td><td align="center">16 15.0 cM</td></tr><tr><td align="left">#67</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AK003182">AK003182</ext-link></td><td align="center">myosin, light polypeptide 1</td><td align="center">Myl1</td><td></td><td align="center">F</td><td></td><td align="center">1 34.1 cM</td></tr><tr><td align="left">#68</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BG806300">BG806300</ext-link></td><td align="center">inactive X specific transcripts</td><td align="center">Xist</td><td align="center">(3.34)</td><td align="center">(13.22)</td><td align="center">(2.41)</td><td align="center">X 42.0 cM</td></tr><tr><td align="left">#69</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AV224521">AV224521</ext-link></td><td align="center">Gelsolin</td><td align="center">Gsn</td><td align="center">3.27</td><td></td><td></td><td align="center">2 24.5 cM</td></tr><tr><td align="left">#70</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_025711">NM_025711</ext-link></td><td align="center">Asporin</td><td align="center">Aspn</td><td align="center">2.72</td><td></td><td></td><td></td></tr><tr><td align="left">#71</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BC002136">BC002136</ext-link></td><td align="center">coronin, actin binding protein 1A</td><td align="center">Coro1a</td><td></td><td></td><td align="center">(2.48)</td><td align="center">7 62.5 cM</td></tr><tr><td align="left">#72</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AK017440">AK017440</ext-link></td><td align="center">RNA imprinted and accumulated in nucleus</td><td align="center">Rian</td><td></td><td align="center">(2.27)</td><td></td><td align="center">12 54.5 cM</td></tr><tr><td align="left">#73</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AV156640">AV156640</ext-link></td><td align="center">expressed in non-metastatic cells 1, protein</td><td align="center">Nme1</td><td align="center">2.23</td><td></td><td></td><td></td></tr><tr><td align="left">#74</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AF291655">AF291655</ext-link></td><td align="center">Tenomodulin</td><td align="center">Tnmd</td><td align="center">2.12</td><td></td><td></td><td></td></tr><tr><td align="left">#75</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BB745314">BB745314</ext-link></td><td align="center">male-specific lethal-2 homolog (Drosophila)</td><td align="center">Msl2</td><td></td><td align="center">(2.11)</td><td></td><td></td></tr><tr><td align="left">#76</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BB490338">BB490338</ext-link></td><td align="center">calponin 3, acidic</td><td align="center">Cnn3</td><td align="center">2.04</td><td></td><td></td><td></td></tr><tr><td align="left">#77</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BB475271">BB475271</ext-link></td><td align="center">LUC7-like 2 (S. cerevisiae)</td><td align="center">Luc7l2</td><td></td><td align="center">1.96</td><td></td><td></td></tr><tr><td align="left">#78</td><td align="left"><ext-link ext-link-type="gen" xlink:href="NM_053123">NM_053123</ext-link></td><td align="center">SWI/SNF related, matrix associated, actin dependent regulator of chromatin, subfamily a, member 1</td><td align="center">Smarca1</td><td></td><td align="center">1.93</td><td></td><td></td></tr><tr><td align="left">#79</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BM207360">BM207360</ext-link></td><td align="center">unc-50 homolog (C. elegans)</td><td align="center">Unc50</td><td></td><td></td><td align="center">1.91</td><td></td></tr><tr><td align="left">#80</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AV011848">AV011848</ext-link></td><td align="center">malate dehydrogenase 1, NAD (soluble)</td><td align="center">Mdh1</td><td align="center">1.89</td><td></td><td></td><td align="center">11 12.0 cM</td></tr><tr><td align="left">#81</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AA270173">AA270173</ext-link></td><td align="center">Lamin B1</td><td align="center">Lmnb1</td><td></td><td align="center">(1.89)</td><td></td><td align="center">18 29.0 cM</td></tr><tr><td align="left">#82</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BB192717">BB192717</ext-link></td><td align="center">protein phosphatase 2 (formerly 2A), reg. subunit A (PR 65), alpha isoform</td><td align="center">Ppp2r1a</td><td></td><td align="center">(1.88)</td><td></td><td></td></tr><tr><td align="left">#83</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BE372352">BE372352</ext-link></td><td align="center">ARP3 actin-related protein 3 homolog (yeast)</td><td align="center">Actr3</td><td align="center">1.77</td><td></td><td></td><td></td></tr><tr><td align="left" colspan="8"><bold>Unknowns</bold></td></tr><tr><td align="left">#84</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AA717264">AA717264</ext-link></td><td align="center">Transcribed sequences</td><td></td><td></td><td align="center">2.51</td><td></td><td></td></tr><tr><td align="left">#85</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BC021831">BC021831</ext-link></td><td align="center">CDNA clone MGC:67258 IMAGE:6413648, complete cds</td><td></td><td></td><td align="center">(2.14)</td><td align="center">(2.36)</td><td></td></tr><tr><td align="left">#86</td><td align="left"><ext-link ext-link-type="gen" xlink:href="AK018316">AK018316</ext-link></td><td align="center">DNA segment, Chr 2, ERATO Doi 145, expressed</td><td align="center">D2Ertd145e</td><td></td><td align="center">2.31</td><td></td><td align="center">2 31.0 cM</td></tr><tr><td align="left">#87</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BE865094">BE865094</ext-link></td><td align="center">expressed sequence AI448196</td><td align="center">AI448196</td><td></td><td></td><td align="center">(2.14)</td><td></td></tr><tr><td align="left">#88</td><td align="left"><ext-link ext-link-type="gen" xlink:href="BB151477">BB151477</ext-link></td><td align="center">RIKEN cDNA 2810468K05 gene</td><td align="center">2810468K05Rik</td><td></td><td align="center">1.88</td><td></td><td></td></tr></tbody></table><table-wrap-foot><p>- Identification numbers of genes found in this study. <sup>†</sup>- Genes are presented in only one functional category although some of them could have been placed in more than one category. <sup>‡</sup>- Values with no bracket correspond to higher levels for males, while values in brackets correspond to higher levels for females. The letter M (Male) or F (Female) replace the fold change value when the gene was expressed in one sex only. <sup>§</sup>- The number of genes differentially expressed at each gestational day (GD) is indicated.</p></table-wrap-foot></table-wrap><p>The expression patterns of surfactant associated protein C (SP-C) obtained from microarray experiments (Figure <xref ref-type="fig" rid="F1">1A</xref>) and real time PCR (Figure <xref ref-type="fig" rid="F1">1B</xref>) using the same RNA preparations are presented. The reported increase in SP-C mRNA was observed in both cases.</p><fig position="float" id="F1"><label>Figure 1</label><caption><p><bold>Relative levels of SP-C gene expression on GDs 15.5, 16.5 and 17.5 </bold>Relative levels of SP-C gene expression (± SD) on GDs 15.5, 16.5 and 17.5 obtained by microarray hybridization (A) and real-time PCR (B). RNAs corresponding to samples "a" and "b" in panel B were pooled prior to preparation of probes used in panel A. "a" and "b" refer to different litters.</p></caption><graphic xlink:href="1477-7827-4-25-1"/></fig></sec><sec><title>Several genes of interest exhibiting no sex difference in expression</title><p>Several genes involved in lung development and expressed with no sex difference in our study are presented in Table <xref ref-type="table" rid="T3">3</xref>. These genes encode members of the fibroblast growth factor (FGF) [<xref ref-type="bibr" rid="B28">28</xref>] and the TGFβ [<xref ref-type="bibr" rid="B29">29</xref>] gene families, their receptors, epidermal growth factor-receptor (EGF-R) [<xref ref-type="bibr" rid="B30">30</xref>], insulin-like growth factor I (IGF-I) [<xref ref-type="bibr" rid="B31">31</xref>] and members of the surfactant-associated protein family [<xref ref-type="bibr" rid="B32">32</xref>].</p><table-wrap position="float" id="T3"><label>Table 3</label><caption><p>Several genes involved in lung development and expressed with no sex difference</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left"><bold>FGF signaling pathway</bold></td><td></td><td align="center"><bold>TGF-beta signaling pathway</bold></td><td align="center"><bold>Surfactant associated proteins</bold></td><td align="center"><bold>Others</bold></td></tr></thead><tbody><tr><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Spry</td><td align="center">Fgfbp1</td><td align="center">Tgfb1</td><td align="center">Sftpa</td><td align="center">Egfr</td></tr><tr><td align="left">Fgf1</td><td align="center">Fgfr1op2</td><td align="center">Tgfb2</td><td align="center">Sftpb</td><td align="center">Igf1</td></tr><tr><td align="left">Fgf7</td><td align="center">Fgfrap1</td><td align="center">Tgfb3</td><td align="center">Sftpc</td><td></td></tr><tr><td align="left">Fgf13</td><td align="center">Akr1b8</td><td align="center">Tgfbr1</td><td align="center">Sftpd</td><td></td></tr><tr><td align="left">Fgf18</td><td align="center">Fgfrl1</td><td align="center">Tgfbr2</td><td></td><td></td></tr><tr><td align="left">Fgfr1</td><td></td><td align="center">Tgfbr3</td><td></td><td></td></tr><tr><td align="left">Fgfr2</td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Fgfr3</td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Fgfr4</td><td></td><td></td><td></td><td></td></tr></tbody></table></table-wrap></sec><sec><title>Genes with sexually dimorphic expression</title><p>Genes with sexually dimorphic expression in mouse lungs on GDs 15.5, 16.5, and 17.5 are presented in Table <xref ref-type="table" rid="T2">2</xref>. These genes are distributed according to an adaptation of the functional categories defined by the Gene Ontology Consortium [<xref ref-type="bibr" rid="B33">33</xref>]. Only 33 genes displayed a sex difference on GD 16.5 compared with 62 and 38 on GD 15.5 and 17.5, respectively. Of these, similar numbers of genes showed higher expression levels for each sex on GD 15.5, while 21 of the 33 genes identified on GD 16.5, and 32 of the 38 genes identified on GD 17.5 presented higher expression in the female lung. Hierarchical clustering and expression profile of these genes are presented in Figure <xref ref-type="fig" rid="F2">2</xref>.</p><fig position="float" id="F2"><label>Figure 2</label><caption><p><bold>Hierarchical clustering and expression profile of differentially expressed genes </bold>Hierarchical clustering and expression profile of genes presented in Table 2. The day of gestation is indicated at top. Accession numbers and gene symbols are presented. For each gene individually, each value was normalized by division with the root mean square calculated from the six values. As a consequence, the relative values obtained for one gene cannot be compared to those of other genes. Expression levels are presented from black (no expression) to yellow (high relative expression level). The clustering was generated using the Genesis 1.6.0 Beta1 software (agglomeration rule of average linkage and Euclidean measurement distance) [26]. M, male; F, female.</p></caption><graphic xlink:href="1477-7827-4-25-2"/></fig></sec></sec><sec><title>Discussion</title><p>This study targeted elucidation of a sex difference in the fetal lung during a three-day gestation period overlapping PTII cell maturation. This sex difference results in more concern for a poorer prognosis for respiratory distress in premature male infants. The genes identified here, with a sexually dimorphic expression during the gestation period examined in the mouse, likely contain key genes involved in PTII cell maturation. Further studies are essential to identify and to characterize these key genes, as our data indicate that PTII cell maturation is not the only aspect of lung development affected by sex during this period of gestation.</p><p>Analysis of the data has to be performed in the context of lung development where a transient delay in the surge of surfactant synthesis is observed. Therefore, as in the case of the surge of surfactant synthesis, some genes can be subject to a transient delay in expression for one sex. The case of the Cyp21a1 gene clearly illustrates this occurrence. Recently, we showed that Cyp21a1, a gene involved in corticosteroid synthesis, is expressed specifically on GD 15 in the fetal mouse lung [<xref ref-type="bibr" rid="B27">27</xref>]. Of the six litters studied on GD 15.5 and obtained using the same mating window of ± 8 hours, two litters presented high expression of Cyp21a1 in females only, whereas one litter had high expression of this gene only in male fetuses. The three remaining litters did not show any elevated levels of expression for this gene (Figure <xref ref-type="fig" rid="F3">3</xref>). The exact gestation time at which the pregnant females were sacrificed varied from litter to litter according to the time where the females were mated during the mating window of ± 8h. Knowing that Balb/C is an inbred strain, variations from litter to litter should represent the expression pattern of Cyp21a1 gene at different times on GD 15. Consequently, our results strongly suggested that Cyp21a1 should present a narrow peak of expression on GD 15 with a delay for one sex [<xref ref-type="bibr" rid="B27">27</xref>]. In the present study, expression of Cyp21a1 (Id. #30) was detected by gene profiling only in males on GD 15.5. Real-time PCR analysis of the two litters pooled to prepare probes on GD15.5 revealed that Cyp21a1 was expressed at high levels only in male lungs (one out of two litters) (data not shown). Therefore, our microarray results are compatible with our previous report [<xref ref-type="bibr" rid="B27">27</xref>]. In this example, the fact that we detected expression only in male fetuses by microarrays depended only on the litters used. Therefore, for the analysis of our microarray results, the identity of the sex where expression is higher compared to the other sex has to be considered with caution. Other animals, mated some hours before or after the pregnant females used in this study within the same mating window, could show elevated expression levels for the other sex, or even present no sex difference at all, as a consequence of a delayed expression in one sex.</p><fig position="float" id="F3"><label>Figure 3</label><caption><p><bold>Hypothesis concerning the expression profile ofsome differentially expressed genes </bold>Our previous results suggest that Cyp21a1 is expressed at different precise gestation time on GD 15 for each gender [27] as shown in the present figure. This hypothesis would explain why the expression profile of Cyp21a1 on GD15.5 varied from litter to litter and presented one of the following patterns : 1) high expression in female fetal lungs only; 2) high expression in male fetal lungs only; or 3) no elevated expression in male and female fetal lungs. Because Balb/C is an inbred strain, the mating window of 16h would explain the variation between litters. Genes presenting such an expression profile cannot be studied by microarrays from a pool of many litters because the sex difference in expression could be lost.</p></caption><graphic xlink:href="1477-7827-4-25-3"/></fig><p>Recently, we reported that all the genes involved in corticosteroid synthesis from cholesterol in the mature adrenal gland are transiently expressed in the developing mouse lung on GD 15 [<xref ref-type="bibr" rid="B27">27</xref>]. Hsd3b1 (Id. #28) encodes the 3β-hydroxysteroid dehydrogenase type 1 enzyme that is involved in this cascade of reactions (pregnenolone → progesterone) along with Cyp21a1 (Id. #30) (progesterone → deoxycorticosterone) as mentioned already. Both genes were also found by microarray analysis to be expressed transiently on GD 15 with a sex difference.</p><p>Some other genes related to steroid hormone metabolism/regulation are expressed also with a sex difference in fetal mouse lungs. This is the case for Akr1b7 (Id. #29) which showed a sex difference in expression on GD 15.5. This gene encodes an aldose reductase-like protein whose major function is detoxification of isocaproaldehyde generated by conversion of cholesterol to pregnenolone by the enzyme P450scc [<xref ref-type="bibr" rid="B34">34</xref>], the latter also being expressed on GD 15.5 in the mouse fetal lung [<xref ref-type="bibr" rid="B27">27</xref>]. Scgb1a1 gene (Id. #53) is expressed also on GD 15.5 with a sex difference. This gene encodes the secretoglobin family 1 A member 1, also called Clara cell secretory protein (CCSP), which is required for the appearance of Clara cells secretory granules and thus participates to the composition of the epithelial lining fluid [<xref ref-type="bibr" rid="B35">35</xref>]. Scgb1a1 gene can be induced by glucocorticoids [<xref ref-type="bibr" rid="B36">36</xref>] and CCSP can act as an anti-inflammatory protein [<xref ref-type="bibr" rid="B37">37</xref>]. It was also shown to bind progesterone [<xref ref-type="bibr" rid="B38">38</xref>]. Some other genes were not detected by microarrays although they are known to be expressed with a sex difference in lung development as evidenced by real-time PCR. These genes include 17β-hydroxysteroid dehydrogenase (HSD) type 2 and type 5 [<xref ref-type="bibr" rid="B18">18</xref>] which are expressed at low levels in the lung, probably below the minimal threshold of sensitivity of the microarray technology.</p><p>Some genes involved in the metabolism of non-steroidal hormones were also found to be expressed differentially between genders. This is the case for Ttr (Id. #32), which codes for transthyretin, a common plasma carrier protein for thyroid hormones and vitamin A metabolites [<xref ref-type="bibr" rid="B39">39</xref>]. Interestingly, retinoids are important regulators of normal epithelial cell differentiation and proliferation [<xref ref-type="bibr" rid="B40">40</xref>] and are involved in lung development [<xref ref-type="bibr" rid="B41">41</xref>]. Ren1 (Id. #33) encodes for renin 1 and presented a sex difference in expression on GD 15.5. Genes of the renin-angiotensin system are known to be expressed in the fetal lung and, interestingly, angiotensin II can present mitogenic effects on human lung fibroblasts through the activation of the type 1 angiotensin II receptor [<xref ref-type="bibr" rid="B42">42</xref>]. In our study, Ace (angiotensin II converting enzyme), Agtr1, and Agtr2 (angiotensin II receptor type 1 and type 2) are expressed with no sex difference (data not shown).</p><p>Our results demonstrate that genes coding for apolipoprotein (apo) AI (Id. #45), apoAII (Id. #47), apoCII (Id. #46), and apoH (Id. #44) are co-expressed in the developing lung. To our knowledge, none of these genes has been found expressed in the lung. All of these genes present a sex difference in favor of females in the developing lung on GD 15.5 and 17.5, with no detectable expression on GD 16.5. As demonstrated by real-time PCR, 17βHSD types 2 and 5 were also expressed on GDs 15.5 and 17.5 with a sex difference for the majority of litters, but not on GD16.5 [<xref ref-type="bibr" rid="B18">18</xref>]. Our results suggest that the expression of these four apolipoprotein genes in the developing lung is under active modulation. 17β-HSD types 5 and 2 genes are involved in androgen synthesis and inactivation, respectively. Such a similar pattern of expression for these apolipoproteins and 17β-HSD genes may suggest a common regulatory mechanism, or an effect of androgens on expression of these apolipoprotein genes. Apolipoproteins are constituents of circulating lipoproteins. ApoCII is an essential cofactor/activator of lipoprotein lipase (LPL) [<xref ref-type="bibr" rid="B43">43</xref>,<xref ref-type="bibr" rid="B44">44</xref>] while apoH was shown to enhance LPL [<xref ref-type="bibr" rid="B45">45</xref>]. This enzyme catalyses the hydrolysis of the triacylglycerol component of circulating lipoproteins (chylomicrons and very low density lipoproteins (VLDLs)). It was reported that the presence and activity of LPL in the lung may be important for surfactant production [<xref ref-type="bibr" rid="B46">46</xref>]. LPL expression was detected with no sex difference. ApoAI [<xref ref-type="bibr" rid="B47">47</xref>] and apoAII [<xref ref-type="bibr" rid="B48">48</xref>] are major protein components of high density lipoproteins (HDLs). ApoAI is a potent activator of lecithin:cholesterol acyltransferase (LCAT), an HDL-associated enzyme playing a role in reverse cholesterol transport. It is possible that the sex difference in expression of four apolipoprotein genes in the lung could explain in part the sex difference in surfactant lipids observed on GD 17.5; although such a role in the surge of surfactant synthesis has never been suspected for these genes before the present study. It is also noteworthy that many genes involved in lipid metabolism were expressed with no sex difference during each gestational day studied (data not shown). For example, this is the case for genes encoding for LDL receptor, LDL receptor-related protein 1, phospholipid transfer protein, and HMG-CoA-reductase (3-hydroxy-3-methylglutaryl-Coenzyme A reductase), the latter being known to catalyze the rate-limiting step of cholesterol synthesis.</p><p>Along with other pathways, fibroblast growth factors (FGFs) and TGFβ signaling pathways have been shown to be involved in many aspects of lung development. We observed expression of many genes that belong to these pathways (Table <xref ref-type="table" rid="T3">3</xref>). However, in the present study, these genes exhibit no sexually dimorphic expression. This is in line with our previous study by real-time PCR that showed expression of 5 genes involved in lung maturation on GDs 16.5 and 17.5 with no sex difference, namely, LIF, IGF-I, KGF (FGF-7), EGF-R, and neuregulin-1 [<xref ref-type="bibr" rid="B18">18</xref>]. Although it is possible that these genes could be expressed with a sex difference within a time window not covered by our studies, it seems that the sex difference in PTII cell maturation could rely on other genes.</p><p>Tissue remodeling and structure also seem subject to sexual differences in gene expression. Indeed, two genes involved in apoptosis, namely, Rock1 (Rho-associated coiled-coil forming kinase 1) (Id. #1) and Casp8ap2 (caspase 8 associated protein 2) (Id. #2) are differentially expressed on GD 16.5 between genders, while many genes involved in cell adhesion are expressed with a sex difference on GD 15.5. In addition, IGF signaling pathway has been shown to be important in mouse lung development where it is involved in regulation of cell proliferation and differentiation [<xref ref-type="bibr" rid="B31">31</xref>,<xref ref-type="bibr" rid="B49">49</xref>,<xref ref-type="bibr" rid="B50">50</xref>]. Igfbp1 (Id. #9) presented a sex difference in expression on GD 15.5, whereas Igfbp1 deficiency was associated with massive hepatocyte apoptosis [<xref ref-type="bibr" rid="B51">51</xref>].</p><p>In preparation for important immunological challenges related to its function, the developing lung acquires many elements associated with innate and adaptive immunity. Our data suggest that modulation of several genes involved in immune functions of the lung is subject to sex differences. Indeed, the predominating functional category of genes showing sexual dimorphism on GD 16.5 concerns "immune functions". It has been shown that many signaling pathways involved in lung morphogenesis and immune responses are crosslinked [<xref ref-type="bibr" rid="B52">52</xref>]. These include TTF-1, GATA6 and HNF-3β transcription factors; and FGF- and NF-κ B-dependant signaling pathways. However, we did not detect any sex difference in expression for these genes.</p></sec><sec><title>Conclusion</title><p>This study revealed that many genes are expressed with a gender difference in the fetal lung. Although we focused on a brief gestational period overlapping the surge of surfactant synthesis, our data demonstrates that PTII cell maturation is not the only aspect of lung development under the influence of fetal sex. We suggest that, among the genes identified here, some are related to the transient delay in lung maturation observed for males. This would help identify novel genes involved in the physiopathology of respiratory distress of the neonate.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>MS participated in the design of the study, performed the data analysis and wrote the manuscript. PRP participated in the design of the study and helped to draft the manuscript. YT conceived the study, participated in its design and in its coordination. All authors read and approved the final manuscript.</p></sec>
Chronic obstructive pulmonary disease (COPD) and occupational exposures	<p>Chronic obstructive pulmonary disease (COPD) is one of the leading causes of morbidity and mortality in both industrialized and developing countries.</p><p>Cigarette smoking is the major risk factor for COPD. However, relevant information from the literature published within the last years, either on general population samples or on workplaces, indicate that about 15% of all cases of COPD is work-related.</p><p>Specific settings and agents are quoted which have been indicated or confirmed as linked to COPD. Coal miners, hard-rock miners, tunnel workers, concrete-manufacturing workers, nonmining industrial workers have been shown to be at highest risk for developing COPD.</p><p>Further evidence that occupational agents are capable of inducing COPD comes from experimental studies, particularly in animal models.</p><p>In conclusion, occupational exposure to dusts, chemicals, gases should be considered an established, or supported by good evidence, risk factor for developing COPD. The implications of this substantial occupational contribution to COPD must be considered in research planning, in public policy decision-making, and in clinical practice.</p>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Boschetto</surname><given-names>Piera</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Quintavalle</surname><given-names>Sonia</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Miotto</surname><given-names>Deborah</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Lo Cascio</surname><given-names>Natalina</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Zeni</surname><given-names>Elena</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Mapp</surname><given-names>Cristina E</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	Journal of Occupational Medicine and Toxicology (London, England)	<sec><title/><p>1. Definition</p><p>2. Occupational exposures and COPD: epidemiologic evidence</p><p>3. Occupational exposures and COPD: experimental evidence</p><p>4. Occupationally-related COPD: diagnosis</p><p>5. Occupationally-related COPD: management and prevention</p></sec><sec><title>1. Definition</title><p>Chronic obstructive pulmonary disease (COPD) is a disease state characterized by airflow limitation that is not fully reversible. The airflow limitation is usually both progressive and associated with an abnormal inflammatory response of the lungs to noxious particles and gases [<xref ref-type="bibr" rid="B1">1</xref>].</p><p>Many previous definitions of COPD have emphasized the terms "emphysema" and "chronic bronchitis" which are no longer included in the definition of COPD [<xref ref-type="bibr" rid="B1">1</xref>]. Emphysema, or destruction of the gas-exchanging surface of the lung (alveoli), is a pathological term that is often (but incorrectly) used clinically and describes only one of several structural abnormalities present in patients with COPD. Chronic bronchitis, or the presence of cough and sputum production for at least 3 months in each of two consecutive years, remains a clinically and epidemiologically useful term. However, it does not reflect the major impact of airflow limitation on morbidity and mortality in COPD patients. It is also important to recognize that cough and sputum production may precede the development of airflow limitation; conversely, some patients develop airflow limitation without chronic cough and sputum production.</p><p>COPD does not have a clinical subcategory that is clearly identified as occupational, largely because the condition develops slowly and, given that the airflow limitation is chronic, does not reverse when exposure is discontinued. Thus, a clinical diagnosis of occupational COPD, using methods similar to those employed for occupational asthma, is not feasible. Epidemiologically, the identification of occupational COPD is based on observing excess occurrence of COPD among exposed workers [<xref ref-type="bibr" rid="B2">2</xref>-<xref ref-type="bibr" rid="B4">4</xref>].</p><p>Some work-related obstructive airway disorders have been classified as COPD but do not neatly fit into this category. For example, work-related variable airflow limitation may occur with occupational exposure to organic dusts such as cotton, flax, hemp, jute, sisal, and various grains. Such organic dust-induced airway disease is sometimes classified as an asthma-like disorder [<xref ref-type="bibr" rid="B5">5</xref>], but both chronic bronchitis and poorly reversible airflow limitation can develop with chronic exposure. Bronchiolitis obliterans and irritant-induced asthma are two other conditions that may overlap clinically with work-related COPD.</p></sec><sec><title>2. Occupational exposures and COPD: epidemiologic evidence</title><p>COPD is a major cause of chronic morbidity and mortality throughout the world. Many people suffer from this disease for years and die prematurely from it or its complications. COPD is currently the fourth leading cause of death in the world [<xref ref-type="bibr" rid="B6">6</xref>], and further increases in its prevalence and mortality can be predicted in the coming decades [<xref ref-type="bibr" rid="B7">7</xref>].</p><p>Cigarette smoking is undoubtedly the main cause of COPD in the population. A dose-response relationship between the amount smoked and an observed accelerated decline in ventilatory function have been consistently found in longitudinal epidemiological studies [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B8">8</xref>-<xref ref-type="bibr" rid="B11">11</xref>]; however, there is a huge individual variation. Fletcher and Peto [<xref ref-type="bibr" rid="B12">12</xref>], in an 8-yr prospective study of working men in West London, showed that the average decline in FEV<sub>1 </sub>in smokers is faster (60 ml/yr) than in non-smokers (30 ml/yr). However, smokers who develop COPD have an average decline in FEV<sub>1 </sub>of greater than 60 ml/yr, and only 15 to 20% of smokers develop clinically significant COPD. In addition, an estimated 6% of persons who had COPD in the United States are never smokers [<xref ref-type="bibr" rid="B13">13</xref>]. Cigarette smoke is analogous to a mixed inhalation exposure at a workplace because it is a complex mixture of particles and gases.</p><p>Despite the difficulty of disentangling the effect of cigarette smoke from those of other exposures, there is growing evidence from large population based studies suggesting that a sizeable proportion of the cases of COPD in a society may be attributable to workplace exposures to dusts, noxious gases/vapours, and fumes (DGVFs). The fraction of cases in a population that arise because of certain exposures is called the attributable fraction in the population or the population attributable risk (PAR). The American Thoracic Society (ATS) recently produced a consensus statement based on an evaluation of a number of large scale general population studies, and calculated that PAR for COPD was about 15% [<xref ref-type="bibr" rid="B14">14</xref>]. Several recent papers published since the completion of the ATS statement provide further evidence in support of a major contribution of occupational exposure to the burden of COPD. Hnizdo and coworkers from the National Institute for Occupational Safety and Health used data collected in the US population-based Third National Health and Nutrition Examination Survey on more than 9800 subjects to estimate the PAR for COPD attributable to work [<xref ref-type="bibr" rid="B15">15</xref>]. The analysis was adjusted for multiple factors, including smoking history. The industries with increased risk include rubber, plastics, and leather manufacturing, utilities, building services, textile manufacturing, and construction. The PAR for COPD attributable to work was estimated at 19% overall and 31% among never smokers. A second US population-based study conducted by Trupin and coworkers [<xref ref-type="bibr" rid="B16">16</xref>] obtained survey information on more than 2000 subjects. Occupational exposures were associated with increased risk of COPD after adjustment for smoking history and demographic variables. The PAR for COPD caused by these exposures was 20%. In this study, the PAR for combined current and former smokers was 56%. Smoking and occupational exposures to dusts, gases, and/or fumes had greater than additive effects. A third study from Sweden was designed to determine whether occupational exposure to dust, fumes, or gases, especially among never-smokers, increased the mortality from COPD [<xref ref-type="bibr" rid="B17">17</xref>]. A cohort of more than 317000 Swedish male construction workers was followed from 1971 to 1999. Exposure to inorganic dusts, gases and irritant chemicals, fumes, and wood dusts was based on a job-exposure matrix. An internal control group with "unexposed" construction workers was used, and the analyses were adjusted for age and smoking. There was a statistically significant increase mortality from COPD among those with any airborne exposure (relative risk 1.12). In a Poisson regression model, including smoking, age and the four major exposure groups listed previously, exposure to inorganic dust was associated with an increased risk, especially among never-smokers. The fraction of COPD among the exposed attributable to any airborne exposure was estimated as 10.7% overall and 52.6% among never-smokers. Thus, occupational exposure among construction workers increases mortality due to chronic obstructive pulmonary disease, even among never-smokers.</p><p>The determination of the PAR% due to occupational exposure has been complicated until recently by the lack of standardization of definition for COPD. Moreover, relatively few studies have been conducted with the specific purpose of determining the occupational contribution to COPD in the general population. In the studies that have been performed, there has been no consistency in terms of a strict definition of COPD. Some have presented data on symptoms and diseases, others have presented data on lung function, and a few have done both. Although a certain degree of standardization has been accomplished for cough and phlegm, dyspnea has been defined more variably among the studies.</p><p>While cigarette smoking and occupational exposures appear to account in combination for the major proportion of the population attributable risk of COPD, other influences are potentially important. The understanding of genetic susceptibility to this condition is still in its relative infancy, but certain data do suggest that genetics influences may be important [<xref ref-type="bibr" rid="B18">18</xref>], when considering both the established disease and the accelerated annual decline in FEV<sub>1</sub>. Furthermore, interactions have been noted between α<sub>1 </sub>anti-trypsin deficiency and environmental exposures in the development of COPD [<xref ref-type="bibr" rid="B19">19</xref>].</p></sec><sec><title>3. Occupational exposures and COPD: experimental evidence</title><p>The airflow limitation that defines COPD is associated with lesions that obstruct the small conducting airways, produce emphysematous destruction of the lung's elastic recoil force with closure of small airways, or both [<xref ref-type="bibr" rid="B20">20</xref>]. Experimental studies have demonstrated that several agents, including sulphur dioxide, mineral dusts, vanadium and endotoxin, are capable of inducing chronic obstructive bronchitis in animal models [<xref ref-type="bibr" rid="B21">21</xref>-<xref ref-type="bibr" rid="B24">24</xref>]. The list of agents that can cause emphysema in animals includes several for which there is also epidemiological evidence in exposed occupational cohorts, such as cadmium, coal, endotoxin, and silica [<xref ref-type="bibr" rid="B25">25</xref>]. The clearest human model of emphysema is that of α<sub>1 </sub>anti-trypsin deficiency [protease inhibitor phenotype Z (PI*Z)] [<xref ref-type="bibr" rid="B26">26</xref>]. This phenotype affects only a small percentage of the general population and is responsible for a correspondingly small fraction of the total burden of COPD. Although smoking is the most potent and well-established cofactor in emphysema related to α<sub>1 </sub>anti-trypsin deficiency, occupational exposure are linked to such disease as well [<xref ref-type="bibr" rid="B27">27</xref>,<xref ref-type="bibr" rid="B28">28</xref>].</p><p>Because α<sub>1 </sub>anti-trypsin is the endogenous inhibitor of neutrophil elastase and neutrophil elastase is capable to cause alveolar destruction, it has long been considered the major player in the development of emphysema. Yet, despite these evidences, it has been difficult to convincingly establish a role for neutrophil elastase in emphysema. The association of neutrophil elastase with human emphysema has been inconsistent, the extracellular release of neutrophil elastase has been questioned, and other proteinases have been shown to play a role in experimental models of emphysema. The finding that a murine knockout model lacking macrophage metalloelastase (MME) is resistant to the development of cigarette smoke-induced emphysema has created great interest in this enzyme and in the potential importance of other proteases [<xref ref-type="bibr" rid="B29">29</xref>-<xref ref-type="bibr" rid="B31">31</xref>].</p><p>The occupationally relevant agents that can cause emphysema (cadmium, coal, endotoxin, and silica), all cause the centrilobular form of the disease rather than the panacinar form that is associated with α<sub>1 </sub>anti-trypsin deficiency so mechanisms other than uninhibited neutrophil elastase activity are likely operative. The recent evidence about MME suggests a potential mechanism by which inhaled dusts or fumes could cause emphysema since macrophages have a primary role in the clearance of these materials from the terminal airways and alveoli.</p></sec><sec><title>4. Occupationally-related COPD: diagnosis</title><p>Cigarette smoking is by far the predominant risk factor for COPD. Till today, diagnostic assessments able to calculate the relative contribution of work exposures in a smoker with COPD are not available. However, adjustment of associations between occupational exposure and COPD for smoking status has been performed in epidemiological studies, showing that occupational risks likely play a role on their own. Thus, physicians must be aware of the potential occupational aetiologies for obstructive airway disease and should consider them in every patient with COPD. An occupational history should be the first step in the initial evaluation of the patient. A proper occupational history consists of a chronological list of all jobs, including job title, a description of the job activities, potential toxins at each job, and an assessment of the extent and duration of exposure. The length of time exposed to the agent, the use of personal protective equipment such as respirators, and a description of the ventilation and overall hygiene of the workplace are helpful in attempting to quantify exposure from the patient's history.</p><p>Additional information can be obtained from a visit to the workplace by experts in occupational hygiene, from material safety data sheets for workplace chemicals, and from the manufacturers of the workplace substances.</p><p>Identifying occupational risk factors on the individual level is important for prevention of disease before it is advanced and for modifying disability risk once disease is established [<xref ref-type="bibr" rid="B32">32</xref>]. In addition, the clinician has a critical role in case identification for the purposes of public health surveillance and appropriate work-related insurance compensation.</p></sec><sec><title>6. Occupationally-related COPD: management and prevention</title><p>Directions about the management and prevention of work-related diseases [<xref ref-type="bibr" rid="B33">33</xref>-<xref ref-type="bibr" rid="B35">35</xref>], can be applied to COPD as well. Physicians should attempt to understand the patient's occupational exposure and whether he/she has been adequately trained in the dangers of these exposures and how to manage them. Removal of the respiratory irritants and substitution of non-toxic agents are the best approach because they eliminate the work-related COPD hazard. If substitution is not possible, ongoing maintenance of engineering controls, such as enclosure of the industrial process and improving work area ventilation, are useful. Administrative controls (e.g., transfer to another job or change in work practices), and personal protective equipment (e.g., masks or respirators) should be mentioned, although less effective in decreasing exposures to respiratory tract irritants.</p><p>Guidelines for identification and management of individuals with work-related asthma have been recently published [<xref ref-type="bibr" rid="B36">36</xref>] and are relevant to work-related COPD. Unlike workers with sensitizer-induced asthma, workers with irritant-induced asthma or COPD may continue to work in their usual jobs if their exposure to the inciting agent is diminished via proper engineering controls or respiratory protective equipment if engineering controls are not feasible.</p><p>Prevention must be the primary tool for decreasing the incidence of morbidity and disability from work-related COPD, which can become severely disabling disease.</p><p>Primary prevention is designed to abate hazards before any damage or injury has occurred. Primary prevention strategies encompass the same exposure controls (elimination, engineering controls, administrative controls, personal protective equipment) described for management of work-related asthma and COPD due to irritant exposure. As cigarette smoking is the main risk factor for COPD, we wish to stress that smoking should be discouraged outside the workplace as well as inside the workplace.</p><p>Secondary prevention addresses early detection of the disease so that its duration and severity can be minimized. Medical surveillance programs are a type of secondary prevention. For medical surveillance of COPD, short symptom questionnaires can be administered before employment and repeated annually. They should include items such as improvement in respiratory symptoms on week-ends and holidays [<xref ref-type="bibr" rid="B37">37</xref>-<xref ref-type="bibr" rid="B39">39</xref>]. In addition, spirometry can be performed on an annual basis and compared to baseline spirometric testing at the time of hire. Review of peak expiratory flow rate records over several weeks can also detect workers at risk for developing irritant-induced COPD.</p><p>Tertiary prevention aims at the prevention of permanent COPD. It includes institution of appropriate health care. Furthermore, early recognition of the disease and early removal from, or reduction of, exposure, make it more likely that the patient will avoid permanent COPD.</p><p>Public policy needs to be better informed about the roles of occupational factors in obstructive airway disease. This will require active education and outreach on the part of the medical-scientific community. Specific public policy issues to be re-examined in light of the magnitude of the occupational contribution to the burden of airway disease include standard setting for exposure in and out of the workplace, attribution criteria for compensation, health care costs and their assignment, and health care resources allocation.</p><p>The clinician must be aware of the potential occupational aetiologies for obstructive airway disease and consider them in every patients with asthma or COPD. Identifying occupational risk factors on the individual level is important for prevention of disease before it is advanced and for modifying disability risk once disease is established [<xref ref-type="bibr" rid="B32">32</xref>]. In addition, the clinician has a critical role in case identification for the purposes of public health surveillance and appropriate work-related insurance compensation.</p></sec><sec><title>Conclusion</title><p>Careful review of the literature demonstrated that approximately 15% of COPD is work-related and that new agents causing COPD, as well as new cases with persistent airflow limitation associated with work, are still being reported. It definitely supports the concept that in a new classification of risk factors for COPD, occupational exposure to dusts, chemicals, gases should be considered an established, or supported by good evidence, risk factor.</p><p>Besides epidemiological studies, further experimental studies can lead to a better understanding of the occupational hazards which may cause COPD and establish a stronger link between the severity of COPD and specific occupations. Experimental studies may actually serve as models from which to derive basic insights of COPD and to identify a cellular basis of the work-related disease.</p></sec><sec><title>Authors' contributions</title><p>PB, SQ, DM, NLC, EZ and CEM have all been involved in drafting the article or revising it critically for important intellectual content and have given final approval of the version to be published.</p></sec><sec><title>Declaration of competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec>
Occupational health for an ageing workforce: do we need a geriatric perspective?	<p>Extending retirement ages and anti-age discrimination policies will increase the numbers of older workers in the future. Occupational health physicians may have to draw upon the principles and experience of geriatric medicine to manage these older workers. Examples of common geriatric syndromes that will have an impact on occupational health are mild cognitive impairment and falls at the workplace. Shifts in paradigms and further research into the occupational health problems of an ageing workforce will be needed.</p>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Koh</surname><given-names>Gerald Choon-Huat</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Koh</surname><given-names>David</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	Journal of Occupational Medicine and Toxicology (London, England)	<sec><title>Introduction – the ageing workforce</title><p>The world is undergoing unprecedented ageing and in many developed countries, the workforce is contracting due to falling birthrates, longer life expectancies and changing population demographics [<xref ref-type="bibr" rid="B1">1</xref>]. Experts have warned that if society continues to reduce the number of people over the age of 50 who are not actively working, economies will suffer a cumulative annual loss of GDP [<xref ref-type="bibr" rid="B2">2</xref>]. Some countries like the UK are already introducing anti-age discrimination policies laws and retirement ages are projected to increase in the coming years [<xref ref-type="bibr" rid="B3">3</xref>]. Employers now have to face the prospect of having workers in their sixties. In New Zealand, the number of older persons aged 45 to 65 years is expected to increase from 35% to 45% within the working-age population between 2001 and 2051 [<xref ref-type="bibr" rid="B4">4</xref>]. The International Labour Organisation estimates that the number of economically active persons aged 65 years and above will increase from 83.2 million persons in the world in 2000 to 136 million persons by 2020 [<xref ref-type="bibr" rid="B5">5</xref>]. Occupational physicians are accustomed to managing middle-aged workers and their associated health problems but are we ready to manage elderly-related illnesses that may impact worker performance and health?</p></sec><sec><title>What does geriatric medicine has to offer?</title><p>Geriatrics is the branch of medicine that is devoted to the care of older people [<xref ref-type="bibr" rid="B6">6</xref>]. The relatively young discipline addresses the unique needs and circumstances of the elderly and is characterized by recognition of geriatric syndromes. Examples of conditions that affect the elderly include falls, impaired cognition, disability, malnutrition, incontinence and iatrogenesis. At first glance, most of these syndromes are associated with advanced age and it is unlikely that such an old person would still be working and hence be seen by an occupational physician. However, when one considers that many geriatric syndromes can present in fifth decade of life, it becomes apparent that knowledge of geriatric syndromes may be relevant to occupational health. This paper will use 2 common geriatric syndromes that may impact on the occupational health of older workers to illustrate this.</p></sec><sec><title>Dementia and mild cognitive impairment</title><p>Dementia is often thought of as a psychiatric disease of the old. However, a paper by McMurtray et al found that 30% of patients presenting at the Veteran's Affairs Medical Center Memory Disorders clinic between 2001 and 2004 for evaluation of memory or cognitive decline had an age of onset of less than 65 years (early onset dementia [EOD]) [<xref ref-type="bibr" rid="B7">7</xref>]. Compared to the late-onset dementia [LOD] group, the EOD patients were less severely impaired on presentation. Hence, it is possible that an older worker may present with onset of dementia before retirement which can interfere with work or endanger the lives of fellow co-workers. It is interesting to note that the EOD group had significantly more dementia attributed to traumatic brain injury, alcohol abuse, human immunodeficiency virus (HIV) and frontotemporal lobe degeneration than the LOD patients which had significantly more Alzheimer's disease compared to the EOD group. With the exception of the last condition, the causes of EOD are largely preventable. Hence, occupational physicians can play an important role in the prevention, early detection and treatment of EOD.</p><p>One of the earliest cognitive domains lost in dementia is executive functioning involving understanding complex material, and this can occur before memory loss [<xref ref-type="bibr" rid="B8">8</xref>]. This has implications because most clinical diagnostic criteria for dementia involve subjective and objective memory impairment and functional decline. Even the clinical diagnostic criteria for mild cognitive impairment (MCI) requires subjective or objective memory loss but without functional impairment (Table <xref ref-type="table" rid="T1">1</xref>) [<xref ref-type="bibr" rid="B9">9</xref>]. An older worker in a job requiring high-level mental functioning may be making poor decisions and losing millions of dollars for the company long before anyone perceives any impairment of memory. Clinically, the distinction between benign senescent forgetfulness (normal process of ageing) and mild cognitive impairment is subtle and this makes the detection of early loss of executive functioning extremely difficult to detect.</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Various definitions of mild cognitive impairment (Adapted from Chong and Sahadevan [9])</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="center"><bold>Amnestic MCI</bold></td><td align="center"><bold>AACD</bold></td><td align="center"><bold>AAMI</bold></td><td align="center"><bold>CIND</bold></td><td align="center"><bold>CDR = 0.5</bold></td></tr></thead><tbody><tr><td align="left"><bold>Subjective memory impairment</bold></td><td align="center">+</td><td align="center">+</td><td align="center">+</td><td align="center">NR</td><td align="center">+</td></tr><tr><td align="left"><bold>Subjective non-memory impairment</bold></td><td align="center">-</td><td align="center">NR</td><td align="center">NR</td><td align="center">NR</td><td align="center">NR</td></tr><tr><td align="left"><bold>Objective memory impairment</bold></td><td align="center">+ <sup>a</sup></td><td align="center">+ <sup>b</sup></td><td align="center">+ <sup>c</sup></td><td align="center">+</td><td align="center">+</td></tr><tr><td align="left"><bold>Objective non-memory impairment</bold></td><td align="center">-</td><td align="center">NR</td><td align="center">NR</td><td align="center">NR</td><td align="center">NR</td></tr><tr><td align="left"><bold>Functional decline</bold></td><td align="center">NR</td><td align="center">NR</td><td align="center">NR</td><td align="center">NR</td><td align="center">+/-</td></tr><tr><td align="left"><bold>Functional impairment</bold></td><td align="center">-</td><td align="center">NR</td><td align="center">NR</td><td align="center">NR</td><td align="center">-</td></tr></tbody></table><table-wrap-foot><p><underline>Abbreviations</underline> MCI = mild cognitive impairment; AACD = age-associated cognitive decline; AAMI = age-associated memory impairment, CIND = cognitive impairment no dementia; CDR = clinical dementia rating scale; the score of 0.5 is used to denote, MCI + = must be present for diagnosis; - = must be absent for diagnosis; +/- = may or may not be present for diagnosis; NR = not required (or not mentioned as criteria for diagnosis); a: >1.5 SD below age-matched controls; b: within normal limits given person's age; c: >1 SD below mean for young adults.</p></table-wrap-foot></table-wrap><p>Fitness for work for workers which require intact cognition will continue to be a challenge with older workers. The earliest an occupational health physician can hope to detect cognitive decline would be when a worker has MCI. This intermediate stage between normal ageing and dementia has received increasing attention because current therapies for dementia are most effective at the early stages and 12% of cases with MCI convert to dementia annually, reaching 80% at 6 years follow-up [<xref ref-type="bibr" rid="B10">10</xref>]. Unfortunately, there is currently no consensus guideline for the diagnosis of mild cognitive impairment but there is evidence for its continued monitoring and treatment [<xref ref-type="bibr" rid="B11">11</xref>]. Current cognitive screening tools to detect dementia have not been validated to detect MCI and clinicians have to rely on special cognitive tests. Prospective studies of people with memory-loss MCI have shown that tests involving episodic memory (such as delayed recall of word lists [<xref ref-type="bibr" rid="B12">12</xref>] and associative learning [<xref ref-type="bibr" rid="B13">13</xref>]), semantic memory [<xref ref-type="bibr" rid="B14">14</xref>], attention processing [<xref ref-type="bibr" rid="B15">15</xref>] and mental speed can consistently predict which patients will develop dementia. Conversely, in a retrospective study of people with MCI who later developed Alzheimer's dementia, verbal and visual memory, associative learning, vocabulary, executive function and other verbal tests of general intelligence were impaired at baseline [<xref ref-type="bibr" rid="B16">16</xref>]. Such tests should be administered by trained personnel and occupational physicians may need training in such assessments.</p></sec><sec><title>Falls and injuries at the workplace</title><p>Falls and injuries are common in the workplace but for older persons, they are associated with greater morbidity and mortality [<xref ref-type="bibr" rid="B17">17</xref>,<xref ref-type="bibr" rid="B18">18</xref>]. Slips, trips and falls are more common among older workers [<xref ref-type="bibr" rid="B19">19</xref>] and the resulting occupational injuries are more likely to result in hospitalization [<xref ref-type="bibr" rid="B20">20</xref>], fatalities [<xref ref-type="bibr" rid="B21">21</xref>] and fractures, particularly among older women [<xref ref-type="bibr" rid="B22">22</xref>]. However, falls in older persons are different from the younger population because there is a higher prevalence of medical problems that predispose older persons to falls. Examples of such medical problems that increase the risk of falls and injuries include strokes, dementia, cataracts, age-related macular degeneration, Stokes-Adam attacks from cardiac arrhythmias, vertebro-basilar insufficiency from cervical spondylosis, anaemia, medications with anti-cholinergic properties (e.g. anti-histamines, tricyclic anti-depressants) and postural hypotension from anti-hypertensives or dehydration.</p><p>When an older worker falls often, there is a need to move beyond treating injuries and improving workplace safety and towards a thorough assessment of the older worker to ascertain why a previously well worker is now sustaining falls and injuries at the workplace. There have been few published studies on the assessment of risk factors for falls among older workers at the workplace. Evidence from e geriatric medicine literature has consistently shown that multi-factorial assessment for falls risk factors, followed by interventions targeted at identified risk factors, have been effective in preventing further falls [<xref ref-type="bibr" rid="B23">23</xref>-<xref ref-type="bibr" rid="B25">25</xref>]. Such targeted assessment and management strategies have been found by a Cochrane Database Systematic Review to reduce occurrence of falls among older persons in the community by 25 to 39% [<xref ref-type="bibr" rid="B26">26</xref>]. Specific recommendations for fall risk factor assessment are summarized in Table <xref ref-type="table" rid="T2">2</xref>.</p><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Recommended Components of a Clinical Assessment and Management of Older Persons with Previous Falls (Adapted from Tinetti [27])</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left"><bold>Risk Factor</bold></td><td align="left"><bold>Management</bold></td></tr></thead><tbody><tr><td align="left">Circumstances of previous falls</td><td align="left">Changes in environment to reduce the likelihood of recurrent falls.</td></tr><tr><td align="left">Medication use<break/>- High risk medications (e.g. benzodiazepines, sedatives, neuroleptics, anti-depressants, anti-convulsants, Class 1A anti-arrhythmics)<break/>- Polypharmacy (4 or more medications)</td><td align="left">Review and reduction of medications</td></tr><tr><td align="left">Vision<break/>- Acuity <20/60<break/>- Decreased depth perception<break/>- Decreased contrast sensitivity<break/>- Cataracts</td><td align="left">- Ample lighting<break/>- Avoidance of multifocal glasses while walking<break/>- Referral to ophthalmologist</td></tr><tr><td align="left">Postural blood pressure (after 5 mins in a supine position, immediately after standing and 2 mins after standing)<break/>- >20 mmHg or (>20%) drop in systolic pressure, with or without symptoms, either immediately or after 2 min of standing, is significant</td><td align="left">Diagnosis and treatment of underlying cause, if possible. Review and reduction of medications; modification of salt restriction, adequate hydration, pressure stockings; fludrocortisone therapy if above strategies fail</td></tr><tr><td align="left">Balance and gait<break/>- Patient's report or observed unsteadiness.<break/>- Impairment on brief assessment (e.g. Get-Up-And-Go test)</td><td align="left">Diagnosis and treatment of underlying cause, if possible. Review and reduction of medications; referral to physical therapist for assistive devices and gait, balance and strength training</td></tr><tr><td align="left">Targeted neurological examination<break/>- Impaired proprioception<break/>- Impaired cognition<break/>- Decreased muscle strength</td><td align="left">Diagnosis and treatment of underlying cause, if possible; increase proprioceptive input (e.g. with assistive device or appropriate footwear that encases the foot and has a low heel and thin sole); review and reduction of medications; referral to physical therapist for assistive devices and gait, balance and strength training</td></tr><tr><td align="left">Targeted musculoskeletal examination<break/>- examination of legs<break/>- examination of feet</td><td align="left">Diagnosis and treatment of underlying cause, if possible; referral to physical therapist for assistive devices and gait, balance and strength training; use appropriate footwear, referral to podiatrist</td></tr><tr><td align="left">Targeted cardiovascular examination<break/>- Syncope<break/>- Arrhythmia</td><td align="left">Diagnosis and treatment of underlying cause, if possible; referral to cardiologist</td></tr></tbody></table></table-wrap><p>To date, there is no randomized control trial to determine effectiveness of interventional strategies to reduce the occurrence of falls among older persons in the workplace, so occupational physicians may need to turn to past studies on older persons in the community. Successful interventions to reduce falls include review and possible reduction of medications, balance and gait training, muscle-strengthening exercises, evaluation and strategies to reduce postural hypotension and targeted cardiovascular assessment and treatment. (Table <xref ref-type="table" rid="T2">2</xref>)</p><p>The role of laboratory testing and other investigations in fall assessment has not been well studied [<xref ref-type="bibr" rid="B27">27</xref>]. Laboratory tests that may be reasonable in the assessment of an older worker who has fallen include a complete blood count (to detect anaemia or a raised total white count suggesting a sub-clinical infection), serum electrolytes, glucose, vitamin B12, blood urea nitrogen and creatinine (to detect serum abnormalities that can cause impaired judgement or muscle weakness) and thyroid function (to detect hypothyroidism which can cause confusion and muscle weakness). In occupational settings with exposure to neurotoxins that can cause cognitive impairment, neuropathy and muscle weakness, such as metals (e.g. arsenic, lead, manganese), solvents (e.g. carbon disulphide, n-hexane and methyl-n-butyl ketones) and pesticides (e.g. organochlorine and organophosphate compounds), screening for these chemicals would be vital. Neuro-imaging is only needed when there is history of head injury with loss of consciousness, focal neurological findings on physical examination or when a central nervous system process is suspected from history or examination.</p><p>More studies are needed to determine if the risk factors for falls among older workers are similar to older persons in the community. However, until more information is known, an older worker who falls, whether at work or not, deserves a full fall risk factor assessment and appropriate intervention to improve workplace safety and maintain employability.</p></sec><sec><title>Conclusion</title><p>The future increase in numbers and age of older persons in the workplace will impact the practice of occupational medicine. To better manage these older workers, occupational physicians may increasingly need to draw upon the principles and experience of geriatric medicine. Mild cognitive impairment and falls in the workplace are two examples of syndromes associated with ageing that can have impact to the occupational health of older workers. Further research into the occupational health problems of older workers is also needed.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>GCHK and DK conceived and drafted the manuscript. Both authors read and approved the final manuscript.</p></sec>
Stress analysis in a layered aortic arch model under pulsatile blood flow	<sec><title>Background</title><p>Many cardiovascular diseases, such as aortic dissection, frequently occur on the aortic arch and fluid-structure interactions play an important role in the cardiovascular system. Mechanical stress is crucial in the functioning of the cardiovascular system; therefore, stress analysis is a useful tool for understanding vascular pathophysiology. The present study is concerned with the stress distribution in a layered aortic arch model with interaction between pulsatile flow and the wall of the blood vessel.</p></sec><sec sec-type="methods"><title>Methods</title><p>A three-dimensional (3D) layered aortic arch model was constructed based on the aortic wall structure and arch shape. The complex mechanical interaction between pulsatile blood flow and wall dynamics in the aortic arch model was simulated by means of computational loose coupling fluid-structure interaction analyses.</p></sec><sec><title>Results</title><p>The results showed the variations of mechanical stress along the outer wall of the arch during the cardiac cycle. Variations of circumferential stress are very similar to variations of pressure. Composite stress in the aortic wall plane is high at the ascending portion of the arch and along the top of the arch, and is higher in the media than in the intima and adventitia across the wall thickness.</p></sec><sec><title>Conclusion</title><p>Our analysis indicates that circumferential stress in the aortic wall is directly associated with blood pressure, supporting the clinical importance of blood pressure control. High stress in the aortic wall could be a risk factor in aortic dissections. Our numerical layered aortic model may prove useful for biomechanical analyses and for studying the pathogeneses of aortic dissection.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Gao</surname><given-names>Feng</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Watanabe</surname><given-names>Masahiro</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" corresp="yes" contrib-type="author"><name><surname>Matsuzawa</surname><given-names>Teruo</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib>	BioMedical Engineering OnLine	<sec><title>Background</title><p>The aorta is the main blood artery that delivers blood from the left ventricle of the heart to the rest of the body, and many cardiovascular diseases, such as aortic dissection, often occur on the aortic arch. It has been well established that many diseases are closely associated with the flow conditions in the blood vessels [<xref ref-type="bibr" rid="B1">1</xref>] and the blood flow in the aortic arch has been widely studied in the past [<xref ref-type="bibr" rid="B2">2</xref>-<xref ref-type="bibr" rid="B4">4</xref>].</p><p>Fluid-structure interactions play an important role in the cardiovascular system. There have been many recent studies of fluid-structure interaction in the aortic valve [<xref ref-type="bibr" rid="B5">5</xref>], aortic aneurysm [<xref ref-type="bibr" rid="B6">6</xref>], and in stented aneurysm models [<xref ref-type="bibr" rid="B7">7</xref>]. However the interaction between a pulsatile blood flow and the aortic wall in an aortic arch model has not yet been studied.</p><p>From the mechanical point of view, ruptures appear if the stresses acting on the wall rise above the ultimate value for the aorta wall tissue [<xref ref-type="bibr" rid="B6">6</xref>]. Mechanical stress plays a crucial role in the functioning of the cardiovascular system; therefore, stress analysis is a useful tool for understanding vascular pathophysiology [<xref ref-type="bibr" rid="B8">8</xref>]. Thubrikar et al. [<xref ref-type="bibr" rid="B9">9</xref>] used finite element analysis to determine the stresses in an aneurysm of the aorta; they found that longitudinal stress in the bulb is the only stress that increases significantly and could be responsible for the tear in an aortic dissection. Stress analysis of the aortic arch was implemented to study the effects of aortic motion on aortic dissection [<xref ref-type="bibr" rid="B10">10</xref>].</p><p>The present study is concerned with the stress distribution in a layered aortic arch model with interaction between a pulsatile flow and the wall of the blood vessel. Circumferential and longitudinal stress, as well as composite stress in the wall plane, is presented.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Geometry and wall properties</title><p>The radius of the arch was set at 30 mm [<xref ref-type="bibr" rid="B11">11</xref>] and the diameter of the vessel was assumed to be uniform (24 mm). The average diameter of the aorta is 20–25 mm [<xref ref-type="bibr" rid="B12">12</xref>]. The thickness of the whole wall was chosen to be 2 mm according to the characteristics of various types of blood vessels in [<xref ref-type="bibr" rid="B13">13</xref>]. The arch angle <bold><italic>a </italic></bold>is measured at the circumference of the median longitudinal cross-section, and ranges from 0° and 180° (Fig. <xref ref-type="fig" rid="F1">1</xref>), so that the wall position of the arch, as denoted by the arch angle, can be discussed.</p><fig position="float" id="F1"><label>Figure 1</label><caption><p><bold>The finite element model of the aortic arch</bold>. The branches of the arch were neglected as a first order approximation. Angle <bold><italic>a </italic></bold>represents the wall position in the median longitudinal cross-section. The inset shows the thickness of each layer.</p></caption><graphic xlink:href="1475-925X-5-25-1"/></fig><p>An average thickness ratio of intima/media/adventitia of 13/56/31 for arteries was observed in Schulze-Bauer's studies [<xref ref-type="bibr" rid="B14">14</xref>]. The thickness ratio of media/adventitia is 2/1 in the computational model for the arterial wall presented by Driessen et al. [<xref ref-type="bibr" rid="B15">15</xref>]. In this three-layered wall model, the intima/media/adventitia thickness ratio was set to 1/6/3. Therefore, the thicknesses of the intima, media, and adventitia were <bold><italic>t</italic></bold><sub><bold><italic>i </italic></bold></sub>= 0.2 mm, <bold><italic>t</italic></bold><sub><bold><italic>m </italic></bold></sub>= 1.2 mm, and <bold><italic>t</italic></bold><sub><bold><italic>a </italic></bold></sub>= 0.6 mm, respectively.</p><p>In Mosora's experiments [<xref ref-type="bibr" rid="B16">16</xref>], the Young's modulus of the thoracic ascending aorta was 2 MPa to 6.5 MPA. In vivo, the surrounding connective tissue and muscle can support the aortic arch. In this study, the aortic arch is a self-supporting structure. The maximum <bold><italic>E </italic></bold>= 6.5MPa was chosen for the Young's modulus of the aorta wall. Xie et al. [<xref ref-type="bibr" rid="B17">17</xref>] performed bending experiments and showed the Young's modulus of the inner layer (intima and media) was three to four times larger than that of the outer layer (adventitia), and we can deduce from Fischer's experimental data [<xref ref-type="bibr" rid="B18">18</xref>] that the Young's modulus of the intima is smaller than that of the media. In the present study, the Young's modulus of the media is assumed to be three times larger than that of the adventitia and the intima. Since the mean Young's modulus of the vessel wall across whole wall volume is invariable, we assume that the Young's modulus of each layer is in inverse proportion to the area of the layer in the cross section. Based on the area equation, the Young's modulus of the intima, media, and adventitia layers is 2.98MPa, 8.95MPa, and 2.98MPa respectively. Fig. <xref ref-type="fig" rid="F2">2</xref> shows the three-layered aortic wall.</p><fig position="float" id="F2"><label>Figure 2</label><caption><p><bold>Three-layered models</bold>. The Young's modulus of layers is shown.</p></caption><graphic xlink:href="1475-925X-5-25-2"/></fig></sec><sec><title>Fluid-structure interaction</title><p>Fig. <xref ref-type="fig" rid="F1">1</xref> shows the circulatory system; the blowup is the FEM model of the aorta. The finite element analysis model of the aorta was made of 38400 eight-node brick elements for the solid domain and 26880 eight-node brick elements for the fluid domain. There are 20 elements through the thickness of the aorta wall.</p><p>In a fluid-structure interaction problem, fluid affects structure and structure also affects fluid. In this study, the procedure is based on the loose coupling of three fields of problems: the flow, the elastic body, and the mesh movement – that is, CFD (computational fluid dynamics), CSD (computational structural dynamics), and CMD (computational mesh dynamics) procedures [<xref ref-type="bibr" rid="B19">19</xref>].</p><p>The FSI (Fluid Structure Interaction) algorithm treats the equations that solve the fluid problem, the structural problem, and the re-meshing problem by a staggered approach; that is to say, the algorithm solves the equations in sequence. The overall computational procedure adopted to solve FSI problems is depicted in Fig. <xref ref-type="fig" rid="F3">3</xref>. Operatively, a ALE (Arbitrary Lagrangian Eulerian) algorithm is used which seeks, at each step increment, the convergence of the three blocks of equations, fluid (CFD), solid (CSD) and mesh movements (CMD). These must then converge altogether before a new step is initiated. The detailed equations are described in the Appendix.</p><fig position="float" id="F3"><label>Figure 3</label><caption><p>Strategy of the computational procedure.</p></caption><graphic xlink:href="1475-925X-5-25-3"/></fig><p>The code <italic>Fidap </italic>(Fluent Inc., Lebanon, NH) was used to carry out the simulation. Bar-Yoseph et al. [<xref ref-type="bibr" rid="B20">20</xref>] gave many examples to demonstrate the validity of the method used in this study. Marzo et al. [<xref ref-type="bibr" rid="B21">21</xref>] also used the same method to do a FSI simulation and got excellent results.</p><p>Typically, 160–190 iterations were required per time step to reduce the residuals. The numerical convergence is a relative change in the solution from one iteration to the next of less than 0.0001.</p><p>Circumferential stress is the normal stress that points along the tangent direction to a cross-section circle in the cross-section plane. Longitudinal stress is the normal stress that points in the direction of the axis of the vessel. Composite stress is the composition of the circumferential stress vector and the longitudinal stress vector, which is the resultant stress in the wall plane.</p></sec><sec><title>Boundary condition and properties of fluid</title><p>The boundary conditions are time-dependent. At the aortic inlet, a flat flow velocity profile was used together with a pulsatile waveform based on experimental data reported by Pedley [<xref ref-type="bibr" rid="B22">22</xref>]. This waveform is shown in Fig. <xref ref-type="fig" rid="F4">4</xref>. The assumption of a flat velocity profile at the aortic inlet is justified by in vivo measurements using hot film anemometry on various animal models, which have demonstrated that the velocity profile distal to the aortic valve are relatively flat [<xref ref-type="bibr" rid="B23">23</xref>]. The Reynolds number for blood flow is approximately 4000 in the aorta [<xref ref-type="bibr" rid="B24">24</xref>]. In our calculations, the Reynolds number is fixed at Re = 4000 based on the inlet velocity at peak systole of the cardiac cycle. There is pressure both at the inside and the outside of the blood vessel and the pressure difference between the inside and the outside of the blood vessel tends to deform the blood vessel [<xref ref-type="bibr" rid="B24">24</xref>]. This study was concerned mainly with the effect of the relative pressure on the wall stress. At the aortic outlet, a zero pressure condition was applied. The surface of the inlet was fixed. The outer edge of the outlet was constrained in the axial direction and permitted to move in the other directions. There was no constraint for the other parts of the aorta model.</p><fig position="float" id="F4"><label>Figure 4</label><caption><p><bold>Inlet velocity profile</bold>. t is the time, T is the total time in one cycle. For the final run, the cardiac cycle starts at A and ends at B.</p></caption><graphic xlink:href="1475-925X-5-25-4"/></fig><p>The simulation reached nearly steady-state oscillation after approximately the third cycle. The fourth cycle was used as the final periodic solution and it is presented in this paper.</p><p>The fluid is Newtonian with a density of 1050 kg/m<sup>3 </sup>and a viscosity of 0.0035 Pa s. Blood is essentially a suspension of erythrocytes in plasma and shows anomalous viscous properties at low velocities. However, the Newtonian assumption is considered acceptable since minor differences in the basic flow characteristics are introduced through the non-Newtonian hypothesis [<xref ref-type="bibr" rid="B25">25</xref>].</p></sec></sec><sec><title>Results</title><sec><title>Result pressure</title><p>The pressure waveform at the inlet corresponding to the inlet velocity was extracted from result data (Fig. <xref ref-type="fig" rid="F5">5</xref>). We see the pressure rises above zero (1330Pa) at the start of systole during the accelerative phase and then becomes negative at peak systole, reaching a minimum value (-1100Pa) at the max reverse flow. There is a transient increase in pressure after the max reversed flow phase. A negative pressure represents the decelerative phase of the aortic flow.</p><fig position="float" id="F5"><label>Figure 5</label><caption><p><bold>Pressure waveform at inlet corresponding to inlet velocity</bold>.</p></caption><graphic xlink:href="1475-925X-5-25-5"/></fig><p>Fig. <xref ref-type="fig" rid="F6">6</xref> shows the variations of pressure along the arch at centerline during the cardiac cycle. The variations of pressure as a function of time are similar to the results shown in Fig. <xref ref-type="fig" rid="F5">5</xref>. Positive pressure values decrease along the arch portion.</p><fig position="float" id="F6"><label>Figure 6</label><caption><p><bold>Pressure at centerline as a function of the function of arch angle <italic>a </italic>and time t/T</bold>. End diastole is at t/T = 4.0. Angle <bold><italic>a </italic></bold>represents the wall position in the arch portion.</p></caption><graphic xlink:href="1475-925X-5-25-6"/></fig></sec><sec><title>Variations of stresses along arch during cardiac cycle</title><p>Tears and dissections usually involve the outer wall of the arch [<xref ref-type="bibr" rid="B26">26</xref>], so the stress distribution in the outer wall was investigated. Fig. <xref ref-type="fig" rid="F7">7</xref> presents the variations of circumferential stress and longitudinal stress on the outer wall along the arch portion during the cardiac cycle. The stresses are averaged across the aortic wall thickness.</p><fig position="float" id="F7"><label>Figure 7</label><caption><p><bold>Variation of circumferential and longitudinal stress at arch outer wall as a function of the function of arch angle a and time t/T</bold>. (a) Circumferential stress variation. (b) Longitudinal stress variation. End diastole is at t/T = 4.0. Angle a represents the wall position in the arch portion.</p></caption><graphic xlink:href="1475-925X-5-25-7"/></fig><p>The circumferential stress on the outer wall initially increases at the start of systole and then decreases below zero during the deceleration phase. After the reverse flow, there is a transient increase in mean circumferential stress. The circumferential stress is higher in the ascending portion than in the descending portion. The variation of the mean circumferential stress in the outer wall of the arch is similar to the variation of the pressure at the arch outer wall.</p><p>Longitudinal stress initially increases at the acceleration phase of systole, then deceases, and then increases a little at the reverse flow. The high stress regions along the arch are at the entrance to the ascending portion, the top of the arch, and the distal end of the arch.</p><p>Since the stress in the acceleration phase is high, a transient time t/T = 0.12 in the middle of the acceleration phase was selected to characterize the variations of stresses along the arch in detail. The circumferential stress and the longitudinal stress, as well as the composite stress along the arch at selected transient times, are shown in Fig. <xref ref-type="fig" rid="F8">8</xref>.</p><fig position="float" id="F8"><label>Figure 8</label><caption><p><bold>Stress distribution on the outer wall along the arch portion</bold>. (a) Circumferential stress distribution. (b) Longitudinal stress distribution. (c) Composite stress distribution.</p></caption><graphic xlink:href="1475-925X-5-25-8"/></fig><p>The circumferential stress decreases along the arch. The longitudinal stress along the arch gets peak values at the entrance to the ascending portion, the top of the arch, and the distal end of arch. The composite stress decreases, with a peak at the top of the arch along the arch portion.</p></sec><sec><title>Variations of stresses across the wall</title><p>Four positions, <bold><italic>a </italic></bold>= 22.5°, <bold><italic>a </italic></bold>= 67.5°, <bold><italic>a </italic></bold>= 112.5°, and <bold><italic>a </italic></bold>= 157.5°, were selected in the arch portion to illustrate the variation of stresses across the wall.</p><p>The variations of circumferential, longitudinal, and composite stress across the three-layer model wall at four positions at systolic acceleration (t/T = 3.12) are shown in Fig. <xref ref-type="fig" rid="F9">9</xref>. The stress is much higher in the media layer than in the intima and adventitia layers.</p><fig position="float" id="F9"><label>Figure 9</label><caption><p>Variation of the stresses across the wall at four positions (<italic>a </italic>= 22.5°, <italic>a </italic>= 67.5°, <italic>a </italic>= 112.5°, and <italic>a </italic>= 157.5°) at selected time of systolic acceleration (t/T = 3.12).</p></caption><graphic xlink:href="1475-925X-5-25-9"/></fig></sec></sec><sec><title>Discussion</title><p>We have simulated the complex mechanical interaction between blood flow and wall dynamics by means of computational coupled fluid-structure interaction analyses and shown the results for mechanical stress in the layered aortic arch model. The composite stress in the aortic wall plane is high at the ascending portion and along the top of the arch and is higher in the media than in the intima and adventitia across the wall thickness.</p><p>Circumferential stress was considered an important parameter for mechanosensitive receptors. Under a pressure <italic>P</italic>, the circumferential and longitudinal stresses in a cylinder of thickness <italic>t </italic>and radius <italic>R </italic>are <italic>PR/t </italic>and <italic>PR/2t</italic>, respectively [<xref ref-type="bibr" rid="B9">9</xref>]. Our resulting pressure waveform at the inlet in the present study agrees with the results in Tyszka's [<xref ref-type="bibr" rid="B27">27</xref>] paper. This demonstrates that circumferential stress depends on the systolic blood pressure. Furthermore, this explains the fact that 70–90% of patients with aortic dissection have high blood pressure [<xref ref-type="bibr" rid="B26">26</xref>,<xref ref-type="bibr" rid="B28">28</xref>-<xref ref-type="bibr" rid="B30">30</xref>] and supports the clinical importance of blood pressure control in reducing the risk of tearing and rupturing in aortic dissection.</p><p>The pressure P decreased along the arch portion, and the radius R and thickness t did not change very much. However, the longitudinal stress depicted in Fig. <xref ref-type="fig" rid="F8">8(b)</xref> along the arch gets peak values at the entrance to the ascending portion, the top of the arch, and the distal end of arch, so the variation would be unpredictable using <italic>PR/2t; </italic>thus, the shape of the arch and fluid force may play crucial roles in aorta mechanics. Similar work on the stress distribution on the aortic arch has been presented by Beller et al. [<xref ref-type="bibr" rid="B10">10</xref>], in which circumferential and longitudinal stress were found to be maximum at the top of the arch. Unfortunately, they only used a load condition of 120 mmHg luminal pressure and the pulsatile flow condition was not considered. In recognizing the influential effects of pulsatile flow in the aortic mechanism, a more realistic wall stress distribution was simulated in our coupled model. Bellers et al. [<xref ref-type="bibr" rid="B10">10</xref>] included branch vessels in their model. Our study was concerned mainly with the effects of the aortic arch on stress distribution. Therefore, the branches along the top of the aortic arch were not included in the model.</p><p>Thubrikar et al. [<xref ref-type="bibr" rid="B9">9</xref>] have proposed that longitudinal stress could be responsible for transverse tears in the aortic dissection. Roberts [<xref ref-type="bibr" rid="B26">26</xref>] reported that in 62% of patients the tear is located in the ascending aorta, usually about 2 cm cephalad to the sinotubular junction. This means the location is about <bold><italic>a </italic></bold>= 0°, at the entrance to the ascending portion of the arch. After the ascending aorta, about 20% of tears are at the aortic isthmus portion [<xref ref-type="bibr" rid="B26">26</xref>], which is near the top of the arch. In the present study, the longitudinal stress achieved peak values at the entrance to the ascending portion and the top of the arch. Yet, the longitudinal stress also peaked in the distal end of the arch, and tears do not often occur at this location.</p><p>Composite stress was high at the entrance to the ascending portion and at the top of the arch; the value at the entrance to the ascending portion was a little higher. The location of this high composite stress value is consistent with the tear locations identified in the Roberts report [<xref ref-type="bibr" rid="B26">26</xref>]. This implies that both circumferential and longitudinal stress may contribute to the tear in an aortic dissection and high composite stress in the aortic wall could be a risk factor for tears in aortic dissections.</p><p>The stress variation across wall thickness was due to the non-homogeneous wall properties. Our results showed that the stress was highest in the media across the aortic wall thickness and this agreed with the results of Maltzahn et al. [<xref ref-type="bibr" rid="B31">31</xref>], that the media was subject to much higher stresses. These results might partially explain why the location of the dissection is in the aortic media [<xref ref-type="bibr" rid="B26">26</xref>].</p><p>Previous studies [<xref ref-type="bibr" rid="B32">32</xref>,<xref ref-type="bibr" rid="B33">33</xref>] examined the properties of the aorta and showed the aortic wall is anisotropic. The smooth muscle cells in aortic wall were torn loose from their attachments to each other and to the adjacent elastin [<xref ref-type="bibr" rid="B33">33</xref>], so the dissection may propagate rather than tearing back through the intima or out through the adventitia. Residual stresses can affect stress distribution through the arterial wall. Circumferential stress distributions including residual stress have shown decreased inner wall stress [<xref ref-type="bibr" rid="B34">34</xref>], which may be induced by the negative value of residual stress in inner side of wall [<xref ref-type="bibr" rid="B17">17</xref>]. The residual stress is high in the outer side of media and positive at adventitia [<xref ref-type="bibr" rid="B17">17</xref>] and this may increase the stress at the outer side of media and at adventitia.</p><p>The aortic arch model used here only partially simulated the real situation because it neglected the effect of the branches of the arch [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B4">4</xref>] and the aortic arch's non-planarity [<xref ref-type="bibr" rid="B35">35</xref>]. Because this was an initial attempt; the model was purposely kept simple in order to give an insight into the stress distribution on the aortic arch with interaction between a pulsatile flow and the wall. Furthermore, more than 80% of tears and dissections are at the aorta, not at branches of the arch [<xref ref-type="bibr" rid="B26">26</xref>,<xref ref-type="bibr" rid="B36">36</xref>]. In the future investigations, the branches will be added to the model. The mechanical properties of arteries have been reported extensively in the literature. The type of nonlinearity is a consequence of the curvature of the strain-stress function, which shows that an artery becomes stiffer as the distending pressure increases. Horsten et al. [<xref ref-type="bibr" rid="B37">37</xref>] showed that elasticity dominates the nonlinear mechanical properties of arterial tissues. In numerical simulation of the biomechanics in arteries [<xref ref-type="bibr" rid="B6">6</xref>-<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B10">10</xref>], the wall was assumed to be an elastic constitutive model. Of course, a non-linear constitutive model could provide more biological aspects of the biomechanics. In future work we plan to implement the nonlinear properties of aortic wall within the code by means of user-subroutines.</p></sec><sec><title>Conclusion</title><p>In summary, our analysis indicates that circumferential stress in the aortic wall is directly associated with blood pressure, supporting the clinical importance of blood pressure control. High stress in the aortic wall could be a risk factor for tearing in aortic dissections. This numerical layered aortic model may prove useful for biomechanical analyses and for studying the pathogeneses of aortic dissection.</p></sec><sec><title>Appendix</title><p>Detailed equations of FSI algorithm</p><p>1. Computational Fluid Dynamics (CFD)</p><p>The ALE (Arbitrary Lagrangian Eulerian) form of the Navier-Stokes equations are used to solve for fluid flow for FSI problems. After spatial discretization by the finite element method, the Navier-Stokes equations are expressed in the ALE formulation:</p><p><bold>Ma </bold>+ <bold>N</bold>(<bold>v </bold>- <inline-graphic xlink:href="1475-925X-5-25-i1.gif"/>)<bold>v </bold>+ <bold>Dv </bold>- <bold>Cp </bold>= <bold>f </bold>    in <sup><italic>F </italic></sup>Ω(<italic>t</italic>)     (1)</p><p><bold>C</bold><sup><bold>T </bold></sup><bold>v </bold>    in <sup><italic>F </italic></sup>Ω(<italic>t</italic>)     (2)</p><p>where <bold>M </bold>represents the fluid mass matrix; <bold>N</bold>, <bold>D</bold>, and <bold>C </bold>are, respectively, the convective, diffusive, and divergence matrices; <bold>f </bold>is an external body force; vectors <bold>a</bold>, <bold>v</bold>, and <bold>p </bold>contain the unknown values of acceleration, velocity, and pressure, respectively; <inline-graphic xlink:href="1475-925X-5-25-i1.gif"/> is the mesh velocity, calculated using Eq. (13) in CMD; and <sup><italic>F </italic></sup>Ω(<italic>t</italic>) is the moving spatial domain upon which the fluid is described.</p><p>The following compatibility conditions are imposed on the interface between fluid and structure:</p><p><inline-graphic xlink:href="1475-925X-5-25-i2.gif"/></p><p>where (<sup><italic>F</italic></sup>) indicates values related to nodes placed in the fluid; (<sup><italic>S</italic></sup>) indicates values related to the nodes placed in the structure; and <sup><italic>I</italic></sup>Γ(<italic>t</italic>) is the interface between fluid and structure at time <italic>t</italic>. <sup><italic>s </italic></sup><bold>v </bold>and <sup><italic>s</italic></sup><bold>a </bold>are calculated at the previous iteration loop in CSD.</p><p>The fluid velocity and pressure field are solved after initial conditions, boundary conditions and compatibility conditions are imposed on the fluid domain. For calculating the traction at the interface between fluid and structure vectors <bold>a</bold>, <bold>v</bold>, and <bold>f </bold>are decomposed into:</p><p><inline-graphic xlink:href="1475-925X-5-25-i3.gif"/></p><p>where (<sup><italic>U</italic></sup>) indicates values related to nodes on the velocity boundary, (<sup><italic>I</italic></sup>) indicates values related to the fluid nodes on the interface between fluid and structure, and the symbol (<sup>-</sup>) denotes the prescribed values. The load applied by the fluid on the structure along the interface between fluid and structure is obtained according to partitioned Eq. (1):</p><p><inline-graphic xlink:href="1475-925X-5-25-i4.gif"/></p><p>where <sup><italic>I </italic></sup><bold>f </bold>is the external force vector at the interface applied by the structure on the fluid; (<sup><italic>IF</italic></sup>) indicates values related to interface nodes related to the fluid; (<sup><italic>IU</italic></sup>) indicates values related to the interface nodes related to the velocity boundary; (<sup><italic>II</italic></sup>) indicates values related to the interface nodes only related to the interface; <inline-graphic xlink:href="1475-925X-5-25-i5.gif"/> is the traction applied by the fluid on the structure along the interface; <bold>M</bold>, <bold>N</bold>, <bold>D</bold>, <bold>C, </bold><bold>a</bold>, <bold>v</bold>, and <bold>p </bold>are known values solved in the previous process in this equation. The traction <inline-graphic xlink:href="1475-925X-5-25-i5.gif"/> is imposed on the structure in CSD.</p><p>2. Computational Structural Dynamics (CSD)</p><p>The overall structural behavior is described by the matrix equation:</p><p><sup><italic>s </italic></sup><bold>M </bold><sup><italic>s </italic></sup><bold>a</bold>+<sup><italic>s </italic></sup><bold>K </bold><sup><italic>s </italic></sup><bold>d</bold>= <inline-graphic xlink:href="1475-925X-5-25-i6.gif"/>+ <inline-graphic xlink:href="1475-925X-5-25-i5.gif"/> - <sup>σ </sup><bold>f </bold>    in <sup><italic>s</italic></sup>Ω(<italic>t</italic>)     (6)</p><p>where <sup><italic>S </italic></sup><bold>M </bold>and <sup><italic>S </italic></sup><bold>K </bold>are the structural mass and nonlinear stiffness matrices; <inline-graphic xlink:href="1475-925X-5-25-i6.gif"/> is the external body force vector; <inline-graphic xlink:href="1475-925X-5-25-i5.gif"/>, calculated using the Eq. (5) in CFD, is the traction applied by the fluid on the structure along the interface; <sup>σ </sup><bold>f </bold>is the force due to the internal stresses at the most recently calculated configuration; <sup><italic>S </italic></sup><bold>d </bold>is the vector of increments in the nodal point displacement; <sup><italic>S </italic></sup><bold>a </bold>is the vector of nodal point acceleration; <sup><italic>s </italic></sup>Ω(<italic>t</italic>) is the structure domain at time <italic>t</italic>. <inline-graphic xlink:href="1475-925X-5-25-i6.gif"/> and <sup>σ </sup><bold>f </bold>are equal to zero in this study. The equation of motion can be integrated and the displacement, velocity, and acceleration vectors can be calculated.</p><p>The displacement at the interface between fluid and structure can be calculated and the surface location is updated:</p><p><sup><italic>I </italic></sup><bold>d </bold>= <sup><italic>s </italic></sup><bold>d </bold>    on <sup><italic>I </italic></sup>Γ(<italic>t</italic>)     (7)</p><p>The mesh displacement can be obtained at the interface between fluid and structure:</p><p><inline-graphic xlink:href="1475-925X-5-25-i7.gif"/> = <sup><italic>I </italic></sup><bold>d </bold>    on <sup><italic>I </italic></sup>Γ(<italic>t</italic>)     (8)</p><p>The mesh displacement <inline-graphic xlink:href="1475-925X-5-25-i7.gif"/> is imposed at the interface between fluid and structure in CMD.</p><p>3. Computational mesh dynamics (CMD)</p><p>An elasticity-based meshing algorithm was employed to solve the remeshing problem. The mesh was treated as a pseudo-elastostatic medium and the same algorithm Eq. (6), which was used to calculate the displacement of the structural body, was employed. In the algorithm, the mesh is modelled as a pseudo-elastic structure the deformation of which is based on the boundary condition resulting form Eq. (8) of the structural problem. Such displacements are obtained from</p><p><inline-graphic xlink:href="1475-925X-5-25-i8.gif"/></p><p>where <inline-graphic xlink:href="1475-925X-5-25-i9.gif"/> is the stiffness matrix, and <inline-graphic xlink:href="1475-925X-5-25-i10.gif"/> is the mesh displacement vector defined by</p><p><inline-graphic xlink:href="1475-925X-5-25-i11.gif"/></p><p>where <inline-graphic xlink:href="1475-925X-5-25-i12.gif"/> s the mesh position vector, and <italic>t </italic><sub><italic>ref </italic></sub>is the reference time. Eq. (9) can be rewritten in a partitioned form as</p><p><inline-graphic xlink:href="1475-925X-5-25-i13.gif"/></p><p>On the interface between fluid and structure, the mesh displacement given by Eq. (8) from CSD is imposed. So we obtain</p><p><inline-graphic xlink:href="1475-925X-5-25-i14.gif"/></p><p>which shows that the fluid mesh motion is driven by the elastic body motion. This equation, together with the other boundary condition, was solved for the fluid mesh displacement.</p><p>The fluid mesh displacement was solved and the mesh geometry was updated. The mesh velocity field can be obtained:</p><p><inline-graphic xlink:href="1475-925X-5-25-i15.gif"/></p><p>The mesh velocity is imposed in CFD.</p></sec>
In-vivo coronary flow profiling based on biplane angiograms: influence of geometric simplifications on the three-dimensional reconstruction and wall shear stress calculation	<sec><title>Background</title><p>Clinical studies suggest that local wall shear stress (WSS) patterns modulate the site and the progression of atherosclerotic lesions. Computational fluid dynamics (CFD) methods based on in-vivo three-dimensional vessel reconstructions have recently been shown to provide prognostically relevant WSS data. This approach is, however, complex and time-consuming. Methodological simplifications are desirable in porting this approach from bench to bedside. The impact of such simplifications on the accuracy of geometry and wall shear stress calculations has to be investigated.</p></sec><sec sec-type="methods"><title>Methods</title><p>We investigated the influence of two methods of lumen reconstruction, assuming circular versus elliptical cross-sections and using different resolutions for the cross-section reconstructions along the vessel axis. Three right coronary arteries were used, of which one represented a normal coronary artery, one with "obstructive", and one with "dilated" coronary atherosclerosis. The vessel volume reconstruction was performed with three-dimensional (3D) data from a previously validated 3D angiographic reconstruction of vessel cross-sections and vessel axis.</p></sec><sec><title>Results</title><p>The difference between the two vessel volumes calculated using the two evaluated methods is less than 1 %. The difference, of the calculated pressure loss, was between 2.5% and 8.5% for the evaluated methods. The distributions of the WSS histograms were nearly identical and strongly cross-correlated (0.91–0.95). The good agreement of the results was confirmed by a Chi-square test.</p></sec><sec><title>Conclusion</title><p>A simplified approach to the reconstruction of coronary vessel lumina, using circular cross-sections and a reduced axial resolution of about 0.8 mm along the vessel axis, yields sufficiently accurate calculations of WSS.</p></sec>	<contrib id="A1" equal-contrib="yes" contrib-type="author"><name><surname>Wellnhofer</surname><given-names>Ernst</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" equal-contrib="yes" corresp="yes" contrib-type="author"><name><surname>Goubergrits</surname><given-names>Leonid</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Kertzscher</surname><given-names>Ulrich</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Affeld</surname><given-names>Klaus</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib>	BioMedical Engineering OnLine	<sec><title>Introduction</title><p>Based on the hypothesis that information on local wall shear stress (WSS) patterns has a prognostic value with respect to the progression and risk of coronary artery disease, in vivo profiling of the endothelial shear stress in coronary arteries has been performed recently in several studies [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B7">7</xref>]. Published serial invasive investigations in the last years support the prognostic impact of local WSS evaluations [<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B9">9</xref>]. Most of these studies were performed by three-dimensional (3D) reconstruction of coronary artery segments fusing intravascular ultrasound images (IVUS) and angiograms with subsequent numerical flow simulation studies. The IVUS is used because it provides detailed information of the circumferential endo-luminal border and also additional information of the vessel wall that is derived from imaging the local plaque. Numerical flow simulation is used since measurements of velocity profiles, and especially of wall shear stress distribution in the coronary arteries, are not feasible.</p><p>3D-reconstruction of coronary artery segments fusing IVUS-images and angiograms with subsequent numerical flow simulation studies is an invasive, expensive and time consuming approach that is limited to dedicated studies and rather small numbers of investigated vessel lesions. Further limitations of this method are size of the segments (> 1 mm), which may be assessed by the IVUS catheter. Thus the assessment of side branches and distal coronary arteries is not feasible. We do not know whether the costs of the accuracy of the reconstruction approach translate into the amount of additional clinical diagnostic impact. Simpler approaches may provide clinically relevant prognostic information [<xref ref-type="bibr" rid="B9">9</xref>]. Thus, research on simpler techniques is required. Furthermore, we would like to characterize coronary artery disease in the coronary trees as is shown in figures <xref ref-type="fig" rid="F2">2</xref> and <xref ref-type="fig" rid="F3">3</xref>. We propose to use 3D-reconstruction of coronary trees based on biplane angiograms with subsequent numerical flow simulations.</p><fig position="float" id="F1"><label>Figure 1</label><caption><p>Example of a pair of corresponding LAO and RAO projections. The three straight lines in each image show corresponding projection lines in the two projections. These lines are used to match the two projections. Numerated squares mark nodes used for segmentation and identification of vascular branches.</p></caption><graphic xlink:href="1475-925X-5-39-1"/></fig><fig position="float" id="F2"><label>Figure 2</label><caption><p>Wire representation of the three right coronary arteries investigated in our study. From left to right: normal right coronary artery, right coronary artery with "obstructive" atherosclerotic disease and right coronary artery with "dilated" atherosclerotic disease. Circular cross-sections are supposed. The reconstructions were done by use of the software Gambit™.</p></caption><graphic xlink:href="1475-925X-5-39-2"/></fig><fig position="float" id="F3"><label>Figure 3</label><caption><p>Geometry of the reconstructed endo-luminal surface of the normal coronary artery using Method 1 and the software SolidWorks™. A normal right coronary artery is presented.</p></caption><graphic xlink:href="1475-925X-5-39-3"/></fig><p>Coronary artery disease is clinically diagnosed by symptoms related to impaired myocardial perfusion and invasively diagnosed by detection of wall irregularities or local obstructions in selective angiograms (luminograms). The luminal contour is the net result of the encroachment of plaque into the vascular lumen and compensatory vessel wall remodeling [<xref ref-type="bibr" rid="B10">10</xref>,<xref ref-type="bibr" rid="B11">11</xref>]. Luminal remodeling is localized and preserves a circular lumen even in the majority of eccentric atherosclerotic lesions [<xref ref-type="bibr" rid="B12">12</xref>]. The irregular lumina occur rarely, generally at severely diseased sites. These advanced lesions imply a failure of local remodeling. This in turn suggests a loss of endothelial function and of WSS responsiveness [<xref ref-type="bibr" rid="B13">13</xref>]. Luminal geometry and flow determine wall shear stress. IVUS-data shows that wall thickness and wall composition have no direct impact on WSS. Thus, wall shear stress estimation from standard luminograms using fast semi-automatic geometric reconstructions and flow calculations should be feasible and may be an important step from bench to bedside in providing clinically relevant WSS-data in vivo.</p><p>The proposed approach implies simplifications of geometric reconstructions and model assumptions with respect to flow simulation in geometries reconstructed from biplane angiograms. A methodological investigation of the impact of these simplifications on WSS calculations using in vivo data is necessary. The goal of this paper is an assessment of the impact of two different time saving simplifications of geometric reconstruction on cycle-averaged WSS. Furthermore, a new method of WSS distribution characterization is proposed – analysis of WSS histograms. A further goal of this paper is to investigate the impact of reconstruction simplifications on the proposed method of the WSS characterization.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Data</title><p>Three distinct cases of coronary artery luminal geometries reconstructed from biplane patient angiograms were used: a normal right coronary artery (control), and right coronary arteries (RCA) with atherosclerosis with "dilated" versus "obstructive" remodeling. The concept of "dilated" versus "obstructive" coronary atherosclerosis was introduced by Schoenhagen et al. [<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B27">27</xref>]. The concept of Schoenhagen addresses the fact that remodeling is an important factor affecting the luminal width which is only loosely related to plaque growth. He says: "Traditionally, the development of coronary artery disease was described as a gradual growth of plaques within the intima of the vessel. The outer boundaries of the intima, the media and the external elastic membrane, were thought to be fixed in size. However, histologic studies demonstrated that certain plaques do not reduce luminal size because of expansion of the media and the external elastic membrane during atheroma development. This phenomenon of "arterial remodeling" was confirmed in necropsy specimens of human coronary arteries. [<xref ref-type="bibr" rid="B14">14</xref>]" Even though in his review Schoenhagen focuses on local remodeling assessed by IVUS and IVUS-specific definitions of remodeling, he sees a clinical analogy between positive remodelling and coronary ectasia. We apply his concept on atherosclerotic coronary ectasia ("dilated") and non-ectatic coronary artery disease ("obstructive"). The underlying hypothesis of our choice of these three particular coronary arteries is that vascular remodeling is intrinsically related to atherosclerotic inflammation and affects environments at multiple sites rather than localized foci. Thus profiles of WSS within whole segments or vessels might identify different patterns of remodeling associated with characteristic changes in the distribution of WSS. This is why three different vessels were used to study the impact of reconstruction simplifications on WSS characterization.</p></sec><sec><title>Three-dimensional reconstruction</title><p>Biplane angiograms (25 frames/s) had been acquired on cine-film by a standard biplane angiographic X-ray device (Philips DCI-System) during end diastole. Rotation and angulations of the C-arms, distances of image intensifiers to X-ray sources and size of image intensifiers had been recorded. Frame numbers were used to find the corresponding images in the left anterior oblique (LAO) and in the right anterior oblique (RAO) projections. These protocol data were used to estimate the three-dimensional geometry. A two-dimensional (2D) model was reconstructed for each projection by a combination of interactive topology marking and automatic vessel detection. Vessel bifurcations were manually identified and used for both segmentation and vessel detection (see figure <xref ref-type="fig" rid="F1">1</xref>). 2D data is organized in segments consisting of coordinates of centerline and related radii, defining both edge points of the vessel projection for the corresponding 2D models of LAO and RAO projections. The 3D reconstruction is calculated from 2D projections in the three-dimensional space estimated from protocol data. Each data set representing a particular 3D reconstruction segment consists of a discrete set of vector triplets that represent the 3D coordinates of the segment centerline and the two radii R1 and R2 obtained from two projections. For a more detailed description of the 2-D models and the 3-D reconstruction procedures, please refer to [15, 16]. The accuracy of the used reconstruction software was tested and validated in phantom studies with well defined geometries and were described elsewhere [<xref ref-type="bibr" rid="B17">17</xref>]. The diameter and volume measurements in this model were performed by three-dimensional calipers (spheres and generalized cones). The resulting accuracy yields an error for diameters of <3%. The inter- and intra-observer variability is <5%. In these phantom studies a reconstruction procedure using elliptical cross-sections was applied.</p><p>There are two simplifications suggested by these considerations that speed up the reconstruction and circumvent the reconstruction problems with vector triplets representing degenerated ellipses and non-orthogonal (tilted) and unevenly spaced cross-sections. The first one is a reduction of the spatial resolution by using only every second vector triplet for interpolation. The cross-sections defined by the original vector triplets are tilted, because the radius vectors are not orthogonal to the longitudinal axis of the vessel. Thus neighboring cross-sections may intersect and unevenly spaced tilted cross-sections occur. The analysis of the 3000 cross-sections of the coronary artery represented in figure <xref ref-type="fig" rid="F2">2</xref> left demonstrated that the mean angle between cross-section planes defined by vector triplets and longitudinal vessel axis was 90° ± 6.5° (SD). However, some cross-sections were angulated by less than 30°. A 3D caliper approach is necessary to construct cross-sections orthogonal to the longitudinal axis from the original vector triplets. As most of the cross-sections defined by vector triplets are nearly orthogonal, an approximate solution is to neglect one of two intersecting cross-sections or respective vector triplets. The axial resolution is only slightly reduced by this approach. The resulting models were defined as high fidelity models (HF). Two further resolution models were also studied – low fidelity models (LF) with a relative to HF models halved number of reconstructed cross-sections and double low fidelity model (DLF) with a relative to LF models halved number of cross-sections. The provided 3D reconstruction data defines only 4 points on the vessel cross-section. Interpolation algorithms are necessary to reconstruct surface and volume geometry and to construct a sufficiently fine grid of this vessel lumen for CFD study.</p><p>Although the biplane views were chosen orthogonal or near orthogonal, the radii are not orthogonal due to foreshortening. Vector triplets may even represent ellipses with high eccentricity. The data provides only an estimate of the real vascular eccentricity, however, since vessel foreshortening causes fake eccentricity. The analysis of the 3000 cross-sections reconstructed for the coronary artery presented in figure <xref ref-type="fig" rid="F2">2</xref> left showed that the mean eccentricity of elliptical cross-sections, defined as the relationship between the two radii R1 and R2, was relatively small with 1.13 ± 0.34 (SD). The assumption of a circular lumen is supported by this data and data from IVUS studies that report the preservation of nearly circular lumina due to highly localized remodeling [<xref ref-type="bibr" rid="B12">12</xref>]. The second simplification is the assumption of a circular cross-section (Method 1) instead of an elliptic one (Method 2). These two methods proposed for cross-section reconstruction procedure are described in detail below:</p><sec><title>Method 1 (circular lumen)</title><p>The radius R of the circles defining cross-sections was chosen as the geometric mean radius (R = (R1·R2)<sup>1/2</sup>). In this case, the cross-sectional area of the resulting circle is equal to the area of the ellipse where the major and minor axes equal to the two radii (R1 and R2). To convert data with original radii into the circular model the radii vectors R1 and R2 had to be scaled by factors R/R1 and R/R2 respectively. This calculation was done automatically using a macro written in MS Excel™. With the macro, obtained data is rewritten in a journal file format. This allows an automatic generation of all cross-section circles in 3-D-space with three defined points (center point and two points on the circle) by the software Gambit™ (Fluent Inc., Lebanon, USA). Figure <xref ref-type="fig" rid="F2">2</xref> shows the resulting wire grid reconstructions of the coronary lumina. The resulting geometries were subsequently exported in commonly used IGES format.</p></sec><sec><title>Method 2 (elliptic lumen)</title><p>These vectors can be used directly as the axes of elliptic contours only in the case of orthogonality of two radius vectors. Otherwise, the shape of the ellipses is ambiguous due to a loss of spatial information. Ambiguity increases with a decreasing angle between the two radii. Three specific points are needed for an unambiguous definition of elliptic contours in 3-D space. The first point is the center of the ellipse. The second point together with first point must define the major axis of the ellipse, whereas the third point is located on the elliptical contour and does not coincide with the second point. Since both radius vectors were defined using the projection edges of vessel lumina, each may initially be considered as the major axis. If neither of both projections is orthogonal to the major axis of the real elliptic cross-section, then neither of the radius vectors represents the direction or length of the real major axis radius. The larger radius of R1 and R2 is the best approximation, however. For non-orthogonal radius vectors including an angle of less than 45°, a circular cross-section was assumed by Wahle et al. [<xref ref-type="bibr" rid="B15">15</xref>]. In this case, the procedure is the same as for Method 1 and was automatically performed using an MS Excel macro. Finally, a Gambit journal file was generated and exported in IGES file format.</p><p>The ultimate generation of endo-luminal surface and volume geometry from cross-sections defined in IGES file format was done with the CAD-program SolidWorks™ (Solidworks Inc., Concord, USA). The IGES files were imported into SolidWorks™. The volume of each segment was then generated by interpolation using the "Loft" tool in SolidWorks™ with the segment centerline as a guide line. Bifurcations were modeled by extrusion. The branch segment was extruded toward the parent segment starting from the first cross-section of the branch segment until the start of the branch segment was completely inside of the parent segment. Figure <xref ref-type="fig" rid="F3">3</xref> shows an example of a complete coronary artery reconstructed according to Method 1.</p><p>The CFD study described in this paper was limited to the reconstruction of the right coronary artery without branches (see figure <xref ref-type="fig" rid="F4">4</xref>). In order to smooth the outlet geometry for CFD and to eliminate the influence of outlet boundary conditions, we generated an additional 10 mm long segment by extruding the last cross-section of the reconstructed vessel in the direction normal to itself.</p><fig position="float" id="F4"><label>Figure 4</label><caption><p>Endo-luminal surfaces of the three right coronary arteries reconstructed without branches: (a) – normal right coronary artery, (b) –, right coronary artery with "obstructive" atherosclerotic disease, and (c) – right coronary artery with "dilated" atherosclerotic disease. These geometries correspond to images shown in figure 1.</p></caption><graphic xlink:href="1475-925X-5-39-4"/></fig><p>For the comparative study presented in this paper the following models were reconstructed: three HF models of three different coronary vessels with circular cross-sections, three HF models of three different coronary vessels with elliptical cross-sections, three LF models of three different coronary vessels with circular cross-sections and one DLF model of the normal coronary artery with a circular cross-section. Table <xref ref-type="table" rid="T1">1</xref> gives an overview of the main geometric parameters of the reconstructed segments.</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Geometric parameters of the three reconstructed right coronary arteries. D stands for diameter.</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="center">Parameters</td><td align="center">Volume, mm<sup>3</sup></td><td align="center">Length, mm</td><td align="center">Inlet D: mm</td><td align="center">Outlet D: mm</td><td align="center">Range of D: mm</td><td align="center">Mean D: mm</td></tr></thead><tbody><tr><td align="center">"normal"</td><td align="center">483</td><td align="center">62</td><td align="center">3.78</td><td align="center">2.58</td><td align="center">2.22 – 3.78</td><td align="center">3.14</td></tr><tr><td align="center">"stenosed"</td><td align="center">413</td><td align="center">94</td><td align="center">2.97</td><td align="center">2.19</td><td align="center">1.78 – 2.97</td><td align="center">2.36</td></tr><tr><td align="center">"dilating"</td><td align="center">1586</td><td align="center">120</td><td align="center">4.56</td><td align="center">3.87</td><td align="center">2.88 – 5.86</td><td align="center">4.10</td></tr></tbody></table></table-wrap></sec></sec><sec><title>Computational fluid dynamics</title><p>The numerical solution of steady Navier-Stokes equations for momentum and mass conservation governing fluid motion under defined boundary conditions were solved by a control volume finite element method (FEM) implemented in FLUENT 6 (Fluent Inc., Lebanon, USA). For finite element numerical simulation the vessel volume had to be represented by a mesh grid. This transformation of the volume data was done by Gambit (software). The surface of the vessel was triangulated with a node distance between 0.1 and 0.2 mm (1:20 of the mean diameter). Based on this surface mesh, a grid composed of tetrahedral elements was generated in the reconstructed vessels. The total number of nodes exceeded 50,000. The average number of elements per cross-section was 350. Recently, some detailed studies were performed regarding the mesh resolution required to appropriately simulate the blood flow in coronary arteries using finite element methods [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B19">19</xref>]. The authors found that a high mesh resolution near the walls was needed in order to get accurate values of WSS. Based on the results of these studies [<xref ref-type="bibr" rid="B19">19</xref>,<xref ref-type="bibr" rid="B20">20</xref>], we generated a mesh which was refined in the near wall region. A boundary layer consisting of 4 rows, with a growth factor of 1.2 (ratio between two consecutive layers near the wall) and a total depth of 0.2 mm, was generated (see figure <xref ref-type="fig" rid="F5">5</xref>). The quality of the generated mesh grid was assessed using different approaches. The maximal skew of their distribution was, for example, below 0.75, which is fully satisfactory. The resulting number of grid volume elements ranged between 270,000 and 390,000 for the different vessels. Stationary laminar flow was simulated presuming rigid motionless walls. A no-slip condition was assumed at the wall. The pressure value was not imposed at the outlet. Blood was modeled as a Newtonian fluid with a kinematic viscosity of 3.5 10<sup>-6 </sup>m<sup>2</sup>/s. A second order discretization scheme and a SIMPLE model for pressure flow coupling were used. A plug velocity profile was assumed at the inlet, because coronary arteries originate from a large compartment (sinus of the aortic root). The mean flow rate for each investigated vessel was estimated based on our flow rate measurements. These were recorded by the simultaneous measurement of pressure and velocity in these patients by a miniaturized ultrasound Doppler probe positioned within the coronary artery. The mean inlet velocities, 0.17 m/s for the normal patient, 0.27 m/s for the patient with "obstructive" coronary atherosclerosis and 0.09 m/s for the patient with "dilated" coronary atherosclerosis, were used, which resulted in the Reynolds numbers of 184, 228, and 117 respectively, for the three investigated geometries. The mass flows were 112.9 ml/min for the patient with a normal coronary artery, 111.8 ml/min for the patient with "obstructive" coronary artery disease and 88.5 ml/min for the patient with "dilated" coronary artery disease. The mass flows were calculated by multiplying the mean inlet velocity with the inlet cross-section areas obtained as the result of reconstructions. The smaller mass flow in the last patient is due to the smaller perfusion territory of the right coronary artery in this patient. The convergence criteria for relative errors in velocity components and pressure were set as 10<sup>-5</sup>.</p><fig position="float" id="F5"><label>Figure 5</label><caption><p>Grid in the inlet region of the right coronary model of the normal coronary artery.</p></caption><graphic xlink:href="1475-925X-5-39-5"/></fig></sec><sec><title>Statistics</title><p>In order to quantify the differences between the resulting WSS distributions, we generated distribution histograms. The whole range of calculated wall shear stress values was divided into 100 classes. For each class, the area corresponding to the WSS range was calculated and normalized as a percent of the total wall surface area. The sum of the calculated values from all classes was then equal to 100%. Cross-correlations and Chi-square tests were used to compare distribution histograms.</p></sec></sec><sec><title>Results</title><p>Three different coronary vessels were reconstructed using two different reconstruction methods (I and II) with a higher resolution – HF models. The reconstruction method I (circular cross-sections) was also used to generate LF models for all three coronary arteries. A DLF model was generated for the normal coronary vessel. Altogether, 10 models of three different coronary vessels were studied. The analysis of the reconstructed models shows that the number of cross-sections used in generating the high fidelity (HF) models resulted in a rather fine resolution of about 0.3–0.4 mm. This equals the resolution of the original data obtained by a reconstruction of biplane angiograms with software developed at the DHZB (German Heart Institute of Berlin). The resolution of low fidelity (LF) models was about 0.8 mm. However, the comparison of volumes and surfaces between HF and LF models yields an error of <1%.</p><p>Steady numerical flow simulations of steady flow were done for the 10 reconstructed models of the three different coronary vessels. The corresponding distributions of WSS are visualized in figures <xref ref-type="fig" rid="F6">6</xref>, <xref ref-type="fig" rid="F7">7</xref>, and <xref ref-type="fig" rid="F8">8</xref>. The distributions of WSS for the reconstructions with high and low resolution appear identical.</p><fig position="float" id="F6"><label>Figure 6</label><caption><p>WSS distributions in the reconstructed normal coronary artery. From the left to the right: HF model reconstructed by Method 1, LF model reconstructed by Method 1, DLF model reconstructed by Method 1 and HF model reconstructed by Method 2 of the normal right coronary artery. The Reynolds number was Re = 184.</p></caption><graphic xlink:href="1475-925X-5-39-6"/></fig><fig position="float" id="F7"><label>Figure 7</label><caption><p>WSS distributions in the reconstructed right coronary artery with "obstructive" atherosclerotic disease. From the left to the right: HF model reconstructed by Method 1 (left), LF model reconstructed by Method 1 (middle), and HF model reconstructed by Method 2 (right) of the right coronary artery with "obstructive" atherosclerotic disease. The Reynolds number was Re = 228.</p></caption><graphic xlink:href="1475-925X-5-39-7"/></fig><fig position="float" id="F8"><label>Figure 8</label><caption><p>WSS distributions in the reconstructed right coronary artery with "dilated" atherosclerotic disease. From left to right: HF model reconstructed by Method 1 (left), LF model reconstructed by Method 1 (middle), and HF model reconstructed by Method 2 (right) of the right coronary artery with "dilated" atherosclerotic disease. The Reynolds number was Re = 117.</p></caption><graphic xlink:href="1475-925X-5-39-8"/></fig><p>Histogram curves were generated from calculated WSS distributions. The distribution curves agree well (see figures <xref ref-type="fig" rid="F9">9a, b</xref> and <xref ref-type="fig" rid="F9">9c</xref>), and demonstrate a strong cross-correlation (0.91 – 0.95) for each of the investigated three coronary arteries. In the Chi-square test for the comparison of histograms, no significant difference was found (p = 0.995).</p><fig position="float" id="F9"><label>Figure 9</label><caption><p>Images (a), (b) and (c) show histograms of the WSS distributions depicted in figures 6, 7 and 8 respectively. Image (d) shows a comparison of three histograms of the three different vessels with normalized WSS ranges for LF models reconstructed with Method 1. Cr stands for models with circular cross-sections. El stands for models with elliptic cross-sections.</p></caption><graphic xlink:href="1475-925X-5-39-9"/></fig><p>The calculated pressure loss was slightly lower (2.5–8.5%) in LF models. Pressure drops strongly depended on the type of geometry (normal control: 6 mmHg, "obstructive" disease: 25.4 mmHg, or "dilated" disease: 1.83 mmHg).</p><p>The effect on the distribution of WSS, if a circular cross-section (Method 1) is assumed as opposed to an elliptic cross-section (Method 2), is also displayed in figures <xref ref-type="fig" rid="F6">6</xref>, <xref ref-type="fig" rid="F7">7</xref>, and <xref ref-type="fig" rid="F8">8</xref>. The comparison was performed for HF models. The distribution histograms shown in figures <xref ref-type="fig" rid="F9">9a, b</xref> and <xref ref-type="fig" rid="F9">9c</xref> are similar, and also demonstrate a high cross-correlation (0.87 – 0.94). Again, in the Chi-square test for the comparison of histograms, no significant difference was found (p = 0.995). The difference between the volumes was less than 0.017 ml (< 2%), the difference in wall area was less than 13 mm<sup>2 </sup>(< 1%) and the pressure loss was slightly higher (3.6–5.3%) with the elliptical cross-sections in coronary artery disease. It should be noted that there are also some differences between the inlet diameters of models reconstructed with different methods (< 1.5 %). The differences between the outlet diameters of models reconstructed with different methods were even higher but remained below 5 %.</p><p>The three distinct varieties of coronary vessel geometry are characteristically reflected by the WSS histograms. The mean WSS was 4.6 Pa in the normal patient and the range was between 0 Pa and 10 Pa. The mean WSS was higher in the sample with "obstructive" atherosclerotic disease with 8.8 Pa, and had a wide range (0 Pa to 20 Pa). On the contrary, the mean WSS in the sample with "dilated" atherosclerotic disease was lower with 1.3 Pa, and had a rather narrow range (0 Pa to 6 Pa). However, the obtained WSS histogram curves also revealed characteristic shape differences (see figure <xref ref-type="fig" rid="F9">9d</xref>) which were shown by normalizing the WSS ranges. The histogram of the control patient is a symmetrically distributed curve with a single peak of WSS at 4 Pa that is nearly in the middle of the WSS range and is close to the mean value of 4.6 Pa (see figure <xref ref-type="fig" rid="F9">9a</xref>). These small differences between mean, median and peak values are also reflected in the low skew value for the histograms of the normal patient – 0.16. The histogram of the patient with "obstructive" atherosclerotic disease demonstrates an asymmetric distribution where the peak is at 5 Pa, which is located near to the peak seen in the histogram of the patient without coronary disease. However, there is a strong right-sided hump in the histogram curve (see figure <xref ref-type="fig" rid="F9">9b</xref>) which is assumed to correspond to stenotic parts of the vessel. This results in the rather large difference between peak (5 Pa), mean (8.8 Pa), and median (10 Pa) values of WSS. The asymmetry of the histogram curve of the patient with "obstructive" atherosclerotic disease is also reflected by a higher skew value – 0.7. The wide range of WSS values reflects multifocal disease and inhomogeneous remodeling. For the patient with "dilated" atherosclerotic disease, the histogram of WSS distribution is very asymmetric (one-sided), with the peak of WSS having shifted to low values at 0.9 Pa. This results in the strong difference between peak (0.9 Pa), mean (1.3 Pa) and median (3 Pa) values of WSS. The excessive asymmetry of the histogram curve of the patient with "dilated" atherosclerotic disease is reflected by a very high skew value – 0.97. The confinement of WSS to a narrow range of low values means diffuse negative remodeling. It should be noted that the simplified reconstruction approaches had no impact on the shape differences of these curves.</p><p>In order to quantify the effect of geometric simplifications on the local WSS distributions, we investigated the WSS distributions in the coronary artery of the normal patient, reconstructed by different methods, in more detail. A part of the wall surface of the normal patient coronary artery was divided by x-constant planes into 36 sections (see figure <xref ref-type="fig" rid="F10">10</xref>) for each of the three models (HF and LF models reconstructed with Method 1 and HF model reconstructed with Method 2). The distance between the planes was 1 mm. The mean WSS was calculated for each section. Figure <xref ref-type="fig" rid="F11">11</xref> shows a comparison of the curves of the mean WSS values for different models along the x-axis (x-position). The difference in WSS values between HF and LF models (see figure <xref ref-type="fig" rid="F11">11b</xref>) reconstructed by Method 1, for all evaluated 36 sections, was 2.2% ± 1.7%. The maximal difference for one site was 6.6 %. The difference in WSS values between HF models reconstructed by Method 1 and Method 2 (see figure <xref ref-type="fig" rid="F11">11a</xref>) was 4.6% ± 5.3%. The maximal difference for one site was 14.3 %.</p><fig position="float" id="F10"><label>Figure 10</label><caption><p>Normal coronary artery of the normal patient used for the analysis of the impact of evaluated simplifications on the local WSS distribution. The letter A marks the x position (x = 18 mm) of the first x-constant plane whereas the letter D marks the x position (x = -18 mm) of the last x-constant plane defining the region of the vessel wall used for the analysis of the local WSS distribution. Letters B and C mark two neighboring x-constant planes which define one of the 36 evaluated vessel wall parts.</p></caption><graphic xlink:href="1475-925X-5-39-10"/></fig><fig position="float" id="F11"><label>Figure 11</label><caption><p>Comparison of the distribution of the WSS averaged over 36 local vessel wall sites for four of the models (the HF model reconstructed with Method 1 – HFcr; the LF model reconstructed with Method 1 – LFcr; the DLF model reconstructed with Method 1 – DLFcr and the HF model reconstructed with Method 2 – HFel) of the coronary artery of the control (normal) patient.</p></caption><graphic xlink:href="1475-925X-5-39-11"/></fig><p>In order to assess the effect of further reduction of the spatial resolution of geometric reconstruction, we investigated the WSS distributions in the double low fidelity model (DLF) of the coronary artery of the normal patient reconstructed with Method 1. The mean distance between two cross-sections in this model was 1.6 mm. The comparison of volumes and surfaces between HF and DLF models yielded an error of about 1%. The corresponding distribution of WSS is visualized in figure <xref ref-type="fig" rid="F6">6</xref>. The distribution of WSS in the reconstructions with high and double low resolution appear identical in the proximal and middle parts of the coronary model. The differences in the distal part are more impressive. However, the distribution curves agree well (see figure <xref ref-type="fig" rid="F9">9a</xref>), and demonstrate a strong cross-correlation (0.93). In the Chi-square test for comparison of histograms, no significant difference was found (p = 0.995). The more detailed comparison of the local WSS values of the 36 vessel wall parts (middle part of the coronary artery) from the HF and DLF models revealed higher differences than between the HF and LF models reconstructed with Method 1 or than between the models reconstructed with Method 1 and Method 2. The difference in WSS values between HF and DLF models (see figure <xref ref-type="fig" rid="F11">11b</xref>) reconstructed by Method 1 for all 36 evaluated sections was 5.3% ± 5.2%. The maximal difference at any site was 26 %.</p></sec><sec><title>Discussion</title><p>Wall shear stress is the most important mechanical regulatory signal which links flow to adaptive changes of the vascular wall and atherosclerotic lesions [<xref ref-type="bibr" rid="B21">21</xref>]. Three-dimensional reconstruction of coronary artery segments, with subsequent numerical flow simulation studies based on individual patient data, are currently the standard approach to in vivo flow profiling and WSS measurements in coronary arteries [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B9">9</xref>]. The prognostic clinical value of this approach is supported by the results of two recent clinical serial studies [<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B9">9</xref>]. We do not know whether the costs of the enhanced accuracy of the complex modeling approach to WSS estimation proposed in [<xref ref-type="bibr" rid="B8">8</xref>] translates into an additional clinical diagnostic impact as compared to the simplified approaches [<xref ref-type="bibr" rid="B9">9</xref>]. A calculation based on a simplified geometric reconstruction would be an important step beyond exemplary studies from bench to bedside. Moreover, routine catheterization data, which do not supply the high resolution circumferential data (e.g. by IVUS) necessary for the complex approach, should be sufficient for a simplified model with a lower resolution. We investigated the impact of two simplifications of the geometric reconstruction of 3-D vessel lumina on the accuracy of flow simulation in coronary arteries.</p><p>Using a LF model reduces the spatial resolution by a factor of two; however, it also reduces the time for the 3-D geometry reconstruction by nearly a factor of 3. This time saving results from the fact that reconstruction problems due to intersecting cross-sections are avoided. The low resolution (LF) model demonstrated a negligible impact on the vessel wall area and the WSS distribution as compared to the results for the HF model. The differences in pressure drops between the HF and LF model were small compared to the differences related to the type of geometry (normal, obstructive or dilated). The deviation of the distribution histograms, caused by the reduced resolution in the LF model, was not significant and small compared to the differences due to the type of geometry. Hence, the HF resolution neither improves the results nor adds additional clinical information. Consequently, LF models are preferred. The LF models smooth the surface geometry. As a result, some local information is lost that may be important. Since the difference between HF and LF models is equivalent to a reduction of the spatial resolution by a factor of two, we studied the impact on the accuracy of the local geometric variability (see figure <xref ref-type="fig" rid="F11">11</xref>) and found that a sufficient accuracy of reconstruction is preserved. The LF model has a mean distance of 0.8 mm between the two cross-sections used for reconstruction. This resolution of the vessel reconstruction along the vessel axis is of the same order as the slice thickness of the standard CT devices and better than the resolution of IVUS and MRI imaging. Further reduction of the spatial resolution by a factor of four in the DLF model revealed a significant loss of local information about the WSS distribution.</p><p>The comparison of HF models of coronary arteries, reconstructed by Method 1 and Method 2, demonstrated only minor differences in the vessel wall area and the WSS distribution. The maximal error of the calculated pressure drop was 6.8 % for simulations in the coronary artery of the normal patient, which is much lower than the variability of the pressure drop in the different types of geometry. The deviations in the histograms of WSS were much larger between the three coronary arteries with a different geometry than the variability caused by the reconstruction method. On the other hand, the differences caused by the reconstruction method were higher than the differences caused by the reduction of model resolution. This is reflected by the lower cross-correlation coefficients (0.91–0.95) that were found by analyzing the effect of resolution and 0.87–0.94 that were found by analyzing the effect of the reconstruction method. One reason for the deviation of the pressure drop between different models of the same vessel is the use of hydraulic radii (radii calculated from the cross-sectional area) in the elliptical cross-sections with non-orthogonal radii. The geometric mean radius is equal to the hydraulic radius only if the radii are orthogonal. For example, the inlet diameters of the coronary models reconstructed with Method 1 were 3.78 mm in the coronary artery of the control patient, 2.96 mm in the coronary artery of the patient with "obstructive" disease and 4.54 mm in the coronary artery of the patient with "dilated" disease. The hydraulic diameters of the coronary models reconstructed with Method 2 were 3.73 mm, 2.97 mm, and 4.53 mm respectively. Note that a difference of 1% in the vessel radius causes a difference of about 2% in the pressure drop for the same inlet velocity. The second, and more important, reason for the differences in pressure drop and WSS caused by reconstruction Method 2, as opposed to Method 1, are irregularities in the cross-sections reconstructed with Method 2. For non-orthogonal radius vectors including an angle of less than 45°, a circular cross-section was assumed by Wahle et al. without any interpolation or smoothing with respect to cross-sections in the neighborhood [<xref ref-type="bibr" rid="B15">15</xref>]. A circular cross-section, instead of an elliptic one, occurs in about 30% of all the cross-sections (3000 cross-sections were analyzed). These artificial wall irregularities cause artifactual local flow disturbances and WSS artifacts that are avoided by the smoother reconstruction technique (Method 1), which uses a circular shape for all reconstructed cross-sections. Thus, the use of Method 1 for geometry reconstruction should provide fewer artifacts, and hence, more accurate and realistic results from the CFD calculations. Furthermore, Method 1 is more suitable for further automatization of the reconstruction algorithm and may be implemented by macro programming in SolidWorks™. Last but not least, Method 1 reflects the physiology of luminal remodeling which is highly localized and preserves a circular lumen even in the majority of eccentric atherosclerotic lesions [<xref ref-type="bibr" rid="B12">12</xref>].</p><p>The problems of the geometry reconstruction using the software developed by DHZB (intersecting cross-sections and irregularities in cross-sections and therefore irregularities in the interpolated surface caused by the use of elliptical and circular shapes) are mainly due to the fact that the primary goal of the software was an assessment of the vessel diameters and volumes, and that a further use for numerical flow simulations was not considered at that stage. The above mentioned problems may be alternatively resolved by an algorithmic approach. However, this implies a time-consuming modification of the existing software. The proposed simplifications for studying WSS profiling in a coronary reconstruction, allows immediate and statistically relevant profiling in a larger sample of available data (about 60 coronary arteries).</p><p>Since the geometry is the main factor influencing the WSS distribution, it is important to consider the impact of neglecting branch flows. Surprisingly, it was shown that branch flows are of minor importance in determining WSS patterns in the trunk of the right coronary artery [<xref ref-type="bibr" rid="B18">18</xref>].</p><p>One of the important issues of WSS profiling is the validation of WSS profiling based on biplane angiograms. Unfortunately, we can not consider 3-D-IVUS as a gold standard for our method. None of these methods provides a direct measurement. IVUS provides a better resolution for an assessment of the vessel cross-section. However, the resolution is worse with regard to the reconstruction of geometry along the vessel axis in the main flow direction. Furthermore, IVUS is not suitable for studying complex vessel trees similar to the vessels shown in figure <xref ref-type="fig" rid="F2">2</xref>.</p><p>Further simplifications in our study concerned the numerical model. The momentum and mass conservation equations have to be solved under difficult boundary conditions in order to fully model the blood flow in arteries. All physiological parameters have to be accounted for, i.e. wall compliance, pulsatile flow, and non-Newtonian behavior of the blood. In addition, in a coronary artery, cardiac contraction induces a continuous, site specific motion and deformation of the vessel. All these aspects may affect flow patterns and were thoroughly studied in the last years [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B22">22</xref>-<xref ref-type="bibr" rid="B26">26</xref>]. The impact of the assumption of rigid or non-rigid arterial walls has been well investigated and discussed in the literature [<xref ref-type="bibr" rid="B22">22</xref>,<xref ref-type="bibr" rid="B23">23</xref>]. The authors of these studies agree that the assumption of a rigid wall is sufficiently accurate for WSS profiling in investigating atherosclerosis for clinical purposes. This judgment is based on their calculations with regard to reconstructions of the first bifurcation of the left coronary artery. However, these results are also valid for right coronary arteries, since there are no significant differences between these two arteries from a hemodynamic point of view (Reynolds and Strouhal numbers, Womersley parameter, pressure pulse wave and vessel thickness). Among the deformations of the coronary arteries due to cardiac surface motion, only torsion is assumed to have a small effect on local WSS. The effect of pulsatility was also found to be small and to have a limited effect on local WSS. No differences were found between steady-state calculations of the WSS distribution and time-averaged calculations over a whole heart cycle [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B26">26</xref>]. Average calculated Womersly parameters (Wo = R(2πf/ν)<sup>0.5</sup>, where R is the artery radius, f is the heart frequency and ν is the kinematic viscosity of blood) were found to be low in coronary arteries (Wo = 3.05 ± 1.00, N = 117) due to the small radii of the coronary arteries (own unpublished results based on in-vivo coronary hemodynamic data and quantitative coronary angiography data acquired by cardiac catheterization). Since flow unsteadiness associated with pulsatility has a significant impact on the local WSS only, if Wo>>1 (cut-off 5). Pulsatile flow modeling is not necessary in coronary arteries, if we study time-averaged WSS. The WSS distribution averaged over one heart cycle is considered as the main hemodynamic parameter that links flow to adaptive changes of the vascular wall and atherosclerotic lesions. There are other parameters, which are held to be important: e.g. temporal (WSSGt) and spatial WSS gradients (WSSG), and the oscillating WSS index (OSI) [<xref ref-type="bibr" rid="B18">18</xref>]. Many of the above-mentioned aspects (e.g. pulsatility, wall elasticity, non-Newtonian blood behavior), do apply for these parameters (OSI, WSSGt, WSSG). Last but not least, it should be noted that atherosclerotic wall alterations reduce wall elasticity and eventually lead to a rigid wall model.</p><p>The most interesting findings of this study are the differences between the WSS histogram curve shapes. These differences seem to be characteristic for these three different coronary arteries which represent three different entities of coronary pathology (normal patients, patients with "obstructive atherosclerosis" and patients with "dilated" atherosclerosis) and might have diagnostic value. However, further studies with a larger number of coronary arteries are necessary to assess the clinical value of these findings. Local information about WSS distribution may be smoothed by our approach. Such local information is thought to be very important for the study of the correlation between WSS distribution and the distribution of atherosclerotic wall alterations or intimal thickening. However, a low WSS value does not imply that wall alterations will necessarily be present at the corresponding site. The relation between WSS and biologic vascular response is modulated by many other factors. The assessment of this relation needs a probabilistic and multi-factorial approach. We know that clinically symptomatic coronary atherosclerosis usually shows a multimodal, or even diffuse distribution, and demonstrates distinct varieties of positive or negative remodeling associated with different geometries. According to the statistical principle of stratified sampling, extreme varieties of coronary geometries were selected to identify characteristic features of the impact of global geometry on flow and WSS. A characteristic global parameter is mean WSS. The mean WSS may be estimated from the mean vessel radius R and the mean velocity V as τ = μ(4 V/R) (Eq. 1), where μ is the dynamic blood viscosity. Using this equation, the mean WSS may be calculated as 1.49 Pa for the coronary of the normal patient, 2.95 Pa for the coronary from the patient with "obstructive" atherosclerotic disease and 0.6 Pa for the coronary from the patient with "dilated" atherosclerotic disease. The average WSS values calculated from the CFD results are 4.6 Pa, 8.8 Pa and 1.3 Pa respectively. Thus a simplified approach based on the mean radius and mean velocity results in significantly lower values of WSS as compared to values based on the CFD calculations. Moreover, the information of the WSS histograms on scatter, distribution, skew, and peak values of WSS provides quantitative information on the diffuseness, the inhomogeneity and the progress of the atherosclerotic disease and the extent and type of associated remodeling as reflected by the resultant luminal geometry. It should be emphasized that the histograms obtained from the CFD solution characterize a very complex flow pattern, including flow separations and flow recirculations caused by the vessel curvature and local narrowings or enlargements. Such histograms cannot be obtained by a simplified approach as the calculation of WSS for each reconstructed cross-sectional volume slice is based on the Hagen-Poiseuille equation (Eq. 1). Furthermore, the WSS estimation, by using the Hagen-Poiseuille equation, is not able to predict the existence of the wall areas with low wall shear stress values (τ < 0.5 Pa), as it is done by the CFD results. These values are included in our histograms. These low WSS values are very important for our study, since they correlate with the loss of endothelial function and WSS responsiveness [<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B28">28</xref>].</p><p>The distinct differences in the distribution of WSS values obtained from histograms may help to distinguish between, and assess the severity of, different coronary artery diseases. Further analysis is currently being performed in our 3-D coronary database consisting of right and left coronary arteries from 6 control patients, 10 patients with "obstructive" atherosclerotic disease, and 8 patients with "dilated" atherosclerotic disease.</p><p>There were a lot of investigations which considered the problem of reconstruction procedure on the accuracy of the numerical flow simulation in vessels, and especially for coronary arteries [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B20">20</xref>,<xref ref-type="bibr" rid="B22">22</xref>]. The novel aspects of the presented study are the following: we used a non-dimensional allometry approach (WSS histogram curve) to characterize and compare WSS in different coronary arteries, we studied the impact of two reconstruction simplifications on wall shear stress characterization using the WSS histogram curve, we used circular instead of elliptical cross-sections, and we used a reduced number of cross-sections for volume reconstruction. The study was applied to three different right coronary arteries representing three different geometries of this vessel: a right coronary artery of a control patient, a right coronary of a patient with "obstructive" atherosclerotic disease and a right coronary of a patient with "dilated" atherosclerotic disease. The study of three different vessels allowed us to show that the proposed simplifications are not significant for the differentiation of these coronary arteries by the proposed method.</p><p>As we mentioned above, there are no significant differences between right and left coronary arteries from a hemodynamic point of view. Hence, the results obtained in this study with right coronary arteries should also be valid for studies with left coronary arteries.</p></sec><sec><title>Conclusion</title><p>A simplified approach to the reconstruction of coronary vessel lumina from biplane angiograms, by assuming circular cross-sections and using only every second vector triplet of the original data with the mean distance of 0.8 mm between cross-sections, yields sufficiently accurate calculations of vessel volumes, surfaces, wall shear stress distributions, and pressure drops. The resolution of 0.8 mm is accurate enough for the global (histogram) and local characterization of the wall shear stress in coronary arteries. This is valid also for other geometry reconstruction methods (3-D-IVUS or magnet resonance imaging). Lower resolution results in non significant deviations for the global characterization parameter of the WSS and in significant local alterations in WSS calculations. The issue, which remains to be validated, is: May we use biplane angiograms for WSS profiling? A final decision on this subject requires a study with a phantom of a real coronary artery, for which an exact computer model exists. This study is under way in our group.</p><p>Profiles of WSS within whole segments or vessels might identify different patterns of remodeling associated with characteristic changes in the distribution of WSS and quantify the extent and diffuseness of coronary artery disease. However, this issue needs further investigation in a larger number of patients with different coronary geometries in each group (controls, "obstructive" and "dilated" coronary atherosclerosis).</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>Dr. Wellnhofer from German Heart Institute of Berlin performed biplane patient angiograms and velocity measurements with a miniaturized ultrasound Doppler probe. He is also one of the developers of the reconstruction software. Dr. Goubergrits developed the reconstruction software further that allowed performing a volume reconstruction, which may be used for numerical simulations with a CFD code. He performed also the numerical simulations with CFD code FLUENT and proposed the analysis of the WSS distributions by the WSS histograms. Dr. Kertzscher was involved in the statistical analysis of the obtained results. Prof. Dr. Affeld is a supervisor of the project and was also involved in the analysis and the discussion of the obtained results.</p></sec>
Secular trends of antimicrobial resistance of blood isolates in a newly founded Greek hospital	<sec><title>Background</title><p>Antimicrobial resistance is one of the most challenging issues in modern medicine.</p></sec><sec sec-type="methods"><title>Methods</title><p>We evaluated the secular trends of the relative frequency of blood isolates and of the pattern of their in vitro antimicrobial susceptibility in our hospital during the last four and a half years.</p></sec><sec><title>Results</title><p>Overall, the data regarding the relative frequency of blood isolates in our newly founded hospital do not differ significantly from those of hospitals that are functioning for a much longer period of time. A noteworthy emerging problem is the increasing antimicrobial resistance of Gram-negative bacteria, mainly <italic>Acinetobacter baumannii </italic>and <italic>Klebsiella pneumoniae </italic>to various classes of antibiotics. <italic>Acinetobacter baumannii </italic>isolates showed an increase of resistance to amikacin (p = 0.019), ciprofloxacin (p = 0.001), imipenem (p < 0.001), and piperacillin/tazobactam (p = 0.01) between the first and second period of the study.</p></sec><sec><title>Conclusion</title><p>An alarming increase of the antimicrobial resistance of <italic>Acinetobacter baumannii </italic>isolates has been noted during our study.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Falagas</surname><given-names>Matthew E</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Kasiakou</surname><given-names>Sofia K</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Nikita</surname><given-names>Dimitra</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Morfou</surname><given-names>Panayiota</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Georgoulias</surname><given-names>George</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Rafailidis</surname><given-names>Petros I</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib>	BMC Infectious Diseases	<sec><title>Background</title><p>Increasing antimicrobial resistance among bloodstream isolates is considered a significant problem worldwide [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B2">2</xref>]. This is especially true in some areas including the countries of Southern Europe where a considerable proportion of pathogens are resistant to antibiotics of several classes [<xref ref-type="bibr" rid="B3">3</xref>]. Although antimicrobial resistance is noted in all pathogens, some phenotypes of resistance such as methicillin resistant <italic>Staphylococcus aureus </italic>(MRSA), vancomycin resistant enterococci (VRE), methicillin resistant coagulase negative staphylococci (MRCNS), and carbapenem resistant enterobacteriacae, <italic>Pseudomonas aeruginosa</italic>, and <italic>Acinetobacter baumannii </italic>are of particular concern. We sought to study the secular trends of the relative frequency and antimicrobial resistance of blood isolates in a newly founded hospital in Greece.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Patient population</title><p>The patient population comprised of patients admitted to Henry Dunant Hospital, Athens, Greece in the period of 01/01/2001–30/06/2005. Henry Dunant Hospital was founded in October 2000. It is a general tertiary hospital with 450 beds covering most medical specialties with the exception of pediatrics, obstetrics, and transplant surgery. It has 3 combined medical and surgical intensive care units with a total of 38 beds.</p></sec><sec><title>Microbiological studies</title><p>Identification of the microorganisms to the species level was performed with the automated system Vitek 2 (Biomérieux) according to the manufacturer's instructions. Not all, but only the first blood isolate per patient was included in the study. The Bactec system (Becton-Dickinson) was used during 2001, 2002, and 2003, and the BacT Alert 3D (Biomérieux) was used during 2004 and 2005. Isolation of bacteria was followed by susceptibility testing that was performed with the Vitek 2 system, applying the criteria suggested by the Clinical and Laboratory Standards Institute (CLSI) [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B5">5</xref>]. The identification and antimicrobial susceptibility of viridans Streptococci was preformed by the use of API (BioMérieux) and the use of the Kirby-Bauer method. Fungi were identified with the use of the specific card for the Vitek 2 system. Susceptibility to colistin was tested by the Vitek method and the E-test. Pulsed field gel electrophoresis and ribotyping were not performed to exclude secondary outbreak strains.</p></sec><sec><title>Statistical analysis</title><p>Differences in proportions were compared by x<sup>2 </sup>test or Fisher's exact test where appropriate. Statistical significance was set for p < 0.05. All statistical analyses were performed using SPSS 11.0 and S-PLUS 6.1 Professional.</p></sec></sec><sec><title>Results</title><p>The frequency of isolation of bacteria from cultures of blood specimens was 182 per 12,593 admissions during 2001 (14.4 per 1,000 admissions), 507 per 25,865 admissions during 2002 (19.6 per 1,000 admissions), 693 per 30,597 admissions during 2003 (22.6 per 1,000 admissions), 566 per 30,599 admissions during 2004 (18.4 per 1,000 admissions) and 208 per 15,683 admissions during 2005 (13.2 per 1,000 admissions). There was a significant difference in the proportion of isolates identified over the 5year period (p < 0.001). The percentage of positive blood cultures to the total number of blood cultures was 4.84% for the year 2001, 6.46% for 2002, 7.91% for 2003, 6.71% for 2004 and 5.7% for 2005. Regarding the relative frequency of the bacteria isolated from blood specimens, Gram-positive bacteria were more common than Gram-negative bacteria throughout the study period (Table <xref ref-type="table" rid="T1">1</xref>). Coagulase negative staphylococci were the commonest blood isolates (52.5 % of total). The relative frequency of other Gram-positive and Gram-negative microorganisms was the following, in descending order: <italic>Escherichia coli </italic>(8.9 %), <italic>Staphylococcus aureus </italic>(5.9 %), <italic>Pseudomonas aeruginosa </italic>(5.2 %), <italic>Klebsiella spp </italic>(4.8 %), <italic>Acinetobacter baumannii </italic>(4.1 %), <italic>Enterococcus faecalis </italic>(2.2 %), and <italic>Enterococcus faecium </italic>(1.8 %). In Table <xref ref-type="table" rid="T2">2</xref> we summarized the relative frequency of blood isolates by service, namely wards and intensive care unit.</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Relative frequency of blood isolates in Henry Dunant Hospital, Athens, Greece (01/01/2001–30/6/2005).</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="center" colspan="7"><bold>Number of isolates (proportion within the year)</bold></td></tr></thead><tbody><tr><td align="left"><bold><italic>Microorganisms</italic></bold></td><td align="center"><bold>2001 (n = 182)</bold></td><td align="center"><bold>2002 (n = 506)</bold></td><td align="center"><bold>2003 (n = 693)</bold></td><td align="center"><bold>2004 (n = 566)</bold></td><td align="center"><bold>2005* (n = 208)</bold></td><td align="center"><bold>Total</bold></td><td align="center"><bold>p-value</bold></td></tr><tr><td colspan="8"><hr></hr></td></tr><tr><td align="left"><bold><italic><underline>Gram-positive</underline></italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left"><italic>Coagulase-negative staphylococci</italic></td><td align="center">84 (46.2)</td><td align="center">282 (55.6)</td><td align="center">377 (54.4)</td><td align="center">297 (52.5)</td><td align="center">94 (45.1)</td><td align="center">1134 (52.5)</td><td align="center">0.03</td></tr><tr><td align="left">Staphylococcus aureus</td><td align="center">12 (6.6)</td><td align="center">40 (7.9)</td><td align="center">42 (6.1)</td><td align="center">27 (4.8)</td><td align="center">8 (3.8)</td><td align="center">129 (5.9)</td><td align="center">0.16</td></tr><tr><td align="left"><italic>Enterococcus faecalis</italic></td><td align="center">2 (1.1)</td><td align="center">15 (3.0)</td><td align="center">21 (3.0)</td><td align="center">6 (1.1)</td><td align="center">4 (1.9)</td><td align="center">48 (2.2)</td><td align="center">0.09</td></tr><tr><td align="left"><italic>Enterococcus faecium</italic></td><td align="center">4 (2.2)</td><td align="center">14 (2.8)</td><td align="center">8 (1.2)</td><td align="center">11 (1.9)</td><td align="center">2 (0.9)</td><td align="center">39 (1.8)</td><td align="center">0.25</td></tr><tr><td align="left"><bold><italic><underline>Gram negative</underline></italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Escherichia coli</td><td align="center">15 (8.2)</td><td align="center">36 (7.1)</td><td align="center">57 (8.2)</td><td align="center">54 (9.5)</td><td align="center">30 (14.4)</td><td align="center">192 (8.9)</td><td align="center">0.03</td></tr><tr><td align="left">Pseudomonas aeruginosa</td><td align="center">13 (7.1)</td><td align="center">29 (5.7)</td><td align="center">35 (5.0)</td><td align="center">26 (4.6)</td><td align="center">11 (5.3)</td><td align="center">114 (5.2)</td><td align="center">0.72</td></tr><tr><td align="left">Acinetobacter baumannii</td><td align="center">10 (5.5)</td><td align="center">10 (2.0)</td><td align="center">15 (2.2)</td><td align="center">38 (6.7)</td><td align="center">17 (8.2)</td><td align="center">90 (4.1)</td><td align="center"><0.001</td></tr><tr><td align="left">Proteus mirabilis</td><td align="center">3 (1.7)</td><td align="center">1 (0.2)</td><td align="center">1 (0.1)</td><td align="center">4 (0.7)</td><td align="center">3 (1.4)</td><td align="center">12 (0.5)</td><td align="center">0.03</td></tr><tr><td align="left">Klebsiella spp.</td><td align="center">5 (2.8)</td><td align="center">17 (3.4)</td><td align="center">39 (5.6)</td><td align="center">33 (5.7)</td><td align="center">11 (5.3)</td><td align="center">105 (4.8)</td><td align="center">0.18</td></tr><tr><td align="left">Enterobacter spp.</td><td align="center">4 (2.2)</td><td align="center">8 (1.6)</td><td align="center">24 (3.5)</td><td align="center">9 (1.6)</td><td align="center">3 (1.4)</td><td align="center">48 (2.2)</td><td align="center">0.11</td></tr><tr><td align="left">Salmonella spp.</td><td align="center">2 (1.1)</td><td align="center">1 (0.2)</td><td align="center">4 (0.6)</td><td align="center">9 (1.6)</td><td align="center">0 (0)</td><td align="center">16 (0.7)</td><td align="center">0.05</td></tr><tr><td align="left"><bold><italic>Others </italic></bold>(include other Gram positive and Gram negative)</td><td align="center">22 (12)</td><td align="center">25 (4.9)</td><td align="center">34 (4.9)</td><td align="center">26 (4.6)</td><td align="center">15 (7.2)</td><td align="center">122 (5.6)</td><td align="center">0.002</td></tr><tr><td align="left"><bold><italic>Fungi</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left"><italic>Candida</italic></td><td align="center">6 (3.3)</td><td align="center">29 (5.7)</td><td align="center">36 (5.2)</td><td align="center">26 (4.6)</td><td align="center">10 (4.8)</td><td align="center">107 (4.9)</td><td align="center">0.74</td></tr></tbody></table></table-wrap><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Relative frequency of blood isolates by service in Henry Dunant Hospital, Athens, Greece (01/01/2001–30/6/2005).</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="center" colspan="8"><bold>Number of isolates by service/total number of same species isolates within the year</bold></td></tr></thead><tbody><tr><td align="left"><bold><italic>Microorganisms</italic></bold></td><td align="center"><bold>2002</bold></td><td></td><td align="center"><bold>2003</bold></td><td></td><td align="center"><bold>2004</bold></td><td></td><td align="center"><bold>2005</bold></td><td></td></tr><tr><td colspan="9"><hr></hr></td></tr><tr><td align="left"><bold><italic><underline>Gram-positive</underline></italic></bold></td><td align="center">Wards</td><td align="center">ICU</td><td align="center">Wards</td><td align="center">ICU</td><td align="center">Wards</td><td align="center">ICU</td><td align="center">Wards</td><td align="center">ICU</td></tr><tr><td align="left"><italic>Coagulase-negative staphylococci</italic></td><td align="center">158/282</td><td align="center">124/282</td><td align="center">191/377</td><td align="center">186/377</td><td align="center">159/297</td><td align="center">138/297</td><td align="center">49/94</td><td align="center">45/94</td></tr><tr><td align="left">Staphylococcus aureus</td><td align="center">26/40</td><td align="center">14/40</td><td align="center">21/42</td><td align="center">21/42</td><td align="center">17/27</td><td align="center">10/27</td><td align="center">6/8</td><td align="center">2/8</td></tr><tr><td align="left"><italic>Enterococcus faecalis</italic></td><td align="center">10/15</td><td align="center">5/15</td><td align="center">13/21</td><td align="center">8/21</td><td align="center">3/6</td><td align="center">3/6</td><td align="center">3/4</td><td align="center">1/4</td></tr><tr><td align="left"><italic>Enterococcus faecium</italic></td><td align="center">10/14</td><td align="center">4/14</td><td align="center">6//8</td><td align="center">2/8</td><td align="center">8/11</td><td align="center">3/11</td><td align="center">2/2</td><td align="center">0/2</td></tr><tr><td align="left"><bold><italic><underline>Gram negative</underline></italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Escherichia coli</td><td align="center">27/36</td><td align="center">9/36</td><td align="center">48/57</td><td align="center">9/57</td><td align="center">49/54</td><td align="center">5/54</td><td align="center">28/30</td><td align="center">2/30</td></tr><tr><td align="left">Pseudomonas aeruginosa</td><td align="center">17/29</td><td align="center">12/29</td><td align="center">18/35</td><td align="center">17/35</td><td align="center">14/26</td><td align="center">12/26</td><td align="center">6/11</td><td align="center">5/11</td></tr><tr><td align="left">Acinetobacter baumannii</td><td align="center">2/10</td><td align="center">8/10</td><td align="center">7/15</td><td align="center">8/15</td><td align="center">9/38</td><td align="center">29/38</td><td align="center">7/17</td><td align="center">10/17</td></tr><tr><td align="left">Klebsiella spp.</td><td align="center">12/17</td><td align="center">5/17</td><td align="center">25/39</td><td align="center">14/39</td><td align="center">14/33</td><td align="center">19/33</td><td align="center">4/11</td><td align="center">7/11</td></tr></tbody></table></table-wrap><p>We compared the antimicrobial resistance of blood isolates of two periods: the first period was 1/1/2002–31/12/2003 and the second period 1/1/2004–30/6/2005. The year 2001 was not included in the comparison of the antimicrobial resistance because the in vitro susceptibility data were not readily available. The antimicrobial resistance of Gram-negative bacteria isolated from blood in our hospital showed some interesting trends (Table <xref ref-type="table" rid="T3">3</xref>). In Table <xref ref-type="table" rid="T4">4</xref> we present data regarding the in vitro susceptibility patterns and the respective MIC<sub>90 </sub>of the isolated bacteria. <italic>Acinetobacter baumannii </italic>isolates showed an increase of resistance to amikacin (p = 0.019), ciprofloxacin (p = 0.001), imipenem (p < 0.001), and piperacillin/tazobactam(p = 0.01) between the first and second period of the study. In addition, we noted the appearance of resistance to polymyxins in one <italic>Acinetobacter baumannii </italic>isolate. Regarding the secular changes of the antimicrobial resistance of <italic>Pseudomonas aeruginosa </italic>isolates during our study, there was only one statistically significant association, namely increased resistance to ceftazidime (p = 0.016).</p><table-wrap position="float" id="T3"><label>Table 3</label><caption><p>Trends of antimicrobial resistance of Gram-positive and Gram-negative bacteria.</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="center" colspan="7"><bold>Number of resistant isolates/total isolates tested (proportion) within the year</bold></td></tr></thead><tbody><tr><td align="left"><bold><italic>Microorganisms</italic></bold></td><td align="center"><bold>2002</bold></td><td align="center"><bold>2003</bold></td><td align="center"><bold>2004</bold></td><td align="center"><bold>2 005</bold></td><td align="center"><bold>2002/2003</bold></td><td align="center"><bold>2004/2005</bold></td><td align="center"><bold>p-value</bold></td></tr><tr><td colspan="8"><hr></hr></td></tr><tr><td align="left"><bold><italic><underline>Gram-positive</underline></italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left"><bold><italic>Staphylococcus epidermidis</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Oxacillin</td><td align="center">171/202 (84.7)</td><td align="center">215/261 (82.4)</td><td align="center">179/219 (81.2)</td><td align="center">75/94 (79.8)</td><td align="center">386/463 (83.3%)</td><td align="center">254/313 (81.2%)</td><td align="center">0.10</td></tr><tr><td align="left">Gentamicin</td><td align="center">144/202 (71.3)</td><td align="center">168/261 (64.4)</td><td align="center">148/219 (67.6)</td><td align="center">54/94 (57.5)</td><td align="center">312/463 (67.4%)</td><td align="center">202/313 (64.5%)</td><td align="center">0.37</td></tr><tr><td align="left">Vancomycin</td><td align="center">0/202 (0)</td><td align="center">0/261 (0)</td><td align="center">0/219 (0)</td><td align="center">0/94 (0)</td><td align="center">0/463 (0%)</td><td align="center">0/313 (0%)</td><td align="center">NA</td></tr><tr><td align="left"><bold><italic>Staphylococcus aureus</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Oxacillin</td><td align="center">25/38 (65.8)</td><td align="center">21/30 (70)</td><td align="center">12/25 (48)</td><td align="center">6/8 (75)</td><td align="center">46/68 (67.6%)</td><td align="center">18/33 (54.5%)</td><td align="center">0.19</td></tr><tr><td align="left">Gentamicin</td><td align="center">23/38 (60.5)</td><td align="center">4/30 (13.3)</td><td align="center">6/25(24)</td><td align="center">3/8 (37.5)</td><td align="center">27/68 (39.7%)</td><td align="center">9/33 (27.3%)</td><td align="center">0.22</td></tr><tr><td align="left">Vancomycin</td><td align="center">0/38 (0)</td><td align="center">0/30 (0)</td><td align="center">0/25 (0)</td><td align="center">0/8 (0)</td><td align="center">0/68 (0%)</td><td align="center">0/33 (0%)</td><td align="center">1</td></tr><tr><td align="left"><bold><italic>Enterococcus faecalis</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Vancomycin</td><td align="center">0/14 (0)</td><td align="center">2/19 (10.5)</td><td align="center">0/7 (0)</td><td align="center">0/4 (0)</td><td align="center">2/33 (6.1%)</td><td align="center">0/11 (0%)</td><td align="center">1</td></tr><tr><td align="left"><bold><italic>Enterococcus faecium</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Vancomycin</td><td align="center">1/15 (6.7)</td><td align="center">0/8 (0)</td><td align="center">3/11 (27.3)</td><td align="center">0/2 (0)</td><td align="center">1/22 (4.5%)</td><td align="center">3/13 (23.1%)</td><td align="center">0.13</td></tr><tr><td align="left"><bold><italic><underline>Gram negative</underline></italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold><italic>Acinetobacter baumannii</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Amikacin</td><td align="center">6/9 (66.6)</td><td align="center">10/15 (66.6)</td><td align="center">32/35 (91.4)</td><td align="center">15/17 (88.2)</td><td align="center">16/24 (66.7%)</td><td align="center">47/52 (90.4%)</td><td align="center">0.019</td></tr><tr><td align="left">Ceftazidime</td><td align="center">8/9 (88.8)</td><td align="center">13/15 (86.6)</td><td align="center">34/35 (97.1)</td><td align="center">17/17 (100)</td><td align="center">21/24 (87.5%)</td><td align="center">51/52 (98.1%)</td><td align="center">0.09</td></tr><tr><td align="left">Ciprofloxacin</td><td align="center">6/9 (66.6)</td><td align="center">11/15 (73.3)</td><td align="center">34/35 (97.1)</td><td align="center">17/17 (100)</td><td align="center">17/24 (70.8%)</td><td align="center">51/52 (98.1%)</td><td align="center">0.001</td></tr><tr><td align="left">Colistin</td><td align="center">1/9 (11.1)</td><td align="center">0/15 (0)</td><td align="center">0/35 (0)</td><td align="center">1/17 (5.8)</td><td align="center">0/24 (0%)</td><td align="center">1/52 (0%)</td><td align="center">1</td></tr><tr><td align="left">Gentamicin</td><td align="center">4/9 (44.4)</td><td align="center">10/15 (66.6)</td><td align="center">16/35 (45.7)</td><td align="center">12/17 (70.5)</td><td align="center">14/24 (58.3%)</td><td align="center">28/52 (53.8%)</td><td align="center">0.71</td></tr><tr><td align="left">Imipenem</td><td align="center">4/9 (44.4)</td><td align="center">8/15 (53.3)</td><td align="center">34/35 (97.1)</td><td align="center">17/17 (100)</td><td align="center">12/24 (50%)</td><td align="center">51/52 (98.1%)</td><td align="center"><0.001</td></tr><tr><td align="left">Piperacillin/Tazobactam</td><td align="center">7/9 (77.7)</td><td align="center">12/15 (80)</td><td align="center">34/35 (97.1)</td><td align="center">17/17 (100)</td><td align="center">19/24 (79.2%)</td><td align="center">51/52 (98.1%)</td><td align="center">0.01</td></tr><tr><td align="left"><bold><italic>Pseudomonas aeruginosa</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Amikacin</td><td align="center">15/32 (46.9)</td><td align="center">15/33 (45.5)</td><td align="center">17/25 (68)</td><td align="center">4/11 (36.4)</td><td align="center">30/65 (46.2%)</td><td align="center">21/36 (58.3%)</td><td align="center">0.24</td></tr><tr><td align="left">Ceftazidime</td><td align="center">18/32 (56.3)</td><td align="center">19/33 (57.6)</td><td align="center">21/25 (84)</td><td align="center">8/11 (72.7)</td><td align="center">37/65 (56.9%)</td><td align="center">29/36 (80.6%)</td><td align="center">0.016</td></tr><tr><td align="left">Ciprofloxacin</td><td align="center">20/32 (62.5)</td><td align="center">14/33 (42.4)</td><td align="center">17/25 (68)</td><td align="center">5/11 (45.4)</td><td align="center">34/65 (52.3%)</td><td align="center">22/36 (61.1%)</td><td align="center">0.39</td></tr><tr><td align="left">Colistin</td><td align="center">0/32 (0)</td><td align="center">0/33 (0)</td><td align="center">0/25 (0)</td><td align="center">0/11 (0)</td><td align="center">0/65 (0%)</td><td align="center">0/36 (0%)</td><td align="center">1</td></tr><tr><td align="left">Imipenem</td><td align="center">20/32 (62.5)</td><td align="center">14/33 (42.4)</td><td align="center">15/25 (60)</td><td align="center">5/11 (45.4)</td><td align="center">34/65 (52.3%)</td><td align="center">20/36 (55.5%)</td><td align="center">0.75</td></tr><tr><td align="left">Piperacillin/Tazobactam</td><td align="center">3/32 (9.4)</td><td align="center">7/33 (21.2)</td><td align="center">18/25 (72)</td><td align="center">2/11 (18.2)</td><td align="center">10/65 (15.4%)</td><td align="center">20/36 (55.5%)</td><td align="center">0.10</td></tr><tr><td align="left"><bold><italic>Klebsiella pneumoniae</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Ceftazidime</td><td align="center">2/7 (28.6)</td><td align="center">11/34 (32.4)</td><td align="center">15/29 (51.7)</td><td align="center">9/11 (81.8)</td><td align="center">13/41 (31.7%)</td><td align="center">24/40 (60%)</td><td align="center">0.010</td></tr><tr><td align="left">Ciprofloxacin</td><td align="center">1/7 (14.3)</td><td align="center">12/34 (35.3)</td><td align="center">16/29 (55.2)</td><td align="center">4/11 (36.4)</td><td align="center">13/41 (31.7%)</td><td align="center">20/40 (50%)</td><td align="center">0.006</td></tr><tr><td align="left">Meropenem</td><td align="center">0/7 (0)</td><td align="center">1/34 (2.9)</td><td align="center">8/29 (27.6)</td><td align="center">9/11 (81.8)</td><td align="center">1/41 (2.4%)</td><td align="center">17/40 (42.5%)</td><td align="center"><0.001</td></tr><tr><td align="left">Cefepime</td><td align="center">0/7 (0)</td><td align="center">11/34 (32.4)</td><td align="center">15/29 (51.7)</td><td align="center">9/11 (81.8)</td><td align="center">11/41 (26.8%)</td><td align="center">24/40 (60%)</td><td align="center"><0.001</td></tr><tr><td align="left">Cefoxitin</td><td align="center">0/7 (0)</td><td align="center">10/34 (29.4)</td><td align="center">18/29 (62.1)</td><td align="center">9/11 (81.8)</td><td align="center">10/41 (24.4%)</td><td align="center">27/40 (67.5%)</td><td align="center"><0.001</td></tr><tr><td align="left">Tobramycin</td><td align="center">0/7 (0)</td><td align="center">8/34 (23.5)</td><td align="center">17/29 (58.6)</td><td align="center">9/11 (81.8)</td><td align="center">8/41 (19.5%)</td><td align="center">26/40 (65%)</td><td align="center"><0.001</td></tr><tr><td align="left">Piperacillin/Tazobactam</td><td align="center">0/7 (0)</td><td align="center">8/34 (23.5)</td><td align="center">15/29 (51.7)</td><td align="center">9/11 (81.8)</td><td align="center">8/41 (19.5%)</td><td align="center">24/40 (60%)</td><td align="center"><0.001</td></tr><tr><td align="left"><bold><italic>Escherichia coli</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Amikacin</td><td align="center">0/35 (0)</td><td align="center">0/57 (0)</td><td align="center">4/47(8.5)</td><td align="center">1/30(0)</td><td align="center">0/92 (0%)</td><td align="center">5/77 (6.5%)</td><td align="center">0.05</td></tr><tr><td align="left">Ciprofloxacin</td><td align="center">7/35 (0.2)</td><td align="center">3/57 (5.2)</td><td align="center">4/47(8.5)</td><td align="center">9/30(30)</td><td align="center">10/92 (10.9%)</td><td align="center">13/77 (16.9%)</td><td align="center">0.006</td></tr><tr><td align="left">Piperacillin/Tazobactam</td><td align="center">1/35 (2.8)</td><td align="center">1/57 (1.7)</td><td align="center">4/47(8.5)</td><td align="center">2/30(6.6)</td><td align="center">2/92 (2.2%)</td><td align="center">6/77 (7.8%)</td><td align="center">0.37</td></tr><tr><td align="left">Ceftazidime</td><td align="center">2/35 (5.7)</td><td align="center">0/57 (0)</td><td align="center">4/47(8.5)</td><td align="center">5/30(16.6)</td><td align="center">2/92 (2.2%)</td><td align="center">9/77 (11.7%)</td><td align="center">0.02</td></tr><tr><td align="left">Cefoxitin</td><td align="center">4/35 (11.4)</td><td align="center">1/57 (1.7)</td><td align="center">4/47(8.5)</td><td align="center">4/30(13.3)</td><td align="center">5/92 (5.4%)</td><td align="center">8/77 (10.4%)</td><td align="center">0.18</td></tr><tr><td align="left">Meropenem</td><td align="center">1/35 (2.8)</td><td align="center">0/57 (0)</td><td align="center">1/47(2.1)</td><td align="center">0/30(0)</td><td align="center">1/92 (1.1%)</td><td align="center">1/77 (1.3%)</td><td align="center">0.52</td></tr></tbody></table><table-wrap-foot><p> Data for the year 2005 refer to the period from January 2005 through June 2005.</p><p>** P-values refer to the comparison of proportions between 2002/2003 and 2004/2005.</p></table-wrap-foot></table-wrap><table-wrap position="float" id="T4"><label>Table 4</label><caption><p>Data on antimicrobial resistance patterns and respective MIC<sub>90 </sub>of isolated bacteria.</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="center" colspan="4"><bold>2002</bold></td><td align="center" colspan="4"><bold>2003</bold></td><td align="center" colspan="4"><bold>2004</bold></td><td align="center" colspan="4"><bold>2005</bold></td></tr></thead><tbody><tr><td align="left"><bold><italic>Microorganisms</italic></bold></td><td align="center"><bold>S</bold></td><td align="center"><bold>I</bold></td><td align="center"><bold>R*</bold></td><td align="center"><bold>MIC</bold><sub><bold>90</bold></sub><sup>#</sup></td><td align="center"><bold>S</bold></td><td align="center"><bold>I</bold></td><td align="center"><bold>R</bold></td><td align="center"><bold>MIC</bold><sub><bold>90</bold></sub></td><td align="center"><bold>S</bold></td><td align="center"><bold>I</bold></td><td align="center"><bold>R</bold></td><td align="center"><bold>MIC</bold><sub><bold>90</bold></sub></td><td align="center"><bold>S</bold></td><td align="center"><bold>I</bold></td><td align="center"><bold>R</bold></td><td align="center"><bold>MIC</bold><sub><bold>90</bold></sub></td></tr><tr><td colspan="17"><hr></hr></td></tr><tr><td align="left"><bold><italic><underline>Gram-positive</underline></italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left"><bold><italic>Staphylococcus epidermidis</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Oxacillin</td><td align="center">31/202</td><td align="center">0/202</td><td align="center">171/202</td><td align="center">4</td><td align="center">46/261</td><td align="center">0/261</td><td align="center">215/261</td><td align="center">4</td><td align="center">36/214</td><td align="center">0/214</td><td align="center">179/214</td><td align="center">4</td><td align="center">15/94</td><td align="center">0/94</td><td align="center">75/94</td><td align="center">4</td></tr><tr><td align="left">Gentamicin</td><td align="center">58/202</td><td align="center">18/202</td><td align="center">126/202</td><td align="center">16</td><td align="center">87/261</td><td align="center">20/261</td><td align="center">154/261</td><td align="center">16</td><td align="center">71/219</td><td align="center">0/219</td><td align="center">148/219</td><td align="center">16</td><td align="center">40/94</td><td align="center">10/94</td><td align="center">44/94</td><td align="center">16</td></tr><tr><td align="left">Vancomycin</td><td align="center">202/202</td><td align="center">0/202</td><td align="center">0/202</td><td align="center">2</td><td align="center">261/261</td><td align="center">0/261</td><td align="center">0/261</td><td align="center">2</td><td align="center">219/219</td><td align="center">0/219</td><td align="center">0/219</td><td align="center">2</td><td align="center">94/94</td><td align="center">0/94</td><td align="center">0/94</td><td align="center">4</td></tr><tr><td align="left"><bold><italic>Staphylococcus aureus</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Oxacillin</td><td align="center">13/38</td><td align="center">0/38</td><td align="center">25/38</td><td align="center">4</td><td align="center">9/30</td><td align="center">0/30</td><td align="center">21/30</td><td align="center">4</td><td align="center">13/25</td><td align="center">0/25</td><td align="center">12/25</td><td align="center">4</td><td align="center">2/8</td><td align="center">0/8</td><td align="center">6/8</td><td align="center">4</td></tr><tr><td align="left">Gentamicin</td><td align="center">16/38</td><td align="center">13/38</td><td align="center">9/38</td><td align="center">8</td><td align="center">21/30</td><td align="center">5/30</td><td align="center">4/30</td><td align="center">8</td><td align="center">19/25</td><td align="center">0/25</td><td align="center">6/25</td><td align="center">4</td><td align="center">5/8</td><td align="center">1/8</td><td align="center">2/8</td><td align="center">0.5</td></tr><tr><td align="left">Vancomycin</td><td align="center">38/38</td><td align="center">0/38</td><td align="center">0/38</td><td align="center">1</td><td align="center">30/30</td><td align="center">0/30</td><td align="center">0/30</td><td align="center">2</td><td align="center">25/25</td><td align="center">0/25</td><td align="center">0/25</td><td align="center">1</td><td align="center">8/8</td><td align="center">0/8</td><td align="center">0/8</td><td align="center">1</td></tr><tr><td align="left"><bold><italic>Enterococcus faecalis</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Vancomycin</td><td align="center">14/14</td><td align="center">0/14</td><td align="center">0/14</td><td align="center">2</td><td align="center">17/19</td><td align="center">0/19</td><td align="center">2/19</td><td align="center">2</td><td align="center">6/6</td><td align="center">0/6</td><td align="center">0/6</td><td align="center">2</td><td align="center">4/4</td><td align="center">0/4</td><td align="center">0/4</td><td align="center">2</td></tr><tr><td align="left"><bold><italic>Enterococcus faecium</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Vancomycin</td><td align="center">13/14</td><td align="center">0/14</td><td align="center">1/14</td><td align="center">1</td><td align="center">8/8</td><td align="center">0/8</td><td align="center">0/8</td><td align="center">1</td><td align="center">2/2</td><td align="center">0/2</td><td align="center">0/2</td><td align="center">32</td><td align="center">2/2</td><td align="center">0/2</td><td align="center">0/2</td><td align="center">1</td></tr><tr><td align="left"><bold><italic><underline>Gram negative</underline></italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold><italic>Acinetobacter baumannii</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center">Amikacin</td><td align="center">3/9</td><td align="center">1/9</td><td align="center">5/9</td><td align="center">64</td><td align="center">5/15</td><td align="center">0/15</td><td align="center">10/15</td><td align="center">64</td><td align="center">3/35</td><td align="center">2/35</td><td align="center">30/35</td><td align="center">64</td><td align="center">2/17</td><td align="center">2/17</td><td align="center">13/17</td><td align="center">64</td></tr><tr><td align="center">Ceftazidime</td><td align="center">1/9</td><td align="center">2/9</td><td align="center">6/9</td><td align="center">64</td><td align="center">2/15</td><td align="center">1/15</td><td align="center">12/15</td><td align="center">64</td><td align="center">1/35</td><td align="center">0/35</td><td align="center">34/35</td><td align="center">64</td><td align="center">0/17</td><td align="center">0/17</td><td align="center">17/17</td><td align="center">64</td></tr><tr><td align="center">Ciprofloxacin</td><td align="center">3/9</td><td align="center">0/9</td><td align="center">6/9</td><td align="center">4</td><td align="center">3/15</td><td align="center">0/15</td><td align="center">12/15</td><td align="center">4</td><td align="center">1/35</td><td align="center">0/35</td><td align="center">34/35</td><td align="center">4</td><td align="center">0/17</td><td align="center">0/17</td><td align="center">17/17</td><td align="center">4</td></tr><tr><td align="center">Colistin</td><td align="center">8/9</td><td align="center">0/9</td><td align="center">1/9</td><td align="center">2</td><td align="center">15/15</td><td align="center">0/15</td><td align="center">0/15</td><td align="center">1</td><td align="center">35/35</td><td align="center">0/35</td><td align="center">0/35</td><td align="center">2</td><td align="center">16/17</td><td align="center">0/17</td><td align="center">1/17</td><td align="center">0.5</td></tr><tr><td align="center">Gentamicin</td><td align="center">5/9</td><td align="center">0/9</td><td align="center">4/9</td><td align="center">16</td><td align="center">5/15</td><td align="center">5/15</td><td align="center">5/15</td><td align="center">16</td><td align="center">19/35</td><td align="center">6/35</td><td align="center">10/35</td><td align="center">16</td><td align="center">5/17</td><td align="center">4/17</td><td align="center">8/17</td><td align="center">16</td></tr><tr><td align="center">Imipenem</td><td align="center">5/9</td><td align="center">2/9</td><td align="center">2/9</td><td align="center">16</td><td align="center">6/15</td><td align="center">4/15</td><td align="center">5/15</td><td align="center">16</td><td align="center">1/35</td><td align="center">5/35</td><td align="center">29/35</td><td align="center">16</td><td align="center">0/17</td><td align="center">4/17</td><td align="center">13/17</td><td align="center">16</td></tr><tr><td align="center">Piperacillin/Tazobactam</td><td align="center">2/9</td><td align="center">1/9</td><td align="center">6/9</td><td align="center">128</td><td align="center">4/15</td><td align="center">1/15</td><td align="center">10/15</td><td align="center">128</td><td align="center">1/35</td><td align="center">1/35</td><td align="center">33/35</td><td align="center">128</td><td align="center">0/17</td><td align="center">3/17</td><td align="center">14/17</td><td align="center">128</td></tr><tr><td align="center"><bold><italic>Pseudomonas aeruginosa</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center">Amikacin</td><td align="center">15/29</td><td align="center">2/29</td><td align="center">12/29</td><td align="center">64</td><td align="center">18/33</td><td align="center">1/33</td><td align="center">14/33</td><td align="center">64</td><td align="center">8/25</td><td align="center">0/25</td><td align="center">17/25</td><td align="center">64</td><td align="center">7/11</td><td align="center">0/11</td><td align="center">4/11</td><td align="center">64</td></tr><tr><td align="center">Ceftazidime</td><td align="center">13/29</td><td align="center">2/29</td><td align="center">15/29</td><td align="center">64</td><td align="center">14/33</td><td align="center">5/33</td><td align="center">14/33</td><td align="center">64</td><td align="center">4/25</td><td align="center">7/25</td><td align="center">14/25</td><td align="center">64</td><td align="center">3/11</td><td align="center">1/11</td><td align="center">7/11</td><td align="center">64</td></tr><tr><td align="center">Ciprofloxacin</td><td align="center">11/29</td><td align="center">0/29</td><td align="center">18/29</td><td align="center">4</td><td align="center">14/33</td><td align="center">0/33</td><td align="center">19/33</td><td align="center">4</td><td align="center">8/25</td><td align="center">0/25</td><td align="center">17/25</td><td align="center">4</td><td align="center">6/11</td><td align="center">0/11</td><td align="center">5/11</td><td align="center">4</td></tr><tr><td align="center">Colistin</td><td align="center">29/29</td><td align="center">0/29</td><td align="center">0/29</td><td align="center">2</td><td align="center">33/33</td><td align="center">0/33</td><td align="center">0/33</td><td align="center">2</td><td align="center">25/25</td><td align="center">0/25</td><td align="center">0/25</td><td align="center">2</td><td align="center">11/11</td><td align="center">0/11</td><td align="center">0/11</td><td align="center">2</td></tr><tr><td align="center">Imipenem</td><td align="center">11/29</td><td align="center">14/29</td><td align="center">4/29</td><td align="center">16</td><td align="center">19/33</td><td align="center">9/33</td><td align="center">5/33</td><td align="center">16</td><td align="center">10/25</td><td align="center">6/25</td><td align="center">9/25</td><td align="center">16</td><td align="center">6/11</td><td align="center">4/11</td><td align="center">1/11</td><td align="center">16</td></tr><tr><td align="center">Piperacillin/Tazobactam</td><td align="center">26/29</td><td align="center">0/29</td><td align="center">3/29</td><td align="center">64</td><td align="center">26/33</td><td align="center">1/33</td><td align="center">6/33</td><td align="center">128</td><td align="center">7/25</td><td align="center">11/25</td><td align="center">7/25</td><td align="center">128</td><td align="center">9/11</td><td align="center">0/11</td><td align="center">2/11</td><td align="center">128</td></tr><tr><td align="center"><bold><italic>Klebsiella pneumoniae</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center">Ceftazidime</td><td align="center">5/7</td><td align="center">0/7</td><td align="center">2/7</td><td align="center">64</td><td align="center">23/34</td><td align="center">0/34</td><td align="center">11/34</td><td align="center">64</td><td align="center">14/29</td><td align="center">1/29</td><td align="center">14/29</td><td align="center">64</td><td align="center">2/11</td><td align="center">0/11</td><td align="center">9/11</td><td align="center">64</td></tr><tr><td align="center">Ciprofloxacin</td><td align="center">6/7</td><td align="center">0/7</td><td align="center">1/7</td><td align="center">1</td><td align="center">22/34</td><td align="center">1/34</td><td align="center">11/34</td><td align="center">4</td><td align="center">13/29</td><td align="center">0/29</td><td align="center">16/29</td><td align="center">4</td><td align="center">7/11</td><td align="center">0/11</td><td align="center">4/11</td><td align="center">4</td></tr><tr><td align="center">Meropenem</td><td align="center">7/7</td><td align="center">0/7</td><td align="center">0/7</td><td align="center">0.25</td><td align="center">33/34</td><td align="center">0/34</td><td align="center">1/34</td><td align="center">0.5</td><td align="center">21/29</td><td align="center">4/29</td><td align="center">4/29</td><td align="center">8</td><td align="center">2/11</td><td align="center">1/11</td><td align="center">8/11</td><td align="center">16</td></tr><tr><td align="center">Cefepime</td><td align="center">7/7</td><td align="center">0/7</td><td align="center">0/7</td><td align="center">2</td><td align="center">23/34</td><td align="center">1/34</td><td align="center">10/34</td><td align="center">64</td><td align="center">14/29</td><td align="center">7/29</td><td align="center">8/29</td><td align="center">64</td><td align="center">2/11</td><td align="center">0/11</td><td align="center">9/11</td><td align="center">64</td></tr><tr><td align="center">Cefoxitin</td><td align="center">7/7</td><td align="center">0/7</td><td align="center">0/7</td><td align="center">4</td><td align="center">24/34</td><td align="center">0/34</td><td align="center">10/34</td><td align="center">64</td><td align="center">11/29</td><td align="center">0/29</td><td align="center">18/29</td><td align="center">64</td><td align="center">2/11</td><td align="center">0/11</td><td align="center">9/11</td><td align="center">64</td></tr><tr><td align="center">Tobramycin</td><td align="center">7/7</td><td align="center">0/7</td><td align="center">0/7</td><td align="center">1</td><td align="center">26/34</td><td align="center">0/34</td><td align="center">8/34</td><td align="center">16</td><td align="center">12/29</td><td align="center">1/29</td><td align="center">16/29</td><td align="center">16</td><td align="center">2/11</td><td align="center">2/11</td><td align="center">7/11</td><td align="center">16</td></tr><tr><td align="center">Piperacillin/Tazobactam</td><td align="center">7/7</td><td align="center">0/7</td><td align="center">0/7</td><td align="center">8</td><td align="center">26/34</td><td align="center">1/34</td><td align="center">7/34</td><td align="center">128</td><td align="center">14/29</td><td align="center">2/29</td><td align="center">13/29</td><td align="center">128</td><td align="center">2/11</td><td align="center">0/11</td><td align="center">9/11</td><td align="center">128</td></tr><tr><td align="center"><bold><italic>Escherichia coli</italic></bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center">Amikacin</td><td align="center">35/35</td><td align="center">0/35</td><td align="center">0/35</td><td align="center">2</td><td align="center">57/57</td><td align="center">0/57</td><td align="center">0/57</td><td align="center">2</td><td align="center">43/47</td><td align="center">4/47</td><td align="center">0/47</td><td align="center">4</td><td align="center">29/30</td><td align="center">0/30</td><td align="center">1/30</td><td align="center">4</td></tr><tr><td align="center">Ciprofloxacin</td><td align="center">28/35</td><td align="center">0/35</td><td align="center">7/35</td><td align="center">4</td><td align="center">54/57</td><td align="center">0/57</td><td align="center">3/57</td><td align="center">0.5</td><td align="center">43/47</td><td align="center">0/47</td><td align="center">4/47</td><td align="center">4</td><td align="center">21/30</td><td align="center">0/30</td><td align="center">9/30</td><td align="center">4</td></tr><tr><td align="center">Piperacillin/Tazobactam</td><td align="center">34/35</td><td align="center">0/35</td><td align="center">1/35</td><td align="center">4</td><td align="center">56/57</td><td align="center">1/57</td><td align="center">0/57</td><td align="center">4</td><td align="center">43/47</td><td align="center">1/47</td><td align="center">3/47</td><td align="center">8</td><td align="center">28/30</td><td align="center">1/30</td><td align="center">1/30</td><td align="center">4</td></tr><tr><td align="center">Ceftazidime</td><td align="center">33/35</td><td align="center">0/35</td><td align="center">2/35</td><td align="center">2</td><td align="center">57/57</td><td align="center">0/57</td><td align="center">0/57</td><td align="center">1</td><td align="center">43/47</td><td align="center">3/47</td><td align="center">1/47</td><td align="center">1</td><td align="center">25/30</td><td align="center">0/30</td><td align="center">5/30</td><td align="center">2</td></tr><tr><td align="center">Cefoxitin</td><td align="center">31/35</td><td align="center">1/35</td><td align="center">3/35</td><td align="center">16</td><td align="center">56/57</td><td align="center">1/57</td><td align="center">0/57</td><td align="center">4</td><td align="center">43/47</td><td align="center">4/47</td><td align="center">0/47</td><td align="center">4</td><td align="center">26/30</td><td align="center">1/30</td><td align="center">3/30</td><td align="center">16</td></tr><tr><td align="center">Meropenem</td><td align="center">34/35</td><td align="center">1/35</td><td align="center">0/35</td><td align="center">0.25</td><td align="center">57/57</td><td align="center">0/57</td><td align="center">0/57</td><td align="center">0.25</td><td align="center">46/47</td><td align="center">0/47</td><td align="center">1/47</td><td align="center">0.25</td><td align="center">30/30</td><td align="center">0/30</td><td align="center">0/30</td><td align="center">2</td></tr></tbody></table><table-wrap-foot><p>Abbreviations: <bold>S </bold>= susceptible, <bold>I </bold>= intermediate, <bold>*R </bold>= resistant, <sup>#</sup><bold>MIC</bold><sub><bold>90 </bold></sub>= minimum inhibitory concentration for 90% of the corresponding microbial population</p></table-wrap-foot></table-wrap><p>The antimicrobial susceptibility pattern of <italic>Klebsiella pneumoniae </italic>isolates changed significantly during our study. Increased resistance of <italic>Klebsiella pneumoniae </italic>isolates was noted for all beta lactams tested [specifically to piperacillin/tazobactam (p < 0.001), ceftazidime (p = 0.01), cefepime (p < 0.001), cefoxitin (p < 0.001) and meropenem (p < 0.001)] between the first and second period of the study. There was also increased resistance of <italic>Klebsiella pneumoniae </italic>to ciprofloxacin (p = 0.006) and tobramycin (p < 0.001).</p><p>Regarding the antimicrobial susceptibility pattern of Gram-positive bacteria during our study there was a considerable proportion of staphylococci with resistance against oxacillin (Table <xref ref-type="table" rid="T1">1</xref>); however, the difference of the proportions of oxacillin resistant staphylococci between the two study periods was not statistically significant (Table <xref ref-type="table" rid="T2">2</xref>). We did not isolate any staphylococci with resistance to vancomycin. <italic>Enterococcus faecalis </italic>and <italic>Enterococcus faecium </italic>were generally susceptible to vancomycin although some strains were resistant; however the difference of the proportions of VRE between the two study periods was not statistically significant (Table <xref ref-type="table" rid="T2">2</xref>).</p></sec><sec><title>Discussion</title><p>Patients with bacteremia have remained a challenge to treat. Knowledge of the hospital epidemiology and antimicrobial susceptibility pattern of blood isolates helps physicians to effectively manage blood stream infections. This is because considerable differences of the frequency of blood isolates are reported even from hospitals of similar size and mixture of patients of the same country [<xref ref-type="bibr" rid="B6">6</xref>].</p><p>In this study we evaluated the secular trends of the relative frequency and antimicrobial resistance of blood isolates in a newly founded Greek hospital. Gram-positive microorganisms are the most common blood isolates. Among them, coagulase negative staphylococci are the commonest blood isolates. The percentage of coagulase negative staphylococci (%) is higher in our study than that reported in large series (31.6%) [<xref ref-type="bibr" rid="B7">7</xref>-<xref ref-type="bibr" rid="B9">9</xref>]. It is possible that the proportion of coagulase negative staphylococci that were contaminants was considerable in our study. The interpretation of blood cultures that are positive for coagulase negative staphylococci has inherent difficulties and requires careful reasoning [<xref ref-type="bibr" rid="B10">10</xref>]. The observed relative frequency of MRSA was considerable high during the studied period. Data from the WHONET Greece (antimicrobial surveillance system) regarding the period from January 2005 through June 2005 showed that a significant proportion of <italic>S. aureus </italic>blood isolates are resistant to methicillin (MRSA strains). Specifically, 32.6%, 55.6%, and 69% of <italic>S. aureus </italic>blood isolates from medical wards, surgical wards, and ICUs respectively were MRSA.</p><p>In general, our results about the relative frequency of blood isolates in our newly founded hospital are not substantially different from those of hospitals that are functioning for a much longer period of time. Similar data have been reported in studies performed in hospitals elsewhere in Europe as well as in North America [<xref ref-type="bibr" rid="B7">7</xref>-<xref ref-type="bibr" rid="B9">9</xref>]. An explanation may be that it is not the microbial ecology of the structure (our newly founded tertiary urban hospital compared to hospitals that are functioning for longer time) but rather the characteristics of the admitted patients like comorbidity, medications, and other host factors that play the most important role in the relative frequency of blood isolates.</p><p>It is also noteworthy that the isolation of Candida spp from the blood was not uncommon during the study period. This is in agreement with the reports from all over the world regarding a considerable prevalence of fungemia due to extensive use of antibiotics, aggressive treatment of neoplastic disease, an expanding population of patients with AIDS with prolonged survivors, use of indwelling devices for ICU monitoring, and many other factors that predispose to fungal infections [<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B9">9</xref>]. Although our hospital does not have a transplant unit, the observed high frequency of Candida isolates is probably explained by the fact that oncology patients and thus neutropenic patients constitute a significant portion of our patients.</p><p>We also evaluated in our study the trends of the antimicrobial resistance of the blood bacteria isolates in our newly founded hospital. The antimicrobial resistance of <italic>Acinetobacter baumannii </italic>showed an alarming increase during the study. <italic>Acinetobacter baumannii </italic>remained susceptible to colistin during the two periods, although the recovery of one resistant strain is of note [<xref ref-type="bibr" rid="B11">11</xref>,<xref ref-type="bibr" rid="B12">12</xref>]. Unfortunately, antimicrobial resistance increased also for <italic>Klebsiella pneumoniae </italic>for all of the 7 antibiotics it was tested for. These results are in concordance with data of the literature about the increasing antimicrobial resistance of Gram-negative bacteria [<xref ref-type="bibr" rid="B13">13</xref>-<xref ref-type="bibr" rid="B15">15</xref>]. In addition, it is noteworthy that the majority of bloodstream <italic>K. pneumoniae </italic>strains recovered in 2005 were resistant to meropenem, however this would probably reflect a nosocomial outbreak of a carbapenem-resistant <italic>K. pneumoniae </italic>clone.</p><p>We should acknowledge several limitations of our study. First, the results obtained from the Vitek II were confirmed by the E-test methodology only for colistin. Second, we did not proceed to the interpretation of the results of this study in terms of culture contamination or clinically relevant bloodstream infection. Third, pulsed-field gel electrophoresis was not performed to identify epidemic clones. Since molecular typing was not performed some of the studied isolates with antimicrobial resistance may be clonally related. Fourth, the number of patients visit the outpatient clinic of the hospital was not readily available. However, the number of positive blood cultures in the ambulatory outpatients is relatively small [<xref ref-type="bibr" rid="B16">16</xref>].</p></sec><sec><title>Conclusion</title><p>Our data suggest that the relative frequency and the antimicrobial resistance pattern of the blood isolates in a newly founded hospital is not very different from those data described in the literature from other older hospitals around the world. In addition, an alarming increase of antimicrobial resistance was noted during our study for Gram-negative bacteria, especially <italic>Acinetobacter baumannii </italic>and <italic>Klebsiella pneumoniae</italic>.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>MEF conceived the idea for the study. SKK, PM, GG, DN, and PIF collected the data. MEF and PIF drafted the manuscript. All authors made revisions of the manuscript and approved its final version.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1471-2334/6/99/prepub"/></p></sec>
Adherence with isoniazid for prevention of tuberculosis among HIV-infected adults in South Africa	<sec><title>Background</title><p>Tuberculosis (TB) is the most common opportunistic infection in HIV-infected adults in developing countries. Isoniazid (INH) is recommended for treatment of latent TB infection, however non-adherence is common. The purpose of this study was to apply in-house prepared isoniazid (INH) urine test strips in a clinical setting, and identify predictors of positive test results in an adherence questionnaire in HIV-infected adults taking INH for prevention of TB.</p></sec><sec sec-type="methods"><title>Methods</title><p>Cross-sectional study of adherence using a questionnaire and urine test strips for detection of INH metabolites at two hospitals in Pietermaritzburg, South Africa. Participants were aged at least 18 years, HIV positive, and receiving INH for prevention of tuberculosis disease. Univariate and multivariate analyses are used to identify factors relevant to adherence.</p></sec><sec><title>Results</title><p>301 consecutive patients were recruited. 28% of participants had negative urine tests. 32 (37.2%, 95% CI25.4, 45.0) of the 86 patients who received INH from peripheral pharmacies said the pharmacy had run out of INH at some time, compared with central hospital pharmacies (p = 0.0001). In univariate analysis, a negative test was associated with self-reported missed INH doses (p = 0.043). Each 12-hour increment since last reported dose increased the likelihood of a negative test by 34% (p = 0.0007). Belief in INH safety was associated with a positive test (p = 0.021). In multivariate analysis, patients who believed INH is important for prevention of TB disease were more likely to be negative (p = 0.0086).</p></sec><sec><title>Conclusion</title><p>Adequate drug availability at peripheral pharmacies remains an important intervention for TB prevention. Key questions may identify potentially non-adherent patients. In-house prepared urine tests strips are an effective and cheap method of objectively assessing INH adherence, and could be used an important tool in TB control programs.</p></sec>	<contrib id="A1" equal-contrib="yes" corresp="yes" contrib-type="author"><name><surname>Szakacs</surname><given-names>Tom A</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" equal-contrib="yes" contrib-type="author"><name><surname>Wilson</surname><given-names>Douglas</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" equal-contrib="yes" contrib-type="author"><name><surname>Cameron</surname><given-names>D William</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A4" equal-contrib="yes" contrib-type="author"><name><surname>Clark</surname><given-names>Michael</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Kocheleff</surname><given-names>Paul</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Muller</surname><given-names>F James</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A7" equal-contrib="yes" contrib-type="author"><name><surname>McCarthy</surname><given-names>Anne E</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Infectious Diseases	<sec><title>Background</title><p>Tuberculosis (TB) remains a major global epidemic despite widespread awareness and effective prevention and therapy. Human immunodeficiency virus (HIV) infection has had a devastating effect on the TB epidemic. In South Africa an estimated two million people are co-infected, as are 60% of all patients with newly diagnosed TB [<xref ref-type="bibr" rid="B1">1</xref>].</p><p>Accordingly, the World Health Organization and American Thoracic Society now recommend INH for treatment of latent TB infection (LTBI) in HIV-infected patients who are TST positive [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B3">3</xref>]. If skin testing is not feasible, INH is still recommended in regions with high TB prevalence [<xref ref-type="bibr" rid="B2">2</xref>].</p><p>Studies have shown rates of non-adherence to be typically 8–33% [<xref ref-type="bibr" rid="B4">4</xref>-<xref ref-type="bibr" rid="B13">13</xref>]. One objective measure of adherence is a biochemical test called the Arkansas method, where a chemical reaction with urinary INH metabolites produces a visible blue colour change [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B15">15</xref>]. This method has a high sensitivity (>99%) and specificity (>96%) [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B15">15</xref>]. While commercial test strips are not readily available in developing countries, they can be reproduced in-house at very little cost, without compromising quality [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B15">15</xref>]. Compared to commercial test strips, in-house prepared strips have a sensitivity and specificity of 99.5% and 96.4% respectively [<xref ref-type="bibr" rid="B6">6</xref>]. Their low cost, ease of use, and accuracy make them an ideal evaluative component of TB control programs.</p><p>The aim of our study was to apply in-house prepared INH urine test strips in a clinical setting in HIV-infected adults taking INH for prevention of TB, and to identify predictors of positive urine test results in an adherence questionnaire at two South African hospitals.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Setting</title><p>The study sample consisted of consecutive outpatients from HIV clinics at two hospitals in Pietermaritzburg, South Africa. Hospital A provides tertiary care and is located in an affluent suburban area; attendees pay US $3 for a consultation or medication. Hospital B is a district hospital in an impoverished suburban area, with limited access to specialist care and significantly less modern facilities. Patients may attend either hospital's HIV clinic; the same physicians staff clinics at both sites. Few patients were on antiretroviral therapy at the time of the study as the anti-retroviral rollout had not yet occurred.</p><p>Patients fill their INH prescription at the hospital pharmacy or at community pharmacies that are funded by the provincial government.</p><p>Regular clinic attendees were asked to participate in the study if they were 18 years or greater, HIV infected and prescribed INH (300 mg daily) with pyridoxine for prevention of TB. Clinic policy was not to perform TST prior to initiating therapy, and to continue treatment indefinitely. Patients receiving INH as part of multi-drug therapy for active TB and those with frank hematuria were excluded from the study. All participants gave written informed consent. The questionnaire and urine test were performed once in each patient with no follow up testing.</p></sec><sec sec-type="methods"><title>Procedures</title><p>Consecutive patients who met entry criteria were invited to participate by a study nurse. For those accepting, a nurse administered the questionnaire and performed the urine test prior to their appointment with the physician. One question asked participants to explain why they were taking INH. A correct response was if they indicated the INH was either to prevent or treat TB.</p><p>The preliminary questionnaire was developed from review of previous studies of factors affecting adherence [<xref ref-type="bibr" rid="B6">6</xref>-<xref ref-type="bibr" rid="B9">9</xref>,<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B16">16</xref>-<xref ref-type="bibr" rid="B20">20</xref>]. Additionally relevant questions were subsequently added after review by all the authors. The questionnaire was available in English or a translated Zulu form depending on patient preference.</p><p>Test strips were prepared via the description by Kilburn et al [<xref ref-type="bibr" rid="B15">15</xref>]. Filter paper [<xref ref-type="bibr" rid="B21">21</xref>] (cut into strips 6 × 0.8 cm) was marked at one end with a graphite pencil. The following stock solutions were prepared: 5% barbituric acid (pH adjusted to 5.2 using 50% sodium hydroxide), 60% potassium thiocyanate in 8% citric acid, and 50% chloramine T. Distilled water was the base medium in all solutions. The barbituric acid and chloramine T solutions were warmed prior to spotting to aid in solubilization. Each strip was spotted with 25 μl in three separate bands, with chloramine T and barbituric acid at the ends and potassium thiocyanate in the middle. The barbituric acid was spotted over the end of the strip previously marked; the strips were then air-dried and stored. In the present study, a single individual (TAS) made 500 test strips in four hours. There were no special laboratory needs; only the chemicals and standard lab equipment were required. Each strip cost less than 1 US cent to prepare.</p><p>Midstream urine was collected in a standard urine bottle. To detect INH, 0.5 ml of patient urine was pipetted from the urine bottle into a test tube, a test strip was added and a stopper applied. The strips were placed with the barbituric acid end downward. After one hour the urine and test strip were read. Patients' urine was deemed positive (INH metabolite present) if the urine or test strip turned a blue-green to dark blue colour, and negative (no INH metabolite present) if the urine and test strip remained yellow. Urine testing was performed just prior to the questionnaire. The questionnaire and urine test were completed on each participant once, with no further follow up testing. For the purpose of this study, patients were deemed to be adherent or non-adherent based on the result of the urine test. To ensure that the test strips were grossly effective, prior to initiating patient recruitment, urine test strips were pilot tested by 5 volunteers from the Department of Medicine at Hospital A. Each volunteer took INH 300 mg once, and had urine samples checked at time 0, 24, 36 and 72 hours. The color of the urine and test strip was recorded.</p><p>This study is in compliance with the Helsinki Declaration and received ethical approval from the Ottawa Hospital Regional Ethics Board and the Research Ethics Committee, Nelson R. Mandela School of Medicine, University of KwaZulu Natal.</p></sec><sec><title>Statistical analysis</title><p>Descriptive statistics were generated for the overall patient population, including means and standard deviations for continuous variables, and proportions and 95% confidence limits for categorical variables. These statistics were then calculated for each hospital. Means and proportions were compared using t-tests and chi-square testing, respectively. Exploratory analyses were conducted on variables relevant to medication availability and provision in dispensing pharmacies. Associations between variables and urine test results (positive or negative) were assessed with univariate and multivariate logistic regression analyses. In multivariate analyses, a forward selection procedure was used, beginning with variables that were found to be significant predictors in univariate analyses. Variables were entered into the model if p < 0.15, and taken out if p > 0.15. The model's overall fit to the data was assessed with Hosmer and Lemeshow goodness-of-fit testing. All of the above analyses were conducted using SAS 8 statistical software (SAS Institute, Cary, NC, USA).</p></sec></sec><sec><title>Results</title><p>During the urine test strip pilot testing by 5 volunteers, all samples were positive at 24 hours, and became negative at 72 hours, consistent with the expected pattern.<sup>14 </sup>The individual results of the tests are shown in <xref ref-type="supplementary-material" rid="S1">Additional file 1</xref>.</p><p>Between February and May 2004, 304 consecutive patients meeting entry criteria were identified. Consent was obtained from 301 of these patients, while 3 patients refused to participate. All of the questionnaire items and their responses are shown in <xref ref-type="supplementary-material" rid="S2">Additional file 2</xref>. The full English version of the questionnaire is shown in <xref ref-type="supplementary-material" rid="S3">Additional file 3</xref>.</p><p>The mean age of study participants was 35, with 76.7% being female. Each household had an average of 6.3 people and a total monthly income of US $238. The mean duration since HIV diagnosis was 33.3 months (SD 26) and mean time on INH therapy was 18.7 months (SD 17). On average, participants had told 3.0 people they were taking INH and 3.1 people that they have HIV, while a substantial portion (20.7% and 17.3% respectively) had not told anyone. A total of 18 patients (6%) were on anti-retrovirals.</p><p>Overall, 28.0% of urine tests were negative. There was no significant difference in urine results between the two hospitals; however, there were a number of qualitative differences (see <xref ref-type="supplementary-material" rid="S2">Additional file 2</xref>). In particular, there was considerable variation in where study participants get their INH. At Hospital A only 4 patients (4.0%, 95% CI 0.2, 7.9) attended peripheral pharmacies, while the remainder attended the hospital pharmacy. In contrast, at Hospital B, 86 patients (42.8%, 95% CI 35.9, 49.6) attended peripheral pharmacies. For patients getting their INH at Hospital A, 42.1% (95% CI 32.2, 52.0) said they sometimes could not afford to go to the pharmacy. This occurred in only 3.5% (95% CI 0.001, 0.0069) attending the Hospital B pharmacy, and 1.1% attending peripheral pharmacies (χ<sup>2 </sup>= 80.3, p = 0.0001). The proportion of patients who ran out of INH between hospital visits was 7.3%, 19.3%, and 26.4%, among patients receiving INH from Hospital A, Hospital B, and peripheral pharmacies, respectively (χ<sup>2 </sup>= 12.1, p = 0.0024). Only 1 patient at Hospital B and none at Hospital A reported their pharmacy had previously run out of INH. However, this occurred for 32 patients (37.2%, 95% CI 25.4, 45.0) attending peripheral pharmacies (χ<sup>2 </sup>= 78.3, p = 0.0001). Participants attended 25 different peripheral pharmacies; 8 of these serviced 73 of the total 86 patients (80.2%). No particular pharmacy was statistically associated with urine test results; however 41.1% of patients attending these 8 pharmacies said it had run out of medications at some time. At the most frequented pharmacy, 10 of 18 patients (55.6%) reported this had occurred.</p><p>In univariate logistic regression analysis, three factors were found to have a statistically significant impact on urine test results (Table <xref ref-type="table" rid="T1">1</xref>). Longer time since last self-reported INH dose showed a stepwise increase in probability of having a negative urine test (OR 1.33; 95% CI, 1.13 to 1.57; p = 0.0007). This relationship is shown in Figure <xref ref-type="fig" rid="F1">1</xref>. Every 12-hour increase in reported time resulted in a 34% (95% CI 13%, 57%) increase in the probability of a negative urine test. Those who believed the INH was safe were less likely to have a negative urine test (OR 0.47; 95% CI, 0.25 to 0.90; p = 0.021). There was also a non-significant trend towards increased compliance in those who believed INH was not safe (OR 0.71; 95% CI, 0.29 to 1.75; p = 0.45). Finally, patients who reported they sometimes forget to take the INH were more likely to have negative tests (OR 2.0; 95% CI, 1.02–3.99; p = 0.043).</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Univariate logistic regression analyses of study variables showing a significant association with a negative urine test result</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left"><bold>Variable</bold></td><td align="center"><bold>OR (95% CI)</bold></td><td align="center"><bold>p value</bold></td></tr></thead><tbody><tr><td align="left"><underline>How long since last INH dose? (12-hour increments)?</underline></td><td align="center">1.33 (1.13, 1.57)</td><td align="center">0.0007</td></tr><tr><td align="left"><underline>Statement: INH is dangerous to your health</underline></td><td></td><td></td></tr><tr><td align="left">Strongly agree/agree</td><td align="center">0.71 (0.29, 1.75)</td><td align="center">0.45</td></tr><tr><td align="left">Disagree/strongly disagree</td><td align="center">0.47 (0.25, 0.90)</td><td align="center">0.021</td></tr><tr><td align="left">Don't know</td><td align="center">1.00</td><td></td></tr><tr><td align="left"><underline>Do you ever forget to take INH?</underline></td><td align="center">2.0 (1.02, 3.99)</td><td align="center">0.043</td></tr></tbody></table></table-wrap><fig position="float" id="F1"><label>Figure 1</label><caption><p>Urine result based on time since last self-reported INH dose (95% CI shown in bars).</p></caption><graphic xlink:href="1471-2334-6-97-1"/></fig><p>Time since last reported INH dose and probability of a negative test remained unchanged in all multivariate analyses (Table <xref ref-type="table" rid="T2">2</xref>). As well, belief that INH is safe continued to decrease the likelihood of a negative urine test (p = 0.0098). Overall, only nine patients in the study answered that their chance of getting TB without INH were average or below average. Those who felt their chances of getting sick from TB without INH were above average or high were more likely to have negative urine results, when compared to those who didn't know (OR 3.00; 95% CI, 1.32 to 6.79; p = 0.0086). The variable of patient forgetfulness in taking the INH was not found to be significant in the multivariate analysis.</p><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Multivariate logistic regression model of study variables predicting a negative urine test result</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left"><bold>Variable</bold></td><td align="center"><bold>Parameter estimate</bold></td><td align="center"><bold>OR (95% CI)</bold></td><td align="center"><bold>p value</bold></td></tr></thead><tbody><tr><td align="left">Intercept</td><td align="center">-1.84</td><td></td><td align="center">0.0064</td></tr><tr><td align="left"><underline>How long since last INH dose? (12-hour increments)?</underline></td><td align="center">0.30</td><td align="center">1.34 (1.13, 1.59)</td><td align="center">0.0007</td></tr><tr><td align="left"><underline>Statement: INH is dangerous to your health</underline></td><td></td><td></td><td></td></tr><tr><td align="left">Strongly agree/agree</td><td align="center">-0.48</td><td align="center">0.62 (0.21, 1.81)</td><td align="center">0.38</td></tr><tr><td align="left">Disagree/strongly disagree</td><td align="center">-1.08</td><td align="center">0.34 (0.15, 0.77)</td><td align="center">0.0098</td></tr><tr><td align="left">Don't know</td><td align="center">0</td><td align="center">1.00</td><td></td></tr><tr><td align="left"><underline>Statement: Without INH, your chance of getting sick from TB is:</underline></td><td></td><td></td><td></td></tr><tr><td align="left">High/above average</td><td align="center">1.10</td><td align="center">3.00 (1.32, 6.79)</td><td align="center">0.0086</td></tr><tr><td align="left">Average/below average</td><td align="center">0.96</td><td align="center">2.61 (0.49, 13.8)</td><td align="center">0.26</td></tr><tr><td align="left">Don't know</td><td align="center">0</td><td align="center">1.00</td><td></td></tr></tbody></table><table-wrap-foot><p>Goodness-of-fit statistic = 4.28 (p = 0.83)</p></table-wrap-foot></table-wrap></sec><sec><title>Discussion</title><p>Urine testing measured the overall rate of non-adherence at 28%. This is similar to other studies that have shown rates varying between 8–33% [<xref ref-type="bibr" rid="B4">4</xref>-<xref ref-type="bibr" rid="B13">13</xref>]. However, there are few consistent predictors of adherence between studies. Some show that homelessness, alcoholism, adherence counselling, race, prior BCG, sex, and age had significant influences [<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B16">16</xref>-<xref ref-type="bibr" rid="B18">18</xref>]. Others point to social factors [<xref ref-type="bibr" rid="B19">19</xref>,<xref ref-type="bibr" rid="B20">20</xref>], and still others have found no correlative factors [<xref ref-type="bibr" rid="B5">5</xref>,<xref ref-type="bibr" rid="B6">6</xref>]. In the present study a number of predictive factors were found, including the self reported time since last INH dose. When asked how frequently they take the INH, 95.3% of participants said daily, reflecting their understanding that INH should be taken daily. If broadly asked about adherence in judgmental or threatening contexts, one may not be straightforward. However, a less threatening question, such as time since last dose is valid in comparison with urine testing. Use of 12-hour time increments provides an easy estimate of the probability of a negative urine test. This confirms the same relationship that was found in a study of LTBI by Perry et al [<xref ref-type="bibr" rid="B8">8</xref>].</p><p>Patients who report they sometimes forget to take their medication were more likely to have a negative urine test. This was previously identified as a predictor in another study of patients on INH for LTBI [<xref ref-type="bibr" rid="B9">9</xref>]. There, 33% admitted to forgetting to take the INH, while this occurred in 14% of our study.</p><p>Those who expressed any opinion as to the safety of INH were less likely to test negative (i.e., more likely adherent to therapy). Participants who disagreed that INH was a danger to their health were significantly less likely to test negative in univariate and multivariate analyses, than those who stated they did not know. However, it is interesting that those who agreed that INH is dangerous were also less likely to test negative, without statistical significance. This may reflect the small number of respondents who "strongly agreed" or "agreed" with the statement (1.7% and 8.7% of respondents, respectively), or may represent lack of understanding of the question.</p><p>Patients were on INH for an average of 18.7 months, longer than the typical 6–12 month course [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B3">3</xref>]. In Pietermaritzburg, longer therapy is used as primary and secondary prophylaxis to prevent new infection from ongoing exposure. Although this contrasts to current international guidelines, there are studies supporting this approach [<xref ref-type="bibr" rid="B22">22</xref>-<xref ref-type="bibr" rid="B24">24</xref>]. One study in South Africa examined indefinite secondary therapy in HIV-infected patients with prior active TB and found a 55% reduction in the incidence of TB recurrence [<xref ref-type="bibr" rid="B22">22</xref>].</p><p>Unlike other studies where INH side effects impacted adherence [<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B16">16</xref>] we did not find this relationship. In fact, only 6% reported this, compared to up to 74% in other studies [<xref ref-type="bibr" rid="B13">13</xref>]. While this may be due to selection of patients that had been able to tolerate INH for a prolonged time, univariate and multivariate testing did not find that duration of treatment affected adherence.</p><p>In multivariate analysis, those believing they have a high or above average chance of getting active TB without INH were more likely non-adherent. While this seems paradoxical, it might be explained by social factors. Patients may perceive stigma or conceal their condition and its treatment, which could interfere with adherence. Alternatively, knowledge of the danger of TB may motivate them to share or sell their INH to others who are also at risk. As this relationship is counter-intuitive it merits further study.</p><p>To our knowledge, ours is the first study to identify location of drug supply as a factor in adherence. An alarming number of peripheral pharmacies are reported to run out of medications, which may impair overall adherence. Although pharmacy location did not significantly impact urine test results, the study was not designed to detect this difference. Regardless, this identification of impaired access to medicines may represent an important barrier to overall adherence.</p><p>In areas where patients attend peripheral pharmacies, TB programs should include resources to routinely audit and address supply problems. Alternatively, it may be necessary to subsidize or facilitate transportation to hospital pharmacies. These conclusions have additional relevance in the context of the South African antiretroviral rollout, which ultimately will be delivered by primary health care facilities.</p><p>To our knowledge, only the present study and that by Meissner et al [<xref ref-type="bibr" rid="B6">6</xref>] have applied in-house prepared test strips. However, in our study we identified qualitative issues not described previously. Only when the barbituric acid end was placed in the urine would the reaction proceed appropriately, hence that end was marked with a pencil. Otherwise, strips with the chloramine T end downward were uniformly negative. Also, previous studies say results were available within 10–15 minutes [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B15">15</xref>]. While this was true for the majority, we found a small portion of results remained ambiguous, and when we extended the evaluation time to one hour some samples then became more clearly positive.</p><p>Some factors may have an impact on use of the urine test. The rate of INH metabolism is affected by liver acetylator status, with some patients exhibiting rapid drug metabolism. No study has attempted to address the impact of this status on the urine test results. Also, there are conflicting reports on impaired INH absorption in HIV and TB co-infection [<xref ref-type="bibr" rid="B25">25</xref>-<xref ref-type="bibr" rid="B27">27</xref>]. Lastly, it may seem attractive to extrapolate the urine test for assessing compliance in patients with active tuberculosis receiving multi-drug therapy. However, no study has formally examined the biochemical impact that other anti-tuberculosis medications may have on urine testing for INH metabolites.</p><p>Although the urine test is a single point measurement and does not necessarily reflect overall adherence, its biochemical characteristics make it a useful predictor. 48 hours after the last dose 76% of patients will have a positive result, while after 72 hours only 6% will remain positive [<xref ref-type="bibr" rid="B14">14</xref>]. In fact, one study showed that urine testing is an accurate estimate of overall adherence when compared to electronic monitoring of pill-bottle lid opening [<xref ref-type="bibr" rid="B28">28</xref>]. Another study performed random urine testing between scheduled appointments and found similar rates of adherence between visits as at appointments [<xref ref-type="bibr" rid="B9">9</xref>]. So, urine testing may give a realistic estimate of overall adherence and is not necessarily affected prior to clinic visits due to a "white coat effect".</p></sec><sec><title>Conclusion</title><p>Commercially available test strips use the same reaction but are expensive and difficult to acquire, while in-house prepared strips are easily produced and at very little cost. As the procedure is readily reproduced, thousands of strips could be routinely made to meet the testing needs of a given clinic population without significant time or cost.</p><p>Monitoring and enhancement of adherence should be part of TB treatment and control programs. In-house prepared urine test strips are inexpensive, effective and rapid, and could be used as a major evaluative tool in TB control programs.</p></sec><sec><title>Authors' contributions</title><p>TS led the design, on-site implementation, and drafting of the manuscript. PK and JM helped in study conception, design, and institutional facilitation. AM and WC were involved in study conception, design, and data analysis. DW participated in study design, analysis and on-site supervision. MC performed the statistical analysis. All authors contributed to drafting of the manuscript. All authors read and approved the final manuscript.</p></sec><sec><title>Declaration of competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1471-2334/6/97/prepub"/></p></sec><sec sec-type="supplementary-material"><title>Supplementary Material</title><supplementary-material content-type="local-data" id="S1"><caption><title>Additional File 1</title><p>Time-based urine result among volunteers taking a single-dose of INH 300 mg. Provides the results among 5 volunteers taking INH 300 mg once with urine testing at set intervals thereafter.</p></caption><media xlink:href="1471-2334-6-97-S1.doc" mimetype="application" mime-subtype="msword"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S2"><caption><title>Additional File 2</title><p>Responses to questionnaire overall and by hospital. Lists all of the questions of the questionnaire with the corresponding proportion of responses, separated based on hospital site.</p></caption><media xlink:href="1471-2334-6-97-S2.doc" mimetype="application" mime-subtype="msword"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S3"><caption><title>Additional File 3</title><p>Questionnaire. Copy of the actual questionnaire used in the study.</p></caption><media xlink:href="1471-2334-6-97-S3.doc" mimetype="application" mime-subtype="msword"><caption><p>Click here for file</p></caption></media></supplementary-material></sec>
Policy challenges from the "White" Senate inquiry into workplace-related health impacts of toxic dusts and nanoparticles	<p>On 22 June 2005 the Senate of the Commonwealth of Australia voted to establish an inquiry into workplace harm related to toxic dust and emerging technologies (including nanoparticles). The inquiry became known as the "White" Inquiry after Mr Richard White, a financially uncompensated sufferer of industrial sandblasting-induced lung disease who was instrumental in its establishment. The "White" Inquiry delivered its final report and recommendations on 31 May 2006. This paper examines whether these recommendations and their implementation may provide a unique opportunity not only to modernize relevant monitoring standards and processes, but related compensation systems for disease associated with workplace-related exposure to toxic dusts. It critically analyzes the likely role of the new Australian Safety and Compensation Council (ASCC) in this area. It also considers whether recommendations related to potential workplace related harm from exposure to nanoparticles could commence a major shift in Australian healthcare regulation.</p>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Faunce</surname><given-names>Thomas A</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Walters</surname><given-names>Haydn</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Williams</surname><given-names>Trevor</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Bryant</surname><given-names>David</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Jennings</surname><given-names>Martin</given-names></name><xref ref-type="aff" rid="I5">5</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Musk</surname><given-names>Bill</given-names></name><xref ref-type="aff" rid="I6">6</xref><email>[email protected]</email></contrib>	Australia and New Zealand Health Policy	<sec><title>Background</title><p>On 22 June 2005 the Senate of the Commonwealth of Australia voted to establish an inquiry into workplace harm related to toxic dust and emerging technologies (including nanoparticles). The inquiry has become known as the "White" Inquiry after Mr Richard White, a financially uncompensated sufferer of industrial sandblasting-induced lung disease whose sentinel case and advocacy through the Australian Sandblasting Diseases Coalition (ASDC) was instrumental in its establishment. Its final report was delivered on 31 May 2006 [<xref ref-type="bibr" rid="B1">1</xref>].</p><p>Numerous government inquiries into toxic dust exposure in Australia have been held since the turn of the century. They have focused, for example, on worker health problems associated with the Western Australia and Victorian goldmines and NSW and Queensland coalmines, as well as the Sydney sandstone industries (through the NSW Silicosis Board (now the Dust Diseases Board)) [<xref ref-type="bibr" rid="B2">2</xref>]. In 1993, the National Occupational Health and Safety Commission reviewed the occupational exposure standard for crystalline silica (silicon dioxide (SiO2) chiefly from abrasive blasting, excavating, quarrying or tunneling quartz and granite, but including the aggregates, sand, mortar, concrete and stone)[<xref ref-type="bibr" rid="B3">3</xref>].</p><p>There are, however, many reasons to consider that the level of disease associated with workplace disease from toxic dust continues to be much greater than revealed by such inquiries or reviews. First, the long latency period of the relevant diseases and the insidious nature of their acquisition, mean that the actual incidence and prevalence of workplace injury from toxic dusts remains largely unknown. Second, it remains unclear what exposure levels (mg/m3) effectively minimize injury[<xref ref-type="bibr" rid="B4">4</xref>]. Occupational exposure limits for crystalline silica, for example, are almost constantly under review worldwide because of the large numbers of exposed people[<xref ref-type="bibr" rid="B5">5</xref>]. The position is further complicated because accurate exposure monitoring and hence enforcement of standards, is generally more difficult at lower exposure levels[<xref ref-type="bibr" rid="B6">6</xref>].</p><p>Third, associated cigarette smoking often complicates causation analysis in workers regularly exposed to toxic dusts[<xref ref-type="bibr" rid="B7">7</xref>]. This is particularly true where the disease manifestation is lung cancer [<xref ref-type="bibr" rid="B8">8</xref>] and many potentially carcinogenic compounds are used in the workplace[<xref ref-type="bibr" rid="B9">9</xref>]. Fourth, many workers suffering disease associated with exposure to toxic dusts during employment will not seek, let alone receive compensation[<xref ref-type="bibr" rid="B10">10</xref>]. Fifth, the infrastructure for successful implementation of national standards (including the number of occupational hygenists in government employment) has been eroded and few if any Australian companies have been prosecuted for exposing workers to the risk of dust-related disease [<xref ref-type="bibr" rid="B2">2</xref>].</p><p>A sentinel case highlighting such problems was provided by Mr Richard White.</p></sec><sec><title>Richard White's silicosis: a sentinel case for policy change</title><p>As a teenager, Mr Richard White was employed, between 1971 and 1974 in the Northern Territory, to sandblast clean barges and fuel tanks, then spray them with an epoxy resin containing many of the carcinogens listed above. He did this without being issued respiratory protection. In the 1990's, Mr White began to notice severe dyspnoea, fatigue, paroxysmal coughing and sputum production, particularly when playing with his children. He suffered repeated chest infections and initiated a compensation claim in the Northern Territory Supreme Court. This alleged that as a result of this employment, he had developed silicosis and/or emphysema and/or chronic air flow limitation[<xref ref-type="bibr" rid="B11">11</xref>]. He was initially met with legal defence arguments about the liability of subsidiary companies, similar to those raised in the James Hardie asbestos litigation[<xref ref-type="bibr" rid="B12">12</xref>]. It is a matter of record that Mr White lost the first instance trial, the subsequent appeal to the Supreme Court of the Northern Territory and an appeal to the High Court of Australia. His history of mild smoking (reported as < 5 pack years) was a significant factor in the case against him, as was a defence-proffered allegation of asthma, for which there was no known medical history [<xref ref-type="bibr" rid="B13">13</xref>].</p><p>The histopathology of Mr White's subsequent open lung biopsy, revealed, as well as some adjacent emphysema:</p><p>"<italic>Scant brightly birefringent, needle-shaped crystalline material, consistent with inhaled silicate crystals, are noted within these macrophages. In addition, similar material is noted in macrophages within the lymphoid aggregates of the dust sumps, with a miniscule amount of associated interstitial fibrosis</italic>".</p><p>These results strongly suggested Mr White had industrially-related silica injury to his lungs, in the form of interstitial fibrosis, small airways disease and emphysema, giving rise to mainly fixed airflow obstruction. Mr White thereafter placed a newspaper advertisement. It requested people to contact him who knew or suspected they had acquired lung or other disease through working for companies that used sandblasting techniques. By Christmas 2004, Mr White had obtained almost a thousand names[<xref ref-type="bibr" rid="B13">13</xref>].</p><p>Many of these people claimed to have experienced symptoms consistent with debilitating lung diseases or cancer related to workplace exposure to toxic dusts. Very few had received or sought any compensation. Given the relatively informal way in which the list was been prepared, Mr White began to believe that many other Australians has suffered potentially harmful exposure to toxic workplace dust, without ever seeking more specific diagnosis or financial compensation[<xref ref-type="bibr" rid="B13">13</xref>].</p><p>Mr White and the first author commenced lobbying for a Senate Inquiry into the workplace risks of toxic dusts in Australia as the best way to initiate a change in health policy in this area. Late in June 2005 Senators Lyn Allison (Dem.) and Gary Humphries (Lib.) successfully obtained the majority required to initiate that Inquiry. Its terms of reference were expanded to include the potential of emerging technologies, including nanoparticles, to result in workplace related harm.[<xref ref-type="bibr" rid="B14">14</xref>] The recommendations of the "White" Senate Inquiry urge the new Australian Safety and Compensation Council (ASCC) toward some far-reaching changes in the system of prevention and monitoring of dust-related disease in Australia.</p></sec><sec><title>Policy recommendations for general workplace toxic dust exposure</title><p>Chronic Obstructive Pulmonary Disease ("COPD") is currently the fourth leading cause of death worldwide and the only chronic lung disease whose incidence in the developed world continues to increase. An analysis of 10 large scale studies (taking into account tobacco smoking status) in the US, France, Spain, Norway, the Netherlands, Italy, China and New Zealand, indicates that approximately 15% of the burden of illness from COPD arises from workplace exposure to toxic dust[<xref ref-type="bibr" rid="B15">15</xref>]. This is unlikely to be a "worst-case" estimate, rather, the studies reveal that much COPD is undiagnosed[<xref ref-type="bibr" rid="B16">16</xref>]. In Australia, recent epidemiological studies in middle aged Melbourne residents have shown that a relatively significant amount of COPD is not related to smoking, and industrial exposures are a significant contribution[<xref ref-type="bibr" rid="B17">17</xref>]. There is now good pathological evidence in both humans and animals of the capacity of crystalline silica to cause emphysema[<xref ref-type="bibr" rid="B18">18</xref>]. Australian healthcare policy and related monitoring and compensation systems appear, however, not to have fully appreciated the regulatory significance of this advance in causal understanding.</p><p>The British Coal litigation, on the other hand, was a watershed in the development of clinical and legal theories about causative relationships between industrial dust exposure and COPD. One of its major conclusions was that disability in a toxic dust-exposed cigarette smoker should not be regarded for compensation purposes as if it was entirely due to one cause or the other. Rather the courts in that nation decided they should attempt to estimate, as far as possible, the contribution of each such cause and then award compensation proportionally[<xref ref-type="bibr" rid="B19">19</xref>]. A related recommendation, posing an obvious challenge to Australian healthcare policy, was that compensation should prima facie be paid to any worker with COPD who has worked underground for 20 years, even in the absence of pneumoconiosis on chest x-ray[<xref ref-type="bibr" rid="B20">20</xref>].</p><p>The "White" Senate Inquiry recommended that the ASCC review the National Data Action Plan, to increase the availability of relevant data (Rec. 1). It required the ASCC to extend the Surveillance of Australian Work-Based Respiratory Events (SABRE) program Australia-wide, to provide mandatory reporting of dust-related disease (Rec. 2). Most importantly, it required the ASCC in conjunction with the Heads of Workplace Safety Authorities consider mechanisms to increase the number of occupational hygienists being trained and employed by regulators (Rec.8). The Minister for Employment and Workplace Relations is asked to raise with the Workplace Relations Ministers' Council the need to enact nationally consistent standards for identification, assessment and compensation for sufferers and their families based on at least the standard on the NSW <italic>Workers Compensation </italic>(<italic>Dust Diseases Act</italic>) <italic>1942</italic>, as well as removing restrictive statutes of limitation (Rec. 7, 9, 10 & 11)[<xref ref-type="bibr" rid="B1">1</xref>].</p><p>These recommendations, particularly the last, provide strong policy opportunities for an Attorney General with a firm interest in unifying areas of legal regulation in Australia and a Federal government willing to display its genuine concern for protecting worker safety in a period of considerable upheaval in workplace relations. They set a challenge for the ASCC that should be called to account for the steps taken toward their implementation by March 2007. State governments displaying a lack of willingness to become involved in this long overdue regulatory rationalization should also have their credentials on workplace safety called into question.</p></sec><sec><title>Policy challenges from workplace exposure to nanotechnology</title><p>Nanoparticles are very small. A nanometer is one-billionth of a metre (a human hair is 80,000 nm wide)[<xref ref-type="bibr" rid="B21">21</xref>]. Nanoparticles particularly used in transparent sunscreens and cosmetics, "smart" surveillance equipment, fertilizers and packaging, nutritionally enhanced foods, long-lasting paints and as industrial catalysts. They may also arise from thermal spraying, metal production and refining, welding, soldering and high speed metal grinding[<xref ref-type="bibr" rid="B22">22</xref>]. Medical nanotechnology involves the development of drug/invasive therapeutic device products controllable at atomic, molecular or macromolecular levels of approximately 1–100 nanometers. It is a rapidly expanding area of research globally with revolutionary implications for disease detection and analysis,[<xref ref-type="bibr" rid="B23">23</xref>] drug delivery,[<xref ref-type="bibr" rid="B24">24</xref>] and reconstructive,[<xref ref-type="bibr" rid="B25">25</xref>] neurological[<xref ref-type="bibr" rid="B26">26</xref>] and cardiac surgery[<xref ref-type="bibr" rid="B27">27</xref>]. Australian companies are already working, for example, on nanotechnology-based sunscreen, anticoagulant and drug delivery products[<xref ref-type="bibr" rid="B28">28</xref>].</p><p>Nanoparticles present unique health risks, being extremely reactive whilst readily penetrating mucosal membranes, entering blood vessels and impacting on the coagulation system. There are currently no effective methods to measure and assess exposure risks to nanoparticles in patients or healthcare workers. Nanoparticle exposure limits do not exist and manufacturers currently have no obligation to publish details of the safety checks imposed on their nanoproducts. A long latency period for disease from exposure to nanoparticles and the insidious symptom development, mean causation will be difficult to legally prove and compensation difficult to obtain.[<xref ref-type="bibr" rid="B29">29</xref>] Sketchy nanomedicine safety and toxicity profiles thus may create major policy challenges, not only for Australian Therapeutic Goods Administration (TGA) marketing approvals, but for Pharmaceutical Benefits Advisory Committee (PBAC) and Medical Services Advisory Committee (MSAC) cost-effectiveness evaluations, as well as the horizon scanning program of the Health Policy Advisory Committee on Technology (HealthPACT). Interdepartmental meetings, industry consultations and discussions on international harmonisation, have occurred on the health risks of nanotechnology generally (and will continue under the interdepartmental committee of the National Nanotechnology Strategy Taskforce ("NNST") within the Department of Industry, Tourism and Resources)[<xref ref-type="bibr" rid="B30">30</xref>]. A 2005 report to the Prime Minister's Science Engineering and Innovation Council warned: "The early introduction and explanation of regulation reduces the risk that public concern will prevent acceptance of nanotechnology. Industry also tends to prefer certainty in regulation"[<xref ref-type="bibr" rid="B31">31</xref>].</p><p>The "White" Senate Inquiry recommended that the National Nanotechnology Strategy be finalized as a matter of priority (Rec. 12). It required establishment of a widely consulting working party on nanotechnology regulation, comprising representatives of the Therapeutic Goods Administration, the National Industrial Chemicals Notification and Assessment Scheme (NICNAS), and the ASCC. This is to consider (with consideration of international models) the appropriateness of existing regulations, how gaps and uncertainties in that regulatory framework can be addressed and risk management incorporated, possible reassessment of safety and whether a permanent nanotechnology regulatory body needs to be established. (Rec. 12, 13 and 14)[<xref ref-type="bibr" rid="B1">1</xref>].</p><p>These recommendations present an opportunity for Australia to develop an innovative and practical regulatory framework that will not only facilitate the development of an important industry sector, but ensure the safety of workers and those members of the public associated with its products. Implementing these policy recommendations becomes a matter of national urgency, given the burgeoning level of research already taking place in this area in Australia.</p></sec><sec><title>Conclusion. Policy challenges for the ASCC</title><p>The new ASCC, replacing the National Occupational Health and Safety Commission (NOHSC), first met on 20 October 2005. The ASCC comprises representatives from Federal, State and Territory Governments, the Australian Council of Trade Unions (ACTU) and the Australian Chamber of Commerce and Industry (ACCI). One of its main aims is to provide policy advice to the Workplace Relations Minister's Council on OHS and workers' compensation arrangements. A related purpose is to deliver nationally consistent frameworks providing leadership and coordination to prevent workplace death, injury and disease through two working groups: an OHS Working Group, and a Workers Compensation Working Group[<xref ref-type="bibr" rid="B32">32</xref>].</p><p>There are many core policy challenges that the ASCC will have to address in implementing the recommendations of the "White" Senate Inquiry. The first is that the crucial problem of workplace-related disease from toxic dusts, both world-wide and in Australia, has not been one of creation of standards, but of their implementation[<xref ref-type="bibr" rid="B33">33</xref>]. Standards, for example, prohibiting abrasive sand blasting and providing recommended exposure limits for respirable crystalline silica, as well as respiratory protection, worker education and regular medical examinations have been in place in Australia since the late 1960's and early 1970's[<xref ref-type="bibr" rid="B34">34</xref>]. Since that time, funding has been reduced for inspectors and insufficient attention paid to increasing their powers. It is imperative that the ASCC give priority to evaluating and recommending to State and Federal governments the required numbers of occupational health and safety inspectors capable of enforcing any new national standards.</p><p>Second, the ASCC will need to address evidence presented to the "White" Senate Inquiry suggesting that toxic dust workplace exposure presents much greater health problems to the Australian community than is currently recognized. Community exposure through wind-borne dust and rainwater, for example, clearly has been insufficiently investigated.</p><p>Third, there is currently almost no credible research on the health impacts of nanoparticles despite increasing use of such technology in Australian industry. This data and regulatory gap must also be filled, thoroughly and competently and after broad consultations, by the ASCC.</p><p>Fourth, the ASCC needs to prioritise streamlining (and equity of access issues) for compensation processes linked with best scientific practice in the understanding of disease causation. Resolution of this issue will provide an important backdrop to implementation of existing and improved safety standards in the area of dust-related disease.</p></sec>
Colorectal cancer screening awareness among physicians in Greece	<sec><title>Background</title><p>Data comparison between SEER and EUROCARE database provided evidence that colorectal cancer survival in USA is higher than in European countries. Since adjustment for stage at diagnosis markedly reduces the survival differences, a screening bias was hypothesized. Considering the important role of primary care in screening activities, the purpose of the study was to investigate the colorectal cancer screening awareness among Hellenic physicians.</p></sec><sec sec-type="methods"><title>Methods</title><p>211 primary care physicians were surveyed by mean of a self-reported prescription-habits questionnaire. Both physicians' colorectal cancer screening behaviors and colorectal cancer screening recommendations during usual check-up visits were analyzed.</p></sec><sec><title>Results</title><p>Only 50% of physicians were found to recommend screening for colorectal cancer during usual check-up visits, and only 25% prescribed cost-effective procedures. The percentage of physicians recommending stool occult blood test and sigmoidoscopy was 24% and 4% respectively. Only 48% and 23% of physicians recognized a cancer screening value for stool occult blood test and sigmoidoscopy. Colorectal screening recommendations were statistically lower among physicians aged 30 or less (p = 0.012). No differences were found when gender, level and type of specialization were analyzed, even though specialists in general practice showed a trend for better prescription (p = 0.054).</p></sec><sec><title>Conclusion</title><p>Contemporary recommendations for colorectal cancer screening are not followed by implementation in primary care setting. Education on presymptomatic control and screening practice monitoring are required if primary care is to make a major impact on colorectal cancer mortality.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Xilomenos</surname><given-names>Apostolos</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Mauri</surname><given-names>Davide</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" corresp="yes" contrib-type="author"><name><surname>Kamposioras</surname><given-names>Konstantinos</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Gkinosati</surname><given-names>Athanasia</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Zacharias</surname><given-names>Georgios</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Sidiropoulou</surname><given-names>Varvara</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A7" contrib-type="author"><name><surname>Papadopoulos</surname><given-names>Panagiotis</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A8" contrib-type="author"><name><surname>Chatzimichalis</surname><given-names>Georgios</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A9" contrib-type="author"><name><surname>Golfinopoulos</surname><given-names>Vassilis</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A10" contrib-type="author"><name><surname>Peponi</surname><given-names>Christina</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A11" contrib-type="author"><on-behalf-of>Panhellenic Association for Continual Medical Research (PACMeR)</on-behalf-of><email>[email protected]</email></contrib>	BMC Gastroenterology	<sec><title>Background</title><p>Data comparison between EUROCARE and SEER database provided evidence that colorectal cancer survival in United States of America is higher than in European countries [<xref ref-type="bibr" rid="B1">1</xref>]. Survival differences were maintained irrespectively of which European Nation was compared, and were much higher when Eastern European countries were considered [<xref ref-type="bibr" rid="B1">1</xref>]. Correction for stage at diagnosis consistently reduced survival differences and the reduction was substantially unrelated to the European geographic area analyzed [<xref ref-type="bibr" rid="B2">2</xref>]. The presence of a diagnostic colorectal cancer screening bias was therefore hypothesized since early diagnostic procedures might be much less available in Europe than in USA [<xref ref-type="bibr" rid="B2">2</xref>]. Deficiencies in European colorectal cancer screening guideline implementation, and inadequacy of screening test advising in primary care setting were highlighted in a recent systematic review of literature, but only data from Italian and French physicians were available for the USA vs European data comparison [<xref ref-type="bibr" rid="B3">3</xref>].</p><p>Colorectal cancer screening survival benefit had been strongly documented by randomized-controlled trials [<xref ref-type="bibr" rid="B4">4</xref>-<xref ref-type="bibr" rid="B8">8</xref>] and a meta-analysis [<xref ref-type="bibr" rid="B9">9</xref>]. In this setting, time to diagnosis and, more precisely, stage at surgery, play the major role for patient outcome. In addition to detecting early stage cancers, screening can detect pre-cancerous polyps as well, which can be removed, thereby preventing cancer. Colorectal cancer screening is therefore strongly recommended. Currently guidelines generally include: stool occult blood test (SOBT) yearly and sigmoidoscopy every 3–5 years, both beginning at the age 50 [<xref ref-type="bibr" rid="B10">10</xref>-<xref ref-type="bibr" rid="B13">13</xref>]. There are two major ways to implement these recommendations: organized screening programs principally endorsed with active invitations by the national health services; and opportunistic screening recommendation by general practitioners. Despite organised screening programmes are based on a more coherent structure, offering a standardised system of care and are at a recognized better level of evidence, National CCS (colorectal cancer screening) programs are actually at the beginning in European countries, and the ongoing pivotal experiences such as the Italian (region-wide programs), UK and Finnish (started in 2000, 2001 and 2004 respectively) are too recent to impact on survival [<xref ref-type="bibr" rid="B14">14</xref>-<xref ref-type="bibr" rid="B16">16</xref>]. Consequently, up to date opportunistic screening implementation in European primary care setting was mandatory. Indeed, physicians involved in primary care both implement programs with active invitation [<xref ref-type="bibr" rid="B17">17</xref>] and recommend screening tests where invitation programs are lacking. This is of particular importance in Greece where organized colorectal cancer screening programs are totally absent.</p><p>Since primary care physicians have a key role in screening practice, and considering that little is known about their screening recommendations in Europe [<xref ref-type="bibr" rid="B3">3</xref>], we surveyed a random sample of Hellenic physicians employed in primary care activities. Both physicians' colorectal cancer screening behaviors and colorectal cancer screening recommendations during usual check-up visits were analyzed.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>The Hellenic trial</title><p>This study is part of a large ongoing program of research on cancer screening practice in Greece, which is organized by the PACMeR (Panhellenic Association for Continual Medical Research). The study's aim is to indicate the current rate of cancer screening among the Hellenic adult population and to identify possible barriers to early diagnosis. In PACMeR_03 trial a medical questionnaire for face-to-face interview was employed to investigate primary care physicians' screening prescriptions habits. The project was approved by PACMReR's Scientific Committee and conformed to the ethical guidelines of the 1975 Declaration of Helsinki.</p><sec><title>Enrollment</title><p>From August 2001 to December 2002, 600 medical doctors employed in primary care activities were recruited from medical lists of 14 health centers, 16 general hospital and 31 rural ambulatory departments in 9 Hellenic provinces (Achaia, Attikis, Chania, Cephalonia, Drama, Etolocarnania, Evro, Lesbo, Serres) and were invited to answer the prescription-habits questionnaire in a self-reported and nominal form during a face to face interview with a PACMeR physician. Personal data privacy was warranted for all physicians who refused to enter the trial.</p><p>Two hundred eleven (211) physicians agreed to participate: 105 male, 94 female, 12 did not declare gender (privacy was guaranteed in spite of the face-to-face nature of the survey). The mean age of physicians involved in the study was 34 (24 – 62 years old); 31 were specialists in General Medical Practice (GMP), 22 trainees in GMP, 25 Internists, 53 trainees in Internal Medicine (employed in internal medicine ambulatory department with activities overlapping primary care), 78 unspecialized physicians (employed in the compulsory rural primary care medical service), 1 who did not answer this item.</p></sec><sec><title>Medical questionnaire</title><p>Physicians were invited to answer two main questions: one for usual screening practice and one for specifically indicated cancer screening practice.</p><p>1<sup>st </sup>Question: "When a patient older than 50 years old comes to your office and asks for a general check up what do you recommend?"</p><p>2<sup>nd </sup>Question: "Which of the following examinations do you think should be prescribed to people older than 50 for cancer screening practice?"</p><p>The medical questionnaire contained a board of multiple choices for 58 possible (related or unrelated) screening procedures (i.e. digital rectal examination, chest radiography, Papanicolaou smear, urinalysis, urinary culture, prostate specific antigen).</p><p>For the 2<sup>nd </sup>Question the prescription frequency (monthly, twice a year, yearly, every 2 years, 3 years, 5 years and no prescription) should have been specified. Data obtained were further analyzed for sigmoidoscopy, stool occult blood test and digital rectal examination.</p><p>Due to the phrasing of the two questions, and considering that DRE may be implemented for both colorectal and prostate cancer screening, the determination of the implemented proportion of the test in each setting was not possible, and related analysis should be therefore considered a secondary outcome.</p></sec><sec><title>Subgroups analysis and statistics</title><p>Since it might be guessed that trainees and recently trained physicians would be more likely to be aware of evidence-based and cost-effectiveness studies, screening habits were further analyzed for physicians' subgroups by age, sex, level and type of specialization. Since the digital rectal examination is not a cost effective screening procedure, we considered its prescription of not value [<xref ref-type="bibr" rid="B18">18</xref>]. Colorectal cancer screening habits were therefore divided in two categories: no prescription OR prescription of at least one documented cost-effective test (sigmoidoscopy or stool occult blood test) at any frequency. Statistical analysis was performed using Pearson's Chi-square test and Fisher's exact test.</p></sec></sec></sec><sec><title>Results</title><p>We found that the 50% (106/211) of physicians recommended colorectal cancer screening during usual check-up visits. After exclusion of non-cost-effective screening procedure (digital rectal examination) from the analysis, the percentage of physicians recommending colorectal screening (stool occult blood test and/or sigmoidoscopy) shrank to 25% (52/211). Only the 3% (7/211) of physicians was found to recommend both sigmoidoscopy and fecal occult blood test; while the 1% (2/211) advised sigmoidoscopy alone and the 20% (43/211) fecal occult blood test alone.</p><p>When physicians were specifically asked about which examinations they think should be prescribed to people older than 50 for cancer screening practice, only the 77% recognized a value for colorectal cancer screening tests (any prescription considered: stool occult blood test and/or sigmoidoscopy and/or digital rectal examination). This percentage dropped to 53% (112/211) when only cost-effective procedures were considered (stool occult blood test and/or sigmoidoscopy). In this setting, we found that 18% (39/211) of physicians think that both sigmoidoscopy and stool occult blood test should be prescribed for cancer screening activities, while 4% (9/211) recommended sigmoidoscopy alone, and 30% (64/211) for stool occult blood test alone. Frequencies by which physicians prescribe the abovementioned tests for screening purposes are reported in Table <xref ref-type="table" rid="T1">1</xref>.</p><sec><title>Subgroups analysis</title><p>The colorectal cancer screening recommendations rate during usual check-up visits were markedly higher among specialists in general medical practice (45.2%) even though the difference was not statistically significant (exact p = 0.054). It was quite homogeneous among the other specialization subgroups (trainees in general medical practice 18.2%, internists 20%, trainees in internal medicine 22,6% and doctors employed in the post-law compulsory rural primary care medical service 20,5%). Colorectal cancer screening recommendation rate during usual check-up visits was statistically lower among physicians aged 30 or less versus older ones (p = 0.012).</p><p>When cancer-screening behaviors were considered, we found no belief differences among physicians' subgroups for any of the analyzed parameters: specialization (exact p = 0.749), age (p = 0.057), gender (p = 0.191).</p></sec></sec><sec><title>Discussion</title><p>Little is known in peer-reviewed literature about colorectal cancer screening recommendations in European primary care. Up to date only 4 surveys had been reported (three from France and one from Italy) [<xref ref-type="bibr" rid="B19">19</xref>-<xref ref-type="bibr" rid="B23">23</xref>] and relative results had been recently analyzed in a systematic review of literature [<xref ref-type="bibr" rid="B3">3</xref>]. Present study is therefore the first report from Greece and from a third European nation.</p><p>In spite of the documented survival benefit from colorectal cancer screening [<xref ref-type="bibr" rid="B4">4</xref>-<xref ref-type="bibr" rid="B9">9</xref>], regardless of its recommendation by health authorities [<xref ref-type="bibr" rid="B10">10</xref>-<xref ref-type="bibr" rid="B13">13</xref>], and in sharp contrast with the high rate of CCS recommendation by US physicians [<xref ref-type="bibr" rid="B24">24</xref>], the rate of CCS implementation was not satisfactory in any of the European studies but one [<xref ref-type="bibr" rid="B21">21</xref>] and particularly in the present survey (table <xref ref-type="table" rid="T2">2</xref>). Indeed in Ganry study (2004)[<xref ref-type="bibr" rid="B21">21</xref>] a significant higher rate of CCS advising (95%) was observed in comparison with previous French reports (1996, 2003)[<xref ref-type="bibr" rid="B19">19</xref>,<xref ref-type="bibr" rid="B20">20</xref>]; this may be explained by the fact that time from guideline implementation (1998 for France)[<xref ref-type="bibr" rid="B25">25</xref>], related medical education and putting guidelines in to practice are sequentially time dependent processes. In our trial 46% of physicians screened their healthy patients with digital rectal examination during regular general practice activities. This percentage was notably higher (68%) when cancer-screening purpose was considered. Surprisingly, recommendation of this not evidence-based test had been previously reported among French physicians [<xref ref-type="bibr" rid="B19">19</xref>,<xref ref-type="bibr" rid="B21">21</xref>]. Indeed, despite the "myth" of digital-rectal examination is alive among physicians, its implementation as screening procedure is not cost effective [<xref ref-type="bibr" rid="B18">18</xref>] and therefore not recommended by major authorities [<xref ref-type="bibr" rid="B10">10</xref>-<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B26">26</xref>,<xref ref-type="bibr" rid="B27">27</xref>].</p><p>We further observed that a large proportion of physicians did not consider cancer-screening practice as a part of the periodic health examination (only 50% of physicians advise it during usual check-up visits, while the 77% believe it should be prescribed in case of cancer screening activities). The more discomforting finding in our trial was that only the 25% of the surveyed physicians recommended evidence-based screening tests during usual check-up visits.</p><p>These results might be in part explained by the actual composition of the Hellenic health system. Indeed, to date primary care services were mainly guaranteed by non-specialized physicians employed in the post law compulsory medical service (for free service supported by the Health system) and by private family physicians (for fee services); and secondarily by internal medicine ambulatory division of general hospital and by professional funds (for free services). Thus, non-specialized young physicians are the key of the actual Hellenic primary care system. Therefore, the low rate of screening recommendation observed should not be surprising and explains partially why in our study colorectal screening recommendations were statistically lower among physicians aged 30 or less. In fact we found that specialists in general practice showed a trend for better prescription (p = 0.054).</p><p>Fortunately, the newborn sanitary system is progressively substituting the unspecialized post law physicians with specialized GPs. The whole primary care composition, as well as services provided, may consequently radically change in next 5–10 years. We therefore hope that a specialized system will provide both the hopeful changes and a cost-effective screening coverage.</p><p>Anyway, our study presents many limitations and may not be applicable to the whole population of primary care physicians. Firstly we have data of only 211 physicians. Secondly, practices of responders may be systematically different from those of non-responders; a high rate of non-responders is anyhow common in this kind of cross-sectional survey [<xref ref-type="bibr" rid="B28">28</xref>-<xref ref-type="bibr" rid="B30">30</xref>]. Finally, lack in the availability of dedicated local endoscopic services may discourage physicians in recommending screening test; something that was not assessed in our study.</p><p>The Hellenic need of a new redefinition for the primary care setting should not be considered a local phenomenon. Poorly defined role of primary care physicians in screening services delivery and the need of a new re-definition of primary care activities have been evidenced even in a recent survey of English practitioners [<xref ref-type="bibr" rid="B31">31</xref>]; and screening data coming from French and Italian surveys were quite discouraging too [<xref ref-type="bibr" rid="B19">19</xref>-<xref ref-type="bibr" rid="B23">23</xref>]. This may be in part attributed to the history of guidelines implementation in both single Nation and European setting. Indeed, the date of guideline release is of a great importance since their dissemination, their cultural re-elaboration and the process of putting them into practice are time-dependent processes [<xref ref-type="bibr" rid="B3">3</xref>]. Actually in Greece there are not national guidelines for colorectal cancer screening and European guidelines were first time implemented only in 2003 [<xref ref-type="bibr" rid="B10">10</xref>].</p></sec><sec><title>Conclusion</title><p>Since early diagnosis of colorectal cancer and stage of disease at surgery are crucial for patients' survival, we conclude that a huge work has to be done in Hellenic and, more widely, in European primary care setting if screening is to make a major impact on colorectal cancer mortality. Need for national guideline implementation and continual medical education are imperatively highlighted.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>AX: Main coordinator, he was actively involved in the discussion of the project & study planning, and manuscript writing. DM: He was actively involved in the discussion of the project, realization of the draft & study planning, review of data abstraction and manuscript writing. KK: was actively involved in the discussion of the project & study planning, and in the data collection during the survey. AG: She was involved in study planning and she was responsible for data collection in Crete Island. She was still actively involved in the discussion of the project and manuscript. GZ: He was involved in the study planning and was responsible for data collection in the wide area of Attika. He was still actively involved in the discussion of the project and manuscript. V S: She was involved in the study planning and was responsible for data collection in northern Greece. She was still actively involved in the discussion of the project and manuscript. PP: He was involved in the study planning and was responsible for data collection. He was still actively involved in the discussion of the project and manuscript. GC: He was involved in the study planning, in data collection and in the discussion of the results. VG: He was actively involved in the discussion of the project, realization of the draft, discussion of the outcomes, reviewing and formatting the manuscript. CP: Main data-manager. She was involved in the study planning and was responsible for data collection in the northwestern part of Greece. She was responsible of data entering in the database. She was involved in manuscript discussion.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1471-230X/6/18/prepub"/></p></sec>
Variation in recruitment across sites in a consent-based clinical data registry: lessons from the Canadian Stroke Network	<sec><title>Background</title><p>In earlier work, we found important selection biases when we tried to obtain consent for participation in a national stroke registry. Recognizing that not all registries will be exempt from requiring consent for participation, we examine here in greater depth the reasons for the poor accrual of patients from a systems perspective with a view to obtaining as representative sample as possible.</p></sec><sec sec-type="methods"><title>Methods</title><p>We determined the percent of eligible patients who were approached to participate and, among those approached, the percent who actually consented to participate. In addition we examined the reasons why people were not approached or did not consent and the variation across sites in the percent of patients approached and consented. We also considered site variation in restrictions on the accrual and data collection process imposed by either the local research ethics board or the hospital.</p></sec><sec><title>Results</title><p>Seventy percent of stroke patients were approached, with wide variations in approach rates across sites (from: 41% to 86%), and considerable inter-site variation in hospital policies governing patient accrual. Chief reasons for not approaching were discharge or death before being approached for consent. Seventeen percent of those approached refused to participate (range: 5% to 75%). Finally, 11% of those approached did not participate due to language or communication difficulties.</p></sec><sec><title>Conclusion</title><p>We found wide variation in approach and agree rates across sites that were accounted for, in part, by different approaches to accrual and idiosyncratic policies of the hospitals. This wide variation in approach and agree rates raises important challenges for research ethics boards and data protection authorities in determining when to waive consent requirements, when to press for increased quality control, when to permit local adaptation of the consent process, and when to permit alternatives to individual express consent. We offer several suggestions for those registries that require consent for participation.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Willison</surname><given-names>Donald J</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Kapral</surname><given-names>Moira K</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Peladeau</surname><given-names>Pierrot</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Richards</surname><given-names>Janice A</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Fang</surname><given-names>Jiming</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Silver</surname><given-names>Frank L</given-names></name><xref ref-type="aff" rid="I5">5</xref><email>[email protected]</email></contrib>	BMC Medical Ethics	<sec><title>Background</title><p>The need for individual consent for the secondary analysis of existing data or for the use of data in clinical registries for a broad, long-range research agenda is highly contentious. Some researchers have called for a waiver of consent requirements for minimal risk research, arguing that obtaining individual consent would be impracticable and allowing individuals to opt-out would introduce bias into analyses [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B5">5</xref>]. Others, however, warn of the blurring of the distinction between research and clinical care, continual expansion of secondary uses of data for non-clinical purposes, and pressures to weaken human subject protections [<xref ref-type="bibr" rid="B6">6</xref>]. The responsibility of weighing the competing demands of scientific rigour and the protection of human subjects' rights falls squarely upon research ethics boards (REBs). In examining how to minimize analytic bias, waiver of consent is not the only option. In particular, REBs may first wish to consider what efforts have been taken to ensure quality control in the recruitment process.</p><p>Earlier, we reported selection biases associated with attempts to obtain consent for participation in a consent-based acute stroke registry, the Registry of the Canadian Stroke Network [<xref ref-type="bibr" rid="B5">5</xref>]. However, we also noted significant variation in approach and consent rates across sites, suggesting possible recruitment process issues that deserve attention. In this paper, we examine more closely the variation in recruitment across sites, and attempt to understand the reasons for variations in patient approach and agree rates. We do this to assist others who will be developing consent-based registries, in obtaining a sample that is as representative as possible of the larger population.</p></sec><sec sec-type="methods"><title>Methods</title><p>The Canadian Stroke Network (CSN) is a collaborative effort of academic researchers, government, industry, and the non-profit sector, dedicated to decreasing the physical, social and economic consequences of stroke on the individual and on society. The Registry of the Canadian Stroke Network (RCSN) is a clinical database of patients with acute stroke patients seen at selected acute care hospitals across Canada. In this paper, we focus on "Phase 2" of the RCSN which took place between June 2002 and December 2002. Patients were recruited into the registry by experienced research nurses. Data collected included information about patient demographics and clinical symptoms, their hospital encounter, and quality of life and functional status (through a follow-up telephone interview).</p><p>We determined the percent of potentially eligible patients who were approached to participate in the Registry. We also examined what percent of those who were approached to participate actually consented to participate. Nurse-coordinators maintained a log documenting whether non-participation was due to patient refusal, inability to consent due to language or another communication barrier, or inability to approach the patient due to early discharge or other factors. In addition, we examined the variation across sites in the percent of patients who were approached to participate in the Registry and, of those, the percent who agreed to participate. Finally, we summarized the barriers encountered at individual sites through a survey of site coordinators.</p></sec><sec><title>Results</title><sec><title>Reasons patients were not accrued into the Registry (Figure <xref ref-type="fig" rid="F1">1</xref>)</title><fig position="float" id="F1"><label>Figure 1</label><caption><p>Patient accrual process – phase 2.</p></caption><graphic xlink:href="1472-6939-7-6-1"/></fig><p>Overall, 70% of potential participants were approached and 72% of these were enrolled in the Registry, with an overall accrual rate of 50.5%. Logistical challenges in approaching patients accounted for 60% of non-accrual. Major reasons included: death before the patient could be approached (10%); discharge from hospital before being approached (10%); and inability to make contact with the patient or surrogate after more than 3 attempts (20%).</p><p>Of the patients approached, approximately 17% refused to participate. This is approximately 2.5 times greater than that encountered in pilot work (unpublished).</p><p>Eight percent of patients were unable to consent due to communication difficulties with no surrogate available. Another 3% of those approached were not administered the consent form because their mother tongue (and that of their surrogate) was not English or French.</p></sec><sec><title>Variation in approach, agree, and overall participation rates (Figure <xref ref-type="fig" rid="F2">2</xref>)</title><fig position="float" id="F2"><label>Figure 2</label><caption><p>Approach and agree rates by site: sorted by agree rate.</p></caption><graphic xlink:href="1472-6939-7-6-2"/></fig><p>The percent of eligible patients approached varied across sites from 41% to 86%, with a mean of 70%. The agree rate averaged 74% of those approached (lowest 25%, highest 95%). Approach and consent rates were not correlated across sites.</p></sec><sec><title>Supports and constraints on data collection (Table <xref ref-type="table" rid="T1">1</xref>)</title><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Supports and barriers to patient accrual and mean approach rates</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td></td><td align="center" colspan="2"><bold>Mean approach rate (n)</bold></td><td align="center"><bold>t-test p value</bold></td></tr><tr><td></td><td></td><td colspan="3"><hr></hr></td></tr><tr><td></td><td></td><td align="center"><bold>Sites answering "Yes"</bold></td><td align="center"><bold>Sites answering "No"</bold></td><td></td></tr></thead><tbody><tr><td align="left">Supports Available for the Registry</td><td align="left">1. Do you work with the "Stroke Team" (attend ward rounds, discharge planning meetings, receive referrals or patient lists from the team, etc.)</td><td align="center">70.4 (11)</td><td align="center">69.4 (9)</td><td align="center">0.8651</td></tr><tr><td></td><td align="left">2. Do your physicians (attending/investigator/residents/fellows) help you obtain patient consents?</td><td align="center">73.0 (9)</td><td align="center">71.6 (9)</td><td align="center">0.7503</td></tr><tr><td></td><td align="left">3. Is there a neurology/stroke prevention clinic where TIA/stroke patients are seen following emergency department visits at your hospital?</td><td align="center">68.9 (15)</td><td align="center">73.0 (5)</td><td align="center">0.5084</td></tr><tr><td></td><td align="left">4. Do you have support from your emergency department (notification of new patients, providing brochures for patients not admitted to hospital, etc.)</td><td align="center">80.5 (5)</td><td align="center">66.9 (14)</td><td align="center">0.0183</td></tr><tr><td align="left">Barriers to Recruitment</td><td align="left">1. Will your institution allow you to collect the minimal dataset on all patients?</td><td align="center">70.1 (18)</td><td align="center">69.0 (2)</td><td align="center">0.901</td></tr><tr><td></td><td align="left">2. Will your institution allow you to obtain lists of potential stroke patients from ED and the wards?</td><td align="center">70.4 (19)</td><td align="center">60.6 (1)</td><td align="center">0.417</td></tr><tr><td></td><td align="left">3. Will your institution allow you to directly approach admitted patients to consent to participate in the Registry?</td><td align="center">71.1 (18)</td><td align="center">59.9 (2)</td><td align="center">0.1962</td></tr><tr><td></td><td align="left">4. If a "potential" registry participant has been discharged (or never admitted) will your institution allow you to contact the patient to participate in the registry?</td><td align="center">72.0 (13)</td><td align="center">65.7 (6)</td><td align="center">0.2877</td></tr><tr><td></td><td align="left">5. Will your institution allow you to obtain telephone consent?</td><td align="center">69.2 (6)</td><td align="center">69.7 (12)</td><td align="center">0.9446</td></tr></tbody></table></table-wrap><p>There was considerable variation across sites in supports for and constraints on approaching patients. Only five sites (25%) had active support from the emergency department (e.g. notification of new patients, providing brochures for patients not admitted to hospital). These sites achieved a substantially higher approach rate than did the sites without such support (average 80.5% vs. 67%; p < 0.02).</p><p>We found approach rates to be lower:</p><p>• When lists of potential stroke patients could not be obtained from the emergency department or the wards (60.6% vs. 70.4%);</p><p>• When coordinators could not approach patients directly. In these cases, the physician responsible for care had to first approach the patient. (59.9% vs. 71.1%); and</p><p>• When coordinators could not make contact with the patients after they had left the hospital (65.7% vs. 72.0%).</p><p>None of these results was statistically significant, although this may have been due to inadequate statistical power. Approach rates were no different in sites where the local principal investigator actively participated in the recruitment and where the nurse recruiter worked closely with the stroke team.</p></sec></sec><sec><title>Discussion</title><p>We found wide variation across sites in both the rate at which potentially eligible patients were approached to participate in the Registry and in consent rate. Based on our discussions with study coordinators, we learned that some of the difference in approach rates was due to variations in the interpretation of provincial data protection laws, and by site-specific limitations imposed by hospitals on the conditions under which patients could be approached. In some cases, the restrictions applied by hospital administration went over and above those applied by the research ethics boards or by provincial laws. In addition, sites receiving support from their emergency department (e.g. notification of new patients, providing brochures for patients not admitted to hospital) had substantially higher approach rates.</p><p>We observed an overall refusal rate of 17% with wide variation across sites (5% to 75%). This indicates unevenness in the approach to recruitment across sites. Ideally, it would be helpful to ask those who refused why they refused. This information is not available from the Registry. However, in future research projects, where consent is required, it would be instructive to learn why people refuse to participate, so as to be responsive to concerns raised.</p><sec><title>Regulatory and governance context</title><p>Canadian federal and provincial data protection laws allow for exemptions from consent for research purposes where, among other conditions: (a) the research cannot be achieved without using personal information, and (b) it is impracticable to obtain consent [<xref ref-type="bibr" rid="B7">7</xref>]. However, none of the provinces' legislation provides clear guidance as to the circumstances under which obtaining consent may be deemed impracticable. In Alberta, Saskatchewan, and Ontario, legislation specifically identifies this to be the purview of the research ethics board (REBs) [<xref ref-type="bibr" rid="B8">8</xref>].</p><p>Article 3.4 of the Tri-Council Policy Statement (TCPS) – the document that articulates the standards in Canada for REBs of institutions receiving funding from any of the three major federal granting councils – states that the "REB <bold><italic><underline>may</underline></italic></bold>[<italic>our emphasis</italic>] also require that a researcher's access to secondary use of data involving identifying information be dependent on: (a) the informed consent of those who contributed data or of authorized third parties;..." [<xref ref-type="bibr" rid="B9">9</xref>]. No specific guidance is provided as to the criteria for determining whether or not consent should be required for secondary use of existing personal information for research. However, two of the fundamental guiding ethical principles articulated in the TCPS are respect for free and informed consent and respect for privacy and confidentiality. Therefore, to be consistent with the values, purpose, and protections advanced in the TCPS, the onus for demonstrating a reasonable exception to the requirement for consent should fall on the researcher.</p><p>In 2005, the Canadian Institutes of Health Research (CIHR) issued its <underline>Best Practices for Protecting Privacy in Health Research</underline>[<xref ref-type="bibr" rid="B8">8</xref>]. Element 3 of the document includes detailed guidance as to the factors to consider when determining whether or not a research project should receive exemption from consent for secondary use of personal information. One of the provisions is quite broad – if, due to the size of the population, the proportion likely to have relocated or died, or lack of continuing relationship with the data holder:</p><p>"...there is a risk of introducing bias into the research because of the loss of data from segments of the population that cannot be contacted to seek their consent, thereby affecting the validity of results and/or defeating the purpose of the study."</p><p>We recognize the risk that this provision runs the risk of becoming a "trump card" over privacy concerns. Accordingly, it places a heavy onus on REBs to determine when to allow use of the data without consent and when to press for increased quality control in recruitment. This is relatively simple when it has been demonstrated that the vast majority of the potential research participants would be willing to allow their information to be used. It is much more difficult when, as we found, a non-trivial proportion of people approached refuses to participate.</p><p>Section 39 of the 2004 Ontario Personal Health Information Protection Act permits the disclosure of personal health information without consent to "prescribed registries" for the purpose of statistical analysis [<xref ref-type="bibr" rid="B12">12</xref>]. A handful of registries, including the Registry of the Canadian Stroke Network, are among the prescribed registries [<xref ref-type="bibr" rid="B10">10</xref>].</p></sec><sec><title>Experience elsewhere</title><p>The limited published literature on recruitment suggests that challenges in variation in recruitment faced by the Registry of the CSN are not unique. While researchers associated with the Mayo Clinic in Minnesota were able to achieve consent rates in excess of 95% to participate in a broad cross-section of disease-registries, there was variation both across sites and by diagnosis [<xref ref-type="bibr" rid="B11">11</xref>,<xref ref-type="bibr" rid="B12">12</xref>].</p><p>It appears, though, that consent-seeking alone is not exclusively responsible for incomplete accrual. Across 91 U.K. clinical databases listed under the Directory of Clinical Databases (DoCDat), completeness of patient recruitment appears to be similar for databases that do and do not require individual consent for enrolment [<xref ref-type="bibr" rid="B13">13</xref>].</p><p>In a different context, Gross and colleagues examined patient accrual for 172 clinical trials in four high-impact medical journals. They found very poor reporting of the patient accrual process, with only 31 studies (18%) screening from a consecutive series of patients [<xref ref-type="bibr" rid="B14">14</xref>].</p></sec><sec><title>Lessons learned</title><p>Many researchers will still need to obtain informed consent for patient participation in their registry projects – for example, where there will be direct patient contact, where genetic information will be included or linked, or in particularly stigmatizing medical conditions. Several lessons can be learned from our experience with developing a consent-based registry. These lessons are derived from the data presented in this paper and from our discussions with site coordinators and co-investigators:</p><p>(1) The consent process needs to be thoroughly pilot tested under 'real-world conditions' with gradual roll-out to participating sites. One should anticipate ample lead time to develop, test, and implement the entire concept – particularly the consenting process and staff responsibilities.</p><p>(2) Close communications need to be established early and maintained with research ethics boards and health care institutions. This is probably best accomplished through a single contact-person working with each REB and hospital from the outset of the project.</p><p>(3) Accountability requirements for those responsible for obtaining consent should be as consistent as possible. Nurse coordinators in this study had a dual accountability: to the central coordinator and to local site principal investigators.</p><p>(4) Consider staging the implementation process, so as to build on the successes of the less complicated recruitment scenarios. For example, from the outset, we tried to recruit patients with transient ischemic attacks. This was ambitious, as these particular patients usually were not admitted to the hospital, and they constituted a large proportion of our patients not approached.</p><p>(5) Use a multi-pronged strategy for recruitment when potential registry participants have multiple points of access or care trajectories (e.g. both inpatient and outpatient treatment). Obtaining consent may be more feasible when repeated outpatient visits allow increased contact and trust.</p><p>(6) Obtain firm support of those departments that have first contact with target patients (e.g. Emergency) to identify potential participants and provide them with information and support to implement screening processes.</p><p>(7) Consider random sampling strategies to reduce workload, rather than including all consecutive patients. This was a strategy we implemented in Phase 2 in institutions with particularly high volumes of stroke patients. We found this increased the approach rate.</p><p>(8) Ongoing monitoring and feedback on accrual help to increase and sustain higher accrual rates and interest.</p><p>(9) Consent forms in other languages and access to translators may be required for projects operating in jurisdictions with multi-ethnic populations. Usually, such hospitals have a roster of translators for such situations.</p><p>(10) Elicit ongoing patient feedback – particularly from those who hesitate or refuse to participate – to ascertain what concerns they may have. While some refusers may not wish to share this information, if this is done in a way that does not pressure patients, then it can provide valuable feedback.</p></sec><sec><title>Long-run changes are needed</title><p>Concern has been expressed elsewhere that, for multi-centered studies, the process of research ethics approval is very time consuming, with considerable duplication of effort and local idiosyncratic restrictions that offer little perceived gain [<xref ref-type="bibr" rid="B15">15</xref>-<xref ref-type="bibr" rid="B17">17</xref>]. In some countries, a centralized review process has been implemented for such multi-centered studies. While intended to streamline the review process, in some cases this has simply added another level of bureaucracy [<xref ref-type="bibr" rid="B18">18</xref>]. Even greater standardization of the process would be helpful. Assimilation of the CIHR privacy guidelines into the review process could help to harmonize the interpretation of acceptable recruitment practice.</p><p>In our study, the chief source of variation in administrative requirements came not from the REBs but from the data stewards – acute care hospitals. In particular, we found major differences in (a) ability to coordinate with the emergency department in the recruitment process; and (b) hospital policy as to whether, and at what point, potential registry participants could be approached to participate. In part, this can be resolved through education of health care institutions as to what is permitted by the law.</p><p>Looking forward, numerous registry and linked electronic health record data collection activities are being planned in North America and Europe. Perhaps it is time to re-think how we go about recruiting patients into these registries [<xref ref-type="bibr" rid="B19">19</xref>]. For example, would it be more efficient to shift responsibility for patient accrual from managers of individual projects to the institution and for a network of institutions to develop common protocols across institutions for obtaining consent for use of clinical records, obtaining biological samples, follow-up surveys, and linkage of clinical data with administrative records?</p></sec></sec><sec><title>Conclusion</title><p>We have described numerous challenges in developing and implementing a consent-based registry for stroke patients. We believe that ours is not a unique experience. Our attempts have led to important sampling biases that limit the generalizability of our data. We have also demonstrated important quality control issues in conducting a multi-centered registry. The teasing out of these issues represents a major challenge to research ethics boards and data protection authorities who are charged with the responsibility of determining when to allow collection of information without consent, when to press for increased quality control in recruitment, and when to permit local adaptation of the recruitment process. We hope that the experience of the Registry of the Canadian Stroke Network will contribute to future-oriented solutions.</p></sec><sec><title>Competing interests</title><p>All the authors except Peladeau have been sponsored through the Canadian Stroke Network to attend one or more annual meetings of the Canadian Stroke Network or related conferences. The authors have no other financial or non-financial competing interests.</p></sec><sec><title>Authors' contributions</title><p>DW conceived the study, drafted the manuscript, and produced subsequent revisions. Remaining authors reviewed and revised the manuscript. JF performed the statistical analyses. JR coordinated the data collection. All authors read and contributed to successive drafts of the manuscript.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1472-6939/7/6/prepub"/></p></sec>
Prenatal cocaine exposure alters alpha2 receptor expression in adolescent rats	<sec><title>Background</title><p>Prenatal cocaine exposure produces attentional deficits which to persist through early childhood. Given the role of norepinephrine (NE) in attentional processes, we examined the forebrain NE systems from prenatal cocaine exposed rats. Cocaine was administered during pregnancy via the clinically relevant intravenous route of administration. Specifically, we measured α<sub>2</sub>-adrenergic receptor (α<sub>2</sub>-AR) density in adolescent (35-days-old) rats, using [<sup>3</sup>H]RX821002 (5 nM).</p></sec><sec><title>Results</title><p>Sex-specific alterations of α<sub>2</sub>-AR were found in the hippocampus and amygdala of the cocaine-exposed animals, as well as an upregulation of α<sub>2</sub>-AR in parietal cortex.</p></sec><sec><title>Conclusion</title><p>These data suggest that prenatal cocaine exposure results in a persistent alteration in forebrain NE systems as indicated by alterations in receptor density. These neurochemical changes may underlie behavioral abnormalities observed in offspring attentional processes following prenatal exposure to cocaine.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Booze</surname><given-names>Rosemarie M</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Wallace</surname><given-names>David R</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Silvers</surname><given-names>Janelle M</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Strupp</surname><given-names>Barbara J</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Snow</surname><given-names>Diane M</given-names></name><xref ref-type="aff" rid="I5">5</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Mactutus</surname><given-names>Charles F</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Neuroscience	<sec><title>Background</title><p>Recently, noradrenergic systems have been identified as a potential teratogenic target underlying the functional effects of prenatal cocaine [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B3">3</xref>]. However, information regarding the consequences of prenatal cocaine on the development of noradrenergic receptors is relatively sparse. NE is present early in brain development and regulates important aspects of prenatal brain development, including neural migration and synaptogenesis [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B5">5</xref>]. Thus, the ability of cocaine to inhibit NE reuptake has potentially profound effects on the developing nervous system and function of NE systems.</p><p>Previous investigations into the effects of prenatal cocaine exposure on catecholaminergic receptors have, for the most part, focused on the long-term effects of exposure on dopaminergic [<xref ref-type="bibr" rid="B6">6</xref>-<xref ref-type="bibr" rid="B10">10</xref>] and serotonergic [<xref ref-type="bibr" rid="B11">11</xref>-<xref ref-type="bibr" rid="B14">14</xref>] receptor systems. Fewer studies have examined the noradrenergic receptor family following prenatal cocaine exposure. The neurophysiological effects of NE are mediated by three types of receptors: α<sub>1</sub>, α<sub>2 </sub>and β. The α<sub>2 </sub>adrenergic receptors are present very early in development, in some brain areas as early as E15 [<xref ref-type="bibr" rid="B4">4</xref>]. Prenatal exposure to cocaine has been found to elevate the density of α<sub>2 </sub>adrenergic receptors in the cerebellum and forebrain [<xref ref-type="bibr" rid="B15">15</xref>]. Henderson et al [<xref ref-type="bibr" rid="B16">16</xref>] reported that cortical α<sub>2 </sub>adrenergic receptor density was unchanged in male rat pups following prenatal cocaine exposure. However, these studies did not differentiate between male and female offspring and used homogenate binding techniques. Moreover, cocaine was administered via the subcutaneous route into the dams, and therefore these effects likely occurred in the presence of potential nutritional and stress confounds [<xref ref-type="bibr" rid="B17">17</xref>].</p><p>Previous studies from this laboratory and others have demonstrated that the IV route of cocaine administration to pregnant rats produces functional alterations in attentional processes [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B18">18</xref>-<xref ref-type="bibr" rid="B21">21</xref>]. Subtle, context-specific sex differences in attentional tasks following prenatal cocaine have been reported in a number of these studies [<xref ref-type="bibr" rid="B18">18</xref>-<xref ref-type="bibr" rid="B21">21</xref>]. The neurological basis of such attentional deficits is complex and likely mediated by several neurotransmitter systems. Several studies have assessed the involvement of norepinephrine specifically in attentional processes. The development of the heart rate orienting response in preweaning rats, a task used to measure attention to a novel stimus, is dependent upon norepinephrine, but not dopamine or serotonin [<xref ref-type="bibr" rid="B22">22</xref>,<xref ref-type="bibr" rid="B23">23</xref>]. Alterations in the heart rate orienting response of cocaine-treated offspring suggest early impairments in noradrenergic systems [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B18">18</xref>]. Direct evidence of the effects of cocaine on norepinephrine systems has been provided by Snow et al. [<xref ref-type="bibr" rid="B24">24</xref>], in which cocaine was found to directly inhibit process outgrowth in locus coeruleus (LC) neurons.</p><p>Altered attention has been reported in 6 year old children gestationally exposed to moderate levels of cocaine [<xref ref-type="bibr" rid="B25">25</xref>]. The impairment in automated vigilance task in 6 year olds most likely reflects a deficit in sustained attention and one that also contained an accuracy component (commission vs. omission errors). The NE system is thought to be critically involved in the regulation of attention [<xref ref-type="bibr" rid="B26">26</xref>-<xref ref-type="bibr" rid="B31">31</xref>]. That is, the activation of NE serves to filter out distracting or competing stimuli and plays a role in selective attention in rats [<xref ref-type="bibr" rid="B32">32</xref>]. A recent study reports that rats exposed to prenatal cocaine are more sensitive to impairment of selective attention by idazoxan, an α2 adrenergic receptor agonist [<xref ref-type="bibr" rid="B1">1</xref>]. These findings, and reported deficits in vigilance/orienting performance of young rats prenatally exposed to IV cocaine [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B18">18</xref>] suggest that NE plays an important role in attention and in long-term cocaine impairments [<xref ref-type="bibr" rid="B19">19</xref>-<xref ref-type="bibr" rid="B21">21</xref>]. The mechanisms of cocaine-induced disruption of NE developmental patterns and the relationship between these patterns and the attentional alterations remain to be determined.</p><p>For the most part, the effects of prenatal cocaine exposure have been assessed either immediately, during the preweaning period, or long-term, i.e. into mature adulthood. Recently, the adolescent period has been recognized as a period of vulnerability to the effects of drugs of abuse [<xref ref-type="bibr" rid="B33">33</xref>]. Exposure to drugs during early development may alter critical neural development, producing long-term effects on sexual maturation and sex-specific behaviors which are manifested during the adolescent period [<xref ref-type="bibr" rid="B34">34</xref>,<xref ref-type="bibr" rid="B35">35</xref>]. Thus, the adolescent period may represent a unique developmental time frame in which to examine the neurological effects of prenatal drug exposure.</p><p>In the present studies we examined potential sex-dependent alterations in α<sub>2 </sub>adrenergic receptor density and function in adolescent rats following prenatal cocaine exposure. The idazoxan derivative, RX821002, was used to detect α<sub>2 </sub>adrenergic receptors. RX821002 is a highly selective antagonist that identifies all the α<sub>2 </sub>adrenergic receptor subtypes with similar affinity [<xref ref-type="bibr" rid="B36">36</xref>]. We first used this well-characterized ligand to determine α<sub>2 </sub>receptor density and binding affinity in adolescent rats that received cocaine <italic>in utero</italic>. We then commenced a more detailed receptor autoradiographic study of the hippocampus, parietal cortex, amygdala, pyriform cortex and hypothalamus to determine whether sex differences in α<sub>2 </sub>receptors were present in adolescent rats prenatally exposed to cocaine. These studies were designed as part of a larger effort to understand the neurobiological basis of the previously reported attentional dysfunction in rats prenatally exposed to cocaine [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B19">19</xref>-<xref ref-type="bibr" rid="B21">21</xref>,<xref ref-type="bibr" rid="B37">37</xref>].</p></sec><sec><title>Results</title><sec><title>Weight/growth parameters</title><p>The mean offspring body weights (on P35) were unaffected by prenatal cocaine exposure. Previous reports have shown that this regimen of in utero cocaine treatment had no effects on maternal weight gain, litter size, gestational length, sex ratio, offspring weight on postnatal day 1, or pup survival [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B38">38</xref>].</p></sec><sec><title>α2-adrenergic receptor density: tissue homogenates</title><p>Results of tissue homogenate binding studies are displayed in Table <xref ref-type="table" rid="T1">1</xref> and Figure <xref ref-type="fig" rid="F1">1</xref>. The affinity of RX821002 for α<sub>2</sub>-adrenergic receptors did not differ significantly by sex or prenatal cocaine exposure In all groups, the displacement of [<sup>3</sup>H]RX821002 by unlabeled RX821002 was close to unity (0.94 ± 0.05 – 1.09 ± 0.16) suggesting displacement from a single binding site. Labeling of α<sub>2</sub>-adrenergic receptors by [<sup>3</sup>H]RX821002 was sex-dependent within the saline group with females displaying 36.2% higher density of α<sub>2</sub>-adrenergic receptors than males. The number of α<sub>2</sub>-adrenergic receptors labeled with 2.0 nM [<sup>3</sup>H]RX821002 following prenatal cocaine exposure was significantly increased [F(1, 20) = 5.0; <italic>P </italic>≤ 0.04) in both males and females. This effect was most profound in cocaine-exposed males, as they exhibited a 58.1% increase in α<sub>2</sub>-adrenergic receptor density, relative to saline treated males [F(1, 20) = 6.9; <italic>P </italic>≤ 0.02). Thus, prenatal cocaine produced alterations in α<sub>2</sub>-adrenergic receptor density in the hippocampus without changes in binding affinity.</p></sec><sec><title>α2-adrenergic receptor density: autoradiography</title><p>Representative [<sup>3</sup>H]RX821002 autoradiograms are depicted in Figure <xref ref-type="fig" rid="F2">2</xref>. Tissue sections incubated with 2 nM [<sup>3</sup>H]RX821002 exhibited both sex- and cocaine-induced alterations in α<sub>2</sub>-adrenergic receptor density. Following prenatal exposure to cocaine, the density of α<sub>2</sub>-adrenergic receptors demonstrated a significant interaction between region, sex and cocaine [F(4, 208) = 4.7; <italic>P </italic>≤ 0.007]. The binding of [<sup>3</sup>H]RX821002 to α<sub>2</sub>-adrenergic receptors exhibited significant differences between region and sex within the saline group [F(4, 208) = 10.2; <italic>P </italic>≤ 0.0001]. Prenatal cocaine exposure significantly affected the density of α<sub>2</sub>-adrenergic receptors in male rats across the regions examined [F(4, 208) = 4.0; <italic>P </italic>≤ 0.01].</p><p>In control animals prenatally exposed to saline, the density of hippocampal CA1 α<sub>2</sub>-adrenergic receptors in female rats was significantly [F(1, 52) = 23.2; <italic>P </italic>≤ 0.001] higher than the density of α<sub>2</sub>-adrenergic receptors observed in male controls (Figure <xref ref-type="fig" rid="F3">3</xref>). In the area CA1 of the hippocampus, a significant interaction between drug treatment and sex was observed [F(1, 52) = 20.00; <italic>P </italic>≤ 0.001]. Similar to tissue homogenates, laconosum-moleculare layer of area CA1 hippocampal α<sub>2</sub>-adrenergic receptors were upregulated 21% in male offspring following cocaine exposure [F(1,52) = 6.8; P < 0.01], whereas α<sub>2</sub>-adrenergic receptors did not significantly differ in females.</p><p>Of the other brain regions examined, only the parietal cortex and central nucleus of the amygdala exhibited alterations in α<sub>2</sub>-adrenergic receptor density following prenatal cocaine exposure. In the parietal cortex, a significant effect of drug was observed [F(1, 52) = 4.1; <italic>P </italic>≤ 0.05], with no effect of sex or interaction between sex and prenatal drug treatment (Figure <xref ref-type="fig" rid="F4">4</xref>). In the amygdala, the density of α<sub>2</sub>-adrenergic receptors was significantly [F(1, 52) = 10.9; <italic>P </italic>≤ 0.002] higher in males compared to females from the saline group (Figure <xref ref-type="fig" rid="F5">5</xref>). Prenatal cocaine exposure increased the number of α<sub>2</sub>-adrenergic receptors in the amygdala from female rats compared to saline treated females [F(1, 52) = 7.2; <italic>P </italic>≤ 0.01] such that the density of α<sub>2</sub>-adrenergic receptors in this region was similar to the density observed in male rats (Figure <xref ref-type="fig" rid="F5">5</xref>). There were no effects of either sex or prenatal drug exposure on the density of α<sub>2</sub>-adrenergic receptors in either the pyriform cortex or periventricular nucleus of the hypothalamus.</p></sec></sec><sec><title>Discussion</title><p>The present study found alterations in α<sub>2 </sub>adrenergic receptor density in the adolescent brain subsequent to prenatal cocaine exposure. Adolescence represents a period of vulnerability for substance abuse and age-dependent sensitivity to drugs. Receptor alterations expressed during adolescence may modify essential transitions necessary for producing normal adult brain function [<xref ref-type="bibr" rid="B39">39</xref>]. Prenatal IV cocaine exposure appears to alter normal development of the NE receptor systems, leading to altered adolescent NE brain systems.</p><p>In general, catecholamine receptor density and function appear to be particularly plastic during the adolescent period. For example, dopamine D1 and D2 receptors are overproduced and eliminated (over 40%) in male rats during adolescence [<xref ref-type="bibr" rid="B40">40</xref>]. The distribution of α2 adrenergic receptors has been reported to generally resemble adult patterns by P28 [<xref ref-type="bibr" rid="B41">41</xref>]; however, functional studies have demonstrated decreased depolarized release of NE and a greater capacity for NE reuptake in the hypothalamus (with a potential shift in hypothalamic alpha receptor subtype) is present in rats during the adolescent period [<xref ref-type="bibr" rid="B42">42</xref>]. Sex differences in NE content have also been reported in several brain regions in adolescent animals (P33) [<xref ref-type="bibr" rid="B43">43</xref>]. Prenatal cocaine exposure has been shown to result in increased levels of NE in the preoptic region of male, but not female, adult rats [<xref ref-type="bibr" rid="B44">44</xref>]. Previous reports of α2 adrenergic receptor distribution used pooled male and female rat brains without presenting statistical analysis on sex differences [<xref ref-type="bibr" rid="B41">41</xref>]. Furthermore, the number of animals used from each litter was unclear. Therefore the current studies are the first to report baseline sex differences in α2 adrenergic receptor density in adolescent animals, as well as alterations in density related to prenatal cocaine exposure in this age group.</p><p>Prenatal cocaine exposure may alter α2 adrenergic receptors density by affecting development of the LC. The entire NE input to the rat hippocampus is provided by LC neurons, primarily from the dorsal one-third of the LC [<xref ref-type="bibr" rid="B45">45</xref>,<xref ref-type="bibr" rid="B46">46</xref>]. In early LC development, approximately 7% of the neurons are generated on GD11, 75% on GD12 and 18% on GD13 [<xref ref-type="bibr" rid="B47">47</xref>]. LC neurons and their processes synthesize NE as early as 12–14 days gestation [<xref ref-type="bibr" rid="B48">48</xref>] well before synaptogenesis is underway in the terminal fields. Efferent LC fibers first appear in the neocortical terminal fields on GD16 and these fibers are proposed to play a major role in induction and differentiation of neural tissue [<xref ref-type="bibr" rid="B49">49</xref>]. Hippocampal neurons are generated three days after the LC neurons [<xref ref-type="bibr" rid="B48">48</xref>]. The late gestational period of the rat is distinguished by continued fiber organization and ramification in the LC terminal fields [<xref ref-type="bibr" rid="B50">50</xref>]. A recent study, using the same cocaine dose and route of administration used in the current study, demonstrated that prenatal cocaine exposure during the development of LC neurons inhibits the growth of LC neuritis [<xref ref-type="bibr" rid="B24">24</xref>]. Cocaine binding sites are evident in the fetal brain as early as GD15 and by GD20 the Kd of [<sup>3</sup>H]cocaine binding is similar to the Kd values observed in adulthood [<xref ref-type="bibr" rid="B51">51</xref>]. Thus, our prenatal cocaine exposure covers the period of LC neuronal genesis and NE axonal proliferation in the terminal fields, during which time these systems are sensitive to disruption by cocaine.</p><p>Adolescent brain development may represent a critical developmental stage in which prenatal cocaine effects may be expressed in a sex-dependent manner. Our findings of α<sub>2 </sub>adrenergic receptor upregulation in male rat hippocampus may be a compensatory response to cocaine-mediated increases in norepinephrine concentration during prenatal development. Interestingly, female offspring did not show a similar robust increase in α<sub>2 </sub>adrenergic receptor density in the hippocampus. However, in the amygdala the α<sub>2 </sub>adrenergic receptors were increased in females subsequent to prenatal cocaine exposure. In other regions such as the pyriform cortex, no changes attributable to cocaine were found. Our findings suggest sex-specific expression of cocaine-mediated alterations displayed in adolescence. Sex-specific alterations in receptor density may be restricted to particular brain regions, differ in directionality, and may not reflect global brain alterations.</p></sec><sec><title>Conclusion</title><p>In summary, IV exposure of pregnant rats to cocaine produced persistent, sex-specific alterations in the NE systems of adolescent offspring. Disruption of forebrain NE systems during the prenatal period might be the neurobiological basis for a number of functional disturbances occurring as a consequence of prenatal cocaine exposure.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Animals</title><p>Nulliparous female Sprague-Dawley rats were obtained from Harlan Sprague-Dawley, Inc. (Indianapolis, IN) at approximately 10–12 weeks of age (225–249 g), placed into quarantine for one week and subsequently moved to the animal colony. The animals were maintained according to NIH guidelines in AAALAC accredited facilities. Food (Pro-Lab Rat, Mouse, Hamster Chow No. 3000) and water were available ad libitum. The animal colony was maintained at 21 ± 2°C and 50 ± 10% relative humidity and a 12 hr light: 12 hr dark cycle with lights on at 07:00 h (EST).</p></sec><sec><title>Surgery</title><p>Half of the animals were surgically implanted with vascular catheters (as described below) and the remaining animals served as surrogate dams for the prenatal vehicle and cocaine treated pups. Catheterization was performed as previously described [<xref ref-type="bibr" rid="B38">38</xref>]. In brief, the animals were anesthetized with a mixture of ketamine hydrochloride (100 mg/kg/ml) and xylazine (3.3 mg/kg/ml) and a sterile Intracath IV catheter with a Luer-lock injection cap (Medex) was implanted dorsally in a subcutaneous pouch. The distal end of the catheter was inserted into the left jugular vein and advanced towards the heart. Animals were kept under periodic postoperative observation and returned to the vivarium upon recovery from anesthesia. Beginning on the day following surgery, the catheters were flushed daily with approximately 0.2 ml of 2.5% heparinized saline. The animals were observed for any signs of discomfort or behavioral distress. Complete anesthesia, surgical, recovery and postoperative records were maintained for each animal.</p></sec><sec><title>Animal mating</title><p>At 10–12 days following surgery, the females were group housed with males for breeding. Females were checked daily for vaginal cytology and the presence of sperm. Sperm positive females were considered at gestation day 0 (GD0) and individually housed in plastic cages with Sani-chip™ bedding throughout pregnancy and lactation.</p></sec><sec><title>Drug treatment</title><p>The catheterized, pregnant animals were randomly assigned to one of two groups that received either saline or 3.0 mg/kg cocaine. This dose of cocaine, delivered IV, was chosen based on reports that 1) it produces a peak arterial level in the male rat not significantly different from peak levels following administration of 32 mg of cocaine IV to humans [<xref ref-type="bibr" rid="B52">52</xref>,<xref ref-type="bibr" rid="B53">53</xref>], 2) the acute heart rate and blood pressure responses in late gestation pregnant rats are similar to those produced in a variety of other species (dog, monkey, humans; [<xref ref-type="bibr" rid="B54">54</xref>]), and 3) under experimental conditions, this dose is self-administered by users multiple times in a 2.5 hr session [<xref ref-type="bibr" rid="B55">55</xref>] Cocaine was administered as an IV bolus injection delivered in a volume of 1.0 ml/kg (15 sec) followed by flushing (15 sec) with 0.2 ml heparinized (2.5%) saline (i.e., the approximate volume of the catheter). Surrogate dams received neither drug treatment nor the daily handling associated with drug treatment. Cocaine and saline injections were administered 1x day from GD8-GD14 and 2x day from GD15-GD21.</p></sec><sec><title>Offspring treatment</title><p>All pregnant rats were checked twice daily for pups. Day of birth was defined as postnatal day 0 (P0). On P1, litters were weighed and culled to 8 pups with an equal number of males and females. Each pup was tattooed for identification and the culled litters of all catheterized dams were fostered to surrogate dams that had delivered within 24 h. Thus, no pups were raised by their biological mother or exposed to drug during the postnatal period. Pups were reared in their surrogate mothers' cage until weaning at P21, at which time offspring were group housed in same sex pairings. One male and one female from each litter were sacrificed on P35 for homogenate binding and autoradiographic analysis. Thus, each experiment included pups from 14 different litters.</p></sec><sec><title>α2-adrenergic receptor homogenate binding</title><p>Hippocampal tissue was collected for α<sub>2</sub>-adrenergic receptor binding from one male and one female from each litter on P35. Animals were killed by rapid decapitation and the brains removed. The hippocampi were dissected and immediately frozen in liquid nitrogen. Frozen hippocampal tissue was weighed and homogenized in 20 volumes/weight ice-cold 25 mM glycylglycine (GLYGLY) buffer (pH = 7.6). Crude homogenates were centrifuged at 20,000X g for 15 min at 4°C. The pellet was resuspended in 200 volumes 25 mM GLYGLY buffer (pH = 7.6) and the resulting crude membrane preparation was used for binding assays at a final protein content of approximately 0.6–0.7 mg/ml. Protein contents were determined by the Bradford [<xref ref-type="bibr" rid="B56">56</xref>] method (BioRad, Richmond, CA).</p><p>Cold saturation homogenate binding assays were performed as previously described [<xref ref-type="bibr" rid="B36">36</xref>]. Briefly, for the binding of [<sup>3</sup>H]RX821002 (58 Ci/mmol; Amersham, Arlington Heights, IL), 50 μl of labeled drug (2 nM) and 50 μl of assay buffer containing one of seventeen concentrations (10<sup>-12 </sup>to 10<sup>-6</sup>) of unlabeled RX821002 were added together and allowed to equilibrate to room temperature (22°C). Binding was initiated by addition of 900 μl of tissue (0.6–0.7 mg of protein) and allowed to incubate to equilibrium at 22°C for 60 min. Nonspecific binding was defined as [<sup>3</sup>H]RX821002 bound in the presence of 10 μM phentolamine. Binding was terminated by filtration under reduced pressure with a Brandel Tissue Harvestor (Gaithersburg, MD) followed by a 15 sec wash with ice-cold GLYGLY buffer onto GF/B glass fiber filters that had been presoaked for 2 hours with buffer containing 0.3% polyethyleneimine. Filters were dried overnight and analysis of radioligand bound was accomplished by scintillation spectrophotometry (40–50% efficiency).</p></sec><sec><title>α2-adrenergic receptor autoradiography</title><p>For preparation of tissue sections, animals were killed by rapid decapitation and the brains were carefully removed from the cranium, blocked and immediately frozen on powdered dry ice. The frozen tissue blocks were cryostat-sectioned (-20°C, 20 μm thick) in the standard coronal plane. Sections were collected at -3.3–3.8 mm relative to Bregma (plates 31–33 in Paxinos and Watson, [<xref ref-type="bibr" rid="B57">57</xref>]). Two additional adjacent sections were collected either for Nissl staining or acetylcholinesterase (AchE) staining [<xref ref-type="bibr" rid="B58">58</xref>] to aid in identification of the subregional structures. All sections were stored desiccated at -80°C prior to processing.</p><p>Frozen tissue sections were thawed and brought to room temperature (22°C; 5 min) followed by preincubation in 25 mM GLYGLY buffer (pH = 7.6) for 5 min. Slides were transferred into incubation vials containing 2 nM [<sup>3</sup>H]RX821002 and incubated at 22°C for 90 min. Nonspecific binding was defined by 10 μM phentolamine. Binding was terminated by transfer of slides into 4°C buffer and washing 2 × 2 min. Slides were then quickly rinsed in 4°C distilled water to remove buffer salts, dried under a stream of cool air and stored desiccated overnight under vacuum. Dried, labeled, tissue sections and [<sup>3</sup>H]microscales (Amersham, Arlington, IL) were exposed to tritium sensitive film (Hyperfilm, Amersham) for 14 days at room temperature in light-tight X-ray film cassettes. Films were developed using Kodak D-19 developer and Kodak rapid fixer.</p><p>Autoradiographic images were examined using the MCID-4 computerized image analysis system (Imaging Research, Ontario, Canada). The use of adjacent Nissl and AChE stained slides allowed for the determination of [<sup>3</sup>H]RX821002 binding in specific subregions of the hippocampus (stratum lacunosum-moleculare of CA1 near the hippocampal fissure). Binding density was also determined for the parietal cortex (layer II), amygdala (central nucleus), pyriform cortex (layer II-III) and hypothalamus (periventricular nucleus). These brain regions have high levels of α<sub>2 </sub>adrenergic receptors [<xref ref-type="bibr" rid="B59">59</xref>]. Adjacent sections, incubated with 10 μM phentolamine (nonspecific binding; < 10% total binding), were aligned with total binding sections and digitally subtracted to obtain "specific" binding images. Regional optical density data are expressed as fmol/mg weight [<xref ref-type="bibr" rid="B60">60</xref>,<xref ref-type="bibr" rid="B61">61</xref>].</p></sec><sec><title>Data analysis</title><p>Data were examined by ANOVA ([<xref ref-type="bibr" rid="B62">62</xref>]; BMDP statistical software, release 7, Los Angeles, CA, 1993). The Greenhouse-Geiser df correction factor was used for violations of compound symmetry [<xref ref-type="bibr" rid="B63">63</xref>]. An α level of <italic>p </italic>< 0.05 was the significance level set for rejection of the null hypothesis. Analysis of K<sub>D </sub>and B<sub>MAX </sub>values were determined using the GraphPAD-PRISM nonlinear curve fitting program (GraphPAD, San Diego, CA).</p></sec><sec><title>Chemicals</title><p>Cocaine HCl was purchased from Sigma Chemical Co. (St. Louis, MO) and was dissolved in sterile isotonic saline at the indicated concentrations based on the weight of the salt. All cocaine solutions were prepared immediately prior to use in a volume of 1 ml/kg. All other buffer chemicals were obtained from Sigma Chemical Co.</p></sec></sec><sec><title>Authors' contributions</title><p>RMB was responsible for the conception and design of the study, interpretation of the data, and drafting the manuscript. DRW participated in the conception and design of the study and carried out the binding studies. JMS participated in data analysis and manuscript preparation. BJS and DMS participated in the conception and interpretation of the study. CFM was responsible for experimental design, statistical analyses, data interpretation and drafting the manuscript. All authors read and approved the final manuscript.</p></sec>
Response of spontaneously hypertensive rats to inhalation of fine and ultrafine particles from traffic: experimental controlled study	<sec><title>Background</title><p>Many epidemiological studies have shown that mass concentrations of ambient particulate matter (PM) are associated with adverse health effects in the human population. Since PM is still a very crude measure, this experimental study has explored the role of two distinct size fractions: ultrafine (<0.15 μm) and fine (0.15- 2.5 μm) PM. In a series of 2-day inhalation studies, spontaneously hypersensitive (SH) rats were exposed to fine, concentrated, ambient PM (fCAP) at a city background location or a combination of ultrafine and fine (u+fCAP) PM at a location dominated by traffic. We examined the effect on inflammation and both pathological and haematological indicators as markers of pulmonary and cardiovascular injury. Exposure concentrations ranged from 399 μg/m<sup>3 </sup>to 3613 μg/m<sup>3 </sup>for fCAP and from 269μg/m<sup>3 </sup>to 556 μg/m<sup>3 </sup>for u+fCAP.</p></sec><sec><title>Results</title><p>Ammonium, nitrate, and sulphate ions accounted for 56 ± 16% of the total fCAP mass concentrations, but only 17 ± 6% of the u+fCAP mass concentrations. Unambiguous particle uptake in alveolar macrophages was only seen after u+fCAP exposures. Neither fCAP nor u+fCAP induced significant changes of cytotoxicity or inflammation in the lung. However, markers of oxidative stress (heme oxygenase-1 and malondialdehyde) were affected by both fCAP and u+fCAP exposure, although not always significantly. Additional analysis revealed heme oxygenase-1 (HO-1) levels that followed a nonmonotonic function with an optimum at around 600 μg/m<sup>3 </sup>for fCAP. As a systemic response, exposure to u+fCAP and fCAP resulted in significant decreases of the white blood cell concentrations.</p></sec><sec><title>Conclusion</title><p>Minor pulmonary and systemic effects are observed after both fine and ultrafine + fine PM exposure. These effects do not linearly correlate with the CAP mass. A greater component of traffic CAP and/or a larger proportion ultrafine PM does not strengthen the absolute effects.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Kooter</surname><given-names>Ingeborg M</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Boere</surname><given-names>A John F</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Fokkens</surname><given-names>Paul HB</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Leseman</surname><given-names>Daan LAC</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Dormans</surname><given-names>Jan AMA</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A6" corresp="yes" contrib-type="author"><name><surname>Cassee</surname><given-names>Flemming R</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	Particle and Fibre Toxicology	<sec><title>Background</title><p>Epidemiological studies have shown that exposure to ambient particulate air pollution (particulate matter or PM) is associated with many health effects [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B3">3</xref>], which include premature death, increased hospitalization for cardiopulmonary diseases, airway complaints, and reduced lung function. Although estimates of relative risks are small, there is a public-health concern because many people are exposed and there are high-risk groups, such as the elderly, very young children, and people with cardiopulmonary diseases. Although the PM-associated adverse health effects have been found all over the world, a more closer look reveals that there seem to be heterogeneous across locations [<xref ref-type="bibr" rid="B4">4</xref>-<xref ref-type="bibr" rid="B8">8</xref>], which might be due to differences of anthropogenic sources such as traffic [<xref ref-type="bibr" rid="B9">9</xref>]. Particulate matter consists of many chemicals, but it is not very likely that some of them (sea salt, sulphate, and nitrate [<xref ref-type="bibr" rid="B10">10</xref>] in ambient air affect health adversely. Aerosolized combustion products from traffic, shipping, industry, and domestic heating are believed to be far more relevant. The risk can be effectively reduced by reducing the PM fraction that is most likely to cause adverse health effects.</p><p>Due to its the complexity, the best way to study PM is by studying the effects of inhaling it. Systems designed to deliver controlled amounts of concentrated ambient particles now exist and allow a mechanistic approach to determining the effect of inhaled PM in different size ranges [<xref ref-type="bibr" rid="B11">11</xref>-<xref ref-type="bibr" rid="B13">13</xref>]. Recently published studies have shown that exposing rodents [<xref ref-type="bibr" rid="B14">14</xref>-<xref ref-type="bibr" rid="B20">20</xref>], dogs [<xref ref-type="bibr" rid="B21">21</xref>] and human volunteers [<xref ref-type="bibr" rid="B22">22</xref>-<xref ref-type="bibr" rid="B26">26</xref>] to concentrated ambient particles (CAP) indicate that PM has the potential to cause adverse effects. Biological responses to high concentrations of PM (which were often well above ambient PM concentrations) were observed. The sensitivity of these toxicological studies is low because of the small number of observations, as well as the fact that exposure levels and PM composition vary from day to day. However, data from homogeneous populations, as well as the use of specific disease models that mimic human risk groups, should increase study sensitivity to the effects of CAP exposures. The results of a series of 1-day (6 h/day) inhalation exposures of compromised rats to fCAP [<xref ref-type="bibr" rid="B27">27</xref>] revealed that CAP can increase inflammation [polymorphonuclear leukocytes (PMNs)] and toxicity [protein and albumin in bronchoalveolar lavage fluid (BALF)], and it can also increase the risk of thrombotic vascular disorders (fibrinogen). Nonetheless, we were unable to prove consistent relationships between PM mass and biological effects. While alterations of biological endpoints were occasionally statistically significant and potentially biologically relevant, we found no convincing proof that ambient PM exposures (up to 3500 μg/m<sup>3</sup>) can modify homeostasis. In another study in which rats pre-treated with ozone or with induced pulmonary hypertension were exposed for 6 h to concentrated freshly generated diesel exhaust particles up to 9000 μg/m<sup>3</sup>, no noteworthy pulmonary toxicity was observed, though increased glutathione levels (in the case of ozone-treated rats) and increased blood fibrinogen levels in rats with existing pulmonary hypertension were observed [<xref ref-type="bibr" rid="B20">20</xref>]. Although these studies and those of others [<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B15">15</xref>] have shown that CAP exposure has adverse effects, no studies have yet been published that convincingly prove relationships between mass concentration and these biologically relevant outcomes [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B27">27</xref>]. Bearing the available evidence of studies exposing animal or human subjects to CAP, PM mass concentrations do not seem to be the optimal metric to be associated with the adverse health effects. In that respect it is noteworthy that monotonic functions are usually sufficient to describe the relationship between air pollution and health effects in the epidemiological studies. However, Seagreave and colleagues [<xref ref-type="bibr" rid="B28">28</xref>] have recently shown that at least some parameters may respond in a way that achieves an optimum concentration beyond which the effects are reduced again.</p><p>Most toxicological studies that have used CAP exposures have focused on the accumulation mode (0.15 μm – 2.5 μm) [<xref ref-type="bibr" rid="B15">15</xref>,<xref ref-type="bibr" rid="B17">17</xref>,<xref ref-type="bibr" rid="B19">19</xref>] or coarse mode (2.5 μm – 10 μm) [<xref ref-type="bibr" rid="B17">17</xref>] of ambient PM. In recent years, more and more information has shown that the ultrafine fraction within PM might be more toxic than the fine mode. It has been suggested that the large surface area or particle number, or perhaps just the particle size, may play an important role in such differential responses [<xref ref-type="bibr" rid="B29">29</xref>,<xref ref-type="bibr" rid="B30">30</xref>]. Other studies have shown that the ultrafine fraction of urban ambient aerosols is not necessarily more potent than the fine or coarse fractions in inducing inflammatory and toxic effects in lung cells [<xref ref-type="bibr" rid="B31">31</xref>].</p><p>In order to compare the potential of PM fractions to induce adverse biological effects, we performed a series of studies of spontaneously hypertensive (SH) rats in which we used both concentrated fine PM (fCAP) and ultrafine + fine PM (u+fCAP). We compared the effects of PM that included ultrafines at a site with traffic as the major source with toxicity of accumulation mode (fine PM) at an urban background site with no dominant source of PM emission. A strain of SH rats was selected as a strain that would be more sensitive to PM [<xref ref-type="bibr" rid="B19">19</xref>] and to allow interstudy comparison [<xref ref-type="bibr" rid="B27">27</xref>,<xref ref-type="bibr" rid="B15">15</xref>].</p></sec><sec><title>Results</title><sec><title>Exposure characteristics</title><p>Integrated and continuous exposure characterization techniques were applied to determine the concentration and the composition of the test atmospheres at the different exposures (Table <xref ref-type="table" rid="T1">1</xref>). Temperature and relative humidity were 22 ± 2°C and 40 ± 10% for the fCAP exposure and both control exposures; they were 22 ± 2°C and 70 ± 10% for the u+fCAP exposure. The mean overall levels of some gaseous pollutants before passing the air through the concentrator were: 12 μg/m<sup>3 </sup>for ozone, 39 μg/m<sup>3 </sup>for carbon monoxide, 29 μg/m<sup>3 </sup>for sulphur dioxide, 19 μg/m<sup>3 </sup>for nitrogen monoxide, 13 μg/m<sup>3 </sup>for nitrogen dioxide and 28 μg/m<sup>3 </sup>for total nitrogen oxides (NOx). Ambient ozone is known to be efficiently removed in the concentrator due to the large metal surface, whereas all other gaseous components will remain at the same concentration or will show a little decrease. Table <xref ref-type="table" rid="T1">1</xref> shows that the sum of ammonium, nitrate, and sulphate ions accounts for 56 ± 16% of the total fCAP mass concentrations for the Bilthoven site (I), whereas it only accounts for 17 ± 6% of the u+fCAP mass concentrations at the HIA site (II). Most of the remaining mass is very likely associated with carbonaceous material, and the difference between site I and site II reflects the enrichment in organic and elemental carbon, as expected, given its proximity to mobile sources.</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Exposure characteristics for studies in spontaneously hypertensive rats exposed to concentrated ambient particulate matter</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left">Exposure</td><td></td><td align="left" colspan="5">fCAP (Site I = Bilthoven)</td><td align="left" colspan="8">u+fCAP (Site II = HIA)</td></tr></thead><tbody><tr><td align="left">Date</td><td></td><td align="left">6-1-03</td><td align="left">13-1-03</td><td align="left">20-1-03</td><td align="left">27-1-03</td><td align="left">11-2-03</td><td align="left">8-7-03</td><td align="left">29-9-03</td><td align="left">6-10-03</td><td align="left">13-10-03</td><td align="left">4-10-04</td><td align="left">6-10-04</td><td align="left">11-10-04</td><td align="left">13-10-04</td></tr><tr><td align="left">Mass<sup>1)</sup></td><td align="left">μg/m<sup>3</sup></td><td align="left">1067.5</td><td align="left">456.5</td><td align="left">399.0</td><td align="left">609.5</td><td align="left">3613.0</td><td align="left">269.0</td><td align="left">409.0</td><td align="left">366.5</td><td align="left">379.0</td><td align="left">501.2</td><td align="left">448.2</td><td align="left">555.8</td><td align="left">534.7</td></tr><tr><td align="left">Diameter<sup>2)</sup></td><td align="left">μm</td><td align="left">0.65</td><td align="left">0.74</td><td align="left">0.71</td><td align="left">0.75</td><td align="left">0.71</td><td align="left">0.68</td><td align="left">0.72</td><td align="left">0.75</td><td align="left">0.67</td><td align="left">1.09</td><td align="left">1.41</td><td align="left">0.58</td><td align="left">1.08</td></tr><tr><td align="left">gsd<sup>2)</sup></td><td></td><td align="left">1.14</td><td align="left">1.34</td><td align="left">1.35</td><td align="left">1.30</td><td align="left">1.17</td><td align="left">1.32</td><td align="left">1.33</td><td align="left">1.56</td><td align="left">1.35</td><td align="left">0.21</td><td align="left">0.28</td><td align="left">0.25</td><td align="left">0.22</td></tr><tr><td align="left">Number counts<sup>2)</sup></td><td></td><td></td><td></td><td></td><td></td><td></td><td align="left">320,000</td><td align="left">470,000</td><td align="left">66,000</td><td align="left">45,000</td><td align="left">998,750</td><td align="left">1,310,320</td><td align="left">1,624,000</td><td align="left">1,282,000</td></tr><tr><td align="left" colspan="15"> <italic>Compounds</italic><sup>3)</sup></td></tr><tr><td align="left"> SO<sub>4</sub><sup>2-</sup></td><td align="left">μg/m<sup>3</sup></td><td align="left">240</td><td align="left">86</td><td align="left">63</td><td align="left">98</td><td align="left">908</td><td align="left">19</td><td align="left">27</td><td align="left">25</td><td align="left">17</td><td align="left">14</td><td align="left">28</td><td align="left">26</td><td align="left">41</td></tr><tr><td align="left"> NO<sub>3</sub><sup>-</sup></td><td align="left">μg/m<sup>3</sup></td><td align="left">349</td><td align="left">86</td><td align="left">66</td><td align="left">112</td><td align="left">860</td><td align="left">16</td><td align="left">37</td><td align="left">16</td><td align="left">37</td><td align="left">33</td><td align="left">50</td><td align="left">33</td><td align="left">53</td></tr><tr><td align="left"> NH<sub>4</sub><sup>+</sup></td><td align="left">μg/m<sup>3</sup></td><td align="left">194</td><td align="left">53</td><td align="left">40</td><td align="left">68</td><td align="left">684</td><td align="left">8</td><td align="left">18</td><td align="left">10</td><td align="left">22</td><td align="left">9</td><td align="left">17</td><td align="left">11</td><td align="left">25</td></tr><tr><td align="left"> Cl<sup>-</sup></td><td align="left">μg/m<sup>3</sup></td><td align="left">22</td><td align="left">51</td><td align="left">42</td><td align="left">96</td><td align="left">65</td><td align="left">38</td><td align="left">33</td><td align="left">45</td><td align="left">35</td><td align="left">1</td><td align="left">1</td><td align="left">1</td><td align="left">1</td></tr><tr><td align="left" colspan="15"> <italic>Gases</italic><sup>4)</sup></td></tr><tr><td align="left"> O<sub>3</sub></td><td align="left">μg/m<sup>3</sup></td><td align="left">22</td><td align="left">16</td><td align="left">22</td><td align="left">22</td><td align="left">10</td><td align="left">nm</td><td align="left">nm</td><td align="left">nm</td><td align="left">nm</td><td align="left">10</td><td align="left">0</td><td align="left">0</td><td align="left">0</td></tr><tr><td align="left"> CO</td><td align="left">μg/m<sup>3</sup></td><td align="left">32</td><td align="left">42</td><td align="left">32</td><td align="left">31</td><td align="left">48</td><td align="left">nm</td><td align="left">nm</td><td align="left">nm</td><td align="left">nm</td><td align="left">39</td><td align="left">40</td><td align="left">41</td><td align="left">37</td></tr><tr><td align="left"> SO<sub>2</sub></td><td align="left">μg/m<sup>3</sup></td><td align="left">40</td><td align="left">42</td><td align="left">36</td><td align="left">26</td><td align="left">12</td><td align="left">nm</td><td align="left">nm</td><td align="left">nm</td><td align="left">nm</td><td align="left">6</td><td align="left">14</td><td align="left">14</td><td align="left">13</td></tr><tr><td align="left"> NO</td><td align="left">μg/m<sup>3</sup></td><td align="left">6</td><td align="left">5</td><td align="left">5</td><td align="left">5</td><td align="left">6</td><td align="left">nm</td><td align="left">nm</td><td align="left">nm</td><td align="left">nm</td><td align="left">28</td><td align="left">32</td><td align="left">50</td><td align="left">27</td></tr><tr><td align="left"> NO<sub>2</sub></td><td align="left">μg/m<sup>3</sup></td><td align="left">9</td><td align="left">11</td><td align="left">9</td><td align="left">9</td><td align="left">21</td><td align="left">nm</td><td align="left">nm</td><td align="left">nm</td><td align="left">nm</td><td align="left">17</td><td align="left">19</td><td align="left">15</td><td align="left">17</td></tr><tr><td align="left"> NO<sub>x</sub></td><td align="left">μg/m<sup>3</sup></td><td align="left">10</td><td align="left">12</td><td align="left">9</td><td align="left">9</td><td align="left">13</td><td align="left">nm</td><td align="left">nm</td><td align="left">nm</td><td align="left">nm</td><td align="left">41</td><td align="left">45</td><td align="left">55</td><td align="left">41</td></tr></tbody></table><table-wrap-foot><p>Experiments A-E concentrated PM in size range of 0.15–2.5 μm, F-N in size range <2.5μm including PM<0.15 μm. fCAP, fine concentrated ambient particulate matter; u+fCAP, ultrafine plus fine concentrated ambient particulate matter</p><p>1) Time-integrated 2-day mean mass concentrations at the outlet of the particle concentrator</p><p>2) Diameter, gsd (geometric standard deviation) and number are measured by an aerodynamic particle sizer which measures only particles larger than 0.5 μm. Results are based on numbers</p><p>3) Chemical analyses (ion chromatography) of filter used for mass concentration</p><p>4) nm, not measured</p></table-wrap-foot></table-wrap></sec><sec><title>Bronchoalveolar lavage</title><p>Table <xref ref-type="table" rid="T2">2</xref> presents the overall results of the BALF biochemical analyses. The contribution to the day-to-day variation was calculated in a two-way ANOVA for the biological indices of both the filtered air and the CAP- exposed animals. Some of the parameters showed substantial day-to-day differences that might be attributed to the assays and need to be taken into account as a confounding factor in the overall statistical analysis. The CAP exposures, with either the combined fCAP exposures (at site I Bilthoven) or the combined u+fCAP exposures (at site II HIA) did not result in statistically significant changes of most biochemical analyses measured in BALF compared to their filter-air controls. No signs of cytotoxicity as indicated by unchanged LDH and ALP levels were observed. No significant changes were found either in the number of cells or cell differentiation in the BALF. However, clear particle uptake was observed in macrophages lavaged from the rat lungs after exposure to fCAP or u+fCAP. MDA, a parameter for lipid peroxidation, is significantly lowered by u+fCAP exposures (Table <xref ref-type="table" rid="T2">2</xref>).</p><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Health effect parameters measured in lung lavage fluid of spontaneously hypertensive rats 8 h afterexposure to concentrated ambient particulate matter or clean air as a control.</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td></td><td align="left" colspan="5"><bold>fCAP (Site I)</bold></td><td align="left" colspan="5"><bold>u+fCAP (Site II)</bold></td></tr></thead><tbody><tr><td></td><td></td><td align="left" colspan="2">Control (<italic>n </italic>= 40)</td><td align="left" colspan="3">CAPs (<italic>n </italic>= 40)</td><td align="left" colspan="2">Control (<italic>n </italic>= 64)</td><td align="left" colspan="3">CAPs (<italic>n </italic>= 64)</td></tr><tr><td colspan="12"><hr></hr></td></tr><tr><td align="left"><bold>Parameter</bold></td><td align="left">Units</td><td align="left"><bold>Mean</bold></td><td align="left">95% CI</td><td align="left"><bold>Mean</bold></td><td align="left">95% CI</td><td align="left">Sign.</td><td align="left"><bold>Mean</bold></td><td align="left">95% CI</td><td align="left"><bold>Mean</bold></td><td align="left">95% CI</td><td align="left">Sign.</td></tr><tr><td align="left"><bold>ALP</bold></td><td align="left">U/L</td><td align="left"><bold>57.0</bold></td><td align="left">52.0–62.1</td><td align="left"><bold>51.6</bold></td><td align="left">44.9–58.3</td><td></td><td align="left"><bold>26.7</bold></td><td align="left">23.4–30.0</td><td align="left"><bold>26.1</bold></td><td align="left">22.9–29.2</td><td></td></tr><tr><td align="left"><bold>LDH</bold></td><td align="left">U/L</td><td align="left"><bold>66.3</bold></td><td align="left">62.2–70.4</td><td align="left"><bold>67.6</bold></td><td align="left">63.3–72.0</td><td></td><td align="left"><bold>55.7</bold></td><td align="left">49.5–61.9</td><td align="left"><bold>52.5</bold></td><td align="left">50.3–54.8</td><td></td></tr><tr><td align="left"><bold>Protein</bold></td><td align="left">mg/L</td><td align="left"><bold>314</bold></td><td align="left">277.9–350.4</td><td align="left"><bold>349</bold></td><td align="left">295.5–402.2</td><td></td><td align="left"><bold>274.5</bold></td><td align="left">242–308</td><td align="left"><bold>282.0</bold></td><td align="left">256–288</td><td></td></tr><tr><td align="left"><bold>NAG</bold></td><td align="left">U/L</td><td align="left"><bold>1.31</bold></td><td align="left">1.20–1.43</td><td align="left"><bold>1.43</bold></td><td align="left">1.32–1.54</td><td></td><td align="left"><bold>0.79</bold></td><td align="left">0.70–0.87</td><td align="left"><bold>0.81</bold></td><td align="left">0.74–0.88</td><td></td></tr><tr><td align="left"><bold>UA-B</bold></td><td align="left">umol/L</td><td align="left"><bold>0.29</bold></td><td align="left">0.22–0.35</td><td align="left"><bold>0.45</bold></td><td align="left">0.26–0.63</td><td></td><td align="left"><bold>0.46</bold></td><td align="left">0.36–0.56</td><td align="left"><bold>0.42</bold></td><td align="left">0.34–0.51</td><td></td></tr><tr><td align="left"><bold>Total Glut.</bold></td><td align="left">umol/L</td><td align="left"><bold>0.92</bold></td><td align="left">0.789–1.058</td><td align="left"><bold>1.00</bold></td><td align="left">0.836–1.165</td><td></td><td align="left"><bold>1.34</bold></td><td align="left">1.02–1.65</td><td align="left"><bold>1.22</bold></td><td align="left">1.12–1.31</td><td></td></tr><tr><td align="left"><bold>GSH</bold></td><td align="left">umol/L</td><td align="left"><bold>0.155</bold></td><td align="left">0.099–0.211</td><td align="left"><bold>0.244</bold></td><td align="left">0.158–0.329</td><td></td><td align="left"><bold>0.39</bold></td><td align="left">0.21–0.57</td><td align="left"><bold>0.31</bold></td><td align="left">0.22–0.40</td><td></td></tr><tr><td align="left"><bold>GSSG:GSH</bold></td><td align="left">ratio</td><td align="left"><bold>2.61</bold></td><td></td><td align="left"><bold>1.70</bold></td><td></td><td></td><td align="left"><bold>4.8</bold></td><td></td><td align="left"><bold>2.7</bold></td><td></td><td></td></tr><tr><td align="left"><bold>TNF-α </bold></td><td align="left">ng/ml</td><td align="left"><bold>46.4</bold></td><td align="left">42.5–50.2</td><td align="left"><bold>47.7</bold></td><td align="left">43.2–52.1</td><td></td><td align="left"><bold>45.9</bold></td><td align="left">37.6–54.2</td><td align="left"><bold>46.4</bold></td><td align="left">37.9–54.9</td><td></td></tr><tr><td align="left"><bold>MIP-2</bold></td><td align="left">ng/ml</td><td align="left"><bold>333</bold></td><td align="left">321–344</td><td align="left"><bold>311</bold></td><td align="left">297–325</td><td></td><td align="left"><bold>192.4</bold></td><td align="left">183–202</td><td align="left"><bold>192.2</bold></td><td align="left">184–200</td><td align="left"><sup>a</sup></td></tr><tr><td align="left"><bold>IL-1b</bold></td><td align="left">ng/ml</td><td align="left"><bold>268</bold></td><td align="left">237–299</td><td align="left"><bold>248</bold></td><td align="left">210–286</td><td></td><td align="left"><bold>n.m.</bold></td><td></td><td align="left"><bold>n.m.</bold></td><td></td><td></td></tr><tr><td align="left"><bold>CC16</bold></td><td align="left">ng/ml</td><td align="left"><bold>1.78</bold></td><td align="left">1.41–2.14</td><td align="left"><bold>1.98</bold></td><td align="left">1.56–2.40</td><td></td><td align="left"><bold>3.31</bold></td><td></td><td align="left"><bold>3.31</bold></td><td></td><td></td></tr><tr><td align="left"><bold>MDA</bold></td><td align="left">μmol/l</td><td align="left"><bold>0.268</bold></td><td align="left">0.200–0.335</td><td align="left"><bold>0.237</bold></td><td align="left">0.201–0.273</td><td></td><td align="left"><bold>0.407</bold></td><td align="left">0.341–0.473</td><td align="left"><bold>0.341</bold></td><td align="left">0.281–0.401</td><td align="left"></td></tr><tr><td align="left"><bold>HO-1</bold></td><td align="left">ng/ml</td><td align="left"><bold>0.393</bold></td><td align="left">0.296–0.491</td><td align="left"><bold>0.521</bold></td><td align="left">0.347–0.696</td><td></td><td align="left"><bold>0.517</bold></td><td align="left">0.418–0.615</td><td align="left"><bold>0.664</bold></td><td align="left">0.547–0.781</td><td align="left"></td></tr><tr><td align="left"><bold>Macrophages</bold></td><td align="left">%</td><td align="left"><bold>90.5</bold></td><td align="left">89.2–91.8</td><td align="left"><bold>88.7</bold></td><td align="left">87.2–90.1</td><td></td><td align="left"><bold>94.4</bold></td><td align="left">93.1–95.6</td><td align="left"><bold>94.2</bold></td><td align="left">93.0–95.3</td><td></td></tr><tr><td align="left"><bold>Neutrophils</bold></td><td align="left">%</td><td align="left"><bold>4.95</bold></td><td align="left">3.70–6.20</td><td align="left"><bold>4.76</bold></td><td align="left">3.54–5.99</td><td></td><td align="left"><bold>1.95</bold></td><td align="left">1.26–2.64</td><td align="left"><bold>1.95</bold></td><td align="left">1.37–2.54</td><td></td></tr><tr><td align="left"><bold>Eosinophils</bold></td><td align="left">%</td><td align="left"><bold>0.60</bold></td><td align="left">0.44–0.76</td><td align="left"><bold>0.44</bold></td><td align="left">0.28–0.61</td><td></td><td align="left"><bold>0.42</bold></td><td align="left">0.23–0.61</td><td align="left"><bold>0.34</bold></td><td align="left">0.21–0.46</td><td></td></tr><tr><td align="left"><bold>Lymfocytes</bold></td><td align="left">%</td><td align="left"><bold>1.05</bold></td><td align="left">0.83–1.27</td><td align="left"><bold>1.43</bold></td><td align="left">1.08–1.77</td><td></td><td align="left"><bold>2.60</bold></td><td align="left">1.93–3.28</td><td align="left"><bold>2.99</bold></td><td align="left">2.11–3.88</td><td></td></tr><tr><td align="left"><bold>Cell count</bold></td><td align="left">10e<sup>3</sup></td><td align="left"><bold>541</bold></td><td align="left">481–600</td><td align="left"><bold>528</bold></td><td align="left">473–583</td><td></td><td align="left"><bold>469</bold></td><td align="left">381–556</td><td align="left"><bold>417</bold></td><td align="left">349–485</td><td></td></tr><tr><td align="left"><bold>Macrophages</bold></td><td align="left">10e<sup>3</sup></td><td align="left"><bold>489</bold></td><td align="left">435–543</td><td align="left"><bold>461</bold></td><td align="left">415–507</td><td></td><td align="left"><bold>445</bold></td><td align="left">358–532</td><td align="left"><bold>394</bold></td><td align="left">327–461</td><td></td></tr><tr><td align="left"><bold>Neutrophils</bold></td><td align="left">10e<sup>3</sup></td><td align="left"><bold>25.70</bold></td><td align="left">17.2–34.2</td><td align="left"><bold>26.80</bold></td><td align="left">17.8–35.9</td><td></td><td align="left"><bold>7.88</bold></td><td align="left">4.36–11.41</td><td align="left"><bold>7.39</bold></td><td align="left">5.23–9.52</td><td></td></tr><tr><td align="left"><bold>Eosinophils</bold></td><td align="left">10e<sup>3</sup></td><td align="left"><bold>3.04</bold></td><td align="left">2.24–3.84</td><td align="left"><bold>2.60</bold></td><td align="left">1.47–3.73</td><td></td><td align="left"><bold>2.02</bold></td><td align="left">0.73–3.31</td><td align="left"><bold>1.46</bold></td><td align="left">0.80–2.11</td><td></td></tr><tr><td align="left"><bold>Lymfocytes</bold></td><td align="left">10e<sup>3</sup></td><td align="left"><bold>5.50</bold></td><td align="left">4.20–6.84</td><td align="left"><bold>7.70</bold></td><td align="left">5.60–9.81</td><td align="left">P = 0.06</td><td align="left"><bold>10.41</bold></td><td align="left">7.43–13.40</td><td align="left"><bold>11.87</bold></td><td align="left">8.01–15.73</td><td></td></tr></tbody></table><table-wrap-foot><p>*<italic>p </italic>< 0.01</p><p><sup>a</sup><italic>n </italic>= 32</p><p>ALP, alkaline phosphatase; LDH, lactate dehydrogenase; NAG, N-acetyl glucosaminidase; UA, uric acid, Total Glut., total glutathione; GSH, reduced glutathione;GSSG, oxidized glutathione; TNF-α, tumor necrotic factor; MIP, macrophage inhibiting factor;IL, interleukine, CC16, Clara cell protein; MDA, malondialdehyde; HO-1, heme oxygenase; CI, confidence interval; fCAP, fine concentrated ambient particulate matter; n.m., not measured; u+fCAP, ultrafine plus fine concentrated ambient particulate matter.</p></table-wrap-foot></table-wrap><p>Levels of the overall HO-1 values in BALF, as a measure for oxidative stress in the lungs, were increased by u+fCAP and fCAP exposures (after omitting data of the two experiments with the highest exposure concentrations, e.g. January 6<sup>th </sup>2003 and February 11 2003) (Table <xref ref-type="table" rid="T2">2</xref>). For the u+fCAP exposure, there was a significant increase to 0.664 ng/ml compared to the 0.517 ng/ml value of the control animals. However, more detailed data analyses of the fCAP data shows a clear nonmonotonic concentration-effect relationship with the HO-1. Maximum levels of HO-1 were observed at around 600 μg/m<sup>3 </sup>measured both in BALF and in lung homogenate (Figure <xref ref-type="fig" rid="F2">2</xref>). There is a correlation of <italic>r</italic><sup>2</sup>= 0.79 for the HO-1 measured in BALF and in lung homogenate (Figure <xref ref-type="fig" rid="F3">3</xref>). Data corrected for the amount of protein present were very comparable to Figure <xref ref-type="fig" rid="F2">2</xref> (data not shown). In none of the fCAP exposures did the LDH content in BALF change significantly upon exposure.</p><p>The only other parameter which showed significance at the individual fCAP exposures was CC16, which was significantly decreased at 457 μg/m<sup>3 </sup>fCAP and increased at the greatest exposure of 3613 μg/m<sup>3 </sup>fCAP (Figure <xref ref-type="fig" rid="F4">4</xref>).</p></sec><sec><title>Blood</title><p>After exposure to fCAP and u+fCAP, there was a significant decrease of WBCs (Table <xref ref-type="table" rid="T3">3</xref>). Consequently, absolute numbers of neutrophils and lymphocytes are decreased (although the decrease is not statistically significant for the number of neutrophils after u+fCAP exposure). Significant changes in haematological parameters, i.e., the mean platelet volume (MPV) and the mean platelet component (MPC), were found after exposure to u+fCAP, but these effects are not observed for the fCAP-exposed rats (the MPC parameter has not been measured for the fCAP exposures). No changes in concentrations of fibrinogen and vWF were observed in plasma after exposure to u+fCAP or fCAP. However, a significant increase of the greatest concentration of the fCAP exposure was observed for vWF (Figure <xref ref-type="fig" rid="F5">5</xref>).</p><table-wrap position="float" id="T3"><label>Table 3</label><caption><p>Health effect parameters measured in blood of spontaneously hypertensive rats 8 h after exposure to concentrated ambient particulate matter or clean air as a control.</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td></td><td align="left" colspan="5"><bold>fCAP (Site I)</bold></td><td align="left" colspan="6"><bold>u+fCAP (Site II)</bold></td></tr></thead><tbody><tr><td></td><td></td><td align="left" colspan="2">Control (<italic>n </italic>= 40)</td><td align="center" colspan="3">CAP (<italic>n </italic>= 40)</td><td align="left" colspan="2">Control (<italic>n </italic>= 32)</td><td align="center" colspan="3">CAP (<italic>n </italic>= 32)</td><td align="center">Variation among</td></tr><tr><td colspan="13"><hr></hr></td></tr><tr><td></td><td></td><td align="left">Control (<italic>n </italic>= 40)</td><td></td><td align="left">CAP (<italic>n </italic>= 40)</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left"><bold>Marker</bold></td><td align="left">Units</td><td align="left">Mean</td><td align="left">95% CI</td><td align="left">Mean</td><td align="left">95% CI</td><td></td><td align="left">Statistical significance</td><td align="left">95% CI</td><td align="center">Mean</td><td align="center">95% CI</td><td></td><td align="center">experiments</td></tr><tr><td align="left"><bold>WBC</bold></td><td align="left"><sup> </sup>e9/L</td><td align="left"><bold>3.20</bold></td><td align="left">3.01–3.38</td><td align="left"><bold>2.89</bold></td><td align="left">2.68–3.10</td><td align="left"></td><td align="left"><bold>3.82</bold></td><td align="left">3.61–4.02</td><td align="center"><bold>3.51</bold></td><td align="center">3.31–3.72</td><td align="center"></td><td align="center">**</td></tr><tr><td align="left"><bold>RBC</bold></td><td align="left"><sup> </sup>e12/L</td><td align="left"><bold>8.67</bold></td><td align="left">8.59–8.76</td><td align="left"><bold>8.74</bold></td><td align="left">8.66–8.82</td><td></td><td align="left"><bold>8.88</bold></td><td align="left">8.78–8.97</td><td align="center"><bold>8.94</bold></td><td align="center">8.86–9.03</td><td></td><td align="center">*</td></tr><tr><td align="left"><bold>HGB</bold></td><td align="left">mmol/L</td><td align="left"><bold>8.69</bold></td><td align="left">8.60–8.77</td><td align="left"><bold>8.75</bold></td><td align="left">8.67–8.82</td><td></td><td align="left"><bold>8.66</bold></td><td align="left">8.56–8.76</td><td align="center"><bold>8.69</bold></td><td align="center">8.60–8.78</td><td></td><td align="center"></td></tr><tr><td align="left"><bold>HCT</bold></td><td align="left">L/L</td><td align="left"><bold>0.404</bold></td><td align="left">0.400–0.408</td><td align="left"><bold>0.408</bold></td><td align="left">0.404–0.412</td><td align="left"></td><td align="left"><bold>0.42</bold></td><td align="left">0.410–0.419</td><td align="center"><bold>0.42</bold></td><td align="center">0.411–0.418</td><td></td><td align="center">*</td></tr><tr><td align="left"><bold>MCV</bold></td><td align="left">fL</td><td align="left"><bold>46.61</bold></td><td align="left">46.22–46.99</td><td align="left"><bold>46.74</bold></td><td align="left">46.35–47.14</td><td></td><td align="left"><bold>46.72</bold></td><td align="left">46.53–46.91</td><td align="center"><bold>46.33</bold></td><td align="center">46.13–46.53</td><td align="center"></td><td></td></tr><tr><td align="left"><bold>MCH</bold></td><td align="left">fmol</td><td align="left"><bold>1.002</bold></td><td align="left">0.994–1.010</td><td align="left"><bold>1.002</bold></td><td align="left">0.995–1.010</td><td></td><td align="left"><bold>0.98</bold></td><td align="left">0.971–0.980</td><td align="center"><bold>0.97</bold></td><td align="center">0.967–0.976</td><td></td><td align="center">*</td></tr><tr><td align="left"><bold>MCHC</bold></td><td align="left">mmol/L</td><td align="left"><bold>21.50</bold></td><td align="left">21.42–21.58</td><td align="left"><bold>21.42</bold></td><td align="left">21.35–21.50</td><td></td><td align="left"><bold>20.90</bold></td><td align="left">20.82–20.94</td><td align="center"><bold>20.97</bold></td><td align="center">20.90–21.04</td><td align="center"></td><td align="center">**</td></tr><tr><td align="left"><bold>RDW</bold></td><td align="left">%</td><td align="left"><bold>11.83</bold></td><td align="left">11.69–11.98</td><td align="left"><bold>11.68</bold></td><td align="left">11.55–11.80</td><td></td><td align="left"><bold>11.64</bold></td><td align="left">11.53–11.76</td><td align="center"><bold>11.67</bold></td><td align="center">11.51–11.82</td><td></td><td></td></tr><tr><td align="left"><bold>HDW</bold></td><td align="left">mmol/L</td><td align="left"><bold>1.44</bold></td><td align="left">1.411–1.4778</td><td align="left"><bold>1.44</bold></td><td align="left">1.408–1.474</td><td></td><td align="left"><bold>1.52</bold></td><td align="left">1.506–1.539</td><td align="center"><bold>1.53</bold></td><td align="center">1.515–1.544</td><td></td><td align="center"></td></tr><tr><td align="left"><bold>PLT</bold></td><td align="left"><sup>* </sup>e9/L</td><td align="left"><bold>942</bold></td><td align="left">913–972</td><td align="left"><bold>929</bold></td><td align="left">899–959</td><td></td><td align="left"><bold>992</bold></td><td align="left">964–1019</td><td align="center"><bold>992</bold></td><td align="center">946–1038</td><td></td><td align="center">*</td></tr><tr><td align="left"><bold>MPV</bold></td><td align="left">fL</td><td align="left"><bold>7.39</bold></td><td align="left">7.29–7.50</td><td align="left"><bold>7.38</bold></td><td align="left">7.31–7.46</td><td></td><td align="left"><bold>7.10</bold></td><td align="left">7.01–7.19</td><td align="center"><bold>7.24</bold></td><td align="center">7.16–7.32</td><td align="center"></td><td align="center"></td></tr><tr><td align="left"><bold>MPC</bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td align="left"><bold>23.30</bold></td><td align="left">23.06–23.56</td><td align="center"><bold>23.86</bold></td><td align="center">23.61–24.11</td><td align="center"><sup>a </sup></td><td align="center">*</td></tr><tr><td align="left"><bold>Reticulo-</bold></td><td align="left">abs</td><td align="left"><bold>75.8</bold></td><td align="left">71.3–80.4</td><td align="left"><bold>72.0</bold></td><td align="left">66.6–77.5</td><td></td><td align="left"><bold>75.8</bold></td><td align="left">71.3–80.4</td><td align="center"><bold>72.0</bold></td><td align="center">66.6–77.5</td><td></td><td></td></tr><tr><td align="left"> <bold>cytes</bold></td><td align="left">%</td><td align="left"><bold>0.87</bold></td><td align="left">0.77–0.96</td><td align="left"><bold>0.71</bold></td><td align="left">0.64–0.78</td><td align="left"></td><td align="left"><bold>0.86</bold></td><td align="left">0.815–0.92</td><td align="center"><bold>0.81</bold></td><td align="center">0.75–0.87</td><td></td><td align="center"></td></tr><tr><td align="left"><bold>Neutrophils</bold></td><td align="left">abs</td><td align="left"><bold>0.42</bold></td><td align="left">0.39–0.46</td><td align="left"><bold>0.33</bold></td><td align="left">0.31–0.36</td><td align="left">*</td><td align="left"><bold>0.51</bold></td><td align="left">0.47–0.55</td><td align="center"><bold>0.47</bold></td><td align="center">0.44–0.52</td><td></td><td align="center"></td></tr><tr><td align="left"><bold>Lymphocytes</bold></td><td align="left">abs</td><td align="left"><bold>2.65</bold></td><td align="left">2.48–2.83</td><td align="left"><bold>2.45</bold></td><td align="left">2.26–2.65</td><td align="left"></td><td align="left"><bold>3.13</bold></td><td align="left">2.96–3.30</td><td align="center"><bold>2.87</bold></td><td align="center">2.69–3.04</td><td align="center"></td><td align="center"></td></tr><tr><td align="left"><bold>Monocytes</bold></td><td align="left">abs</td><td align="left"><bold>0.075</bold></td><td align="left">0.067–0.083</td><td align="left"><bold>0.0687</bold></td><td align="left">0.060–0.077</td><td></td><td align="left"><bold>0.103</bold></td><td align="left">0.092–0.114</td><td align="center"><bold>0.098</bold></td><td align="center">0.0869–0.109</td><td></td><td align="center"></td></tr><tr><td align="left"><bold>Eosinophils</bold></td><td align="left">abs</td><td align="left"><bold>0.0262</bold></td><td align="left">0.023–0.029</td><td align="left"><bold>0.0262</bold></td><td align="left">0.022–0.029</td><td></td><td align="left"><bold>0.013</bold></td><td align="left">0.0073–0.0190</td><td align="center"><bold>0.013</bold></td><td align="center">0.0098–0.0165</td><td></td><td align="center"></td></tr><tr><td align="left"><bold>Basophils</bold></td><td align="left">abs</td><td align="left"><bold>0.00498</bold></td><td align="left">0.00347–0.00649</td><td align="left"><bold>0.00517</bold></td><td align="left">0.00381–0.00653</td><td></td><td align="left"><bold>0.060</bold></td><td align="left">0.0496–0.0699</td><td align="center"><bold>0.062</bold></td><td align="center">0.0506–0.0736</td><td></td><td align="center"></td></tr><tr><td align="left"><bold>LUC</bold></td><td align="left">abs</td><td align="left"><bold>0.0129</bold></td><td align="left">0.0109–0.0149</td><td align="left"><bold>0.0119</bold></td><td align="left">0.0100–0.0137</td><td></td><td align="left"><bold>0.008</bold></td><td align="left">0.0043–0.0107</td><td align="center"><bold>0.009</bold></td><td align="center">0.0054–0.0117</td><td></td><td align="center">*</td></tr><tr><td align="left"><bold>Neutrophils</bold></td><td align="left">%</td><td align="left"><bold>13.54</bold></td><td align="left">12.21–14.88</td><td align="left"><bold>11.90</bold></td><td align="left">10.82–12.98</td><td align="left"></td><td align="left"><bold>13.37</bold></td><td align="left">12.55–14.20</td><td align="center"><bold>13.48</bold></td><td align="center">12.44–14.51</td><td></td><td></td></tr><tr><td align="left"><bold>Lymphocytes</bold></td><td align="left">%</td><td align="left"><bold>82.62</bold></td><td align="left">81.22–84.02</td><td align="left"><bold>84.23</bold></td><td align="left">83.11–85.36</td><td align="left"></td><td align="left"><bold>81.90</bold></td><td align="left">80.98–82.76</td><td align="center"><bold>81.40</bold></td><td align="center">80.23–82.58</td><td></td><td></td></tr><tr><td align="left"><bold>Monocytes</bold></td><td align="left">%</td><td align="left"><bold>2.39</bold></td><td align="left">2.14–2.63</td><td align="left"><bold>2.33</bold></td><td align="left">2.15–2.52</td><td></td><td align="left"><bold>2.66</bold></td><td align="left">2.46–2.87</td><td align="center"><bold>2.72</bold></td><td align="center">2.51–2.93</td><td></td><td align="center"></td></tr><tr><td align="left"><bold>Eosinophils</bold></td><td align="left">%</td><td align="left"><bold>0.836</bold></td><td align="left">0.754–0.918</td><td align="left"><bold>0.935</bold></td><td align="left">0.823–1.048</td><td></td><td align="left"><bold>0.340</bold></td><td align="left">0.204–0.475</td><td align="center"><bold>0.353</bold></td><td align="center">0.276–0.430</td><td></td><td align="center"></td></tr><tr><td align="left"><bold>Basophils</bold></td><td align="left">%</td><td align="left"><bold>0.203</bold></td><td align="left">0.169–0.236</td><td align="left"><bold>0.192</bold></td><td align="left">0.161–0.223</td><td></td><td align="left"><bold>1.550</bold></td><td align="left">1.33–1.79</td><td align="center"><bold>1.790</bold></td><td align="center">1.47–2.12</td><td></td><td align="center"></td></tr><tr><td align="left"><bold>LUC</bold></td><td align="left">%</td><td align="left"><bold>0.384</bold></td><td align="left">0.340–0.427</td><td align="left"><bold>0.391</bold></td><td align="left">0.339–0.443</td><td></td><td align="left"><bold>0.218</bold></td><td align="left">0.152–0.285</td><td align="center"><bold>0.254</bold></td><td align="center">0.190–0.318</td><td></td><td align="center"></td></tr><tr><td align="left"><bold>Fibrinogen</bold></td><td align="left">mg/ml</td><td align="left"><bold>2.18</bold></td><td align="left">2.13–2.22</td><td align="left"><bold>2.16</bold></td><td align="left">2.13–2.19</td><td></td><td align="left"><bold>2.236</bold></td><td align="left">2.199–2.274</td><td align="center"><bold>2.267</bold></td><td align="center">2.232–2.303</td><td align="center"><sup>b</sup></td><td></td></tr><tr><td align="left"><bold>vWF</bold></td><td align="left">ratio</td><td align="left"><bold>1.49</bold></td><td align="left">1.34–1.65</td><td align="left"><bold>1.64</bold></td><td align="left">1.36–1.92</td><td></td><td align="left"><bold>0.67</bold></td><td align="left">0.65–0.70</td><td align="center"><bold>0.65</bold></td><td align="center">0.61–0.69</td><td align="center"><sup>a</sup></td><td></td></tr></tbody></table><table-wrap-foot><p><sup>a</sup><italic>n </italic>= 32; <sup>b</sup><italic>n </italic>= 61; <sup>c</sup><italic>n </italic>= 63; <italic>p </italic>< 0.05; <italic>p </italic>< 0.01; <italic>p </italic>< 0.001. For abbreviations of the markers, see Materials and methods</p></table-wrap-foot></table-wrap></sec><sec><title>Pathology</title><p>There were no differences in the lung and body weights of sham and CAP-exposed animals (data not shown). The characteristic hallmarks of the strain of rats used as controls were clearly present in their lungs: small alveolar haemorrhages and extensive bronchus-associated lymphoid tissue (BALT) at many bifurcations of the airways. These signs were not affected by the CAP exposures (Table <xref ref-type="table" rid="T4">4</xref>).</p><table-wrap position="float" id="T4"><label>Table 4</label><caption><p>Summary of histological lung changes due to 2-day filtered air control or exposure to concentrated ambient particulate matter</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="center" colspan="2"><bold>uCAP </bold><bold>(Site I)</bold></td><td align="center" colspan="2"><bold>u+fCAP </bold><bold>(Site II)</bold></td></tr></thead><tbody><tr><td align="left">Parameter</td><td align="left">Control<break/>(<italic>n </italic>= 40)</td><td align="left">CAP<break/>(n = 40)</td><td align="center">Control<break/>(<italic>n </italic>= 64)</td><td align="center">CAP<break/>(<italic>n </italic>= 64)</td></tr><tr><td align="left">Alveolar macrophages</td><td></td><td></td><td></td><td></td></tr><tr><td align="left"> minimal</td><td align="left">29</td><td align="left">33</td><td align="center">33</td><td align="center">41</td></tr><tr><td align="left"> slight</td><td align="left">0</td><td align="left">3</td><td align="center">2</td><td align="center">2</td></tr><tr><td align="left">Foci thick septa + macrophages</td><td></td><td></td><td></td><td></td></tr><tr><td align="left"> minimal</td><td align="left">22</td><td align="left">23</td><td align="center">31</td><td align="center">28</td></tr><tr><td align="left"> slight</td><td align="left">2</td><td align="left">0</td><td align="center">2</td><td align="center">5</td></tr><tr><td align="left">Perivascular infiltrate</td><td align="left">29</td><td align="left">22</td><td align="center">51</td><td align="center">58</td></tr><tr><td align="left">Foci interstitial pneumonia with alveolitis (macrophages + lymphocytes)</td><td></td><td></td><td></td><td></td></tr><tr><td align="left">minimal</td><td align="left">0</td><td align="left">0</td><td align="center">1</td><td align="center">0</td></tr><tr><td align="left">Peribronchitis+hypertrophy bronchial epithelium</td><td align="left">0</td><td align="left">0</td><td align="center">0</td><td align="center">1</td></tr><tr><td align="left">Macrophages loaded with Particulate matter</td><td align="left">0</td><td align="left">0</td><td align="center">0</td><td align="center">52</td></tr><tr><td align="left">Erythrocytes in alveoli</td><td align="left">28</td><td align="left">24</td><td align="center">41</td><td align="center">39</td></tr><tr><td align="left">BrdU score number examined</td><td align="left">40</td><td align="left">39</td><td align="center">64</td><td align="center">64</td></tr><tr><td align="left"> Minimal</td><td align="left">27</td><td align="left">28</td><td align="center">39</td><td align="center">35</td></tr><tr><td align="left"> Slight</td><td align="left">1</td><td align="left">0</td><td align="center">5</td><td align="center">4</td></tr></tbody></table><table-wrap-foot><p>BrdU, bromodeoxyuridine; CAP, concentrated ambient particulate matter;</p></table-wrap-foot></table-wrap><p>No noticeable pathological changes could be observed for either the fCAP or the u+fCAP exposure. Deposition of PM was noted in most of the animals exposed to u+fCAP (52 of 64). This was not observed for animals exposed to fCAP only. Groups of alveolar macrophages were present in nearly all CAP-exposed animals, as in controls.</p><p>The lymphocytes in the BALT areas and the perivascular infiltrate of all SH rats were fairly strongly labelled as noted by BrdU incorporation in the DNA, and there was a slight labelling in the alveolar area, which reflects the background turnover. There was slightly more BrdU labelling in all components of those areas of the H&E slides where inflammatory foci were present: an increased proliferation rate in the bronchiolar and alveolar epithelium, as well as in alveolar macrophages. No change of cell proliferation was seen from the labelling-frequency data of nuclei in the control and CAP-exposed groups, and there were no differences between u+fCAP and fCAP groups in this respect (Table <xref ref-type="table" rid="T4">4</xref>). The number of foci detected by the immunocytochemical BrdU procedure runs nearly parallel to the observed inflammatory foci with thickened septa in the H&E-stained sections.</p></sec></sec><sec><title>Discussion</title><p>Although some epidemiological studies suggest that ultrafine particles have serious health effects [<xref ref-type="bibr" rid="B32">32</xref>-<xref ref-type="bibr" rid="B34">34</xref>], others fail to prove that this PM fraction is more relevant to health than the fine fraction of PM [<xref ref-type="bibr" rid="B35">35</xref>,<xref ref-type="bibr" rid="B36">36</xref>], or they fail to separate the effects of ultrafine particles from other air pollutants [<xref ref-type="bibr" rid="B37">37</xref>]. The present study focuses on the hypothesis that the ultrafine fraction of PM2.5 dominates the biological responses of rats [<xref ref-type="bibr" rid="B32">32</xref>,<xref ref-type="bibr" rid="B38">38</xref>,<xref ref-type="bibr" rid="B39">39</xref>]. Although there are some small differences between repeated studies with fCAP and u+fCAP, these studies do not fully support our hypothesis.</p><p>Besides the normal pathological pulmonary characteristics of the SH rat, no differences (including differences in cell proliferation) were observed between the control rats and the fCAP or u+fCAP exposed rats. However, in contrast to the animals exposed to fCAP (in which more than half of the mass consists of sulphate, nitrate, and ammonium), black particles appeared in the alveolar macrophages of the rats exposed to u+fCAP. This indicates that significant amounts of insoluble particles were deposited in the lungs. It is likely that they originated from combustion processes. Since these exposures were carried out next to a busy freeway, a substantial part of the pollution will have been produced by traffic. Most of the biological endpoints were not affected at all, and the biological relevance of markers that were influenced by CAP exposures remains questionable. This is in line with a previous study [<xref ref-type="bibr" rid="B27">27</xref>], in which fCAP (1-day exposures) produced little effect in the bronchoalveolar lavage fluid of SH rats, with the exception of an increase of PMNs.</p><p>We added two additional parameters for oxidative stress to the present study: MDA as a measure of lipid peroxidation and HO-1 as a measure of antioxidant response. An increase of lipid peroxidation is expected after exposure to air pollution with oxidative capacity, such as pollution by ozone or PM containing polycyclic aromatic hydrocarbon (PAH) and metals. However, exposure to CAP in the present study resulted in either no significant change (fCAP) or a significant decrease (u+fCAP) of MDA levels in bronchiolar lavage fluid. The biological relevance of these observations is questionable, given the smallness of the decline. Inconsistent data regarding both humans and rats have been reported both as increases [<xref ref-type="bibr" rid="B40">40</xref>,<xref ref-type="bibr" rid="B41">41</xref>] and decreases [<xref ref-type="bibr" rid="B42">42</xref>] of lipid peroxidation after exposure to particles. Where the increase is often seen as a direct result of the oxidative stress, the decrease is assigned to the adaptive capacity of the human organism after prolonged exposure [<xref ref-type="bibr" rid="B42">42</xref>].</p><p>Exposure to fCAP (after omitting the two highest exposure concentrations) or u+fCAP resulted in a significant increase in HO-1. The enzyme HO-1 is regulated by oxidative stress, catalysing heme oxidation into biliverdin, CO, and iron. The increase results in augmented production of the antioxidant biliverdin and CO, which acts as an anti-inflammatory agent [<xref ref-type="bibr" rid="B43">43</xref>]. Indeed, the present study finds no sign of developing inflammation. Various agents, such as endotoxins, cytokines, and heavy metals, as well as CO itself, are known to induce HO-1 [<xref ref-type="bibr" rid="B44">44</xref>,<xref ref-type="bibr" rid="B45">45</xref>]. The use of HO-1 as a biologically relevant indicator of PM-induced stress has been exemplified in <italic>in vitro </italic>studies in which the PAH content derived from airborne PM positively correlates with increased HO-1 expression [<xref ref-type="bibr" rid="B46">46</xref>,<xref ref-type="bibr" rid="B29">29</xref>]. It also has been proven that the oxidative potential of CAP in <italic>in vitro </italic>studies correlate well with HO-1 induction [<xref ref-type="bibr" rid="B47">47</xref>]. <italic>In vitro </italic>studies that use a murine macrophage cell line [<xref ref-type="bibr" rid="B47">47</xref>], have shown that ultrafine PM is a more potent inducer of HO-1 and depleter of intracellular glutathione (anti-oxidant) than ultrafine+fine PM. However, the present study gives no clear indications that, per unit mass, u+fCAP has a greater impact on HO-1 or glutathione than fCAP. Although this might occur at higher exposure levels for u+fCAP than were achieved in this study (550 μg/m<sup>3</sup>), the oxidative stress potency as measured by the HO-1 of ultrafine particles deposited in the lungs may not be significantly greater than that of fine particles.</p><p>A statistically significant decrease of WBCs after exposure to fCAP and u+fCAP was noted. There was a concomitant decrease of absolute numbers of neutrophils and lymphocytes. It has previously been reported that these systemic responses, such as a decrease in WBC, are observed 24 h after exposure [<xref ref-type="bibr" rid="B25">25</xref>], but not immediately after exposure to CAP [<xref ref-type="bibr" rid="B25">25</xref>,<xref ref-type="bibr" rid="B26">26</xref>]. Ghio states that the extravasion of neutrophils from the blood into the lung would account for the time dependency of the decrement of the WBC count, since an inflammatory influx after PM exposure is not immediate, but certainly becomes evident at 24 h [<xref ref-type="bibr" rid="B25">25</xref>]. However, others [<xref ref-type="bibr" rid="B48">48</xref>] report that haematological changes (increase in blood neutrophils and a decrease in lymphocytes 3 h after exposure of rats to CAP (3 h for 110–350 μg/m<sup>3</sup>), but these changes were absent at 24 h after exposure. Such an increase in WBC counts purportedly reflects the inflammatory state after inhalation of PM. In a study similar to ours, Kodavanti et al. [<xref ref-type="bibr" rid="B15">15</xref>] report that there were no significant changes in haematological parameters 18–20 h after exposure.</p><p>In contrast to previous findings [<xref ref-type="bibr" rid="B15">15</xref>,<xref ref-type="bibr" rid="B27">27</xref>], we do not observe changes in plasma fibrinogen concentrations as a result of exposure to both u+fCAP and fCAP. The fibrinogen concentration, a risk factor for cardiovascular disease, has been shown to increase after PM exposure in rodent studies [<xref ref-type="bibr" rid="B49">49</xref>-<xref ref-type="bibr" rid="B51">51</xref>] as well as in human studies [<xref ref-type="bibr" rid="B25">25</xref>,<xref ref-type="bibr" rid="B52">52</xref>], although decreases have also been observed [<xref ref-type="bibr" rid="B49">49</xref>]. A strain-dependent effect in the increase of fibrinogen has also been observed: the effects seem more dominant in SH rats than in Wistar-Kyoto (WKY) rats. This supports the hypothesis that humans with cardiovascular diseases may be more susceptible to increased pulmonary and cardiac impairments [<xref ref-type="bibr" rid="B51">51</xref>,<xref ref-type="bibr" rid="B53">53</xref>]. A significant increase of fibrinogen levels in SH rats (but not WKY rats) exposed to CAPs has been found [<xref ref-type="bibr" rid="B15">15</xref>]. Sela et al [<xref ref-type="bibr" rid="B54">54</xref>] proved that oxidative stress resulted in increased plasma MDA, fibrinogen, and PMN counts before hypertension developed in the rat. In this respect, the absence of increased serum fibrinogen levels is rather unexpected. Differences in CAP composition might explain these differences.</p><p>The significant changes on the haematological indices corpuscular volume (increased MPV and decreased MCV) in combination with unchanged platelet numbers found after exposure to u+fCAP suggests that the ultrafine particles might have effects on the platelet state. These observations are therefore in agreement with the hypothesis that PM can affect haematological indices [<xref ref-type="bibr" rid="B55">55</xref>]. It has been observed that a reduction of MPC may be used to detect <italic>in vitro </italic>platelet activation [<xref ref-type="bibr" rid="B56">56</xref>]. However, since the observed differences are very small and opposite in sign, it is questionable whether the changes are biologically relevant.</p><p>The fact is that u+fCAP contains less of the soluble inorganic aerosols, sulphates, nitrates, and ammonium than fCAP. Indeed, u+fCAP is very likely to be enriched with carbonaceous (combustion derived) PM since these exposures took place right next to a traffic tunnel. However, except for the black particles in the lung tissue and the changed haematological parameters concerning the platelet state, the effects observed for the fCAP and u+fCAP exposures do not differ.</p><p>Studies using concentrated PM are confronted with the fact that air pollution is a complex mixture that varies from day to day. As a consequence, duplication of experiments is virtually impossible. Epidemiological studies have consistently demonstrated that health effects can be predicted with monotonic functions of particle mass concentrations. In turn, CAP studies can be used to verify this relationship under more controlled conditions. In other words, this allows for testing the hypothesis that effects on biological systems due to exposure to PM are always linearly related to the mass concentration. However, the results of the present study indicate that, for example, HO-1 levels show a significant nonlinear relationship with particle mass concentrations. Similar response patterns were seen for CC16 in lavage fluid. In both cases, combined analysis of all experiments did not reveal a statistically significant effect due to CAP exposures. Some other studies report that linear regression shows correlations between a biomarker and the exposure concentrations [<xref ref-type="bibr" rid="B27">27</xref>,<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B28">28</xref>], but the correlations are usually rather poor. Unfortunately, many of the other markers of biological effect in the present study were not affected at all, which precluded a similar analysis. Therefore, analysis of the combined data of multiple CAPs experiments should always consider that nonmonotonic relationships of concentration effects may be better descriptors of physiological processes. The phenomenon of nonmonotonic relationships of concentration effects applies to many biological processes, for example, enzymatic activity as a function of temperature.</p><p>The results of the present series of experiments indicate that minor pulmonary and systemic effects are due to exposure to fine and ultrafine+fine particles at concentrations well above ambient levels. No clear CAP mass correlation has been found for these effects based on the assumption that this relation is linear as shown in epidemiological studies. We even provide evidence that effects due to the oxidative potential of PM might be masked at greater than ambient concentrations, so that prudence is called for when sets of exposures of various concentrations are combined in order to increase the group number of observations in a statistical analysis. In addition, this study shows no proof that change of location, resulting in a larger traffic CAP component, results in noteworthy and biological relevant pulmonary or systemic effects.</p></sec><sec sec-type="materials\|methods"><title>Materials and methods</title><sec><title>Animals</title><p>Male SH rats, 11–13 weeks old, were purchased from Charles River Laboratories were assigned to 13 studies (Table <xref ref-type="table" rid="T1">1</xref>). Immediately after arrival, the animals were weighed, randomized and then allowed to acclimatize for at least 7 days. The animals were housed in macrolon cages (type III) and fed with SSP-TOX pellets of a cereal-based rodent diet (SMR-A; Hope Farms, Woerden, the Netherlands) and tap water via the automatic drinking-water system, both ad libitum during nonexposure periods. The room temperature was maintained at 22 ± 2°C, the relative humidity at 40–70%, and a 12-h light/dark cycle was maintained.</p><p>Each study used eight animals for CAP exposure and eight animals for filtered air exposure (control group). Shortly before exposure, the animals were transported to the mobile exposure laboratory equipped with an ambient particle concentrator, where they were housed in macrolon type III cages, equipped with water bottles. The housing facilities were ventilated with HEPA filters and chemically (activated carbon and purafil) filtered air. Directly after the animals were exposed to CAP or clean air (which was done simultaneously), they were returned to the housing facilities. During exposure, the animals were deprived of water and food.</p></sec><sec><title>Study design</title><p>A total of 13 studies (which were identical apart from the CAP exposure) were performed at two locations. The SH rats were either exposed to fCAP (five replicate studies located in a city background in Bilthoven = site I) or to u+fCAP (eight replicate studies located in a freeway tunnel near Hendrik Ido Ambacht (HIA; site II)). Each study consisted of a control group of eight animals exposed to HEPA-filtered air, and a group of eight animals exposed to CAP. Exposures lasted 6 h on 2 consecutive days (Figure <xref ref-type="fig" rid="F1">1</xref>).</p></sec><sec><title>Generation and characterization of the test atmosphere</title><p>To obtain fCAP, ambient PM was generated by drawing ambient air through a size selective inlet that removes particles larger than 2.5 μm and subsequently through a four-stage set-up of the ambient fine particle concentrator (AFPC) [<xref ref-type="bibr" rid="B11">11</xref>]. The AFPC operates at an air-intake rate of 5000 l/min, and the output flow for PM is 10 l/min. The size distribution of the ambient aerosol after the air passed the PM2.5 size selective inlet was determined during the exposure period with a multi-orifice-impactor (MOI, MSP, Minneapolis, Minn., USA).</p><p>The PM mass concentrations were measured continuously at the inlet (ambient) and once an hour (for 5 min) at the outlet of a concentrator during the exposure with a nephelometer (DATARAM, MIE, Billerica, Mass., USA). The time-integrated mass concentrations were also measured at both the inlet and the outlet by means of collection on two 47-mm filters placed in parallel (polytetrafluoroethylene (PTFE) and Quartz), with sampling at a flow rate of 2 l/min. Ozone, carbon monoxide, sulphur dioxide, nitrogen oxides, and the particle number concentrations were recorded every minute in the ambient air behind the PM2.5 impactor. The size distribution of the concentrated aerosol in the range of 0.15 – 2.5 μm was determined once an hour with an aerodynamic particle sizer for particles greater than 0.5 μm (APS, TSI, St. Paul, Minn., USA).</p><p>The u+fCAP exposure atmospheres were generated by drawing ambient air through a modified single-stage set-up of a versatile aerosol concentration enrichment system (VACES) [<xref ref-type="bibr" rid="B12">12</xref>,<xref ref-type="bibr" rid="B13">13</xref>]. The design of the VACES is such that particles greater than 2.5 μm are not concentrated. The VACES operates at an air intake flow rate of 500 l/min, and the output flow for PM is 15 l/min. The temperature of the coolers was kept at a constant -4°C. The temperature of the humidifiers was kept between 25°C and 30°C, depending on the ambient air conditions. The size distribution of the concentrated aerosol in the range of 0.01μm -2.5 μm was determined once an hour with an aerodynamic particle sizer (particles >0.5 μm) (APS, TSI, St. Paul, Minn., USA) for the fCAP experiments.</p><p>A condensation particle counter (CPC, TSI, St. Paul, Minn., USA) was used to determine the particle number concentrations after the particles had passed through the concentrator. PM was collected on three 47-mm filters (2 × PTFE and 1 × Quartz) placed in parallel at flow rate of 10 l/min. Similar, PM was collected at a flow rate of 1 l/min after the air passed the concentrator. A carbon sampler tube was placed downstream of one of the PTFE filters at the outlet to collect the VOCs. Ozone, carbon monoxide, sulphur dioxide, nitrogen oxides, and the particle number concentrations were recorded every minute in the ambient air upstream of the humidifier impactor. The size distribution of the concentrated aerosol in the range of 0.02 – 2.5 μm was determined once an hour with an aerodynamic particle sizer (for particles >0.5 μm) (APS, TSI, St. Paul, Minn., USA).</p><p>Temperature and relative humidity was recorded once every 5 minutes in the exposure chambers and control exposure chambers, as well as in the ambient air during the exposures. A Sartorius MC-5 microbalance (Sartorius, Goettingen, Germany) was used in controlled relative humidity (40 – 45%) and temperature (22 – 24°C) conditions to do the mass measurements, and the PTFE filters were weighed before and after each field test. Laboratory and field blanks were used for quality assurance. We then analysed the PTFE filters by means of ion chromatography to determine the concentrations of particulate sulphate, nitrate, and ammonium ions. The Quartz filters and activated carbon samplers were stored for future use. At this stage, no efforts have been made for chemical characterization of the PM samples since hardly any effect due to CAP exposure was observed.</p></sec><sec><title>Exposure chamber</title><p>Rats were exposed to the test atmosphere in a nose-only exposure chamber placed inside an inhalation unit, which was lighted with tubular fluorescent lamps. The animals were placed in nose-only tubes (Novoplast Tube T, Münster, Muttenz, Switzerland), restrained, and attached to the exposure chamber. The animals could breathe a continuous supply of test atmosphere during exposure (about 8 l/min). Control groups were exposed to air drawn from a concentrator down stream to the size selective inlet, and a HEPA filter filtered the air to provide a particle-free exposure atmosphere. To minimize stress, the animals were allowed to become accustomed to the tubes for 3 days in advance of the exposure, 1 h each day without exposure to the test atmosphere.</p></sec><sec><title>Necropsy</title><p>Eighteen hours after exposure, the rats were anaesthetized with Ketamine/Rompun (0.1 ml/100 g body weight of a mix of 0.85 ml 100 mg/ml Ketamine (Aesculaap, Boxtel, The Netherlands) and 0.65 ml of 20 mg/ml Rompun (Bayer, Leverkusen, Germany) and sacrificed by exsanguination via the abdominal aorta. A cannula was inserted in the trachea, and bronchoalveolar lavage fluid (BALF) was taken from the right lung after ligation of the left bronchus. The right lungs were lavaged (three times up and down) with a volume of saline corresponding with 27 ml/kg body weight at 37°C. The fluid recovered from the lavage was placed on ice. The left lung was dissected, weighed, and fixed for 1 h under a constant pressure of 20 cm H<sub>2</sub>O with 10% phosphate-buffered formalin. Five μm paraplast lung sections were stained with haematoxylin and eosin (H&E) and examined under a light microscope.</p></sec><sec><title>Morphometry</title><p>To measure cumulative cell proliferation, the animals were injected prior to CAPs exposure and 2 h prior to necropsy with bromodeoxyuridine (BrdU) measured to100 mg/kg body weight (Sigma-Aldrich, Zwijndrecht, The Netherlands). Lung sections from these animals were immunohistochemically stained with anti-BrdU antibody (Boehringer, Mannheim, Germany) and labelled with peroxidase. Since only the inflammatory foci displayed a focally increased proliferation rate, only a semi quantitative labelling score for the size of areas with an increased labelling was assigned. Labelling the frequency in square millimetres of bronchiolar epithelium, for example, makes no sense, as this procedure is fully dependant on how much epithelium is measured both inside and outside an inflammatory focus.</p><p>The analySIS soft imaging system (SIS, Münster, Germany) was used to quantify the BrdU-stained cells per millimetre of terminal bronchiolar epithelium. Terminal bronchioles were defined as those bronchioles flowing into alveolar ducts, as well as bronchioles smaller than 250 μm in diameter that are situated in the periphery of the lung and in the close vicinity of a centriacinar area. The total length of examined terminal bronchioles per animal varied between 15 mm and 25 mm.</p></sec><sec><title>Bronchoalveolar lavage analyses</title><p>The BALF collected from each animal was centrifuged at 400 <italic>g </italic>and 4°C for 10 min. The cell-free fluid from the lavage was used for biochemical assays. We used a commercial reagent kit (Roche Nederland, Mijdrecht, The Netherlands) to determine the activities of lactate dehydrogenase (LDH), N-acetyl glucosaminidase (NAG), alkaline phosphatase (ALP). The levels of uric acid (UA-B) were determined using a reagent kit obtained from Roche (Almere, the Netherlands) and malondialdehyde (MDA) was determined using a HPLC kit obtained from Chromsystems (Munich, Germany). We determined the total protein levels with a reagent kit obtained from Pierce (Oud-Beijerland, The Netherlands). Methods for the determination of glutathione, both its reduced (GSH) and oxidized (GSSG) forms, and Clara cell secretory protein (CC16) have been described previously [<xref ref-type="bibr" rid="B27">27</xref>].</p><p>We determined the total cell number by mixing 0.5 ml of the cell suspension with 9.5 ml of Isoton II (Beckman Coulter, Mijdrecht, The Netherlands) and then counting them in a Coulter Counter Z1 and Z2 (Beckman Coulter, Mijdrecht, The Netherlands). For differential cell counts, cytospin preparations were made and stained with the May-Grünwald and Giemsa method. Each cytospin preparation counted 400 cells, and the proportion of each cell type (macrophages, neutrophilic granulocytes, eosinophilic granulocytes, and lymphocytes) was calculated on the basis of total cells per BALF sample.</p></sec><sec><title>Blood analysis</title><p>Fibrinogen was determined as a risk factor for thrombotic vascular disorder and von Willibrand factor (vWf) as a marker for early endothelial injury, both as previously described [<xref ref-type="bibr" rid="B27">27</xref>].</p><p>Cell differentials were determined in ethylenediaminetetraacetic acid (EDTA; Terumo Europe, Leuven, Belgium), and anticoagulated blood was analysed in an H1-E Multi Species Haematology Analyser (Bayer, Mijdrecht, The Netherlands). The following parameters were measured: white blood cell (WBC) and red blood cell (RBC) concentrations, haemoglobin (HGB) and platelet concentrations (PLT), the mean platelet volume (MPV), and the haematocrit value (HCT). The mean corpuscular volume (MCV), mean platelet component (MPC), mean cell haemoglobin (MCH), mean cell haemoglobin concentration (MCHC), red blood cell distribution width (RDW) and haemoglobin distribution width (HDW) are also provided.</p></sec><sec><title>Statistical analysis</title><p>All effect parameters were log-transformed before two-way analysis of variance (ANOVA) was performed. Log-transformation is used to account for the increased variation in groups of animals exposed to CAPs versus the animals that were sham exposed. Two-way ANOVA techniques (simple factorial) were used to assess differences due to the factors "CAP exposure", "day-to-day variation" and their interaction while treating "CAP exposure" as a binary term. For those biological parameters that showed a significant effect of "CAP exposure" factor between control animals and CAP-exposed animals, the binary exposure factor was replaced by particle mass CAP as a continuous grouping factor, after which another two-way ANOVA and univariate regression analysis were performed. S-Plus software was used for all statistical analyses. The criterion for significance was set at <italic>p </italic>< 0.05.</p></sec></sec><sec><title>Abbreviations</title><p>ANOVA – analysis of variance; ALP – alkaline phosphatase; BALF – bronchoalveolar lavage fluid; BALT – bronchoalveolar lymphoid tissue; BrdU – 5-bromo-2-deoxyuridine; b.w. – body weight; CI, confidence interval; CC16 – Clara cell protein; ethylenediaminetetraacetic acid (EDTA; ELISA – enzyme-linked immunosorbent assay; ET-1 – endothelin-1; fCAP, fine concentrated ambient particulate matter;gsd – geometric standard deviation; GSH – reduced glutathione; GSSG – oxidized glutathione; HCT – haematocrit value; HDW – haemoglobin distribution width; HE – hematoxylin-eosin; HIA- Hendrik-Ido-Ambacht; LDH – lactate dehydrogenase; HGB – haemoglobin; HO-1, heme oxygenase; MDA, malondialdehyde; MIP-2 – macrophage inflammatory protein-2; MPC – mean platelet component; MPV – mean platelet volume; MCV – mean corpuscular volume; MCH – mean cell haemoglobin; MCHC – mean cell haemoglobin concentration; NAG – N-acetyl glucosaminidase; n.m., not measured; PM – particulate matter; PBS – phosphate buffered saline; PLT- platelet concentrations; PMN – polymorph nuclear neutrophil; PUF – polyurethane foam; RBC -red blood cell; RDW- red blood cell distribution width; SH – spontaneously hypertensive; TNF-α – tumour necrosis factor α; u+fCAP, ultrafine plus fine concentrated ambient particulate matter; UA – uric acid; vWF – von Willebrand factor; WBC- white blood cell.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>IMK has designed, coordinated and supervised the experimental work of this study, interpreted the results and drafted the manuscript. AJFB participated in the design and coordination of the study, carried out the in vivo experiments including sample handling, and participated in the statistical analysis. PHBF performed CAPs exposures and carried out the data analysis of the exposures. DLACL participated in the design, supported the in vivo experiments and collection of blood and tissue samples, carried out several BALF and blood analysis. JAMAD supported collection of lung tissue, performed histopathology and the statistical analysis for this part of the study. FRC is project leader, participated in conceiving the study, its design, interpretation of the results and is co-writer of the manuscript. All authors have read, reviewed, commented and approved the final manuscript.</p><fig position="float" id="F1"><label>Figure 1</label><caption><p>Schematic representation of the study design. Animals were exposed to concentrated ambient particulate matter on day 1 and 2. Necropsy took place 18 h after exposure. SHR, spontaneously hypertensive rat.</p></caption><graphic xlink:href="1743-8977-3-7-1"/></fig><fig position="float" id="F2"><label>Figure 2</label><caption><p>Relative heme oxygenase-1 present in bronchoalveolar lavage fluid (<inline-graphic xlink:href="1743-8977-3-7-i1.gif"/>) and in lung homogenate (<inline-graphic xlink:href="1743-8977-3-7-i2.gif"/>) versus the mass of fine, concentrated, ambient particulate matter. Relative heme oxygenase-1 is defined as ([HO-1]<sub>CAP</sub>- [HO-1]<sub>Control</sub>)/[HO-1]<sub>Control</sub>, where [HO-1] is the mean value of <italic>n </italic>= 8.</p></caption><graphic xlink:href="1743-8977-3-7-2"/></fig><fig position="float" id="F3"><label>Figure 3</label><caption><p>Correlation between heme oxygenase-1 present in bronchoalveolar lavage fluid versus heme oxygenase-1 present in lung homogenate. All animals were exposed to fine, concentrated, ambient particulate matter. Regression: Y = -0.15 + 0.217X, correlation coefficient = 0.79.</p></caption><graphic xlink:href="1743-8977-3-7-3"/></fig><fig position="float" id="F4"><label>Figure 4</label><caption><p>Clara cell secretory protein present in bronchoalveolar lavage fluid after rats were exposed to fine, concentrated, ambient particulate matter (fCAP) (site I) <inline-graphic xlink:href="1743-8977-3-7-i3.gif"/> control, ■ fCAP exposed.</p></caption><graphic xlink:href="1743-8977-3-7-4"/></fig><fig position="float" id="F5"><label>Figure 5</label><caption><p>Von Willebrand factor present in citrate plasma after rats were exposed to fine, concentrated, ambient particulate matter (fCAP) (site I) <inline-graphic xlink:href="1743-8977-3-7-i3.gif"/> control, ■ fCAP exposed.</p></caption><graphic xlink:href="1743-8977-3-7-5"/></fig></sec>
Inter-relationship between microsatellite instability, thymidylate synthase expression, and p53 status in colorectal cancer: implications for chemoresistance	<sec><title>Background</title><p>Studies indicate that thymidylate synthase (TS) expression, p53 and mismatch repair status have potential to influence colorectal cancer (CRC) outcome. There is, however, little data on the inter-relationship between these three markers. We sought to investigate whether relationships exist between these markers that might contribute to CRC phenotypes.</p></sec><sec sec-type="methods"><title>Methods</title><p>Four hundred and forty-one stage I-III CRCs were investigated. p53 status and TS expression were assessed by standard immunohistochemistry methods. Mismatch repair status was determined by assessment of microsatellite instability (MSI) using radiolabelled microsatellite genotyping.</p></sec><sec><title>Results</title><p>244 tumours (55%) over-expressed p53, and 259 (58%) expressed high TS levels. 65 tumours (15%) had MSI. A significant relationship between p53 over-expression and high TS expression was observed (p = 0.01). This was independent of MSI status. A highly significant inverse relationship between MSI and p53 status was observed (p = 0.001). No relationship was seen between MSI status and TS expression (p = 0.59).</p></sec><sec><title>Conclusion</title><p>Relationships exist between p53 status and TS expression, and MSI and p53 status. These inter-relationships may contribute to the clinical phenotype of CRCs associated with each of the molecular markers. High TS expression is unlikely to account for the clinical behaviour of CRCs with MSI.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Popat</surname><given-names>Sanjay</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Wort</surname><given-names>Richard</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Houlston</surname><given-names>Richard S</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Cancer	<sec><title>Background</title><p>Colorectal cancer (CRC) is one of the commonest malignancies of developed countries [<xref ref-type="bibr" rid="B1">1</xref>], with approximately 19,000 and 160,000 new cases in the UK and US, respectively, each year, and around 500,000 new cases world-wide [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B3">3</xref>]. CRC tumourigenesis is a multi-step phenomenon, typified by a series of genomic events that parallel development of invasive malignancy from normal epithelium through formation of pre-malignant adenomas [<xref ref-type="bibr" rid="B4">4</xref>]. Whilst most (~85%) CRCs develop through the chromosomal instability pathway, in which adenoma formation is typified by loss of APC function, and development of invasive malignancy by <italic>TP53 </italic>mutation [<xref ref-type="bibr" rid="B4">4</xref>], a smaller number (~15%) develop as a consequence of mismatch repair (MMR) deficiency. These tumours are characterised by high frequency microsatellite instability (MSI), proximal colonic distribution, poor differentiation, mucinous appearance, and lymphocytic infiltration [<xref ref-type="bibr" rid="B5">5</xref>]. Additionally, these tumours tend to retain the native diploid state [<xref ref-type="bibr" rid="B5">5</xref>]. By contrast, chromosomally unstable tumours tend to be aneuploid and have no site predilection [<xref ref-type="bibr" rid="B6">6</xref>].</p><p>Several studies have demonstrated that tumour molecular phenotype is a determinant of CRC prognosis [<xref ref-type="bibr" rid="B7">7</xref>-<xref ref-type="bibr" rid="B9">9</xref>]. Although many markers of prognosis have been investigated in recent years, the most promising to date are mismatch repair status, thymidylate synthase (TS) expression level, and p53 status.</p><p>Tumours developing through the mismatch repair pathway are characterised by loss of mismatch repair gene function (primarily <italic>hMLH1</italic>), through either mutation or more commonly, epigenetic change [<xref ref-type="bibr" rid="B5">5</xref>]. This results in somatic hypermutability most pronounced at short tandem repeat sequences such as microsatellites, termed microsatellite instability (MSI), readily detectable by the presence/absence of novel alleles or the shortening of at least 2–3 repeat units by electrophoresis. Several studies of clinical datasets have indicated that CRCs with MSI are associated with improved survival [<xref ref-type="bibr" rid="B10">10</xref>]. However, <italic>in vitro </italic>data has indicated that these tumours are paradoxically characterised by 5FU chemoresistance [<xref ref-type="bibr" rid="B11">11</xref>,<xref ref-type="bibr" rid="B12">12</xref>]. Although the mechanism for these observations remain unclear, the notion that fluoropyrimidine sensitivity might be modulated through either TS expression or p53 mutation is plausible [<xref ref-type="bibr" rid="B13">13</xref>]. Thymidylate synthase (TS) is a major protein involved in CRC development and outcome. As well as providing the sole intracellular source of thymidine for DNA synthesis [<xref ref-type="bibr" rid="B14">14</xref>], TS is also a target for a number of drugs used for CRC treatment, including 5-fluorouracil (5FU), whose mechanism of action is primarily mediated through competitive TS inhibition. Thymidylate synthase expression has been shown to be a key determinant of 5FU resistance <italic>in vitro </italic>[<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B15">15</xref>] and several patient series have confirmed poorer outcomes in those with tumours expressing high TS levels [<xref ref-type="bibr" rid="B16">16</xref>]. Mutation in the tumour suppressor <italic>TP53 </italic>has also been associated with chemoresistance and poorer survival in CRC [<xref ref-type="bibr" rid="B17">17</xref>,<xref ref-type="bibr" rid="B18">18</xref>]. <italic>TP53 </italic>maps to chromosome 17p13.1 [<xref ref-type="bibr" rid="B19">19</xref>] and is one of the commonest genes mutated in CRC, encoding a transcription factor (p53) critically involved in control of cell cycle, apoptosis, and carcinogenesis [<xref ref-type="bibr" rid="B20">20</xref>-<xref ref-type="bibr" rid="B25">25</xref>].</p><p>Whilst each of these markers impact on CRC prognosis, the inter-relationship between each has been the subject of few studies. Hence, the contribution to the clinical behaviour of CRCs associated with either p53, TS or mismatch repair status of any such inter-relationship is unclear. Here we report analysis of the largest series of early stage CRC assessed for these three clinically relevant molecular co-variates.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Patients</title><p>Four hundred and forty-one paired paraffin-embedded formalin fixed tissue blocks (one tumour tissue and one normal tissue) from patients with stage I-III CRC collected at the time of potentially curative surgery were assessed. The study was conducted in accordance with the tenets of the Declaration of Helsinki.</p><p>Fifteen-micron sections of formalin-fixed paraffin embedded tumours were cut onto double-sided clear adhesive tape on glass slides. Regions of ≥ 70% tumour were microdissected after light staining with toluidine blue. Ten-micron whole sections of normal mucosa from a separate paired block of normal colorectal mucosa from each patient were provided as a source of germline DNA extraction. DNA was extracted using standard commercially available methods (Qiagen, West Sussex, UK) and genotyping was performed at the highly sensitive and specific mononucleotide microsatellite locus BAT26 [<xref ref-type="bibr" rid="B26">26</xref>,<xref ref-type="bibr" rid="B27">27</xref>]. In cases where no paired normal tissue was available, tumour DNA was genotyped alone. Target DNA sequences were amplified using <sup>32</sup>P end-labelled primers. Mismatch repair status was assigned as MSI or microsatellite stable (MSS) by presence or absence of novel alleles or the shortening of at least 2–3 repeat units at autoradiography. Genotyping was performed at least twice per sample. Only tumours with unambiguous genotypes were assigned MSI status.</p></sec><sec><title>p53 and TS evaluation</title><p>Three micron sections from a representative part of the primary tumour were cut onto silane coated glass slides and assessed for TS and p53 expression by the avidin-biotin complex immunohistochemical technique (Vectastain Elite ABC kit, Vector Laboratories Inc, Burlingame, CA, USA) [<xref ref-type="bibr" rid="B28">28</xref>]. Negative and positive control slides were included in each staining run. Primary antibody was replaced with phosphate buffered saline, pH 7.6/0.1% (v/v) Tween solution for negative controls, and tumour sections known to stain heavily for target antigens were used for positive controls.</p><p>Tumour sections were deparaffinized in Histoclear (National Diagnostics, Hull, UK) and hydrated in decreasing concentrations of ethanol. Endogenous peroxidase activity was quenched with 5% hydrogen peroxide in methanol for 20 minutes. Antigen retrieval was required for p53 only and was by a microwave oven-based method. Specifically, sections were incubated in boiling 10 mmol/L citric acid buffer (pH 6.0) for 10 minutes and then cooled in running water. For both TS and p53, non-specific background staining was blocked with 20% goat serum for 20 minutes. Sections were then incubated with appropriate primary antibodies at a 1:100 dilution, in a humidified chamber at room temperature for 60 minutes, using either a validated rabbit polyclonal antibody to recombinant human TS [<xref ref-type="bibr" rid="B29">29</xref>], or an anti-p53 mouse monoclonal (clone DO-7, Dako Corp, Denmark). After rinsing, a biotinylated anti-rabbit secondary antibody was applied for 30 minutes followed by further washing, and avidin-biotin-peroxidase complexes (Vectastain Elite ABC kit; Vector Laboratories Inc, Burlingame, CA, USA). Immunostaining was developed by applying freshly prepared 0.05% 3, 3'-diaminobenzidine tetrahydrochloride (Vectastain Elite ABC kit, Vector laboratories). Sections were counterstained in Mayer's Haematoxylin (Sigma Chemical Co., St Louis, MO, USA), dehydrated in a series of ethanols, cleared in Histoclear (National Diagnostics, Hull, UK) and mounted with glass coverslips using DePeX (BDH, Poole, UK).</p></sec><sec><title>Immunohistochemistry evaluation</title><p>All slides were randomly allocated for independent assessment by two observers (RSH and SP), blinded to marker status. TS expression was assessed using the commonest reported method; a semiquantative grading of tumour tissue for chromagen intensity from 0 to 3, with the highest tumour staining detected as the reference for classification. Grades 0 and 1, representing none and minimal staining respectively, were defined as the "low" expression group, whereas grades 2 and 3 were defined as the "high" expression group. p53 immunoreactivity was dichotomised into positive or negative based on staining of malignant nuclei, with a threshold of 10%.</p><p>Level of scoring agreement between the two observers was recorded. In cases of disagreement, marker status was discussed and determined by consensus.</p></sec><sec><title>Statistical analysis</title><p>All statistical manipulations were preformed using STATA (Version 7, Stata Corp. TX77840, USA). Relationships between TS, p53 expression, and MSI status were assessed with Fisher's exact test, stratified by relevant marker.</p></sec></sec><sec><title>Results</title><sec><title>Tumour phenotypes</title><p>Genotype data was available from all 441 specimens. Sixty-five (15%) tumours demonstrated MSI, whereas 376 had MSS. Figure <xref ref-type="fig" rid="F1">1</xref> shows representative tumour genotypes, with two examples of MSI.</p><p>One quarter of the tumours stained for TS with grade 3 (101, 23%) or grade 0 (15, 3%) levels of chromagen intensity, whilst three-quarters stained with either grade 1 (167, 38%) or grade 2 (158, 36%) intensity. When dichotomised into high and low levels of TS expression, just over half of the samples expressed high TS levels (259, 59%), with the remainder having low-level expression (182, 41%). Just over half demonstrated p53 over-expression (244, 55%), with the remainder showing neither over-expression nor staining (197, 45%). Figures <xref ref-type="fig" rid="F2">2</xref> and <xref ref-type="fig" rid="F3">3</xref> show representative sections stained for TS and p53, respectively, demonstrating levels of chromagen intensity required to allocate expression status.</p></sec><sec><title>Inter-relationship between p53, TS, and mismatch repair status</title><p>A significant association between p53 status and TS chromagen intensity was observed (p = 0.01). This was maintained when dichotomising TS expression into high and low groups. Specifically, tumours with p53 over-expression were significantly associated with high TS levels (p = 0.005, Table <xref ref-type="table" rid="T1">1</xref>). When stratified by MSI status, a relationship between p53 status and TS level was again observed (p = 0.005), with tumours over-expressing p53 (p53 positive) tending to express high TS levels. This relationship was maintained in both subsets of tumours with MSS (p = 0.01, Table <xref ref-type="table" rid="T1">1</xref>), and MSI (p = 0.30, Table <xref ref-type="table" rid="T1">1</xref>). Although this relationship did not reach formal statistical significance in MSI tumours, this was likely due to the small numbers of tumours assessed, since an over-representation of p53 positive tumours with high TS levels was again observed.</p><p>As expected, a highly significant inverse relationship between mismatch repair status and p53 status was observed; MSI tumours tending to be p53 negative (p = 0.001, Table <xref ref-type="table" rid="T2">2</xref>). This was maintained and not influenced by TS status (Table <xref ref-type="table" rid="T2">2</xref>).</p><p>No relationship was observed between mismatch repair status and TS level (p = 0.59, Table <xref ref-type="table" rid="T3">3</xref>) and p53 status did not impact on these findings (Table <xref ref-type="table" rid="T3">3</xref>).</p><p>In order to exclude any bias that may have resulted from erroneous classification, we reassessed p53 status using thresholds of 5% and 15%. This resulted in no reclassification, and therefore no change in associations. Our results were further re-analysed using the Pearson χ<sup>2 </sup>test. All significant and non-significant associations were maintained. Our results have therefore been reported using the Fisher test, which gives the exact p value, rather than the asymptotic value calculated by the χ<sup>2 </sup>test.</p></sec></sec><sec><title>Discussion</title><p>We have investigated the relationship between MSI, TS and p53 status using standard genotyping and immunohistochemical methods in early stage CRC. Although only loosely correlating with <italic>TP53 </italic>mutation, p53 nuclear over-expression detected by IHC has been found to be a marker of worse prognosis in many previously published analyses of CRC datasets [<xref ref-type="bibr" rid="B21">21</xref>,<xref ref-type="bibr" rid="B30">30</xref>,<xref ref-type="bibr" rid="B31">31</xref>]. Our results indicate a highly significant association between p53 status and TS expression, with CRCs expressing high TS levels more likely to over-express p53, regardless of MSI status.</p><p>In normal cells, regulation of both TS and p53 are independently tightly controlled. In addition to it's role in enzyme catalysis, TS also functions as a RNA binding protein [<xref ref-type="bibr" rid="B32">32</xref>,<xref ref-type="bibr" rid="B33">33</xref>], regulating it's own expression by a negative autoregulatory mechanism [<xref ref-type="bibr" rid="B33">33</xref>,<xref ref-type="bibr" rid="B34">34</xref>], as well as binding to it's own RNA, to form TS-ribonucleoprotein complexes with several RNA species including <italic>c-myc </italic>and <italic>TP53 </italic>[<xref ref-type="bibr" rid="B35">35</xref>]. Although <italic>in vitro </italic>data indicates that p53 and TS have the ability to regulate each-other in non-malignant cells [<xref ref-type="bibr" rid="B32">32</xref>,<xref ref-type="bibr" rid="B33">33</xref>,<xref ref-type="bibr" rid="B36">36</xref>], evidence for a relationship in CRC has been conflicting, with some studies reporting that TS negatively regulates p53 expression [<xref ref-type="bibr" rid="B37">37</xref>], whilst others have shown no such relationship [<xref ref-type="bibr" rid="B38">38</xref>,<xref ref-type="bibr" rid="B39">39</xref>]. In addition, the potential for wild-type p53 to regulate TS expression has also been demonstrated using a luciferase-based system, which showed that p53 expression can inhibit <italic>TYMS </italic>promoter activity [<xref ref-type="bibr" rid="B40">40</xref>]. These results are, however, based on <italic>in vitro </italic>analysis, and no studies have characterised the role of mutant p53 or whether aberrant mismatch repair impacts on the relationship. Our results suggest a relationship between <italic>TP53 </italic>status and TS expression implying that the poor prognosis and chemoresistance observed in studies of CRC patients with either high TS expression or <italic>TP53 </italic>mutation/p53 over-expression, may have been impacted on by either co-variate.</p><p>A number of potential mechanisms may account for our findings. <italic>TP53 </italic>mutation is associated with loss of transcriptional activity control resulting in up- or down-regulation of downstream p53 effectors. Thus, inactive p53 might disinhibit an inhibitory role of p53 on TS expression. Alternatively, according to the gain-of-function paradigm [<xref ref-type="bibr" rid="B41">41</xref>], mutant p53 might acquire novel activities that promote cell growth and survival [<xref ref-type="bibr" rid="B42">42</xref>], perhaps through enhanced TS expression. An example of the latter case is the 273 Arg-His mutation, which has strong transactivating activity [<xref ref-type="bibr" rid="B43">43</xref>]. Although the role of this specific mutation in regulating TS expression is unknown, it is feasible that specific p53 mutations retain transcriptional regulatory activity, which may be partially responsible for control of transcriptional activity of proteins such as TS. Supporting this, Lenz et al. investigated the relationship between <italic>TP53 </italic>mutation and TS expression in a series of 36 CRCs and demonstrated that CRCs with p53 mutations affecting the zinc-binding domains had higher TS expression than those with mutation outside these domains [<xref ref-type="bibr" rid="B44">44</xref>]. Zinc domains are involved in direct DNA contact, protein stabilization, and structural activity, indicating that these mutations may have a severe impact on the transcriptional activity of p53.</p><p>Nine other studies have investigated the relationship between p53 and TS expression, based on immunohistochemistry, using p53 over-expression as a surrogate of <italic>TP53 </italic>mutation [<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B45">45</xref>-<xref ref-type="bibr" rid="B52">52</xref>]. However, most have been based on small sample sizes (median 66, range 22[<xref ref-type="bibr" rid="B51">51</xref>]–691[<xref ref-type="bibr" rid="B48">48</xref>]). Our results are consistent with four [<xref ref-type="bibr" rid="B44">44</xref>-<xref ref-type="bibr" rid="B46">46</xref>,<xref ref-type="bibr" rid="B50">50</xref>] of these studies, which also demonstrated a relationship between p53 and TS expression. Moreover, our study is the only one to assess this relationship stratified by MSI status. The relationship between p53 and TS was observed both in tumours with MSI and those with MSS, implying that aberrant mismatch repair is unlikely to impact on any mechanism relating p53 to TS expression. However, given the small number of samples with MSI, this cannot be entirely excluded.</p><p>As expected a highly significant inverse relationship between p53 status and MSI was observed. This relationship was independent of TS status, observed in both tumours expressing low and high TS levels (p ≤ 0.03), and is consistent with the concept that most CRCs develop either along the chromosomal instability pathway associated with <italic>TP53 </italic>mutation and MSS tumours or the aberrant mismatch repair pathway associated with wild-type p53 and MSI tumours [<xref ref-type="bibr" rid="B53">53</xref>-<xref ref-type="bibr" rid="B56">56</xref>]. This association may also explain why tumours with MSI seem to have an improved prognosis compared to those with intact mismatch repair. However, we demonstrated no relationship between MSI and TS status. This finding was independent of p53 status. Whilst the precise mechanisms by which cells with MSI seem resistant to 5FU <italic>in vitro </italic>has been poorly defined, Ricciardiello et al. [<xref ref-type="bibr" rid="B57">57</xref>] have suggested that this may, in part, be due to TS over-expression TS. In their analysis of 192 CRCs the authors demonstrated an association between CRCs with MSI and high TS expression [<xref ref-type="bibr" rid="B57">57</xref>]. This observation was, however, contrary to an earlier study based on an analysis of 53 CRCs, which also observed no relationship between TS expression and the MSI phenotype [58]. Moreover, the study by Ricciardiello et al. [<xref ref-type="bibr" rid="B57">57</xref>] was based on only 24 CRCs with MSI and the rate of high TS expression in their study was at the lower end of that previously reported (21%) [<xref ref-type="bibr" rid="B16">16</xref>] and may have biased findings. Our data, based on a sample size over two times larger, precludes at least a 16% difference between MMR status and high TS expression at the 5% threshold, with 90% power.</p><p>Our results provide little evidence that TS expression plays a major role in defining chemoresistance in microsatellite unstable CRCs, but gives support to previous reports of an inverse relationship between MSI and p53 status. In addition, we demonstrated that TS expression is related to p53 status, and this relationship may in part account for the poorer prognosis and relative chemoresistance seen in these tumours.</p></sec><sec><title>Conclusion</title><p>Relationships exist between TS expression and p53 status, and MSI and p53 status in CRC that may account for the clinical phenotypes of these tumours. High TS expression is unlikely to play a major role in the clinical behaviour of CRCs characterised by MMR deficiency.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>RW and SP carried out the molecular and immunohistochemical studies. SP and RSH reviewed results and assigned tumour categories. RW participated in study design. SP and RSH conceived the study, performed the statistical analysis, and drafted the manuscript. All authors read and approved the final manuscript.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1471-2407/6/150/prepub"/></p></sec>
Challenging the concept of microinvasive carcinoma of the vulva: report of a case with regional lymph node recurrence and review of the literature	<sec><title>Background</title><p>It is widely accepted that vulvar carcinoma with a depth of invasion of less than one millimeter is sufficiently treated by vulvectomy or wide local excision without inguinal lymphadenectomy.</p></sec><sec><title>Case presentation</title><p>However, a patient with inguinal lymph node recurrence 21 months after radical vulvectomy for stage IA squamous cell carcinoma was observed.</p></sec><sec><title>Conclusion</title><p>According to a review of the literature, there are five additional cases of metastasizing vulvar cancer with a depth of invasion of less than one millimeter. Therefore, the definition of microinvasive carcinoma of the vulva based on depth of invasion alone may not be as reliable as previously thought and does not rule out inguinal lymph node involvement or recurrence. Consequently, the necessity of inguinal node dissection for microinvasive carcinoma needs to be discussed on an individual basis taking into account the age of the patient as well as the potential morbidity of extended surgery.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Sidor</surname><given-names>Jutta</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Diallo-Danebrock</surname><given-names>Raihana</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Eltze</surname><given-names>Elke</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Lellé</surname><given-names>Ralph J</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib>	BMC Cancer	<sec><title>Background</title><p>Invasive vulvar carcinoma of all stages is usually treated by radical vulvectomy and bilateral lymphadenectomy. However, such an extensive surgical procedure has a high morbidity. In younger women, severe psychologic stress is caused by this hugely deforming operation, as sexual function and body image are disturbed. Therefore, over the past twenty years, efforts have been made to individualize treatment and to define a subgroup of patients which may be treated by less radical procedures [<xref ref-type="bibr" rid="B1">1</xref>]. It is widely thought that tumors with a depth of invasion of less than one millimeter according to criteria established by Wilkinson et al. [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B3">3</xref>] are sufficiently treated by vulvectomy or wide local excision without lymphadenectomy. However, there are five reports of lymph node metastases in patients with so-called "<italic>microinvasive carcinoma of the vulva</italic>". The patient reported here also developed inguinal lymph node metastasis after local treatment for squamous cell carcinoma of the vulva with a depth of invasion of less than one millimeter.</p></sec><sec><title>Case presentation</title><p>An 80-year-old white female was referred for evaluation of a suspicious vulvar lesion in September of 1997. At that time, the patient was free of complaints. On physical exam, atrophy of the vulva was seen but without evidence of lichen sclerosus. Approximately five millimeters from the clitoris, on the inner aspect of the left labium majus, a 1.5 cm indurated exophytic lesion was noted bleeding on touch (Figure <xref ref-type="fig" rid="F1">1</xref>). There was no involvement of the urethra or vagina. No regional lymphadenopathy was noted at this time. Biopsy revealed keratinizing squamous cell carcinoma. The patient had a previous history of recurrent adenocarcinoma within the oral cavity but had been free of disease for ten years.</p><p>Anterior radical vulvectomy without inguinal lymph node dissection was performed with margins of at least 2 cm from the well defined lesion. Frozen section biopsy revealed a depth of invasion of less than 1 mm. Histopathologic evaluation of the entire specimen showed a well differentiated 1.7 × 1.5 cm superficially invasive keratinizing squamous cell carcinoma (Figure <xref ref-type="fig" rid="F2">2</xref>). The exophytic tumor had a thickness of 5 mm. The greatest depth of invasion was 0.08 mm using the criteria established by Wilkinson et al. [<xref ref-type="bibr" rid="B2">2</xref>]. No lymphvascular space involvement was seen and margins were tumor-free (at least 7 mm at the urethral margin as measured on the paraffin-embedded specimen).</p><p>The patient remained free of disease until twenty-one months later when she noticed a painful swelling of the left inguinal area (Figure <xref ref-type="fig" rid="F3">3</xref>). A five centimeter tumor was noted. A CT scan of the abdomen and pelvis did not show any enlarged pelvic or paraaortic lymph nodes. Subsequently, bilateral inguinal lymphadenectomy was performed. The left inguinal tumor was found to be a 4.3 cm metastasis of a well to moderately differentiated keratinizing squamous cell carcinoma consistent with the previous diagnosis of vulvar carcinoma (Figure <xref ref-type="fig" rid="F4">4</xref>). At that time, there was no evidence of vulvar recurrence and radiation therapy was given to the groin and to the pelvis. The patient remained free of disease for 69 months. She then developed extensive vulvar recurrence and was treated with palliative radiation therapy. She died 8 months later.</p></sec><sec><title>Conclusion</title><p>So far, five cases of regional lymph node recurrences following treatment for FIGO stage Ia squamous cell carcinoma of the vulva have been reported in the literature [<xref ref-type="bibr" rid="B4">4</xref>-<xref ref-type="bibr" rid="B8">8</xref>]. One of two cases described by Thangavelu et al [<xref ref-type="bibr" rid="B5">5</xref>] cannot be considered true stage IA disease as the tumor was multifocal at initial presentation, the official FIGO classification stating that "Stage IA carcinoma of the vulva is defined as <underline>a single lesion </underline>measuring 2 cm or less in diameter with a depth of invasion of 1.0 mm or less ..." [<xref ref-type="bibr" rid="B9">9</xref>].</p><p>Age at initial diagnosis was between 39 and 84 years. Tumors were well differentiated (G1) in three patients [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B8">8</xref>] including the present case. None of the tumors showed lymphvascular space involvement. In four patients, vulvar intraepithelial neoplasia (VIN) was also present [<xref ref-type="bibr" rid="B4">4</xref>-<xref ref-type="bibr" rid="B6">6</xref>]. All patients with the exception of the case presented here had had a long-standing history of local complaints such as pruritus or suffered from vulvar diseases such as lichen sclerosus prior to the diagnosis of vulvar cancer.</p><p>The case reported by Volgger et al [<xref ref-type="bibr" rid="B4">4</xref>] is unusual as the patient had received immunosuppressive treatment for twelve years following a renal transplant. Long-standing Immunosuppression may have influenced the course of disease leading to groin node recurrence which occurred only four months after initial diagnosis, whereas, in the other five cases of microinvasive carcinoma, recurrence free survival was reported between 12 and 36 months.</p><p>All inguinal recurrences were unilateral. Five patients [<xref ref-type="bibr" rid="B4">4</xref>-<xref ref-type="bibr" rid="B6">6</xref>] including the present case died from progressive disease and/or distant metastases. In the other two cases [<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B8">8</xref>], no follow-up information was provided.</p><p>In summary, no pattern of particular risk factors can be deducted from this small number of cases. Theoretically, HPV status of the tumor tissue may play a role. However, Pinto et al [<xref ref-type="bibr" rid="B10">10</xref>] did not find an association between HPV DNA in the primary tumor and clinical outome in patients with vulvar carcinoma, whereas viral presence in the inguinal lymph node metastases was indeed associated with a worse prognosis.</p><p>As the salvage rate for patients with inguinal recurrence is low, the development of groin metastases might have been avoided in some of these patients if inguinal lymphadenectomy had been performed at the time of initial cancer diagnosis [<xref ref-type="bibr" rid="B11">11</xref>]. In retrospect, the patient discussed in this article had been brought into remission through secondary lymph node dissection and radiation therapy. At the time of recurrence she was already 82 years old. In a younger patient, radiation therapy with the concurrent administration of cytotoxic treatment using cisplatin should have been considered.</p><p>A more extensive surgical approach including pelvic lymphadenectomy should be reserved for the debulking of grossly positive nodes. In all other cases the routine removal of pelvic lymph nodes is obsolete [<xref ref-type="bibr" rid="B12">12</xref>].</p><p>In conclusion, the definition of microinvasive carcinoma of the vulva based on depth of invasion alone is not as reliable as previously thought for ruling out inguinal lymph node involvement or recurrence. Although the present standard of care cannot be revised on the basis of a very small number of cases, the necessity of inguinal surgery for microinvasive carcinoma needs to be discussed on an individual basis taking into account the age of the patient as well as the potential morbidity. In addition, immunocompromised patients may be at a significantly higher risk for inguinal spread of vulvar cancer.</p><p>Thus, unilateral inguinal lymph node dissection through a separate incision should not be ruled out entirely in patients with microinvasive vulvar cancer. Certainly, radical groin node dissection with removal of the fascia lata and resection of the saphenous vein is not justified. Instead, the surgical procedure should be tailored to the exact localization of inguinal lymph nodes as defined by several anatomical studies [<xref ref-type="bibr" rid="B13">13</xref>-<xref ref-type="bibr" rid="B15">15</xref>].</p><p>Nevertheless, lymph node involvement is rare if depth of invasion is less than one millimeter, and omission of inguinal lymph node dissection may be discussed individually with selected patients in order to avoid the considerable morbidity following inguinal surgery. In the future, these patients should undergo sentinel lymph node biopsy. Although not yet applied routinely in vulvar cancer patients, preliminary studies show its feasibility: Sliutz et al [<xref ref-type="bibr" rid="B16">16</xref>] have been able to identify one or more sentinel lymph nodes in all 26 patients studied. Furthermore, the survival of patients with early vulvar cancer treated with sentinel lymph node biopsy and radical local excision appears to be excellent [<xref ref-type="bibr" rid="B17">17</xref>].</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>JS performed the review of the literature and participated in the draft of the manuscript.</p><p>RD-D collected the specimen and performed the histopathological analysis.</p><p>EE participated in the histopathological analysis.</p><p>RJL took care of the patient and drafted the manuscript.</p><p>All authors have read and approved the final manuscript.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1471-2407/6/157/prepub"/></p></sec>
Analysis of folylpoly-γ-glutamate synthetase gene expression in human B-precursor ALL and T-lineage ALL cells	<sec><title>Background</title><p>Expression of folylpoly-γ-glutamate synthetase (FPGS) gene is two- to three-fold higher in B-precursor ALL (Bp- ALL) than in T-lineage ALL (T-ALL) and correlates with intracellular accumulation of methotrexate (MTX) polyglutamates and lymphoblast sensitivity to MTX. In this report, we investigated the molecular regulatory mechanisms directing FPGS gene expression in Bp-ALL and T-ALL cells.</p></sec><sec sec-type="methods"><title>Methods</title><p>To determine FPGS transcription rate in Bp-ALL and T-ALL we used nuclear run-on assays. 5'-RACE was used to uncover potential regulatory regions involved in the lineage differences. We developed a luciferase reporter gene assay to investigate FPGS promoter/enhancer activity. To further characterize the FPGS proximal promoter, we determined the role of the putative transcription binding sites NFY and E-box on FPGS expression using luciferase reporter gene assays with substitution mutants and EMSA.</p></sec><sec><title>Results</title><p>FPGS transcription initiation rate was 1.6-fold higher in NALM6 <italic>vs</italic>. CCRF-CEM cells indicating that differences in transcription rate led to the observed lineage differences in FPGS expression between Bp-ALL and T-ALL blasts. Two major transcripts encoding the mitochondrial/cytosolic and cytosolic isoforms were detected in Bp-ALL (NALM6 and REH) whereas in T-ALL (CCRF-CEM) cells only the mitochondrial/cytosolic transcript was detected. In all DNA fragments examined for promoter/enhancer activity, we measured significantly lower luciferase activity in NALM6 <italic>vs</italic>. CCRF-CEM cells, suggesting the need for additional yet unidentified regulatory elements in Bp-ALL. Finally, we determined that the putative transcription factor binding site NFY, but not E-box, plays a role in FPGS transcription in both Bp- and T-lineage.</p></sec><sec><title>Conclusion</title><p>We demonstrated that the minimal FPGS promoter region previously described in CCRF-CEM is not sufficient to effectively drive FPGS transcription in NALM6 cells, suggesting that different regulatory elements are required for FPGS gene expression in Bp-cells. Our data indicate that the control of FPGS expression in human hematopoietic cells is complex and involves lineage-specific differences in regulatory elements, transcription initiation rates, and mRNA processing. Understanding the lineage-specific mechanisms of FPGS expression should lead to improved therapeutic strategies aimed at overcoming MTX resistance or inducing apoptosis in leukemic cells.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Leclerc</surname><given-names>Guy J</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Leclerc</surname><given-names>Gilles M</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Kinser</surname><given-names>Ting Ting Hsieh</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" corresp="yes" contrib-type="author"><name><surname>Barredo</surname><given-names>Julio C</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Cancer	<sec><title>Background</title><p>Folate antimetabolites play a central role as anticancer agents. In mammalian tissues, intracellular folates and antifolates exist as poly-γ-glutamates with typical chains ranging from five to nine residues [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B3">3</xref>]. Polyglutamation is catalyzed by folylpoly-γ-glutamate synthetase (FPGS) and results in increased intracellular concentration and cytoxicity of classical antifolates [<xref ref-type="bibr" rid="B4">4</xref>]. Furthermore, when polyglutamated, some antifolates (e.g., raltitrexed, lometrexol) increase their <italic>Ki </italic>against targeted enzymes by over 100-fold [<xref ref-type="bibr" rid="B5">5</xref>,<xref ref-type="bibr" rid="B6">6</xref>]. In childhood acute lymphoblastic leukemia (ALL) a strong correlation exists between FPGS expression, intracellular methotrexate (MTX) polyglutamate accumulation and treatment outcome [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B7">7</xref>].</p><p>The FPGS gene is controlled by at least two mechanisms: one tissue/lineage-specific and a second proliferation-dependent [<xref ref-type="bibr" rid="B8">8</xref>-<xref ref-type="bibr" rid="B11">11</xref>]. FPGS activity is distributed to both cytosolic and mitochondrial compartments of mammalian cells. In humans, these two isoforms are encoded by a single locus in chromosome region 9q (34.1–34.2) [<xref ref-type="bibr" rid="B12">12</xref>], and differ by the use of two alternative translational start sites within exon 1 [<xref ref-type="bibr" rid="B13">13</xref>]. Use of these alternative start sites translates the FPGS protein with or without the addition of a mitochondrial leader sequence. Alternative FPGS exon 1 variants (exons 1, 1A, 1B, 1C, and 2A), all spliced to exon 2, have been described [<xref ref-type="bibr" rid="B12">12</xref>,<xref ref-type="bibr" rid="B14">14</xref>]. We have demonstrated no lineage-specific differences in the expression of these alternative transcripts in human leukemia and normal tissues [<xref ref-type="bibr" rid="B15">15</xref>].</p><p>In mice, two promoters spaced by 10 kb were shown to express distinct functional tissue-specific FPGS isoenzymes [<xref ref-type="bibr" rid="B16">16</xref>]. The upstream transcript (exon A1a) was expressed only in few differentiated tissues such as liver, whereas the downstream transcript (exon 1) was expressed in dividing normal and neoplastic tissues. This dual promoter mechanism directing expression of murine isoenzymes is not conserved in humans. In contrast, the human FPGS exon 1 transcript is present in both dividing and differentiated tissues [<xref ref-type="bibr" rid="B14">14</xref>]. In human leukemia cells, the enzyme translated from exon 1 transcript was reported as the only catalytically active form. Transcription of the human FPGS gene appears to be controlled by a TATA-less promoter driven by a set of 8 concatameric Sp1 sites spaced within a 150 bp region upstream of exon 1 [<xref ref-type="bibr" rid="B17">17</xref>]. Several additional transcription factors, including NFY (Y-box) and E-box motifs have been identified within the human minimal FPGS promoter region [<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B17">17</xref>].</p><p>Our laboratory first demonstrated that constitutive levels of FPGS mRNA, protein, and enzyme activity are two- to three-fold higher in B-precursor (Bp) ALL cells compared to T-lineage ALL [<xref ref-type="bibr" rid="B9">9</xref>,<xref ref-type="bibr" rid="B11">11</xref>]. We now report studies investigating the molecular mechanisms for this differential FPGS gene expression. Our results demonstrate that FPGS transcriptional start sites (+1) are the same in all hematopoietic lineages studied. To analyze lineage differences in FPGS promoter activity we used a FPGS promoter-luciferase gene reporter assay. All DNA fragments examined for promoter/enhancer activity exhibited higher levels of luciferase activity in CCRF-CEM <italic>vs</italic>. NALM6 cells. Finally, we determined the role of the putative NFY and E-box transcription factor binding sites on FPGS gene transcription using the same reporter gene assay with substitution mutants and EMSA analysis.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Leukemia cell lines</title><p>The human leukemia cell lines CCRF-CEM (T-ALL) and REH (Bp-ALL t(12;21)) were obtained from the American Type Culture Collection. NALM6 (Bp-ALL) cell line was obtained from DSMZ (Germany). RCH-ACV (Bp- ALL t(1;19)) was kindly provided by Dr. Stephen Hunger (UFL, Gainesville, FL). All cell lines were grown in RPMI 1640 (Sigma) supplemented with 10% FBS at 37°C and 5% CO<sub>2</sub>. Normal bone marrow (BM) was extracted from normal volunteers. We obtained institutional review board approval, and IRB approved informed consent was obtained from normal volunteers prior to participation.</p></sec><sec><title>RNA isolation and real-time RT-PCR</title><p>Total RNA was isolated using the RNeasy kit (Qiagen Inc). FPGS exons 14 and 15 were RT-PCR amplified [<xref ref-type="bibr" rid="B15">15</xref>] and quantitated using pFPGS-cDNA, a pCR2.1-TOPO plasmid containing the human FPGS cDNA gene (1926 bp; Table <xref ref-type="table" rid="T1">1</xref>). Briefly, the 1926 pb FPGS cDNA gene was amplified by PCR using primers 1exonF and 15exonR-c1934 (Table <xref ref-type="table" rid="T1">1</xref>) from total RNA reverse transcribed with primer 15exonR-c2107 (5'-GGCCAGGCAGCGCACACAAT). Results were normalized to β-actin mRNA expression. All real-time PCR reactions (SYBR green) were performed using the BIO-RAD iCycler iQ system (Bio-Rad) [<xref ref-type="bibr" rid="B18">18</xref>].</p></sec><sec><title>Nuclear run-on assay</title><p>Nuclear run-on assay was performed as described by Hanson et al. [<xref ref-type="bibr" rid="B19">19</xref>]. One μg of FPGS and 18S RNA cDNAs were blotted, UV cross-linked and hybridized with nascent labeled RNA transcripts. After washing, membranes were exposed to XAR-5 film (Kodak). Densitometric scan analysis was performed using the GelPro Analyzer program. Results were normalized to the level of 18S RNA and statistic achieved using one-tailed, Student paired <italic>t</italic>-test (GraphPad Prism, version 2.01). All data are expressed as mean ± S.E.M.</p></sec><sec><title>5'-Rapid amplification of cDNA ends (RACE)</title><p>Amplification of the 5'-termini of FPGS mRNA was performed using the 5'-RACE system Version 2.0 (Invitrogen Corporation). Briefly, polyA+ RNA isolated from cells using the Oligotex Direct mRNA mini kit (Qiagen) were reverse transcribed using the oligonucleotide Exon5R-F8 (5'-CTTGGTGAAGAGCTCAGGACTG), a gene-specific primer in exon 5. The poly dC-tailed cDNA products were amplified using a nested primer in exon 2, Exon2R-F4 (5'- ACTCCGTGCCAGGTACAGTTCCATG), and the abridged anchor primer. Primary PCR products were re-amplified using an exon 2 upstream nested primer, 2exonR-2381 (5'-CAGGTAGCCGGCATTGGTCTG), and the abridged universal amplification primer. 5'-RACE products were separated on a 2.5% agarose gel, purified and cloned into the pCR2.1-TOPO vector (Invitrogen). Identity of the 5'-RACE products was determined by nucleotide sequence.</p></sec><sec><title>Construction of FPGS-luciferase reporter gene fusions</title><p>Regions of the FPGS gene promoter were generated by PCR, cloned into pCR2.1-TOPO vector and sub-cloned into pGL3-basic vector (Promega) or pGL2628-ATGm/c. Plasmids and primers used in this study are listed in Table <xref ref-type="table" rid="T1">1</xref>. The BAC clone RCPI-11 465E22 contains 188,098 bp of human chromosome 9 including 104,600 bp upstream of FPGS exon 1 (obtained from Dr. P.J. de Jong, Children's Hospital Oakland). PCR conditions were optimized for each of these fragments.</p></sec><sec><title>Nucleofection and luciferase reporter gene assays</title><p>The CCRF-CEM, NALM6 and REH cell lines were transfected by nucleofection (Amaxa Biosystems) following manufacturer's protocol. Briefly, 5 × 10<sup>6 </sup>cells were resuspended in 100 μl of solution V<sup>® </sup>(CCRF-CEM and NALM6) or R<sup>® </sup>(REH) and mixed with 2.5 μg of plasmid pGL1374 (FPGS promoter::luc) or equimolar concentration of other FPGS promoter::luc plasmids, and 3 μg of pCMVβ. Cells were incubated at 37°C/5% CO<sub>2 </sub>for 24 hrs, harvested, washed twice with cold PBS 1X and resuspended in dual-light lysis buffer to yield cellular extracts. Luciferase and β-galactosidase activities were assayed using the dual-light reporter gene assay system (Tropix, Inc.). Transfection efficiencies and cell viability were monitored by flow cytometry. Cell viability was 70–75% and 78–85% for CCRF-CEM and NALM6 cells, respectively. Statistical tests were achieved using one-tailed, Student paired <italic>t</italic>-test (GraphPad Prism, version 2.01).</p></sec><sec><title>Methylation-specific PCR analysis</title><p>Genomic DNA was isolated from CCRF-CEM and NALM6 cells using the DNeasy kit (Qiagen, Inc.) and treated with sodium bisulfite to convert unmethylated cytosine to uracil residues while 5-methylcytosines remain unaltered (CpGenome DNA modification kit [Chemicon International]). Identification of CpG islands and design primer sets were determined using the MethPrimer program [<xref ref-type="bibr" rid="B33">33</xref>]. Chemically converted genomic DNA was PCR amplified using specific primer sets for methylated and unmethylated forms (Table <xref ref-type="table" rid="T1">1</xref>). Methyl-specific PCR (MSP) products were resolved on a 3% agarose gel.</p></sec><sec><title>Site-directed mutagenesis</title><p>The QuikChange II XL Site-Directed Mutagenesis Kit (Stratagene) was used to generate substitution mutants of the human FPGS mitochondrial and cytosolic initiation codons (ATG) from the pGL2628 plasmid using oligonucleotides QC19934F/QC19934R and QC20060F/QC20060R, respectively (Table <xref ref-type="table" rid="T1">1</xref>). Similarly, FPGS putative NFY868 and E-box952 binding sites were mutagenized using the plasmid pGL2628-ATGm/c plasmid DNA template. Primers NFY-868F/NFY-868R and E47-952F/E47-952R were used for the NFY and E-box mutagenesis, respectively (Table <xref ref-type="table" rid="T1">1</xref>). Mutations were confirmed by nucleotide sequencing.</p></sec><sec><title>Nuclear extracts and electrophoretic mobility shift assay (EMSA)</title><p>Nuclear extracts were prepared from CCRF-CEM and NALM6 cells using the NE-PER Nuclear and Cytoplasmic Extraction Reagents (Pierce, Biotechnology, Inc.). DNA-protein interactions were carried out and detected using the LightShift Chemiluminescent EMSA kit (Pierce, Biotechnology, Inc.). Each EMSA reactions contained 25 nM labeled FPGS -32/-14 (5'-CTGCGCTGATTGGCTGGGG) oligonucleotides, 50 ng Poly (dI-dC) and 10 μg of nuclear protein. When required, competitive unlabeled NFY DNA oligomer, unlabeled Epstein-Barr nuclear antigen (EBNA) DNA, and NFY antibody (CBF-A (C20) or CBF-B (H209) (Santa Cruz, Biotechnology, Inc.)) were added to EMSA. The DNA-protein complexes were resolved on non-denaturing 5% TBE polyacrylamide gels (Bio-Rad).</p></sec></sec><sec><title>Results</title><sec><title>FPGS transcription rate and mRNA transcription start sites in Bp- and T-ALL</title><p>To investigate whether differences in FPGS mRNA expression in Bp-ALL and T-ALL resulted from differences in FPGS promoter transcription rate, we determined the frequency of transcription initiation in CCRF-CEM and NALM6 cells using nuclear run-on assays. As shown in Figure <xref ref-type="fig" rid="F1">1A</xref>, ratio of FPGS/18S mRNA transcription rate was 1.64-fold higher in NALM6 (2.26 ± 0.768) <italic>vs</italic>. CCRF-CEM (1.37 ± 0.416) cells (<italic>p </italic>< 0.05). These results are consistent with the observed two- to three-fold higher levels of FPGS mRNA, protein expression and activity in NALM6 <italic>vs</italic>. CCRF-CEM [<xref ref-type="bibr" rid="B9">9</xref>,<xref ref-type="bibr" rid="B11">11</xref>,<xref ref-type="bibr" rid="B15">15</xref>].</p><p>To uncover potential regulatory regions involved in the lineage differences in expression of the FPGS gene, we localized the FPGS promoter in Bp-ALL <italic>vs</italic>. T-ALL cell lines by mapping FPGS transcription initiation sites using 5'-RACE. These experiments detected a long (~280 bp) and a short (~180 bp) fragment, which were individually characterized by nucleotide sequence analysis to encode mitochondrial/cytosolic and cytosolic FPGS, respectively. Long fragments were detected in CCRF-CEM (T-ALL) whereas both short and long fragments were amplified from NALM6 and REH (Bp-ALL) (Figure <xref ref-type="fig" rid="F1">1B</xref>). Therefore, both Bp-ALL and T-ALL cells use the same promoter to transcribe FPGS mRNA but lineage differences in mRNA transcripts exist.</p></sec><sec><title>FPGS gene promoter activity in Bp- and T-ALL lineages</title><p>The human FPGS minimal promoter has been characterized in CCRF-CEM cells and encompassed a region starting -43 bp from the main transcriptional start site to +150 bp of exon 1 [<xref ref-type="bibr" rid="B17">17</xref>] (see Figure <xref ref-type="fig" rid="F2">2</xref>). To further investigate the mechanisms that control FPGS transcription in human hematopoietic cells, we analyzed and compared DNA fragments located upstream of exon A1b and in the 5'-flanking region of exon 1 (encompassing the previously described minimal promoter region) for promoter/enhancer transcriptional activity in CCRF-CEM and NALM6 cells. FPGS-luciferase transcriptional gene fusions were constructed and assayed as described in Material and Methods. Under our experimental conditions, transfection efficiencies were 38–50% and 42–50% in CCRF-CEM and NALM6 cells, respectively. As shown in Figure <xref ref-type="fig" rid="F3">3</xref>, DNA fragments located upstream of exon A1b (pGL2256 and pGL3588-2628) yielded no promoter or enhancer activity in both cell lines suggesting that the 5'-flanking region of exon A1b exerts no regulatory activity on human FPGS expression. The higher level of luciferase activity detected in pGL3588-2628 <italic>vs</italic>. pGL2256 is likely the result of the presence of the additional fragment of 2628 bp contained in pGL3588-2628 that included the FPGS minimal promoter region. When DNA fragments from the 5'-flanking region of exon 1 (pGL1374, pGL2628-ATGm/c, pGL4689) encompassing the described minimal promoter were analyzed, we found 4- to 12-fold higher level of luciferase/β-galactosidase activity in CCRF-CEM compared to NALM6 cells (Figure <xref ref-type="fig" rid="F3">3</xref>). To validate our reporter gene assay and rule out any technical differences in luciferase activity in NALM6 <italic>vs</italic>. CCRF-CEM cells, we assayed a CMV promoter-luciferase driven vector (pCMV-luc) and found no significant differences in both cell lines (data not shown). Therefore, these data indicate that DNA fragments containing the previously described minimal promoter region are not sufficient to effectively drive FPGS transcription in Bp-ALL (NALM6) compared to T-ALL (CCRF-CEM). Similar experiments with other Bp-ALL cell lines such as RCH-ACV (t(1:19)) and REH (t(12;21)) were performed and yielded same results (data not shown). To exclude that differences in nucleotide sequence were responsible for low promoter activity observed in Bp-ALL, we PCR amplified and sequenced the 1374 bp fragment containing the minimal promoter from normal bone marrow (BM), CCRF-CEM and NALM6 cells. No difference in nucleotide sequence was detected.</p><p>The inability of the minimal promoter region to efficiently drive FPGS transcription in NALM6 cells implied that additional unidentified regulatory regions may be required in Bp-ALL cells. To test this hypothesis we then investigated the presence of enhancers or transcription factors required for FPGS transcription in NALM6 <italic>vs</italic>. CCRF-CEM cells, by analyzing the effect in <italic>cis </italic>of DNA fragments located within the exons A1b-1 and in the 3'-UTR of the FPGS gene on the FPGS-luc expression of pGL2628-ATGm/c. As shown in Figure <xref ref-type="fig" rid="F3">3</xref>, plasmids pGL2299-2628, pGL2158-2628, pGL1163-2628, and pGL2881-2628 constructs yielded 1.8- to 4.2-fold lower level of luciferase activity in NALM6 <italic>vs</italic>. CCRF-CEM cells. Therefore, the regulatory elements required for FPGS gene expression in Bp-ALL appear not to be localized within a region encompassing 12 kb upstream of exon 1 nor within the 3'-untranslated region.</p></sec><sec><title>Methylation status of the FPGS promoter in Bp- and T-lineage</title><p>DNA methylation has been associated with transcriptional inactivation and gene silencing [<xref ref-type="bibr" rid="B20">20</xref>,<xref ref-type="bibr" rid="B21">21</xref>]. Nucleotide sequence analysis of the 1374 bp fragment (-791 to +582) containing the FPGS minimal promoter predicted one CpG island (54% GC content) at position -330 to +294. We determined whether DNA methylation could contribute to the lineage-specific differences in FPGS expression in ALL cell lines by methylation-specific PCR (MSP) analysis using bisulfite-treated DNA. Primers sets were designed to anneal to unmethylated DNA (U) and methylated templates (M). As shown in Figure <xref ref-type="fig" rid="F4">4A</xref>, amplification products (142 bp) were detected in both CCRF-CEM and NALM6 bisulfite-treated nuclear DNA with unmethylated primers (U) indicating a preferentially unmethylated CpG island in both cell lines. Control experiments with untreated DNA or absence of template yielded no products (data not shown). Quantitative analysis of cytosine methylation was determined in both cell lines by sequencing of the 142 bp unmethylated PCR amplicon. This analysis revealed identical number of unmethylated cytosines in both cell lines. Therefore, DNA methylation of the CpG island region which contains the described FPGS minimal promoter does not play a role in the observed lineage-specific differences in FPGS expression in ALL cell lines.</p></sec><sec><title>Role of putative NFY-box and E-box binding sites on FPGS expression</title><p>To identify specific regulatory elements involved in FPGS gene transcription, we analyzed the nucleotide sequence of the minimal FPGS gene promoter for presence of known transcription factor recognition motifs using the MatInspector program (Genomatix, release 7.3.1). As shown in Figure <xref ref-type="fig" rid="F2">2</xref>, two Sp1 (GGGCGG; -10, -15), one reverse Sp1 (+5), one inverted NFY-box (CCAAT; -20), and one E-box (CANNTG; +61) transcription factor binding sites were identified within the FPGS minimal promoter (minP). We examined the role of putative NFY and E-box DNA binding sites by generating mutants at each site with substitutions reported to reduce or abolish gene promoter activity [<xref ref-type="bibr" rid="B22">22</xref>,<xref ref-type="bibr" rid="B23">23</xref>]. Mutant constructs for NFY (NFY-868; CCAAT → C<bold>TTT</bold>T) and E-box (Ebox-952; CACCTG → CA<bold>TTC</bold>G) were co-transfected with pCMVβ in both CCRF-CEM and NALM6 cells and assayed for luciferase and β-galactosidase activities. As shown in Figure <xref ref-type="fig" rid="F4">4B</xref>, mutation in the NFY site (NFY-868) reduced the level of FPGS transcription by 45% in NALM6 (p < 0.001) and 40% in CCRF-CEM cells (p < 0.005) when compared to the wild type construct (pGL2628-ATGm/c). Normalized luciferase activities were 1.9 × 10<sup>-2 </sup>± 2.0 × 10<sup>-3 </sup>(pGL2628-ATGm/c) and 1.0 × 10<sup>-2 </sup>± 1.2 × 10<sup>-3 </sup>(pGL868NFY-ATGm/c) in NALM6, and 2.4 × 10<sup>-1 </sup>± 2.4 × 10<sup>-2 </sup>(pGL2628-ATGm/c) and 1.5 × 10<sup>-1 </sup>± 1.4 × 10<sup>-2 </sup>(pGL868NFY-ATGm/c) in CCRF-CEM cells. These data suggest that the NFY binding site is required to activate FPGS gene transcription in lymphoid cells. In contrast, no significant differences were observed in either cell line with the mutant Ebox-952 construct, suggesting a negligible role in FPGS expression (Figure <xref ref-type="fig" rid="F4">4B</xref>). In NALM6 and CCRF-CEM cells, the normalized level of luciferase activity was 1.6 × 10<sup>-2 </sup>± 8.4 × 10<sup>-4 </sup>(pGL952Ebox-ATGm/c) and 2.2 × 10<sup>-1 </sup>± 1.8 × 10<sup>-2 </sup>(pGL952Ebox-ATGm/c), respectively.</p><p>To establish NFY protein interaction with the NFY-box element (CCAAT) at position -20 upstream the FPGS gene promoter, we performed electrophoretic mobility shift assays (EMSAs) using an oligonucleotide probe containing FPGS gene sequence from -32 to -14. The sequences of double-stranded biotinylated oligonucleotide probes and competitors were incubated with nuclear extracts prepared from CCRF-CEM or NALM6 cells. The EMSA revealed the formation of two complexes with the NFY oligonucleotide probe (Figure <xref ref-type="fig" rid="F4">4C</xref>, C1 and C2). To further ascertain the specificity of binding, we demonstrated that the formation of complexes C1 and C2 were inhibited by excess amounts of unlabeled NFY oligonucleotides (Figure <xref ref-type="fig" rid="F4">4C</xref>, lanes 3 and 8) but not with unlabeled non-specific Epstein-Barr nuclear antigen (EBNA) oligonucleotides (Figure <xref ref-type="fig" rid="F4">4C</xref>, lane 6). To directly evaluate the proposed participation of NFY in the shifted complexes, anti-NFY peptide (CBF-A and CBF-B subunits) antibodies were incubated with EMSA binding reactions before electrophoresis. Addition of anti NFY-A (CBF-A) antibody to these reactions resulted in strongly retarded supershift complex (SSC) (Figure <xref ref-type="fig" rid="F4">4C</xref>, lanes 4 and 9). In contrast, addition of NFY-B (CBF-B) antibody abolished the formation of the complex C2 (Figure <xref ref-type="fig" rid="F4">4C</xref>, lanes 5 and 10) without formation of a supershifted band. These data clearly demonstrate that NFY transcription factor is present in the complex that binds to the putative NFY binding site (CCAAT) at position -20.</p></sec></sec><sec><title>Discussion</title><p>Folate antimetabolites such as MTX are essential chemotherapeutic drugs in the treatment of children with acute lymphoblastic leukemia (ALL). MTX is retained within the cell by cellular metabolism catalyzed by the enzyme FPGS. The human FPGS promoter has been previously characterized using CCRF-CEM cells and it encompasses a GC-rich region without a typical TATA sequence, usually a characteristic of housekeeping genes and proto-oncogenes. The minimal portion of the promoter required to drive transcription in CCRF-CEM cells consists of a region starting -43 to +150 of the main transcription start site including part of exon 1 [<xref ref-type="bibr" rid="B17">17</xref>]. When 5'-flanking sequences upstream of exon 1 were analyzed for promoter and/or enhancer activity in Bp- <italic>vs</italic>. T-cells, we found significantly lower luciferase activity in the Bp-ALL cell line NALM6 compared to the T-ALL cell line CCRF-CEM. These results are opposite of what one would predict based on the known lineage differences resulting in two- to three-fold higher FPGS expression Bp- <italic>vs</italic>. T-ALL [<xref ref-type="bibr" rid="B9">9</xref>,<xref ref-type="bibr" rid="B11">11</xref>]. These low levels of FPGS-luciferase activity observed with our constructs in NALM6 <italic>vs</italic>. CCRF-CEM cells, lead us to conclude that the minimal promoter region is not sufficient to effectively drive FPGS transcription in Bp-cells (NALM6). Consequently, we hypothesize that additional regulatory elements are required to drive FPGS expression in Bp-ALL. Our analysis failed to demonstrate promoter and/or enhancer activity in the immediate 5'-flanking region of exon A1b, between exons A1b and 1, and in the 3'-UTR. We propose that Bp-cell specific enhancer(s) are required for FPGS gene expression in Bp-ALL cells and that distant yet unidentified regulatory loci exist.</p><p>The human FPGS transcription start sites have been mapped previously in human CCRF-CEM and HepG2 (hepatoma) cells and shown that FPGS transcription was initiated from multiple start sites spread over 80 bp clustered in two major regions differing by the presence of the mitochondrial <italic>vs</italic>. cytosolic initiation codons [<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B17">17</xref>]. The first transcription initiation site generates a long transcript encoding both the mitochondrial and cytosolic isoforms of the enzyme, while the second start site generates a shorter transcript encoding only the cytosolic protein. Under our conditions, these two major transcripts were detected only in Bp-ALL (NALM6 and REH) whereas in T-ALL (CCRF-CEM) cells only the longer transcript was detected. NALM6 cells which exhibited higher level of FPGS mRNA expressed both transcripts, while in REH and RCH-ACV which expressed intermediate levels of FPGS mRNA the longer transcript (mitochondrial/cytosolic) was either faint or absent, and in CCRF-CEM cells expressing the lowest level of FPGS mRNA the shorter transcript (cytosolic) was absent. This finding is consistent with and underscores regulatory differences in the expression of the FPGS gene between Bp- and T-lineage ALL cells. It has also led us to hypothesize that the additional expression of the shorter transcript encoding for additional cytosolic protein in NALM6 cells could contribute to higher FPGS mRNA expression leading to higher cytosolic protein expression in these cells. Although speculative at this time, this hypothesis which is currently being tested in our laboratory is consistent with the higher protein and enzymatic activity observed in Bp-ALL <italic>vs</italic>. T-ALL. If confirmed, the responsible regulatory elements could be used as targets for molecular or pharmacological interventions aimed at increasing FPGS expression and overcome <italic>de novo </italic>or acquired resistance in selected leukemic phenotypes were low FPGS expression mediates unresponsiveness to MTX. An alternative, although unlikely explanation is that failure from our 5'-RACE conditions to amplify the short transcript or enriched amplification of the longer form in CCRF-CEM cells resulted in these findings.</p><p>The 5'-flanking region of the downstream FPGS promoter contains eight forward/one reverse Sp1, two inverted NFY-boxes, and one E-box putative binding sites (see Figure <xref ref-type="fig" rid="F2">2</xref>) in which one NFY, one E-box, and three Sp1 binding sites are present within the FPGS minimal promoter region [<xref ref-type="bibr" rid="B17">17</xref>]. Functional activity of these Sp1 sites was previously demonstrated [<xref ref-type="bibr" rid="B17">17</xref>]. Herein, we determined that the putative transcription factor binding site NFY, but not E-box, plays a positive role in FPGS transcription in both Bp- and T-lineages. Several studies have demonstrated that the CCAAT box plays a role in gene transcription by binding specific transcription factors such as C/EBP, NF-1, and NFY [<xref ref-type="bibr" rid="B24">24</xref>-<xref ref-type="bibr" rid="B26">26</xref>]. EMSA experiments confirmed that the NFY transcription factor is part of the complex that binds the CCAAT sequence within the FPGS promoter. The NFY transcription factor is involved in the regulation of many TATA-less promoter genes such as RAG-1, [<xref ref-type="bibr" rid="B22">22</xref>,<xref ref-type="bibr" rid="B27">27</xref>], EPHX1 [<xref ref-type="bibr" rid="B25">25</xref>], and Gfi-1B [<xref ref-type="bibr" rid="B28">28</xref>]. Therefore, it is not unexpected that NFY plays a role in the regulation of the TATA-less FPGS promoter. In addition, it has been shown that NFY cooperates with many adjacent transcription factors such as GATA-1 [<xref ref-type="bibr" rid="B28">28</xref>,<xref ref-type="bibr" rid="B29">29</xref>], C/EBPα [<xref ref-type="bibr" rid="B25">25</xref>], and Sp1 [<xref ref-type="bibr" rid="B30">30</xref>], through protein-protein interactions to mediate gene transcription. Ongoing studies are investigating the interactions of NFY and some of these proteins as regulators of FPGS gene expression.</p><p>Within the FPGS promoter region we identified one CpG island at position -330 to +294 encompassing Sp1 and NFY sites. We found no role of promoter specific methylation in differential FPGS gene expression between CCRF-CEM and NALM6 cells. Promoter regions with adjacent CpG islands were found to initiate multiple transcripts similar to those identified in FPGS transcription start sites [<xref ref-type="bibr" rid="B31">31</xref>]. Therefore, nuclear factor(s) such as Sp1 that binds to GC-rich motifs within the CpG island and NFY could operate in conjunction to regulate human FPGS gene transcription.</p></sec><sec><title>Conclusion</title><p>We demonstrated that the minimal FPGS promoter region previously described in CCRF-CEM is not sufficient to effectively drive FPGS transcription in NALM6 cells, suggesting that different regulatory elements are required for FPGS gene expression in Bp-cells. Our 5'-RACE analysis showed that two major transcripts encoding the mitochondrial/cytosolic and cytosolic isoforms were expressed in Bp-ALL (NALM6 and REH) whereas T-ALL (CCRF-CEM) cells expressed only the mitochondrial/cytosolic transcript. We determined that the putative transcription factors binding site NFY, but not E-box, plays a role in FPGS transcription in both Bp- and T-lineages. Taken together, our data indicate that the control of FPGS expression in human hematopoietic cells is complex and involves lineage-specific differences in regulatory elements, transcription initiation rates, and mRNA processing. Understanding the lineage-specific mechanisms of FPGS expression should lead to improved therapeutic strategies aimed at overcoming MTX resistance in leukemic cells by upregulating FPGS and increasing accumulation of MTX-PGs in those phenotypes with low FPGS expression. Physiologically FPGS is also known to be essential for eukaryotic cell survival although its targeted inhibition as a novel anticancer strategy has been elusive [<xref ref-type="bibr" rid="B32">32</xref>]. Further understanding of the genetic mechanisms that control FPGS expression could also lead to the development of novel molecular strategies capable of inducing apoptosis in leukemic cells by selectively turning off FPGS expression and could translate to better treatment outcomes for children with high-risk or refractory ALL.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>GJL conceived of the study, participated in its design, carried out the molecular cloning, 5'-RACE, site-directed mutagenesis, EMSA, methylation-specific PCR, transfection and luciferase/β-galactosidase assays, and drafted the manuscript. GML participated in the primers design, site-directed mutagenesis studies, EMSA, and statistical analysis. TTHS participated in the nuclear run-on assays. JCB conceived of the study, participated in its design and coordination, and drafted the manuscript. All authors read and approved the final manuscript.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1471-2407/6/132/prepub"/></p></sec>
Expression analysis of mammaglobin A (<italic>SCGB2A2</italic>) and lipophilin B (<italic>SCGB1D2</italic>) in more than 300 human tumors and matching normal tissues reveals their co-expression in gynecologic malignancies	<sec><title>Background</title><p>Mammaglobin A (<italic>SCGB2A2</italic>) and lipophilin B (<italic>SCGB1D2</italic>), two members of the secretoglobin superfamily, are known to be co-expressed in breast cancer, where their proteins form a covalent complex. Based on the relatively high tissue-specific expression pattern, it has been proposed that the mammaglobin A protein and/or its complex with lipophilin B could be used in breast cancer diagnosis and treatment. In view of these clinical implications, the aim of the present study was to analyze the expression of both genes in a large panel of human solid tumors (n = 309), corresponding normal tissues (n = 309) and cell lines (n = 11), in order to evaluate their tissue specific expression and co-expression pattern.</p></sec><sec sec-type="methods"><title>Methods</title><p>For gene and protein expression analyses, northern blot, dot blot hybridization of matched tumor/normal arrays (cancer profiling arrays), quantitative RT-PCR, non-radioisotopic RNA <italic>in situ </italic>hybridization and immunohistochemistry were used.</p></sec><sec><title>Results</title><p>Cancer profiling array data demonstrated that mammaglobin A and lipophilin B expression is not restricted to normal and malignant breast tissue. Both genes were abundantly expressed in tumors of the female genital tract, i.e. endometrial, ovarian and cervical cancer. In these four tissues the expression pattern of mammaglobin A and lipophilin B was highly concordant, with both genes being down-, up- or not regulated in the same tissue samples. In breast tissue, mammaglobin A expression was down-regulated in 49% and up-regulated in 12% of breast tumor specimens compared with matching normal tissues, while lipophilin B was down-regulated in 59% and up-regulated in 3% of cases. In endometrial tissue, expression of mammaglobin A and lipophilin B was clearly up-regulated in tumors (47% and 49% respectively). Both genes exhibited down-regulation in 22% of endometrial tumors. The only exceptions to this concordance of mammaglobin A/lipophilin B expression were normal and malignant tissues of prostate and kidney, where only lipophilin B was abundantly expressed and mammaglobin A was entirely absent. RNA <italic>in situ </italic>hybridization and immunohistochemistry confirmed expression of mammaglobin A on a cellular level in endometrial and cervical cancer and their corresponding normal tissues.</p></sec><sec><title>Conclusion</title><p>Altogether, these data suggest that expression of mammaglobin A and lipophilin B might be controlled in different tissues by the same regulatory transcriptional mechanisms. Diagnostic assays based on mammaglobin A expression and/or the mammaglobin A/lipophilin B complex appear to be less specific for breast cancer, but with a broader spectrum of potential applications, which includes gynecologic malignancies.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Zafrakas</surname><given-names>Menelaos</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Petschke</surname><given-names>Beate</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Donner</surname><given-names>Andreas</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Fritzsche</surname><given-names>Florian</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Kristiansen</surname><given-names>Glen</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Knüchel</surname><given-names>Ruth</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A7" corresp="yes" contrib-type="author"><name><surname>Dahl</surname><given-names>Edgar</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Cancer	<sec><title>Background</title><p>Mammaglobin A (secretoglobin, family 2A, member 2 – <italic>SCGB2A2</italic>) and lipophilin B (secretoglobin, family 1D, member 2 – <italic>SCGB1D2</italic>) are members of the secretoglobin superfamily, a group of small, secretory, rarely glycosylated, dimeric proteins with unclear physiologic functions, mainly expressed in mucosal tissues [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B2">2</xref>]. The rabbit uteroglobin is the founder member of this family of mammalian proteins [<xref ref-type="bibr" rid="B1">1</xref>], which has expanded to more than 25 members in recent years, currently including nine human secretoglobins. Mammaglobin A, lipophilin B, and most of the human secretoglobins are localized on chromosome 11q13, where they form a dense cluster [<xref ref-type="bibr" rid="B1">1</xref>].</p><p>The mammaglobin A gene (<italic>SCGB2A2</italic>) encodes a 93-amino acid protein with a predicted molecular mass of 10.5 kDa [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B4">4</xref>]. In breast tissue it exists in two main forms with approximate molecular masses of 18 and 25 kDa, due to posttranslational modifications [<xref ref-type="bibr" rid="B5">5</xref>]. Mammaglobin A is considered to be a highly specific breast tissue marker; initially it was found to be overexpressed in breast cancer, and its expression was restricted to normal and malignant breast tissue [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B4">4</xref>]. No gene amplification or gene rearrangement was detected in tumors overexpressing mammaglobin A, suggesting changes in transcriptional regulation as the cause of overexpression [<xref ref-type="bibr" rid="B4">4</xref>]. In contrast to other members of the secretoglobin family [<xref ref-type="bibr" rid="B6">6</xref>], its expression does not appear to be influenced by steroid hormones [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B7">7</xref>].</p><p>Due to its tissue specificity, mammaglobin A has drawn much attention with more than 70 relevant publications in the last five years. More than 30 studies have evaluated its role in detection of minimal residual disease in breast cancer patients, while others investigated its role as a diagnostic and prognostic marker, and its potential use as a therapeutic target (see Ref. 8 for review). Recently however, some studies have shown that it is also expressed in tissues other than the breast [<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B9">9</xref>-<xref ref-type="bibr" rid="B14">14</xref>]. In breast cancer mammaglobin A is overexpressed in a high proportion of primary tumors [<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B14">14</xref>-<xref ref-type="bibr" rid="B17">17</xref>], and it is associated with estrogen receptor positive tumors, a less aggressive tumor phenotype [<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B15">15</xref>,<xref ref-type="bibr" rid="B17">17</xref>], and relapse-free survival [<xref ref-type="bibr" rid="B7">7</xref>].</p><p>Lipophilin B (<italic>SCGB1D2</italic>) has not been studied as extensively as mammaglobin A. The secreted lipophilins A, B, and C should not be confused with the family of lipophilins described as hydrophobic integral membrane proteins in myelin [<xref ref-type="bibr" rid="B1">1</xref>]. Lipophilin B is expressed in a high proportion of breast carcinomas [<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B18">18</xref>], it is more frequently expressed in estrogen receptor positive tumors [<xref ref-type="bibr" rid="B14">14</xref>], but it shows a lower degree of tissue-specificity [<xref ref-type="bibr" rid="B18">18</xref>]. Recently, two studies independently showed that in breast cancer the mammaglobin A and lipophilin B proteins form a covalent complex, and that the two proteins are bonded in a head-to-tail orientation [<xref ref-type="bibr" rid="B19">19</xref>,<xref ref-type="bibr" rid="B20">20</xref>]. Moreover, the expression levels of mammaglobin A in breast tumors were significantly correlated with those of lipophilin B [<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B19">19</xref>,<xref ref-type="bibr" rid="B20">20</xref>].</p><p>The association between mammaglobin A and lipophilin B in breast cancer, the controversy about tissue-specificity of mammaglobin A, and the clinical implications by the use of both genes in cancer early detection, diagnosis, and treatment gave us the impetus to systematically analyse their expression in a large panel of normal and malignant human tissues and cell lines. We report herein that mammaglobin A expression and its co-expression with lipophilin B are not restricted to breast cancer, and that their applications in cancer diagnosis and treatment could also include malignancies of the female genital tract.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Tissue specimens and cell lines</title><p>Formalin-fixed paraffin-embedded tissue from cervical, endometrial and breast cancer and corresponding normal tissue specimens were obtained from patients treated at the Gynecology Departments of the Charité Berlin and the University Hospital of Aachen, with institutional review board approval. Cell lines were obtained from ATCC and cultured as described in the ATCC cell biology catalogue (LGC Promochem, Teddington, England). The following 11 cell lines were analyzed by RT-PCR: HaCat (human keratinocytes), MCF-10A (breast tissue, fibrocystic disease), T47D (breast cancer), ZR75.1 (breast cancer), MDA-MB 468 (breast cancer), MDA-MB 231 (breast cancer), PC3 (prostate cancer), LnCaP (prostate cancer), DU 145 (prostate cancer), MaTu (breast cancer), and A375 (malignant melanoma).</p></sec><sec><title>Multiple tissue northern blot in malignant and normal breast tissue</title><p>Mammaglobin A and lipophilin B expression was analyzed on a Clontech (Heidelberg, Germany) multiple tissue northern blot containing four pairs of invasive ductal carcinoma and matched normal tissue from four female patients (51, 36, 47, and 45 years old). Hybridization was performed as described in the following section for the tumor/normal cDNA arrays.</p></sec><sec><title>Expression analysis using tumor/normal cDNA arrays</title><p>Mammaglobin A and lipophilin B expression were each analyzed using two different nylon filter arrays from Clontech (Heidelberg, Germany), each containing spotted cDNAs from tumor and corresponding normal tissue of the same patient. The "Matched Tumor/Normal Expression Array" (MTNA) (Clontech, Product number 7840) consisted of 136 cDNAs, synthesized from 68 tumor and 68 matched normal tissue specimens. The "Cancer Profiling Array" (CPA) (Clontech, Product number 7841) consisted of 511 dots with 494 cDNAs synthesized from 241 primary tumor, 241 matched normal tissue, and 12 cDNAs from metastases corresponding to 12 of the tumor/normal pairs. Each cDNA pair was independently normalized based on the expression of housekeeping genes used as controls and immobilized in separate dots [<xref ref-type="bibr" rid="B22">22</xref>]. Data for controls and clinicopathological parameters for each specimen can be found on the provider's website [<xref ref-type="bibr" rid="B23">23</xref>,<xref ref-type="bibr" rid="B24">24</xref>].</p><p>For both the MTNA and CPA, hybridization was performed using 25 ng of a gene-specific <sup>32</sup>P-labeled cDNA probe derived from Unigene cDNA clones (<italic>SCGB2A2</italic>: AA513640; <italic>SCGB1D2</italic>: AJ224172). These gene-specific cDNA fragments were radiolabelled using a Megaprime labelling kit (Amersham Biosciences, Braunschweig, Germany), hybridized overnight at 68°C using ExpressHyb Hybridization Solution (Clontech, Heidelberg, Germany), washed, and exposed to Kodak XAR-5 X-ray film with an intensifying screen (Eastman Kodak Co, Rochester, NY, USA). The tumor/normal intensity ratio was calculated using a Typhoon 9410 High Performance Imager (GE-Healthcare, Chalfont St. Giles, UK) and normalized against the background.</p><p>The specificity of the mammaglobin A and lipophilin B hybridization probes was determined by co-hybridization of nylon membranes containing different concentrations of spotted mammaglobin A and lipophilin B cDNAs in plasmid clones: 1 ng, 100 pg, 10 pg and 1 pg of cDNA from each gene were diluted in 100 ul of 15XSSC buffer, heat-denatured for 5 min by boiling and then quenched on ice. Denatured cDNAs were spotted on Hybond N+ membranes (Amersham Biosciences, Freiburg, Germany) using a vacuum manifold (Millipore, Eschborn, Germany). These membranes were treated during filter hybridization, washing and exposition exactly like the tumor/normal arrays</p></sec><sec><title>Quantitative RT-PCR</title><p>Mammaglobin A and lipophilin B expression were analyzed with real-time RT-PCR in a panel of 11 cell lines (see above) and 23 normal human tissues (see Figures <xref ref-type="fig" rid="F5">5</xref> and <xref ref-type="fig" rid="F6">6</xref>) using commercially available RNA (Clontech, Heidelberg, Germany). For each cDNA synthesis, 1μg of RNA was reverse transcribed using the Superscript II Reverse Transcription System (Invitrogen, Karlsruhe, Germany), according to the instructions of the manufacturer.</p><p>Real-time RT-PCR was performed with the Gen Amp<sup>® </sup>5700 sequence detection system (PE Applied Biosystems, Weiterstadt, Germany), using intron-spanning primers and FAM (5' end)/TAMRA (3' end) – labeled specific oligonucleotides. The housekeeping gene <italic>GAPDH </italic>was used as reference. Primers and probes used in this study are presented in Table <xref ref-type="table" rid="T1">1</xref>. Each PCR reaction was performed in a 25μl volume, which included 12.5μl 2XTaqMan Universal PCR-Mastermix (PE Applied Biosystems, Weiterstadt, Germany), 1 ng of cDNA template, 300 nM of forward and 900 nM of reverse primer, and the specific probe for each gene (150 nM for mammaglobin A and 100 nM for lipophilin B). Gene expression was quantified by the comparative C<sub>T </sub>method, normalizing C<sub>T</sub>-values to the housekeeping gene <italic>GAPDH </italic>and calculating the relative expression values of tumor and normal tissues [<xref ref-type="bibr" rid="B21">21</xref>].</p></sec><sec><title>Non-radioisotopic RNA in situ hybridization</title><p>Non-radioisotopic RNA <italic>in situ </italic>hybridization in cervical and endometrial cancer and matched normal tissue was performed as previously described [<xref ref-type="bibr" rid="B25">25</xref>].</p></sec><sec><title>Immunohistochemistry</title><p>Formalin-fixed paraffin embedded tissue was freshly cut (4μm). The sections were mounted on superfrost slides (Menzel Gläser, Braunschweig, Germany), deparaffinized with xylene and gradually hydrated. We used a monoclonal anti-mammaglobin A antibody (BioPrime, NY, USA, MAM001-05, dilution 1:100). Antigen retrieval for mammaglobin A was achieved by heat and citrate buffer using the Ventana immunostainer and all slides were stained with the BenchMark<sup>® </sup>XT autostainer (Ventana, Tucson AZ, USA).</p></sec></sec><sec><title>Results</title><sec><title>Expression analysis using multiple tissue northern blots</title><p>Mammaglobin A (<italic>SCGB2A2</italic>) and lipophilin B (<italic>SCGB1D2</italic>) expression was analyzed by northern blot in a panel of 4 matched breast cancer/normal breast tissue pairs (Figure <xref ref-type="fig" rid="F1">1</xref>). Transcripts of approximately 600 bp in size for both genes were expressed in the same two out of four tumor samples, with hybridization signals of similar intensity in each sample (compare Figure <xref ref-type="fig" rid="F1">1A</xref> with Figure <xref ref-type="fig" rid="F1">1B</xref>).</p></sec><sec><title>Expression analysis using Cancer Profiling Arrays (CPAs)</title><p>Mammaglobin A (<italic>SCGB2A2</italic>) and lipophilin B (<italic>SCGB1D2</italic>) expression were analyzed by dot blot analysis using Clontech's "Matched Tumor/Normal Array" (MTNA) and "Cancer Profiling Array I" (CPA) for each gene. The two expression arrays together contained 630 cDNAs synthesized from 309 human tumor and 309 matched normal tissue specimens, and 12 cDNAs from human metastases corresponding to 12 of the tumor/normal pairs. The specificity of the mammaglobin A and lipophilin B hybridizations on the arrays was established by co-hybridization of two dot blots, containing spotted plasmid cDNAs of either mammaglobin A or lipophilin B (see Figure <xref ref-type="fig" rid="F2">2</xref>). The radiolabelled mammaglobin A probe efficiently hybridized only to mammaglobin A cDNA (not to lipophilin B cDNA), and it was detectable up to a concentration of 1 pg. Likewise, the radiolabelled lipophilin B probe efficiently hybridized only to lipophilin B cDNA, and it was detectable up to a concentration of 10 pg. Thus, cross-hybridization was excluded between the two related genes of the secretoglobin family.</p><p>Overall, abundant expression of at least one of the two genes was detected in six of the 13 tested primary tumor entities and corresponding normal tissues. These results are summarized in Table <xref ref-type="table" rid="T2">2</xref>. Mammaglobin A expression was abundant in malignant and normal samples from the breast (Figure <xref ref-type="fig" rid="F2">2A</xref>), uterus (Figure <xref ref-type="fig" rid="F2">2C</xref>), ovaries (Figure <xref ref-type="fig" rid="F2">2E</xref>) and uterine cervix (Figure <xref ref-type="fig" rid="F2">2G</xref>), and absent in the majority of samples from the other nine tissues (prostate, kidney, colon, rectum, small intestine, stomach, pancreas, lung, and thyroid). In the small number of metastatic samples tested its expression was heterogeneous (Figure <xref ref-type="fig" rid="F2">2</xref>). Mammaglobin A expression was also detectable in one out of 25 gastric, one out of 24 lung, one out of 34 kidney, and two out of 25 rectal tumors, but not in the corresponding normal samples (data not shown). As in the case of mammaglobin A, lipophilin B expression was abundant in malignant and normal samples from the breast (Figure <xref ref-type="fig" rid="F2">2B</xref>), uterus (Figure <xref ref-type="fig" rid="F2">2D</xref>), ovaries (Figure <xref ref-type="fig" rid="F2">2F</xref>) and uterine cervix (Figure <xref ref-type="fig" rid="F2">2H</xref>). In addition, abundant lipophilin B expression was found in matched samples from the kidney and the prostate (Table <xref ref-type="table" rid="T2">2</xref>), while it was absent in most samples from the other seven tissues. Lipophilin B expression was also detectable in one out of 25 gastric, one out of 24 lung, and two out of 25 rectal tumors, but not in the corresponding normal samples (data not shown).</p><p>The expression pattern of mammaglobin A in malignant and normal samples from different tissues was in general highly concordant to that of lipophilin B (compare e.g. Figure <xref ref-type="fig" rid="F2">2A</xref> and <xref ref-type="fig" rid="F2">2B</xref> for breast tissue), except for kidney and prostate samples, where only lipophilin B but not mammaglobin A was expressed. The two genes exhibited an identical pattern of differential expression (i.e. both down-, up- or non-regulation) in the majority of matched pairs from the breast (78%), uterus (78%), ovaries (56%), and the uterine cervix (100%) (Table <xref ref-type="table" rid="T2">2</xref>), without marked disparities (i.e. no cases with one gene up- and the other down-regulated) in the remaining cases. According to the Spearman rank correlation test, co-expression of mammaglobin A and lipophilin B was highly significant in breast, uterine and ovarian tissues (each p < 0.001) but failed to reach significance in cervical tissues due to the small sample size (n = 2). A very interesting finding was that mammaglobin A and lipophilin B were both up-regulated in the same one out of 25 gastric, and the same two out of 25 rectal tumor samples, in which expression of the two genes was detectable. However, this was not the case with lung tumors, in which mammaglobin A and lipophilin B were each expressed in one out of 24, but not in the same sample (data not shown). No correlation was found between the (co)expression pattern of mammaglobin A and lipophilin B in various tumors and available clinicopathological data.</p></sec><sec><title>Quantitative RT-PCR</title><p>Mammaglobin A (<italic>SCGB2A2</italic>) and lipophilin B (<italic>SCGB1D2</italic>) expression were analyzed with real-time RT-PCR in a panel of 11 cell lines and 23 normal human tissues using commercially available RNA (Clontech, Heidelberg, Germany). Among the cell lines tested, mammaglobin A was expressed only in the breast cancer cell line ZR-75.1, and negative in the other five breast and five non breast cell lines (Figure <xref ref-type="fig" rid="F3">3</xref>). Lipophilin B was expressed in the same cell line, as well as in the T-47D (breast cancer cell line) and LnCaP cells (prostate cancer cell line)(Figure <xref ref-type="fig" rid="F4">4</xref>). There was no major difference in expression of both genes between tumor cells grown under confluent and subconfluent conditions.</p><p>Among all normal tissues tested, mammaglobin A expression was highest in normal tissue from the uterine cervix, followed in descending order by normal breast tissue, thymus, uterus, testis, trachea, and stomach. No mammaglobin A expression was detected in the other 16 normal tissues (Figure <xref ref-type="fig" rid="F5">5</xref>). As with mammaglobin A, lipophilin B expression was highest in normal tissue from the uterine cervix. Lipophilin B was also expressed in descending order in the uterus, breast, kidney, colon, pancreas, heart, placenta, and testis. There was no detectable lipophilin B expression in the other 14 normal tissues (Figure <xref ref-type="fig" rid="F6">6</xref>).</p></sec><sec><title>Non-radioisotopic RNA in situ hybridization and immunohistochemistry</title><p>In order to further establish the expression of mammaglobin A in gynecologic malignancies, we have analyzed its expression in tissue sections from cervical and endometrial cancer and normal tissue using non-radioisotopic RNA <italic>in situ </italic>hybridization. Consistently with the dot blot hybridization and quantitative RT-PCR results presented above, mammaglobin A was expressed in normal cervical glands (Figure <xref ref-type="fig" rid="F7">7A</xref>) as well as in cervical and endometrial cancer (Figure <xref ref-type="fig" rid="F7">7 D, G</xref>). Representative sections are presented in Figure <xref ref-type="fig" rid="F7">7</xref>. Furthermore we performed immunohistochemistry using a mammaglobin A-specific antibody (BioPrime, NY, USA,) on paraffin-embedded tissue from breast cancer, as well as cervical and endometrial cancer. Mammaglobin A was clearly detectable in invasive ductal (Figure <xref ref-type="fig" rid="F8">8A</xref>) and invasive lobular (Figure <xref ref-type="fig" rid="F8">8B</xref>) carcinoma of the breast. Mammaglobin A protein was also detectable in squamous cell carcinoma of the cervix (Figure <xref ref-type="fig" rid="F8">8C</xref>) and in endometrioid adenocarcinoma of the endometrium (Figure <xref ref-type="fig" rid="F8">8D</xref>).</p></sec></sec><sec><title>Discussion</title><p>In initial reports mammaglobin A appeared to be an almost ideal tissue marker, since its expression was restricted to normal and malignant breast tissue (see Ref. 8 for review). Subsequently, mammaglobin A was evaluated for detection of minimal residual disease in breast cancer patients [<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B26">26</xref>,<xref ref-type="bibr" rid="B27">27</xref>], differential diagnosis of metastases and malignant pleural effusions [<xref ref-type="bibr" rid="B26">26</xref>-<xref ref-type="bibr" rid="B29">29</xref>], and as an immunotherapeutic target in <italic>in vitro </italic>experiments [<xref ref-type="bibr" rid="B30">30</xref>-<xref ref-type="bibr" rid="B33">33</xref>] and <italic>in vivo </italic>animal models [<xref ref-type="bibr" rid="B34">34</xref>,<xref ref-type="bibr" rid="B35">35</xref>]. However, in later reports, its expression was detected, rarely and/or in low levels, in various normal and malignant tissues: the normal uterine cervix [<xref ref-type="bibr" rid="B10">10</xref>], uterus [<xref ref-type="bibr" rid="B9">9</xref>-<xref ref-type="bibr" rid="B11">11</xref>,<xref ref-type="bibr" rid="B36">36</xref>], ovary [<xref ref-type="bibr" rid="B10">10</xref>,<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B36">36</xref>], thymus, testis, trachea, skeletal muscle, kidney [<xref ref-type="bibr" rid="B36">36</xref>], skin [<xref ref-type="bibr" rid="B18">18</xref>], sweat glands [<xref ref-type="bibr" rid="B13">13</xref>], salivary glands [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B36">36</xref>], prostate [<xref ref-type="bibr" rid="B10">10</xref>], and nasal mucosa [<xref ref-type="bibr" rid="B37">37</xref>], and tumors of the sweat glands [<xref ref-type="bibr" rid="B13">13</xref>], lungs [<xref ref-type="bibr" rid="B12">12</xref>] and ovaries [<xref ref-type="bibr" rid="B14">14</xref>].</p><p>Our results confirm that mammaglobin A is expressed in various normal and malignant tissues other than the breast, and thus it is rather not an ideal breast-specific marker. Expression of mammaglobin A in normal tissues certainly limits its potential use as an immunotherapeutic target, due to concerns about autoimmune toxicity, particularly since autoimmunity is not a concern with other immunotherapeutic targets [<xref ref-type="bibr" rid="B38">38</xref>]. Our results, also confirm previous reports that mammaglobin A is not expressed in all breast cancer cell lines and breast tumors [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B16">16</xref>,<xref ref-type="bibr" rid="B39">39</xref>], and thus does not have a 100% sensitivity as a diagnostic marker. An important finding of our study was that mammaglobin A is commonly expressed in normal and malignant tissue of the female genital tract, and only rarely or at low levels in other normal and malignant tissues. It should be noted that expression in gynecologic tissues was demonstrated by four independent methods (dot blot hybridization of matched tumor/normal arrays, real time RT-PCR, non-radioisotopic RNA <italic>in situ </italic>hybridization and immunohistochemistry). Thus, given the limitations in specificity and sensitivity, mammaglobin A could be also used in diagnostic assays for detection of gynecologic malignancies.</p><p>The expression pattern of lipophilin B in our study, as well as in previous reports, appeared to be even less tissue specific than that of mammaglobin A, and thus its use as a diagnostic marker seems very unlikely. In the present study, lipophilin B was abundantly expressed in normal and malignant tissue from the breast, cervix, uterus, ovary, kidney and prostate. Lower or rare lipophilin B expression was found in normal colon, pancreas, heart, in gastric and rectal tumors, and as previously reported in normal testis and placenta [<xref ref-type="bibr" rid="B36">36</xref>] and lung tumors [<xref ref-type="bibr" rid="B12">12</xref>]. In previous reports, lipophilin B expression was also detected in the normal anterior pituitary and pituitary adenomas [<xref ref-type="bibr" rid="B40">40</xref>], in normal adrenal gland, cartilage, retina, [<xref ref-type="bibr" rid="B18">18</xref>], skin [<xref ref-type="bibr" rid="B19">19</xref>], and salivary gland [<xref ref-type="bibr" rid="B19">19</xref>,<xref ref-type="bibr" rid="B36">36</xref>].</p><p>The most important finding regarding lipophilin B expression in the present study was that it is concordant to that of mammaglobin A in most tissues tested. It has been previously reported, that mammaglobin A and lipophilin B are significantly co-expressed in breast cancer [<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B19">19</xref>,<xref ref-type="bibr" rid="B20">20</xref>], and their proteins are bonded in an antiparallel manner forming a covalent complex [<xref ref-type="bibr" rid="B19">19</xref>,<xref ref-type="bibr" rid="B20">20</xref>]. In the present study we found that their co-expression is not restricted to breast tumors, but is also present in normal breast tissue, as well as normal and malignant tissue from the uterus, ovaries, and uterine cervix. On the other hand, in normal and malignant prostate and kidney tissue lipophilin B was abundantly expressed while mammaglobin A was entirely absent. Interestingly, the only gastric and two rectal tumors expressing mammaglobin A expressed lipophilin B as well, but this was not seen in lung cancer. Altogether, these data suggest that expression of the two genes, which are both localized on the same cluster on chromosome 11q13, is probably regulated by common transcriptional mechanisms. It is also reasonable to expect, that serum antibodies against lipophilin B or against its complex with mammaglobin A, as those previously detected in breast cancer patients [<xref ref-type="bibr" rid="B18">18</xref>], could also be found in patients with gynecologic tumors.</p></sec><sec><title>Conclusion</title><p>Systematic expression analysis of a panel of solid tumors and normal tissues showed that mammaglobin A and lipophilin B are abundantly expressed in malignant and normal tissues of the breast and the female genital tract, namely the cervix, uterus, and ovary, while lower expression levels were rarely found in other tumors and normal tissues. Intriguingly, the expression pattern of the two genes was highly concordant in most tissues tested, suggesting common regulatory transcriptional mechanisms. Use of mammaglobin A and its complex with lipophilin B in breast cancer diagnosis might lead to less specific results than previously expected, but these markers could also be used in diagnosis of gynecologic cancer. The potential use of mammaglobin A as an immunotherapeutic target might be limited, due to the possibility of autoimmune toxicity.</p></sec><sec><title>Abbreviations</title><p><italic>SCGB2A2</italic>: secretoglobin, family 2A, member 2; <italic>SCGB1D2</italic>: secretoglobin, family 1D, member 2; MTNA: Matched Tumor/Normal Array; CPA: Cancer Profiling Array; RT-PCR: reverse transcription – polymerase chain reaction; GAPDH: Glyceraldehyde-3-phosphate dehydrogenase.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>MZ: participated in design of the study, data analysis, data interpretation and drafted the manuscript; BP: carried out the molecular studies, and critically revised the manuscript; AD: supported with pathological expertise in data interpretation and critically revised the manuscript; FF: established and performed the mammaglobin A immunohistochemistry analysis; GK: pathologist that analyzed the mammaglobin A immunohistochemistry study and critically revised the manuscript; RK: participated in design and coordination of the study, and critically revised the manuscript; ED conceived the study, participated in study design and coordination, molecular and data analysis, data interpretation and drafting of the manuscript.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1471-2407/6/88/prepub"/></p></sec>
PPARδ status and mismatch repair mediated neoplasia in the mouse intestine	<sec><title>Background</title><p>Therapeutic regulation of PPARδ activity using selective agonists has been proposed for various disorders. However, the consequences of altered peroxisome proliferator-activated receptor delta (PPARδ) activity in the context of intestinal tumourigenesis remain somewhat unclear. Contradictory evidence suggesting PPARδ either attenuates or potentiates intestinal neoplasia. To further investigate the PPARδ dependency of intestinal tumourigenesis, we have analysed the consequences of PPARδ deficiency upon intestinal neoplasia occurring in mice with impaired mismatch DNA repair.</p></sec><sec sec-type="methods"><title>Methods</title><p>Mice deficient for both PPARδ and the mismatch repair gene Mlh1 were produced and the incidence and severity of intestinal neoplasia recorded.</p></sec><sec><title>Results</title><p>No significant differences between the control genotypes and the double mutant genotypes were recorded indicating that deficiency of PPARδ does not modify impaired mismatch repair induced neoplasia.</p></sec><sec><title>Conclusion</title><p>In contrast with the previously observed acceleration of intestinal neoplasia in the context of the <italic>Apc</italic><sup><italic>Min</italic>/+ </sup>mouse, PPARδ deficiency does not alter the phenotype of mismatch repair deficiency. This data supports the notion that PPARδ is not required for adenoma formation and indicate that any pro-tumourigenic effect of PPARδ inactivation may be highly context dependent.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Reed</surname><given-names>Karen R</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Sansom</surname><given-names>Owen J</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Hayes</surname><given-names>Anthony J</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Gescher</surname><given-names>Andreas J</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Peters</surname><given-names>Jeffrey M</given-names></name><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Clarke</surname><given-names>Alan R</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Cancer	<sec><title>Background</title><p>Peroxisome proliferator-activated receptors (PPARs) are lipid-activated transcription factors exerting several functions in development and metabolism. There are 3 major PPAR isoforms; α,β/δ and γ and each has distinct agonist binding properties and different regulation of expression resulting in distinct distributions [<xref ref-type="bibr" rid="B1">1</xref>]. The roles for PPARδ appear diverse and are not fully characterised, but include the regulation of lipid uptake, metabolism, and regulation of proliferation and differentiation within many different cell types. Consequently, the therapeutic regulation of PPARδ activity using selective agonists has been proposed for many varied disorders including: lung cancer [<xref ref-type="bibr" rid="B2">2</xref>], experimental autoimmune encephalomyelitis [<xref ref-type="bibr" rid="B3">3</xref>], skin disorders such as psoriasis and cancer [<xref ref-type="bibr" rid="B4">4</xref>], type 2 diabetes [<xref ref-type="bibr" rid="B5">5</xref>], metabolic syndrome [<xref ref-type="bibr" rid="B6">6</xref>] and dyslipidemias [<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B8">8</xref>].</p><p>Although a great deal of evidence exists to show that PPARδ is potentially important in intestinal tumourigenesis, it is currently unclear whether PPARδ attenuates or potentiates this condition. Several studies have shown that activation of PPARδ or increased PPARδ levels are associated with increased intestinal neoplasia in a variety of tissues [<xref ref-type="bibr" rid="B9">9</xref>-<xref ref-type="bibr" rid="B13">13</xref>]. Also, two studies have shown that PPARδ deficiency suppressed or had no role upon tumourigenesis [<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B15">15</xref>]. Taken together these analyses suggest that PPARδ potentiates colon carcinogenesis. However, two independent approaches have recently suggested that PPARδ expression in vivo is not up-regulated in intestinal adenomas. Firstly a study of matched human tumour and normal intestinal tissues found reduced levels of PPARδ expression in the tumours [<xref ref-type="bibr" rid="B16">16</xref>]. Second several studies using the classical mouse model of intestinal neoplasia (the Apc<sup>Min </sup>mouse), found either no change, or reduced PPARδ expression in colonic adenomas compared to normal tissues in this model [<xref ref-type="bibr" rid="B17">17</xref>-<xref ref-type="bibr" rid="B20">20</xref>], while a recent publication has shown that ligand activation of PPARδ attenuates chemically induced colon carcinogenesis [<xref ref-type="bibr" rid="B20">20</xref>]. Furthermore, it has been demonstrated that PPARδ deficiency does not suppress intestinal tumourigensis in <italic>Apc</italic><sup><italic>Min</italic>/+ </sup>mice, but indeed promotes some aspects of intestinal neoplasia [<xref ref-type="bibr" rid="B17">17</xref>,<xref ref-type="bibr" rid="B20">20</xref>,<xref ref-type="bibr" rid="B21">21</xref>]. These data suggest that PPARδ attenuates colon carcinogenesis.</p><p>Thus, given the obvious disparity within the published literature, we have utilized another mouse neoplasia model to further investigate the role of PPARδ in intestinal tumourigenesis. Mice possessing null mutations in the mismatch repair (MMR) gene Mlh1 are prone to develop different types of neoplasia and present with lymphomas and intestinal tumours but show increased mutation in all tissues examined [<xref ref-type="bibr" rid="B22">22</xref>]. Likewise, germline mutations in the human MLH1 gene are involved in Hereditary non-polyposis colorectal cancer [<xref ref-type="bibr" rid="B23">23</xref>]. We have inter-crossed PPARδ null mice [<xref ref-type="bibr" rid="B24">24</xref>] to the mismatch repair defective <italic>Mlh1 </italic>null mice [<xref ref-type="bibr" rid="B25">25</xref>] and produced cohorts for the different combinations of the genotypes in order to investigate the consequences of impaired MMR induced tumourigenesis in the context of PPARδ deficiency.</p></sec><sec sec-type="methods"><title>Methods</title><p>Mice were generated from sixth generation C57BL 6 backcrossed mice. All experiments were performed according to UK Home Office regulations. Inter-crossing the PPARδ null mice (<italic>PPAR</italic>δ<sup>-/-</sup>) to the mismatch repair defective Mlh1 null mice (<italic>Mlh1</italic><sup>-/-</sup>) produced cohorts containing a minimum of 16 animals for each genotype combination of interest (<italic>Mlh1</italic><sup>-/-</sup><italic>PPAR</italic>δ<sup>+/+</sup><italic>, Mlh1</italic><sup>-/-</sup><italic>PPAR</italic>δ<sup>+/-</sup><italic>, Mlh1</italic><sup>-/-</sup><italic>PPAR</italic>δ<sup>-/-</sup>). Littermates were genotyped by PCR on DNA from tail biopsy and allowed to age and monitored for signs of intestinal tumours. Animals were harvested when they displayed overt symptoms of disease, and tumour burden was ascertained upon dissection.</p></sec><sec><title>Results and discussion</title><p>Through inter-crossing the PPARδ null mice (<italic>PPAR</italic>δ<sup>-/-</sup>) to the mismatch repair defective Mlh1 null mice (<italic>Mlh1</italic><sup>-/-</sup>) we produced cohorts containing a minimum of 16 animals for each genotype combination of interest and examined survival, intestinal adenoma multiplicity and tumour size at death for each of the cohorts (Figure <xref ref-type="fig" rid="F1">1</xref>). We find that, although the mean age at death of <italic>Mlh1</italic><sup>-/-</sup><italic>PPAR</italic>δ<sup>-/- </sup>mice was 248.1 days compared to 203.5 days in controls, this was not statistically different (Figure <xref ref-type="fig" rid="F1">1a</xref>, p = 0.34 Log-Rank test). Furthermore, the predisposition to lymphomagenesis was not significantly altered between the <italic>Mlh1</italic><sup>-/-</sup><italic>PPAR</italic>δ<sup>+/+ </sup>and <italic>Mlh1</italic><sup>-/-</sup><italic>PPAR</italic>δ<sup>-/- </sup>cohorts (Figure <xref ref-type="fig" rid="F1">1b</xref>). The possibility remains that there may be subtle effects of PPARδ deficiency that lie below the detection threshold of the present study, a possibility that would be resolved by a substantially increased cohort analysis.</p><p>Contrary to the finding from the APC<sup>Min/+ </sup><italic>PPAR</italic>δ<sup>-/- </sup>intercross, which indicated that PPARδ deficiency promotes some aspects of intestinal neoplasia [<xref ref-type="bibr" rid="B17">17</xref>,<xref ref-type="bibr" rid="B20">20</xref>,<xref ref-type="bibr" rid="B21">21</xref>], no significant differences were discovered between either number or size of intestinal adenomas in the <italic>Mlh1</italic><sup>-/-</sup><italic>PPAR</italic>δ<sup>+/+ </sup>and <italic>Mlh1</italic><sup>-/-</sup><italic>PPAR</italic>δ<sup>-/- </sup>cohorts (Figure <xref ref-type="fig" rid="F1">1c, d</xref>). This was confirmed in both the small intestinal and large intestinal tumours (P > 0.1, Mann Whitney U test), although notably the group sizes in both these analyses were small.</p><p>To assess the nature of the <italic>Mlh1</italic><sup>-/-</sup><italic>PPAR</italic>δ<sup>+/+ </sup>and <italic>Mlh1</italic><sup>-/-</sup><italic>PPAR</italic>δ<sup>-/- </sup>aberrant intestinal tissues, β-catenin immunohisochemistry was performed and again no differences were observed between the two genotypes (data not shown). Up-regulation of β-catenin in large intestinal adenomas was observed in both genotypes. Furthermore in the small intestine, in addition to lesions displaying increased β-catenin levels, it was also possible to identify a small subset of lesions that maintain normal levels of β-catenin, a phenomenon previously described within defective MMR intestine [<xref ref-type="bibr" rid="B26">26</xref>]. Thus, the induction of dysplastic lesions and the deregulation of β-catenin in the intestine occurring as a consequence of defective MMR is not dependent upon PPARδ status and has no requirement for PPARδ.</p><p>We therefore find that loss of PPARδ does not alter tumour incidence or morphology of tumourigenesis on the <italic>Mlh1</italic><sup>-/- </sup>background. Given that MMR deficiency is considered to lead to a mutator phenotype which thereby increases the rate of mutation of elements of the Wnt pathway, our findings argue that PPARδ deficiency does not enhance or diminish such a mutator phenotype. By implication, the increased adenoma formation seen in the <italic>Apc</italic><sup><italic>Min</italic>/+</sup><italic>PPAR</italic>δ<sup>-/- </sup>mutants may arise through PPARδ dependent modification of the frequency of gene conversion events that are known to underlie the majority of polyp formation in the <italic>Apc</italic><sup><italic>Min</italic>/+ </sup>mouse [<xref ref-type="bibr" rid="B27">27</xref>]. Alternatively, subsequent to the gene conversion events, PPARδ deficiency may alter the resulting cellular signalling/gene expression pathways which permit cell survival and tumour progression. Our data again apparently contradict the observed acceleration of adenoma formation following agonist activation of PPARδ[<xref ref-type="bibr" rid="B11">11</xref>], as this predicts reduced adenoma burden in the absence of PPARδ. However, PPARδ deficiency (as assessed through the PPARδ null mutation) may not necessarily be functionally opposite of ectopic PPARδ activation as it is possible that any alteration in the levels of PPARδ activity, whether a reduction or increase, have different consequences dependant on the genetic background or indeed tissue being studied. In terms of evaluating the potential deleterious pro-tumourigenic effect of PPARδ deficiency, our data suggests that this only accelerates adenoma formation in certain defined genetic settings (e.g. <italic>Apc</italic><sup><italic>Min</italic>/+</sup>), and does not generally enhance adenoma formation <italic>per se</italic>. This genetic dependency may reflect differences in the mutational events occurring in the Mlh1 and Apc mutant backgrounds. By analogy, any potential danger in the therapeutic use of PPARδ agonists to activate PPARδ may critically depend upon subtle changes in the underlying genetic predisposition.</p></sec><sec><title>Conclusion</title><p>In summary, we show that PPARδ deficiency does not alter either lymphomagenesis or adenoma formation in mice with defective MMR. This data again support the notion that PPARδ is not required for adenoma formation and indicate that any pro-tumourigenic effect of inactivation may be highly context dependent. Thus, in the context of a defective MMR environment, PPARδ agonism is unlikely to be pro tumourigenic.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>KRR, OJS and AJH participated in the animal studies; KRR and OJS carried out the statistical analysis and data presentation; ARC, JMP and AJG conceived the study, and participated in its design and coordination. KRR drafted the manuscript and all authors read and approved the final manuscript.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1471-2407/6/113/prepub"/></p></sec>
Nucleotide supplementation: a randomised double-blind placebo controlled trial of IntestAidIB in people with Irritable Bowel Syndrome [ISRCTN67764449]	<sec><title>Background</title><p>Dietary nucleotide supplementation has been shown to have important effects on the growth and development of cells which have a rapid turnover such as those in the immune system and the gastrointestinal tract. Work with infants has shown that the incidence and duration of diarrhoea is lower when nucleotide supplementation is given, and animal work shows that villi height and crypt depth in the intestine is increased as a result of dietary nucleotides. Dietary nucleotides may be semi-essential under conditions of ill-health, poor diet or stress. Since people with Irritable Bowel Syndrome tend to fulfil these conditions, we tested the hypothesis that symptoms would be improved with dietary nucleotide supplementation.</p></sec><sec sec-type="methods"><title>Methods</title><p>Thirty-seven people with a diagnosis of Irritable Bowel gave daily symptom severity ratings for abdominal pain, diarrhoea, urgency to have a bowel movement, incomplete feeling of evacuation after a bowel movement, bloating, flatulence and constipation for 28 days (baseline). They were then assigned to either placebo (56 days) followed by experimental (56 days) or the reverse. There was a four week washout period before crossover. During the placebo and experimental conditions participants took one 500 mg capsule three times a day; in the experimental condition the capsule contained the nutroceutical substances. Symptom severity ratings and psychological measures (anxiety, depression, illness intrusiveness and general health) were obtained and analysed by repeated measures ANOVAs.</p></sec><sec><title>Results</title><p>Symptom severity for all symptoms (except constipation) were in the expected direction of baseline>placebo>experimental condition. Symptom improvement was in the range 4 – 6%. A feeling of incomplete evacuation and abdominal pain showed the most improvement. The differences between conditions for diarrhoea, bloating and flatulence were not significant at the p < .05 level. There were no significant differences between the conditions for any of the psychological measures.</p></sec><sec><title>Conclusion</title><p>Dietary nucleotide supplementation improves some of the symptoms of irritable bowel above baseline and placebo level. As expected, placebo effects were high. Apart from abdominal pain and urgency to have a bowel movement, the improvements, while consistent, are modest, and were not accompanied by improvements in any of the psychological measures. We suggest that the percentage improvement over and above the placebo effect is a physiological effect of the nucleotide supplement on the gut. The mechanisms by which these effects might improve symptoms are discussed.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Dancey</surname><given-names>CP</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Attree</surname><given-names>EA</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Brown</surname><given-names>KF</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	Nutrition Journal	<sec><title>Background</title><p>Irritable Bowel Syndrome (IBS) is a chronic disorder affecting an estimated 15 – 22% of western populations, and is the major cause of referrals to gastroenterology clinics in the western world [<xref ref-type="bibr" rid="B1">1</xref>]. The symptoms may include abdominal pain, diarrhoea, urgency to have a bowel movement, a feeling of incomplete evacuation after a bowel movement, flatulence and bloating. Some sufferers experience constipation, or an alternation between constipation and diarrhoea. IBS is more common among women than men, with a 2:1 female: male ratio [<xref ref-type="bibr" rid="B2">2</xref>]. Current medical treatments are directed at symptomatic relief, and although these can give some relief, there is no one treatment which has been shown to be lastingly effective. Although IBS is not a life-threatening disease, the symptoms and the effects of the symptoms on daily life can have a great impact on sufferers [<xref ref-type="bibr" rid="B3">3</xref>]. IBS is also associated with large healthcare and economic costs in terms of hospital investigations, repeated visits to GPs, prescription medicines, and loss of time from work [<xref ref-type="bibr" rid="B4">4</xref>]. Although hospital investigations for more serious diseases such as cancer or Inflammatory Bowel Disease are negative in people with IBS, some abnormalities in the gut have been found. For instance, some patients have been found to have a degree of mucosal inflammation, which may be in response to some foods [<xref ref-type="bibr" rid="B5">5</xref>]. It is possible that people with IBS have immunological reactions to dietary antigens as food elimination based on serum immunoglobulin G antibodies has been found to result in a significant decrease in symptoms of IBS. Both numbers of mast cells and their mediators have been shown to be increased in intestinal mucosa in patients with irritable bowel syndrome, especially in the close proximity of intestinal nerves [<xref ref-type="bibr" rid="B6">6</xref>]. Kalliomäki [<xref ref-type="bibr" rid="B6">6</xref>] suggests that food antigens induce mast cells to secrete mediators which regulate gastrointestinal motility, resulting in alterations in peristalsis and an increase in abdominal pain and discomfort. Furthermore, the mast cell-derived mediators have effects on immune cell functions. It may be then, that the nutrition of people with IBS is more important than has been traditionally thought. As people with IBS tend to believe that their symptoms are affected by diet, they often attempt to manage their disorder by dietary control. However, the only consistent advice given to people with IBS is usually simply to eat a "healthy" diet which includes fruit, vegetables and fibre. In an early study of people with IBS, Dancey & Backhouse [<xref ref-type="bibr" rid="B7">7</xref>] found that although the majority of their sample of 148 people (70%) stated that they were trying to follow a "healthy" diet with large amounts of fruit and vegetables; for many of these people, such a strategy had not led to symptom improvement, and in an attempt to control their IBS, 14% were eating very restricted diets. Some of these diets involved avoiding complete groups of foods, e.g. carbohydrates. Whilst such a strategy may reduce bloating, for instance, such a diet is not likely to enhance wellbeing. As well as eating a sufficient quantity of a wide variety of foods from each food group, micronutrients and nucleotides may also be important for health, especially in the sub-well. It is nucleotides which are the focus of this study.</p><p>Nucleotides are substances which are synthesised endogenously – they have important effects on the growth and development of cells which have a rapid turnover, such as those in the immune system and the gastrointestinal tract. The intestinal epithelium is a rapidly proliferating tissue with a high cellular turnover rate. A complete cell cycle in humans is 24 hours, with a replacement of the entire enteric epithelium within 3–6 days. In healthy people, dietary nucleotides are probably not essential, and in fact most will be metabolised and rapidly excreted from the system. However, under certain circumstances (e.g. in the sub-well, diseased, or under conditions of stress or poor diet) dietary nucleotides may be what Maldonado, Navorro, Narbona, & Gil [<xref ref-type="bibr" rid="B8">8</xref>] call "semi-essential", optimising the function of the gastrointestinal and immune systems. In relation to the gastrointestinal system work has shown that dietary nucleotides enhance the intestinal absorption of iron [<xref ref-type="bibr" rid="B9">9</xref>]. Dietary sources of nucleotides are nucleoproteins and nucleic acids, and these are found to varying degrees in many foods – lamb, liver, mushrooms (but not fruit and other vegetables) all are rich in nucleotides. Rapidly dividing tissue requires a constant supply of nucleotides in order to manufacture essential nucleic acids. Exogenous supplies of nucleotides may optimise tissue function particularly during recovery from mucosal injuries when the endogenous supply may limit the synthesis of nucleic acids.</p><p>Holen & Jonsson [<xref ref-type="bibr" rid="B10">10</xref>] found that dietary nucleotides had beneficial effects, especially when the nutrition supply was inadequate. Work with infants has shown that the incidence and duration of acute diarrhoea is lower in infants when dietary nucleotides are included in their diets [<xref ref-type="bibr" rid="B11">11</xref>]. Previous work on the effect of nucleotide supplementation in animals has found that such supplements are important for the repair mechanism of immune cells [<xref ref-type="bibr" rid="B12">12</xref>]. In piglets, nucleotide supplementation had effects on the gastrointestinal system by increasing villi height and crypt depth. [<xref ref-type="bibr" rid="B13">13</xref>]. Evans, Tian, Gu, Jones & Ziegler [<xref ref-type="bibr" rid="B14">14</xref>], using rats to model short-bowel syndrome, found that nucleotide supplementation is associated with increased jejunal adaptive growth after massive small bowel resection in rats. Dietary nucleotides have been found to help athletes by reducing the release of stress related hormones and chemicals in the body, and by maintaining a higher level of antibodies, which enables the immune system to work more effectively [<xref ref-type="bibr" rid="B15">15</xref>]. In people with a chronic illness such as IBS whose primary symptoms relate to the gastrointestinal tract, nucleotide supplementation may improve symptoms via improved gut function or by an enhancement of the immune system.</p><p>There are particular problems in assessing the benefits of treatments of people with IBS, and some of the problems of this patient group in relation to clinical trials have been discussed in detail by Spiller [<xref ref-type="bibr" rid="B16">16</xref>]. People with IBS show great variability in frequency and severity of symptoms, both when compared to others and also from day-to-day in their own symptoms. Spiller [<xref ref-type="bibr" rid="B16">16</xref>] has shown there are clear benefits to participating in clinical trials; people with IBS tend to be helped by placebo alone. This is thought to be due to a reduction in anxiety and/or depression as a result of help and reassurance given by the people running the trial. If this is the case, then anxiety and depression should reduce over the length of the trial. One would also expect that, if symptoms improve, then psychological well-being should in some way improve also. Thus although our primary aim was to determine whether nucleotide supplementation improved the symptoms of irritable bowel, we also wished to determine whether ratings of anxiety and/or depression would show any change as a result of symptom change.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Design</title><p>A double-blind randomised placebo cross-over study in which participants rated symptoms daily for six months in baseline, placebo, and experimental conditions. This study was approved by our university ethics committee and was in accordance with the code of conduct for psychologists (ethical principles for conducting research with human participants) produced by the British Psychological Society. The research was conducted only when written consent was received from the participant, and their G.P. had provided written confirmation of their diagnosis, and confirmation of no other co-existing conditions.</p></sec><sec><title>Participants</title><p>Thirty seven people with a diagnosis of Irritable Bowel Syndrome took part in the study. The inclusion criteria were that participants should be aged 18 – 65, should have been diagnosed as having IBS by a qualified medical practitioner, and should have diarrhoea as a main symptom. The exclusion criteria were any other co-existing illnesses, and non-confirmation of the diagnosis by G.P. The participant flow is shown below (see Figure <xref ref-type="fig" rid="F1">1</xref>).</p><fig position="float" id="F1"><label>Figure 1</label><caption><p>participant flow through the trial.</p></caption><graphic xlink:href="1475-2891-5-16-1"/></fig><p>Thirty-seven participants completed the trial, during which they completed daily symptom diaries for four weeks (baseline) following by either the experimental condition (eight weeks) or the placebo condition (eight weeks). There was a four week washout period between conditions. The mucosal cell turn-over for the whole tract is approximately six days, with the majority of other cells in the body are also replaced over 30 days. For this reason a 30 day washout was selected. Although some of the participants rated some symptoms as not particularly bothersome over the baseline condition no person was excluded by us.</p></sec><sec><title>IBS symptoms</title><p>The symptoms of interest were: abdominal pain, diarrhoea, urgency to have a bowel movement, incomplete evacuation after a bowel movement, flatulence, bloating, and constipation. Participants rated symptoms every day on a scale of 1 ("no discomfort at all today") to 7 ("very severe discomfort today"). Reliability for these symptoms has been previously established [<xref ref-type="bibr" rid="B17">17</xref>].</p><p>Participants were given a pack containing a month's supply of daily diaries in which to record their symptoms. The next month's supply were sent to the participants once the previous completed pack had been returned.</p></sec><sec><title>Neutroceutical product</title><p>Participants took one 500 mg capsule three times a day with meals – in the experimental condition these contained the neutroceutical product IntestAidIB which consists of the following active substances: nucleotides & RNA (concentrated extracts of Saccharomyces cerevisae), hydroxypropyl methycellulose, FOS (fructo-olligosaccharides), Methionine, Glutamine, Inositol, Lysine, Pantothenic acid (Vitamin B5 as Calcium d-pantothenate), Sodium citrate, Riboflavin (Vitamin B2), Vanillin, Folic acid, and Biotin. Flow agents are magnesium stearate and silicic acid. The capsules are carbohydrate based. RNA and the specific, purified nucleotides are the natural extracts of yeast. There are no yeast cells carried over from the extraction process into the product. The daily diaries contained space where participants confirmed that they had taken each capsule, and at what time of day.</p></sec><sec><title>Procedure</title><p>Recruitment was by a variety of sources. Notices asking for people with diagnosed IBS interested in taking part in a trial for a neutroceutical product were sent to people with IBS who had previously expressed an interest in IBS-related research; a request was included in one issue of Gut Reaction (a quarterly journal sent to people with IBS by The IBS Network, a British self-help organisation), and notices were placed in doctors' surgeries and libraries. Such a recruitment strategy was considered necessary, as strict inclusion and exclusion criteria, plus the burden of completing daily diaries for six months meant recruitment might be difficult. We wished to have a mixed group of people with IBS, representative of the general population of people with IBS. Recruiting from tertiary centres does not allow generalisation to the wider population of people with this condition.</p><p>Participants were invited to respond to these recruitment methods by telephone or e-mail. During this initial contact, the researcher ensured that potential participants met the inclusion criteria. Eligible individuals were then given full information relating to the study. They were advised at this stage of their right to withdraw their participation and/or any data already provided, at any time, without giving any notice or reason to the research team.</p><p>Participants were then asked to provide contact details for their GP, and a consent form was sent by post to the participant for them to complete and return, as well as a pack providing further information about the trial and the supplement. Their GP was also sent a letter at this point, summarising the trial and requesting confirmation of diagnosis. Upon receipt of the confirmation of diagnosis and participant consent forms, participants were randomised to each of the conditions.</p><p>Initially participants were sent their first symptom diary pack, which contained detailed instructions on how to complete the diary as well as the first diary itself. The diary required participants to rate the severity of their symptoms (see above) and to specify the times at which they had taken each of their three capsules; to name any medications which they were prescribed, had purchased (including price) and had actually taken (including time taken) that day; and to note any visits to health professionals made that day (including who they visited, reason for visit and advice given).</p><p>Participants were contacted by telephone or e-mail a week after this first pack was sent out (week 1), to confirm that they understood how to complete the questionnaires and the diary. Participants were sent their first set of capsules (a "set" comprised four sealed tubs) – all tubs were marked "nutritional supplement, contents 42 capsules (500 mg), two weeks supply", but with A or B clearly marked both on the label and on the cap) one week before they were to need them (week 3), with a letter stating the date on which they should begin taking the capsules. They were contacted by telephone or e-mail a few days after this date (week 5) to ensure that they were taking the capsules at the right dosage (one capsules three times a day, preferably with meals), that they were completing the symptom diary with the required information (time capsules taken, any missed capsules noted), and that they were aware of how to contact the research team if they had concerns about any effect which the capsules may have upon them.</p><p>Following this, participants were sent a symptom diary pack every month, their second set of capsules for week 17, a set of questionnaires (described below) at week 13 and week 25. All documentation and capsules were sent one week before they were required to ensure that they arrived in good time. Every letter sent to the participants included contact details for the research team and urged participants to telephone or e-mail with any questions or worries, at any stage of the trial. Participants were sent questionnaire packs at baseline, washout, and end of trial.</p><p>Every participant forgot to take at least one capsule across the duration of the trial, but this was usually one capsule only on any given day. There were two exceptions – one participant took capsules erratically over the last six weeks of the experimental condition and after the first five weeks of the placebo condition. This participant had a change of personal circumstances during this time (she underwent a hysterectomy). The other participant took no capsules for 5 days in the last week of the experimental condition as she went away and forgot to take the capsules with her.</p></sec><sec><title>Psychological measures</title><p>Measures were taken at the beginning of the baseline period, the end of the experimental condition and the end of the placebo condition.</p><p>Depression was measured by the CES-D [<xref ref-type="bibr" rid="B18">18</xref>] and anxiety was measured by the Stait-Trait Anxiety inventory [<xref ref-type="bibr" rid="B19">19</xref>]; a specific measure of health anxiety was provided by the Health Anxiety Questionnaire [<xref ref-type="bibr" rid="B20">20</xref>]. General health and happiness were measured by the total of the GHQ-60 [<xref ref-type="bibr" rid="B21">21</xref>] and the Affect Balance Scale [<xref ref-type="bibr" rid="B22">22</xref>]. The extent to which IBS intrudes into various aspects of everyday life was measured by the 13-item Illness Intrusiveness Rating Scale [<xref ref-type="bibr" rid="B23">23</xref>].</p></sec></sec><sec><title>Results</title><p>The average person with IBS spent approximately £15.00 during the six months trial on medications, supplements, minerals and vitamins and visited the G.P.'s surgery approximately seven times, the majority of the consultation times being for reasons other than IBS.</p><sec><title>Symptoms</title><p>For symptom recording, three participants had some sequential data missing, i.e. for the first participant this was for seven days in the experimental condition; for the second it was three days in the experimental condition, and for the third, three days during baseline. For each missing data point frequency data were obtained for the particular condition in which the missing data occurred, and the most appropriate (representative) measure of central tendency was inserted.</p></sec><sec><title>Analysis of symptoms</title><p>Mean ratings for the symptom series for each condition with 95% C.I.'s for the symptoms are represented below (see Figure <xref ref-type="fig" rid="F2">2</xref>).</p><fig position="float" id="F2"><label>Figure 2</label><caption><p>mean symptom severity ratings with 95% C.Is.</p></caption><graphic xlink:href="1475-2891-5-16-2"/></fig><p>A repeated measures ANOVA on each of the symptoms was carried out. Sphericity was not assumed and therefore the Greenhouse-Geisser correction for degrees of freedom was used. The difference between conditions for abdominal pain (F<sub>2,67 </sub>= 3.71; Eta<sup>2 </sup>= .10), urgency to have a bowel movement (F<sub>2,64 </sub>= 3.82; Eta<sup>2 </sup>= .10) and a feeling of incomplete evacuation (F = <sub>2,67 </sub>= 3.52; Eta<sup>2 </sup>= .09) were significant at p < .05. Diarrhoea (F<sub>2,58 </sub>= 3.08; Eta<sup>2 </sup>= .08), Flatulence (F<sub>2,70 </sub>= 2.89; Eta<sup>2 </sup>= .07), Bloating (F<sub>2,68 </sub>= 2.61; Eta<sup>2</sup>= .07) and Constipation (F<sub>2,49 </sub>= .31; Eta<sup>2 </sup>= .01) were not statistically significant at p < .05.</p><p>The percentage improvement of both placebo and IntestAidIB over baseline is shown below (see Figure <xref ref-type="fig" rid="F3">3</xref>).</p><fig position="float" id="F3"><label>Figure 3</label><caption><p>bar chart showing percentage change from baseline to placebo and experimental conditions.</p></caption><graphic xlink:href="1475-2891-5-16-3"/></fig><p>Symptom severity for all symptoms (except constipation) are in the expected direction of baseline>placebo>IntestAidIB.</p></sec><sec><title>Psychological measures</title><p>Repeated measures ANOVA were carried out on the psychological variables. There were no significant differences between conditions on these measures (p > .05).</p><p>It might be expected that due to the benefits of being in a clinical trial, anxiety and depression would decrease over time. We thus tested this by a repeated measures analysis for CES-D and Health Anxiety and Stait-Trait Anxiety comparing measures at baseline, washout and end of trial without considering condition (placebo or experimental). None of these measures were statistically significant (all p > .05)</p></sec></sec><sec><title>Discussion</title><p>The study has shown that there is a consistent improvement in most of the symptoms of irritable bowel syndrome following nucleotide supplementation with IntestAidIB. There was a very low drop-out rate (18%); none of the participants who completed the study reported any side-effects. A feeling of urgency to have a bowel movement and abdominal pain showed the most improvement over baseline and placebo, and since abdominal pain is the symptom most likely to prompt people to seek medical help, this is a important finding. A feeling of incomplete evacuation after a bowel movement also improved following treatment by the neutroceutical product. However, whilst statistically significant, improvements are modest, and this may be due to several factors. Firstly, people with IBS are not a homogenous group, and show great variability in frequency and severity of symptoms, both from each other, and individually. This makes it difficult to detect effects of interventions. Secondly, the placebo response was strong – Spiller [<xref ref-type="bibr" rid="B16">16</xref>] states that it takes approximately 12 weeks for the placebo effect to reduce. Perhaps a longer trial – and with a larger dose – effects might have been stronger. However, this is speculative and further trials need to take these factors into consideration.</p><p>Participants were a mixed group of people who have a current diagnosis of IBS, some of whom still attend a gastroenterological clinic, and some of whom do not. They were not obtained from tertiary centres and carried on with their normal life taking their usual medications. Although their symptoms were not perhaps as severe as tertiary patients, their symptoms were bothersome enough for them to enter the clinical trial, and the symptoms were severe enough for the participants to buy a range of products aimed at relieving their symptoms. We expect that the effects found in this simple would be stronger if replicated with participants from tertiary centres.</p><p>The benefits over placebo compare favourably with benefits found in some drug trials. For example, Tegaserod (a drug which acts as a selective agonist at 5HT receptors in the gastrointestinal tract) was found to produce 4.7% advantage over placebo in the participants' assessment of global relief of IBS [<xref ref-type="bibr" rid="B24">24</xref>]. Although many such drugs are well tolerated, there were no side effects at all reported in the present study, which is a considerable advantage. The percentage improvement varied according to the symptoms – between 4 and 6%. The symptom improvements shown are unlikely to be the result of a decrease in anxiety or depression which has sometimes been cited as a reason for any improvement in trials, as anxiety and depression in the present study did not decrease significantly as the trial progressed (irrespective of condition). The effect – at least at this dosage – does not seem strong enough to lead to an improvement in psychological state since depression, anxiety, general health and illness intrusiveness did not differ between the conditions.</p><p>The strong placebo effects found in the present study are similar to those in other IBS- studies [<xref ref-type="bibr" rid="B16">16</xref>]. We suggest that the percentage improvement over and above the placebo effect is a physiological effect of the nucleotide supplement on the gut. However, whilst results were consistent and some symptoms were statistically significant, these effects may not be strong enough to be perceived as a great improvement by the participants, who will try to assess the benefits of such a supplement against a background of extremely variable symptoms. The mechanism by which nucleotide supplementation might improve gut function could be via increased mucosal protein, DNA and villus height – as has been found in animal studies [<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B25">25</xref>] Evans et al [<xref ref-type="bibr" rid="B14">14</xref>] state that rodent models are useful in translational research to identify potential new treatments to increase gut mucosal growth that is potentially relevant to humans with short bowel syndrome (this is not related to IBS). They state "...an increase in the surface area that would correlate with the increased villus height and crypt depth may conceivably correspond to an increase in available nutrient transporters, which may translate into increased nutrient uptake" (Although people with IBS have not been shown to have damage to the gut, nucleotide supplementation may improve gut function nonetheless by such a mechanism. However, this is purely speculative, and the present study set out only to determine whether symptoms of IBS were improved with nucleotide supplementation. Further studies, preferably dose-dependent studies, will need to be carried out to determine the mechanism by which any improvements occur.</p></sec><sec><title>Conclusion</title><p>A neutroceutical product, IntestAidIB, was found to improve six of the measured seven symptoms of IBS compared to both baseline and placebo. However, only abdominal pain and urgency to have a bowel movement showed statistically significant effects at the p < .05 level. Although the improvements in symptoms were consistent, the effects were not strong, and psychological measures showed no improvement either as a result of the experimental condition, or due to the benefits of taking part of a clinical trial. Further studies need to replicate and extend these results, seeking to clarify the mechanism by which improvements occur.</p></sec><sec><title>Competing interests</title><p>This one-year study was funded by Wyreside Products Ltd, who provided us with information relating to the product and nucleotides, and the product itself. Wyreside Products Ltd were not involved in the design, running, analysis or write up of the study.</p><p>CPD and EA designed the study; KFB was employed as a research assistant and was in charge of the day-to-day running of the trial. All contributed to entering the data, data analyses, and writing the article.</p></sec>
Public sector nurses in Swaziland: can the downturn be reversed?	<sec><title>Background</title><p>The lack of human resources for health (HRH) is increasingly being recognized as a major bottleneck to scaling up antiretroviral treatment (ART), particularly in sub-Saharan Africa, whose societies and health systems are hardest hit by HIV/AIDS. In this case study of Swaziland, we describe the current HRH situation in the public sector. We identify major factors that contribute to the crisis, describe policy initiatives to tackle it and base on these a number of projections for the future. Finally, we suggest some areas for further research that may contribute to tackling the HRH crisis in Swaziland.</p></sec><sec sec-type="methods"><title>Methods</title><p>We visited Swaziland twice within 18 months in order to capture the HRH situation as well as the responses to it in 2004 and in 2005. Using semi-structured interviews with key informants and group interviews, we obtained qualitative and quantitative data on the HRH situation in the public and mission health sectors. We complemented this with an analysis of primary documents and a review of the available relevant reports and studies.</p></sec><sec><title>Results</title><p>The public health sector in Swaziland faces a serious shortage of health workers: 44% of posts for physicians, 19% of posts for nurses and 17% of nursing assistant posts were unfilled in 2004. We identified emigration and attrition due to HIV/AIDS as major factors depleting the health workforce. The annual training output of only 80 new nurses is not sufficient to compensate for these losses, and based on the situation in 2004 we estimated that the nursing workforce in the public sector would have been reduced by more than 40% by 2010. In 2005 we found that new initiatives by the Swazi government, such as the scale-up of ART, the introduction of retention measures to decrease emigration and the influx of foreign nurses could have the potential to improve the situation. A combination of such measures, together with the planned increase in the training capacity of the country's nursing schools, could even reverse the trend of a diminishing health workforce.</p></sec><sec><title>Conclusion</title><p>Emigration and attrition due to HIV/AIDS are undermining the health workforce in the public sector of Swaziland. Short-term and long-term measures for overcoming this HRH crisis have been initiated by the Swazi government and must be further supported and increased. Scaling up antiretroviral treatment (ART) and making it accessible and acceptable for the health workforce is of paramount importance for halting the attrition due to HIV/AIDS. To this end, we also recommend exploring ways to make ART delivery less labour-intensive. The production of nurses and nursing assistants must be urgently increased. Although the migration of HRH is a global issue requiring solutions at various levels, innovative in-country strategies for retaining staff must be further explored in order to stem as much as possible the emigration from Swaziland.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Kober</surname><given-names>Katharina</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Van Damme</surname><given-names>Wim</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	Human Resources for Health	<sec><title>Background</title><p>For a long time a rather neglected resource of health systems, the health workforce – or human resources for health (HRH) – has recently been receiving increased attention from the international health community. The HRH shortage is now being identified as one of the major challenges for improving health in low-income countries. The Joint Learning Initiative estimates that: "sub-Saharan countries must nearly triple their current numbers of workers by adding the equivalent of one million workers [...] if they are to come close to approaching the Millennium Development Goals for health" [<xref ref-type="bibr" rid="B1">1</xref>].</p><p>HIV/AIDS has increased the burden on existing health facilities and is increasingly becoming a major direct and indirect cause for health worker attrition. In many countries of sub-Saharan Africa, people with HIV-related illnesses occupy more than 50% of hospital beds and there is abundant evidence that health workers are overwhelmed by the demand for care [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B3">3</xref>]. At the same time, the lack of health workers in sub-Saharan Africa is regarded by many as the key bottleneck for scaling up antiretroviral treatment (ART) for the millions in need of it [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B5">5</xref>].</p><p>Swaziland, a small, lower middle-income country of just over one million inhabitants, is, like most other countries in southern Africa, facing a HRH crisis that is exacerbated by the impact of HIV/AIDS. With an adult HIV prevalence of approximately 42% in 2004, Swaziland is reckoned to have had around 220 000 people living with HIV/AIDS in 2003 and more than 36 000 requiring ART by the end of 2005 [<xref ref-type="bibr" rid="B6">6</xref>]. Swaziland's <italic>Health sector response to HIV/AIDS plan 2003–2005 </italic>identifies human resource shortages at all levels of the health sector as a serious constraint for scaling up ART. It identifies the problem of brain drain and the impact of HIV/AIDS as major factors contributing to the dire situation and calls for urgent measures to be taken to tackle the HRH crisis [<xref ref-type="bibr" rid="B7">7</xref>].</p><p>The present country case study of HRH is a more in-depth follow-up of the authors' previous exploration, in January 2004, of the impact of HIV/AIDS on health systems and the main issues related to scaling up ART in four countries in southern Africa [<xref ref-type="bibr" rid="B4">4</xref>]. Having identified the lack of HRH as the main bottleneck for scaling up ART in the region, we conducted a two-week rapid assessment study in Swaziland in June 2004 to describe and analyse the HRH situation, to make projections for the future based on the available data and to identify major factors contributing to the HRH bottlenecks for scaling up ART in Swaziland. We updated the 2004 study 18 months later, in December 2005, in order to illustrate the responses and their effects on the HRH situation in the country.</p></sec><sec sec-type="methods"><title>Methods</title><p>Our rapid-assessment study comprised two visits to Swaziland. During the first two weeks in June 2004 we collected both quantitative and qualitative data.</p><p>However, with no information system readily available in Swaziland that integrates information about numbers, deployment, educational levels and attrition of the health workforce, much of our quantitative information is from secondary sources, which probably results in a number of inaccuracies.</p><p>The follow-up visit, in December 2005, served the main purpose of finding out what steps had been taken to deal with the HRH situation. Due to the mentioned limitations of the quantitative data from 2004, we relied mainly on qualitative methods for the second purpose, instead of trying to quantify minor staff changes with the help of incomplete documents.</p><p>The HRH movements are described in a flow diagram (Figure <xref ref-type="fig" rid="F1">1</xref>), yet the absence of a comprehensive HRH information system did not allow an accurate quantification of the flow of health workers, whether between regions or sectors or internationally. Since it was not possible to obtain reliable data on employment in the private, for-profit sector, we focused our analysis on the public and mission sectors. We looked at three categories of staff – physicians, nurses and nursing assistants – and distinguished two inventory levels: current post establishment and actual employment. No further distinction was made between professional specializations within the three staff categories.</p><fig position="float" id="F1"><label>Figure 1</label><caption><p>Map of HRH flows in Swaziland.</p></caption><graphic xlink:href="1478-4491-4-13-1"/></fig><p>In order to obtain our data, we analysed a variety of primary sources, such as staff establishments and lists of medical students, as well as secondary sources, such as reports and studies of the HRH situation in the country [<xref ref-type="bibr" rid="B8">8</xref>]. We triangulated the written data by conducting semi-structured interviews with key informants from the Ministry of Health and Social Welfare (MOHSW) and the National Emergency Response Council on HIV/AIDS (NERCHA) and group interviews with management staff of health facilities, nursing schools and the nursing association.</p><p>Where no hard data were available, we worked with estimates based on this variety of sources. Thus, findings on the attrition of nurses are based on the information obtained during our interviews and discussions and on an extrapolation of population data to the health workforce. Since the results of an HIV/AIDS impact study on the health workforce in Swaziland, conducted in 2005, were not yet published at the time of our study, we based our attrition estimate on findings from other impact studies in a number of sub-Saharan countries, such as Botswana, Mozambique, South Africa and Zambia. The HIV prevalence and AIDS-related mortality among nurses in these countries have been estimated to be at least as high as in the general population [<xref ref-type="bibr" rid="B9">9</xref>-<xref ref-type="bibr" rid="B11">11</xref>]. We therefore supposed that the HIV prevalence among nurses in Swaziland is similar to the 42% among the general adult population recorded in the national 2004 antenatal clinic survey [<xref ref-type="bibr" rid="B12">12</xref>]. In the absence of ART, we assumed an approximate 10% annual mortality rate among the HIV-infected population. We used these fractions to project an attrition scenario among the nursing workforce.</p><p>Due to our rapid assessment approach, we could neither visit all health facilities nor interview all stakeholders, a fact that may have introduced bias in the sample visited and in the key informants selected.</p></sec><sec><title>Results</title><sec><title>The health care system, HRH, HIV/AIDS and ART scale-up: general findings in 2004 and 2005</title><p>Swaziland's health care system comprises public, private not-for-profit, private for-profit and industry-owned facilities. The majority of the private not-for-profit facilities are owned by missions but receive most of their subsidies from the Swazi government. The public and mission sectors operate six hospitals, eight public health units and five health centres offering both preventive and curative services, with between them a total of 1851 beds. Community-based care is offered in these sectors by 89 health clinics and 174 outreach clinics. In the private for-profit sector there are more than 50 clinics with and without beds; various industries own between them around 40 facilities, ranging from small health posts to clinics with more than 35 beds.</p><p>As Swaziland has no medical school, students must go abroad to obtain a medical degree. Nurses are trained at two sites in the country. One school is located within the compound of a mission hospital; the other was upgraded in 1997 to be part of the University of Swaziland. Basic nursing training lasts three years, but the majority of students do either one additional year for midwifery or a five-year course to obtain a Bachelor in Nursing Sciences degree. Some 125 registered nurses have been trained for a total of six years to become Family Nurse Practitioners. There is one school for nursing assistants in the country, where the training lasts two years.</p><p>At the time of our two visits, several ministries were involved in HRH-related decision-making. Overall responsibility for policy, management and planning for HRH in the public services lay with the Ministry of Public Works and Information. The Civil Service Board was in charge of technical recruitment matters; the Ministry of Education dealt with the pre-service training of all health workers. The MOHSW was responsible for delivering health services. The four Regional Management Committees of the MOHSW and the regional hospitals were supposed to submit their staffing plans to the central level of the MOHSW, which in turn made a request to the Civil Service Board.</p><p>Referring to a study commissioned by the MOHSW in 2000, WHO gives the total number of health workers in Swaziland as 3726, spread over 17 professional groups [<xref ref-type="bibr" rid="B8">8</xref>]. At the time of our visit in 2004, public and mission health facilities had 1481 posts and the actual number of staff reported to us as currently working in these facilities was 1184. The zero growth in the public sector over the past few years meant that spending on HRH could not be increased [<xref ref-type="bibr" rid="B8">8</xref>] despite the shortage of health workers confirmed by the Personnel Officer at the MOHSW. He reported requesting 200 additional health workers for the fiscal year 2003/2004 and being granted only two new medical staff. According to several informants from public and mission hospitals, the number of new posts has not increased since 1985, despite an increased burden of care.</p><p>The impact of HIV/AIDS on the health sector is most visible in the hospitals, where it is estimated that 80% of bed occupancy in the medical and paediatric wards is HIV/AIDS-related. Doctors from a mission hospital estimated that five years ago they would spend an average of five minutes per patient on a ward round, while presently this was more likely to be 20 to 30 minutes. This was seen as a consequence of the increasing number of terminally ill patients needing time-intensive care. There is an increased demand for health services, and health workers speak of feeling overwhelmed and burnt-out.</p><p>The government of Swaziland committed itself to providing ART to 12 000 people by the end of 2005, while WHO's "3 by 5" target would have been 16 000 [<xref ref-type="bibr" rid="B13">13</xref>]. The provision of ART in the public sector started in late 2001 in Mbabane hospital, and free-of-charge ART has been offered since November 2003. By June 2004 the programme had been extended to five public (government and mission) hospitals treating a total of nearly 3200 people. At the same time, two facilities in the private for-profit and commercial sectors were providing ART to more than 700 people.</p><p>By December 2005, ART was provided by all six public hospitals in the country, by five public health centres and by six facilities in the private sector. We were informed by the director of the Swazi National AIDS Programme that close to 12 000 people were alive and on ART by December 2005. According to the <italic>Second national multisectoral HIV and AIDS strategic plan 2006 – 2008</italic>, the government is planning to increase this number by approximately 13 000 new ART patients per year. The plan also states that staff shortages seriously compromise the effective delivery of ART to those in need. It estimates that 410 nurses and 247 nursing assistants were needed to support ART services in 2005 [<xref ref-type="bibr" rid="B6">6</xref>].</p></sec><sec><title>Physicians and nurses: findings and estimates in 2004</title><sec><title>Numbers and distribution</title><sec><title>Physicians</title><p>The doctor:population ratio, placed by WHO at 17.6 per 100 000, is based on the total of 182 doctors who were registered with the medical and dental council in Swaziland in 2004 [<xref ref-type="bibr" rid="B14">14</xref>]. The public and mission sectors have 90 posts, of which 50 were filled in June that year (Table <xref ref-type="table" rid="T1">1</xref>). Of the 92 remaining registered doctors, 80 were working in the private for-profit sector and 12 in the industry-owned sector. More than 70% of doctors in the government and mission sector were of non-Swazi nationality. For the private sector this figure was estimated at 60%. In the public sector, two urban hospitals with a total of 844 beds employed 34 doctors, while the remaining four rural hospitals, with 826 beds, employed 16 doctors.</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>HRH inventory of the public and mission health sectors in Swaziland, June 2004</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="center"><bold>Total registered (per 100 000 population)</bold></td><td align="center"><bold>Official posts in public sector</bold></td><td align="center"><bold>Actually working in public facilities (per 100 000 population)</bold></td><td align="center"><bold>Shortfall (in %)</bold></td></tr></thead><tbody><tr><td align="left">Physicians</td><td align="center">182 (17)</td><td align="center">90</td><td align="center">50 (5)</td><td align="center">40 (44%)</td></tr><tr><td align="left">Nurses</td><td align="center">3261 (296)</td><td align="center">944</td><td align="center">758 (69)</td><td align="center">183 (19%)</td></tr><tr><td align="left">Nursing Assistants</td><td align="center">700 (63)</td><td align="center">454</td><td align="center">376 (34)</td><td align="center">78 (17%)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left"><bold>Total</bold></td><td align="center"><bold>4143</bold></td><td align="center"><bold>1488</bold></td><td align="center"><bold>1184</bold></td><td align="center"><bold>301</bold></td></tr></tbody></table></table-wrap></sec><sec><title>Nurses and nursing assistants</title><p>In 2003 the nursing council had registered 3261 nurses, which translates into a nurse/population ratio of 296 per 100 000. Some 758 of the 944 established nursing posts in the public and mission sectors were filled by June 2004. The nurse/population ratio in the public and mission sectors is therefore around 70 per 100 000. The official number of nursing posts was widely regarded as inadequate for the actual workload in the health facilities.</p><p>The chief nursing officer at the MoH told us in December 2005 that often nursing staff is seconded from the hospitals to specific programmes or research projects. This practice makes it look as if all posts in the establishment were filled, while in fact the hospital wards are short of staff. In 2004 mission hospitals were generally better staffed, apparently because of their ability to employ non-Swazi nurses, a strategy that was not permitted in the government sector until July 2004. Due to the lack of data for the private sector, we could not clarify which proportion of the remaining 2317 registered nurses are employed in the private and commercial sectors. At the biggest industry-owned health centre we visited in 2004, the staffing ratio was more favourable than in the public sector, with 34 nurses, 3 doctors and 10 nursing assistants being responsible for 37 beds.</p><p>The total number of registered nursing assistants could not be established, but was estimated by the Swazi nursing association at around 700. Of 454 established posts in the public and mission sectors, 376 could be confirmed as filled in 2004 (Table <xref ref-type="table" rid="T1">1</xref>).</p></sec></sec><sec><title>Inflow, outflow and attrition</title><sec><title>Physicians</title><p>Most medical students go to South Africa, where the majority study at Medunsa University. In June 2004, a total of 72 students were following medical studies abroad; there are around 15 graduates each year. Students receive a government bursary which, theoretically, they are to pay back in the years after graduation. According to respondents from the MHOSW, only a minority of graduates return to Swaziland, and even fewer enter the public sector.</p><p>The outflow of Swazi doctors from the public service has long been recognized as problematic, as is shown by a 1996 report requested by the Parliament and submitted to the Principal Secretary of the MOHSW: <italic>Report on why medical doctors leave government employment for work without the Swazi public service </italic>[<xref ref-type="bibr" rid="B15">15</xref>]. Its findings include issues such as a lack of proper coordination and planning in the recruitment system of newly trained doctors, perceived preferential treatment of foreign doctors to Swazi doctors by the MOHSW and an inadequate career structure. Several of the physicians we interviewed described these issues as being still unresolved.</p></sec><sec><title>Nurses and nursing assistants</title><p>The two nursing schools receive between them around 500 annual applications; they are able to take 80 to 90 entrants. The school for nursing assistants has an annual output of 20 to 30. In June 2004 the directors and lecturers at both schools told us they were already working to full capacity and could not take more students. Insufficient student bursaries from the MOHSW, lack of teaching staff, lack of accommodation and insufficient possibilities for the practical stages of the training were given as the main reasons. According to the directors of the training sites, usually all students are recruited into public service soon after their graduation.</p><p>Hospital managers found it difficult to estimate the number of nurses who died or left the service because of terminal illness in 2003 and 2004. However, our respondents from the hospitals told us about high absentee rates among the staff because of their own illness, care commitments in the family or funerals. In one hospital it was reckoned that on average only half of all nurses could be counted on to be able to do their full duty at any one time. In the smallest hospital, with 28 nursing posts, seven nurses were thought to have died between January 2002 and June 2004. Management staff from the government hospital in Mbabane, with 110 filled nursing posts, estimated that each year three or four nurses had died over the past couple of years.</p><p>The most striking recent development observed by our informants in 2004 was clearly the extent of the emigration of nurses from Swaziland. The two major hospitals, with 203 and 125 established nursing posts, respectively, reckoned that each of them had lost between 25 and 35 nurses in 2003. In the smallest hospital in the country, a mission facility with 28 posts, 10 nurses had come to get their transcripts to apply for a position abroad in the first six months of 2004. The Department of Training at the MOHSW estimated that four to five nurses had been leaving the country weekly in the same period and the nursing association reported of a batch of 27 nurses who left for the United Kingdom in June 2004.</p><p>The extent of the nurses' desire to find work abroad is illustrated by the finding that there were at least four private personnel recruitment offices in Swaziland in June 2004. The manager of one of them told us that his office alone had recruited more than 30 nurses to the United Kingdom in the first two months after starting in the business. A nurse tutor, who had asked her students about the motivation for their career choice, told us that more than 50% of first-year students hoped it would give them the opportunity to go abroad.</p></sec></sec><sec><title>Projections 2004 to 2010</title><p>Swaziland's approximate 38% adult HIV prevalence made it possible to estimate that 288 of the 758 nurses employed in the public and mission sectors in 2004 could have been HIV-positive and that 29, or 10% of all HIV-positive nurses, might have died. We could therefore project that around 3% to 4% of the entire nursing workforce would die from AIDS annually. In the cases of the above hospitals with 203, 125 and 28 established nursing posts, this would have meant that 77, 48 and 11 nurses, respectively, might have been HIV-positive in 2004, and that the hospitals would lose a minimum of 8, 5 and 1 nurses each in that year. The impact of HIV/AIDS on the entire public health workforce would be considerable, with 20 to 30 deaths of nurses annually over the next five years.</p><p>This yearly loss of 3% to 4% of the public health workforce due to AIDS is alarming. Still, in the perception of our respondents the attrition of the health workforce due to HIV/AIDS seemed to pale in comparison with the exodus of nurses, which was believed to have grown to this massive scale only over the course of 2003. Hospital managers found it difficult to estimate the number of nurses who had died over the same period of time. Without doubt, emigration was a huge issue in 2004, and based on the information from our informants we made a conservative estimate that 100 nurses were leaving Swaziland every year.</p><p>Our 2004 projections show that both emigration and attrition due to HIV/AIDS pose a serious double threat for Swaziland's health system, which, if not effectively tackled, would mean the loss of more than 330, or 44%, of the nursing workforce in the public sector up to 2010 (Table <xref ref-type="table" rid="T2">2</xref>).</p><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Projections of the nursing workforce in the public and mission health sectors until 2010, based on findings in 2004</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="center"><bold>Mid-2004</bold></td><td align="center"><bold>Mid-2005</bold></td><td align="center"><bold>Mid-2006</bold></td><td align="center"><bold>Mid-2007</bold></td><td align="center"><bold>Mid-2008</bold></td><td align="center"><bold>Mid-2009</bold></td><td align="center"><bold>Mid-2010</bold></td></tr></thead><tbody><tr><td align="left">Total nurses in the public and mission sectors</td><td align="center">758</td><td align="center">694</td><td align="center">634</td><td align="center">577</td><td align="center">524</td><td align="center">473</td><td align="center">426</td></tr><tr><td align="left">Total inflow</td><td align="center">80</td><td align="center">80</td><td align="center">80</td><td align="center">80</td><td align="center">80</td><td align="center">80</td><td align="center">80</td></tr><tr><td align="left">Training output</td><td align="center">80</td><td align="center">80</td><td align="center">80</td><td align="center">80</td><td align="center">80</td><td align="center">80</td><td align="center">80</td></tr><tr><td align="left">Immigration</td><td align="center">0</td><td align="center">0</td><td align="center">0</td><td align="center">0</td><td align="center">0</td><td align="center">0</td><td align="center">0</td></tr><tr><td align="left">Total outflow</td><td align="center">144</td><td align="center">140</td><td align="center">137</td><td align="center">133</td><td align="center">130</td><td align="center">127</td><td align="center">125</td></tr><tr><td align="left">Normal retirement</td><td align="center">15</td><td align="center">14</td><td align="center">13</td><td align="center">12</td><td align="center">10</td><td align="center">9</td><td align="center">9</td></tr><tr><td align="left">Retirement/death due to AIDS</td><td align="center">29</td><td align="center">26</td><td align="center">24</td><td align="center">22</td><td align="center">20</td><td align="center">18</td><td align="center">16</td></tr><tr><td align="left">Emigration</td><td align="center">100</td><td align="center">100</td><td align="center">100</td><td align="center">100</td><td align="center">100</td><td align="center">100</td><td align="center">100</td></tr><tr><td align="left">Net loss per year</td><td align="center">64</td><td align="center">60</td><td align="center">57</td><td align="center">53</td><td align="center">50</td><td align="center">47</td><td align="center">45</td></tr></tbody></table><table-wrap-foot><p> Rate calculated at 2%</p><p>** Based on 38% HIV prevalence and estimating that 10% of HIV+ people have AIDS</p></table-wrap-foot></table-wrap></sec></sec><sec><title>HRH: findings and estimates, 2005</title><sec><title>Numbers and distribution</title><p>During our second visit in December 2005, we could not obtain an updated list of filled posts of either doctors or nurses in the public sector. The total number of doctors registered with the medical and dental council had fallen from 182 to 167. In 2004 as well as in 2005, informants from the health facilities and the central MoH criticized the establishment, saying that its adjustment to the increased workload was overdue.</p></sec><sec><title>HRH policy initiatives between 2004 and 2005</title><p>By June 2004, the shortage of health workers was acknowledged by all stakeholders in Swaziland as a serious problem, compromising not only the ART scale-up but also the quality of the general health services. However, the absence of a HRH monitoring system made it difficult to quantify the problem, let alone plan an adequate response based on anticipated developments in the HRH sector. Therefore, the development and subsequent implementation of a HRH policy and strategic plan, as initiated by the MoH with technical support from WHO, are important steps in response to the health workforce crisis. According to WHO, the HRH policy foresees an improved integration of HRH planning activities both between ministries and between the regional health management teams and the central level of the MoH.</p><p>According to information from WHO in December 2005, the government is also planning to simplify the current HRH policy and planning structures. The aim is to establish one body, the Health and Social Welfare Service Commission (HSWSC), that will be responsible for the recruitment, deployment, development, motivation, retention and discipline of all health care staff.</p></sec><sec><title>Policy initiatives and their effects to increase the inflow</title><sec><title>Influx of foreign staff</title><p>The lack of medical doctors in the public health system, particularly in view of the needs for the ART scale-up, has prompted NERCHA to include a request for 19 general doctors and 4 paediatric specialists in Swaziland's proposals to the Global Fund to Fight AIDS, TB and Malaria (GFATM) in rounds two and five. According to NERCHA, nine foreign doctors could be recruited with Global Fund money up to December 2005. The MoH has also negotiated with the Ministry of Public Works to lift the ban on the recruitment of foreign nurses into government service in July 2004. Between then and December 2005, 32 additional foreign nurses could be recruited for the government health sector with funds from the GFATM. This measure can of course be criticized on the grounds that, while the recruitment of foreign staff helps overcome staff shortages in Swaziland within a relatively short time, it aggravates the HRH situation in the neighbouring countries.</p></sec><sec><title>Increased training output</title><p>With regard to nursing training, WHO informed us of MoH plans to assess pre-service training needs and requirements, taking into account the high attrition rate among health workers. At the Faculty of Health Sciences we were told that they intend to double their intake of students and were waiting for MoH approval of their plans to expand the infrastructure and increase the number of tutors accordingly. According to the deputy nursing officer at the MoH, increasing the number of tutors or their salaries at the faculty is problematic because they are not employed by the MoH but by the faculty itself.</p></sec></sec><sec><title>Policy initiatives and their effects to decrease emigration</title><sec><title>Retention strategies</title><p>In order to counterbalance the losses due to emigration, the MoH has started to develop new retention strategies. In April 2005, the government raised the salaries of the civil servants by around 60%. Our interviews in December 2005 show that the salary rise may indeed help to keep nurses and doctors in the public sector, possibly even draw some back from the private for-profit sector. According to the MoH, there are also plans to improve retention with non-monetary motivation strategies for all health staff, such as better accommodation, child care facilities and easier access to car and housing loans for nurses.</p><p>Informants from the MoH and the Nursing Association told us that the salary increase in the government sector had substantially changed the situation of the internal market for health workers. Apparently it has become more attractive for nurses to work in the public sector instead of the private sector. The personnel manager of a private clinic in Mbabane confirmed this by saying the clinic could not afford salaries as high as the increased ones in the public sector, nor could it provide additional loan schemes. The only chance of keeping its nurses was to offer better working conditions than the public sector. According to the director of the Swazi Nursing Association, the private sector is now facing double competition: from abroad and from the higher salaries in the public service.</p><p>However, these observations did not conform to the comments from the doctors and nurses we talked with in the government hospitals. While people appreciated the salary increase, the main source of dissatisfaction in 2004 and 2005 remained the working conditions in the public sector. According to the nursing association, 21 nurses left Mbabane Hospital in the month of the salary increase.</p><p>Doctors and nurses felt overworked and complained to us about the inadequate number of nurses, the lack of essential material and equipment and the poor condition of the hospital infrastructure. Several doctors voiced their frustration about not being able to provide medical care of high quality under the existing conditions. Some also mentioned the lack of medical specialists in the country as unsatisfactory for a conscientious medical professional who sees many patients who would need to be referred. Nurses also mentioned again in 2005 their fear of infection with HIV because of the insufficient and poor-quality material and equipment in the public hospitals. We could not find out whether the retention plans of the MoH also included an improvement of the general working conditions in the public sector.</p><p>Still, in December 2005 the deputy director of nursing at the MoH provided us with a list of nurses who had left the government service since 2003, based on figures from the emigration office. The data in this list are neither complete nor 100% accurate, yet they may give an approximation of the real extent of emigration, which somehow differs from the estimations made by our informants. According to the emigration office, 91 nurses left government service in the 22 months between March 2003 and October 2005. Of the 84 who went abroad, 65 went to the United Kingdom and 19 destinations remain unconfirmed. The remaining seven left for various destinations within the country (private, retirement). Of the 65 nurses who departed for the United Kingdom, only five did so in 2005. All 19 nurses with unconfirmed destinations left before 2005.</p><p>These official numbers do not match those we obtained from the Director of the Swazi Nursing Association, according to whom the number of nurses leaving the country in 2005 was around six per month: roughly half the number in 2004. This estimation is based on the number of nurses who approach the Nursing Association for an official letter they are required to show to the emigration office. These figures, while leaving a large margin of uncertainty, point to a possible decrease in emigration, which could be an early effect of the retention measures introduced by the government.</p></sec><sec><title>ART scale-up to reduce attrition</title><p>There are currently no official estimates of the attrition rates due to illness or death in the public health sector. In December 2005 we were informed that an HIV/AIDS impact study had been conducted but that the draft version with the findings was not yet ready for distribution.</p><p>Regarding nurses' access to ART, it is possible that the proportion of nurses on treatment equals the proportion among the general HIV positive population on treatment. Yet, we learnt from our interviews that there is a specific Swazi dimension to the problem of health workers' access to ART.</p><p>As the country is very small, there is very little chance of anonymity in general, but even less so for health workers. The short distances and good infrastructure make it possible for most people to travel with relative ease and to receive ART in facilities away from home. Health workers, however, are quite a small professional group, in which it is hardly possible to obtain ART anonymously in any facility. Therefore it can be assumed that the proportion of nurses on ART lies below the proportion among the general population in need of ART. Still, it is very likely that the health workforce has also benefited from the progress of the general ART scale-up, with one third of the people in need of ART on treatment by the end of 2005.</p></sec></sec><sec><title>Projections 2005 to 2010</title><p>We use the described HRH policy initiatives as variables for establishing a number of future projections different from those based on our findings from 2004.</p><sec><title>Variable 1. ART scale-up</title><p>Starting from the latest estimate of an approximate 42% adult HIV prevalence, we calculate that 318 of the 758 nurses employed in the public and mission sectors may currently be HIV-positive. Assuming that the proportion of nurses on ART is roughly equivalent to the proportion of the general HIV-positive population on ART, around one third of those in need of ART would receive it. If the ART scale-up is maintained at that rate (i.e. one third of those in need annually on ART), with every other variable remaining unchanged, the loss of nurses up to 2010 would fall to 300 nurses, or 40% of the entire public nursing workforce (Table <xref ref-type="table" rid="T3">3</xref>).</p><table-wrap position="float" id="T3"><label>Table 3</label><caption><p>Projections of the nursing workforce in the public and mission health sectors in 2010, based on findings in 2005</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="left"><bold>Scenario mid-2010 based on 2004 estimates</bold></td><td align="left"><bold>ART scale-up</bold></td><td align="left"><bold>...plus reduced emigration</bold></td><td align="left"><bold>...plus increased foreign recruitment</bold></td><td align="left"><bold>...plus doubled training output</bold></td><td align="left"><bold>All measures combined</bold></td></tr></thead><tbody><tr><td align="left">Total nurses in the public and mission sectors by mid 2010</td><td align="center"><italic>426</italic></td><td align="center">474</td><td align="center">734</td><td align="center">618</td><td align="center">538</td><td align="center">974</td></tr><tr><td align="left">Total inflow</td><td align="center"><italic>560</italic></td><td align="center">480</td><td align="center">480</td><td align="center">660</td><td align="center">560</td><td align="center">740</td></tr><tr><td align="left">Training output</td><td align="center"><italic>560</italic></td><td align="center">480</td><td align="center">480</td><td align="center">480</td><td align="center">560</td><td align="center">560</td></tr><tr><td align="left">Immigration</td><td align="center"><italic>0</italic></td><td align="center">0</td><td align="center">0</td><td align="center">180</td><td align="center">0</td><td align="center">180</td></tr><tr><td align="left">Total outflow</td><td align="center"><italic>937</italic></td><td align="center">764</td><td align="center">504</td><td align="center">800</td><td align="center">780</td><td align="center">524</td></tr><tr><td align="left">Normal retirement¶</td><td align="center"><italic>82</italic></td><td align="center">77</td><td align="center">87</td><td align="center">84</td><td align="center">75</td><td align="center">95</td></tr><tr><td align="left">Retirement/death due to AIDS§</td><td align="center"><italic>155</italic></td><td align="center">87</td><td align="center">122</td><td align="center">117</td><td align="center">105</td><td align="center">134</td></tr><tr><td align="left">Emigration</td><td align="center"><italic>700</italic></td><td align="center">600</td><td align="center">295</td><td align="center">600</td><td align="center">600</td><td align="center">295</td></tr><tr><td align="left">Total loss between 2004 and 2010</td><td align="center"><italic>377</italic></td><td align="center">284</td><td align="center">24</td><td align="center">140</td><td align="center">220</td><td align="center">-216</td></tr></tbody></table><table-wrap-foot><p> Column taken from Table 2 (under the 2004 assumption that no HRH measures are taken)</p><p>¶ Rate calculated at 2%</p><p>§ Based on the assumption of 42% HIV prevalence; 10% of HIV+ people have AIDS and need ART; one third of those in need are put on ART (i.e. the current rate of ART scale-up)</p></table-wrap-foot></table-wrap></sec><sec><title>Variable 2. ART scale-up plus reduced emigration</title><p>If we accept the view of our Swazi informants that emigration had reached its peak in 2004 and has fallen since, we take the mean of the lowest (46) and highest (144) estimates of annual emigration before 2005 provided by the emigration office and the Nursing Association, respectively, and arrive at 95 nurses who left the country in 2004. The mean from the same sources for 2005 would be 39 nurses. Assuming that emigration will continue at around 40 nurses per year, our projection for 2010 shows a loss of 24, or 3%, of the total public nursing workforce (Table <xref ref-type="table" rid="T3">3</xref>).</p></sec><sec><title>Variable 3. ART scale-up plus import of foreign nurses</title><p>Having lifted the recruitment ban for foreign nurses, Swaziland could consider employing an increasing number of nurses from other African countries. Until now, the recruitment of foreign nurses seems to have been a one-off initiative financed with Global Fund money. Continuing with the employment of around 30 foreign nurses annually (as in 2004/2005), the total workforce would have lost 140, or 19%, of its nurses by 2010 (Table <xref ref-type="table" rid="T3">3</xref>).</p></sec><sec><title>Variable 4. ART scale-up plus increased training capacity</title><p>If the MoH can manage to double the training capacity of the nursing training facilities by mid-2006, the results in terms of nursing graduates cannot be expected before 2009. Thus the impact of this measure would be felt only in the last two years of our projections but would still reduce the loss to 220, or 29%, of nurses until 2010 (Table <xref ref-type="table" rid="T3">3</xref>). However, in the long run a doubled training output would slowly reverse the trend of annual losses, as from 2009 onwards the total annual inflow would exceed the annual outflow.</p></sec><sec><title>Variable 5. All measures combined</title><p>If the HRH policy in Swaziland combined all the measures discussed above, this would result in quite a spectacular reverse of the HRH depletion scenario as projected based on our observations in 2004. In 2010, the current baseline of 758 nurses would have increased by 216 nurses, or 28% (Table <xref ref-type="table" rid="T3">3</xref>).</p></sec></sec></sec></sec><sec><title>Conclusion</title><sec><title>Summary of findings</title><p>Our findings from June 2004 showed that the public health sector in Swaziland was losing its health workers at an alarming rate. Emigration and attrition due to HIV/AIDS were the major causes of these losses. Without new HRH policy initiatives it looked as if the public health sector could lose up to 44% of its entire nursing workforce by 2010. During our second visit, in 2005, we realized that measures had indeed been taken to overcome the HRH crisis, and we could observe some possible early effects of such measures that allowed us to develop more optimistic future scenarios.</p><p>Each of the four HRH policy measures being discussed or already implemented by the government of Swaziland would slightly reduce the losses of the health workforce. The reduction of emigration is the measure with potentially the biggest impact. However, it is the combination of all measures in a comprehensive HRH policy that would result in a spectacular reversal of the trend and lead to growth of the health workforce in Swaziland.</p></sec><sec><title>Limitations of findings</title><p>The absence of a comprehensive HRH information system in Swaziland made it necessary to work with estimates based on a variety of sometimes-contradictory sources. Therefore our projections depart from baseline calculations that may themselves provide a slightly inaccurate quantification of the actual situation of the health workforce in Swaziland.</p><p>We are also aware that our projections may not capture all potential effects of HRH developments and policies, nor do they include a quantification of the workload and possible future changes to it. Still, they provide us with a rough but very illustrative idea of future developments that may result from present actions taken or not taken.</p><p>Our rapid assessment approach did not allow us to visit all health facilities or interview all stakeholders. This may have introduced bias in the sample visited and in the key informants selected.</p></sec><sec><title>Implications of findings</title><p>We argue that the trend of a diminishing nursing workforce in Swaziland can be reversed if as many as possible of the planned policy initiatives are swiftly and jointly implemented.</p><p>We would further stress the particular importance of implementing as soon as possible the planned extension of training capacity of the country's nursing schools. Depending on the scale of the extension, such an initiative could substantially increase the number of nurses available for public health service in Swaziland in the medium and long term. With more than 500 nursing applicants per year, the lack of accommodation, grants and tutors should be overcome as obstacles for accepting more students into nursing school. Yet, training more nurses alone may not necessarily improve the HRH situation in the Swazi health system as long as so many of the new students make their career choice with the intention of working abroad.</p><p>Therefore, there is a need for qualitative research into the right mix of retention measures, beyond financial incentives such as salary increases. To our knowledge, no comprehensive job satisfaction survey has been conducted so far in the Swazi health sector. Some retention measures, such as easier access to different kinds of loans, seem to be much appreciated by nurses, but other strategies being discussed at the MoH, such as child care facilities and subsidized housing, should also be further investigated and implemented as swiftly as possible if deemed useful.</p><p>Yet, in our interviews with nurses and doctors, the actual working conditions in the public sector figured more prominently than the issue of salaries and benefit packages. The importance of good working conditions is illustrated by the example of a private clinic in Mbabane with 32 beds and 20 full-time nurses. This clinic pays the same salary as the government and does not offer any extra loan or pension schemes, yet has not lost any of its staff to emigration in the past couple of years. The clinic's personnel manager attributes this to some features that distinguish the clinic from public sector hospitals, such as a lower workload, a different shift system, many training opportunities and very close tutoring and guidance on good care practices.</p><p>While we assume that the ART scale-up in Swaziland is also reducing attrition of the health workforce due to HIV/AIDS, there seem still to be problems of access due to the perceived lack of anonymity for health workers when seeking treatment. While the solution will lie ultimately in reducing the stigma around HIV/AIDS, we would recommend exploring other, more intermediate solutions to the problem of health workers' access to ART.</p><p>The lack of HRH in Swaziland is widely regarded as the main bottleneck to scaling up ART and maintaining thousands of patients on ART over time. Therefore we would also recommend intensified research into the potential of making the current ART delivery model less HRH-intensive. Several studies in other sub-Saharan African countries with severe HRH shortages have drawn attention to the need for context-specific ART delivery models that require considerably less time from doctors and nurses [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B5">5</xref>,<xref ref-type="bibr" rid="B16">16</xref>].</p><p>Some of our respondents at the MoH and NERCHA regard the stronger involvement of PLWHA as crucial for the successful scale-up of ART in Swaziland. Yet up to now PLWHA only assist with logistics and patient flow in a few ART clinics, an activity for which they are paid a small incentive from Global Fund money. A 2004 directory lists almost 50 PLWHA associations in Swaziland [<xref ref-type="bibr" rid="B17">17</xref>] and we know that more than 11 000 people are on ART. We would therefore suggest exploring in more depth the potential and the capacity-strengthening needs of these groups and people on treatment and identify areas for the involvement of PLHA beyond mere support functions but for tasks up to now reserved for doctors and nurses [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B19">19</xref>].</p><p>However, while increasing the production of health workers, improving the retention strategies, fighting the stigma of AIDS and adjusting ART delivery models are necessary and important measures for dealing with the HRH crisis, emigration of HRH remains a complex issue that cannot be tackled at the national level alone [<xref ref-type="bibr" rid="B20">20</xref>,<xref ref-type="bibr" rid="B21">21</xref>]. It is unlikely that the problem of emigration can be tackled in the long term without addressing the structural and political factors that affect HRH at national and global levels [<xref ref-type="bibr" rid="B20">20</xref>,<xref ref-type="bibr" rid="B22">22</xref>-<xref ref-type="bibr" rid="B24">24</xref>].</p></sec></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>Katharina Kober and Wim Van Damme designed the study. KK conducted the interviews in Swaziland, reviewed the available documentation of HRH in Swaziland and wrote the initial draft of this article. Both authors analysed the obtained data and WvD revised the subsequent drafts of this article.</p></sec>
Identification of multi-drug resistant <italic>Pseudomonas aeruginosa </italic>clinical isolates that are highly disruptive to the intestinal epithelial barrier	<sec><title>Background</title><p>Multi-drug resistant <italic>Pseudomonas aeruginosa </italic>nosocomial infections are increasingly recognized worldwide. In this study, we focused on the virulence of multi-drug resistant clinical strains <italic>P. aeruginosa </italic>against the intestinal epithelial barrier, since <italic>P. aeruginosa </italic>can cause lethal sepsis from within the intestinal tract of critically ill and immuno-compromised patients via mechanisms involving disruption of epithelial barrier function.</p></sec><sec sec-type="methods"><title>Methods</title><p>We screened consecutively isolated multi-drug resistant <italic>P. aeruginosa </italic>clinical strains for their ability to disrupt the integrity of human cultured intestinal epithelial cells (Caco-2) and correlated these finding to related virulence phenotypes such as adhesiveness, motility, biofilm formation, and cytotoxicity.</p></sec><sec><title>Results</title><p>Results demonstrated that the majority of the multi-drug resistant <italic>P. aeruginosa </italic>clinical strains were attenuated in their ability to disrupt the barrier function of cultured intestinal epithelial cells. Three distinct genotypes were found that displayed an extreme epithelial barrier-disrupting phenotype. These strains were characterized and found to harbor the <italic>exoU </italic>gene and to display high swimming motility and adhesiveness.</p></sec><sec><title>Conclusion</title><p>These data suggest that detailed phenotypic analysis of the behavior of multi-drug resistant <italic>P. aeruginosa </italic>against the intestinal epithelium has the potential to identify strains most likely to place patients at risk for lethal gut-derived sepsis. Surveillance of colonizing strains of <italic>P. aeruginosa </italic>in critically ill patients beyond antibiotic sensitivity is warranted.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Zaborina</surname><given-names>Olga</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Kohler</surname><given-names>Jonathan E</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Wang</surname><given-names>Yingmin</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Bethel</surname><given-names>Cindy</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Shevchenko</surname><given-names>Olga</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Wu</surname><given-names>Licheng</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A7" equal-contrib="yes" contrib-type="author"><name><surname>Turner</surname><given-names>Jerrold R</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A8" equal-contrib="yes" corresp="yes" contrib-type="author"><name><surname>Alverdy</surname><given-names>John C</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	Annals of Clinical Microbiology and Antimicrobials	<sec><title>Background</title><p>The human opportunistic pathogen, <italic>Pseudomonas aeruginosa</italic>, is a major cause of infectious-related mortality among the critically ill patients, and carriers the highest case fatality rate of all gram-negative infections [<xref ref-type="bibr" rid="B1">1</xref>]. Although the lungs have been traditionally considered to be a major site of <italic>P. aeruginosa </italic>infection among critically ill patients, a significant number of these infections arise as a result of direct contamination of the airways by the gastrointestinal flora or by hematogenous dissemination from the intestine to the lung parenchyma [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B3">3</xref>]. Yet even in the absence of established extraintestinal infection and bacteremia, the presence of highly virulent strains of <italic>P. aeruginosa </italic>within the intestinal tract alone can be a major source of systemic sepsis and death among immuno-compromised patients [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B5">5</xref>]. Extensive studies on the endemicity and prevalence of <italic>P. aeruginosa </italic>in the critically ill patients have identified the intestinal tract to be the single most important reservoir for this pathogen in cases of severe life-threatening sepsis [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B7">7</xref>]. Work from our laboratory has demonstrated that a major mechanism of the lethal effect of intestinal <italic>P. aeruginosa </italic>lies in its ability to adhere to and disrupt the intestinal epithelial barrier [<xref ref-type="bibr" rid="B8">8</xref>].</p><p>Within as little as 3 days in an intensive care unit, the feces of more than 50% of patients will culture positive for <italic>P. aeruginosa </italic>with up to 30% of these strains being antibiotic resistant [<xref ref-type="bibr" rid="B6">6</xref>]. In such patients, intestinal colonization by <italic>P. aeruginosa </italic>alone has been associated with a 3-fold increase in mortality in critically ill patients [<xref ref-type="bibr" rid="B4">4</xref>]. In fact the importance of intestinal <italic>P. aeruginosa </italic>as a cause of mortality in critically ill patients was recently demonstrated by a randomized prospective study in which selective antibiotic decontamination of the digestive tract (SDD) in critically ill patients with oral non-absorbable antibiotics decreased mortality associated with a decrease in fecal <italic>P. aeruginosa </italic>[<xref ref-type="bibr" rid="B9">9</xref>].</p><p>How multi-drug resistant (MDR) <italic>P. aeruginosa </italic>clinical isolates behave against the human intestinal epithelium is unknown. Therefore the purpose of this study was to determine the ability of MDR <italic>P. aeruginosa </italic>to disrupt epithelial integrity of Caco-2 monolayers and to correlate these findings to other relevant virulence features of <italic>P. aeruginosa </italic>including adhesiveness, motility, ability to form biofilm, and the presence of specific type III secretion related genes <italic>exoU </italic>and <italic>exoS</italic>.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Bacterial isolates</title><p>Under IRB protocol #11646B, University of Chicago, 35 strains of <italic>P. aeruginosa </italic>were consecutively obtained from the clinical microbiology laboratory from those selectively screened for gentamicin (Gm) resistance. We initially screened consecutive <italic>P. aeruginosa </italic>isolates that were resistant to Gm since Gm resistance has been shown to be the most common feature of MDR <italic>P. aeruginosa </italic>[<xref ref-type="bibr" rid="B10">10</xref>]. Among the 35 strains, three (# 3, 5, and 32) lost their resistance to Gm and one (#24) was re-identified not to be <italic>P. aeruginosa </italic>on subsequent culture. Therefore 31 clinical strains were available for phenotype and genotype analysis. Most isolates identified as <italic>P. aeruginosa </italic>were oxidase positive, hydrolyzed acetamide and arginine, oxidized glucose, and grew on cetrimide agar. Remaining isolates were identified by the Vitek 2 system (bioMérieux, Inc. Durham, NC). Additionally, isolates were verified by amplification of 16S DNA using primers forward 5'-GGACGGGTGAGTAATGCCTA-3' and reverse 5'-CGTAAGGGCCATGATGACTT-3', and genome DNAs of clinical isolates as templates. Susceptibility testing was performed by testing on the Vitek 2 or by disk diffusion. Susceptibility results were interpreted using Clinical Laboratory Standards Institute (CLSI) guidelines. Single colonies were picked up from Columbia SB agarized plates (Beckton Dickinson, Cockeysville, MD), grown in <italic>Pseudomonas </italic>broth containing Gm, 50 μg.ml<sup>-1 </sup>and kept at -80°C as frozen stocks containing 8% glycerol. The isolates were routinely subcultured from frozen stocks on <italic>Pseudomonas </italic>isolation agar (PIA) containing Gm, 50 μg.ml<sup>-1</sup>. <italic>P. aeruginosa </italic>strains PAOI, ATCC 27853, PA103, and the environmental isolates PA190 and PA180 [<xref ref-type="bibr" rid="B11">11</xref>-<xref ref-type="bibr" rid="B13">13</xref>] were used as reference strains.</p></sec><sec><title>DNA fingerprint analysis</title><p>The clonality of <italic>P. aeruginosa </italic>isolates was determined using the random amplified polymorphic DNA (RAPD) PCR fingerprinting, described previously [<xref ref-type="bibr" rid="B14">14</xref>-<xref ref-type="bibr" rid="B16">16</xref>]. Primers 208 (5'-ACGGCCGACC-3') and 272 (5'-AGCGGGCCAA-3') were synthesized and used in PCR amplifications. Intact bacteria were used as a source of template chromosomal DNA. The following protocol was used: 45 cycles of 1 min at 94°C, 1 min at 45°C and 1 min at 72°C. After the last cycle, samples were maintained at 72°C for 10 min. The resulting amplified DNA fragments were separated on agarose gels (0.8%, w/v) containing ethidium bromide (0.5 μg.ml<sup>-1</sup>) and visualized using UV radiation. Fingerprints were considered distinct if they differed by at least three bands.</p></sec><sec><title>Human epithelial cells and transepithelial resistance (TER) assay</title><p>The Caco-2bbe (brush border-expressing) cell line was used in bacterial-cell culture experiments. Caco-2 cells were grown in 0.3 cm<sup>2 </sup>transwells (Costar) in HEPES buffered (15 mM) DMEM media containing 10% FBS for 20 days, and electrophysiological measurements were done using agar bridges and Ag-AgCl-calomel electrodes and a voltage clamp (University of Iowa Bioengineering, Iowa City, IA) as previously described [<xref ref-type="bibr" rid="B17">17</xref>]. Fixed currents of 50 μA were passed across Caco-2 monolayers, and transepithelial resistance (TER) was calculated using Ohm's law. Fluid resistance was subtracted from all values. In order to assess the disrupting ability of <italic>P. aeruginosa </italic>strains against Caco-2 monolayers, overnight culture was added to the apical well (volume = 200 μl) to achieve a final bacterial concentration of ~10<sup>7 </sup>CFU/ml. Media from the apical wells was then quantitatively cultured on PIA plates to determine the final bacterial count. Caco-2 monolayers were co-incubated with bacteria for up to 8 hours at 37°C, 5% CO<sub>2</sub>, and TER was measured each hour. All experiments were performed in triplicate.</p></sec><sec><title>Swimming motility</title><p>Swimming assay was performed as previously described by Rashid and Kornberg [<xref ref-type="bibr" rid="B18">18</xref>]. Briefly, swim plates prepared by using of 1% tryptone, 0.5% NaCl and 0.3% (wt/vol) agarose, were inoculated with bacteria using a sterile toothpick. The plates were then wrapped to prevent dehydration and incubated at 37°C, overnight. The ability to swim was assessed by the radius of colony. All experiments were performed in triplicate.</p></sec><sec><title>Twitching motility</title><p>Twitching motility was determined by the method of Rashid and Kornberg [<xref ref-type="bibr" rid="B18">18</xref>]. Fresh prepared and briefly dried twitch plates (Tryptic soy broth solidified with 1% (wt/vol) Difco granulated agar) were stab inoculated with a sharp toothpick into the bottom of the Petri dish. After incubation at 37°C for 24 h, the halo zone of growth at the interface between the agar and the polystyrene surface was measured. All motility experiments were performed in triplicate.</p></sec><sec><title>Ability to form biofilm</title><p>Biofilm formation was assayed as described with modifications [<xref ref-type="bibr" rid="B19">19</xref>]. Briefly, <italic>P. aeruginosa </italic>strains were grown in 96-well plates in M63 supplemented with 0.5% casamino acids and 0.2% glucose. Plates were incubated at 37°C under mild shaking at 50 rpm (C24 Incubator Shaker, New Brunswick Scientific, Edison, NJ) for 8 hrs. The wells were then rinsed thoroughly with water and the attached material was stained with 0.1% crystal violet, washed with water, and solubilized in ethanol. Solubilized fractions were collected and absorbance measured at 550 nm with a Plate Reader. All experiments were performed in triplicate.</p></sec><sec><title>Adhesiveness</title><p>Caco-2 cells were grown to confluence in 24-well plates using HEPES-buffered DMEM media containing 10% fetal bovine serum. Overnight cultures of <italic>P. aeruginosa </italic>were added to the apical side of Caco-2 cells to a final concentration of 10<sup>7 </sup>CFU/ml and co-incubated for 1 hour at 37°C, 5% CO<sub>2</sub>. Following the one hour incubation, the media was removed and ten-fold dilutions were plated on PIA plates to quantify non-adherent bacteria. Wells were then washed with a continuous flow of 35 ml of PBS. A final single washing with 200 μl was diluted and plated on PIA to quantify the final amount of remaining non-adherent bacteria. Caco-2 cells were then trypsinized with 200 μl Trypsin-EDTA (Gibco), incubated for 20 min at 37°C, 5% CO<sub>2</sub>, and lysed with 400 μl of a lysis mixture (PBS, EDTA 10 mM, Triton X-100 0.25%) [<xref ref-type="bibr" rid="B20">20</xref>] added directly to the trypsinized Caco-2 cells. The cells were vigorously pipetted for one minute, and released bacteria were plated on PIA to quantify adherent cells. The proportion of bacterial cells adhering to Caco-2 cells was calculated as (adherent cells - cells in last washing)/non-adherent + adherent cells. All experiments were performed in triplicate.</p></sec><sec><title>Effect of exposure of MDR <italic>P. aeruginosa </italic>clinical isolates to Gm on growth rate</title><p>Overnight culture of <italic>P. aeruginosa </italic>clinical isolate #1 was diluted as 1:100 in fresh M63 media supplemented with 0.5% casamino acids and 0.2% glucose and grown for 2 hours. After that, culture was spitted for control (no Gm) and Gm-variant that was added by Gm to a desirable concentration. 300 μl aliquots (in triplicates) were loaded in 96-well plate, and absorbance at OD550 nm was measured dynamically during growth at 37°C, 200 rpm. All experiments were performed in triplicate.</p></sec><sec><title>The <italic>exoU </italic>and <italic>exoS </italic>gene detection by PCR</title><p>PCR assays for detection of the <italic>exoU </italic>and <italic>exoS </italic>genes were performed using intact <italic>P. aeruginosa </italic>grown on PIA as a source of template chromosomal DNA as described [<xref ref-type="bibr" rid="B16">16</xref>]. Amplification was performed in the presence of primers for <italic>exoU</italic>: <italic>exoU</italic>2998, 5'-GCTAAGGCTTGGCGGAATA-3' and <italic>exoU</italic>3182, 5'-AGATCACACCCAGCGGTAAC-3'; for <italic>exoS</italic>: <italic>exoS </italic>1106, 5'-ATGTCAGCGGGATATCGAAC-3', and <italic>exoS </italic>1335, 5'-CAGGCGTACATCCTGTTCCT-3'.</p></sec><sec><title>Cytotoxicity assay</title><p>Caco-2 cells were grown to confluence in 96-well plates, and inoculated apically by <italic>P. aeruginosa </italic>to the final concentration of 10<sup>7 </sup>CFU/ml. Cells were incubated at 37°C, 5% CO<sub>2</sub>, for 8 hours, and released lactate dehydrogenase was determined by CytoTox 96 assay (Promega). All experiments were performed in triplicate.</p></sec><sec><title>Statistical analysis</title><p>Statistical analysis of the data was performed using Student t-test. Regression analysis was performed using Sigmaplot software.</p></sec></sec><sec><title>Results</title><sec><title>Morphological and demographic analyses of MDR <italic>P. aeruginosa </italic>clinical isolates</title><p>Morphological and demographic data are displayed in Table <xref ref-type="table" rid="T1">1</xref>. <italic>P. aeruginosa </italic>strains were consecutively collected based on their resistance to gentamicin (Gm), however most clinical isolates displayed multiple antibiotic resistances to various antibiotics clinical used against <italic>P. aeruginosa</italic>. Most strains were obtained from sputum and tracheal aspirates while few were from tissues and urine. Significant variation was noted in colony morphology among the various strains. Environmental strains PA190 and PA180 were also tested for antibiotic resistance. Results indicated that PA190 was sensitive to all of the antibiotics routinely used for <italic>P. aeruginosa </italic>infection, whereas PA180 was resistant to Gm.</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Demographic and morphological data of MDR <italic>P. aeruginosa </italic>isolates</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left">##</td><td align="left">Morphology of colony on PIA</td><td align="left">Antibiotic resistance <sup>a</sup></td><td align="left">Source</td><td align="left">Patient location</td></tr></thead><tbody><tr><td align="left">1</td><td align="left">Yellow, smooth, flat edge</td><td align="left">IMI 11, Ptaz14, Cefr 16, Ctaz 17, Gm 6, Tobr 6, Amik 18, Cipr 6 [b]</td><td align="left">DN<sup>c</sup></td><td align="left">DN<sup>c</sup></td></tr><tr><td align="left">2</td><td align="left">Green, smooth, flat edge</td><td align="left">Tobr 16, Cipr 4, Gm 16, Ptaz 128 [d]</td><td align="left">Sputum</td><td align="left">ICU</td></tr><tr><td align="left">4</td><td align="left">Slightly green, rough edge</td><td align="left">IMI 16, Ctaz 64, Gm 16, Ptaz 128 [d]</td><td align="left">Tracheal aspirate</td><td align="left">ICU</td></tr><tr><td align="left">6</td><td align="left">Bright greenish-blue, smooth, flat edge</td><td align="left">Gm 16, Ptaz 128, Tobr 16 [d]</td><td align="left">Tracheal aspirate</td><td align="left">Burn ICU</td></tr><tr><td align="left">7</td><td align="left">Green, smooth, flat edge</td><td align="left">Gm 16, Cipr 4, Tobr 16 [d]</td><td align="left">Wound</td><td align="left">Floor</td></tr><tr><td align="left">8</td><td align="left">Green, rough edge</td><td align="left">IMI 21, Ptaz 28, Cefr 24, Ctaz 27, Gm 6, Tobr 9, Amik 6, Cipr 26 [b]</td><td align="left">Maxillary sinus</td><td align="left">ENT clinic</td></tr><tr><td align="left">9</td><td align="left">Greenish-blue, slightly roug,</td><td align="left">Gm 16, Cipr 4, Tobr 16, Ptaz 128, Levo 8 [d]</td><td align="left">Clean void urine</td><td align="left">Floor</td></tr><tr><td align="left">10</td><td align="left">White, smooth, flat edge, mucoid</td><td align="left">Gm 16, Cipr 4, Tobr 16 [d]</td><td align="left">Sputum</td><td align="left">Burn ICU</td></tr><tr><td align="left">11</td><td align="left">Slightly green, flat edge</td><td align="left">Gm 10, Amik 13, Tobr 11, IMI 6, Ptaz 14 [b]</td><td align="left">Sputum, CFRC</td><td align="left">Pulmonary</td></tr><tr><td align="left">12</td><td align="left">Green, rough, nonflat edge</td><td align="left">Gm 11 [b]</td><td align="left">Sputum, CFRC</td><td align="left">Floor</td></tr><tr><td align="left">13</td><td align="left">Bright yellow, smooth, flat edge</td><td align="left">Gm 16, Cipr 4, Tobr 16 [d]</td><td align="left">Catheter tip</td><td align="left">Floor</td></tr><tr><td align="left">14</td><td align="left">Bright yellow, smooth, flat edge</td><td align="left">Gm 16, Cipr 4, Tobr 16 [d]</td><td align="left">Catheter tip</td><td align="left">Floor</td></tr><tr><td align="left">15</td><td align="left">Green, slightly rough, nonflat edge</td><td align="left">Gm 16, Cipr 4, Tobr 16, Ptaz 128, Levo 8, Amik 64, Ctaz 64, IMI 16 [d]</td><td align="left">Urine</td><td align="left">Nursing home</td></tr><tr><td align="left">16</td><td align="left">Green, slightly rough, nonflat edge</td><td align="left">Gm 6, Cipr 6, Tobr 10, Ptaz 17, Amik 6, Ctaz 10 [b]</td><td align="left">Sputum, CFRC</td><td align="left">Pulmonary</td></tr><tr><td align="left">17</td><td align="left">White, mucoid</td><td align="left">Gm 8, Cipr 15, Ptaz 15, Amik 8 [b]</td><td align="left">Sputum, CFRC</td><td align="left">Pulmonary</td></tr><tr><td align="left">18</td><td align="left">Yellow, smooth, flat edge</td><td align="left">Gm 16, Tobr 16 [d]</td><td align="left">Catheter tip</td><td align="left">Burn ICU</td></tr><tr><td align="left">19</td><td align="left">Slightly green, smooth, flat edge</td><td align="left">IMI 23, Ptaz 35, Cefr 24, Ctaz 30, Gm 9, Tobr 15, Amik 11 [b]</td><td align="left">ET tube</td><td align="left">Burn ICU</td></tr><tr><td align="left">20</td><td align="left">Slightly green, smooth, flat edge</td><td align="left">IMI 8, Ptaz 21, Cefr 20, Ctaz 22, Gm 12, Tobr 17, Amik 17 [b]</td><td align="left">Tracheal aspirate</td><td align="left">ICU</td></tr><tr><td align="left">21</td><td align="left">Slightly green, smooth, flat edge</td><td align="left">Gm 16, IMI 16 [d]</td><td align="left">Tracheal aspirate</td><td align="left">ICU</td></tr><tr><td align="left">22</td><td align="left">Green, smooth, flat edge</td><td align="left">Gm16 [d]</td><td align="left">Wound</td><td align="left">Floor</td></tr><tr><td align="left">23</td><td align="left">Slightly green, smooth, flat edge</td><td align="left">Gm16 [d]</td><td align="left">Tracheal aspirate</td><td align="left">Burn ICU</td></tr><tr><td align="left">25</td><td align="left">Rough, nonflat edge, slightly green</td><td align="left">Ctaz 64, IMI 16, Gm 16 [d]</td><td align="left">Tracheal aspirate</td><td align="left">ICU</td></tr><tr><td align="left">26</td><td align="left">Rough, nonflat edge, slightly green</td><td align="left">Ctaz 64, IMI 16, Gm 16 [d]</td><td align="left">Tracheal aspirate</td><td align="left">ICU</td></tr><tr><td align="left">27</td><td align="left">Rough, nonflat edge, slightly green</td><td align="left">Ctaz 64, IMI 16, Gm 16, Ptaz 128 [d]</td><td align="left">Urine</td><td align="left">ICU</td></tr><tr><td align="left">28</td><td align="left">Rough, nonflat edge, slightly green</td><td align="left">Ctaz 64, IMI 16, Gm 16 [d]</td><td align="left">Foley catheter urine</td><td align="left">ICU</td></tr><tr><td align="left">29</td><td align="left">Green, slightly rough, nonflat edge</td><td align="left">Amik 6, Tobr 12, Gm 6 [b]</td><td align="left">Sputum, CFRC</td><td align="left">Pulmonary</td></tr><tr><td align="left">30</td><td align="left">Pink, smooth, flat edge, mucoid</td><td align="left">IMI 29, Ptaz 22, Cefr 6, Ctaz 24, Gm 6, Tobr 6, Amik 6, Cipr 19 [b]</td><td align="left">Sputum, CFRC</td><td align="left">Pulmonary</td></tr><tr><td align="left">31</td><td align="left">Pink, smooth, flat edge, mucoid</td><td align="left">IMI 27, Ptaz 26, Cefr 19, Ctaz 27, Gm 6, Tobr 6, Amik 6, Cipr 26 [b]</td><td align="left">Sputum, CFRC</td><td align="left">Pulmonary</td></tr><tr><td align="left">33</td><td align="left">Slightly green, smooth, flat edge</td><td align="left">Amik 13, IMI 6, Gm 9, Cipr 6 [b]</td><td align="left">Clean void urine</td><td align="left">Floor</td></tr><tr><td align="left">34</td><td align="left">Slightly green, smooth, flat edge</td><td align="left">Gm 16, Cipr 4, Tobr 16, Ptaz 128, Ctaz 64, IMI 16 [d]</td><td align="left">Tissue</td><td align="left">Floor</td></tr><tr><td align="left">35</td><td align="left">Rough, nonflat edge, slightly green</td><td align="left">Gm 16, Cipr 4, Tobr 16, Ptaz 128, Ctaz 64, IMI 16 [d]</td><td align="left">Tissue</td><td align="left">Floor</td></tr></tbody></table><table-wrap-foot><p><sup>a </sup>Cephems: ceftazidime (Ctaz), cefoperazone (Cefr); carbapenems: imipenem (IMI); aminoglycosides: amikacin (Amik), tobramycin (Tobr), gentamicin (Gm); fluoroquinolones: ciprofloxacin (Cipr); and b-lactam/b-lactamase inhibitor combinations: piperacillin/tazobactam (Ptaz); [b], performed by disc diffusion method; <sup>c </sup>DN: demographic data are not available; [d], performed by MIC on Vitek 2.</p></table-wrap-foot></table-wrap></sec><sec><title>RAPD fingerprinting of consecutively obtained MDR <italic>P. aeruginosa </italic>clinical isolates</title><p>A total of 31 <italic>P. aeruginosa </italic>clinical isolates were typed by RAPD analysis with primers 208 (Fig. <xref ref-type="fig" rid="F1">1A</xref>) and 272 (Fig. <xref ref-type="fig" rid="F1">1B</xref>) [<xref ref-type="bibr" rid="B15">15</xref>]. RAPD fingerprints demonstrated that most clinical strains were of distinct RAPD type. More detailed demographic analysis of strains with similar RAPD revealed that strains 13 and 14 (G13) were from a single patient, strains 30 and 31 (G30) were also from a single patient, and 34 and 35 (G34) were also from a single patient. RAPD fingerprint G20 was similar for strains 4, 20, 21, and 25–28. All of these strains were obtained from specimens of tracheal aspirate, urine, and Foley catheter urine from the same patient during a 4 month period. As such, the total 31 clinical isolates contained 22 different genotypes.</p><fig position="float" id="F1"><label>Figure 1</label><caption><p><bold>Random Amplified Polymorphic DNA Typing (RAPD) of multi-drug resistant (MDR) <italic>P. aeruginosa </italic>clinical isolates</bold>. Random Amplified Polymorphic DNA Typings were generated by RAPD primers (A) 208, 5'ACGGCCGACC 3', and (B) 272, 5'AGCGGGCCAA3' [15]. Molecular size markers (Fermentas) were run in left lanes, and DNA sizes (in kilobases) are indicated to the left of the gels.</p></caption><graphic xlink:href="1476-0711-5-14-1"/></fig></sec><sec><title>Effect of multi-drug resistant (MDR) clinical isolates of <italic>P. aeruginosa </italic>on transepithelial resistance (TER) of Caco-2 monolayers</title><p>Among clinical isolates in our study, three isolates, #12, #22, and #23 showed resistance to Gm only, and two isolates, #18 and #21 showed resistance to only two anti-pseudomonas antibiotics (Table <xref ref-type="table" rid="T1">1</xref>). Since multi-drug resistance is generally defined as resistance to three or more antimicrobial agents [<xref ref-type="bibr" rid="B10">10</xref>], we did not include these strains in any further experiments. Strains 13 and 14, 30 and 31, 34 and 35 were found to be repeat isolates based on RAPD analysis and demographic data; therefore, strains 14, 31, and 34 were not included in any further experiments.</p><p>The effect of MDR <italic>P. aeruginosa </italic>clinical isolates on TER of Caco-2 cells following apical inoculation is summarized in Figure <xref ref-type="fig" rid="F2">2</xref>. Dynamic tracking of TER following apical exposure of Caco-2 cells to <italic>P. aeruginosa </italic>(Fig. <xref ref-type="fig" rid="F2">2A</xref>) (Fig. <xref ref-type="fig" rid="F2">2B</xref>) demonstrated that the strains 1, 13, and that of RAPD type G20 induced a rapid and profound decrease in TER similar to the highly cytotoxic strain PA103 [<xref ref-type="bibr" rid="B21">21</xref>]. Three isolates, 29, 7, and 15 had significant yet moderate effect on TER similar to the antibiotic sensitive reference strains ATCC 27853, PA01, and PA190. The remaining strains showed a minimal to negligible effect on TER as did the Gm<sup>R </sup>environmental isolate, PA180. Strain #1 was found to be most virulent strain based on the TER response of Caco-2 cells. TER decreased following apical exposure to as little as 10<sup>3 </sup>CFU/ml (Fig. <xref ref-type="fig" rid="F2">2C</xref>) suggesting a profound ability of the organism to disrupt epithelial barrier function.</p><fig position="float" id="F2"><label>Figure 2</label><caption><p><bold>Effect of multi-drug resistant (MDR) clinical isolates of <italic>P. aeruginosa </italic>on transepithelial resistance (TER) of Caco-2 monolayers</bold>. (A) TER of Caco-2 cells measured dynamically during co-incubation with MDR <italic>P. aeruginosa</italic>. PA103, well known cytotoxic strain; PAO1, well known invasive laboratory strain; ATCC 27853, a prototype laboratory strain used as a susceptible control in the antibiotic resistance assay; 190, a Gm<sup>S </sup>environmental isolate; and 180, a Gm<sup>R </sup>environmental isolate were used as non-MDR controls. TER is expressed as % of control TER in confluent Caco-2 cells. (B) MDR clinical isolates and control non-MDR <italic>P. aeruginosa </italic>strains are arranged in descending order of their ability to affect the TER of Caco-2 cells expressed as ΔTER/hour normalized to the initial bacterial cell density. (C) The most virulent strain, #1, induced a fall in TER even at an extremely low concentration of 10<sup>3 </sup>CFU/ml. Data are mean ± SD (n = 3).</p></caption><graphic xlink:href="1476-0711-5-14-2"/></fig></sec><sec><title>Adherence properties, motility patterns, and biofilm formation in relation to the epithelial barrier-disrupting phenotype</title><p>Regression analysis revealed that adherence (Fig. <xref ref-type="fig" rid="F3">3A</xref>) and swimming motility (Fig. <xref ref-type="fig" rid="F3">3B</xref>) significantly correlated with the TER changes in Caco-2 cells induced by MDR <italic>P. aeruginosa </italic>(r = 0.88, P < 0.0001, r = 0.57, P < 0.01, respectively). There was no correlation however between TER changes and twitching motility (r = 0.44) (Fig. <xref ref-type="fig" rid="F3">3C</xref>), or biofilm formation (r = 0.42) (Fig. <xref ref-type="fig" rid="F3">3D</xref>). High swimming motility and adherence to Caco-2 cells were the main phenotypic features of MDR barrier-disruptive strains 1, 13, and strains of G20 RAPD fingerprint. As a group, strains with a minimal effect on TER were characterized as having attenuated adherence, motility, and biofilm formation although several strains with a minimal effect on TER did display high motility behavior suggesting that motility alone is not predictive of the virulence of MDR <italic>P. aeruginosa </italic>against the intestinal epithelium.</p><fig position="float" id="F3"><label>Figure 3</label><caption><p><bold>Correlation of the ability of MDR <italic>P. aeruginosa </italic>clinical isolates to induce decrease in TER with phenotypic features</bold>. (A) adhesion, (B) swimming motility, (C) twitching motility, and (D) biofilm formation. Strains with numerically close values are grouped into enclosed boxes. Data are mean ± SD (n = 3).</p></caption><graphic xlink:href="1476-0711-5-14-3"/></fig></sec><sec><title>Effect of exposure of MDR <italic>P. aeruginosa </italic>to Gm on growth rate</title><p>Strains #1, 13, and those of G20 RAPD genotype, the most virulent in terms of their effect on TER were tested for their ability to grow in the presence of Gm. We found that as much as 50 μg.ml-1of Gm had no effect on the growth of strains 13 and G20 RAPD genotype strains (data not shown), whereas strain #1 grown in the presence of Gm showed a dose-dependent stimulation (10–20 μg.ml<sup>-1</sup>) of growth (Fig. <xref ref-type="fig" rid="F4">4A</xref>). Dynamic tracking of strain #1 exposed to 20 μg.ml<sup>-1 </sup>of Gm demonstrated this effect to be greatest during the exponential phase of growth (Fig. <xref ref-type="fig" rid="F4">4B</xref>).</p><fig position="float" id="F4"><label>Figure 4</label><caption><p><bold>Effect of exposure to Gm on the growth of <italic>P. aeruginosa </italic>clinical isolate #1</bold>. (A) Cell density measured as absorbance at 550 nm after 5 hours of growth in the presence of varied concentration of Gm. (B) Dynamically tracked cell density of clinical isolate #1 grown in the absence (control) or presence of Gm, 20 μg.ml<sup>-1</sup>. Data are mean ± SD (n = 3).</p></caption><graphic xlink:href="1476-0711-5-14-4"/></fig></sec><sec><title>Cytotoxicity of MDR <italic>P. aeruginosa </italic>clinical isolates, correlation with <italic>exoU/exoS </italic>genotype</title><p>The cytotoxic effect of the various clinical isolates following 8 hours of bacterial exposure is shown in Figure <xref ref-type="fig" rid="F5">5</xref>. Results demonstrated that most MDR clinical isolates with barrier-disruptive phenotypes harbored the <italic>exoU </italic>gene (except strain #33) and displayed cytotoxicity against Caco-2 monolayers. Clinical isolates harboring the <italic>exoS </italic>gene were not cytotoxic to Caco-2 cells.</p><fig position="float" id="F5"><label>Figure 5</label><caption><p><bold>Cytotoxicity of MDR <italic>P. aeruginosa </italic>clinical isolates against Caco-2 monolayers and their correlation to the <italic>exoU/exoS </italic>genotype</bold>. Cytotoxic effect on Caco-2 monolayers was determined after 8 hours of co-incubation and correlated to the <italic>exoU</italic>-containing clinical isolates with the exception of isolate #33. Data are mean ± SD (n = 3).</p></caption><graphic xlink:href="1476-0711-5-14-5"/></fig></sec></sec><sec><title>Discussion</title><sec><title>Effect of MDR <italic>P. aeruginosa </italic>clinical isolates on the intestinal epithelial barrier</title><p>Numerous reports have documented that the rise in multi-drug resistant nosocomial pathogens continues to threaten hospitalized patients despite various countermeasures including isolation techniques and antibiotic de-escalation therapy [<xref ref-type="bibr" rid="B22">22</xref>,<xref ref-type="bibr" rid="B23">23</xref>]. In the present study we focused on the effect of multi-drug resistant strains of <italic>P. aeruginosa </italic>on the intestinal epithelial barrier since intestinal <italic>P. aeruginosa </italic>has been shown to be a major cause of morbidity and mortality among immuno-compromised patients [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B24">24</xref>,<xref ref-type="bibr" rid="B25">25</xref>].</p><p>Caco-2 cells are an ideal cell model for these studies since they express several markers that are characteristic of normal intestinal epithelial cells including the presence of a brush border and the ability to maintain a highly resistant barrier to bacterial pathogens [<xref ref-type="bibr" rid="B17">17</xref>,<xref ref-type="bibr" rid="B26">26</xref>]. As previously mentioned, the ability of microorganisms to adhere to and alter the barrier function of intestinal epithelia is a key feature that defines their pathogenicity within the intestinal tract reservoir [<xref ref-type="bibr" rid="B27">27</xref>,<xref ref-type="bibr" rid="B28">28</xref>]. Conversely, the ability of the epithelium to resist the barrier dysregulating effect of a given pathogen through the release of mucus, IgA, defensins, etc, defines its innate defensive properties [<xref ref-type="bibr" rid="B29">29</xref>-<xref ref-type="bibr" rid="B31">31</xref>]. During host illness, especially under circumstances of critical illness, this delicate balance can be tipped in the favor of the microbe where the potential for a versatile pathogen like <italic>P. aeruginosa </italic>to subvert and erode an already compromised epithelial defense system exists [<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B32">32</xref>].</p><p>Whether MDR <italic>P. aeruginosa </italic>[<xref ref-type="bibr" rid="B33">33</xref>-<xref ref-type="bibr" rid="B36">36</xref>] strains necessarily express a more virulent phenotype continues to remain a controversial issue. While the behavior of MDR <italic>P. aeruginosa </italic>against the intestinal epithelium is unknown, its high prevalence in the intestinal tract of critically ill and immuno-compromised patients begs a better understanding of the degree to which certain strains can disrupt the intestinal epithelial barrier. For example the apical side of the intestinal epithelium is highly resistant to various toxic and cytolytic exoproducts of <italic>P. aeruginosa </italic>including exotoxin A and elastase [<xref ref-type="bibr" rid="B8">8</xref>,<xref ref-type="bibr" rid="B11">11</xref>,<xref ref-type="bibr" rid="B37">37</xref>], whereas the lung is highly susceptible. As such, lung models of <italic>P. aeruginosa </italic>infection and pathogenesis cannot be directly extrapolated to the intestinal model. Interestingly, data from the present study establish that among the MDR <italic>P. aeruginosa </italic>isolates tested in the Caco-2 model, most display a minor to minimal ability to disrupt the intestinal epithelium in both motile and non-motile strains.</p></sec><sec><title>Phenotype and genotype analysis of <italic>P. aeruginosa </italic>isolates highly disruptive to the intestinal epithelium</title><p>We identified 8 MDR clinical isolates with 3 distinct RAPD fingerprints that display a disruptive phenotype against the intestinal epithelial barrier. The presence of such strains within the intestinal tract of critically ill patients has the potential to induce a state of gut-derived sepsis with a high mortality rate as their presence in this site is often difficult to detect and eradicate.</p><p>Common features of these highly disruptive strains include high swimming motility, increased adhesiveness to intestinal epithelium, and the presence of the <italic>exoU </italic>gene. ExoU, an effector protein of the type III secretion machinery, has been previously shown to play a major role in mediating a cytotoxic phenotype of <italic>P. aeruginosa </italic>[<xref ref-type="bibr" rid="B38">38</xref>,<xref ref-type="bibr" rid="B39">39</xref>] against lung epithelial cells and HeLa cells [<xref ref-type="bibr" rid="B40">40</xref>]. That ExoU also plays an important role in disruption of the intestinal epithelial barrier and cellular cytotoxicity in this model suggests that intestinal colonization with MDR <italic>P. aeruginosa </italic>strains harboring the <italic>exoU</italic>-genotype may be associated with poor outcome in patients colonized by such strains. Although the presence of ExoS has been previously reported to play a role in the virulence of <italic>P. aeruginosa </italic>in a lung model [<xref ref-type="bibr" rid="B41">41</xref>], we found no correlation between exoS-genotype and the ability of strains to disrupt the intestinal epithelial barrier among our clinical isolates. As previously reported and confirmed by the results of the present study [<xref ref-type="bibr" rid="B42">42</xref>], motility and adhesion to host cells are important factors that appear to predict virulence.</p><p>As we and others have suggested, bacteria are fully capable of changing their virulence phenotype in direct response to host illness [<xref ref-type="bibr" rid="B43">43</xref>,<xref ref-type="bibr" rid="B44">44</xref>]. The frequent use of multiple antibiotics in the most severely ill patients could lead to the acquisition of, or alternatively the transformation to, highly virulent strains of <italic>P. aeruginosa </italic>that pose a significant threat to the patient. The ability of multi-drug resistant strains to persist for prolonged periods in such patients may allow for the development of extremely virulent phenotypes [<xref ref-type="bibr" rid="B45">45</xref>].</p><p>In conclusion, heterogeneity among critically ill humans, variability in immune response, and antibiotic use could explain the extremely polar phenotypes identified in the series of multi-drug resistant isolates collected in the present study: from phenotypes that are essentially inert with respect to the intestinal epithelium to highly motile, adhesive, and destructive phenotypes. Phenotypic assays such as motility and adhesiveness, and genotyping for the <italic>exoU </italic>gene could provide a significant prognostic input to identify multi-drug resistant <italic>P. aeruginosa </italic>strains most likely to place patients at risk for lethal gut-derived sepsis. Further characterization of strains 1, 13 and those of G20 RAPD genotype will be necessary to better understand the precise mechanism by which these strains disrupt the intestinal epithelium to a degree not previously reported for any intestinal pathogen.</p></sec></sec><sec><title>Abbreviations</title><p>Multi-drug resistance, MDR; transepithelial resistance, TER; random amplified polymorphic DNA PCR fingerprinting, RAPD; phosphate buffered saline, PBS; lactate dehydrogenase, LDH; <italic>Pseudomonas </italic>isolation agar, PIA.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>OZ performed experimental design, most experimental work, and drafting/revising the manuscript. JEK had developed and carried out the adhesiveness assay. YW was responsible for cultivation of Caco-2 cells and growing them on transwells. CB isolated and identified clinical isolates. OS participated in adherence and RAPD analyses. LW participated in adherence analyses. JRT was involved in the experimental design and discussion of experiments and manuscript revision. JCA performed experimental design, experimental data discussion, drafting/revising the manuscript, and is the PI of the NIH funding mechanism of the study. All authors read and approved the final manuscript.</p></sec>
Planning a cluster randomized trial with unequal cluster sizes: practical issues involving continuous outcomes	<sec><title>Background</title><p>Cluster randomization design is increasingly used for the evaluation of health-care, screeening or educational interventions. At the planning stage, sample size calculations usually consider an average cluster size without taking into account any potential imbalance in cluster size. However, there may exist high discrepancies in cluster sizes.</p></sec><sec sec-type="methods"><title>Methods</title><p>We performed simulations to study the impact of an imbalance in cluster size on power. We determined by simulations to which extent four methods proposed to adapt the sample size calculations to a pre-specified imbalance in cluster size could lead to adequately powered trials.</p></sec><sec><title>Results</title><p>We showed that an imbalance in cluster size can be of high influence on the power in the case of severe imbalance, particularly if the number of clusters is low and/or the intraclass correlation coefficient is high. In the case of a severe imbalance, our simulations confirmed that the minimum variance weights correction of the variation inflaction factor (VIF) used in the sample size calculations has the best properties.</p></sec><sec><title>Conclusion</title><p>Publication of cluster sizes is important to assess the real power of the trial which was conducted and to help designing future trials. We derived an adaptation of the VIF from the minimum variance weights correction to be used in case the imbalance can be a priori formulated such as "a proportion (<italic>γ</italic>) of clusters actually recruit a proportion (<italic>τ</italic>) of subjects to be included (<italic>γ </italic>≤ <italic>τ</italic>)".</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Guittet</surname><given-names>Lydia</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Ravaud</surname><given-names>Philippe</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" corresp="yes" contrib-type="author"><name><surname>Giraudeau</surname><given-names>Bruno</given-names></name><xref ref-type="aff" rid="I3">3</xref><xref ref-type="aff" rid="I4">4</xref><email>[email protected]</email></contrib>	BMC Medical Research Methodology	<sec><title>Background</title><p>A cluster randomized trial involves randomizing social units or clusters of individuals rather than the individuals themselves. This design, which is increasingly being used for evaluating healthcare, screening and educational interventions presents specific constraints that must be considered during planning and analysis [<xref ref-type="bibr" rid="B1">1</xref>,<xref ref-type="bibr" rid="B2">2</xref>]. Indeed, the responses of individuals within a cluster tend to be more similar than those of individuals of different clusters, and we thus define the clustering effect as 1 + (<italic>m </italic>- 1)<italic>ρ</italic>, where <italic>m </italic>is the average number of subjects per cluster and <italic>ρ </italic>the intraclass correlation coefficient (ICC). This clustering effect is used during the planning of cluster randomized trials as an inflation factor to increase the sample size required by an individual randomization trial. However, such an approach does not take into account variations in cluster size, which might differ greatly. Indeed, as illustrated by Kerry <italic>et al </italic>[<xref ref-type="bibr" rid="B3">3</xref>], cluster size may depend on, for example, (i) the potential of recruitment of the cluster (i.e., the number of subjects belonging to each cluster), (ii) the eligible fraction of subjects, which may vary among clusters, or (iii) the ability of physicians to recruit subjects within each cluster. Such an imbalance in cluster size reduces the power of the trial and has to be taken into account in the sample size calculation.</p><p>Kerry <italic>et al </italic>[<xref ref-type="bibr" rid="B3">3</xref>] assessed the theoretical efficacy of 3 weightings of the inflation factor but in the context of cluster level analysis, so summary statistics are estimated at the cluster level and the unit of analysis remains the cluster. Manatunga <italic>et al </italic>[<xref ref-type="bibr" rid="B4">4</xref>], however, assessed a correction on the basis of the assumed distribution of cluster sizes in the context of marginal models, but the authors' simulations covered a range of ICCs larger than those usually observed in cluster randomized trials.</p><p>Our aim was therefore to assess these proposed corrections in the framework of cluster randomized trials in which the unit of analysis remains the subject, embedded in the cluster. We first describe the random effects model used to simulate clustered data; then display the simulation design used to evaluate the loss of power due to imbalance in cluster size and the findings. Corrections of the variance inflation factor to allow for cluster size inequality evaluated by simulation and robustness of these corrections to misspecification of the ICC is assessed. practical guidelines for the planning stage of cluster randomized trials are drawn and perspectives for future research.</p></sec><sec sec-type="methods"><title>Methods and results</title><sec><title>Theoretical background</title><sec><title>The mixed effects model</title><p>Let us supposed a continuous outcome distributed according to the following mixed-effects model:</p><p><italic>Y</italic><sub><italic>ijk </italic></sub>= <italic>θ</italic><sub><italic>i </italic></sub>+ <italic>β</italic><sub><italic>ij </italic></sub>+ <italic>ε</italic><sub><italic>ijk</italic></sub>    (1)</p><p>where <italic>Y</italic><sub><italic>ijk </italic></sub>is the observed response for the <italic>k</italic>th subject in the <italic>j</italic>th cluster of the <italic>i</italic>th group, <italic>θ</italic><sub><italic>i </italic></sub>is the overall mean in the <italic>i</italic>th group, <italic>β</italic><sub><italic>ij </italic></sub>is the random effect associated with the cluster effect and <italic>ε</italic><sub><italic>ijk </italic></sub>is the residual effect. The <italic>β</italic><sub><italic>ij </italic></sub>and <italic>ε</italic><sub><italic>ijk </italic></sub>are assumed to be independent and normally distributed as (0; <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M1" name="1471-2288-6-17-i1" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>σ</mml:mi><mml:mi>b</mml:mi><mml:mn>2</mml:mn></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGaemOyaigabaGaeGOmaidaaaaa@30E2@</mml:annotation></mml:semantics></mml:math></inline-formula>) and (0; <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M2" name="1471-2288-6-17-i2" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>σ</mml:mi><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGaem4DaChabaGaeGOmaidaaaaa@310C@</mml:annotation></mml:semantics></mml:math></inline-formula>) respectively.</p><p>The ICC quantifies the degree of similarity between the responses of subjects in the same cluster and is defined as the proportion of the total outcome variation between clusters:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M3" name="1471-2288-6-17-i3" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>ρ</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:msubsup><mml:mi>σ</mml:mi><mml:mi>b</mml:mi><mml:mn>2</mml:mn></mml:msubsup></mml:mrow><mml:mrow><mml:msubsup><mml:mi>σ</mml:mi><mml:mi>b</mml:mi><mml:mn>2</mml:mn></mml:msubsup><mml:mo>+</mml:mo><mml:msubsup><mml:mi>σ</mml:mi><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msubsup></mml:mrow></mml:mfrac><mml:mtext>     </mml:mtext><mml:mrow><mml:mo>(</mml:mo><mml:mn>2</mml:mn><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFbpGCcqGH9aqpdaWcaaqaaiab=n8aZnaaDaaaleaacqWGIbGyaeaacqaIYaGmaaaakeaacqWFdpWCdaqhaaWcbaGaemOyaigabaGaeGOmaidaaOGaey4kaSIae83Wdm3aa0baaSqaaiabdEha3bqaaiabikdaYaaaaaGccaWLjaGaaCzcamaabmaabaGaeGOmaidacaGLOaGaayzkaaaaaa@40F0@</mml:annotation></mml:semantics></mml:math></inline-formula></p></sec><sec><title>Sample size calculations</title><p>Considering <italic>g </italic>clusters of <italic>m </italic>individuals to be randomized in each group, the total number of subject <italic>N </italic>per group is given by [<xref ref-type="bibr" rid="B2">2</xref>]:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M4" name="1471-2288-6-17-i4" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>N</mml:mi><mml:mo>=</mml:mo><mml:mi>m</mml:mi><mml:mi>g</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mn>2</mml:mn><mml:msup><mml:mi>σ</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:msup><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>t</mml:mi><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>α</mml:mi><mml:mo>/</mml:mo><mml:mn>2</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>,</mml:mo><mml:mn>2</mml:mn><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>g</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:msub><mml:mo>+</mml:mo><mml:msub><mml:mi>t</mml:mi><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>β</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>,</mml:mo><mml:mn>2</mml:mn><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>g</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mn>2</mml:mn></mml:msup><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>m</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>]</mml:mo></mml:mrow></mml:mrow><mml:mrow><mml:msup><mml:mi>Δ</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mfrac><mml:mtext>     </mml:mtext><mml:mrow><mml:mo>(</mml:mo><mml:mn>3</mml:mn><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGobGtcqGH9aqpcqWGTbqBcqWGNbWzcqGH9aqpdaWcaaqaaiabikdaYGGaciab=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@66FE@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>where Δ is the absolute mean difference between groups (i.e., Δ = \|<italic>θ</italic><sub>0 </sub>- <italic>θ</italic><sub>1</sub>\|), <italic>σ</italic><sup>2 </sup>is the total variance defined as (<inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M5" name="1471-2288-6-17-i1" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>σ</mml:mi><mml:mi>b</mml:mi><mml:mn>2</mml:mn></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGaemOyaigabaGaeGOmaidaaaaa@30E2@</mml:annotation></mml:semantics></mml:math></inline-formula> + <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M6" name="1471-2288-6-17-i2" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>σ</mml:mi><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGaem4DaChabaGaeGOmaidaaaaa@310C@</mml:annotation></mml:semantics></mml:math></inline-formula>) and <italic>t</italic><sub>(1 - <italic>α</italic>/2),2(<italic>g </italic>- 1) </sub>and <italic>t</italic><sub>(1 - <italic>β</italic>),2(<italic>g </italic>- 1) </sub>is the 100 × (1 - <italic>α</italic>/2) and 100 × (1 - <italic>β</italic>) percentiles of the Student <italic>t</italic>-distribution with 2(<italic>g </italic>- 1) degrees of freedom. Considering the effect size, defined as the relative difference between groups (i.e., <italic>ES </italic>= \|<italic>θ</italic><sub>0 </sub>- <italic>θ</italic><sub>1</sub>\|/<italic>σ </italic>= Δ/<italic>σ</italic>), expression (3) can be re-written as:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M7" name="1471-2288-6-17-i5" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>N</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mn>2</mml:mn><mml:msup><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>t</mml:mi><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>α</mml:mi><mml:mo>/</mml:mo><mml:mn>2</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>,</mml:mo><mml:mn>2</mml:mn><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>g</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:msub><mml:mo>+</mml:mo><mml:msub><mml:mi>t</mml:mi><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>β</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>,</mml:mo><mml:mn>2</mml:mn><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>g</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mn>2</mml:mn></mml:msup><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>m</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>]</mml:mo></mml:mrow></mml:mrow><mml:mrow><mml:mi>E</mml:mi><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mfrac><mml:mtext>     </mml:mtext><mml:mrow><mml:mo>(</mml:mo><mml:mn>4</mml:mn><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@6135@</mml:annotation></mml:semantics></mml:math></inline-formula></p></sec></sec><sec><title>Impact of cluster size inequality</title><sec><title>Simulation study</title><p>Monte Carlo simulations were used to assess the impact of imbalance in cluster size on both power and type I error. A 2 × 4 × 4 factorial plan was used, considering 2 effect sizes (0.25, 0.50) to be detected with fixed numbers of clusters (5, 10, 20, 40) and 4 <italic>a priori </italic>postulated values of the ICC (0.005, 0.02, 0.05, 0.10). The ICC values were chosen according to previously published estimates [<xref ref-type="bibr" rid="B5">5</xref>-<xref ref-type="bibr" rid="B15">15</xref>], and the number of clusters is in agreement with that from a recent review of cluster randomized trials in primary care settings in which the median number of randomized clusters was estimated at 34 [<xref ref-type="bibr" rid="B13">13</xref>]. The <italic>α </italic>and <italic>β </italic>values were fixed at 0.05 and 0.20, respectively, in any case.</p><p>Once the sample size was calculated, correlated data were simulated, according to model (1). From a practical point of view, data were generated as the sum of a fixed effect (<italic>θ</italic><sub>0 </sub>or <italic>θ</italic><sub>1 </sub>if the control or experimental group, respectively) and realizations of the 2 random variables <italic>β</italic><sub><italic>ij </italic></sub>and <italic>ε</italic><sub><italic>ijk</italic></sub>. For convenience and without loss of generality we set <italic>θ</italic><sub>0 </sub>equal to 0 and (<inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M8" name="1471-2288-6-17-i1" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>σ</mml:mi><mml:mi>b</mml:mi><mml:mn>2</mml:mn></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGaemOyaigabaGaeGOmaidaaaaa@30E2@</mml:annotation></mml:semantics></mml:math></inline-formula> + <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M9" name="1471-2288-6-17-i2" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>σ</mml:mi><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGaem4DaChabaGaeGOmaidaaaaa@310C@</mml:annotation></mml:semantics></mml:math></inline-formula>) equal to 1. These constraints then allow for defining <italic>θ</italic><sub>1 </sub>as the effect size ES, <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M10" name="1471-2288-6-17-i1" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>σ</mml:mi><mml:mi>b</mml:mi><mml:mn>2</mml:mn></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGaemOyaigabaGaeGOmaidaaaaa@30E2@</mml:annotation></mml:semantics></mml:math></inline-formula> as <italic>ρ </italic>and <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M11" name="1471-2288-6-17-i2" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>σ</mml:mi><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGaem4DaChabaGaeGOmaidaaaaa@310C@</mml:annotation></mml:semantics></mml:math></inline-formula> as (1 - <italic>ρ</italic>).</p></sec><sec><title>Cluster size</title><p>For any combination of <italic>ES</italic>, <italic>g </italic>and <italic>ρ</italic>, we simulated randomized trials with, on the one hand, constant cluster size and, on the other, imbalance in cluster size. In the absence of cluster sizes publications, three types of imbalance were considered:</p><p>1. A moderate imbalance:</p><p>For each group, each of the <italic>N </italic>subjects had an equiprobability of being in any of the <italic>g </italic>clusters randomized in this group. From a practical point of view, for any of the <italic>N </italic>subjects, we randomly selected with equiprobability the cluster to which it belongs, before adding the appropriate realizations of random variables <italic>β</italic><sub><italic>ij </italic></sub>and <italic>ε</italic><sub><italic>ijk</italic></sub>.</p><p>2. A "Pareto" imbalance</p><p>Following the economic Pareto's principle, we considered the situation in which 80% of the subjects actually belong to only 20% of the clusters. From a practical point of view, we thus defined 2 strata within each group: the strata of large clusters (e.g., 20% of the <italic>g </italic>clusters) and the strata of small clusters. Eighty percent of the <italic>N </italic>subjects were in the large cluster strata, while the 20% remaining were in the small cluster strata. Then, within each stratum, subjects were randomly assigned with equiprobability to one of the clusters.</p><p>3. A Poisson imbalance</p><p>Cluster sizes were finally defined according to a Poisson distribution, which has already been used in such a context [<xref ref-type="bibr" rid="B16">16</xref>,<xref ref-type="bibr" rid="B17">17</xref>]. We thus considered a Poisson distribution with parameter <italic>m </italic>defined as <italic>N</italic>/<italic>g </italic>and defined the cluster size of any cluster before generating the associated observations.</p><p>In this latter situation, and contrary to the 2 previous ones, the total number of patients per group varies and is equal to <italic>N </italic>only on average. Moreover, in the 3 types of cluster size inequality, the actual number of clusters per group could be smaller than <italic>g</italic>, because clusters could be empty.</p><p>For any combination of ES, <italic>g </italic>and ICC, and for any situation (balance or any type of imbalance in cluster size), 5000 replications of data were simulated by use of SAS 8.1 software.</p></sec><sec><title>Analysis</title><p>Data analysis involved no stratification on cluster size. We used the MIXED procedure in SAS [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B19">19</xref>] to assess restricted maximum likelihood (REML) estimates of variance components. The Wald test statistic was then used to test the significance of the intervention effect with the Student t-distribution, with g<sub>0</sub>+g<sub>1</sub>-2 degrees of freedom as the reference distribution, where g<sub>0 </sub>and g<sub>1 </sub>are the actual numbers of nonempty clusters in the control and intervention groups, respectively.</p><p>The empirical type I error and power were calculated as the proportion of significant trials (defined as a p value smaller than the nominal <italic>α </italic>level) when <italic>θ</italic><sub>1 </sub>equals 0 and <italic>ES</italic>, respectively.</p></sec></sec></sec><sec><title>Results</title><p>Results are expressed as absolute bias and mean square error on the one hand, and empirical' type I error and power on the other. Table <xref ref-type="table" rid="T1">1</xref> displays the results associated with an <italic>a priori </italic>postulated effect size of 0.25, while Table <xref ref-type="table" rid="T2">2</xref> displays the results associated with a 0.50 effect size. In 7 situations, data sets could not be generated for the following combinations ES/ICC/g: 0.25/0.020/5, 0.25/0.050/5, 0.25/0.050/10, 0.25/0.100/5, 0.25/0.100/10, 0.25/0.100/20 and 0.50/0.100/5. Indeed, when the number of clusters is small and/or the ICC high, even an infinite cluster size may not allow for achieving 80% power [<xref ref-type="bibr" rid="B20">20</xref>].</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Bias, mean square error, empirical type I error and power in cluster randomized trials according to several types of imbalance in cluster size – Effect size = 0.25</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="center" colspan="3"><bold>Simulation parameters</bold><sup>1</sup></td><td align="left"><bold>Type of imbalance</bold></td><td align="center"><bold>Bias</bold></td><td align="center"><bold>Mean Square Error</bold></td><td align="center"><bold>Empirical type I error</bold><sup>2</sup></td><td align="center"><bold>Empirical power</bold><sup>2</sup></td></tr><tr><td colspan="3"><hr></hr></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>Intraclass correlation coefficient (<italic>ρ</italic>)</bold></td><td align="center"><bold>Number of clusters in each arm (<italic>g</italic>)</bold></td><td align="center"><bold>Total number of subjects in each arm (<italic>N</italic>)</bold></td><td></td><td></td><td></td><td></td><td></td></tr></thead><tbody><tr><td align="center"><bold>0.005</bold></td><td align="center"><bold>5</bold></td><td align="center"><bold>485</bold></td><td align="left"><bold>None</bold></td><td align="center">-0.0020</td><td align="center">0.0062</td><td align="center">0.0328</td><td align="center">0.7756</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">-0.0015</td><td align="center">0.0062</td><td align="center">0.0300</td><td align="center">0.7800</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">0.0003</td><td align="center">0.0061</td><td align="center">0.0368</td><td align="center">0.7814</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">0.0002</td><td align="center">0.0100</td><td align="center">0.0664</td><td align="center">0.6432</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.005</bold></td><td align="center"><bold>10</bold></td><td align="center"><bold>326</bold></td><td align="left"><bold>None</bold></td><td align="center">0.0005</td><td align="center">0.0070</td><td align="center">0.0326</td><td align="center">0.7868</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0009</td><td align="center">0.0073</td><td align="center">0.0402</td><td align="center">0.7838</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0010</td><td align="center">0.0070</td><td align="center">0.0356</td><td align="center">0.7884</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">0.0043</td><td align="center">0.0100</td><td align="center">0.0566</td><td align="center">0.6968</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.005</bold></td><td align="center"><bold>20</bold></td><td align="center"><bold>282</bold></td><td align="left"><bold>None</bold></td><td align="center">-0.0010</td><td align="center">0.0072</td><td align="center">0.0320</td><td align="center">0.7942</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0014</td><td align="center">0.0075</td><td align="center">0.0398</td><td align="center">0.7878</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0010</td><td align="center">0.0076</td><td align="center">0.0408</td><td align="center">0.7802</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">0.0000</td><td align="center">0.0090</td><td align="center">0.0486</td><td align="center">0.7258</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.005</bold></td><td align="center"><bold>40</bold></td><td align="center"><bold>265</bold></td><td align="left"><bold>None</bold></td><td align="center">0.0006</td><td align="center">0.0078</td><td align="center">0.0444</td><td align="center">0.7848</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0011</td><td align="center">0.0082</td><td align="center">0.0458</td><td align="center">0.7936</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0017</td><td align="center">0.0082</td><td align="center">0.0484</td><td align="center">0.7772</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">0.0000</td><td align="center">0.0086</td><td align="center">0.0466</td><td align="center">0.7572</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.020</bold></td><td align="center"><bold>10</bold></td><td align="center"><bold>629</bold></td><td align="left"><bold>None</bold></td><td align="center">-0.0017</td><td align="center">0.0070</td><td align="center">0.0448</td><td align="center">0.8012</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0017</td><td align="center">0.0073</td><td align="center">0.0544</td><td align="center">0.7974</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">0.0009</td><td align="center">0.0074</td><td align="center">0.0510</td><td align="center">0.7992</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">-0.0022</td><td align="center">0.0118</td><td align="center">0.0904</td><td align="center">0.6236</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.020</bold></td><td align="center"><bold>20</bold></td><td align="center"><bold>353</bold></td><td align="left"><bold>None</bold></td><td align="center">-0.0004</td><td align="center">0.0073</td><td align="center">0.0452</td><td align="center">0.8000</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0006</td><td align="center">0.0074</td><td align="center">0.0408</td><td align="center">0.7980</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0007</td><td align="center">0.0075</td><td align="center">0.0458</td><td align="center">0.7968</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">0.0009</td><td align="center">0.0115</td><td align="center">0.0660</td><td align="center">0.6546</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.020</bold></td><td align="center"><bold>40</bold></td><td align="center"><bold>290</bold></td><td align="left"><bold>None</bold></td><td align="center">0.0017</td><td align="center">0.0080</td><td align="center">0.0518</td><td align="center">0.7932</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0001</td><td align="center">0.0077</td><td align="center">0.0466</td><td align="center">0.7944</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0003</td><td align="center">0.0077</td><td align="center">0.0466</td><td align="center">0.7912</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">0.0003</td><td align="center">0.0101</td><td align="center">0.0556</td><td align="center">0.7008</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.050</bold></td><td align="center"><bold>20</bold></td><td align="center"><bold>743</bold></td><td align="left"><bold>None</bold></td><td align="center">-0.0007</td><td align="center">0.0075</td><td align="center">0.0436</td><td align="center">0.7916</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">-0.0018</td><td align="center">0.0077</td><td align="center">0.0540</td><td align="center">0.8026</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0003</td><td align="center">0.0078</td><td align="center">0.0536</td><td align="center">0.7950</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">-0.0022</td><td align="center">0.0115</td><td align="center">0.0562</td><td align="center">0.6256</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.050</bold></td><td align="center"><bold>40</bold></td><td align="center"><bold>361</bold></td><td align="left"><bold>None</bold></td><td align="center">-0.0012</td><td align="center">0.0080</td><td align="center">0.0528</td><td align="center">0.7944</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0031</td><td align="center">0.0080</td><td align="center">0.0510</td><td align="center">0.7926</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0001</td><td align="center">0.0080</td><td align="center">0.0502</td><td align="center">0.7904</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">-0.0023</td><td align="center">0.0121</td><td align="center">0.0604</td><td align="center">0.6242</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.100</bold></td><td align="center"><bold>40</bold></td><td align="center"><bold>652</bold></td><td align="left"><bold>None</bold></td><td align="center">0.0021</td><td align="center">0.0076</td><td align="center">0.0504</td><td align="center">0.7966</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0013</td><td align="center">0.0078</td><td align="center">0.0458</td><td align="center">0.8118</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0022</td><td align="center">0.0078</td><td align="center">0.0506</td><td align="center">0.7946</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">-0.0031</td><td align="center">0.0121</td><td align="center">0.0546</td><td align="center">0.6006</td></tr></tbody></table><table-wrap-foot><p><sup>1</sup>N is the number of subjects per intervention arm, calculated under the assumption of constant cluster size</p><p><sup>2 </sup>The nominal values for type I and type II error rates were fixed at 0.05 and 0.20, respectively.</p></table-wrap-foot></table-wrap><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Bias, mean square error, empirical type I error and power in cluster randomized trials according to several types of imbalance in cluster size – Effect size = 0.50</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="center" colspan="3"><bold>Simulation parameters</bold><sup>1</sup></td><td align="left"><bold>Type of imbalance</bold></td><td align="center"><bold>Bias</bold></td><td align="center"><bold>Mean Square Error</bold></td><td align="center"><bold>Empirical type I error</bold><sup>2</sup></td><td align="center"><bold>Empirical power</bold><sup>2</sup></td></tr><tr><td colspan="3"><hr></hr></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>Intraclass correlation coefficient (<italic>ρ</italic>)</bold></td><td align="center"><bold>Number of clusters in each arm (<italic>g</italic>)</bold></td><td align="center"><bold>Total number of subjects in each arm (<italic>N</italic>)</bold></td><td align="left"><bold>Type of imbalance</bold></td><td></td><td></td><td></td><td></td></tr></thead><tbody><tr><td align="center"><bold>0.005</bold></td><td align="center"><bold>5</bold></td><td align="center"><bold>89</bold></td><td align="left"><bold>None</bold></td><td align="center">0.0025</td><td align="center">0.0238</td><td align="center">0.0190</td><td align="center">0.7648</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0010</td><td align="center">0.0243</td><td align="center">0.0204</td><td align="center">0.7622</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">0.0014</td><td align="center">0.0243</td><td align="center">0.0214</td><td align="center">0.7596</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">-0.0064</td><td align="center">0.0393</td><td align="center">0.0256</td><td align="center">0.6250</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.005</bold></td><td align="center"><bold>10</bold></td><td align="center"><bold>73</bold></td><td align="left"><bold>None</bold></td><td align="center">-0.0011</td><td align="center">0.0288</td><td align="center">0.0328</td><td align="center">0.7660</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">-0.0015</td><td align="center">0.0290</td><td align="center">0.0322</td><td align="center">0.7718</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0007</td><td align="center">0.0298</td><td align="center">0.0318</td><td align="center">0.7662</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">-0.0005</td><td align="center">0.0344</td><td align="center">0.0352</td><td align="center">0.7090</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.005</bold></td><td align="center"><bold>20</bold></td><td align="center"><bold>67</bold></td><td align="left"><bold>None</bold></td><td align="center">-0.0011</td><td align="center">0.0303</td><td align="center">0.0384</td><td align="center">0.7764</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">-0.0038</td><td align="center">0.0296</td><td align="center">0.0318</td><td align="center">0.7700</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0012</td><td align="center">0.0301</td><td align="center">0.0382</td><td align="center">0.7664</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">-0.0004</td><td align="center">0.0323</td><td align="center">0.0334</td><td align="center">0.7322</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.005</bold></td><td align="center"><bold>40</bold></td><td align="center"><bold>65</bold></td><td align="left"><bold>None</bold></td><td align="center">0.0005</td><td align="center">0.0310</td><td align="center">0.0446</td><td align="center">0.7986</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0021</td><td align="center">0.0322</td><td align="center">0.0478</td><td align="center">0.7896</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">0.0028</td><td align="center">0.0305</td><td align="center">0.0396</td><td align="center">0.7860</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">-0.0007</td><td align="center">0.0320</td><td align="center">0.0382</td><td align="center">0.7518</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.020</bold></td><td align="center"><bold>5</bold></td><td align="center"><bold>119</bold></td><td align="left"><bold>None</bold></td><td align="center">0.0025</td><td align="center">0.0238</td><td align="center">0.0190</td><td align="center">0.7648</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">-0.0011</td><td align="center">0.0248</td><td align="center">0.0310</td><td align="center">0.7856</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0021</td><td align="center">0.0250</td><td align="center">0.0306</td><td align="center">0.7786</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">0.0009</td><td align="center">0.0413</td><td align="center">0.0674</td><td align="center">0.6262</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.020</bold></td><td align="center"><bold>10</bold></td><td align="center"><bold>81</bold></td><td align="left"><bold>None</bold></td><td align="center">-0.0031</td><td align="center">0.0273</td><td align="center">0.0320</td><td align="center">0.7798</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0006</td><td align="center">0.0282</td><td align="center">0.0364</td><td align="center">0.7772</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0003</td><td align="center">0.0288</td><td align="center">0.0378</td><td align="center">0.7778</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">0.0078</td><td align="center">0.0394</td><td align="center">0.0550</td><td align="center">0.6910</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.020</bold></td><td align="center"><bold>20</bold></td><td align="center"><bold>70</bold></td><td align="left"><bold>None</bold></td><td align="center">0.0026</td><td align="center">0.0312</td><td align="center">0.0476</td><td align="center">0.7838</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0026</td><td align="center">0.0312</td><td align="center">0.0476</td><td align="center">0.7838</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">0.0032</td><td align="center">0.0306</td><td align="center">0.0436</td><td align="center">0.7894</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">0.0003</td><td align="center">0.0362</td><td align="center">0.0460</td><td align="center">0.7056</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.020</bold></td><td align="center"><bold>40</bold></td><td align="center"><bold>66</bold></td><td align="left"><bold>None</bold></td><td align="center">-0.0026</td><td align="center">0.0314</td><td align="center">0.0498</td><td align="center">0.7878</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0026</td><td align="center">0.0312</td><td align="center">0.0476</td><td align="center">0.7838</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">0.0049</td><td align="center">0.0326</td><td align="center">0.0476</td><td align="center">0.7828</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">-0.0007</td><td align="center">0.0337</td><td align="center">0.0422</td><td align="center">0.7328</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.050</bold></td><td align="center"><bold>5</bold></td><td align="center"><bold>423</bold></td><td align="left"><bold>None</bold></td><td align="center">0.0015</td><td align="center">0.0246</td><td align="center">0.0482</td><td align="center">0.7980</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">-0.0001</td><td align="center">0.0240</td><td align="center">0.0460</td><td align="center">0.8004</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0005</td><td align="center">0.0237</td><td align="center">0.0478</td><td align="center">0.7988</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">0.0026</td><td align="center">0.0337</td><td align="center">0.0768</td><td align="center">0.6808</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.050</bold></td><td align="center"><bold>10</bold></td><td align="center"><bold>103</bold></td><td align="left"><bold>None</bold></td><td align="center">-0.0006</td><td align="center">0.0280</td><td align="center">0.0426</td><td align="center">0.7964</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0012</td><td align="center">0.0286</td><td align="center">0.0446</td><td align="center">0.7952</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0017</td><td align="center">0.0293</td><td align="center">0.0440</td><td align="center">0.7754</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">-0.0022</td><td align="center">0.0466</td><td align="center">0.0770</td><td align="center">0.6342</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.050</bold></td><td align="center"><bold>20</bold></td><td align="center"><bold>76</bold></td><td align="left"><bold>None</bold></td><td align="center">0.0027</td><td align="center">0.0298</td><td align="center">0.0436</td><td align="center">0.8020</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">-0.0018</td><td align="center">0.0308</td><td align="center">0.0452</td><td align="center">0.7784</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0016</td><td align="center">0.0323</td><td align="center">0.0526</td><td align="center">0.7672</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">-0.0056</td><td align="center">0.0396</td><td align="center">0.0528</td><td align="center">0.6620</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.050</bold></td><td align="center"><bold>40</bold></td><td align="center"><bold>67</bold></td><td align="left"><bold>None</bold></td><td align="center">-0.0019</td><td align="center">0.0313</td><td align="center">0.0468</td><td align="center">0.7880</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">-0.0012</td><td align="center">0.0294</td><td align="center">0.0516</td><td align="center">0.7740</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">0.0000</td><td align="center">0.0335</td><td align="center">0.0504</td><td align="center">0.7712</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">0.0022</td><td align="center">0.0376</td><td align="center">0.0516</td><td align="center">0.7026</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.100</bold></td><td align="center"><bold>10</bold></td><td align="center"><bold>213</bold></td><td align="left"><bold>None</bold></td><td align="center">0.0006</td><td align="center">0.0263</td><td align="center">0.0426</td><td align="center">0.8056</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0015</td><td align="center">0.0289</td><td align="center">0.0530</td><td align="center">0.7940</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0007</td><td align="center">0.0287</td><td align="center">0.0538</td><td align="center">0.8004</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">-0.0027</td><td align="center">0.0438</td><td align="center">0.0730</td><td align="center">0.6394</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.100</bold></td><td align="center"><bold>20</bold></td><td align="center"><bold>89</bold></td><td align="left"><bold>None</bold></td><td align="center">-0.0029</td><td align="center">0.0303</td><td align="center">0.0470</td><td align="center">0.7888</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">-0.0020</td><td align="center">0.0324</td><td align="center">0.0530</td><td align="center">0.7760</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">-0.0004</td><td align="center">0.0316</td><td align="center">0.0488</td><td align="center">0.7744</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">0.0064</td><td align="center">0.0492</td><td align="center">0.0674</td><td align="center">0.6276</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.100</bold></td><td align="center"><bold>40</bold></td><td align="center"><bold>70</bold></td><td align="left"><bold>None</bold></td><td align="center">0.0038</td><td align="center">0.0331</td><td align="center">0.0510</td><td align="center">0.7890</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Moderate</bold></td><td align="center">0.0031</td><td align="center">0.0337</td><td align="center">0.0506</td><td align="center">0.7738</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Poisson</bold></td><td align="center">0.0020</td><td align="center">0.0332</td><td align="center">0.0456</td><td align="center">0.7658</td></tr><tr><td></td><td></td><td></td><td align="left"><bold>Pareto</bold></td><td align="center">0.0019</td><td align="center">0.0433</td><td align="center">0.0536</td><td align="center">0.6641</td></tr></tbody></table><table-wrap-foot><p><sup>1</sup>N is the number of subjects per intervention arm, calculated under the assumption of constant cluster size</p><p><sup>2 </sup>The nominal values for type I and type II error rates were fixed at 0.05 and 0.20, respectively.</p></table-wrap-foot></table-wrap><p>No significant bias was induced by inequality in cluster size (since the relative bias was no more than about 1.5%, in absolute value), while the mean square error was barely increased in cases of severe imbalance (Pareto imbalance).</p><p>When the number of clusters is small, type I errors were estimated at a lower level than the nominal one, even with no imbalance in cluster sizes. A symmetrical result was also observed for power, which was estimated at a lower level than the nominal one. This result was of greater magnitude for small ICCs and for greater effect size, which corresponded to situations in which the total number of subjects to be included is reduced. Otherwise, although moderate and Poisson imbalances were of no influence, a Pareto's imbalance was associated with an increase in both type I and type II errors. As an example, if one is willing to detect a 0.25 effect size and plan a randomized trial with 10 clusters per arm with an <italic>a priori </italic>postulated ICC of 0.02, a Pareto imbalance leads to type I and type II errors of 9% and 38%, respectively, and nominal values fixed at 5% and 20%. This result is of greater magnitude for large ICCs and a small number of clusters.</p><p>Thus, while moderate imbalances (based on an equiprobability hypothesis) and Poisson's imbalances can be neglected at the planning stage, a more severe imbalance (such as the Pareto's imbalance) should be taken into account, thus leading to an adjustment in sample size calculations.</p><sec><title>Sample size adjustment for unbalanced trials</title><sec><title>Adjusted variance inflation factors</title><p>The (1 + (<italic>m </italic>- 1)<italic>ρ</italic>) factor in expressions (3) and (4) defines the variance inflation factor (VIF) that takes into account the correlation induced by the cluster randomization. This VIF supposes a constant cluster size (<italic>m</italic>) or is based on the average cluster size in case of imbalance. Kerry <italic>et al </italic>[<xref ref-type="bibr" rid="B3">3</xref>] and Manatunga <italic>et al </italic>[<xref ref-type="bibr" rid="B4">4</xref>] proposed to adjust the VIF in cases of an imbalance in cluster size. Thus, we propose 4 corrections. The first 3 are based on weights derived from the <italic>a priori </italic>postulated distribution of cluster sizes among the g clusters (i.e., the different values of <italic>m</italic><sub><italic>j</italic></sub>, where <italic>m</italic><sub><italic>j </italic></sub>is the size of the <italic>j</italic><sup>th </sup>cluster), and the fourth is based on the expected mean and variance of this latter distribution.</p><p>1. Equal weights (denoted <italic>w</italic><sub>1</sub>)[<xref ref-type="bibr" rid="B3">3</xref>]:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M12" name="1471-2288-6-17-i6" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>g</mml:mi></mml:mfrac><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>g</mml:mi></mml:munderover><mml:mrow><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow></mml:mfrac><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>+</mml:mo><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>ρ</mml:mi></mml:mrow></mml:mstyle></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrdaWgaaWcbaGaem4DaC3aaSbaaWqaaiabigdaXaqabaaaleqaaOGaeyypa0ZaaSaaaeaacuWGTbqBgaqeaaqaaiabdEgaNbaadaaeWbqaamaalaaabaGaeGymaedabaGaemyBa02aaSbaaSqaaiabdQgaQbqabaaaaOWaaeWaaeaacqaIXaqmcqGHsisliiGacqWFbpGCaiaawIcacaGLPaaacqGHRaWkcuWGTbqBgaqeaiab=f8aYbWcbaGaemOAaOMaeyypa0JaeGymaedabaGaem4zaCganiabggHiLdaaaa@4AF5@</mml:annotation></mml:semantics></mml:math></inline-formula> where <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M13" name="1471-2288-6-17-i7" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mi>g</mml:mi></mml:mfrac><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>g</mml:mi></mml:munderover><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow></mml:mstyle></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeaiabg2da9maalaaabaGaeGymaedabaGaem4zaCgaamaaqahabaGaemyBa02aaSbaaSqaaiabdQgaQbqabaaabaGaemOAaOMaeyypa0JaeGymaedabaGaem4zaCganiabggHiLdaaaa@3B51@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>2. Cluster size weights (denoted <italic>w</italic><sub>2</sub>)[<xref ref-type="bibr" rid="B3">3</xref>]:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M14" name="1471-2288-6-17-i8" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mo stretchy="false">(</mml:mo><mml:mover accent="true"><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>A</mml:mi></mml:msub></mml:mrow><mml:mo stretchy="true">¯</mml:mo></mml:mover><mml:mo>−</mml:mo><mml:mn>1</mml:mn><mml:mo stretchy="false">)</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrdaWgaaWcbaGaem4DaC3aaSbaaWqaaiabikdaYaqabaaaleqaaOGaeyypa0JaeGymaeJaey4kaSIaeiikaGYaa0aaaeaacqWGTbqBdaWgaaWcbaGaemyqaeeabeaaaaGccqGHsislcqaIXaqmcqGGPaqkiiGacqWFbpGCaaa@3DCB@</mml:annotation></mml:semantics></mml:math></inline-formula> where <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M15" name="1471-2288-6-17-i9" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mover accent="true"><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>A</mml:mi></mml:msub></mml:mrow><mml:mo stretchy="true">¯</mml:mo></mml:mover><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>g</mml:mi></mml:munderover><mml:mrow><mml:msubsup><mml:mi>m</mml:mi><mml:mi>j</mml:mi><mml:mn>2</mml:mn></mml:msubsup></mml:mrow></mml:mstyle></mml:mrow><mml:mrow><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>g</mml:mi></mml:munderover><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaadaqdaaqaaiabd2gaTnaaBaaaleaacqWGbbqqaeqaaaaakiabg2da9maalaaabaWaaabCaeaacqWGTbqBdaqhaaWcbaGaemOAaOgabaGaeGOmaidaaaqaaiabdQgaQjabg2da9iabigdaXaqaaiabdEgaNbqdcqGHris5aaGcbaWaaabCaeaacqWGTbqBdaWgaaWcbaGaemOAaOgabeaaaeaacqWGQbGAcqGH9aqpcqaIXaqmaeaacqWGNbWza0GaeyyeIuoaaaaaaa@450E@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>3. Minimum variance weights (denoted <italic>w</italic><sub>3</sub>) [<xref ref-type="bibr" rid="B3">3</xref>]:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M16" name="1471-2288-6-17-i10" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>g</mml:mi></mml:mrow><mml:mrow><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>g</mml:mi></mml:munderover><mml:mrow><mml:mfrac><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow></mml:mfrac></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrdaWgaaWcbaGaem4DaC3aaSbaaWqaaiabiodaZaqabaaaleqaaOGaeyypa0ZaaSaaaeaacuWGTbqBgaqeaiabdEgaNbqaamaaqahabaWaaSaaaeaacqWGTbqBdaWgaaWcbaGaemOAaOgabeaaaOqaaiabigdaXiabgUcaRmaabmaabaGaemyBa02aaSbaaSqaaiabdQgaQbqabaGccqGHsislcqaIXaqmaiaawIcacaGLPaaaiiGacqWFbpGCaaaaleaacqWGQbGAcqGH9aqpcqaIXaqmaeaacqWGNbWza0GaeyyeIuoaaaaaaa@4AB9@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>4. Distribution-based correction (denoted d) [<xref ref-type="bibr" rid="B4">4</xref>]:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M17" name="1471-2288-6-17-i11" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mi>d</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mi>E</mml:mi><mml:msup><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mi>m</mml:mi><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mn>2</mml:mn></mml:msup><mml:mo>+</mml:mo><mml:mi>var</mml:mi><mml:mo>⁡</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mi>m</mml:mi><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mrow><mml:mi>E</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mi>m</mml:mi><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mfrac><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrdaWgaaWcbaGaemizaqgabeaakiabg2da9iabigdaXiabgUcaRmaabmaabaWaaSaaaeaacqWGfbqrdaqadaqaaiabd2gaTbGaayjkaiaawMcaamaaCaaaleqabaGaeGOmaidaaOGaey4kaSIagiODayNaeiyyaeMaeiOCai3aaeWaaeaacqWGTbqBaiaawIcacaGLPaaaaeaacqWGfbqrdaqadaqaaiabd2gaTbGaayjkaiaawMcaaaaacqGHsislcqaIXaqmaiaawIcacaGLPaaaiiGacqWFbpGCaaa@4ACE@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>where E(<italic>m</italic>) and var(<italic>m</italic>) are the expected mean and the variance of the cluster size.</p><p>We considered these 4 adjustments when a Pareto's imbalance is <italic>a priori </italic>supposed to be observed. Since moderate imbalances have been shown to be of no influence, we assumed a constant cluster size within each stratum associated with the Pareto's imbalance. The adjusted VIF then becomes (Appendix A):</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M18" name="1471-2288-6-17-i12" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mn>3.25</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>3.25</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi><mml:mtext>     </mml:mtext><mml:mrow><mml:mo>(</mml:mo><mml:mn>5</mml:mn><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrdaWgaaWcbaGaem4DaC3aaSbaaWqaaiabigdaXaqabaaaleqaaOGaeyypa0JaeG4mamJaeiOla4IaeGOmaiJaeGynauJaey4kaSYaaeWaaeaacuWGTbqBgaqeamaaBaaaleaacqWG3bWDdaWgaaadbaGaeGymaedabeaaaSqabaGccqGHsislcqaIZaWmcqGGUaGlcqaIYaGmcqaI1aqnaiaawIcacaGLPaaaiiGacqWFbpGCcaWLjaGaaCzcamaabmaabaGaeGynaudacaGLOaGaayzkaaaaaa@48A4@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>with <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M19" name="1471-2288-6-17-i13" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mn>6.5</mml:mn><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mrow><mml:mi>g</mml:mi><mml:mi>E</mml:mi><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>−</mml:mo><mml:mn>2</mml:mn><mml:mi>ρ</mml:mi><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mfrac></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeamaaBaaaleaacqWG3bWDdaWgaaadbaGaeGymaedabeaaaSqabaGccqGH9aqpdaWcaaqaaiabiAda2iabc6caUiabiwda1maabmaabaGaeGymaeJaeyOeI0ccciGae8xWdihacaGLOaGaayzkaaGaemivaq1aaWbaaSqabeaacqaIYaGmaaaakeaacqWGNbWzcqWGfbqrcqWGtbWudaahaaWcbeqaaiabikdaYaaakiabgkHiTiabikdaYiab=f8aYjabdsfaunaaCaaaleqabaGaeGOmaidaaaaaaaa@471B@</mml:annotation></mml:semantics></mml:math></inline-formula> and <italic>T </italic>= <italic>t</italic><sub>(1 - <italic>α</italic>/2),2(<italic>g </italic>- 1) </sub>+ <italic>t</italic><sub>(1 - <italic>β</italic>),2(<italic>g </italic>- 1)</sub></p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M20" name="1471-2288-6-17-i14" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>3.25</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi><mml:mtext>     </mml:mtext><mml:mrow><mml:mo>(</mml:mo><mml:mn>6</mml:mn><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrdaWgaaWcbaGaem4DaC3aaSbaaWqaaiabikdaYaqabaaaleqaaOGaeyypa0JaeGymaeJaey4kaSYaaeWaaeaacqaIZaWmcqGGUaGlcqaIYaGmcqaI1aqncuWGTbqBgaqeamaaBaaaleaacqWG3bWDdaWgaaadbaGaeGOmaidabeaaaSqabaGccqGHsislcqaIXaqmaiaawIcacaGLPaaaiiGacqWFbpGCcaWLjaGaaCzcamaabmaabaGaeGOnaydacaGLOaGaayzkaaaaaa@46C8@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>with <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M21" name="1471-2288-6-17-i15" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mn>2</mml:mn><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mrow><mml:mi>g</mml:mi><mml:mi>E</mml:mi><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>−</mml:mo><mml:mn>6.5</mml:mn><mml:mi>ρ</mml:mi><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mfrac></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeamaaBaaaleaacqWG3bWDdaWgaaadbaGaeGOmaidabeaaaSqabaGccqGH9aqpdaWcaaqaaiabikdaYmaabmaabaGaeGymaeJaeyOeI0ccciGae8xWdihacaGLOaGaayzkaaGaemivaq1aaWbaaSqabeaacqaIYaGmaaaakeaacqWGNbWzcqWGfbqrcqWGtbWudaahaaWcbeqaaiabikdaYaaakiabgkHiTiabiAda2iabc6caUiabiwda1iab=f8aYjabdsfaunaaCaaaleqabaGaeGOmaidaaaaaaaa@471D@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M22" name="1471-2288-6-17-i16" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>4</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>0.25</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow></mml:mfrac><mml:mtext>     </mml:mtext><mml:mrow><mml:mo>(</mml:mo><mml:mn>7</mml:mn><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrdaWgaaWcbaGaem4DaC3aaSbaaWqaaiabiodaZaqabaaaleqaaOGaeyypa0ZaaSaaaeaadaqadaqaaiabigdaXiabgUcaRmaabmaabaGaeGinaqJafmyBa0MbaebadaWgaaWcbaGaem4DaC3aaSbaaWqaaiabiodaZaqabaaaleqaaOGaeyOeI0IaeGymaedacaGLOaGaayzkaaacciGae8xWdihacaGLOaGaayzkaaWaaeWaaeaacqaIXaqmcqGHRaWkdaqadaqaaiabicdaWiabc6caUiabikdaYiabiwda1iqbd2gaTzaaraWaaSbaaSqaaiabdEha3naaBaaameaacqaIZaWmaeqaaaWcbeaakiabgkHiTiabigdaXaGaayjkaiaawMcaaiab=f8aYbGaayjkaiaawMcaaaqaaiabigdaXiabgUcaRmaabmaabaGafmyBa0MbaebadaWgaaWcbaGaem4DaC3aaSbaaWqaaiabiodaZaqabaaaleqaaOGaeyOeI0IaeGymaedacaGLOaGaayzkaaGae8xWdihaaiaaxMaacaWLjaWaaeWaaeaacqaI3aWnaiaawIcacaGLPaaaaaa@616E@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>with <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M23" name="1471-2288-6-17-i17" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeamaaBaaaleaacqWG3bWDdaWgaaadbaGaeG4mamdabeaaaSqabaaaaa@30F6@</mml:annotation></mml:semantics></mml:math></inline-formula> being the positive solution of the following equation:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M24" name="1471-2288-6-17-i18" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow><mml:mn>2</mml:mn></mml:msubsup><mml:mi>ρ</mml:mi><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mi>g</mml:mi><mml:mi>E</mml:mi><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>−</mml:mo><mml:mn>2</mml:mn><mml:mi>ρ</mml:mi><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mo>]</mml:mo></mml:mrow><mml:mo>+</mml:mo><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mi>g</mml:mi><mml:mi>E</mml:mi><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>−</mml:mo><mml:mn>8.5</mml:mn><mml:mi>ρ</mml:mi><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mo>]</mml:mo></mml:mrow><mml:mo>−</mml:mo><mml:mn>2</mml:mn><mml:msup><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mn>2</mml:mn></mml:msup><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>=</mml:mo><mml:mn>0</mml:mn></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeamaaDaaaleaacqWG3bWDdaWgaaadbaGaeG4mamdabeaaaSqaaiabikdaYaaaiiGakiab=f8aYnaadmaabaGaem4zaCMaemyrauKaem4uam1aaWbaaSqabeaacqaIYaGmaaGccqGHsislcqaIYaGmcqWFbpGCcqWGubavdaahaaWcbeqaaiabikdaYaaaaOGaay5waiaaw2faaiabgUcaRiqbd2gaTzaaraWaaSbaaSqaaiabdEha3naaBaaameaacqaIZaWmaeqaaaWcbeaakmaabmaabaGaeGymaeJaeyOeI0Iae8xWdihacaGLOaGaayzkaaWaamWaaeaacqWGNbWzcqWGfbqrcqWGtbWudaahaaWcbeqaaiabikdaYaaakiabgkHiTiabiIda4iabc6caUiabiwda1iab=f8aYjabdsfaunaaCaaaleqabaGaeGOmaidaaaGccaGLBbGaayzxaaGaeyOeI0IaeGOmaiZaaeWaaeaacqaIXaqmcqGHsislcqWFbpGCaiaawIcacaGLPaaadaahaaWcbeqaaiabikdaYaaakiabdsfaunaaCaaaleqabaGaeGOmaidaaOGaeyypa0JaeGimaadaaa@65C0@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M25" name="1471-2288-6-17-i19" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mi>d</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mn>3.25</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>d</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>]</mml:mo></mml:mrow><mml:mi>ρ</mml:mi><mml:mtext>     </mml:mtext><mml:mrow><mml:mo>(</mml:mo><mml:mn>8</mml:mn><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrdaWgaaWcbaGaemizaqgabeaakiabg2da9iabigdaXiabgUcaRmaadmaabaGaeG4mamJaeiOla4IaeGOmaiJaeGynauJafmyBa0MbaebadaWgaaWcbaGaemizaqgabeaakiabgkHiTiabigdaXaGaay5waiaaw2faaGGaciab=f8aYjaaxMaacaWLjaWaaeWaaeaacqaI4aaoaiaawIcacaGLPaaaaaa@4495@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>with <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M26" name="1471-2288-6-17-i20" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>d</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mn>2</mml:mn><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mrow><mml:mi>g</mml:mi><mml:mi>E</mml:mi><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>−</mml:mo><mml:mn>6.5</mml:mn><mml:mi>ρ</mml:mi><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mfrac></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeamaaBaaaleaacqWGKbazaeqaaOGaeyypa0ZaaSaaaeaacqaIYaGmdaqadaqaaiabigdaXiabgkHiTGGaciab=f8aYbGaayjkaiaawMcaaiabdsfaunaaCaaaleqabaGaeGOmaidaaaGcbaGaem4zaCMaemyrauKaem4uam1aaWbaaSqabeaacqaIYaGmaaGccqGHsislcqaI2aGncqGGUaGlcqaI1aqncqWFbpGCcqWGubavdaahaaWcbeqaaiabikdaYaaaaaaaaa@45CD@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>The distribution-based and cluster size weights correction are equivalent [<xref ref-type="bibr" rid="B21">21</xref>]. We therefore no longer consider the distribution-based correction and focus on the 3 weighted corrections proposed by Kerry <italic>et al </italic>[<xref ref-type="bibr" rid="B3">3</xref>].</p></sec><sec><title>Simulation study</title><p>Monte Carlo simulations were performed to determine to what extent the proposed corrections could lead to adequately powered trials. We thus calculated the sample size needed assuming a Pareto repartition, using each of the adjusted VIFs. For each situation, we then simulated cluster randomized trials with a Pareto imbalance to estimate empirical type I error and power. The same approach as that explained in the preceeding was used.</p></sec><sec><title>Results</title><p>Results are displayed in Tables <xref ref-type="table" rid="T3">3</xref> and <xref ref-type="table" rid="T4">4</xref> for effect sizes of 0.25 and 0.50, respectively. For the cluster size weights correction, several situations existed in which the sample size calculations showed that 80% power could not be reached, thus preventing the generation of associated data sets. If sample size calculations were possible, this correction led to sample sizes barely greater than the sample size obtained with the minimal variance weights correction and empirical type I error and power near the nominal value. This result is consistent for the different values of ES, <italic>ρ </italic>and g in Tables <xref ref-type="table" rid="T3">3</xref> and <xref ref-type="table" rid="T4">4</xref>, except for the combination 0.25/0.02/20. Actually, for fixed values of ES, couples of values for (g, <italic>ρ</italic>) lead to null values of the denominator of <italic>m</italic><sub><italic>w</italic>2</sub>. If ES is fixed at 0.25, the couple (20, 0.0233) is one of these. For <italic>ρ </italic>just under this critical value (0.020 in our case), m<sub>w2 </sub>begins to diverge, and when <italic>ρ </italic>is greater, m<sub>w2 </sub>can no longer be calculated. Equal weights correction led to a much greater sample size than minimum variance weights, particularly when the ICC is small, and the empirical power obtained was therefore much higher than its nominal value: it may even reach 99% if the nominal value were fixed at 80%. The minimum variance weights correction required the smallest increase in sample size and resulted in the smallest difference between empirical and nominal power. Empirical type I errors were also near the nominal 5% level, except when both the number of clusters and the ICC are small.</p></sec></sec><sec><title>Robustness of sample size adjustment for unbalanced trials with misspecification of the ICC</title><sec sec-type="methods"><title>Method</title><p>We assessed the robustness of the different sample size adjustments for Pareto-like unbalanced trials with misspecification of the ICC. We considered an effect size of 0.25, <italic>a priori </italic>postulated ICCs of 0.005 and 0.020 and the combinations of number of clusters and cluster sizes previously used (see sample sizes in Table <xref ref-type="table" rid="T3">3</xref>). Then, for each weighting method, (i.e., for each total number of subjects of each arm N<sub>w1</sub>, N<sub>w2</sub>, N<sub>w3</sub>) we plotted the expected power calculated for a pre-specified ICC as a function of the real ICC (which will be <italic>a posteriori </italic>assessed). This power was calculated by use of the variance inflation factor VIF<sub>w3 </sub>derived from minimum variance weights, because it allows for calculating an expected power that does not differ from the empirical one by more than 3.8% in the situations explored in Table <xref ref-type="table" rid="T3">3</xref> (data not shown). For reference, we also plotted the expected power (calculated with the usual VIF) as a function of the real ICC in cases of no imbalance in cluster size.</p><table-wrap position="float" id="T3"><label>Table 3</label><caption><p>Required sample size and empirical Type I error and power when using corrected variance inflation factors with an <italic>a priori </italic>hypothesized Pareto imbalance in cluster size – Effect size = 0.25</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td></td><td align="center" colspan="3"><bold>No correction</bold></td><td align="center" colspan="3"><bold>Equal weights</bold></td><td align="center" colspan="3"><bold>Cluster size weights</bold><sup>1</sup></td><td align="center" colspan="3"><bold>Minimum variance weights</bold></td></tr></thead><tbody><tr><td align="center"><bold>Intraclass correlation coefficient (<italic>ρ</italic>)</bold></td><td align="center"><bold>Number of clusters in each arm (<italic>g</italic>)</bold></td><td align="center"><bold>Sample size</bold></td><td align="center" colspan="2"><bold>Empirical probabilities</bold></td><td align="center"><bold>Sample size</bold></td><td align="center" colspan="2"><bold>Empirical probabilities</bold></td><td align="center"><bold>Sample size</bold></td><td align="center" colspan="2"><bold>Empirical probabilities</bold></td><td align="center"><bold>Sample size</bold></td><td align="center" colspan="2"><bold>Empirical probabilities</bold></td></tr><tr><td></td><td></td><td></td><td colspan="2"><hr></hr></td><td></td><td colspan="2"><hr></hr></td><td></td><td colspan="2"><hr></hr></td><td></td><td colspan="2"><hr></hr></td></tr><tr><td></td><td></td><td></td><td align="center"><bold>Type I error</bold></td><td align="center"><bold>Power</bold></td><td></td><td align="center"><bold>Type I error</bold></td><td align="center"><bold>Power</bold></td><td></td><td align="center"><bold>Type I error</bold></td><td align="center"><bold>Power</bold></td><td></td><td align="center"><bold>Type I error</bold></td><td align="center"><bold>Power</bold></td></tr><tr><td colspan="14"><hr></hr></td></tr><tr><td align="center"><bold>0.005</bold></td><td align="center"><bold>5</bold></td><td align="center">485</td><td align="center">0.0664</td><td align="center">0.6432</td><td align="center">1569</td><td align="center">0.1028</td><td align="center">0.8606</td><td align="center">-</td><td align="center">-</td><td align="center">-</td><td align="center">1037</td><td align="center">0.0948</td><td align="center">0.7992</td></tr><tr><td></td><td align="center"><bold>10</bold></td><td align="center">326</td><td align="center">0.0566</td><td align="center">0.6968</td><td align="center">1057</td><td align="center">0.0784</td><td align="center">0.9386</td><td align="center">515</td><td align="center">0.0704</td><td align="center">0.8106</td><td align="center">464</td><td align="center">0.0704</td><td align="center">0.7806</td></tr><tr><td></td><td align="center"><bold>20</bold></td><td align="center">282</td><td align="center">0.0486</td><td align="center">0.7258</td><td align="center">917</td><td align="center">0.0624</td><td align="center">0.9770</td><td align="center">336</td><td align="center">0.0450</td><td align="center">0.7934</td><td align="center">331</td><td align="center">0.0458</td><td align="center">0.7850</td></tr><tr><td></td><td align="center"><bold>40</bold></td><td align="center">265</td><td align="center">0.0466</td><td align="center">0.7572</td><td align="center">861</td><td align="center">0.0512</td><td align="center">0.9918</td><td align="center">287</td><td align="center">0.0424</td><td align="center">0.7842</td><td align="center">286</td><td align="center">0.0474</td><td align="center">0.7706</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.02</bold></td><td align="center"><bold>10</bold></td><td align="center">629</td><td align="center">0.0904</td><td align="center">0.6236</td><td align="center">2043</td><td align="center">0.0638</td><td align="center">0.8196</td><td align="center">-</td><td align="center">-</td><td align="center">-</td><td align="center">1731</td><td align="center">0.0624</td><td align="center">0.7968</td></tr><tr><td></td><td align="center">20</td><td align="center">353</td><td align="center">0.0660</td><td align="center">0.6546</td><td align="center">1147</td><td align="center">0.0614</td><td align="center">0.8924</td><td align="center">1852</td><td align="center">0.0576</td><td align="center">0.9432</td><td align="center">677</td><td align="center">0.0752</td><td align="center">0.7976</td></tr><tr><td></td><td align="center"><bold>40</bold></td><td align="center">290</td><td align="center">0.0556</td><td align="center">0.7008</td><td align="center">942</td><td align="center">0.0582</td><td align="center">0.9486</td><td align="center">435</td><td align="center">0.0558</td><td align="center">0.8092</td><td align="center">401</td><td align="center">0.0514</td><td align="center">0.7960</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.05</bold></td><td align="center"><bold>20</bold></td><td align="center">743</td><td align="center">0.0562</td><td align="center">0.6256</td><td align="center">2414</td><td align="center">0.0564</td><td align="center">0.8140</td><td align="center">-</td><td align="center">-</td><td align="center">-</td><td align="center">2165</td><td align="center">0.0480</td><td align="center">0.7976</td></tr><tr><td></td><td align="center"><bold>40</bold></td><td align="center">361</td><td align="center">0.0604</td><td align="center">0.6242</td><td align="center">1173</td><td align="center">0.0500</td><td align="center">0.8598</td><td align="center">-</td><td align="center">-</td><td align="center">-</td><td align="center">770</td><td align="center">0.0550</td><td align="center">0.8048</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td></td></tr><tr><td align="center"><bold>0.10</bold></td><td align="center">40</td><td align="center">652</td><td align="center">0.0546</td><td align="center">0.6006</td><td align="center">2116</td><td align="center">0.0542</td><td align="center">0.8090</td><td align="center">-</td><td align="center">-</td><td align="center">-</td><td align="center">1881</td><td align="center">0.0500</td><td align="center">0.8036</td></tr></tbody></table><table-wrap-foot><p>Sample size calculations were performed considering type I and type II error rates fixed at 0.05 and 0.20, respectively.</p><p><sup>1 </sup>In some cases, 80% power was not reachable</p></table-wrap-foot></table-wrap><table-wrap position="float" id="T4"><label>Table 4</label><caption><p>Required sample size and empirical Type I error and power when using corrected variance inflation factors with an <italic>a priori </italic>hypothesized Pareto imbalance in cluster size – Effect size = 0.50</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td></td><td align="center" colspan="3"><bold>No correction</bold></td><td align="center" colspan="3"><bold>Equal weights</bold></td><td align="center" colspan="3"><bold>Cluster size weights</bold><sup>1</sup></td><td align="center" colspan="3"><bold>Minimum variance weights</bold></td></tr></thead><tbody><tr><td align="center"><bold>Intraclass correlation coefficient (<italic>ρ</italic>)</bold></td><td align="center"><bold>Number of clusters in each arm (<italic>g</italic>)</bold></td><td align="center"><bold>Sample size</bold></td><td align="center" colspan="2"><bold>Empirical probabilities</bold></td><td align="center"><bold>Sample size</bold></td><td align="center" colspan="2"><bold>Empirical probabilities</bold></td><td align="center"><bold>Sample size</bold></td><td align="center" colspan="2"><bold>Empirical probabilities</bold></td><td align="center"><bold>Sample size</bold></td><td align="center" colspan="2"><bold>Empirical probabilities</bold></td></tr><tr><td></td><td></td><td></td><td colspan="2"><hr></hr></td><td></td><td colspan="2"><hr></hr></td><td></td><td colspan="2"><hr></hr></td><td></td><td colspan="2"><hr></hr></td></tr><tr><td></td><td></td><td></td><td align="center"><bold>Type I error</bold></td><td align="center"><bold>Power</bold></td><td></td><td align="center"><bold>Type I error</bold></td><td align="center"><bold>Power</bold></td><td></td><td align="center"><bold>Type I error</bold></td><td align="center"><bold>Power</bold></td><td></td><td align="center"><bold>Type I error</bold></td><td align="center"><bold>Power</bold></td></tr><tr><td colspan="14"><hr></hr></td></tr><tr><td align="center"><bold>0.005</bold></td><td align="center"><bold>5</bold></td><td align="center">89</td><td align="center">0.0256</td><td align="center">0.6250</td><td align="center">288</td><td align="center">0.0536</td><td align="center">0.9330</td><td align="center">111</td><td align="center">0.0270</td><td align="center">0.7156</td><td align="center">108</td><td align="center">0.0324</td><td align="center">0.6906</td></tr><tr><td></td><td align="center"><bold>10</bold></td><td align="center">73</td><td align="center">0.0352</td><td align="center">0.7090</td><td align="center">236</td><td align="center">0.0528</td><td align="center">0.9768</td><td align="center">79</td><td align="center">0.0306</td><td align="center">0.7370</td><td align="center">79</td><td align="center">0.0306</td><td align="center">0.7370</td></tr><tr><td></td><td align="center"><bold>20</bold></td><td align="center">67</td><td align="center">0.0334</td><td align="center">0.7322</td><td align="center">218</td><td align="center">0.0470</td><td align="center">0.9912</td><td align="center">70</td><td align="center">0.0390</td><td align="center">0.7524</td><td align="center">70</td><td align="center">0.0390</td><td align="center">0.7524</td></tr><tr><td></td><td align="center"><bold>40</bold></td><td align="center">65</td><td align="center">0.0382</td><td align="center">0.7518</td><td align="center">210</td><td align="center">0.0394</td><td align="center">0.9970</td><td align="center">66</td><td align="center">0.0400</td><td align="center">0.7558</td><td align="center">66</td><td align="center">0.0400</td><td align="center">0.7558</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.02</bold></td><td align="center"><bold>5</bold></td><td align="center">119</td><td align="center">0.0674</td><td align="center">0.6262</td><td align="center">387</td><td align="center">0.1072</td><td align="center">0.8642</td><td align="center">-</td><td align="center">-</td><td align="center">-</td><td align="center">256</td><td align="center">0.0954</td><td align="center">0.7946</td></tr><tr><td></td><td align="center"><bold>10</bold></td><td align="center">81</td><td align="center">0.0550</td><td align="center">0.6910</td><td align="center">261</td><td align="center">0.0900</td><td align="center">0.9346</td><td align="center">127</td><td align="center">0.0654</td><td align="center">0.7990</td><td align="center">115</td><td align="center">0.0672</td><td align="center">0.7856</td></tr><tr><td></td><td align="center"><bold>20</bold></td><td align="center">70</td><td align="center">0.0460</td><td align="center">0.7056</td><td align="center">226</td><td align="center">0.0684</td><td align="center">0.9752</td><td align="center">83</td><td align="center">0.0492</td><td align="center">0.7680</td><td align="center">82</td><td align="center">0.0482</td><td align="center">0.7680</td></tr><tr><td></td><td align="center"><bold>40</bold></td><td align="center">66</td><td align="center">0.0422</td><td align="center">0.7328</td><td align="center">212</td><td align="center">0.0534</td><td align="center">0.9908</td><td align="center">71</td><td align="center">0.0390</td><td align="center">0.7540</td><td align="center">71</td><td align="center">0.0390</td><td align="center">0.7540</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.05</bold></td><td align="center"><bold>5</bold></td><td align="center">423</td><td align="center">0.0768</td><td align="center">0.6808</td><td align="center">1375</td><td align="center">0.0578</td><td align="center">0.8130</td><td align="center">-</td><td align="center">-</td><td align="center">-</td><td align="center">1311</td><td align="center">0.0556</td><td align="center">0.7962</td></tr><tr><td></td><td align="center"><bold>10</bold></td><td align="center">103</td><td align="center">0.0770</td><td align="center">0.6342</td><td align="center">335</td><td align="center">0.0824</td><td align="center">0.8600</td><td align="center">-</td><td align="center">-</td><td align="center">-</td><td align="center">230</td><td align="center">0.0920</td><td align="center">0.7952</td></tr><tr><td></td><td align="center"><bold>20</bold></td><td align="center">76</td><td align="center">0.0528</td><td align="center">0.6620</td><td align="center">245</td><td align="center">0.0652</td><td align="center">0.9284</td><td align="center">136</td><td align="center">0.0706</td><td align="center">0.8252</td><td align="center">115</td><td align="center">0.0628</td><td align="center">0.7872</td></tr><tr><td></td><td align="center"><bold>40</bold></td><td align="center">67</td><td align="center">0.0516</td><td align="center">0.7026</td><td align="center">217</td><td align="center">0.0590</td><td align="center">0.9712</td><td align="center">83</td><td align="center">0.0578</td><td align="center">0.7694</td><td align="center">81</td><td align="center">0.0488</td><td align="center">0.7772</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="center"><bold>0.15</bold></td><td align="center"><bold>10</bold></td><td align="center">213</td><td align="center">0.0730</td><td align="center">0.6394</td><td align="center">691</td><td align="center">0.0548</td><td align="center">0.8042</td><td align="center">-</td><td align="center">-</td><td align="center">-</td><td align="center">631</td><td align="center">0.0572</td><td align="center">0.8002</td></tr><tr><td></td><td align="center"><bold>20</bold></td><td align="center">89</td><td align="center">0.0674</td><td align="center">0.6276</td><td align="center">290</td><td align="center">0.0644</td><td align="center">0.8646</td><td align="center">-</td><td align="center">-</td><td align="center">-</td><td align="center">193</td><td align="center">0.0638</td><td align="center">0.7888</td></tr><tr><td></td><td align="center"><bold>40</bold></td><td align="center">70</td><td align="center">0.0536</td><td align="center">0.6641</td><td align="center">225</td><td align="center">0.0564</td><td align="center">0.9316</td><td align="center">122</td><td align="center">0.0506</td><td align="center">0.8208</td><td align="center">104</td><td align="center">0.0578</td><td align="center">0.7838</td></tr></tbody></table><table-wrap-foot><p><sup>1</sup>In some cases, 80% power was not reachable</p></table-wrap-foot></table-wrap></sec></sec></sec><sec><title>Results</title><p>Results are displayed in Figures <xref ref-type="fig" rid="F1">1</xref> and <xref ref-type="fig" rid="F2">2</xref> for an effect size of 0.25 and <italic>a priori </italic>postulated ICC values of 0.005 and 0.020, respectively. As expected [<xref ref-type="bibr" rid="B20">20</xref>], in any situation, the power decreases as the ICC increases, and this result is all the more important when the number of clusters is low. In the planning situations explored, minimum variance weights and cluster size weights curves are very close, except when 20 clusters per intervention arm are randomized and the ICC is <italic>a priori </italic>fixed at 0.020, but this latter situation is extreme, as discussed previously. Otherwise, the power associated with equal weights remains greater than that associated with minimum variance weights in any situation. However, this finding probably just reflects that the use of this weighting system leads to higher required sample sizes than the use of a minimum variance weights system (cf Tables <xref ref-type="table" rid="T3">3</xref> and <xref ref-type="table" rid="T4">4</xref>) and therefore higher power. In any case, imbalance in cluster size is associated with a higher sensitivity to the a priori-specified ICC than constant cluster size. For example, let us consider the case of 20 clusters per intervention arm: if the ICC is <italic>a priori </italic>postulated at 0.005, but in reality equals 0.015, the power associated with constant cluster size decreases from 0.80 to 0.75 only, whereas the power associated with Pareto repartition decreases from 0.80 to 0.68 (with the minimum variance weighting system). However, all weighting systems show great sensitivity to the actual value of the ICC. Consider the former example (ES = 0.25, g = 20 and Pareto repartition, increase in ICC from 0.005 to 0.015), the power associated with equal weights will decrease from 0.98 to 0.90, and the power associated with cluster size weights from 0.80 to 0.68. Thus, if little prior knowledge is available concerning the value of the ICC, the sensitivity analysis involving several values of ICC is of major importance, particularly when imbalance in cluster size is expected.</p><sec><title>Practical implications</title><sec><title>General considerations</title><p>Cluster size inequality may induce a loss of power and must be taken into account at the planning stage by using the minimum variance weights correction. From a practical point of view, 2 situations must be distinguished. First, when entire clusters are randomized such as in cluster-cluster trials [<xref ref-type="bibr" rid="B22">22</xref>]. the cluster size distribution is <italic>a priori </italic>known and cluster size inequalities are therefore easy to be taken into account at the planning stage. Second, if physicians have to recruit patients to each cluster according to selection criteria, cluster size distribution cannot <italic>a priori </italic>be known. In this latter situation, a sensitivity analysis must be performed considering several hypotheses on cluster size distribution for an optimal sample size determination.</p></sec><sec><title>Adaptation of the VIF for a Pareto like imbalance</title><p>Let us assume that the cluster size inequality corresponds to a Pareto-like distribution, say that in each arm a proportion (<italic>γ</italic>) of clusters actually recruit the proportion (<italic>τ</italic>) of patients to be recruited (which implies <italic>γ </italic>≤ <italic>τ</italic>). If <italic>γ </italic>and <italic>τ </italic>are fixed at 20% and 80%, respectively, we have the Pareto imbalance defined previously; if <italic>γ </italic>and <italic>τ </italic>are equal, the cluster size imbalance is absent or moderate (and can then be neglected). The sensitivity analysis then consists of varying the parameters (<italic>γ</italic>) and (<italic>τ</italic>), thus allowing for imbalance increases with the absolute difference between the 2 values. The inflation factor calculated with the minimum variance weights correction will be the following (Appendix B):</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M27" name="1471-2288-6-17-i21" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:mi>F</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>τ</mml:mi></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow></mml:mfrac><mml:mi>m</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>]</mml:mo></mml:mrow><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mi>τ</mml:mi><mml:mi>γ</mml:mi></mml:mfrac><mml:mi>m</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>]</mml:mo></mml:mrow></mml:mrow><mml:mrow><mml:mi>τ</mml:mi><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>τ</mml:mi></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow></mml:mfrac><mml:mi>m</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>]</mml:mo></mml:mrow><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>τ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mi>τ</mml:mi><mml:mi>γ</mml:mi></mml:mfrac><mml:mi>m</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>]</mml:mo></mml:mrow></mml:mrow></mml:mfrac><mml:mtext>     </mml:mtext><mml:mrow><mml:mo>(</mml:mo><mml:mn>9</mml:mn><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrcqGH9aqpdaWcaaqaamaadmaabaGaeGymaeJaey4kaSYaaeWaaeaadaWcaaqaaiabigdaXiabgkHiTGGaciab=r8a0bqaaiabigdaXiabgkHiTiab=n7aNbaacqWGTbqBcqGHsislcqaIXaqmaiaawIcacaGLPaaacqWFbpGCaiaawUfacaGLDbaadaWadaqaaiabigdaXiabgUcaRmaabmaabaWaaSaaaeaacqWFepaDaeaacqWFZoWzaaGaemyBa0MaeyOeI0IaeGymaedacaGLOaGaayzkaaGae8xWdihacaGLBbGaayzxaaaabaGae8hXdq3aamWaaeaacqaIXaqmcqGHRaWkdaqadaqaamaalaaabaGaeGymaeJaeyOeI0Iae8hXdqhabaGaeGymaeJaeyOeI0Iae83SdCgaaiabd2gaTjabgkHiTiabigdaXaGaayjkaiaawMcaaiab=f8aYbGaay5waiaaw2faaiabgUcaRmaabmaabaGaeGymaeJaeyOeI0Iae8hXdqhacaGLOaGaayzkaaWaamWaaeaacqaIXaqmcqGHRaWkdaqadaqaamaalaaabaGae8hXdqhabaGae83SdCgaaiabd2gaTjabgkHiTiabigdaXaGaayjkaiaawMcaaiab=f8aYbGaay5waiaaw2faaaaacaWLjaGaaCzcamaabmaabaGaeGyoaKdacaGLOaGaayzkaaaaaa@7B24@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>To illustrate the discrepancy between nominal and real power if an imbalance of the form "<italic>γ </italic>clusters actually recruit <italic>τ </italic>patients" is not taken into account, we performed the following calculations. We used formula (4) (i.e., assuming a constant cluster size) to derive the number of subjects needed. Then, using expression (9), we calculated the expected power with such a sample size, with a proportion of <italic>γ </italic>clusters actually recruiting a proportion <italic>τ </italic>of the patients to be included.</p><fig position="float" id="F1"><label>Figure 1</label><caption><p>Real power of cluster randomized trials according to the discrepancy between the <italic>a priori </italic>postulated and <italic>a posteriori </italic>estimated intraclass correlation coefficients (ICCs). The ICC is <italic>a priori </italic>postulated at 0.005 and sample sizes (N) and associated powers were calculated: 1°) assuming Pareto repartition of cluster sizes and using 3 corrections of the variance inflation factor (equal weights, cluster size weights and minimum variance weights), 2°) assuming constant cluster size (reference).</p></caption><graphic xlink:href="1471-2288-6-17-1"/></fig><fig position="float" id="F2"><label>Figure 2</label><caption><p>Real power of cluster randomized trials according to the discrepancy between the <italic>a priori </italic>postulated and <italic>a posteriori </italic>estimated intraclass correlation coefficients (ICCs). The ICC is <italic>a priori </italic>postulated at 0.020 and sample sizes (N) and associated powers were calculated: 1°) assuming Pareto repartition of cluster sizes and using 3 corrections of the variance inflation factor (equal weights, cluster size weights and minimum variance weights), 2°) assuming constant cluster size (reference).</p></caption><graphic xlink:href="1471-2288-6-17-2"/></fig><p>Figures <xref ref-type="fig" rid="F3">3</xref> and <xref ref-type="fig" rid="F4">4</xref> display the results for several combinations of ES/ICC/g and <italic>γ</italic>/<italic>τ </italic>under the assumption of no empty cluster. The upper part of Figures <xref ref-type="fig" rid="F3">3</xref> and <xref ref-type="fig" rid="F4">4</xref> is empty, since an 80% power cannot be reach for the associated combinations of ICC and g. Moreover, <italic>γ </italic>is smaller than or equal to <italic>τ</italic>, which explains why any upper part of matrices associated with an ICC/g combination is empty. As expected, the bigger the cluster size inequality, the more important the discrepancy between nominal and real power. For example, let us consider a trial aimed at detecting a 0.25 effect size in which 10 clusters are to be randomized in each arm. Assuming an ICC of 0.005 and a balance in cluster size, this study would require 326 subjects to be recruited in each arm to reach 80% power. If 10% of the clusters recruit 50% of the subjects, the power barely declines, to 77%; if a major imbalance such as 90% of the patients are to be recruited by 10% of the clusters, the power would fall to 54%. The latter phenomenon is all the more acute with a low number of clusters; critical situations in which a substantial loss in power may be expected are displayed in Figures <xref ref-type="fig" rid="F3">3</xref> and <xref ref-type="fig" rid="F4">4</xref>. Red levels approximately follow diagonals representing constant <italic>τ</italic>-<italic>γ </italic>differences. It can be shown (appendix C) that the gini coefficient, a quantitative measure of site accrual inequality [<xref ref-type="bibr" rid="B23">23</xref>], comes down to the absolute difference between <italic>τ </italic>and <italic>γ </italic>when a proportion <italic>γ </italic>of clusters actually recruit a proportion <italic>τ </italic>of patients to be recruited. Our results show that varying this summary measure of imbalance is enough for performing a sensitivity analysis and that there is no need to specify both <italic>τ </italic>and <italic>γ</italic>.</p><fig position="float" id="F3"><label>Figure 3</label><caption><p>Power of cluster randomized trials if an imbalance in cluster size is not taken into account when planning. The imbalance is <italic>a priori </italic>hypothesized to be "a proportion of <italic>γ </italic>clusters will actually recruit a proportion <italic>τ </italic>of the subjects to be included" (<italic>γ </italic>≤ <italic>τ</italic>) – The intraclass correlation coefficient is fixed at 0.005 and 0.02.</p></caption><graphic xlink:href="1471-2288-6-17-3"/></fig><fig position="float" id="F4"><label>Figure 4</label><caption><p>Power of cluster randomized trials if an imbalance in cluster size is not taken into account when planning. The imbalance is <italic>a priori </italic>hypothesized to be "a proportion of <italic>γ </italic>clusters will actually recruit a proportion <italic>τ </italic>of the subjects to be included" (<italic>γ </italic>≤ <italic>τ</italic>) – The intraclass correlation coefficient is fixed at 0.05 and 0.10.</p></caption><graphic xlink:href="1471-2288-6-17-4"/></fig><p>Assigning a value of 1 to <italic>τ </italic>creates a situation in which a proportion (1-<italic>γ</italic>) of clusters is empty. In this situation achieving the required sample size supposes to increase the average cluster size of the <italic>γ</italic>g clusters by a factor 1/<italic>γ</italic>. However one has to be aware that such a strategy will indeed allow achieving the pre-specified sample size, but it will not allow to reach the nominal power. Indeed it is known that for a fixed total number of subjects, the higher the number of clusters, the higher the power [<xref ref-type="bibr" rid="B1">1</xref>] which means that reducing the number of clusters will translate in a loss in power even if the pre-specified sample size is achieved. Therefore, in case it is anticipated that empty clusters may occur, sensitivity analyses have to be conducted using formula (4) on the basis of the hypothesized number of active clusters g' = <italic>γ</italic>g.</p></sec></sec></sec><sec><title>Discussion</title><p>A moderate inequality in cluster sizes has little effect on power and can thus be neglected at the planning stage. However, a major imbalance in cluster sizes, like the "Pareto" imbalance, (i.e. 80% of the subjects belong to only 20% of the clusters) is associated with a loss in power, and the phenomenon is all the more important when the number of clusters is low and/or the ICC is high. In these situations, the minimum variance weights correction has good properties and allows for achieving the nominal power. This result, obtained in the extreme situation of a Pareto imbalance, suggests that this correction can be used to derive sample size or power in any situation where, in each group, cluster sizes can be separated in two strata, the small cluster stratum and the big cluster stratum. The higher sensitivity of severely unbalanced trials to the <italic>a priori</italic>-postulated value of the ICC compared to that of balanced trials emphasized the necessity of a sensitivity analysis on this parameter. We derived an adaptation of the VIF, which should be used when the imbalance is <italic>a priori </italic>hypothesized to be "a proportion of <italic>γ </italic>clusters will actually recruit a proportion <italic>τ </italic>of the subjects to be included".</p><p>A limit to this approach remains the degree of imbalance being usually difficult to foresee at the planning stage, except when, for instance, families or practices are randomized and clusters as a whole are included in the trial. In these latter situations, one may <italic>a priori </italic>know precisely the cluster size repartition and therefore use the minimum variance weights correction as initially specified by Kerry et al [<xref ref-type="bibr" rid="B3">3</xref>]. However, if, within each cluster, the physician has to recruit patients to be included in the trial, cluster size distribution may then be difficult to hypothesize. It is all the more difficult since cluster sizes are usually not reported in published clustered randomized trials. We therefore proposed to consider that cluster sizes distribution can be divided in each arm in two strata: a stratum of small clusters, and another of large clusters. This hypothesis may be debatable. However, since a moderate inequality of cluster size is of minor effect, it seems a rather useful and simple way to consider the risk of cluster size inequality at the planning stage, particularly since no precise data on cluster size inequality are available. Another limitation is that our work focused on normally distributed continuous outcomes. More work is needed to extend our results to non-normal distributions, especially with binary variables. Finally, we restricted our work to cases of no differential recruitment between arms, thus considering that imbalance is the same in the two arms. Such a hypothesis may be questionable in cluster randomized trials: since inclusion is posterior to randomization, this may indeed induce differential recruitment and imbalance in patient characteristics, which may lead to questioning the results of the study [<xref ref-type="bibr" rid="B24">24</xref>].</p></sec><sec><title>Conclusion</title><p>In conclusion, our study demonstrates that severely imbalanced trials with continuous outcomes may be highly underpowered. If such imbalance in cluster size can be anticipated at the design stage, minimum variance weights correction should be used to inflate the required sample size. <italic>A priori </italic>estimation of the expectable imbalance would be facilitated if more details on cluster sizes were given in published cluster randomized trials, as was recently advised in the extension of the CONSORT statement for cluster randomized trials [<xref ref-type="bibr" rid="B25">25</xref>]. Moreover, such publication of cluster sizes would be of particular interest to assess the real power of the trial conducted.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>This study was designed by LG, BG and PR. LG performed the statistical analysis and drafted the article, which was then revised by BG and PR.</p></sec><sec><title>Appendix A: corrected variance inflation factor (VIF) for an <italic>a priori </italic>postulated Pareto imbalance</title><p>Four corrections have been proposed for adjusting sample size in cases of imbalance in cluster size. Considering the specific situation of a Pareto imbalance, the general form of these corrections can be simplified.</p><p>Characteristics of the Pareto imbalance</p><table-wrap position="float" id="T5"><label>Table 5</label><table frame="hsides" rules="groups"><thead><tr><td></td><td align="left">Number of clusters by intervention arm</td><td align="center">Number of patients belonging to the clusters</td><td align="center">Mean cluster size</td></tr></thead><tbody><tr><td align="left">Small clusters</td><td align="center">0.8<italic>g</italic></td><td align="center">0.2<inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M28" name="1471-2288-6-17-i22" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeaaaa@2E27@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>g</italic></td><td align="center">(0.2<inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M29" name="1471-2288-6-17-i22" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeaaaa@2E27@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>g</italic>)/(0.8<italic>g</italic>) = 0.25<inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M30" name="1471-2288-6-17-i22" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeaaaa@2E27@</mml:annotation></mml:semantics></mml:math></inline-formula></td></tr><tr><td align="left">Big clusters</td><td align="center">0.2<italic>g</italic></td><td align="center">0.8<inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M31" name="1471-2288-6-17-i22" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeaaaa@2E27@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>g</italic></td><td align="center">(0.8<inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M32" name="1471-2288-6-17-i22" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeaaaa@2E27@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>g</italic>)/(0.2<italic>g</italic>) = 4<inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M33" name="1471-2288-6-17-i22" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeaaaa@2E27@</mml:annotation></mml:semantics></mml:math></inline-formula></td></tr></tbody></table></table-wrap><p><italic>g </italic>refers to the number of clusters within each arm and <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M34" name="1471-2288-6-17-i22" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeaaaa@2E27@</mml:annotation></mml:semantics></mml:math></inline-formula> is the average cluster size</p><p><italic>Equal weights </italic>(<italic>denoted w</italic><sub>1</sub>) [<xref ref-type="bibr" rid="B3">3</xref>]</p><p>With an equal weights correction, the VIF is expressed as:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M35" name="1471-2288-6-17-i6" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>g</mml:mi></mml:mfrac><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>g</mml:mi></mml:munderover><mml:mrow><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow></mml:mfrac><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>+</mml:mo><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>ρ</mml:mi></mml:mrow></mml:mstyle></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrdaWgaaWcbaGaem4DaC3aaSbaaWqaaiabigdaXaqabaaaleqaaOGaeyypa0ZaaSaaaeaacuWGTbqBgaqeaaqaaiabdEgaNbaadaaeWbqaamaalaaabaGaeGymaedabaGaemyBa02aaSbaaSqaaiabdQgaQbqabaaaaOWaaeWaaeaacqaIXaqmcqGHsisliiGacqWFbpGCaiaawIcacaGLPaaacqGHRaWkcuWGTbqBgaqeaiab=f8aYbWcbaGaemOAaOMaeyypa0JaeGymaedabaGaem4zaCganiabggHiLdaaaa@4AF5@</mml:annotation></mml:semantics></mml:math></inline-formula> where <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M36" name="1471-2288-6-17-i7" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mi>g</mml:mi></mml:mfrac><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>g</mml:mi></mml:munderover><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow></mml:mstyle></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeaiabg2da9maalaaabaGaeGymaedabaGaem4zaCgaamaaqahabaGaemyBa02aaSbaaSqaaiabdQgaQbqabaaabaGaemOAaOMaeyypa0JaeGymaedabaGaem4zaCganiabggHiLdaaaa@3B51@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>With a Pareto imbalance, this equation is expressed as:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M37" name="1471-2288-6-17-i23" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mtable><mml:mtr><mml:mtd><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub></mml:mrow><mml:mi>g</mml:mi></mml:mfrac><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mn>0.8</mml:mn><mml:mi>g</mml:mi></mml:mrow><mml:mrow><mml:mn>0.25</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub></mml:mrow></mml:mfrac><mml:mo>+</mml:mo><mml:mfrac><mml:mrow><mml:mn>0.2</mml:mn><mml:mi>g</mml:mi></mml:mrow><mml:mrow><mml:mn>4</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub></mml:mrow></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>+</mml:mo><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mi>ρ</mml:mi></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mo>=</mml:mo><mml:mn>3</mml:mn><mml:mo>.</mml:mo><mml:mn>2</mml:mn><mml:mn>5</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>3.25</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mtd></mml:mtr></mml:mtable><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@6FAB@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>where <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M38" name="1471-2288-6-17-i24" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeamaaBaaaleaacqWG3bWDdaWgaaadbaGaeGymaedabeaaaSqabaaaaa@30F2@</mml:annotation></mml:semantics></mml:math></inline-formula> (the average cluster size for which an equal weights correction is used) is defined as:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M39" name="1471-2288-6-17-i25" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mi>g</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mn>2</mml:mn><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mrow><mml:mi>E</mml:mi><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mfrac><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mn>3.25</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>3.25</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>]</mml:mo></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeamaaBaaaleaacqWG3bWDdaWgaaadbaGaeGymaedabeaaaSqabaGccqWGNbWzcqGH9aqpdaWcaaqaaiabikdaYiabdsfaunaaCaaaleqabaGaeGOmaidaaaGcbaGaemyrauKaem4uam1aaWbaaSqabeaacqaIYaGmaaaaaOWaamWaaeaacqaIZaWmcqGGUaGlcqaIYaGmcqaI1aqncqGHRaWkdaqadaqaaiqbd2gaTzaaraWaaSbaaSqaaiabdEha3naaBaaameaacqaIXaqmaeqaaaWcbeaakiabgkHiTiabiodaZiabc6caUiabikdaYiabiwda1aGaayjkaiaawMcaaGGaciab=f8aYbGaay5waiaaw2faaaaa@4D05@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>which leads to :</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M40" name="1471-2288-6-17-i26" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mn>6.5</mml:mn><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mrow><mml:mi>g</mml:mi><mml:mi>E</mml:mi><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>−</mml:mo><mml:mn>2</mml:mn><mml:mi>ρ</mml:mi><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mfrac><mml:mo>,</mml:mo></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeamaaBaaaleaacqWG3bWDdaWgaaadbaGaeGymaedabeaaaSqabaGccqGH9aqpdaWcaaqaaiabiAda2iabc6caUiabiwda1maabmaabaGaeGymaeJaeyOeI0ccciGae8xWdihacaGLOaGaayzkaaGaemivaq1aaWbaaSqabeaacqaIYaGmaaaakeaacqWGNbWzcqWGfbqrcqWGtbWudaahaaWcbeqaaiabikdaYaaakiabgkHiTiabikdaYiab=f8aYjabdsfaunaaCaaaleqabaGaeGOmaidaaaaakiabcYcaSaaa@4805@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>where <italic>ES </italic>refers to the effect size and <italic>T </italic>= <italic>t</italic><sub>(1 - <italic>α</italic>/2),2(<italic>g </italic>- 1) </sub>+ <italic>t</italic><sub>(1 - <italic>β</italic>),2(<italic>g </italic>- 1)</sub></p><p><italic>Cluster size weights </italic>(<italic>denoted w</italic><sub>2</sub>) [<xref ref-type="bibr" rid="B3">3</xref>]</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M41" name="1471-2288-6-17-i8" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mo stretchy="false">(</mml:mo><mml:mover accent="true"><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>A</mml:mi></mml:msub></mml:mrow><mml:mo stretchy="true">¯</mml:mo></mml:mover><mml:mo>−</mml:mo><mml:mn>1</mml:mn><mml:mo stretchy="false">)</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrdaWgaaWcbaGaem4DaC3aaSbaaWqaaiabikdaYaqabaaaleqaaOGaeyypa0JaeGymaeJaey4kaSIaeiikaGYaa0aaaeaacqWGTbqBdaWgaaWcbaGaemyqaeeabeaaaaGccqGHsislcqaIXaqmcqGGPaqkiiGacqWFbpGCaaa@3DCB@</mml:annotation></mml:semantics></mml:math></inline-formula> where <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M42" name="1471-2288-6-17-i9" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mover accent="true"><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>A</mml:mi></mml:msub></mml:mrow><mml:mo stretchy="true">¯</mml:mo></mml:mover><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>g</mml:mi></mml:munderover><mml:mrow><mml:msubsup><mml:mi>m</mml:mi><mml:mi>j</mml:mi><mml:mn>2</mml:mn></mml:msubsup></mml:mrow></mml:mstyle></mml:mrow><mml:mrow><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>g</mml:mi></mml:munderover><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaadaqdaaqaaiabd2gaTnaaBaaaleaacqWGbbqqaeqaaaaakiabg2da9maalaaabaWaaabCaeaacqWGTbqBdaqhaaWcbaGaemOAaOgabaGaeGOmaidaaaqaaiabdQgaQjabg2da9iabigdaXaqaaiabdEgaNbqdcqGHris5aaGcbaWaaabCaeaacqWGTbqBdaWgaaWcbaGaemOAaOgabeaaaeaacqWGQbGAcqGH9aqpcqaIXaqmaeaacqWGNbWza0GaeyyeIuoaaaaaaa@450E@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>With a Pareto imbalance, we can write the equation as:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M43" name="1471-2288-6-17-i27" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>A</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mn>0.8</mml:mn><mml:mi>g</mml:mi><mml:msup><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>0.25</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:mrow></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mn>2</mml:mn></mml:msup><mml:mo>+</mml:mo><mml:mn>0.2</mml:mn><mml:mi>g</mml:mi><mml:msup><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>4</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:mrow></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mi>g</mml:mi></mml:mrow></mml:mfrac><mml:mo>=</mml:mo><mml:mn>3.25</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:mrow></mml:msub></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@5B15@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>So the VIF is reduced to:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M44" name="1471-2288-6-17-i14" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>3.25</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi><mml:mtext>     </mml:mtext><mml:mrow><mml:mo>(</mml:mo><mml:mn>6</mml:mn><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrdaWgaaWcbaGaem4DaC3aaSbaaWqaaiabikdaYaqabaaaleqaaOGaeyypa0JaeGymaeJaey4kaSYaaeWaaeaacqaIZaWmcqGGUaGlcqaIYaGmcqaI1aqncuWGTbqBgaqeamaaBaaaleaacqWG3bWDdaWgaaadbaGaeGOmaidabeaaaSqabaGccqGHsislcqaIXaqmaiaawIcacaGLPaaaiiGacqWFbpGCcaWLjaGaaCzcamaabmaabaGaeGOnaydacaGLOaGaayzkaaaaaa@46C8@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>and</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M45" name="1471-2288-6-17-i15" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mn>2</mml:mn><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mrow><mml:mi>g</mml:mi><mml:mi>E</mml:mi><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>−</mml:mo><mml:mn>6.5</mml:mn><mml:mi>ρ</mml:mi><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mfrac></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeamaaBaaaleaacqWG3bWDdaWgaaadbaGaeGOmaidabeaaaSqabaGccqGH9aqpdaWcaaqaaiabikdaYmaabmaabaGaeGymaeJaeyOeI0ccciGae8xWdihacaGLOaGaayzkaaGaemivaq1aaWbaaSqabeaacqaIYaGmaaaakeaacqWGNbWzcqWGfbqrcqWGtbWudaahaaWcbeqaaiabikdaYaaakiabgkHiTiabiAda2iabc6caUiabiwda1iab=f8aYjabdsfaunaaCaaaleqabaGaeGOmaidaaaaaaaa@471D@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p><italic>Minimum variance weights </italic>(<italic>denoted w</italic><sub>3</sub>) [<xref ref-type="bibr" rid="B3">3</xref>]</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M46" name="1471-2288-6-17-i10" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>g</mml:mi></mml:mrow><mml:mrow><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>g</mml:mi></mml:munderover><mml:mrow><mml:mfrac><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow></mml:mfrac></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrdaWgaaWcbaGaem4DaC3aaSbaaWqaaiabiodaZaqabaaaleqaaOGaeyypa0ZaaSaaaeaacuWGTbqBgaqeaiabdEgaNbqaamaaqahabaWaaSaaaeaacqWGTbqBdaWgaaWcbaGaemOAaOgabeaaaOqaaiabigdaXiabgUcaRmaabmaabaGaemyBa02aaSbaaSqaaiabdQgaQbqabaGccqGHsislcqaIXaqmaiaawIcacaGLPaaaiiGacqWFbpGCaaaaleaacqWGQbGAcqGH9aqpcqaIXaqmaeaacqWGNbWza0GaeyyeIuoaaaaaaa@4AB9@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>With a Pareto imbalance, the equation can be written as:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M47" name="1471-2288-6-17-i28" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mtable><mml:mtr><mml:mtd><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mi>g</mml:mi></mml:mrow><mml:mrow><mml:mn>0.8</mml:mn><mml:mi>g</mml:mi><mml:mfrac><mml:mrow><mml:mn>0.25</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>0.25</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow></mml:mfrac><mml:mo>+</mml:mo><mml:mn>0.2</mml:mn><mml:mi>g</mml:mi><mml:mfrac><mml:mrow><mml:mn>4</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>4</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow></mml:mfrac></mml:mrow></mml:mfrac></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>4</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>0.25</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow></mml:mfrac></mml:mtd></mml:mtr></mml:mtable><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakqaaeeqaaiabdAfawjabdMeajjabdAeagnaaBaaaleaacqWG3bWDdaWgaaadbaGaeG4mamdabeaaaSqabaGccqGH9aqpdaWcaaqaaiqbd2gaTzaaraWaaSbaaSqaaiabdEha3naaBaaameaacqaIZaWmaeqaaaWcbeaakiabdEgaNbqaaiabicdaWiabc6caUiabiIda4iabdEgaNnaalaaabaGaeGimaaJaeiOla4IaeGOmaiJaeGynauJafmyBa0MbaebadaWgaaWcbaGaem4DaC3aaSbaaWqaaiabiodaZaqabaaaleqaaaGcbaGaeGymaeJaey4kaSYaaeWaaeaacqaIWaamcqGGUaGlcqaIYaGmcqaI1aqncuWGTbqBgaqeamaaBaaaleaacqWG3bWDdaWgaaadbaGaeG4mamdabeaaaSqabaGccqGHsislcqaIXaqmaiaawIcacaGLPaaaiiGacqWFbpGCaaGaey4kaSIaeGimaaJaeiOla4IaeGOmaiJaem4zaC2aaSaaaeaacqaI0aancuWGTbqBgaqeamaaBaaaleaacqWG3bWDdaWgaaadbaGaeG4mamdabeaaaSqabaaakeaacqaIXaqmcqGHRaWkdaqadaqaaiabisda0iqbd2gaTzaaraWaaSbaaSqaaiabdEha3naaBaaameaacqaIZaWmaeqaaaWcbeaakiabgkHiTiabigdaXaGaayjkaiaawMcaaiab=f8aYbaaaaaabaGaeyypa0ZaaSaaaeaadaqadaqaaiabigdaXiabgUcaRmaabmaabaGaeGinaqJafmyBa0MbaebadaWgaaWcbaGaem4DaC3aaSbaaWqaaiabiodaZaqabaaaleqaaOGaeyOeI0IaeGymaedacaGLOaGaayzkaaGae8xWdihacaGLOaGaayzkaaWaaeWaaeaacqaIXaqmcqGHRaWkdaqadaqaaiabicdaWiabc6caUiabikdaYiabiwda1iqbd2gaTzaaraWaaSbaaSqaaiabdEha3naaBaaameaacqaIZaWmaeqaaaWcbeaakiabgkHiTiabigdaXaGaayjkaiaawMcaaiab=f8aYbGaayjkaiaawMcaaaqaaiabigdaXiabgUcaRmaabmaabaGafmyBa0MbaebadaWgaaWcbaGaem4DaC3aaSbaaWqaaiabiodaZaqabaaaleqaaOGaeyOeI0IaeGymaedacaGLOaGaayzkaaGae8xWdihaaaaaaa@964E@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>with</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M48" name="1471-2288-6-17-i29" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mi>g</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mn>2</mml:mn><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mrow><mml:mi>E</mml:mi><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mfrac><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>4</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>0.25</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow></mml:mfrac></mml:mrow><mml:mo>]</mml:mo></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@65CB@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>which leads to <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M49" name="1471-2288-6-17-i17" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeamaaBaaaleaacqWG3bWDdaWgaaadbaGaeG4mamdabeaaaSqabaaaaa@30F6@</mml:annotation></mml:semantics></mml:math></inline-formula> being the positive solution of the following equation:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M50" name="1471-2288-6-17-i18" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow><mml:mn>2</mml:mn></mml:msubsup><mml:mi>ρ</mml:mi><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mi>g</mml:mi><mml:mi>E</mml:mi><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>−</mml:mo><mml:mn>2</mml:mn><mml:mi>ρ</mml:mi><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mo>]</mml:mo></mml:mrow><mml:mo>+</mml:mo><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mi>g</mml:mi><mml:mi>E</mml:mi><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>−</mml:mo><mml:mn>8.5</mml:mn><mml:mi>ρ</mml:mi><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mo>]</mml:mo></mml:mrow><mml:mo>−</mml:mo><mml:mn>2</mml:mn><mml:msup><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mn>2</mml:mn></mml:msup><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>=</mml:mo><mml:mn>0</mml:mn></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeamaaDaaaleaacqWG3bWDdaWgaaadbaGaeG4mamdabeaaaSqaaiabikdaYaaaiiGakiab=f8aYnaadmaabaGaem4zaCMaemyrauKaem4uam1aaWbaaSqabeaacqaIYaGmaaGccqGHsislcqaIYaGmcqWFbpGCcqWGubavdaahaaWcbeqaaiabikdaYaaaaOGaay5waiaaw2faaiabgUcaRiqbd2gaTzaaraWaaSbaaSqaaiabdEha3naaBaaameaacqaIZaWmaeqaaaWcbeaakmaabmaabaGaeGymaeJaeyOeI0Iae8xWdihacaGLOaGaayzkaaWaamWaaeaacqWGNbWzcqWGfbqrcqWGtbWudaahaaWcbeqaaiabikdaYaaakiabgkHiTiabiIda4iabc6caUiabiwda1iab=f8aYjabdsfaunaaCaaaleqabaGaeGOmaidaaaGccaGLBbGaayzxaaGaeyOeI0IaeGOmaiZaaeWaaeaacqaIXaqmcqGHsislcqWFbpGCaiaawIcacaGLPaaadaahaaWcbeqaaiabikdaYaaakiabdsfaunaaCaaaleqabaGaeGOmaidaaOGaeyypa0JaeGimaadaaa@65C0@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p><italic>Distribution-based correction </italic>(<italic>denoted d</italic>) [<xref ref-type="bibr" rid="B4">4</xref>]</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M51" name="1471-2288-6-17-i30" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mi>d</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mi>E</mml:mi><mml:msup><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mi>m</mml:mi><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mn>2</mml:mn></mml:msup><mml:mo>+</mml:mo><mml:mi>var</mml:mi><mml:mo>⁡</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mi>m</mml:mi><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mrow><mml:mi>E</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mi>m</mml:mi><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mfrac><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrdaWgaaWcbaGaemizaqgabeaakiabg2da9iabigdaXiabgUcaRmaabmaabaWaaSaaaeaacqWGfbqrdaqadaqaaiabd2gaTbGaayjkaiaawMcaamaaCaaaleqabaGaeGOmaidaaOGaey4kaSIagiODayNaeiyyaeMaeiOCai3aaeWaaeaacqWGTbqBaiaawIcacaGLPaaaaeaacqWGfbqrdaqadaqaaiabd2gaTbGaayjkaiaawMcaaaaacqGHsislcqaIXaqmaiaawIcacaGLPaaaiiGacqWFbpGCaaa@4ACE@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p><inline-graphic xlink:href="1471-2288-6-17-i31.gif"/></p><p>So we have:</p><p><inline-graphic xlink:href="1471-2288-6-17-i32.gif"/></p><p>with</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M52" name="1471-2288-6-17-i33" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>d</mml:mi></mml:msub><mml:mi>g</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mn>2</mml:mn><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mrow><mml:mi>E</mml:mi><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mfrac><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>3.25</mml:mn><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>d</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>]</mml:mo></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeamaaBaaaleaacqWGKbazaeqaaOGaem4zaCMaeyypa0ZaaSaaaeaacqaIYaGmcqWGubavdaahaaWcbeqaaiabikdaYaaaaOqaaiabdweafjabdofatnaaCaaaleqabaGaeGOmaidaaaaakmaadmaabaGaeGymaeJaey4kaSYaaeWaaeaacqaIZaWmcqGGUaGlcqaIYaGmcqaI1aqncuWGTbqBgaqeamaaBaaaleaacqWGKbazaeqaaOGaeyOeI0IaeGymaedacaGLOaGaayzkaaacciGae8xWdihacaGLBbGaayzxaaaaaa@4887@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>that is to say:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M53" name="1471-2288-6-17-i20" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>d</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mn>2</mml:mn><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mrow><mml:mi>g</mml:mi><mml:mi>E</mml:mi><mml:msup><mml:mi>S</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>−</mml:mo><mml:mn>6.5</mml:mn><mml:mi>ρ</mml:mi><mml:msup><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mfrac></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeamaaBaaaleaacqWGKbazaeqaaOGaeyypa0ZaaSaaaeaacqaIYaGmdaqadaqaaiabigdaXiabgkHiTGGaciab=f8aYbGaayjkaiaawMcaaiabdsfaunaaCaaaleqabaGaeGOmaidaaaGcbaGaem4zaCMaemyrauKaem4uam1aaWbaaSqabeaacqaIYaGmaaGccqGHsislcqaI2aGncqGGUaGlcqaI1aqncqWFbpGCcqWGubavdaahaaWcbeqaaiabikdaYaaaaaaaaa@45CD@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>One then recognizes the results obtained using the cluster size weights correction.</p></sec><sec><title>Appendix B: minimum variance weights-corrected variance inflation factor (VIF) for an <italic>a priori </italic>postulated Pareto-like imbalance</title><p>Characteristics of the Pareto-like imbalance</p><table-wrap position="float" id="T6"><label>Table 6</label><table frame="hsides" rules="groups"><thead><tr><td></td><td align="left">Number of clusters by intervention arm</td><td align="center">Number of patients belonging to the clusters</td><td align="center">Mean cluster size</td></tr></thead><tbody><tr><td align="left">Small clusters</td><td align="center">(1-<italic>γ</italic>)<italic>g</italic></td><td align="center">(1 - <italic>τ</italic>)<inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M54" name="1471-2288-6-17-i22" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeaaaa@2E27@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>g</italic></td><td align="center"><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M55" name="1471-2288-6-17-i34" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mfrac><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>τ</mml:mi></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow></mml:mfrac><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaadaWcaaqaaiabigdaXiabgkHiTGGaciab=r8a0bqaaiabigdaXiabgkHiTiab=n7aNbaacuWGTbqBgaqeaaaa@355F@</mml:annotation></mml:semantics></mml:math></inline-formula></td></tr><tr><td align="left">Big clusters</td><td align="center"><italic>γg</italic></td><td align="center"><italic>τ</italic><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M56" name="1471-2288-6-17-i22" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeaaaa@2E27@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>g</italic></td><td align="center"><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M57" name="1471-2288-6-17-i35" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mfrac><mml:mi>τ</mml:mi><mml:mi>γ</mml:mi></mml:mfrac><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaadaWcaaqaaGGaciab=r8a0bqaaiab=n7aNbaacuWGTbqBgaqeaaaa@31A5@</mml:annotation></mml:semantics></mml:math></inline-formula></td></tr></tbody></table></table-wrap><p><italic>g </italic>refers to the number of clusters within each arm and <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M58" name="1471-2288-6-17-i22" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGTbqBgaqeaaaa@2E27@</mml:annotation></mml:semantics></mml:math></inline-formula> is the average cluster size</p><p>Minimum variance weighs VIF</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M59" name="1471-2288-6-17-i36" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mtable><mml:mtr><mml:mtd><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:msub><mml:mi>F</mml:mi><mml:mrow><mml:msub><mml:mi>w</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>g</mml:mi></mml:mrow><mml:mrow><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>g</mml:mi></mml:munderover><mml:mrow><mml:mfrac><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow></mml:mfrac></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>g</mml:mi></mml:mrow><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>g</mml:mi><mml:mfrac><mml:mrow><mml:mfrac><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>τ</mml:mi></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow></mml:mfrac><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>τ</mml:mi></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow></mml:mfrac><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow></mml:mfrac><mml:mo>+</mml:mo><mml:mi>γ</mml:mi><mml:mi>g</mml:mi><mml:mfrac><mml:mrow><mml:mfrac><mml:mi>τ</mml:mi><mml:mi>γ</mml:mi></mml:mfrac><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mi>τ</mml:mi><mml:mi>γ</mml:mi></mml:mfrac><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow></mml:mfrac></mml:mrow></mml:mfrac></mml:mtd></mml:mtr></mml:mtable><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakqaaeeqaaiabdAfawjabdMeajjabdAeagnaaBaaaleaacqWG3bWDdaWgaaadbaGaeG4mamdabeaaaSqabaGccqGH9aqpdaWcaaqaaiqbd2gaTzaaraGaem4zaCgabaWaaabCaeaadaWcaaqaaiabd2gaTnaaBaaaleaacqWGQbGAaeqaaaGcbaGaeGymaeJaey4kaSYaaeWaaeaacqWGTbqBdaWgaaWcbaGaemOAaOgabeaakiabgkHiTiabigdaXaGaayjkaiaawMcaaGGaciab=f8aYbaaaSqaaiabdQgaQjabg2da9iabigdaXaqaaiabdEgaNbqdcqGHris5aaaaaOqaaiabg2da9maalaaabaGafmyBa0MbaebacqWGNbWzaeaadaqadaqaaiabigdaXiabgkHiTiab=n7aNbGaayjkaiaawMcaaiabdEgaNnaalaaabaWaaSaaaeaacqaIXaqmcqGHsislcqWFepaDaeaacqaIXaqmcqGHsislcqWFZoWzaaGafmyBa0MbaebaaeaacqaIXaqmcqGHRaWkdaqadaqaamaalaaabaGaeGymaeJaeyOeI0Iae8hXdqhabaGaeGymaeJaeyOeI0Iae83SdCgaaiqbd2gaTzaaraGaeyOeI0IaeGymaedacaGLOaGaayzkaaGae8xWdihaaiabgUcaRiab=n7aNjabdEgaNnaalaaabaWaaSaaaeaacqWFepaDaeaacqWFZoWzaaGafmyBa0MbaebaaeaacqaIXaqmcqGHRaWkdaqadaqaamaalaaabaGae8hXdqhabaGae83SdCgaaiqbd2gaTzaaraGaeyOeI0IaeGymaedacaGLOaGaayzkaaGae8xWdihaaaaaaaaa@821D@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>So we obtain:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M60" name="1471-2288-6-17-i37" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>V</mml:mi><mml:mi>I</mml:mi><mml:mi>F</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>τ</mml:mi></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow></mml:mfrac><mml:mi>m</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>]</mml:mo></mml:mrow><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mi>τ</mml:mi><mml:mi>γ</mml:mi></mml:mfrac><mml:mi>m</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>]</mml:mo></mml:mrow></mml:mrow><mml:mrow><mml:mi>τ</mml:mi><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>τ</mml:mi></mml:mrow><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow></mml:mfrac><mml:mi>m</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>]</mml:mo></mml:mrow><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>τ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mi>τ</mml:mi><mml:mi>γ</mml:mi></mml:mfrac><mml:mi>m</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>ρ</mml:mi></mml:mrow><mml:mo>]</mml:mo></mml:mrow></mml:mrow></mml:mfrac></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGwbGvcqWGjbqscqWGgbGrcqGH9aqpdaWcaaqaamaadmaabaGaeGymaeJaey4kaSYaaeWaaeaadaWcaaqaaiabigdaXiabgkHiTGGaciab=r8a0bqaaiabigdaXiabgkHiTiab=n7aNbaacqWGTbqBcqGHsislcqaIXaqmaiaawIcacaGLPaaacqWFbpGCaiaawUfacaGLDbaadaWadaqaaiabigdaXiabgUcaRmaabmaabaWaaSaaaeaacqWFepaDaeaacqWFZoWzaaGaemyBa0MaeyOeI0IaeGymaedacaGLOaGaayzkaaGae8xWdihacaGLBbGaayzxaaaabaGae8hXdq3aamWaaeaacqaIXaqmcqGHRaWkdaqadaqaamaalaaabaGaeGymaeJaeyOeI0Iae8hXdqhabaGaeGymaeJaeyOeI0Iae83SdCgaaiabd2gaTjabgkHiTiabigdaXaGaayjkaiaawMcaaiab=f8aYbGaay5waiaaw2faaiabgUcaRmaabmaabaGaeGymaeJaeyOeI0Iae8hXdqhacaGLOaGaayzkaaWaamWaaeaacqaIXaqmcqGHRaWkdaqadaqaamaalaaabaGae8hXdqhabaGae83SdCgaaiabd2gaTjabgkHiTiabigdaXaGaayjkaiaawMcaaiab=f8aYbGaay5waiaaw2faaaaaaaa@7757@</mml:annotation></mml:semantics></mml:math></inline-formula></p></sec><sec><title>Appendix C: gini coefficient for an <italic>a priori </italic>postulated Pareto-like imbalance</title><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M61" name="1471-2288-6-17-i38" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>g</mml:mi><mml:mi>i</mml:mi><mml:mi>n</mml:mi><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:mn>2</mml:mn><mml:msup><mml:mi>g</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow></mml:mfrac><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>g</mml:mi></mml:munderover><mml:mrow><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>g</mml:mi></mml:munderover><mml:mrow><mml:mrow><mml:mo>\|</mml:mo><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow><mml:mo>\|</mml:mo></mml:mrow></mml:mrow></mml:mstyle></mml:mrow></mml:mstyle></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGNbWzcqWGPbqAcqWGUbGBcqWGPbqAcqGH9aqpdaWcaaqaaiabigdaXaqaaiabikdaYiabdEgaNnaaCaaaleqabaGaeGOmaidaaOGafmyBa0MbaebaaaWaaabCaeaadaaeWbqaamaaemaabaGaemyBa02aaSbaaSqaaiabdMgaPbqabaGccqGHsislcqWGTbqBdaWgaaWcbaGaemOAaOgabeaaaOGaay5bSlaawIa7aaWcbaGaemOAaOMaeyypa0JaeGymaedabaGaem4zaCganiabggHiLdaaleaacqWGPbqAcqGH9aqpcqaIXaqmaeaacqWGNbWza0GaeyyeIuoaaaa@50E0@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>Given the characteristics of the Pareto-like imbalance presented in appendix B, considering that clusters are ordered hierarchically by increasing size, the matrix of the difference \|<italic>m</italic><sub><italic>i </italic></sub>- <italic>m</italic><sub><italic>j</italic></sub>\| can be written as:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M62" name="1471-2288-6-17-i39" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>M</mml:mi><mml:mo>=</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mtable><mml:mtr><mml:mtd><mml:mrow><mml:msub><mml:mn>0</mml:mn><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>γ</mml:mi><mml:mi>g</mml:mi><mml:mo>,</mml:mo><mml:mi>γ</mml:mi><mml:mi>g</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:msub></mml:mrow></mml:mtd><mml:mtd><mml:mrow><mml:msub><mml:mn>1</mml:mn><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>γ</mml:mi><mml:mi>g</mml:mi><mml:mo>,</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>g</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:msub><mml:mrow><mml:mo>\|</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mi>τ</mml:mi><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow><mml:mrow><mml:mi>γ</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mfrac></mml:mrow><mml:mo>\|</mml:mo></mml:mrow></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mrow><mml:msub><mml:mn>1</mml:mn><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>g</mml:mi><mml:mo>,</mml:mo><mml:mi>γ</mml:mi><mml:mi>g</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:msub><mml:mrow><mml:mo>\|</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mi>τ</mml:mi><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow><mml:mrow><mml:mi>γ</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mfrac></mml:mrow><mml:mo>\|</mml:mo></mml:mrow></mml:mrow></mml:mtd><mml:mtd><mml:mrow><mml:msub><mml:mn>0</mml:mn><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>g</mml:mi><mml:mo>,</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>g</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:msub></mml:mrow></mml:mtd></mml:mtr></mml:mtable></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGnbqtcqGH9aqpdaqadaqaauaabeqaciaaaeaacqaIWaamdaWgaaWcbaWaaeWaaeaaiiGacqWFZoWzcqWGNbWzcqGGSaalcqWFZoWzcqWGNbWzaiaawIcacaGLPaaaaeqaaaGcbaGaeGymaeZaaSbaaSqaamaabmaabaGae83SdCMaem4zaCMaeiilaWYaaeWaaeaacqaIXaqmcqGHsislcqWFZoWzaiaawIcacaGLPaaacqWGNbWzaiaawIcacaGLPaaaaeqaaOWaaqWaaeaadaWcaaqaaiab=r8a0jabgkHiTiab=n7aNbqaaiab=n7aNnaabmaabaGaeGymaeJaeyOeI0Iae83SdCgacaGLOaGaayzkaaaaaaGaay5bSlaawIa7aaqaaiabigdaXmaaBaaaleaadaqadaqaamaabmaabaGaeGymaeJaeyOeI0Iae83SdCgacaGLOaGaayzkaaGaem4zaCMaeiilaWIae83SdCMaem4zaCgacaGLOaGaayzkaaaabeaakmaaemaabaWaaSaaaeaacqWFepaDcqGHsislcqWFZoWzaeaacqWFZoWzdaqadaqaaiabigdaXiabgkHiTiab=n7aNbGaayjkaiaawMcaaaaaaiaawEa7caGLiWoaaeaacqaIWaamdaWgaaWcbaWaaeWaaeaadaqadaqaaiabigdaXiabgkHiTiab=n7aNbGaayjkaiaawMcaaiabdEgaNjabcYcaSmaabmaabaGaeGymaeJaeyOeI0Iae83SdCgacaGLOaGaayzkaaGaem4zaCgacaGLOaGaayzkaaaabeaaaaaakiaawIcacaGLPaaaaaa@8068@</mml:annotation></mml:semantics></mml:math></inline-formula></p><p>Where 0<sub>(<italic>γg</italic>,<italic>γg</italic>) </sub>and 0<sub>((1-<italic>γ</italic>)<italic>g</italic>,(1-<italic>γ</italic>)<italic>g</italic>) </sub>are squared matrices of size <italic>γg </italic>and (1-<italic>γ</italic>)<italic>g </italic>respectively and 1<sub>(<italic>γg</italic>,(1-<italic>γ</italic>)<italic>g</italic>) </sub>and 1<sub>((1-<italic>γ</italic>)<italic>g</italic>,<italic>γg</italic>) </sub>are matices of size <italic>γg </italic>× (1-<italic>γ</italic>)<italic>g </italic>and (1-<italic>γ</italic>)<italic>g </italic>× <italic>γg </italic>respectively, containing only 1 s.</p><p>Thus:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M63" name="1471-2288-6-17-i40" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mtable columnalign="left"><mml:mtr><mml:mtd><mml:mi>g</mml:mi><mml:mi>i</mml:mi><mml:mi>n</mml:mi><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:mn>2</mml:mn><mml:msup><mml:mi>g</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow></mml:mfrac><mml:mn>2</mml:mn><mml:msup><mml:mi>g</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mi>γ</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mrow><mml:mo>\|</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mi>τ</mml:mi><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow><mml:mrow><mml:mi>γ</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mfrac></mml:mrow><mml:mo>\|</mml:mo></mml:mrow><mml:mover accent="true"><mml:mi>m</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mi>g</mml:mi><mml:mi>i</mml:mi><mml:mi>n</mml:mi><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mrow><mml:mo>\|</mml:mo><mml:mrow><mml:mi>τ</mml:mi><mml:mo>−</mml:mo><mml:mi>γ</mml:mi></mml:mrow><mml:mo>\|</mml:mo></mml:mrow></mml:mtd></mml:mtr></mml:mtable><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakqaabeqaaiabdEgaNjabdMgaPjabd6gaUjabdMgaPjabg2da9maalaaabaGaeGymaedabaGaeGOmaiJaem4zaC2aaWbaaSqabeaacqaIYaGmaaGccuWGTbqBgaqeaaaacqaIYaGmcqWGNbWzdaahaaWcbeqaaiabikdaYaaaiiGakiab=n7aNnaabmaabaGaeGymaeJaeyOeI0Iae83SdCgacaGLOaGaayzkaaWaaqWaaeaadaWcaaqaaiab=r8a0jabgkHiTiab=n7aNbqaaiab=n7aNnaabmaabaGaeGymaeJaeyOeI0Iae83SdCgacaGLOaGaayzkaaaaaaGaay5bSlaawIa7aiqbd2gaTzaaraaabaGaem4zaCMaemyAaKMaemOBa4MaemyAaKMaeyypa0ZaaqWaaeaacqWFepaDcqGHsislcqWFZoWzaiaawEa7caGLiWoaaaaa@60CF@</mml:annotation></mml:semantics></mml:math></inline-formula></p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1471-2288/6/17/prepub"/></p></sec>
Lessons and implications from a mass immunization campaign in squatter settlements of Karachi, Pakistan: an experience from a cluster-randomized double-blinded vaccine trial [NCT00125047]	<sec><title>Objective</title><p>To determine the safety and logistic feasibility of a mass immunization strategy outside the local immunization program in the pediatric population of urban squatter settlements in Karachi, Pakistan.</p></sec><sec sec-type="methods"><title>Methods</title><p>A cluster-randomized double blind preventive trial was launched in August 2003 in 60 geographic clusters covering 21,059 children ages 2 to 16 years. After consent was obtained from parents or guardians, eligible children were immunized parenterally at vaccination posts in each cluster with Vi polysaccharide or hepatitis A vaccine. Safety, logistics, and standards were monitored and documented.</p></sec><sec><title>Results</title><p>The vaccine coverage of the population was 74% and was higher in those under age 10 years. No life-threatening serious adverse events were reported. Adverse events occurred in less than 1% of all vaccine recipients and the main reactions reported were fever and local pain. The proportion of adverse events in Vi polysaccharide and hepatitis A recipients will not be known until the end of the trial when the code is broken. Throughout the vaccination campaign safe injection practices were maintained and the cold chain was not interrupted. Mass vaccination in slums had good acceptance. Because populations in such areas are highly mobile, settlement conditions could affect coverage. Systemic reactions were uncommon and local reactions were mild and transient. Close community involvement was pivotal for information dissemination and immunization coverage.</p></sec><sec><title>Conclusion</title><p>This vaccine strategy described together with other information that will soon be available in the area (cost/effectiveness, vaccine delivery costs, etc) will make typhoid fever control become a reality in the near future.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Khan</surname><given-names>Mohammad Imran</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Ochiai</surname><given-names>Rion Leon</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Hamza</surname><given-names>Hasan Bin</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Sahito</surname><given-names>Shah Muhammad</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Habib</surname><given-names>Muhammad Atif</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Soofi</surname><given-names>Sajid Bashir</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A7" contrib-type="author"><name><surname>Bhutto</surname><given-names>Naveed Sarwar</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A8" contrib-type="author"><name><surname>Rasool</surname><given-names>Shahid</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A9" contrib-type="author"><name><surname>Puri</surname><given-names>Mahesh K</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A10" contrib-type="author"><name><surname>Ali</surname><given-names>Mohammad</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A11" contrib-type="author"><name><surname>Wasan</surname><given-names>Shafi Mohammad</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A12" contrib-type="author"><name><surname>Khan</surname><given-names>Mohammad Jawed</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A13" contrib-type="author"><name><surname>Abu-Elyazeed</surname><given-names>Remon</given-names></name><xref ref-type="aff" rid="I4">4</xref><xref ref-type="aff" rid="I5">5</xref><email>[email protected]</email></contrib><contrib id="A14" contrib-type="author"><name><surname>Ivanoff</surname><given-names>Bernard</given-names></name><xref ref-type="aff" rid="I6">6</xref><email>[email protected]</email></contrib><contrib id="A15" contrib-type="author"><name><surname>Galindo</surname><given-names>Claudia M</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A16" contrib-type="author"><name><surname>Pang</surname><given-names>Tikki</given-names></name><xref ref-type="aff" rid="I7">7</xref><email>[email protected]</email></contrib><contrib id="A17" contrib-type="author"><name><surname>Donner</surname><given-names>Allan</given-names></name><xref ref-type="aff" rid="I8">8</xref><email>[email protected]</email></contrib><contrib id="A18" contrib-type="author"><name><surname>von Seidlein</surname><given-names>Lorenz</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A19" contrib-type="author"><name><surname>Acosta</surname><given-names>Camilo J</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A20" contrib-type="author"><name><surname>Clemens</surname><given-names>John D</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A21" contrib-type="author"><name><surname>Nizami</surname><given-names>Shaikh Qamaruddin</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A22" corresp="yes" contrib-type="author"><name><surname>Bhutta</surname><given-names>Zulfiqar A</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	Trials	<sec><title>Background</title><p>Despite major breakthroughs in the development of new vaccines over the past two decades, the gap in access to vaccines between wealthy and poorer countries has widened. As a result, immunization schedules offer more vaccines in high-income countries than in those with low income [<xref ref-type="bibr" rid="B1">1</xref>]. Children in low-income countries are also at a disadvantage because vaccine research and development agendas are tailored to the needs of developed countries. The focus the International Vaccine Institute typhoid fever program is to enable people at risk to get access to the vaccines and decrease the burden of typhoid fever [<xref ref-type="bibr" rid="B2">2</xref>]. The program is being conducted in five urban slums of Indonesia, China, Vietnam, India and Pakistan.</p><p>Two typhoid fever vaccines, Ty21a and Vi polysaccharide (PS) are currently licensed for use. A recent Cochrane review [<xref ref-type="bibr" rid="B3">3</xref>] on typhoid fever vaccines found that the two vaccines i.e. Ty 21 and Vi have similar results with Vi having an advantage of heat stability and single dose regimen. ViPS was thus chosen for use in the DOMI trial as it would suit the public health program for immunization in the countries of south east Asia. The use of Vi requiring a single, injectable dose was thought to be logistically easier than use of Ty21 which requires three doses.</p><p>In Pakistan, DOMI program aims to introduce an available, affordable Vi polysaccharide vaccine [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B5">5</xref>] for children 2 to 16 years of age living in urban low socio economic settings. The study setting has a high typhoid fever burden and treatment is increasingly costly [<xref ref-type="bibr" rid="B6">6</xref>] and difficult due to high drug resistance[<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B8">8</xref>]. Vi PS vaccine, which has moderate efficacy (64–77%), seems to be an immediate and affordable option for impoverished populations exposed to typhoid fever[<xref ref-type="bibr" rid="B9">9</xref>]. Unfortunately, there is no evidence about its cost-effectiveness and no delivery strategy has been envisaged that would enable policymakers to make rational decisions about the use of this vaccine as a public health tool.</p><p>To determine the effectiveness of Vi PS in reducing typhoid fever burden in slum areas and the cost-effectiveness of the vaccine, DOMI investigators designed a cluster-randomized double blind trial [<xref ref-type="bibr" rid="B2">2</xref>]. Given the apparent immunological limitation of PS vaccines in early age groups [<xref ref-type="bibr" rid="B10">10</xref>] and the fact that the high-risk group in Karachi is the entire pediatric population, the local Expanded Program of Immunization (EPI), which usually reaches children under age 5 years, is unlikely to be the best delivery structure. Also, national reported immunization coverage rates for Pakistan have been very variable (range of 60–90%) since 1998 [<xref ref-type="bibr" rid="B11">11</xref>]. In addition to a lack of resources, other documented reasons for low vaccine coverage in Pakistan include lack of awareness of need, mothers unable to attend the vaccine posts, and inconvenient immunization sites [<xref ref-type="bibr" rid="B12">12</xref>-<xref ref-type="bibr" rid="B14">14</xref>] For these reasons, the DOMI program decided to implement the mass immunization campaign outside the EPI delivery system. Here we report initial results from this mass vaccination in two slums in Karachi, Pakistan.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Study site</title><p>The vaccination campaign was carried out in two adjacent squatter settlements, Sultanabad and Hijrat Colony, in Karachi. The population is a mix of Punjabi-Pathan ethnic groups from northern Pakistan. The total population of the two areas is 53,738 (project census 2003), with 21,059 children (aged 2–16 years) in the study target pediatric population. There are 8,278 households in the combined settlement areas of 0.54 km<sup>2</sup>. Most health-care is provided by the private sector through small clinics. In the last 7 years the Department of Pediatrics of Aga Khan University [<xref ref-type="bibr" rid="B7">7</xref>] has rendered free clinical services to the pediatric population through health centers, one in each study area. These are staffed with research medical officers (RMO), health assistants, field supervisors, and community health workers (CHW). The study area is considered to be a high endemic area for typhoid fever, especially among children [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B15">15</xref>].</p></sec><sec><title>Sample size</title><p>A total of approximately 24,710 children are needed in order to have 80% power to detect a 50% vaccine protection at a 5% level of significance. Using Hayes and Bennet formula [<xref ref-type="bibr" rid="B16">16</xref>], this sample size calculation assumes a minimum cumulative typhoid incidence of 2.8 per 1000 (during 2 years), assuming alpha = 0.05, minimum power 0.8 (= 1-beta) to achieve a significant difference, Protective Efficacy (PE) of 0.5 for 2 years, between cluster coefficient of variation (CV) below 0.5 an average cluster size of 580.</p></sec><sec><title>Study design</title><p>The cluster-randomized design employed by the trial mimics the way Vi vaccine would be delivered under a public health program in Pakistan. The study area, Sultanabad and Hijrat Colony were divided into 28 and 32 geographic clusters (a group of adjacent households), respectively (figure <xref ref-type="fig" rid="F2">2</xref>). Cluster sizes varied from 162 to 653 children (2–16 years of age) with an average of 350. The unit of randomization (clusters) was stratified by slum (Hijrat or Sultanabad) and cluster size (large of small). The eligible population was children aged 2–16 years who were included in the project census and whose parents/guardians gave consent to participate. These 60 geographic areas (clusters, 32 in Hijrat colony and 28 in Sultanabad) were randomly allocated to Vi PS vaccine (Typherix<sup>®</sup>) or the active control hepatitis A vaccine (Havrix<sup>®</sup>). Randomization was done by an expert statistician who was independent of disease surveillance activities in the study setting. The local investigators were not aware which vaccine was assigned codes (C & M). Labeling of the vaccine was done by the vaccine producer in Rixensart, Belgium. The randomization sequence was not changed at any level once it was initially generated. Each team worked with the single code throughout the campaign to minimize the risk of mix-ups. The schedule of cluster visits were arranged in such a way that the number of vaccine in both groups given per day was approximately equal. Continuous supervisory visits by supervisory team and external monitors ensured that all procedures were followed according to the protocol.</p><fig position="float" id="F1"><label>Figure 1</label><caption><p>Distribution of target population and vaccination coverage Vi trial in Karachi Pakistan.</p></caption><graphic xlink:href="1745-6215-7-17-1"/></fig></sec><sec><title>The vaccines</title><p>Both vaccines, ViPS and HAV, are licensed in Pakistan and were donated by GlaxoSmithKline (GSK). Similar individual-dose-syringe vaccines were labelled with information on: the batch number, expiry date, route of administration and code (C or M). The identification of codes was kept with the Data Safety and Monitoring Board (DSMB). Each 0.5 ml dose of Typherix<sup>® </sup>contained 25 micrograms of the Vi PS of <italic>S</italic>. Typhi. Each 0.5 ml of pediatric dose of Havrix<sup>® </sup>vaccine consists of not less than 720 units of viral antigen, adsorbed on 0.25 mg aluminum hydroxide. The Havrix<sup>® </sup>dosage consisted of a primary course and a booster that will be administered after the study ends (year 2). Both vaccines are for intramuscular injection only. Both groups will ultimately receive the benefits of the Vi vaccine as well as the HAV vaccine as a cross-over vaccination is planned at the end of the surveillance period.</p><p>In case of an adverse event the physician in charge examined the vaccine recipients and assess the severity of the event. In case of an event requiring hospitalization, the clinical monitor was notified who visited the patient and assessed the need of breaking the code after managing the case.</p><p>This project was approved by the AKU (Karachi) ethical committee, the Institutional Review Board of the International Vaccine Institute (IVI), Seoul, and the World Health Organization (WHO) ethical committee. A Data Safety and Monitoring Board (DSMB) was established for the audit, protocol review and take a decision on breaking the codes in case the breaking of code was deemed necessary due to an adverse event.</p></sec><sec><title>Information dissemination and consent</title><p>Information dissemination started 12 weeks before the campaign. Sessions at street level were conducted by Research Medical Officers (RMO) or trained female Community Health Workers (CHW), as appropriate, and focused on the importance of immunization against typhoid fever and other control measures. More intense promotion of the campaign began in June 2003, 4 weeks prior to the campaign, by a team of RMOs and social scientists, who conducted meetings with community and religious leaders, members of local government bodies, and street representatives. Information leaflets were distributed and announcements were made at local mosques. Suitable areas for vaccination posts in each cluster were identified. The day before the vaccination date, households were visited and given formal invitation letters that included the site, date, and time for vaccination. During the campaign repeated visits were made by CHWs to remind and motivate those targeted for vaccination.</p><p>Parents and guardians visiting vaccination posts on the day of immunization were given information about the nature of the trial, expected risks and benefits, and procedural details as part of the informed consent. Trained project personnel provided this information. Upon agreeing to participate, a thumbprint was affixed on the vaccination record book together with the signature of a witness.</p></sec><sec><title>Training and logistics</title><p>Vaccination teams received intensive training that focused on the trial's primary objectives, public health implications, blinding, cold chain maintenance, adverse events (AE), and community mobilization. Good Clinical Practices (GCPs) were emphasized at every step of training and monitored throughout the campaign. An adverse event was defined as a medical incident that takes place after vaccination, causes concern and is believed to be caused by the immunization. Continuous feedback meetings were conducted at intervals throughout the campaign.</p><p>Ten vaccination teams were employed consisting each of 1 physician, 1 vaccinator, 1 recorder, and 2 or 3 assistants. All were recruited specifically for the campaign and vaccinators were hired locally from the study sites. Physicians (team leaders) were responsible for the clinical and logistical components of their respective group. Supervisors (4) were in charge of two or three teams. External observers (2) monitored and documented key aspects (consent process, standards, safety, and cold chain) on designated forms. Seven community members assisted the campaign. Most of the vaccination posts were rooms or entire homes provided by residents and agreed upon by community leaders. Additionally, four local social scientists developed and organized the community awareness of the campaign. AKU drivers (5) assisted with transportation of supplies.</p><p>Vaccines were transported from Belgium to a local warehouse in Karachi. Recommended storage temperature of both vaccines is +2°C to +8°C. The temperature was monitored by one or more of the following methods: conventional thermometer, maximum-minimum thermometer, visual indicators of low temperature (FreezeWatch™, 3 M, USA), wheel recorders, and automated battery-powered devices (COX Technology, Sensitech, Inc., Beverly, MA, USA). Vaccines were stored in several sites: the central warehouse, the AKU warehouse, a field site logistics center, and cool boxes at vaccination posts. At the logistics center an officer distributed the vaccine on a daily basis to each vaccination team. Temperature was monitored and documented as follows: constantly at the central warehouse, once a day at the AKU warehouse, twice or more each day at the logistics center, and in the field (by the team leader) at least twice a day. Alternate power supply was available at all storage locales. Cooling equipment at each storage site consisted of cold rooms, a chest refrigerator at the logistics center, and cool boxes with frozen ice packs for the vaccination teams. Vaccine usage was recorded daily on a logistics form.</p></sec><sec><title>Immunization campaign</title><p>The mass immunization campaign was planned and launched in a way that it would not disturb other regular local health programs. Vaccination took place from Monday through Sunday from 3 pm to 11 pm; this timing allowed the parent/guardian (usually a working male) flexibility to visit after the working hours. Each vaccination team was in charge of one cluster and was assigned to deliver one and only one vaccine code letter (C, M). The chance of breaking the code was reduced by explaining the purpose of blinding to the teams, rechecking by another person at the time of distribution and follow up visits by site supervisors to ensure the same code is being given in the cluster that is assigned after randomization. All efforts were made to ensure blinding.</p><p>Based on a pre-assigned schedule, vaccination teams visited each cluster from August 12<sup>th </sup>to September 12<sup>th</sup>, 2003, to cover the target population of 21,059 children. Any child with fever > 37.5°C at the time of immunization or a female who was married, pregnant, and/or lactating was considered ineligible. Febrile individuals were provided with antipyretics and asked to return if the temperature subsided.</p><p>A vaccine record book (per cluster) containing a page-by-page alphabetical listing of all household members (based on a project census conducted 6 months earlier) was available to each team. Children were identified by their project identification (ID) card; if the card was not available, a computerized ID search system was used. The record book also documented the date of vaccination, eligibility, letter code of the vaccine, and presence or absence of an immediate AE. Team assistants (who belonged to the study area and have been working with AKU for at least one year) repeatedly visited the households in a specific cluster to re-invite and also to update household status such as migration, refusals, temporary absentees, and census duplications. The clusters where the vaccine coverage was less than 60% were visited again in the last 4 days of the campaign by re-establishing the vaccine post.</p><p>Vaccination AE data were obtained by direct observation of each vaccinee at each vaccination center for 30 minutes to detect immediate serious AE; by home visits (once/day, total of 3) in a cluster-based random sub-sample of 240 children (4 per cluster) to detect solicited AE; and by passive surveillance of un-solicited AE in the initial 30 days. Vaccination posts had basic emergency equipment and trained study staff to treat immediate severe AE (SAE); transportation to the AKU hospital was assured for SAEs. WHO guidelines on safe injection practices [<xref ref-type="bibr" rid="B17">17</xref>] were followed. Needlestick injuries were reported to the supervisor and treated in accordance with national guidelines. Disposal boxes for the safe disposal of sharps were provided to each team; the boxes were later incinerated at AKU hospital.</p></sec><sec><title>Outcomes, data management, and statistical analysis</title><p>The primary outcome of the trial is an episode of fever during which S. Typhi is isolated from blood culture or fever ≥ 3 days and a positive serology-proven typhoid fever test (Widal test or Tubex or Typhidot-M) or fever ≥ 3 days and a positive Widal test or fever ≥ 24 hours and body temperature ≥ 38.5C and ≥ 1 of the following symptoms: headache, abdominal pain or constipation and a positive (Widal or Tubex or Typhidot-M) test.</p><p>The denominator for the incidence rate calculations will be the number of subjects at risk and it will be the primary denominator used to measure the outcomes.</p><p>Analysis of the surveillance data will be based on a fixed cohort approach and the incidence of typhoid fever in a two-year period will be calculated. Vaccine coverage was calculated on the basis of the vaccination record books and the proportion immunized from the target population. To assess logistics, the following were described and quantified: (1) resources, including personnel, needed for vaccine storage, transport, and delivery; (2) efficiency of vaccine storage, transport, and handling; and (3) safe vaccination practices, including vaccine administration and disposal of sharps.</p><p>Data were maintained with a FoxPro 6 (Microsoft, Redmond, WA USA) based data management system. Simple descriptive statistics such as frequencies, averages estimates, and proportions with standard deviation were calculated using SPSS version 10 (SPSS Inc., Cary, NC, USA). ArcView 8 (ESRI, USA) was used for geographical information system analysis.</p></sec></sec><sec><title>Results</title><sec><title>Vaccine coverage</title><p>In total, 12,830 (61%) children were vaccinated during the campaign. Of the remaining target population of 8,229 who did not receive vaccine, main reasons were emigration (2,824), refusal to participate (2,613), absence (1,811), ineligible (69), vaccinated in other programs (17), and incorrect census data (895) (table <xref ref-type="table" rid="T1">1</xref>). The final coverage in the study sites was 74% with highest coverage in children aged 2–10 years (77%) and lower in those >10 years (67%). Overall coverage by gender and area was similar (table <xref ref-type="table" rid="T2">2</xref>). Cluster vaccination varied from 37% to 81%, with median cluster coverage of 63% (figure <xref ref-type="fig" rid="F1">1</xref>). Second visit to the low coverage clusters in the last days (mop-up) increased the overall coverage from 70% to 74%.</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Vaccination coverage results from mass vaccination campaign in urban squatter settlements of Karachi, Pakistan – 2003</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left"><bold>Target population</bold></td><td align="left">No. of individuals</td><td align="left">%</td></tr></thead><tbody><tr><td align="left"><bold>Not vaccinated</bold></td><td align="left"><bold>21,059</bold></td><td></td></tr><tr><td align="left"> Migrated</td><td align="left">2,824</td><td align="left">13.4</td></tr><tr><td align="left"> Census programmatic error</td><td align="left">895</td><td align="left">4.2</td></tr><tr><td align="left"><bold>Total population eligible for vaccination</bold></td><td align="left"><bold>17,340</bold></td><td></td></tr><tr><td align="left"> Vaccinated</td><td align="left">12,830</td><td align="left">74.0</td></tr><tr><td align="left"> Ineligible (fever/pregnancy/etc.)</td><td align="left">69</td><td align="left">0.4</td></tr><tr><td align="left"> Absent</td><td align="left">1,811</td><td align="left">10.4</td></tr><tr><td align="left"> Refused</td><td align="left">2,613</td><td align="left">15.1</td></tr><tr><td align="left"> Already received typhoid vaccine</td><td align="left">17</td><td align="left">0.1</td></tr></tbody></table></table-wrap><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Demographic Data of study population present at the time of Mass Vaccination Campaign by area of residence in Karachi, Pakistan – 2003</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td></td><td></td><td></td><td align="left" colspan="3">Hijrat</td><td align="left" colspan="3">Sultanabad</td></tr></thead><tbody><tr><td></td><td align="left"><bold>Total</bold></td><td align="left"><bold>Received vaccine</bold></td><td align="left"><bold>%</bold></td><td align="left"><bold>Total</bold></td><td align="left"><bold>Received vaccine</bold></td><td align="left"><bold>%</bold></td><td align="left"><bold>Total</bold></td><td align="left"><bold>Received vaccine</bold></td><td align="left"><bold>%</bold></td></tr><tr><td colspan="10"><hr></hr></td></tr><tr><td align="left"><bold>Total</bold></td><td align="left">17,340</td><td align="left">12,830</td><td align="left">74</td><td align="left">9,832</td><td align="left">7,160</td><td align="left">73</td><td align="left">7,508</td><td align="left">5,670</td><td align="left">76</td></tr><tr><td align="left"><bold>Gender</bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left"> Male</td><td align="left">9,232</td><td align="left">6,767</td><td align="left">73</td><td align="left">5,139</td><td align="left">3,701</td><td align="left">72</td><td align="left">4,093</td><td align="left">3,066</td><td align="left">75</td></tr><tr><td align="left"> Female</td><td align="left">8,108</td><td align="left">6,063</td><td align="left">75</td><td align="left">4,693</td><td align="left">3,459</td><td align="left">74</td><td align="left">3,415</td><td align="left">2,604</td><td align="left">76</td></tr><tr><td align="left"><bold>Age group, yrs</bold></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td align="left"> < 5</td><td align="left">4,788</td><td align="left">3,585</td><td align="left">75</td><td align="left">2,546</td><td align="left">1,881</td><td align="left">74</td><td align="left">2,242</td><td align="left">1,704</td><td align="left">76</td></tr><tr><td align="left"> 5 – 10</td><td align="left">7,909</td><td align="left">6,149</td><td align="left">78</td><td align="left">4,553</td><td align="left">3,518</td><td align="left">77</td><td align="left">3,356</td><td align="left">2,631</td><td align="left">78</td></tr><tr><td align="left"> > 10</td><td align="left">4,643</td><td align="left">3,096</td><td align="left">67</td><td align="left">2,733</td><td align="left">1,761</td><td align="left">64</td><td align="left">1,910</td><td align="left">1,335</td><td align="left">70</td></tr></tbody></table></table-wrap></sec><sec><title>Safety</title><p>There were 116 children with AE (<1%) of which 108 were detected by passive surveillance by the vaccination teams or health center staff. Of the 116 events, 53 were considered by the study physicians to be probably related to the vaccines. Among 139 persons surveyed three days after vaccination, 5 solicited AE were detected and none was considered a serious event. Three persons were hospitalized post immunization and were managed as an SAE until the DSMB and clinical monitor labeled them not to be related with the study vaccines. One child developed petchial heamorrhges and later was found out to be having a bleeding disorder. Another child was admitted with fever and was diagnosed as having culture proven typhoid. The third one had developed an injection abscess. The main adverse events reported included fever (48), local pain (56) and local swelling (15). No needle stick injuries were reported.</p></sec><sec><title>Cold chain</title><p>There was no important deviation of the cold chain at any storage site. The central warehouse maintained the vaccine from +5°C to +6°C (mean +5.8°C); the AKU warehouse at +2°C to +8°C (mean 4.8°C), and the logistics center at +1°C to +12°C (mean 4.8°C). No vaccines were frozen. Alternate power supply was not needed at any locale.</p><p>The temperatures recorded at the vaccination posts were +3°C to +20°C (mean +4°C). These were maintained by 4 or 5 ice packs per cool box. The highest temperatures were observed during the busy hours when cool boxes were opened frequently. Temperatures were never above 8°C for more than 2 hours and were within the manufacturer's recommended guidelines. Thus, no vaccine had to be discarded because of temperature variation.</p></sec><sec><title>Resources and supplies</title><p>On an average, 389 children were vaccinated per day and each vaccination team worked 7 hours a day for 33 days. A total of 12,837 vaccine doses were opened during the vaccination campaign. Seven doses were not used because the needle was injected in the vein.</p></sec></sec><sec><title>Discussion</title><p>The results of the program in two Karachi squatter settlements show that a large-scale vaccination program has good acceptance (74% vaccine coverage) and poses no major safety problems. The cold chain was maintained throughout the study in acceptable ranges. Thus, a mass vaccination campaign in squatter settlements is logistically feasible and safe.</p><p>Improved sanitation and food hygiene are the long-term solutions for reducing or eliminating typhoid fever [<xref ref-type="bibr" rid="B18">18</xref>] but these approaches are linked to socio-economic progress, which is slow in areas endemic for the disease. Hence large immunization schemes as public health measures are being recommended [<xref ref-type="bibr" rid="B9">9</xref>].</p><p>Some suggestions and plans for adoption based on our trial are considered below:</p><sec><title>Sample size</title><p>The targeted sample size could not be achieved due to refusals and non-response during the vaccination campaign. Since; the target population in the study setting was lower than expected. The effect of decreased coverage on the power of the study and hence on the result was obvious. Therefore in consultation with a team of statisticians the number of clusters was increased to 120 and the study was extended in another setting of Karachi with comparable socio-economic characteristics. In 2004 in a separate vaccination campaign we were able to vaccinate 14406 children of the similar age group from the remaining 60 clusters. In this way the targeted sample size was achieved. However it is very important that careful assessment of factors that affect response in needed in future trials to overcome problems that affect statistical power.</p></sec><sec><title>Community involvement</title><p>First, the community mobilization strategy is pivotal. The information dissemination strategy, which was based on early stage involvement of community members with project personnel, provided frequent opportunities for clarifying misconceptions. This, in turn, brought about unprecedented community rapport and response. Community leaders took children to the vaccination posts when their mothers or female guardians were unable to do so and facilitated the vaccination of teenage girls. More intensive campaigns in areas with low vaccine uptake, areas that are politically and religiously unique from other clusters could increase coverage.</p><p>Although refusals rates were high in this vaccination campaign no direct resistance was seen. Some of the concerns mentioned included; 1) Why has this site been chosen for the trial, 2) Why are vaccines coded? 3) Why is the vaccine given for free? These concerns were dealt with by senior field staff. A senior person from the team would visit the household to answer questions. The questions that were asked for the first time were then added to our list of frequently asked questions and were discussed in the next interaction with the community.</p><p>Religious leaders have a significant influence on the intervention projects in developing countries, especially in Muslim majority populations. We interacted with them during the campaign after realizing the religious concerns of the people. The religious institutes such as mosques should be involved as part of community mobilization plan very early in the project.</p><p>Nevertheless we are aware the trial nature of the vaccination process could have impinged on the coverage results. Collectively those who refused to participate in the trial (15%) and those present and did not show-up (10%) make up a significant proportion of the target population (25%). The parents who refused to vaccinate their children were hesitant to give their children a vaccine with a code, despite detailed information was provided to them by the project staff. Out side trial conditions the refusal rate is therefore expected to be lower.</p></sec><sec><title>Cost-effectiveness</title><p>Cost-effectiveness or cost-benefit of Vi PS for Pakistan is critical to convince decision-makers to finance public sector use of a new vaccine. Data from slums in New Delhi, India, indicate that typhoid fever is a disease with high economic consequences where the direct costs of the disease was more than US $100 per hospitalized typhoid fever episode. More than half of these costs come directly "out of pocket" rather than through government subsidies [<xref ref-type="bibr" rid="B19">19</xref>]. Another vaccine policy analysis in the area also suggests that a vaccine such as Vi PS would provide cost-saving to society [<xref ref-type="bibr" rid="B20">20</xref>]. Current results from this trial suggest the logistic feasibility, but it is understood that quantification and cost-estimation would be an important factor in terms of acceptance by the government.</p></sec><sec><title>Implementation outside of EPI program</title><p>Vi PS vaccine, which does not induce immunological memory [<xref ref-type="bibr" rid="B4">4</xref>], is probably not a long-term vaccine option for Pakistan and other typhoid fever-endemic countries for that matter. Recently a Vi conjugate vaccine was shown to have high efficacy (89%) in Vietnamese children (ages 2–5 years) 46 months after immunization [<xref ref-type="bibr" rid="B21">21</xref>]. This vaccine potentially could provide long-term protection. Until the commercial availability of this vaccine, it will be important to consider alternative delivery scheme using the currently available typhoid vaccine. The trial has shown that approaching the population at risk – children aged 2 to 16 – through a mass immunization campaign was feasible.</p></sec><sec><title>Timing of the campaign</title><p>The reported campaign took place at the end of school vacations when many households were visiting outside the study area. There were a significant (13%) proportion of migration between the census and the vaccine campaign (6-month period) but these findings should not be regarded as an impediment to the introduction of a Vi PS campaign.</p></sec></sec><sec><title>Conclusion</title><p>The introduction of new vaccines into developing countries, in particular to the most impoverished, requires basic research, clinical evaluations, epidemiological assessments, policy and economic research, establishment of production facilities, sound regulatory systems, procurement mechanisms, and distribution capabilities [<xref ref-type="bibr" rid="B22">22</xref>,<xref ref-type="bibr" rid="B23">23</xref>] In areas where typhoid fever is a serious and widespread problem, Vi PS vaccination appears to be the most promising available strategy for the control of typhoid fever in the short and mid term. The mass vaccination campaign described here showed that conducting a mass immunization outside of the EPI program and infrastructure is feasible and acceptable, and could potentially be implemented in a public health system.</p></sec><sec><title>Competing interests</title><p>Vaccines were donated by GSK Biologicals, Rixensart Belgium.</p></sec><sec><title>Authors' contributions</title><p>MIK was involved in the conduct of the mass vaccination campaign, supervision of data computerization and cleaning, analysis and drafting the manuscript RLO conduct of mass vaccination campaign, supervision of data computerization and cleaning – HBH conduct of mass vaccination campaign, supervision of data computerization and cleaning – SMS conduct of mass vaccination campaign, community mobilization and supervision of data computerization and cleaning – MAH coordinated the campaign activities in one site and supported the data management unit in data computerization – SBS was involved in the design and conduct of the trial – NSB coordinated campaign activities – SR supervised the data computerization and assisted in data analysis – MKP designed the data computerization software and assisted in data analysis – MA designed the data computerization software, helped in data analysis plan and assisted in data analysis – SMW coordinated the campaign and supervised the adverse event follow up – MJK supervised the surveillance for adverse events and coordinated the community mobilization – RAE was involved in design and development of procedure manuals – BI was involved in design and planning – CMG was involved in design and also planned the laboratory procedures – TP was involved in design and planning – AD designed the analysis plan, stratification and randomization – LvS was involved in planning, designing, and conduct of the study – CJA was involved in design, conduct, analysis, drafting the manuscript – JC conceived the study and was involved in design and conduct – SQN was involved in design and conduct – ZAB supervised the planning, design and implementation phase.</p><table-wrap position="float" id="T3"><label>Table 3</label><caption><p>Details of logistics used during the vaccination campaign in Karachi – Pakistan 2003</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left"><bold>Forms</bold></td><td align="left"><bold>First Aid Box</bold></td><td align="left"><bold>Basket</bold></td></tr></thead><tbody><tr><td align="left">Record book</td><td align="left">Blood collection tubes</td><td align="left">Resuscitator</td></tr><tr><td align="left">Member list</td><td align="left">Syringes (5 ml)</td><td align="left">Spirit swabs</td></tr><tr><td align="left">Household list</td><td align="left">Tube stand</td><td align="left">Cotton balls</td></tr><tr><td align="left">ID card (undistributed)</td><td align="left">Tourniquit</td><td align="left">Disposable gloves</td></tr><tr><td align="left">Informed consent</td><td align="left">Butterfly needles</td><td align="left">Disposal bags</td></tr><tr><td align="left">Daily logistics</td><td align="left">Inj Epinephrine</td><td align="left">Lamp</td></tr><tr><td align="left">Temperature chart</td><td align="left">Inj Dexomethazone</td><td align="left"><bold>Additional supply</bold></td></tr><tr><td align="left">Transfer sheet</td><td align="left">Syringes (1 ml)</td><td align="left">Umbrella</td></tr><tr><td align="left">Progress sheet</td><td align="left">Scissors</td><td align="left">Fan</td></tr><tr><td align="left">Tally sheet</td><td align="left">Handiplast</td><td align="left">Water cooler</td></tr><tr><td align="left">Attendance sheet</td><td align="left">Thermometer</td><td align="left"><bold>Others</bold></td></tr><tr><td align="left">Immunogenicity</td><td align="left">Extra needles</td><td align="left">Safety box</td></tr><tr><td align="left">Economics</td><td align="left">Soap</td><td align="left">Juices</td></tr><tr><td align="left">IAE</td><td align="left"><bold>Shoulder bag</bold></td><td></td></tr><tr><td align="left">AE definition</td><td align="left">Pen</td><td></td></tr><tr><td align="left"><bold>Icebox</bold></td><td align="left">Marker</td><td></td></tr><tr><td align="left">Icepacks</td><td align="left">Notepad</td><td></td></tr><tr><td align="left">Replacement</td><td align="left">Stamp pad</td><td></td></tr><tr><td align="left">Thermometer</td><td align="left">Duct tape</td><td></td></tr></tbody></table></table-wrap><fig position="float" id="F2"><label>Figure 2</label><caption><p>Cluster wise distribution of vaccine codes in Vi demonstration project Karachi – Pakistan 2003.</p></caption><graphic xlink:href="1745-6215-7-17-2"/></fig><fig position="float" id="F3"><label>Figure 3</label><caption><p>Cluster wise distribution of vaccination coverage during mass vaccination campaign in Vi demonstration project Karachi – Pakistan 2003.</p></caption><graphic xlink:href="1745-6215-7-17-3"/></fig></sec>
Overexpression of connexin 43 using a retroviral vector improves electrical coupling of skeletal myoblasts with cardiac myocytes <italic>in vitro</italic>	<sec><title>Background</title><p>Organ transplantation is presently often the only available option to repair a damaged heart. As heart donors are scarce, engineering of cardiac grafts from autologous skeletal myoblasts is a promising novel therapeutic strategy. The functionality of skeletal muscle cells in the heart milieu is, however, limited because of their inability to integrate electrically and mechanically into the myocardium. Therefore, in pursuit of improved cardiac integration of skeletal muscle grafts we sought to modify primary skeletal myoblasts by overexpression of the main gap-junctional protein connexin 43 and to study electrical coupling of connexin 43 overexpressing myoblasts to cardiac myocytes <italic>in vitro</italic>.</p></sec><sec sec-type="methods"><title>Methods</title><p>To create an efficient means for overexpression of connexin 43 in skeletal myoblasts we constructed a bicistronic retroviral vector MLV-CX43-EGFP expressing the human connexin 43 cDNA and the marker EGFP gene. This vector was employed to transduce primary rat skeletal myoblasts in optimised conditions involving a concomitant use of the retrovirus immobilising protein RetroNectin<sup>® </sup>and the polycation transduction enhancer Transfectam<sup>®</sup>. The EGFP-positive transduced cells were then enriched by flow cytometry.</p></sec><sec><title>Results</title><p>More than four-fold overexpression of connexin 43 in the transduced skeletal myoblasts, compared with non-transduced cells, was shown by Western blotting. Functionality of the overexpressed connexin 43 was demonstrated by microinjection of a fluorescent dye showing enhanced gap-junctional intercellular transfer in connexin 43 transduced myoblasts compared with transfer in non-transduced myoblasts. Rat cardiac myocytes were cultured in multielectrode array culture dishes together with connexin 43/EGFP transduced skeletal myoblasts, control non-transduced skeletal myoblasts or alone. Extracellular field action potential activation rates in the co-cultures of connexin 43 transduced skeletal myoblasts with cardiac myocytes were significantly higher than in the co-cultures of non-transduced skeletal myoblasts with cardiac myocytes and similar to the rates in pure cultures of cardiac myocytes.</p></sec><sec><title>Conclusion</title><p>The observed elevated field action potential activation rate in the co-cultures of cardiac myocytes with connexin 43 transduced skeletal myoblasts indicates enhanced cell-to-cell electrical coupling due to overexpression of connexin 43 in skeletal myoblasts. This study suggests that retroviral connexin 43 transduction can be employed to augment engineering of the electrocompetent cardiac grafts from patients' own skeletal myoblasts.</p></sec>	<contrib id="A1" equal-contrib="yes" corresp="yes" contrib-type="author"><name><surname>Tolmachov</surname><given-names>Oleg</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" equal-contrib="yes" contrib-type="author"><name><surname>Ma</surname><given-names>Yu-Ling</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Themis</surname><given-names>Michael</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Patel</surname><given-names>Pravina</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Spohr</surname><given-names>Hilmar</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>MacLeod</surname><given-names>Kenneth T</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A7" contrib-type="author"><name><surname>Ullrich</surname><given-names>Nina D</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A8" contrib-type="author"><name><surname>Kienast</surname><given-names>Yvonne</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A9" contrib-type="author"><name><surname>Coutelle</surname><given-names>Charles</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A10" contrib-type="author"><name><surname>Peters</surname><given-names>Nicholas S</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib>	BMC Cardiovascular Disorders	<sec><title>Background</title><p>There is only minor potential for cell renewal in the adult myocardium [<xref ref-type="bibr" rid="B1">1</xref>]. Loss of myocardial cells due to cardiac disease results in impairment of cardiac contractile and electrical function, and despite currently available medical therapies the mortality rate remains substantial and worsens with deteriorating myocardial function [<xref ref-type="bibr" rid="B2">2</xref>]. Cardiac transplantation is currently the treatment offered to selected patients with end-stage left ventricular dysfunction, but due to limited donor supply and resources, this mode of therapy will remain confined to a minority of patients [<xref ref-type="bibr" rid="B3">3</xref>].</p><p>Cell transplantation is a promising means of repairing damaged myocardium. A number of different cell types and their combinations are under investigation for transplantation to the ventricular myocardium, including neonatal or fetal cardiomyocytes, autologous skeletal myoblasts, fibroblasts, hematopoetic stem cells, and embryonic stem cell-derived cells [<xref ref-type="bibr" rid="B4">4</xref>-<xref ref-type="bibr" rid="B6">6</xref>]. Transplanted cells have been widely reported to engraft into the host myocardium, but with variability in the degree of differentiation and integration with the host tissue [<xref ref-type="bibr" rid="B7">7</xref>-<xref ref-type="bibr" rid="B9">9</xref>].</p><p>Owing to immunological, ethical and practical advantages over some of the other cell types, transplantation of autologous skeletal myoblasts for myocardial repair was the first to undergo clinical trials [<xref ref-type="bibr" rid="B10">10</xref>,<xref ref-type="bibr" rid="B11">11</xref>]. Although modest improvements of cardiac function have been reported, ventricular tachycardia was observed in a number of patients indicating absence of electrical incorporation of the grafted cells into the host myocardium. In general, the results obtained from cell, animal and human studies have indicated that the implanted myoblasts showed fusion and differentiated into multinucleated myotubes, did not transdifferentiate to cardiac myocytes and did not couple with the host cardiac myocytes. It is possible that lack of electrical coupling of the implanted cells with the host myocytes is the key factor blocking adequate functional incorporation of the grafted skeletal myoblasts into the beating cardiac muscle [<xref ref-type="bibr" rid="B12">12</xref>]. Indeed, the major myocardial gap junctional protein connexin 43 is not expressed in mature skeletal myotubes [<xref ref-type="bibr" rid="B13">13</xref>]. We therefore addressed the hypothesis that connexin 43 overexpression by retroviral transduction of the skeletal myoblasts can enhance their electrical integration with cardiac myocytes in co-culture.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title><italic>E. coli </italic>strain, transformation and bacterial culture conditions</title><p><italic>E. coli </italic>strain DH10B F<sup>- </sup><italic>mcrA </italic>Δ<italic>(mrr-hsdRMS-mcrBC) </italic>ϕ<italic>80lacZ</italic>Δ<italic>M15 </italic>Δ<italic>lacX74 recA1 endA1 araD139 </italic>Δ<italic>(ara, leu)7697 galU galK </italic>λ<sup>- </sup><italic>rpsL(Str<sup>R</sup>) nupG </italic>(Invitrogen) was used as a plasmid host. This strain was transformed by electroporation using the Gene Pulser<sup>® </sup>II apparatus (BioRad). Bacterial clones were cultured in the LB medium supplemented with appropriate antibiotics [<xref ref-type="bibr" rid="B14">14</xref>].</p></sec><sec><title>Plasmids, oligonucleotides, PCR and sequencing</title><p>Plasmid pGF1 containing human connexin 43 cDNA [<xref ref-type="bibr" rid="B15">15</xref>] and primers CX-NHE 5'-TACTGCAGAA GCTAGCAGCC GCCACCATGG GTGACTGGAG CGCCTTAGGC-3' and CX-BAM 5'-TGTTACTATA GGATCCCTAG ATCTCCAGGT CATCAGGCC-3' were employed to amplify the connexin 43 gene by PCR using high fidelity Pfx DNA polymerase (Invitrogen). The forward PCR primer CX-NHE included a recognition sequence for <italic>Nhe</italic>I followed by the consensus extended Kozak sequence 5'-GCCGCCACC-3'. The reverse PCR primer CX-BAM contained a <italic>Bam</italic>HI recognition sequence at its 5' end. The obtained PCR product was cut by <italic>Nhe</italic>I and <italic>Bam</italic>HI and inserted into the <italic>Nhe</italic>I and <italic>Bam</italic>HI digested plasmid pIRES2EGFP (Clontech) to produce the new plasmid pCX43-IRES-EGFP. Then the entire connexin 43 cDNA insert was sequenced (Advanced Biotechnology Centre, Imperial College London) using primers CX-SEQ2 5'-TTGCTGCGAACCTACATCATCAGT-3', CX-SEQ3 5'-TTTCAATGGC TGCTCCTCACCAAC-3' and CX-SEQ4 5'-GAAGATGGTTTTCTCCGTGGGGCGAGA-3' to confirm its faithful reproduction by PCR.</p><p>The CX43-IRES-EGFP bicistronic expression cassette was excised using <italic>Ase</italic>I and <italic>Not</italic>I to include the CMV promoter but to exclude the SV40 polyadenylation signal. This fragment was blunt-ended by the Klenow fragment of <italic>Escherichia coli </italic>DNA polymerase I (KF) and inserted between KF-polished <italic>Eco</italic>RI and <italic>Not</italic>I ends of the digested retroviral MLV vector plasmid pLZRS-LacZ(A) [<xref ref-type="bibr" rid="B16">16</xref>]. The resultant retroviral vector plasmid was called pMLV-CX43-EGFP (Figure <xref ref-type="fig" rid="F1">1</xref>).</p><fig position="float" id="F1"><label>Figure 1</label><caption><p><bold>Plasmid pMLV-CX43-EGFP containing the DNA sequence for the retroviral vector MLV-CX43-EGFP</bold>. A retroviral backbone plasmid, pMLV-CX43-EGFP, co-expressing human connexin 43 cDNA and the EGFP gene, was generated on the basis of the retroviral backbone plasmid pLZRS-LacZ(A). We used the internal ribosome entry site (IRES) to arrange the connexin 43 sequence and the downstream EGFP marker gene within one single transcription unit under control of a tandem of MLV LTR promoter and CMV immediate early promoter. The obtained plasmid pMLV-CX43-EGFP contains the Epstein-Barr virus EBNA-1-oriP segment driving episomal replication in the producer cell line and, thus, improving chances for selection of a high titre virus producing clone. The EGFP marker of the retroviral vector simplifies identification of transduced cells and is useful for evaluation of the viral vector titre.</p></caption><graphic xlink:href="1471-2261-6-25-1"/></fig></sec><sec><title>Tissue culture techniques and construction of the producer cell line generating retroviral vector MLV-CX43-EGFP</title><p>Packaging cells, virus producer cells, and immortalised L6 rat myoblasts (ATCC CRL-1458) were grown in DMEM with glutamax-I supplemented with 10% FCS (Invitrogen) at 37°C with 5% CO<sub>2 </sub>in air.</p><p>To generate retroviral producer cell lines, the amphotropic MLV packaging cell line TEFLYA [<xref ref-type="bibr" rid="B17">17</xref>] was transfected with the obtained plasmid pMLV-CX43-EGFP. Transfection was performed using Lipofectamin<sup>® </sup>(Invitrogen) as recommended by the supplier. Transfected cells were selected in the medium supplemented with puromycin (10 μg/ml). Different dilutions of the trypsinised transfected cells were prepared and seeded onto 96 well plates. 267 individual clones were produced and expanded in 24 well plates. The clones were screened using an inverted fluorescence microscope (Leitz) for EGFP expression and also for viral vector production by transduction of L6 rat skeletal myoblasts. 39 clones (14.6%) were fluorescent of which 17 clones (6.4%) produced virus. Screening of supernatants from 267 potential producer cell lines revealed a broad variation in the titre of produced virus, determined by transduction of L6 rat skeletal myoblasts in the presence of 10 μg/ml of Polybrene™. Clones A113, A189 and A247 produced virus with the highest end-point titre (1 × 10<sup>6 </sup>TU/ml in the freshly collected virus-containing medium). Clone A247 was used as a source of the MLV-CX43-EGFP virus vector for further experiments. Virus vector preparations were produced by centrifugation of culture supernatants from producer cells at 5000 rpm for 10 min and filtration of the obtained fluid through 0.8 μm filters (Sartorius). The virus vector preparations were stored frozen at -80°C.</p></sec><sec><title>Isolation and culture of skeletal myoblasts and cardiac myocytes</title><p>The manipulations of animals in this work conform to UK Home Office guidelines. Primary rat skeletal myoblasts were isolated from the hind leg muscles of adult male Wistar rats. The muscle slices were digested in 0.25% pancreatin, 1% trypsin for 1 hour with occasional agitation. The isolated cells were collected by filtering through 70 μm nylon cell strainers (Falcon). Counter-selection against fibroblasts was accomplished by 2 rounds of differential adhesion on collagen coated tissue culture flasks (40 min at 37°C for each adhesion step). Primary myoblasts were cultured in a CO<sub>2 </sub>incubator at 37°C in DMEM with glutamax-I (Invitrogen) supplemented with 20% FCS and further purified by sorting using paramagnetic beads (Dynal Biotech) coated with antibody H36 against myoblast specific α-7 integrin [<xref ref-type="bibr" rid="B18">18</xref>].</p><p>Cardiac myocytes were isolated from neonatal rats. Freshly excised ventricles were dissociated in trypsin-EDTA (Invitrogen) and the dispersed cells were suspended in the culture medium, a 4:1 mixture of DMEM and M199 media (Invitrogen) supplemented with 15% horse serum and 5% FCS. The cell suspension was pre-plated to separate fibroblasts from myocytes as described for skeletal myoblasts above. The myocytes remaining in the suspension were cultured in a CO<sub>2 </sub>incubator at 37°C.</p></sec><sec><title>Retroviral transduction of primary skeletal myoblasts and FACS analysis</title><p>Filtered preparation of the MLV-CX43-EGFP was poured into RetroNectin<sup>® </sup>(TaKaRa) coated plates for the virions to attach. Myoblasts were loaded onto the immobilised viral vector particles in the DMEM supplemented with 20% FCS and, optionally, 5 μg/ml Transfectam<sup>® </sup>(dioctadecylamidoglycyl spermine, DOGS, Promega) or 10 μg/ml Polybrene™ (hexadimethrine bromide, Sigma). The proportion of the transduced cells was determined by FACS analysis on a FACSCalibur machine (Becton Dickinson) relying on expression of the EGFP transgene. The total number of counted cells was 10000 in all FACS measurements. The transduced cells were sorted using a preparative FACS machine (Becton Dickinson FACS DIVA cell sorter) to produce a cell population with more than 70% of the cells expressing the EGFP marker (Figure <xref ref-type="fig" rid="F2">2D</xref>). The sorted cells were passaged once and their myogenic nature was confirmed by immunoconfocal analysis with a monoclonal anti-desmin antibody (clone DE-U-10, product number D033 from Sigma-Aldrich) as shown in <xref ref-type="supplementary-material" rid="S2">Additional file 2</xref>: Additional_file_2.pdf.</p><fig position="float" id="F2"><label>Figure 2</label><caption><p><bold>Transduction of primary rat skeletal myoblasts by the retroviral vector MLV-CX43-EGFP and resultant overexpression of connexin 43</bold>. Connexin 43 and EGFP were expressed from a single bicistronic transcription unit. <bold>(A) </bold>Transduction of primary rat skeletal myoblasts using various dilutions of the MLV-CX43-EGFP viral vector preparation. Medium in the wells of a 24-well plate containing non-confluent primary myoblasts was aspirated and the wells were filled with 1 ml of the non-diluted and diluted viral vector suspensions supplemented with Polybrene™ (10 μg/ml). After 48 hours the cells were trypsinised, washed in complete DMEM medium, resuspended in PBS and used for FACS analysis to determine the efficiency of transfection relying on EGFP expression. The obtained data were averaged among 3 wells for each virus vector dilution. <bold>(B) </bold>Concomitant use of RetroNectin<sup>®</sup>-mediated viral vector concentration and polycation transduction enhancement for delivery of the connexin 43 and EGFP genes into primary skeletal myoblasts using MLV-CX43-EGFP vector. Transduction experiments were performed in 6-well plates containing non-confluent myoblasts in the presence of the polycation Polybrene™ (10 μg/ml), in the presence of the polycation Transfectam<sup>® </sup>(5 μg/ml) or without addition of polycations. The efficiency of transduction was determined using the EGFP marker from the FACS subtraction plots using non-transduced primary skeletal myoblasts as a control, non-fluorescent cell population. The data were averaged among 6 individual transduction experiments. <bold>(C) </bold>Fractions of EGFP-positive cells in the population of connexin 43/EGFP transduced primary skeletal myoblasts during continuous passaging. The cells were split and re-seeded in the ratio 1:10 each week with concomitant FACS measurement of the fraction of the EGFP-expressing cells. <bold>(D) </bold>Results of FACS analysis of populations of primary skeletal myoblasts plotted as histograms (x-axis – intensity of green fluorescence, y-axis – counts of cells). Purple graph corresponds to non-transduced primary skeletal myoblasts, red graph corresponds to the primary skeletal myoblasts after RetroNectin<sup>®</sup>-mediated Transfectam<sup>®</sup>-enhanced transduction by MLV-CX43-EGFP, green graph represents primary skeletal myoblasts, which were preparatively sorted for EGFP expression after transduction by MLV-CX43-EGFP. <bold>(E) </bold>Western blotting analysis of connexin 43 expression in connexin 43 transduced skeletal myoblasts (after EGFP-based preparative sorting). The cell extracts were analysed by electrophoresis in a SDS-PAGE gel and immunoblotting using anti-connexin 43 antibody. Lane 1 corresponds to non-transduced primary skeletal myoblasts, lane 2 corresponds to connexin 43 transduced primary skeletal myoblasts, lane 3 corresponds to connexin 43 transduced primary skeletal myoblasts, which were passaged 13 weeks in tissue culture.</p></caption><graphic xlink:href="1471-2261-6-25-2"/></fig><p>The obtained population of connexin 43 transduced myoblasts was frozen in liquid nitrogen in the DMEM containing 20% FCS and 10% DMSO. Aliquots of the frozen transduced skeletal myoblasts were used to seed subcultures for Western blotting analysis, dye transfer and electrophysiological experiments.</p></sec><sec><title>Western blotting analysis of connexin 43 expression</title><p>Quantitative immunoblotting was performed to confirm overexpression of connexin 43 in the skeletal myoblasts containing the connexin 43 transgene. Seeded cells were allowed to expand until they reached complete confluence (7 days). Then the cells were scraped off, homogenised in the lysis buffer SB20 (20% SDS and 0.15 M Tris-HCl pH 6.8) and sonicated to shear the genomic DNA. The protein concentration was determined and adjusted to 0.5 mg/ml. Samples with 5.0 μg of total protein were resolved by SDS-PAGE on a 12.5% gel and transferred onto a polyvinylidene difluoride membrane. The mouse primary antibody against connexin 43 (Chemicon), the secondary alkaline phosphatase-conjugated anti-mouse antibody (Pierce) and the proprietary alkaline phosphatase substrate (Promega) were used to detect connexin 43. The protein bands were quantified by densitometry. Two independent cell cultivation experiments were performed and two independent gels were analysed by Western blotting for each cultivation experiment.</p></sec><sec><title>Microinjection and intercellular transfer of a fluorescent dye</title><p>Skeletal myoblasts transduced with the MLV-CX43-EGFP vector, control non-transduced skeletal myoblasts and co-cultures of these cells with cardiac myocytes were grown in culture dishes to ~80% confluence. The culture dishes were then placed onto the stage of an inverted fluorescence microscope. Microelectrodes with resistance 50–60 MΩ were loaded with a 5% solution of Cascade Blue derivative 8-methoxypyrene-trisulphonic acid (Molecular Probes, Oregon, USA) and were back-filled with 1 M LiCl. The dye was iontophoresed into each cell by 4–6 nA current for 2 min following impalement of a cell by the electrode. Dye transfer to the adjacent cells was recorded using a Nikon digital camera (Coolpix 990). Dye transfer images were captured 2 min after injection. The images were used to score the number of neighbouring cells to which the dye was transferred from each injected cell.</p></sec><sec><title>Electrophysiological measurements in cell co-cultures</title><p>Cell culture dishes incorporating a group of 60 embedded unipolar electrodes with diameter 30 μm with interelectrode distances of 100 μm (Multielectrode Array, MCS GmbH, Reutlingen, Germany) were used to study and compare the electrical integration of skeletal myoblasts (connexin 43/EGFP transduced and non-transduced) in co-cultures with cardiac myocytes. Five groups of cell cultures were under investigation: 1) cardiac myocytes alone; 2) skeletal myoblasts alone; 3) connexin 43 transduced skeletal myoblasts alone; 4) cardiac myocytes co-cultured in a ratio of 4:1 with connexin 43 transduced skeletal myoblasts; 5) cardiac myocytes co-cultured in the same ratio with non-transduced skeletal myoblasts. To establish co-cultures, cardiac myocytes were seeded in the multielectrode array dishes (1 million cells per dish) to allow the cells to settle. The cells were cultured in a medium composed of a 4:1 mixture of DMEM and M199 media (Invitrogen) supplemented with 15% horse serum and 5% FCS. At day 2 after the initial seeding, connexin 43 transduced skeletal myoblasts or non-transduced skeletal myoblasts (0.25 million cells per dish) were added to cardiac myocytes. At day 3, the cells reached confluence with both skeletal and cardiac cells distinguishable under the microscope. Then MEA dishes were placed onto the recording system and an extracellular stimulatory current was applied in 10 evenly spaced pulses (80 μA, 5 ms) during 10 s time interval. The stimulatory pulses were delivered to the cells by a pair of electrodes located outside the 60-electrode array. To register FAP, electrograms (potential against time) were recorded for 10 s. The FAP activation rate (that is, frequency of FAP firing) was determined from the discrete spikes on the electrograms with amplitude > 500 μV and duration > 5 ms using the spike sorter of the MC-Rack data analyser (Microcal Software, Northampton, MA, USA). The FAP activation rate data obtained from the 60 electrodes were then averaged.</p></sec><sec><title>Statistical analysis</title><p>The data are presented as the mean value and its standard error (mean ± standard error, M ± SE). The following formula was used to compute average standard error (ASE) of the ratio of two means (M<sub>1 </sub>± SE<sub>1</sub>, M<sub>2 </sub>± SE<sub>2</sub>): ASE = (M<sub>1</sub>/M<sub>2</sub>)*(SE<sub>1</sub><sup>2</sup>/M<sub>1</sub><sup>2</sup>+SE<sub>2</sub><sup>2</sup>/M<sub>2</sub><sup>2</sup>)<sup>1/2</sup>. The significance of the differences between experimental groups was estimated using unpaired t-test.</p></sec></sec><sec><title>Results</title><sec><title>Construction of the retroviral connexin 43 vector MLV-CX43-EGFP and optimisation of myoblast transduction conditions</title><p>To be able to deliver the connexin 43 gene to skeletal myoblasts we constructed an amphotropic retroviral vector MLV-CX43-EGFP, co-expressing human connexin 43 cDNA and the EGFP marker gene (Figure <xref ref-type="fig" rid="F1">1</xref>).</p><p>The passaging time of primary skeletal myoblasts in tissue culture before transplantation is limited because of clinical considerations. Therefore, it is important to achieve maximum cell transduction efficiency with a minimal number of exposures of the myoblasts to the viral vector. Amphotropic MLV vector preparations commonly contain substances, which severely limit the efficiency of transduction. In an attempt to minimise the adverse effect of the transduction inhibitors, we used various dilutions of the MLV-CX43-EGFP viral vector preparation with end-point titre 2 × 10<sup>5 </sup>TU/ml for infection of the primary rat skeletal myoblasts in the presence of 5 μg/ml of the commonly used transduction enhancer polycation Polybrene™ (hexadimethrine bromide). The highest efficiency of transduction (9.94 ± 0.50%) was achieved with 5-fold dilution of the viral vector and not with undiluted vector, confirming contamination of the vector preparation by transduction inhibitors (Figure <xref ref-type="fig" rid="F2">2A</xref>). To increase efficiency of transduction by the MLV-CX43-EGFP vector, we tested RetroNectin<sup>®</sup>-mediated virion capture [<xref ref-type="bibr" rid="B19">19</xref>] to immobilise the vector particles on a plastic surface and, thus, to get rid of the RetroNectin<sup>®</sup>-unbound portion of the transduction inhibitors. We compared the standard RetroNectin<sup>®</sup>-mediated transduction protocol, which is performed without polycation transduction enhancers, to RetroNectin<sup>®</sup>-mediated transduction in the presence of Polybrene™ or, alternatively, in the presence of lipopolyamine Transfectam<sup>® </sup>(dioctadecylamidoglycyl spermine, [<xref ref-type="bibr" rid="B20">20</xref>]). The obtained transduction efficiency data are summarised in Figure <xref ref-type="fig" rid="F2">2B</xref>. Employment of Transfectam<sup>® </sup>to enhance transduction of primary skeletal myoblasts by RetroNectin<sup>®</sup>-immobilized MLV-CX43-EGFP vector allowed transduction of 38.30 ± 0.89% cells. In contrast, the efficiency of transduction was 22.56 ± 0.75% in the RetroNectin<sup>®</sup>/Polybrene™ experiments and 14.34 ± 0.60% when RetroNectin<sup>®</sup>-immobilised virus was used without addition of a polycation substance.</p><p>To estimate the stability of connexin 43 expression, one part of the transduced population of skeletal myoblasts was stored frozen in liquid nitrogen and another part was used for continuous passaging. During 12-week cultivation, the cells were split and re-seeded in the ratio 1:10 each week with simultaneous FACS analysis of the EGFP transgene expression. The percentage of EGFP-positive cells stayed practically constant during passaging <italic>in vitro </italic>indicating absence of the EGFP transgene shut-down and suggesting the absence of shut-down of the linked connexin 43 transgene (Figure <xref ref-type="fig" rid="F2">2C</xref>).</p><p>The passaged and the stored populations of transduced myoblasts were then used for preparative FACS to enrich EGFP expressing cells to more than 70% (Figure <xref ref-type="fig" rid="F2">2D</xref>).</p><p>Western blotting analysis showed a 4.85 ± 0.25 times overexpression of connexin 43 in the population of sorted non-passaged MLV-CX43-EGFP transduced skeletal myoblasts compared with control non-transduced skeletal myoblasts. Confirming longevity of connexin 43 expression in tissue culture, Western blotting analysis showed a 4.71 ± 0.23 times overexpression of connexin 43 in the population of passaged sorted skeletal myoblasts compared with non-transduced skeletal myoblasts (Figure <xref ref-type="fig" rid="F2">2E</xref>).</p></sec><sec><title>Connexin 43 overexpression enhances intercellular dye transfer between skeletal myoblasts</title><p>To show that connexin 43 overexpression has improved intercellular communication in skeletal myoblasts, we injected fluorescent dye 8-methoxypyrene-trisulphonic acid (a Cascade Blue derivative) into individual cells in pure cultures of connexin 43 transduced and non-transduced skeletal myoblasts (Figure <xref ref-type="fig" rid="F3">3</xref>). 13 injections of the dye were carried out in the cultures of connexin 43 transduced skeletal myoblasts, 10 of them (77%) resulted in dye migration to the neighbouring cells. In contrast, just 3 out of 11 (27%) injections led to intercellular dye spread in non-transduced skeletal myoblasts. The number of cells to which the dye permeated from the injected cell varied and was 1.38 ± 0.33 for connexin 43 transduced cells and 0.27 ± 0.14 for non-transduced ones. Thus, enhanced dye spread to neighbouring cells was observed in connexin 43 transduced skeletal myoblasts compared with non-transduced ones (P < 0.05), indicating a higher density of gap junctions in the cells overexpressing connexin 43.</p><fig position="float" id="F3"><label>Figure 3</label><caption><p><bold>Intercellular fluorescent dye microinjection in connexin 43 and EGFP transduced and non-transduced primary skeletal myoblasts</bold>. Cascade Blue derivative 8-methoxypyrene-trisulphonic acid was used as a fluorescent dye. Panel 'ABCD' shows fluorescent dye microinjection experiment in a culture of non-transduced primary skeletal myoblasts. Panel 'EFGH' shows fluorescent dye microinjection experiment in a culture of connexin 43 and EGFP transduced primary skeletal myoblasts. All pictures were taken using 400 times instrumental magnification. Blue fluorescence was visualised using Nikon filter block UV-1A (DM400). Green fluorescence was visualised using a standard 'FITC' filter block. <bold>(A) </bold>A phase contrast image of a group of adjacent non-transduced skeletal myoblasts. <bold>(B) </bold>An image of the injection microelectrode inserted into one of the adjacent cells under UV light. <bold>(C) </bold>An image of the same group of the cells under UV light after 2 min, no dye transfer. <bold>(D) </bold>An overlay of 'A' and 'C'. <bold>(E) </bold>An image of a group of adjacent connexin 43 and EGFP transduced skeletal myoblasts under UV light before dye injection (green fluorescence confirms EGFP expression). <bold>(F) </bold>An image of the injection microelectrode inserted into one of the adjacent cells under UV light. <bold>(G) </bold>An image of the same group of cells under UV light after 2 min, the dye is transferred to adjacent cells. <bold>(H) </bold>An overlay of 'E' and 'G'.</p></caption><graphic xlink:href="1471-2261-6-25-3"/></fig></sec><sec><title>Connexin 43 overexpression in skeletal myoblasts improves electrical coupling in co-cultures of cardiac myocytes and skeletal myoblasts</title><p>Co-cultures of skeletal myoblasts and cardiac myocytes were established to mimic <italic>in vivo </italic>transplantation of skeletal myoblasts to the host myocardium. The cells were grown in multielectrode array (MEA) assemblies, which allowed application of stimulatory current pulses and recording of field action potentials (FAPs) propagating in the cell population using 60 electrodes (Figure <xref ref-type="fig" rid="F4">4AC</xref>). Spontaneous FAPs were observed in 100% (5 out of 5) of cultures of pure cardiac myocytes, 40% (4 out of 10) of co-cultures of connexin 43 transduced skeletal myoblasts with cardiac myocytes and 12.5% (1 out of 8) of co-cultures of non-transduced skeletal myoblasts with cardiac myocytes. In the latter case FAP activation had a sporadic pattern. Stimulation with 10 pulses of current applied during 10 s was sufficient to obtain FAPs in 100% (10 out of 10) of co-cultures of connexin 43 transduced skeletal myoblasts with cardiac myocytes and 25% (2 out of 8) of co-cultures of non-transduced skeletal myoblasts with cardiac myocytes. Again, in the latter case the pattern of FAP activation was always only sporadic. No FAP activation was observed in 10 individual cultures of non-tranduced skeletal myoblasts and 10 individual cultures of transduced skeletal myoblasts. Typical electrograms are presented in Figure <xref ref-type="fig" rid="F4">4BD</xref> (illustration for all 60 electrodes is shown in <xref ref-type="supplementary-material" rid="S1">Additional file 1</xref>: Electrograms.pdf). The mean FAP activation rate for the 60 electrodes over 10 s after the last stimulation pulse was 2.74 ± 0.20 Hz for cardiac myocytes alone, 1.97 ± 0.34 Hz for co-culture of connexin 43 transduced skeletal myoblasts with cardiac myocytes (no significant difference compared with cardiac myocytes alone) and 0.44 ± 0.19 Hz for co-culture of non-transduced skeletal myoblasts with cardiac myocytes (P < 0.001 compared with cardiac myocytes alone, and P < 0.002 compared with co-culture of connexin 43 transduced skeletal myoblasts with cardiac myocytes).</p><fig position="float" id="F4"><label>Figure 4</label><caption><p><bold>Multi-electrode array recording in co-cultures of primary skeletal myoblasts with cardiac myocytes</bold>. Microscopic images of the multi-electrode array (MEA) dishes containing co-cultures of cardiac myocytes with non-transduced <bold>(A) </bold>or connexin 43/EGFP transduced <bold>(C) </bold>primary skeletal myoblasts. The images were taken under UV light with 100 times magnification. Skeletal myoblasts transduced with MLV-CX43-EGFP vector expressed the EGFP marker and, therefore, were fluorescent. Individual recordings (x-axis – time in s, y-axis – potential in μV) from the electrode No. 33 of the MEA in co-culture of cardiac myocytes with non-transduced skeletal myoblasts <bold>(B) </bold>and in co-culture of cardiac myocytes with connexin 43/EGFP transduced skeletal myoblasts <bold>(D)</bold>. These electrograms show the last stimulatory pulse in the series of 10 and the first resultant FAP <bold>(D) </bold>or absence of it <bold>(B)</bold>. Ten stimulatory current pulses were applied with the frequency of 1 Hz.</p></caption><graphic xlink:href="1471-2261-6-25-4"/></fig><p>Thus, a comparison of all the experimental groups indicated that electrical coupling in the co-culture of skeletal myoblasts with cardiac myocytes is significantly enhanced in the presence of overexpressed connexin 43.</p></sec></sec><sec><title>Discussion</title><p>Electric communication between cells is mediated by bursts of the action potential and gap junctions provide the low resistance pathway for its cell-to-cell propagation. In addition gap junctions mediate flux of small molecules that regulate normal tissue development and tissue patterning. It is, therefore, reasonable to hypothesise that gap junctions can play a central role in the electromechanical incorporation of cardiac grafts. Autologous skeletal myoblasts are an attractive source for cardiac transplants because of their immune privileges, availability and non-tumourogenicity (reviewed in [<xref ref-type="bibr" rid="B21">21</xref>]). However, when skeletal myoblasts differentiate into myotubes, they permanently lose expression of the major gap junctional protein, connexin 43, and, thus, do not have the apparatus for gap-junctional coupling. This is most likely to be the reason why, after transplantation of skeletal myoblasts, the newly developed graft has not been observed beating synchronously with the heart tissue [<xref ref-type="bibr" rid="B22">22</xref>]. There was no intercellular transfer and electrical coupling between the cells developed from the transplanted myoblasts and the host cardiac myocytes [<xref ref-type="bibr" rid="B23">23</xref>]. It was also reported that human patients with engrafted skeletal myoblasts suffered from ventricular tachycardia [<xref ref-type="bibr" rid="B24">24</xref>,<xref ref-type="bibr" rid="B25">25</xref>]. Therefore, in pursuit of improved cardiac integration of skeletal muscle grafts we modified primary skeletal myoblasts by overexpression of the main gap-junctional protein connexin 43.</p><p>We have chosen a retroviral MLV vector for delivery and overexpression of the connexin 43 transgene into skeletal myoblasts because retroviral vectors are able to integrate stably into the genome of the transduced cells and thus to provide long-term expression of a transgene [<xref ref-type="bibr" rid="B26">26</xref>]. Indeed, in our study, expression of connexin 43 and the EGFP marker in the transduced cells did not subside after 12 weeks of continuous passaging in tissue culture. However, long term gene expression <italic>in vitro </italic>can be only tentatively projected to long term gene expression <italic>in vivo</italic>, where transgene expression shutdown events are common [<xref ref-type="bibr" rid="B27">27</xref>]. Therefore, one should consider the possible need of 'topping up' connexin 43 gene expression in the transplanted tissue <italic>in vivo</italic>. As the engrafted cells do not actively divide, the choice of the 'topping up' vector is limited to cell division independent vectors, for example lentiviral HIV-based vectors. In this context employment of an MLV vector at the <italic>ex vivo </italic>transduction step is beneficial because it allows subsequent use of efficient and MLV-compatible lentiviral vectors for additional transduction of the grafts <italic>in vivo</italic>. If lentiviral vectors were used for initial transduction <italic>ex vivo</italic>, there would be a possibility of a reduced efficiency of the 'topping up' transgene delivery <italic>in vivo </italic>because of a CD4-independent superinfection interference of the resident lentiviral vector with the incoming one ([<xref ref-type="bibr" rid="B28">28</xref>], reviewed in [<xref ref-type="bibr" rid="B29">29</xref>]).</p><p>Absence of connexin 43 in adult muscle can be due to shutdown of the native connexin 43 promoter during myoblast differentiation. We chose the MLV LTR and the CMV promoters to drive expression of human connexin 43 cDNA because these viral promoters are known to be able to direct transcription in adult muscle and therefore they are unlikely to shut down due to changes in the balance of transcription factors in the course of myoblast differentiation. A tandem arrangement of the two promoters in the MLV-CX43-EGFP vector could reduce the chances of transcription silencing after transplantation.</p><p>The EGFP marker of the generated viral vector MLV-CX43-EGFP was useful for the purification of the transduced cells by FACS and it can also be useful at the post-grafting stage to track the transplants <italic>in vivo</italic>.</p><p>Our observations show that confluent cultures of primary myoblasts can stay alive for at least a month in medium supplemented with FCS (and considerably longer in medium without serum). This property of primary myoblasts was in stark contrast to the transformed rat L6 myoblasts, which often died in a week after achieving confluency. Thus, fortuitously, the number of culture passages (and, therefore, cell divisions) required for our manipulation of primarily myoblasts (magnetic sorting, retroviral transduction, preparative FACS) was lower, than the number of passages needed for analogous manipulations with a permanent myoblast cell line L6.</p><p>To increase the yield of connexin 43 transduced skeletal myoblasts from a single muscle biopsy it is important to achieve a high efficiency of transduction by the MLV vector. However, preparations of amphotropic MLV commonly contain infection inhibiting substances, which reduce maximal transduction efficiencies without reduction of end-point virus titres [<xref ref-type="bibr" rid="B17">17</xref>]. It is, therefore, important to optimise conditions for high efficiency of transduction. In our experiments we have shown for the first time that a combination of viral vector concentration on the plastic surface using the virus-binding protein RetroNectin<sup>® </sup>[<xref ref-type="bibr" rid="B19">19</xref>] and transduction in the presence of lipid polycation Transfectam<sup>® </sup>[<xref ref-type="bibr" rid="B20">20</xref>] is particularly effective for transduction of primary myoblasts by an amphotropic MLV vector. The achieved efficiency of transduction (38.30 ± 0.89%) can be further increased by: 1) improving the viral vector titre, for example by virion production at 32°C [<xref ref-type="bibr" rid="B30">30</xref>]; 2) additional concentration of the viral vector, e.g. by using magnetic nanoparticles [<xref ref-type="bibr" rid="B31">31</xref>] or low speed centrifugation [<xref ref-type="bibr" rid="B19">19</xref>]; 3) increasing transduction competence of the recipient cells, for example by phosphate starvation of the myoblasts [<xref ref-type="bibr" rid="B17">17</xref>] or boosting the myoblast division rate using growth factors [<xref ref-type="bibr" rid="B32">32</xref>].</p><p>We have demonstrated that connexin 43 transduction of skeletal myoblasts and ensuing connexin 43 overexpression significantly improves propagation of action potential (measured as FAP activation rate) in co-culture of cardiac myocytes and skeletal myoblasts <italic>in vitro</italic>. Enhanced gap junction formation between connexin 43 transduced skeletal myoblasts and cardiac myocytes is the most likely mechanism involved. This conjecture is supported by the results of Reinecke <italic>et al </italic>[<xref ref-type="bibr" rid="B33">33</xref>] who reported that transplantation of genetically engineered myoblasts, which were designed to express connexin 43 during differentiation, resulted in close apposition of the skeletal myotubes and the host cardiac myocytes.</p><p>Skeletal myoblasts are non-differentiated muscle cells and, unsurprisingly, we did not observe any FAP activation in pure cultures of skeletal myoblasts, whether overexpressing connexin 43 or not. Thus, although fluorescent dye transfer occurred to a significantly greater extent in connexin 43 transduced skeletal myoblasts, improved gap-junctional communication in these cells did not result in FAP generation. We registered only a limited FAP activation in co-cultures of non-transduced skeletal myoblasts with cardiac myocytes. However, with connexin 43 overexpression in the skeletal myoblasts, the FAP activation rate in the co-cultures of skeletal myoblasts and cardiac myocytes was significantly enhanced, and was close to the FAP activation rate in pure cultures of cardiac myocytes.</p></sec><sec><title>Conclusion</title><p>More than 4 times overexpression of connexin 43 in primary skeletal myoblasts was achieved after retroviral transduction in optimised conditions involving a concomitant use of the retrovirus immobilising protein RetroNectin<sup>® </sup>and the polycation transduction enhancer Transfectam<sup>®</sup>. Connexin 43 overexpression resulted in improvement of electrical coupling between transduced skeletal myoblasts and cardiac myocytes <italic>in vitro</italic>. Thus, retroviral connexin 43 transduction is a useful step for engineering of electrocompetent cardiac grafts.</p></sec><sec><title>Abbreviations</title><p>Ap – ampicillin, ASE – average standard error, Cm – chloramphenicol, DMEM – Dulbecco's modified Eagle's medium, DMSO – dimethylsulphoxide, EDTA – ethylenediaminetetraacetic acid, EGFP – enhanced green fluorescence protein, FACS – fluorescence activated cell sorting, FAP – field action potential, FCS – fetal calf serum, CMV – cytomegalovirus, HIV – human immunodeficiency virus, IRES – internal ribosome entry site, KF – Klenow fragment of <italic>Escherichia coli </italic>DNA polymerase I, LTR – long terminal repeat, M – mean, MEA – multielectrode array, MLV – murine leukemia virus, PBS – phosphate buffered saline, PCR – polymerase chain reaction, rpm – revolutions per minute, SDS-PAGE – sodium dodecyl sulfate polyacrylamide gel electrophoresis, SE – standard error, TU – transduction unit.</p></sec><sec><title>Competing interests</title><p>The authors declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>OT generated the connexin 43 retroviral vector, supervised and performed primary myoblast isolation, retroviral transductions, cell sorting and longevity study of EGFP expression, took part in Western blotting analysis of connexin 43 expression, drafted the manuscript. YM isolated primary myoblasts, transduced them with a retroviral vector, performed dye injection and multielectrode array experiments, drafted the manuscript. PP performed Western blotting analysis. HS participated in MEA experiments. KTM and NDU performed some dye injections experiments. YK performed some of the experiments in longevity study of EGFP expression. MT, CC and NSP conceived the study and revised the manuscript. All authors read and approved the final manuscript.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1471-2261/6/25/prepub"/></p></sec><sec sec-type="supplementary-material"><title>Supplementary Material</title><supplementary-material content-type="local-data" id="S1"><caption><title>Additional File 1</title><p><bold>Multi-electrode array recording in co-cultures of primary skeletal myoblasts with cardiac myocytes (data for all 60 electrodes)</bold>. Recordings from 60 electrodes in the MEA are presented as a collection of 60 individual electrograms (x-axis – time in s, y-axis – potential in μV). The time window frame was chosen to show the last stimulatory current pulse (in the series of 10, delivered with the frequency of 1 Hz). <bold>(A) </bold>A nest of electrograms showing the last stimulatory pulses and absence of any ensuing FAP spikes in co-cultures of cardiac myocytes with non-transduced skeletal myoblasts (recordings from all 60 electrodes of the MEA). <bold>(B) </bold>A nest of electrograms showing the last stimulatory pulses and the ensuing FAP spikes in co-cultures of cardiac myocytes with connexin 43 transduced skeletal myoblasts.</p></caption><media xlink:href="1471-2261-6-25-S1.pdf" mimetype="application" mime-subtype="pdf"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S2"><caption><title>Additional File 2</title><p><bold>Immunoconfocal analysis of a myogenic marker desmin in populations of primary myoblasts at an early stage and a late stage of cultivation</bold>. Immunostaining was performed with anti-desmin mouse monoclonal antibody as a primary antibody and goat anti-mouse Cy3-labelled as a secondary antibody. Cells were grown to form a monolayer on glass cover slips and were fixed with ice-cold methanol before immunostaining. Images were obtained using a Leica TCSNT confocal microscope at an instrumental magnification of 800 times. Phase contrast <bold>(A) </bold>and immunostaining <bold>(B) </bold>micrographs of primary rat myoblasts obtained after magnetic sorting with anti-α-7 integrin antibody. Phase contrast <bold>(C) </bold>and immunostaining <bold>(D) </bold>micrographs of primary rat myoblasts after magnetic sorting with anti-α-7 integrin antibody followed by an additional 4-week passaging to allow transduction with the MLV-CX43-EGFP vector and EGFP-based preparative FACS sorting. Phase contrast <bold>(E) </bold>and immunostaining <bold>(F) </bold>micrographs of NIH3T3 mouse fibroblasts (desmin-negative control). Phase contrast <bold>(G) </bold>and immunostaining <bold>(H) </bold>micrographs of L6 rat myoblasts (desmin-positive control).</p></caption><media xlink:href="1471-2261-6-25-S2.pdf" mimetype="application" mime-subtype="pdf"><caption><p>Click here for file</p></caption></media></supplementary-material></sec>
Behaviour change in perinatal care practices among rural women exposed to a women's group intervention in Nepal [ISRCTN31137309]	<sec><title>Background</title><p>A randomised controlled trial of participatory women's groups in rural Nepal previously showed reductions in maternal and newborn mortality. In addition to the outcome data we also collected previously unreported information from the subgroup of women who had been pregnant prior to study commencement and conceived during the trial period. To determine the mechanisms via which the intervention worked we here examine the changes in perinatal care of these women. In particular we use the information to study factors affecting positive behaviour change in pregnancy, childbirth and newborn care.</p></sec><sec sec-type="methods"><title>Methods</title><p>Women's groups focusing on perinatal care were introduced into 12 of 24 study clusters (average cluster population 7000). A total of 5400 women of reproductive age enrolled in the trial had previously been pregnant and conceived during the trial period.</p><p>For each of four outcomes (attendance at antenatal care; use of a boiled blade to cut the cord; appropriate dressing of the cord; not discarding colostrum) each of these women was classified as BETTER, GOOD, BAD or WORSE to describe whether and how she changed her pre-trial practice. Multilevel multinomial models were used to identify women most responsive to intervention.</p></sec><sec><title>Results</title><p>Among those not initially following good practice, women in intervention areas were significantly more likely to do so later for all four outcomes (OR 1.92 to 3.13). Within intervention clusters, women who attended groups were more likely to show a positive change than non-group members with regard to antenatal care utilisation and not discarding colostrum, but non-group members also benefited.</p></sec><sec><title>Conclusion</title><p>Women's groups promoted significant behaviour change for perinatal care amongst women not previously following good practice. Positive changes attributable to intervention were not restricted to specific demographic subgroups.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Wade</surname><given-names>Angie</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Osrin</surname><given-names>David</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Shrestha</surname><given-names>Bhim Prasad</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Sen</surname><given-names>Aman</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Morrison</surname><given-names>Joanna</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A6" contrib-type="author"><name><surname>Tumbahangphe</surname><given-names>Kirti Man</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A7" contrib-type="author"><name><surname>Manandhar</surname><given-names>Dharma S</given-names></name><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A8" contrib-type="author"><name><surname>de L Costello</surname><given-names>Anthony M</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib>	BMC Pregnancy and Childbirth	<sec><title>Background</title><p>Maternal and newborn mortality rates remain unacceptably high in the developing world. Most births and newborn deaths occur outside health facilities, so behaviour change in relation to home care practices and care-seeking behaviour is an essential component of any strategy to reduce deaths. We reported previously a cluster randomised controlled trial of the effects of participatory women's groups on neonatal outcomes in rural Nepal[<xref ref-type="bibr" rid="B1">1</xref>]. The trial intervention was a woman facilitator (who was not a trained health worker) within each area paid to instigate and guide women's groups focused on care in the perinatal period. The trial showed significant falls in neonatal (30%) and maternal mortality (78%), and appeared to be cost effective[<xref ref-type="bibr" rid="B2">2</xref>].</p><p>Married women of reproductive age (15–49 years) living in the study areas at the time of study inception were eligible. Before the trial started, each eligible woman was asked about her most recent pregnancy. If this resulted in a stillbirth, infant care practices were asked in respect of the most recent live birth. Information as to who was present at the birth and whether it took place in an institution was recorded. In particular it was ascertained whether there was a skilled attendant at the birth. The woman was asked whether she had attended antenatal care, which implement was used to cut the cord, what was applied to the cord after it was cut (the criterion for cleanliness was that either nothing or antiseptic was used) and whether or not she had discarded colostrum before starting to breastfeed, a practice distinct from discarding the foremilk at each feed. The evidence base for deciding which care practices are beneficial for good perinatal outcome is limited[<xref ref-type="bibr" rid="B3">3</xref>]. However, the practices recorded within the trial (antenatal care, skilled birth attendance, measures of cleanliness and good breastfeeding practice) have long been accepted as important[<xref ref-type="bibr" rid="B4">4</xref>-<xref ref-type="bibr" rid="B8">8</xref>].</p><p>After the baseline interview each woman became a member of the closed cohort who were randomised within village development committee areas (VDCs) and followed prospectively. In the original trial the efficacy of women's groups was measured for all women living within intervention areas (compared with control areas), even though many did not attend groups. The use of pre-trial pregnancy data allowed us to investigate the precise patterns of behaviour change within individual women and subgroups of women. Some women, in both arms of the trial, did not have the capacity for positive change attributable to intervention because they followed good practice in a pre-trial pregnancy. For women who did not follow good practice before the trial, our study gives us greater insight into factors affecting positive behaviour change, such as group membership, socioeconomic status, ethnicity and maternal age. The subset of women used for these analyses had by definition a pre-trial pregnancy and the results are not necessarily generalisable to women whose first pregnancy occurred in the trial.</p></sec><sec sec-type="methods"><title>Methods</title><p>Details of the original trial are reported elsewhere[<xref ref-type="bibr" rid="B9">9</xref>,<xref ref-type="bibr" rid="B10">10</xref>]. Briefly, 24 cluster units comprising village development committee areas (VDCs) – existing geopolitical units of population about 7000 – were placed into 12 matched pairings of similar topography, ethnicity and population densities. One VDC area of each matched pair was randomly assigned to receive the intervention and the other formed a control. All eligible women were identified and details of pregnancies, births and deaths were recorded prospectively for 33 months.</p><p>The analysis includes women who had reported a previous pregnancy and who had a subsequent pregnancy during the surveillance period. Twin pregnancies were included only once in the dataset as the process outcomes under consideration mostly related to the delivery or woman at that time rather than the individual child. Repeated pregnancies were included in the analysis with the appropriate clustering to account for within-woman correlation of outcomes and responses.</p><sec><title>Statistical analysis</title><p>The practices undertaken in each trial pregnancy were compared with those practices a woman had reported in her pre-trial pregnancy. Each practice was classified for each pregnancy as:</p><p>1) BETTER – lack of good practice in the preceding pregnancy followed by good practice in the trial pregnancy.</p><p>2) GOOD – good practice in both preceding and trial pregnancies.</p><p>3) BAD – lack of good practice in both preceding and trial pregnancies.</p><p>4) WORSE – good practice in the preceding pregnancy but not in the trial pregnancy.</p><p>We fitted multilevel multinomial models, taking into account the pairing of VDC area clusters, the clustering of women within VDC areas and households, and the correspondence between repeat prospective pregnancies in the same woman, to the 4-category outcomes. Multinomial models were preferred to logistic regressions of the trial practices corrected for pre-trial behaviour since they distinguished between changes from bad to good or from good to bad practice.</p><p>Multinomial models are extensions of logistic models. Associations between the outcomes and various features of the women are quantified and presented as coefficients for the ratios falling into the BETTER category relative to the other categories. This representation of the model results was chosen as being the easiest to interpret clinically. For all 3 ratios thus obtained, larger values were associated with more favourable outcome. All coefficients are presented with 95% confidence intervals adjusted for the clustered nature of the data. For each feature, separate coefficients are given to quantify the ratios:</p><p>1) BETTER relative to GOOD – quantifies the extent to which women following good practice in the trial pregnancy were doing so as a result of positive change (as opposed to continuing the good practices they had adopted pre-trial).</p><p>2) BETTER relative to BAD – quantifies the extent to which women who were following bad practice in the pre-trial pregnancy improved their practice within the trial. This coefficient is of particular interest as it describes the extent to which opportunities for positive change were taken.</p><p>3) BETTER relative to WORSE – quantifies the extent to which those women who changed practice made a positive, as opposed to negative, change.</p><p>A series of multilevel multinomial models were fitted to each of the process variables. Firstly, models were used to quantify differences in patterns of change between control and intervention clusters and the additional effect of attending a women's group for women within intervention clusters.</p><p>Secondly, a series of models were fitted to investigate whether the effect of intervention varied between women of differing ages, literacy levels and education, or between those living within households of differing ethnicity, assets or food sufficiency. For each of these demographic variables, a model which incorporated the demographic variable, a variable representing intervention status, and a term for the interaction between these two variables was used. The intervention and demographic variables were independently significantly associated with outcomes in all models. The coefficients for the fitted interaction terms showed which groups of women were most likely to respond to intervention and these are presented. Coefficients greater than 1 indicate that the women in the intervention clusters within that demographic subgroup had a more favourable distribution compared to the baseline category which was over and above any increase in favourable practices that could be attributed to intervention across all subgroups. Coefficients less than 1 are associated with a less favourable response for that demographic subgroup of women in the intervention compared with control clusters.</p><sec><title>Ethics and consent</title><p>The study was registered as an International Standard Randomised Controlled Trial, number ISRCTN31137309. It was approved by the Nepal Health Research Council and the ethical committee of the Institute of Child Health and Great Ormond Street Hospital for Children, and was conducted in collaboration with His Majesty's Government Ministry of Health, Nepal. The aims and design of the trial were discussed at both national and local meetings, after which consent to cluster involvement was given by chairpersons of VDC areas and the Makwanpur district development committee. Women who chose to participate in the study gave oral consent, were free to decline to be interviewed at any time, and the information they provided remained confidential.</p></sec></sec></sec><sec><title>Results</title><p>Of the women for whom information regarding a previous pregnancy had been recorded, 4929 had one further pregnancy during the trial surveillance period, 228 had 2 pregnancies and 5 had 3 pregnancies. Hence, there were a total of 5400 within-trial pregnancies from women with retrospectively recorded information.</p><p>Most women delivered at home (93%), without a trained attendant (92%) or any government health personnel present (90%), in either the preceding or study pregnancy. The sentinel care practices of antenatal care uptake, use of a clean blade to cut the umbilical cord, appropriate dressing of the cord and feeding of colostrum to the baby were more variably followed. Table <xref ref-type="table" rid="T1">1</xref> shows the demographic breakdown of the women and the extent to which good practice was being followed prior to commencement of the study.</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Practices in pre-trial pregnancies according to demographic variables</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="left" colspan="4">Number following good practice pre-trial (% of total)</td></tr></thead><tbody><tr><td></td><td align="left">Antenatal care attendance</td><td align="left">Boiling the blade</td><td align="left">Appropriate dressing of cord</td><td align="left">Not discarding colostrum</td></tr><tr><td></td><td align="left">n = 5373 (%)</td><td align="left">n = 5216 (%)</td><td align="left">N = 5216 (%)</td><td align="left">n = 5120 (%)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left"><bold>Household</bold></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Ethnicity:</td><td></td><td></td><td></td><td></td></tr><tr><td align="left"> Tamang</td><td align="left">757 (21.2)</td><td align="left">648 (18.6)</td><td align="left">2654 (76.3)</td><td align="left">1543 (45.3)</td></tr><tr><td align="left">Brahmin-Chhetri</td><td align="left">551 (67.0)</td><td align="left">512 (65.0)</td><td align="left">601 (76.3)</td><td align="left">535 (68.8)</td></tr><tr><td align="left"> Magar</td><td align="left">103 (40.9)</td><td align="left">83 (33.5)</td><td align="left">187 (75.4)</td><td align="left">151 (61.9)</td></tr><tr><td align="left"> Other</td><td align="left">212 (29.2)</td><td align="left">258 (36.8)</td><td align="left">526 (74.9)</td><td align="left">407 (59.0)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">No assets listed</td><td align="left">651 (22.1)</td><td align="left">634 (22.2)</td><td align="left">2175 (76.0)</td><td align="left">1333 (47.5)</td></tr><tr><td align="left">Clock, radio, iron, bicycle</td><td align="left">569 (31.9)</td><td align="left">517 (29.8)</td><td align="left">1305 (75.2)</td><td align="left">896 (52.6)</td></tr><tr><td align="left">More costly appliances</td><td align="left">403 (62.5)</td><td align="left">350 (56.6)</td><td align="left">488 (79.0)</td><td align="left">407 (66.8)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left"><bold>Mother</bold></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Illiterate</td><td align="left">704 (18.6)</td><td align="left">713 (19.3)</td><td align="left">2816 (76.3)</td><td align="left">1628 (45.0)</td></tr><tr><td align="left">Reads with difficulty</td><td align="left">320 (45.3)</td><td align="left">285 (41.5)</td><td align="left">522 (76.1)</td><td align="left">418 (62.1)</td></tr><tr><td align="left">Reads with ease</td><td align="left">599 (68.5)</td><td align="left">503 (60.0)</td><td align="left">630 (75.2)</td><td align="left">590 (71.3)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">No formal education</td><td align="left">949 (21.8)</td><td align="left">933 (22.1)</td><td align="left">3219 (76.2)</td><td align="left">1959 (47.3)</td></tr><tr><td align="left">Primary schooling only</td><td align="left">395 (57.2)</td><td align="left">325 (48.7)</td><td align="left">503 (75.3)</td><td align="left">424 (64.4)</td></tr><tr><td align="left">Secondary or higher</td><td align="left">279 (82.8)</td><td align="left">243 (75.7)</td><td align="left">246 (76.6)</td><td align="left">253 (79.3)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">TOTAL</td><td align="left">1623 (30.2)</td><td align="left">1501 (28.8)</td><td align="left">3968 (76.1)</td><td align="left">2636 (51.5)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td></td><td align="center" colspan="4">Median (Interquartile range) for those following good (G) and bad (B) practice retrospectively:</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left"><bold>Household</bold></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Number of months with</td><td align="left">G: 10 (7,12)</td><td align="left">G: 12 (7,12)</td><td align="left">G: 10 (7, 12)</td><td align="left">G: 10 (7, 12)</td></tr><tr><td align="left">sufficient food</td><td align="left">B: 9 (7,12)</td><td align="left">B: 9 (7,12)</td><td align="left">B: 9 (7, 12)</td><td align="left">B: 10 (7, 12)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left"><bold>Mother</bold></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Age (per additional year)</td><td align="left">G: 22.2<break/>(19.9, 26.2)</td><td align="left">G: 22.7<break/>(20.1, 27.0)</td><td align="left">G: 24.6<break/>(20.9, 29.8)</td><td align="left">G: 24.0<break/>(20.8, 28.8)</td></tr><tr><td></td><td align="left">B: 25.8<break/>(21.6, 31.2)</td><td align="left">B: 25.3<break/>(21.3, 30.9)</td><td align="left">B: 24.4<break/>(20.9, 29.9)</td><td align="left">B: 25.2<break/>(21.1, 30.9)</td></tr></tbody></table><table-wrap-foot><p>Note: Numbers are less than 5400 for each outcome since some women did not have pregnancies that progressed to the stage for that outcome to be appropriate. For example, 280 of the eligible pregnancies did not result in a live birth of a surviving mother and hence the discarding of colostrum was only appropriate as an outcome for 5120 women.</p></table-wrap-foot></table-wrap><p>Approximately three quarters of the women were appropriately dressing the cord initially and this proportion was fairly constant across demographic subgroups. Attendance at antenatal care, boiling of the blade and not discarding colostrum were all more prevalent amongst the more highly educated and literate women from wealthier households.</p><sec><title>The effect of being in an intervention VDC</title><p>Table <xref ref-type="table" rid="T2">2</xref> shows the percentages of pregnancy pairings falling into each of the 4 categories (BETTER, GOOD, BAD, WORSE) for the 4 outcomes for women in intervention and control arms of the trial.</p><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Behaviour change over time between pre-trial and trial pregnancies for four perinatal care practices.</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td></td><td align="left">Antenatal care attendance</td><td align="left">Boiling the blade</td><td align="left">Appropriate dressing of cord</td><td align="left">Not discarding colostrum</td></tr><tr><td></td><td align="left">Intervention</td><td align="left">n = 2535</td><td align="left">n = 2454</td><td align="left">n = 2454</td><td align="left">n = 2409</td></tr><tr><td></td><td align="left">Control</td><td align="left">n = 2838</td><td align="left">n = 2762</td><td align="left">n = 2762</td><td align="left">n = 2711</td></tr></thead><tbody><tr><td align="left">% BETTER</td><td align="left">Intervention</td><td align="left">19.1</td><td align="left">21.8</td><td align="left">17.9</td><td align="left">29.7</td></tr><tr><td></td><td align="left">Control</td><td align="left">16.6</td><td align="left">12.1</td><td align="left">16.3</td><td align="left">23.2</td></tr><tr><td colspan="6"><hr></hr></td></tr><tr><td align="left">% GOOD</td><td align="left">Intervention</td><td align="left">36.0</td><td align="left">32.5</td><td align="left">63.2</td><td align="left">41.7</td></tr><tr><td></td><td align="left">Control</td><td align="left">12.5</td><td align="left">12.8</td><td align="left">56.7</td><td align="left">34.7</td></tr><tr><td colspan="6"><hr></hr></td></tr><tr><td align="left">% BAD</td><td>Intervention</td><td align="left">36.8</td><td align="left">39.2</td><td align="left">4.2</td><td align="left">17.5</td></tr><tr><td></td><td align="left">Control</td><td align="left">65.6</td><td align="left">68.1</td><td align="left">9.3</td><td align="left">26.5</td></tr><tr><td colspan="6"><hr></hr></td></tr><tr><td align="left">% WORSE</td><td align="left">Intervention</td><td align="left">8.2</td><td align="left">6.4</td><td align="left">14.7</td><td align="left">11.2</td></tr><tr><td></td><td align="left">Control</td><td align="left">5.2</td><td align="left">7.0</td><td align="left">17.8</td><td align="left">15.5</td></tr><tr><td colspan="6"><hr></hr></td></tr><tr><td align="left">TOTAL (%)</td><td align="left">Intervention</td><td align="left">100</td><td align="left">100</td><td align="left">100</td><td align="left">100</td></tr><tr><td></td><td align="left">Control</td><td align="left">100</td><td align="left">100</td><td align="left">100</td><td align="left">100</td></tr><tr><td colspan="6"><hr></hr></td></tr><tr><td align="left">Intervention/control comparisons : Odds ratios (95% confidence interval)</td><td></td><td></td><td></td><td></td><td></td></tr><tr><td colspan="6"><hr></hr></td></tr><tr><td align="left">%BETTER/%BAD ratio </td><td></td><td align="left">2.04 (1.82, 2.27)</td><td align="left">3.13 (2.78, 3.45)</td><td align="left">2.44 (1.92, 3.13)</td><td align="left">1.92 (1.69, 2.22)</td></tr><tr><td align="left">%BETTER/%WORSE ratio</td><td></td><td align="left">0.73 (0.59, 0.91)</td><td align="left">1.96 (1.59, 2.44)</td><td align="left">1.33 (1.15, 1.54)</td><td align="left">1.79 (1.52, 2.08)</td></tr><tr><td align="left">%BETTER/%GOOD ratio</td><td></td><td align="left">0.40 (0.35, 0.46)</td><td align="left">0.71 (0.61, 0.81)</td><td align="left">0.99 (0.88, 1.10)</td><td align="left">1.06 (0.95, 1.19)</td></tr></tbody></table><table-wrap-foot><p>Results from multilevel multinomial models. The estimates and intervals are adjusted to take account of the correlations between pregnancies within the same women, women from the same household, households from the same VDC and VDCs within the same matched pair. All odds ratios are significantly different to 1. Coefficients greater than 1 indicate that the women in the intervention clusters had a more favourable distribution, those less than 1 are associated with a less favourable response for the women in the intervention compared with control clusters.</p></table-wrap-foot></table-wrap><p>The percentage of women who were following good practice during their trial pregnancies can be obtained by adding together the percentages falling into the BETTER and GOOD categories. For each of the 4 outcomes a greater percentage of the women in the intervention clusters followed good practice during the trial.</p><p>Combining the BETTER and BAD categories gives the percentage of women who were following bad practice pre-trial (and hence had the capacity to change for the better). For all outcomes apart from the discarding of colostrum, control clusters had more women with that capacity than intervention clusters. The percentages of women who recalled discarding colostrum in their preceding births were approximately equal between control and intervention clusters. The percentages lying within the BAD category represent missed opportunities for positive change and there were consistently fewer women within intervention clusters falling into this category for each of the 4 outcomes.</p><p>Women who changed their practice between preceding and study pregnancies fell into the BETTER and WORSE categories. The percentage of women in the intervention clusters falling into the BETTER category was greater than the percentage in the WORSE category, showing that women were more likely to make a positive, as opposed to detrimental, change for all outcomes. Women in the control clusters were more likely to stop, as opposed to start, appropriate dressing of the cord, but otherwise their changes were similarly more likely to be in a positive direction.</p><p>The differences between women in the intervention and control VDCs are further quantified by the fitting of multinomial models to the 4 outcomes with intervention status as a predictor. The coefficients and confidence intervals are given for the BETTER/BAD and BETTER/WORSE ratios. These are all significantly different to 1. For all four practices women who were initially following bad practice were significantly more likely to change to good practice if they lived in an intervention VDC (BETTER/BAD ratios). For example, women who did not attend antenatal care in preceding pregnancies were more than twice as likely to do so during the study period if they lived in an intervention area (odds ratio 2.04 95% ci (1.82, 2.27 times)). Of the women who changed practice these changes were significantly more likely to be in a positive direction for all outcomes except antenatal care attendance (BETTER/WORSE ratio).</p><p>Women attending antenatal care and/or using a boiled blade to cut the cord in pregnancies falling within the study period were significantly less likely to be doing so as a result of a positive change in practice if they lived in an intervention VDC (BETTER/GOOD ratios). These results are not unexpected given the larger percentages of women within the intervention VDCs following good practice for these outcomes pre-trial.</p></sec><sec><title>The independent effect of attending a women's group</title><p>About one in twelve married women of reproductive age, and about one third of newly pregnant women in intervention clusters attended the women's groups. There were few differences between the percentages of women who did and did not attend women's groups falling into each of the 4 categories.</p><p>The effect of attending a group over and above the improvements attributable to living within an intervention area was greatest for antenatal care attendance. The percentages of women who attended the groups falling into the BETTER, GOOD, BAD and WORSE categories were 22.3, 37.3, 33.7 and 6.7 respectively, compared to 17.8, 34.9, 38.9 and 9.1 of those within intervention VDCs who did not attend groups. Hence, a larger percentage of those attending the women's groups improved their practice (22.3 vs 17.8%) or maintained previous good practice (37.3 vs 34.9%). The significantly lower odds of making a positive as opposed to negative change (BETTER/WORSE ratio) in the intervention VDCs were counter-acted in the subgroup who attended the women's groups. The women who attended the groups were significantly more likely to make positive changes than non-attending women within intervention VDCs (BETTER/WORSE ratio 1.77 (1.30, 2.40)). Similarly, the women within intervention VDCs who attended the groups but did not attend antenatal care in their previous pregnancies were significantly more likely to start doing so than the women within those same VDCs who did not attend (BETTER/BAD ratio 1.51 (1.28, 1.79)). This difference was additional to the 2.04 fold increase seen in the intervention VDCs overall. The BETTER/GOOD ratio for attenders vs non-attenders was also significant (1.22 (1.04, 1.45)). Women attending the groups were significantly more likely to make positive changes compared to non-attending women in the same VDCs with respect to discarding colostrum (BETTER/WORSE ratio 1.03 (1.01, 1.06)) and there was some evidence that if they were discarding colostrum previously they were more likely to stop doing so (BETTER/BAD ratio 1.02 (1.00, 1.04)). There were no other significant differences.</p></sec><sec><title>Were specific subgroups of women with the capacity for positive change more likely to respond to intervention?</title><p>Table <xref ref-type="table" rid="T3">3</xref> shows the increase in the BETTER/BAD ratios for the intervention group compared to the women in control areas. Values greater than 1 indicate that the intervention was more successful in those subgroups of women relative to the baseline demographic category. Significant differences in the effects of intervention on the four process outcomes were not consistent across demographic subgroups.</p><table-wrap position="float" id="T3"><label>Table 3</label><caption><p>Coefficients and 95% confidence intervals for the extent to which women in the intervention VDCs, relative to women in the control VDCs, within different demographic subgroups were more (or less) likely to make a positive change, relative to those in the baseline subgroup, if they were not initially following good practice (%BETTER/%BAD ratio)</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="left">Antenatal care attendance N = 5373</td><td align="left">Boiling the blade prior to cord cutting n = 5216</td><td align="left">Appropriate dressing of the cord n = 5216</td><td align="left">Not discarding colostrum n = 5120</td></tr></thead><tbody><tr><td align="left"><bold>Household:</bold></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Ethnicity:</td><td></td><td></td><td></td><td></td></tr><tr><td align="left"> Tamang</td><td align="left">1</td><td align="left">1</td><td align="left">1</td><td align="left">1</td></tr><tr><td align="left">...Brahmin-Chhetri</td><td align="left">1.41 (0.98, 2.04)</td><td align="left"><bold>0.43 (0.28, 0.68)</bold></td><td align="left">1.11 (0.55, 2.22)</td><td align="left">1.16 (0.70, 1.89)</td></tr><tr><td align="left"> Magar</td><td align="left"><bold>4.17 (2.27, 7.69)</bold></td><td align="left">0.95 (0.54, 1.67)</td><td align="left">0.38 (0.12, 1.27)</td><td align="left"><bold>2.50 (1.01, 6.25)</bold></td></tr><tr><td align="left"> Other</td><td align="left"><bold>1.85 (1.23, 2.86)</bold></td><td align="left">0.74 (0.52, 1.06)</td><td align="left"><bold>8.33 (1.92, 33.33)</bold></td><td align="left"><bold>1.61 (1.04, 2.50)</bold></td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">No assets listed</td><td align="left">1</td><td align="left">1</td><td align="left">1</td><td align="left">1</td></tr><tr><td align="left">Clock, radio, iron, bicycle</td><td align="left">0.84 (0.65, 1.08)</td><td align="left">0.96 (0.75, 1.23)</td><td align="left">1.10 (0.65, 1.82)</td><td align="left">1.02 (0.76, 1.37)</td></tr><tr><td align="left">More costly appliances</td><td align="left">0.73 (0.48, 1.11)</td><td align="left">1.06 (0.70, 1.59)</td><td align="left">1.92 (0.85, 4.35)</td><td align="left">0.77 (0.44, 1.33)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">Number of months with sufficient food</td><td align="left">1.01 (0.97, 1.05)</td><td align="left">0.97 (0.93, 1.01)</td><td align="left"><bold>1.10 (1.01, 1.20)</bold></td><td align="left"><bold>1.05 (1.00, 1.10)</bold></td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left"><bold>Mother:</bold></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Age (per additional year)</td><td align="left">0.99 (0.97, 1.01)</td><td align="left">1.00 (0.98, 1.02)</td><td align="left"><bold>0.93 (0.91, 0.97)</bold></td><td align="left">1.00 (0.99, 1.02)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">Illiterate</td><td align="left">1</td><td align="left">1</td><td align="left">1</td><td align="left">1</td></tr><tr><td align="left">Reads with difficulty</td><td align="left"><bold>0.59 (0.41, 0.83)</bold></td><td align="left">1.08 (0.76, 1.54)</td><td align="left">1.09 (0.54, 2.17)</td><td align="left">0.67 (0.42, 1.04)</td></tr><tr><td align="left">Reads with ease</td><td align="left">0.73 (0.50, 1.08)</td><td align="left"><bold>0.68 (0.46, 1.00)</bold></td><td align="left">0.89 (0.44, 1.79)</td><td align="left">1.04 (0.64, 1.69)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">No formal education</td><td align="left"><bold>1</bold></td><td align="left">1</td><td align="left">1</td><td align="left">1</td></tr><tr><td align="left">Primary schooling only</td><td align="left"><bold>0.57 (0.40, 0.83)</bold></td><td align="left"><bold>0.42 (0.29, 0.61)</bold></td><td align="left"><bold>1.47 (1.08, 2.04)</bold></td><td align="left">1.28 (0.81, 2.04)</td></tr><tr><td align="left">Secondary or higher</td><td align="left"><bold>0.25 (0.11, 0.58)</bold></td><td align="left">0.53 (0.21, 1.35)</td><td align="left">0.92 (0.57, 1.47)</td><td align="left">0.50 (0.13, 1.92)</td></tr></tbody></table><table-wrap-foot><p>(Results from multilevel multinomial models. The estimates and intervals are adjusted to take account of the correlations between pregnancies within the same women, women from the same household, households from the same VDC and VDCs within the same matched pair. Significant differences are shown in <bold>bold</bold>.)</p></table-wrap-foot></table-wrap></sec><sec><title>Were women who changed practice more likely to do so positively if they were from specific subgroups?</title><p>The extent to which women made positive, as opposed to negative, changes in practice is quantified by the BETTER/WORSE model coefficients. Patterns were not consistent (Table <xref ref-type="table" rid="T4">4</xref>) but they were based on the smallest groups (Table <xref ref-type="table" rid="T2">2</xref>). Women from households with more assets within intervention VDCs were significantly more likely to make a positive change to dressing the cord but a negative change with respect to the treatment of colostrum. It was the older women, those who were less literate and the less well educated who were significantly more likely to have stopped, as opposed to started, discarding colostrum if they lived in intervention, as opposed to control, VDCs.</p><table-wrap position="float" id="T4"><label>Table 4</label><caption><p>Coefficients and 95% confidence intervals for the extent to which women in the intervention VDCs, relative to women in the control VDCs, within different demographic subgroups were more (or less) likely to make positive as opposed to negative changes, relative to those in the baseline subgroup, (%BETTER/%WORSE ratio)</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="left">Ante-natal care attendance n = 5373</td><td align="left">Boiling the blade prior to cord cutting n = 5216</td><td align="left">Appropriate dressing of the cord n = 5216</td><td align="left">Not discarding colostrum n = 5120</td></tr></thead><tbody><tr><td align="left"><bold>Household:</bold></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Ethnicity:</td><td></td><td></td><td></td><td></td></tr><tr><td align="left"> Tamang</td><td align="left">1</td><td align="left">1</td><td align="left">1</td><td align="left">1</td></tr><tr><td align="left">...Brahmin-Chhetri</td><td align="left"><bold>0.58 (0.39, 0.88)</bold></td><td align="left">0.76 (0.45, 1.28)</td><td align="left"><bold>1.85 (1.14, 2.94)</bold></td><td align="left">0.79 (0.47, 1.35)</td></tr><tr><td align="left"> Magar</td><td align="left">0.67 (0.39, 1.15)</td><td align="left">1.30 (0.37, 4.55)</td><td align="left">0.65 (0.32, 1.30)</td><td align="left"><bold>3.57 (1.32, 10.00)</bold></td></tr><tr><td align="left"> Other</td><td align="left">0.83 (0.59, 1.16)</td><td align="left">0.59 (0.29, 1.19)</td><td align="left"><bold>2.38 (1.52, 3.85)</bold></td><td align="left"><bold>2.27 (1.33, 3.85)</bold></td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">No assets listed</td><td align="left">1</td><td align="left">1</td><td align="left">1</td><td align="left">1</td></tr><tr><td align="left">Clock, radio, iron, bicycle</td><td align="left">0.92 (0.57, 1.47)</td><td align="left">1.15 (0.71, 1.85)</td><td align="left">1.37 (0.98, 1.89)</td><td align="left">0.99 (0.68, 1.43)</td></tr><tr><td align="left">More costly appliances</td><td align="left">1.92 (0.97, 3.85)</td><td align="left">0.89 (0.44, 1.82)</td><td align="left"><bold>2.33 (1.45, 3.85)</bold></td><td align="left"><bold>0.36 (0.18, 0.70)</bold></td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">Number of months with sufficient food</td><td align="left">0.97 (0.89, 1.05)</td><td align="left">0.98 (0.89, 1.06)</td><td align="left">1.02 (0.96, 1.08)</td><td align="left">0.99 (0.93, 1.05)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left"><bold>Mother:</bold></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Age (per additional year)</td><td align="left"><bold>0.96 (0.93, 1.00)</bold></td><td align="left">1.01 (0.97, 1.04)</td><td align="left">0.98 (0.96, 1.01)</td><td align="left"><bold>1.04 (1.01, 1.06)</bold></td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">Illiterate</td><td align="left">1</td><td align="left">1</td><td align="left">1</td><td align="left">1</td></tr><tr><td align="left">Reads with difficulty</td><td align="left">0.91 (0.51, 1.64)</td><td align="left">1.20 (0.66, 2.22)</td><td align="left">1.03 (0.62, 1.72)</td><td align="left"><bold>0.31 (0.18, 0.52)</bold></td></tr><tr><td align="left">Reads with ease</td><td align="left">1.54 (0.89, 2.63)</td><td align="left">1.19 (0.69, 2.04)</td><td align="left">1.47 (0.97, 2.22)</td><td align="left">0.61 (0.36, 1.05)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">No formal education</td><td align="left">1</td><td align="left">1</td><td align="left">1</td><td></td></tr><tr><td align="left">Primary schooling only</td><td align="left">1.08 (0.60, 1.96)</td><td align="left">1.03 (0.56, 1.92)</td><td align="left">1.11 (0.84, 1.47)</td><td align="left"><bold>0.49 (0.27, 0.87)</bold></td></tr><tr><td align="left">Secondary or higher</td><td align="left">2.04 (0.93, 4.55)</td><td align="left">0.99 (0.46, 2.17)</td><td align="left">0.97 (0.95, 1.00)</td><td align="left">0.79 (0.34, 1.85)</td></tr></tbody></table><table-wrap-foot><p>(Results from multilevel multinomial models. The estimates and intervals are adjusted to take account of the correlations between pregnancies within the same women, women from the same household, households from the same VDC and VDCs within the same matched pair. Significant differences are shown in <bold>bold</bold>.)</p></table-wrap-foot></table-wrap></sec><sec><title>Were women from specific subgroups who followed good practice during the trial more likely to be doing so as a result of a positive change?</title><p>These differences are quantified in Table <xref ref-type="table" rid="T5">5</xref> (BETTER/GOOD ratios). This table is presented for completeness. However, it is of the least clinical interest due to the dependence on the variability between groups of the percentages who show no changes but continue good practice throughout.</p><table-wrap position="float" id="T5"><label>Table 5</label><caption><p>Coefficients and 95% confidence intervals for the extent to which women in the intervention VDCs, relative to women in the control VDCs, within different demographic subgroups who were following good practice during the study period were more (or less) likely to be doing so as a result of a positive change in practice (%BETTER/%GOOD ratio)</p></caption><table frame="hsides" rules="groups"><thead><tr><td></td><td align="left">Ante-natal care attendance n = 5373</td><td align="left">Boiling the blade prior to cord cutting n = 5216</td><td align="left">Appropriate dressing of the cord n = 5216</td><td align="left">Not discarding colostrums n = 5120</td></tr></thead><tbody><tr><td align="left"><bold>Household:</bold></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Ethnicity:</td><td></td><td></td><td></td><td></td></tr><tr><td align="left"> Tamang</td><td align="left">1</td><td align="left">1</td><td align="left">1</td><td align="left">1</td></tr><tr><td align="left">...Brahmin-Chhetri</td><td align="left">0.79 (0.44, 1.41)</td><td align="left"><bold>4.35 (2.94, 6.67)</bold></td><td align="left"><bold>1.52 (1.09, 2.08)</bold></td><td align="left"><bold>1.72 (1.25, 2.38)</bold></td></tr><tr><td align="left"> Magar</td><td align="left">0.71 (0.22, 2.22)</td><td align="left"><bold>5.26 (2.70, 10.0)</bold></td><td align="left"><bold>2.44 (1.37, 4.17)</bold></td><td align="left"><bold>3.33 (1.92, 5.88)</bold></td></tr><tr><td align="left"> Other</td><td align="left">1.49 (0.68, 3.33)</td><td align="left"><bold>2.63 (1.69, 4.00)</bold></td><td align="left">0.91 (0.65, 1.28)</td><td align="left"><bold>1.45 (1.02, 2.04)</bold></td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">No assets listed</td><td align="left">1</td><td align="left">1</td><td align="left">1</td><td align="left">1</td></tr><tr><td align="left">Clock, radio, iron, bicycle</td><td align="left">1.28 (0.93, 1.75)</td><td align="left"><bold>1.61 (1.16, 2.22)</bold></td><td align="left"><bold>1.28 (1.01, 1.64)</bold></td><td align="left">1.12 (0.88, 1.45)</td></tr><tr><td align="left">More costly appliances</td><td align="left">1.33 (0.89, 2.00)</td><td align="left"><bold>2.13 (1.43, 3.23)</bold></td><td align="left"><bold>2.94 (2.04, 4.35)</bold></td><td align="left">1.19 (0.82, 1.72)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">Number of months with sufficient food</td><td align="left"><bold>1.05 (1.00, 1.11)</bold></td><td align="left"><bold>1.11 (1.05, 1.19)</bold></td><td align="left"><bold>1.06 (1.02, 1.11)</bold></td><td align="left">0.95 (0.91, 1.00)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left"><bold>Mother:</bold></td><td></td><td></td><td></td><td></td></tr><tr><td align="left">Age (per additional year)</td><td align="left">1.01 (0.99, 1.04)</td><td align="left">1.00 (0.98, 1.03)</td><td align="left">0.99 (0.98, 1.01)</td><td align="left">1.01 (0.99, 1.03)</td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">Illiterate</td><td align="left">1</td><td align="left">1</td><td align="left">1</td><td align="left">1</td></tr><tr><td align="left">Reads with difficulty</td><td align="left">0.94 (0.63, 1.41)</td><td align="left">1.41 (0.94, 2.13)</td><td align="left"><bold>1.67 (1.18, 2.33)</bold></td><td align="left">0.99 (0.71, 1.39)</td></tr><tr><td align="left">Reads with ease</td><td align="left"><bold>1.56 (1.10, 2.22)</bold></td><td align="left"><bold>1.59 (1.12, 2.22)</bold></td><td align="left"><bold>1.56 (1.14, 2.13)</bold></td><td align="left"><bold>1.47 (1.06, 2.04)</bold></td></tr><tr><td colspan="5"><hr></hr></td></tr><tr><td align="left">No formal education</td><td align="left">1</td><td align="left">1</td><td align="left">1</td><td align="left">1</td></tr><tr><td align="left">Primary schooling only</td><td align="left">1.35 (0.94, 1.96)</td><td align="left">1.06 (0.74, 1.54)</td><td align="left">1.01 (0.87, 1.18)</td><td align="left">1.35 (0.96, 1.92)</td></tr><tr><td align="left">Secondary or higher</td><td align="left">1.54 (0.91, 2.56)</td><td align="left"><bold>1.89 (1.14, 3.13)</bold></td><td align="left">1.15 (0.93, 1.41)</td><td align="left">1.59 (0.93, 2.70)</td></tr></tbody></table><table-wrap-foot><p>(Results from multilevel multinomial models. The estimates and intervals are adjusted to take account of the correlations between pregnancies within the same women, women from the same household, households from the same VDC and VDCs within the same matched pair. Significant differences are shown in <bold>bold</bold>.)</p></table-wrap-foot></table-wrap></sec></sec><sec><title>Discussion</title><p>Within a large scale trial of a community group intervention, women were followed prospectively to document patterns of behaviour change for perinatal care. This helps to understand how primary trial outcomes may be explained by changes in the practices of individuals within the communities. Of the 6380 women who became pregnant and were included in the main trial analyses, a subset of 5162 (77.3%) had a pregnancy pre-trial with which to compare their trial pregnancy behaviour. Within this subgroup we have investigated the changes for women undergoing their second or subsequent pregnancies. The findings cannot be extrapolated to women in their first pregnancies.</p><p>As expected, there were strong relationships between past and present behaviour. Those who followed good practice in previous pregnancies were likely to do so again, regardless of whether they were allocated to the intervention or control arm of the trial. Having a skilled birth attendant is known to be an important indicator of outcome. Less than 1 in 12 of the women had such a person present at either their pre-trial or any trial pregnancy. The numbers therefore were too low to investigate any impact the intervention may have had on improving skilled birth attendance. However, it was possible to investigate changes in other factors known to be important: antenatal care, cleanliness of blade and cord, and discarding of colostrum. The intervention effectively promoted significant change in all four care behaviours amongst the group of women not previously following good practice (Table <xref ref-type="table" rid="T2">2</xref>). Positive changes in antenatal care attendance and the discarding of colostrum were more likely to be made by women who attended the groups, but behaviour change in hygienic cutting and dressing is observed generally in the intervention areas. The lack of uniform relationship between group attendance and outcome was expected. The presence of groups in an area has a wider impact than merely on the women who attend. In our study only 8% of married women of reproductive age joined our groups, but 37% of newly pregnant women attended at least once. Whilst group members showed a greater tendency to positive behaviour change than non-group members, this effect is unlikely to explain the overall improved behaviour change in intervention versus control clusters. Our data provides evidence that the activities and existence of the group stimulate wider behaviour change in their communities. The group intervention is a dynamic process that is uniform only in its participatory method, thus further study is necessary to explore these processes of behaviour change. We hoped to bring about behaviour change by giving women and grandmothers the knowledge they need to make informed choices, and by creating favourable social conditions, and an enabling environment in which they could take these decisions[<xref ref-type="bibr" rid="B11">11</xref>-<xref ref-type="bibr" rid="B13">13</xref>]. Preliminary analysis of qualitative data, and the data presented here suggest that this has been the case in the intervention areas.</p><p>Most of the responses to intervention were positive. Significantly greater percentages of women in intervention VDCs who were following bad practice pre-trial stopped doing so after the commencement of the women's groups. It was surprising that significantly more of the intervention area women stopped as opposed to started attending antenatal care compared to the women within control VDCs. This difference was mostly attributable to the greater proportion of women in intervention VDCs who stopped attending. For the other 3 practices a greater percentage of control women stopped previous good practice and for all 4 practices there was a lower percentage who started. It is possible that the women in the intervention VDCs saw women's groups as a replacement for antenatal care. This potentially detrimental effect of the intervention requires further investigation, perhaps via the use of focus groups in similar future initiatives.</p><p>Since allocation was random, we would not expect baseline differences between women in control and intervention VDCs. Despite similar mortality rates at baseline[<xref ref-type="bibr" rid="B1">1</xref>], some differences in practices were found in the subgroup with a previous pregnancy reported here. In particular, women within intervention VDCs were more likely to have attended antenatal care (44.2% intervention, 17.7% control) and to have boiled the blade (38.9% and 19.8% respectively) in their pre-trial pregnancies. The analyses presented in this paper show that multigravid women in intervention VDCs were significantly more likely to continue or begin good practices after accounting for baseline differences. These significant differences in behaviour within this subgroup of just over three-quarters of the women who fell pregnant within the trial period are compatible with the reductions in major outcomes found within the trial.</p><p>We have presented secondary analyses of the dataset. Many comparisons are presented and these are meant to be interpreted in unison and with the main outcome analyses. The study was not originally designed to detect subgroup differences and the results should not be interpreted as though they were primary objectives. What we have aimed to do is to identify patterns that might be clinically relevant and informative to future studies. We have not identified any major consistent patterns, a finding which is itself of interest. Having identified significant differences which could be attributed to the intervention[<xref ref-type="bibr" rid="B1">1</xref>], this analysis investigates the modes via which those differences may have been achieved. We would expect to observe differences in process outcomes since these are known to be related to mortality outcomes. The finding that the intervention was associated with increased uptake of good practices in those previously not following them is both important and as expected: in this paper we attempt to quantify the degree of difference. It is also important to note that no tendency for the intervention to target only subgroups of privileged or non-privileged women was found. Prior to performing these analyses we had no notion of the direction that any intervention bias might fall.</p><p>If women were already following good practice, the capacity for the women's groups to effect positive change was limited. It was important that those following good practices continued to do so. Therefore, we have considered all patterns of change and how they related to features of the mother, the household in which she lived and whether or not she resided in an intervention area. Four pre-trial practices were found to have a large capacity for positive change and for there to have been significant alterations during the study period. The women's groups discussed the prevention of neonatal deaths, home care practices that might help, and the use of health services for either routine or emergency care. The issues of antenatal care, the use of clean cord-cutting implements, avoidance of unhygienic dressings and the benefits of colostrum feeding arose as subjects of discussion on many occasions. These issues could, and were, easily translated into specific actions.</p><p>It was clear that the less educated and illiterate women were less likely to be following good practice initially. Although these women were significantly targeted by the intervention for some outcomes, the differences were not uniform. There were benefits across all of the demographic subgroups of women.</p></sec><sec><title>Conclusion</title><p>In conclusion, peer-education and empowerment of women through women's groups has positive effects on perinatal care practices for women in their second or subsequent pregnancies. Both group members and non-group members in the locality benefit from this intervention.</p></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Authors' contributions</title><p>AW drafted the initial paper, helped design the original study, devised the analysis plan, conducted the analyses and provided an initial interpretation. DO, BPS, AS, JM, KMT, DSM and AMC devised and designed the study, assisted in the interpretation of the data and commented on multiple drafts of the paper. All authors read and approved the final submission.</p></sec><sec><title>Funding detail</title><p>The study was funded by the Department for International Development of the United Kingdom, with important support from the Division of Child and Adolescent Health, World Health Organization, Geneva, the United Nations Children's Fund (UNICEF), Nepal, and the United Nations Fund for Population Activities (UNFPA), Nepal. The Department for International Development can accept no responsibility for any information provided or views expressed.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1471-2393/6/20/prepub"/></p></sec>
Human rights, health and the state in Bangladesh	<sec><title>Background</title><p>This paper broadly discusses the role of the State of Bangladesh in the context of the health system and human rights. The interrelation between human rights, health and development are well documented. The recognition of health as a fundamental right by WHO and subsequent approval of health as an instrument of welfare by the Universal Declaration of Human Rights (UDHR) and the International Covenant on Social, Economic and Cultural Rights (ICSECR) further enhances the idea. Moreover, human rights are also recognized as an expedient of human development. The state is entrusted to realize the rights enunciated in the ICSECR.</p></sec><sec><title>Discussion</title><p>In exploring the relationship of the human rights and health situation in Bangladesh, it is argued, in this paper, that the constitution and major policy documents of the Bangladesh government have recognized the health rights and development. Bangladesh has ratified most of the international treaties and covenants including ICCPR, ICESCR; and a signatory of international declarations including Alma-Ata, ICPD, Beijing declarations, and Millennium Development Goals. However the implementation of government policies and plans in the development of health institutions, human resources, accessibility and availability, resource distribution, rural-urban disparity, the male-female gap has put the health system in a dismal state. Neither the right to health nor the right to development has been established in the development of health system or in providing health care.</p></sec><sec><title>Summary</title><p>The development and service pattern of the health system have negative correlation with human rights and contributed to the underdevelopment of Bangladesh. The government should take comprehensive approach in prioritizing the health rights of the citizens and progressive realization of these rights.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Rahman</surname><given-names>Redwanur M</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib>	BMC International Health and Human Rights	<sec><title>Background</title><p>This paper broadly discusses the role of State in the context of human rights and the health system of Bangladesh. Section I conceptualizes the interrelation between human rights, health and development. Section II analyses the health system development in Bangladesh. Section III discusses how the health system development has not succeeded in progressive realization of health aspect of human rights and further contribution to under-development of Bangladesh. It also gives directions on what should be the government priority to uplift health rights in the context of Bangladesh.</p><p>The constitution of World Health Organization (WHO) focused upon relationships between health and human rights. It stated that "the enjoyment of the highest attainable standard of health is one of the fundamental rights of every human being without distinction of race, religion, political belief, social and economic condition" [<xref ref-type="bibr" rid="B1">1</xref>]. The Declaration of Alma-Ata [<xref ref-type="bibr" rid="B2">2</xref>] of "health for all" in 1978 and the Ottawa Charter for Health Promotion [<xref ref-type="bibr" rid="B3">3</xref>] in 1986 further embraced the need for social and economic inputs to improve the health of the population. The Universal Declaration of Human Rights (UDHR) of 1948 and the International Covenant on Economic, Social and Cultural Rights (ICESCR) in 1966 further enunciate the appropriateness of health and human rights for the well being of individuals and the family [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B5">5</xref>]. So there is profound affiliation between human rights and health.</p><sec><title>From civil & political rights to social, economic rights and development</title><p>The human rights have recognized not only the civil and political rights but also the social, economic and cultural rights by giving importance to the latter through articulating and prioritizing rights to health, education, housing, and employment. Moreover, the fundamental tenet of the human rights is that every individual's dignity should be protected being a human. This dignity merely means not only political liberty but also a guarantee of economic subsistence, cultural freedom and the provision of social services [<xref ref-type="bibr" rid="B6">6</xref>,<xref ref-type="bibr" rid="B7">7</xref>]. In this context, the ICESCR deals with the State's obligation to create affirmative conditions to facilitate human well-being. The entitlements of these services are clearly mentioned in UDHR Articles 21 and 25; and ICESCR articles 2, 6, 9, 11, 12, 13, 14. Article 25 of UDHR indicates that "every one has the right to a standard of living adequate for the health and well-being of himself and of his family, including food, clothing, housing, medical care and necessary social services, and the right to security in the event of unemployment, sickness, disability, widowhood, old age or other lack of livelihood in circumstances beyond his control" [[<xref ref-type="bibr" rid="B4">4</xref>], UDHR Article 25]. At the same time, ICESCR further defines these rights by obligating the state to undertake steps to maximize available resources with a view to achieving progressive realization of these rights [[<xref ref-type="bibr" rid="B5">5</xref>], ICSER Article 2]. Moreover, the state should not only give importance towards economic, social and cultural rights but the same amount of attention and importance also should be given to civil and political rights. In addition, the United Nations declaration on right to development in 1986 further advanced the 'idea' that rights should be renamed as an instrument of development. It is argued that civil and political rights should be fought for on the socio-economic and cultural fronts, and this has been conceptualized as the 'right to development' [<xref ref-type="bibr" rid="B8">8</xref>-<xref ref-type="bibr" rid="B11">11</xref>]. Since then, 'rights' have been focused on as an agent of development that has been affirmed and reaffirmed in different fora. The World Conference on Human rights in 1993 describes right to development as universal and inalienable [<xref ref-type="bibr" rid="B12">12</xref>]. So it can be said that human rights means all aspects of rights which is further linked to development.</p></sec><sec><title>State and social rights</title><p>The ICSCER gives obligation to the state authority to ensure social rights. In a particular socio-political, historical, cultural and economic environment; society, social structure, political process and the power relations try to alleviate human miseries. Moreover, the state structure can facilitate and guarantee the social human rights to every individual in accessing to essential levels of social services.</p><p>There is a common understanding that it is the responsibility of the State to facilitate social rights. Though UDHR did not create any legal binding or obligation on state parties, state parties viewed it as obligation to maintain basic and minimum international human rights standards. In 1966, International Covenants on Civil and Political Rights (ICCPR) and ICESCR imposed binding obligations on state parties. The ICESCR emphasizes that state parties require positive steps towards progressive achievement of the full realization of rights, which is incorporated in the covenant. It further imposes that state parties have an obligation to ensure the satisfaction of at least the minimum level of each right [<xref ref-type="bibr" rid="B13">13</xref>]. The <italic>Limburg Principles </italic>adopted in 1986 clarified that social rights could be guaranteed and implemented in different socio-political and economic settings [<xref ref-type="bibr" rid="B14">14</xref>]. The <italic>Principles </italic>favour ensuring and respecting for minimum subsistence rights for all, regardless of economic level development. The state should utilize its legal, administrative, economic, social, educational and related means to materialize the obligations of the covenant. So the state is entrusted to facilitate social rights to every citizen of a country. Hence the responsibility lies with the state to ensure and to guarantee individual's access to requisite resources to live in a dignified way. These requisite resources may include economic, social, cultural, civil and political rights. The guarantee of these social goods is necessary preconditions of the enjoyment of all human rights and allows individuals to participate fully in all other areas of their lives [<xref ref-type="bibr" rid="B7">7</xref>]. The intervention of the state is required for mobilizing of resources and expenditure for the fulfillment of social rights [<xref ref-type="bibr" rid="B7">7</xref>,<xref ref-type="bibr" rid="B15">15</xref>]. Moreover, the implementation of social rights also depends on state's active involvement and participation in policy-making, policy implementation, budgetary allocation and priorities to ensure that every individual will be able to receive his or her entitlements. So every state should take appropriate action to initiate the 'right to development' in every aspect to ensure equality of opportunity for all in their access to basic resources such as health, education, food, housing, employment, and fair distribution of income, thus recognizing social rights.</p></sec><sec><title>State and health rights</title><p>The role of state in the provision of health rights further ensures states' obligation to provide minimum care to every individual. The right to health imposes three obligations on states, i) to respect ii) to protect and iii) to fulfil and to promote the enjoyment of the right to health [<xref ref-type="bibr" rid="B16">16</xref>]. The state should respect the health rights of every individual through protecting them from illness and diseases, facilitating and providing minimum basic services and health promotion. If there is a natural calamity, or sudden out-break of diseases in part or in the whole country, the government should take necessary measures to protect the citizens. The state should maintain minimum basic conditions of the institutional facilities to provide health services to every individual irrespective of caste, class, creed, religion, and geographical location. Providing basic services also respects the citizens' right to health care. The state should also need to promote health rights to individuals. The promotion of health indicates not only health services but also providing necessary services to ensure safe food, hygienic shelter, potable water, sanitation and drugs. In addition, health promotion also includes legislative, financial, societal, and organizational change to promote healthy life styles for the well-being of the citizens [<xref ref-type="bibr" rid="B3">3</xref>]. So the state is responsible for health promotional activities.</p><p>The state needs to facilitate the availability, accessibility, and to maintain acceptability, quality and standard in the provision of health care. The presence of basic facilities for health and health services are drinking water, sanitation, hospitals, clinics, trained health personnel, and essential drugs. Accessible denotes such a health service that is easily accessible without any barrier. The barrier includes financial, geographical, religious, class and caste. Maintaining standards requires acceptable criteria and scientific and medically tested procedures to ensure quality [<xref ref-type="bibr" rid="B16">16</xref>]. The state needs to maintain standard of quality of care in the provision of health services.</p></sec><sec><title>State and development</title><p>The declaration of the 'right to development' by the United Nations in 1986 gives new impetus to the state to play a new role in advancing development with a new vision. In pursuing development goals, the state should take appropriate action to initiate the 'right to development' in every aspect and must ensure equal opportunity for all. It should not make barriers to the access to basic resources such as health, education, food, housing, employment, and fair distribution of income. It is the State's responsibility to facilitate individual access to requisite resources to live in a dignified way. These requisite resources may include economic, social, cultural, civil and political rights. However, it is observed that in many cases, the state's nature does not allow for enjoyment of basic human rights by providing and ensuring availability and accessibility to health care, sufficient food, basic education, employment and adequate livelihood [<xref ref-type="bibr" rid="B17">17</xref>-<xref ref-type="bibr" rid="B19">19</xref>]. All states should explore viable options to fulfill the goal of human rights, which lead to development. Moreover, the state's role in alleviating and mitigating social disparity, misery and deprivation may bring social change towards development. The State's provision of guarantees for social rights should help the poorer, weaker and disadvantaged groups to get their due share in society, which also leads to 'development'.</p></sec><sec><title>Human rights, health, and development: their interrelationship</title><p>An understanding about the social right to health focused upon a new dimension about the relationship between human rights and health. It gives an idea to maximize the benefit of social good irrespective of social strata, class, race, sex, and religion. Health is universally recognized as an important aspect of human development. Without the development of health, overall development is not possible. On the other hand, human rights are also a part of development. So the interface between human rights and health is towards the development of human welfare (Figure <xref ref-type="fig" rid="F1">1</xref>). We have mentioned how the declarations of UDHR and ICESCR have integrated health as a part of human rights. We also mentioned that the constitution of World Health Organization (WHO) also recognizes that the highest attainable standard of health is a fundamental right of every human being. Article 12 of ICESCR delineates specific goals to attain better health provision which includes 'right to the highest attainable standard of physical and mental health', development of child health, improvement of environmental and industrial hygiene, prevention and treatment of disease, reduction of infant mortality rate, and the creation of health facilities which are easily available and accessible for the sick [[<xref ref-type="bibr" rid="B5">5</xref>], ICSECR Article 12]. Moreover, WHO defined health as "a state of complete physical, mental and social well-being and not merely the absence of disease or infirmity" [<xref ref-type="bibr" rid="B1">1</xref>]. The Alma-Ata declaration of 1978 further enhanced health dimension as treating health as a "social goal whose realization requires the action of many other social and economic sectors in addition to the health sector" [<xref ref-type="bibr" rid="B2">2</xref>]. The WHO definition gives importance of health promotion as "the process of enabling people to increase control over and to improve their health". So the modern concept of health includes not only health care but also embraces the broader societal dimension to include population well being. So the vision of human rights and health is well documented to ensure human welfare [<xref ref-type="bibr" rid="B20">20</xref>-<xref ref-type="bibr" rid="B22">22</xref>]. The international declaration of health rights also proclaims "enjoyment of the highest attainable standard of health is one of the fundamental rights of every human being" and health "is not a privilege reserved for those with power, money and social standing" [<xref ref-type="bibr" rid="B23">23</xref>]. The General Comment 14 further states that "health is a fundamental human right indispensable for the exercise of other human rights. Every human being is entitled to the enjoyment of the highest attainable standard of health conducive to living a life in dignity" [<xref ref-type="bibr" rid="B13">13</xref>]. The inter-linking between human rights and health will not only help to improve the development of health status but human development in general and also embrace equity, solidarity and social justice [<xref ref-type="bibr" rid="B24">24</xref>]. Hence human rights and health are inextricably linked in advancing human welfare and development.</p><fig position="float" id="F1"><label>Figure 1</label><caption><p>The interrelation between health, human rights and development.</p></caption><graphic xlink:href="1472-698X-6-4-1"/></fig></sec></sec><sec><title>Discussion</title><sec><title>State, human rights and health system in Bangladesh</title><p>Bangladesh is located in the north-eastern part of South Asia. Bangladesh became independent from Pakistan in 1971 after a fierce civil war. Administratively, the country is divided into six divisions, 64 districts, 472 <italic>upazillas </italic>(sub-districts), 496 <italic>thanas </italic>(police stations) and 4,451 unions which serve as the basic unit of administration. According to the United Nations Children's Emergency Fund, the estimated population in Bangladesh was 140.36 million in 2001. The per capita gross national product was US$ 460 in 2004. Currently the literacy rate is 65 per cent while 49.8 per cent live below the poverty line and about 30 per cent live on US$ 1 a day. The per capita total expenditure on health was US$ 14 in 1999 [<xref ref-type="bibr" rid="B25">25</xref>]. According to a household expenditure survey of 1995–96, the top five per cent of the households in the urban areas have an income of US$ 641 and in the rural areas, it is US$ 240 [<xref ref-type="bibr" rid="B26">26</xref>]. Another top 20 per cent of households in the urban areas have very good income. The Bangladesh Bank, Central Bank of Bangladesh, disclosed that 45,000 people have deposits over US$ 1,61,290.32 each and 500 have over US$ 8,06,451.61[<xref ref-type="bibr" rid="B27">27</xref>]. In the Bangladesh Parliament, on the basis of professional categories of the parliamentarians, there were about 23 per cent representation from business and industrialists groups in 1973 but their representation has increased to 59 per cent in 1991 [<xref ref-type="bibr" rid="B28">28</xref>] and 74 per cent in 1996 and 81 per cent in 2001 [<xref ref-type="bibr" rid="B29">29</xref>]. It is revealed from Bangladesh political scenario that moneyed men are in centre of politics and they influence in decision-making process. Thus it indicates that political and economic powers are concentrated on a few people who control and maintain public policies to ensure their part in gaining resources and accessing various public sector facilities.</p></sec><sec><title>Structure of public health care sector</title><p>The Ministry of Health and Family Welfare (MOHFW) is the largest institutional health care provider in Bangladesh. Its services range from primary to more complex treatments; its structure is centralized. All decisions regarding the development of personnel and facilities, the allocation of resources and the formulation of policy are made at the central level by the MOHFW.</p><p>Specialized institutions operate at the national level. These institutions provide a wide range of services, including cardiology, cardiac surgery, dentistry, endrocrinology, medicine, nephrology, neuro-surgery, oncology, opthalmology, orthopaedics and psychiatry. They offer both inpatient and outpatient services. Ideally, these facilities provide 'follow-up' care for patients referred by various medical college hospitals and other hospitals. However, the referral system works on a very limited scale in Bangladesh as referral <italic>process </italic>has not been fully developed. Their bed capacity varies from 50 to 600, while there are 3,325 beds in national-level hospitals [<xref ref-type="bibr" rid="B30">30</xref>,<xref ref-type="bibr" rid="B31">31</xref>] (see Table <xref ref-type="table" rid="T1">1</xref>).</p><table-wrap position="float" id="T1"><label>Table 1</label><caption><p>Pattern of Health Care Growth in Bangladesh, 1973 and 2003</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left"><bold>Type of Facility</bold></td><td align="left"><bold>1973</bold></td><td align="left"><bold>1978</bold></td><td align="left"><bold>1983</bold></td><td align="left"><bold>1988</bold></td><td align="left"><bold>1993</bold></td><td align="left"><bold>1998</bold></td><td align="left"><bold>2003</bold></td></tr></thead><tbody><tr><td align="left">Total Hospitals</td><td align="left">308</td><td align="left">424</td><td align="left">724</td><td align="left">875</td><td align="left">903</td><td align="left">1,273</td><td align="left">1,464</td></tr><tr><td align="left">Government hospitals</td><td align="left">NA</td><td align="left">388</td><td align="left">560</td><td align="left">608</td><td align="left">611</td><td align="left">647</td><td align="left">654</td></tr><tr><td align="left">Private Hospitals</td><td align="left">NA</td><td align="left">36</td><td align="left">164</td><td align="left">267</td><td align="left">292</td><td align="left">626</td><td align="left">810</td></tr><tr><td align="left">District Hospitals</td><td align="left">13</td><td align="left">37</td><td align="left">43</td><td align="left">59</td><td align="left">57</td><td align="left">59</td><td align="left">59</td></tr><tr><td align="left"><italic>Upazilla </italic>Health Complexes</td><td align="left">160</td><td align="left">253</td><td align="left">319</td><td align="left">352</td><td align="left">372</td><td align="left">402</td><td align="left">417</td></tr><tr><td align="left" colspan="8"><bold>Number of Beds</bold></td></tr><tr><td align="left">Total Beds</td><td align="left">12,311</td><td align="left">19,538</td><td align="left">25,057</td><td align="left">33,334</td><td align="left">35,280</td><td align="left">41,514</td><td align="left">44,275</td></tr><tr><td align="left">Number of Beds in Public Sector</td><td align="left">10,449</td><td align="left">16,853</td><td align="left">20,286</td><td align="left">26,871</td><td align="left">27,637</td><td align="left">30,143</td><td align="left">32,615</td></tr><tr><td align="left">Number of Beds in Private Sector</td><td align="left">1,862</td><td align="left">2,685</td><td align="left">4,771</td><td align="left">6,463</td><td align="left">7,643</td><td align="left">11,371</td><td align="left">11,660</td></tr><tr><td align="left" colspan="8"><bold>Teaching Facility</bold></td></tr><tr><td align="left">Total Medical Colleges</td><td align="left">8</td><td align="left">8</td><td align="left">9</td><td align="left">9</td><td align="left">16</td><td align="left">19</td><td align="left">33</td></tr><tr><td align="left">Private Medical Colleges</td><td align="left">NA</td><td align="left">NA</td><td align="left">NA</td><td align="left">NA</td><td align="left">3</td><td align="left">6</td><td align="left">20</td></tr><tr><td align="left">Postgraduate Institutes</td><td align="left">1</td><td align="left">3</td><td align="left">6</td><td align="left">6</td><td align="left">6</td><td align="left">6</td><td align="left">6</td></tr><tr><td align="left" colspan="8"><bold>Personnel</bold></td></tr><tr><td align="left">Registered Doctors</td><td align="left">5,001</td><td align="left">7,035</td><td align="left">11,496</td><td align="left">18,030</td><td align="left">21,004</td><td align="left">29,613</td><td align="left">36,553</td></tr><tr><td align="left">Registered Nurses</td><td align="left">765</td><td align="left">2,011</td><td align="left">5,164</td><td align="left">7,390</td><td align="left">9,655</td><td align="left">16,104</td><td align="left">19,066</td></tr><tr><td align="left">Registered Midwives</td><td align="left">764</td><td align="left">1,041</td><td align="left">3,424</td><td align="left">6,556</td><td align="left">7,713</td><td align="left">14,312</td><td align="left">16,553</td></tr></tbody></table><table-wrap-foot><p>Note: NA indicates not available Source:</p><p>[30, 31, 41, 66–68]</p></table-wrap-foot></table-wrap><p>At the regional level, medical college hospitals provide a range of specialized laboratory facilities for the treatment of complicated cases. These hospitals are required to receive cases referred by the <italic>Upazilla </italic>Health Complexes (UHC) and district hospitals. There are 13 public medical college hospitals with bed capacities of 250 to 1,400, with a total bed capacity of 8,000 [<xref ref-type="bibr" rid="B30">30</xref>].</p><p>At the district level, there are 59 hospitals (as shown in Table <xref ref-type="table" rid="T1">1</xref>). These hospitals are expected to receive cases referred by UHCs. They provide specialist, laboratory and diagnostic services. District-level hospitals have bed capacities ranging from 50 to 250 beds, with a total bed capacity of 5,295 [<xref ref-type="bibr" rid="B30">30</xref>].</p><p>At the <italic>upazilla </italic>(sub-district) level, there are 417 UHCs (as shown in Table <xref ref-type="table" rid="T1">1</xref>), each with a minimum bed capacity of 31 to 50. <italic>Upazilla </italic>Health Complexes are designed to provide primary health care services and to function as referral institutions for Union Sub-Centres (USC) or Union Health and Family Welfare Centres (UHFWC). There are approximately 13,000 beds at the <italic>upazilla </italic>level [<xref ref-type="bibr" rid="B32">32</xref>].</p><p>There are a total of 4,400 USCs and UHFWCs. These are smallest and most peripheral health, family planning and mother and child health care units, providing only outpatient services for simple injuries and ailments. They have no surgical or bed facilities. According to the HPSP, the government planned to build one community clinics <italic>per </italic>6,000 people. However, the government has built 11,000 community clinics but they are not yet operating [<xref ref-type="bibr" rid="B33">33</xref>]. At the ward level (with a population of 6,000–7,000 and comprises two-three villages), health workers provide 'doorstep' services. A health worker visits each household once every four to eight weeks to provide domiciliary services.</p></sec><sec><title>Government policy position</title><p>Since the emergence of Bangladesh, the government has given utmost priority to ensure human rights and dignity of the population. Constitutional provisions have been made to protect, respect and to promote individual as well as collective rights in the society. Moreover, Bangladesh is a signatory to most of the international treaties, declarations and ratified covenants to ensure the 'right to development' as a means of promotion of human rights. It also ratified the ICCPR and ICESCR [<xref ref-type="bibr" rid="B34">34</xref>].</p><p>The country recognizes its obligation to protect and promote human rights. Civil and political rights are recognized in the constitution as fundamental rights. Article 11 of the Bangladesh constitution states "the Republic shall be a democracy in which fundamental human rights and freedoms and respect for the dignity and worth of the human person shall be guaranteed" [[<xref ref-type="bibr" rid="B35">35</xref>], Article 11]. Even social, cultural and economic rights are included. The constitution mandates that "it shall be a fundamental responsibility of the state to attain, through planned economic growth, a constant increase in productive forces and a steady improvement in the material and cultural standard of living of the people with a view to securing to its citizens (a) the provision of the basic necessities to life, including food, clothing, shelter, education and medical care" [[<xref ref-type="bibr" rid="B35">35</xref>], Article 15]. Article 16 of the constitution also mentions that the state shall adopt effective measures to reduce disparity in health care progressively. Article 18.1 also depicts that the state shall foster rising levels of nutrition and the improvement of public health measures and Article 19 gives importance towards reducing inequality. Bangladesh has given high priority to the development of social sector including health and education with high level of political support. The constitutional provision also guaranteed employment with reasonable wage, right to social security, and quality of life [[<xref ref-type="bibr" rid="B35">35</xref>], Article 15, 16, 18, & 19]. The constitutional provisions are made to protect, promote and respect health care as a constituent of human rights in Bangladesh.</p><p>The Government of Bangladesh (GOB) has invested substantially in building institution and strengthening of the health care system. It has accepted the goal and reiterated firm political and social commitment to achieve the Primary Health Care (PHC) strategy declared in Alma-Ata in 1978. Moreover, Bangladesh is a signatory of International Conference on Population and Development, Womens' Conference in Beijing and most recently the UN special session on Children's Rights and other important international declarations. Bangladesh signed United Nations Millennium Development Goals which emphasized improvement of maternal health, stemming the spread of HIV/AIDS, malaria and other diseases. In 1998, the Ministry of Health and Family Welfare formulated Health and Population Sector Strategy (HPSS) [<xref ref-type="bibr" rid="B36">36</xref>]. It intends to increase quality, equity, efficiency and integration of health and family planning services. As a part of the implementation process, the government has also formulated a Health and Population Sector Programme (HPSP) [<xref ref-type="bibr" rid="B37">37</xref>]. It prioritized an essential service package for primary care. Even for the first time, the government announced National Health Policy in 2000. The preamble of the policy document states that the provision of health for the people is the constitutional obligation of the state, [<xref ref-type="bibr" rid="B38">38</xref>]. The policy plan on Health, Nutrition, and Population Sector Programme (HNPSP) of 2004 also recognizes health care as basic rights of every citizen [<xref ref-type="bibr" rid="B39">39</xref>]. The HNPSP emphasized the necessity for the improvement of maternal and child health care, and facilitating essential service package. Recently prepared Poverty Reduction Strategy Paper (PRSP) gives importance on the enhancement of health [<xref ref-type="bibr" rid="B40">40</xref>]. All the plans, programmes and documents have amply demonstrated the government's commitment to making essential health care accessible to every individual and community.</p></sec><sec><title>Status of health indicators and institutions</title><p>The Government has given efforts to develop a network of health care systems from village level to national level to cater to the masses. In response, different types of hospitals and other ancillary services have emerged at different levels i.e union level to national level. There were 417 <italic>Upazilla </italic>Health Complexes (UHC), 4,400 Union Sub-Centers or Union Health and Family Welfare Centers (USC/UHFWC), and 1,464 hospitals [<xref ref-type="bibr" rid="B41">41</xref>] (see Table <xref ref-type="table" rid="T1">1</xref>). There are improvements of morbidity, mortality, fertility and life expectancy at birth (LEB) (see Table <xref ref-type="table" rid="T2">2</xref>). But these achievements are much lower than those of the neighbouring countries and far below the regional and global averages [<xref ref-type="bibr" rid="B9">9</xref>,<xref ref-type="bibr" rid="B42">42</xref>].</p><table-wrap position="float" id="T2"><label>Table 2</label><caption><p>Basic Health Indicators of Bangladesh in 1973 and 2001</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left"><bold>Indicator</bold></td><td align="left"><bold>1973</bold></td><td align="left"><bold>2001</bold></td></tr></thead><tbody><tr><td align="left">Infant Mortality Rate (IMR)</td><td align="left">140/1000</td><td align="left">51/1000</td></tr><tr><td align="left">Maternal Mortality Rate (MMR)</td><td align="left">30/1000</td><td align="left">3.5/1000</td></tr><tr><td align="left">Crude Birth Rate (CBR)</td><td align="left">47/1000</td><td align="left">19.9/1000</td></tr><tr><td align="left">Crude Death Rate (CDR)</td><td align="left">17/1000</td><td align="left">4.8/1000</td></tr><tr><td align="left">Life Expectancy at Birth (LEB)</td><td align="left">45 years</td><td align="left">62 years</td></tr><tr><td align="left">Doctor/population ratio</td><td align="left">1:6250</td><td align="left">1:4105</td></tr><tr><td align="left">Doctor Nurse ratio</td><td align="left">-</td><td align="left">2:1.7</td></tr><tr><td align="left">Bed population ratio</td><td align="left">-</td><td align="left">1:3154</td></tr><tr><td align="left">Immunization coverage under one year</td><td align="left">-</td><td align="left">80 per cent</td></tr><tr><td align="left">Proportion of one year old children immunized against measles</td><td align="left">-</td><td align="left">64 per cent</td></tr><tr><td align="left">Total population covered by essential health care</td><td align="left">-</td><td align="left">42 per cent</td></tr><tr><td align="left">Proportion of diarrhoea control</td><td align="left">-</td><td align="left">70 per cent</td></tr><tr><td align="left">Delivery assisted by a trained person</td><td align="left">-</td><td align="left">14 per cent</td></tr><tr><td align="left">Prevalence of Low Birth Weight</td><td align="left">-</td><td align="left">25 per cent</td></tr><tr><td align="left">Prevalence of Child Malnourishment</td><td align="left">-</td><td align="left">48 per cent</td></tr></tbody></table><table-wrap-foot><p>Sources: [43, 69, 70]</p></table-wrap-foot></table-wrap><p>The public policies of Bangladesh are directed towards development to alleviate people's misery and protect their rights. But the poor implementation of these policies has given little benefit to the poorer and disadvantaged groups. Different policy documents of the government of Bangladesh reveal that the government target was to achieve 50 per cent deliveries by a trained person by 1995, but the shocking history is that by 1997, it was only 14 per cent. The first five year plan aimed at establishing one <italic>Upazilla </italic>Health Complex in each <italic>upazilla </italic>and one Health Center in each union but, till today (2003), more than 55 <italic>upazilla a</italic>nd more than 200 unions do not have any health facilities. The government had planned to provide essential health care to 80 per cent of the population by 1995 but they had only provided about 45 per cent [<xref ref-type="bibr" rid="B43">43</xref>-<xref ref-type="bibr" rid="B46">46</xref>]. Moreover, the government has generally emphasized the expansion of the physical infrastructure, like the building of hospitals. Since independence, government has built 654 hospitals, and 4,400 USCs/UHFWCs, 72 dispensaries, 96 maternity clinics, and 32,615 beds. But most of these facilities lack laboratory facilities, required manpower, equipment and furniture. A survey of 16 UHCs, 12 Rural Dispensaries, and 100 USCs shows that 63 per cent had inadequate physical facilities, 60 per cent had inadequate personnel, and 80 per cent faced shortage of supplies or vaccines [<xref ref-type="bibr" rid="B47">47</xref>]. While the government counts the number of buildings constructed in assessing its performance in the sector, the creation of these facilities has not ensured services to the population irrespective of place and class i.e. rich and poor. The media reports a very dismal picture of public health facilities in Bangladesh.</p><p>This Table <xref ref-type="table" rid="T3">3</xref> shows that the privileged patients from the richest quintile are admitted for in-patient care five times more than the patients from the poorest quintile. The urban patients are more than twice advantaged over the rural patients and the male patients are more likely to get adequate and quality treatment than the female patients. The lowest 20 per cent receive only 16 per cent while the highest 20 per cent receive 26 per cent of all health expenditure [<xref ref-type="bibr" rid="B48">48</xref>]. It shows that people from upper echelon is more benefited than lower echelon.</p><table-wrap position="float" id="T3"><label>Table 3</label><caption><p>Use of Public Facilities by Level (in percentage)</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left"><bold>Income Group</bold></td><td align="left"><bold>Hospital Visits</bold></td><td align="left"><bold>UHC Visits</bold></td><td align="left"><bold>Union level facility visits</bold></td></tr></thead><tbody><tr><td align="left">Poorest quintile</td><td align="left">13</td><td align="left">23</td><td align="left">26</td></tr><tr><td align="left">Second quintile</td><td align="left">17</td><td align="left">20</td><td align="left">19</td></tr><tr><td align="left">Third quintile</td><td align="left">25</td><td align="left">23</td><td align="left">21</td></tr><tr><td align="left">Fourth quintile</td><td align="left">23</td><td align="left">20</td><td align="left">17</td></tr><tr><td align="left">Richest quintile</td><td align="left">22</td><td align="left">14</td><td align="left">17</td></tr><tr><td align="left" colspan="4"><bold>Residence</bold></td></tr><tr><td align="left">Rural (82 per cent)</td><td align="left">65</td><td align="left">89</td><td align="left">83</td></tr><tr><td align="left">Urban (18 per cent)</td><td align="left">35</td><td align="left">11</td><td align="left">17</td></tr><tr><td align="left" colspan="4"><bold>Gender</bold></td></tr><tr><td align="left">Male (51.3 per cent)</td><td align="left">48</td><td align="left">53</td><td align="left">55</td></tr><tr><td align="left">Female (48.7 per cent)</td><td align="left">52</td><td align="left">47</td><td align="left">45</td></tr></tbody></table><table-wrap-foot><p>Source: [37].</p></table-wrap-foot></table-wrap></sec><sec><title>Manpower situation: specialists and urban biases</title><p>In Bangladesh public health care is facing a shortage of personnel. Approximately 12,000 physicians work in the public sector [<xref ref-type="bibr" rid="B31">31</xref>]. Chaudhury and Hammer report that more than 26 per cent of positions in all categories of health personnel are vacant in public health facilities. The vacancy rate for doctors is 41 per cent. This figure represents more than 2,000 public physician positions [<xref ref-type="bibr" rid="B49">49</xref>]. The vacancy rate is higher in rural and poor regions. Moreover, the public health services are gradually tending to have more specialists, rather than mid level health personnel including paramedics, nurses and auxiliary health personnel. The following shows that though there is a gradual improvement of doctor and other health personnel ratio, it still is much lower than global averages [<xref ref-type="bibr" rid="B9">9</xref>]. The developed countries have more mid level health personnel compared to doctors but developing countries like Bangladesh have the opposite picture [<xref ref-type="bibr" rid="B9">9</xref>]. It is observed that the state has produced more specialist than mid-level health personnel. During 1973, there were 5,570 doctors including dentists. The number had increased to 17,560 in 1985 and further to 27,646 in 1997. But the number of mid-level health personnel (i.e. Nurses, Medical Assistants, Pharmacists, Radiographers, Laboratory Technician, Sanitary Inspector and Dental Technician) was 3,665, 18,865 and 26,715 respectively, in the same period [<xref ref-type="bibr" rid="B43">43</xref>-<xref ref-type="bibr" rid="B46">46</xref>,<xref ref-type="bibr" rid="B50">50</xref>,<xref ref-type="bibr" rid="B51">51</xref>]. Since independence, the government has invested more on medical colleges or development of specialized medical institutions but little attention has given to develop paramedical and nursing institutions. This leads to development of more physicians than paramedics, nurses and other auxiliary personnel.</p><p>In the early 1970s, there was only one postgraduate, eight graduate and one paramedical institute. Currently, there are six post graduate institutes, 14 medical colleges, three dental colleges, as against two para-medical institutes, five medical training schools, one college of nursing and 38 nursing institutes in the public sector. Every year 1,250 postgraduate and graduate doctors are in the profession but the number of paramedics is less than 200 and that of nurses about 900 [<xref ref-type="bibr" rid="B44">44</xref>,<xref ref-type="bibr" rid="B45">45</xref>,<xref ref-type="bibr" rid="B50">50</xref>,<xref ref-type="bibr" rid="B51">51</xref>]. The turn-out of health technologists, paramedics and nurses is below the national requirement. The revenue expenses for medical colleges and other training schools reveal that from 1985–86 to 1989–90, the share of the medical colleges was 3.3 per cent as against 1.2 per cent share of the training schools. From 1990–91 to 1994–95, the share was the same for both the sectors [<xref ref-type="bibr" rid="B52">52</xref>]. The manpower development plan and the resource allocations indicate the high preference for fully-fledged and specialist doctors. These circumstances show that paramedics and nurses get very little attention.</p><p>Furthermore, the health system is urban biased in facility development and resource distribution. There were 15,706 beds available in the urban areas and the share of the rural areas was 11,297 in 1990 [<xref ref-type="bibr" rid="B53">53</xref>] and the comparative figures in 1998 were 14,037 and 12,292 respectively [<xref ref-type="bibr" rid="B46">46</xref>,<xref ref-type="bibr" rid="B51">51</xref>]. It is also seen that all the specialized and super-specialized hospitals and 14 medical colleges, are located in the city centres only [<xref ref-type="bibr" rid="B28">28</xref>].</p><p>The manpower distribution is also more urban oriented. Generally, physicians prefer to locate their practice in the urban areas where they get better income opportunities, better living facilities and other socio-cultural services. There is shortage of doctors in the union and <italic>upazilla </italic>level health centres. The government is not able to provide even a graduate doctor in all the union level health facilities but there is an over concentration of health personnel in the urban area. It is observed that there were 359 doctors at Institute of Postgraduate Medicine and Research (presently Medical University), 384 in Dhaka medical college, 253 in Salimullah Medical College in 1990 [<xref ref-type="bibr" rid="B54">54</xref>].</p><p>Absenteeism is another problem which is common in the public health care system. A background study of the World Bank Report (2004) found that 42 per cent of all categories of health personnel employed in public facilities are usually absent. For physicians the absentee rate at the <italic>upazilla </italic>level is 40 per cent and at the union level 74 per cent [<xref ref-type="bibr" rid="B49">49</xref>]. The recent media reports also support that many regional and rural districts and <italic>upazilla </italic>have shortage of doctors and other auxiliary health personnel. Various national daily new papers always reports about shortage of manpower and instrument s in public health facilities at UHC and District level. So it can be said that manpower development and resource distribution has been directed to specialists and urban biases.</p></sec><sec><title>Level of patients care</title><p>There has been a decreasing trend of outpatient use at government health facilities. In 1993, there were 23.50 million outpatient attendance in the hospitals that declined to 15.65 million in 1996; for UHCs and for USC, the figures were 24.98 million in 1993, which declined to 17.18 million in 1996 [<xref ref-type="bibr" rid="B51">51</xref>]. Another study confirms the same findings. It is found that in 1984, about 20 per cent of the rural patients suffering from acute illnesses were treated in the public sector. This declined to 13 per cent in 1987 and to 12 per cent in 1994 [<xref ref-type="bibr" rid="B55">55</xref>]. This declining trend can be attributed to the non-availability of doctors, drugs, and inadequate attention by the doctors, and also distance from home. A study indicates that a minimum of 28.1 per cent mentioned inadequate attention from the doctors, 25.7 per cent talked about the non-availability of drugs, 4.9 per cent about the long waiting time, and 9.2 per cent about the very long distance [<xref ref-type="bibr" rid="B56">56</xref>]. It shows that those who go to seek health care from public facilities are directly and indirectly neglected by the functionaries of health facilities. This leads to greater use of private facilities. A World Bank report states that the private sector was used widely: 56 per cent of caesarian; 92 per cent of children with diarrhoea and 75 per cent of "first consultations" were conducted in this sector. The same study reports that, of those who sought health care for illness, 87 per cent of the "urban sample" and 75 of the "rural sample" consulted private providers [<xref ref-type="bibr" rid="B57">57</xref>,<xref ref-type="bibr" rid="B19">19</xref>,<xref ref-type="bibr" rid="B20">20</xref>]. The editorial of an English- language daily newspaper commented that there "has been the erosion of public trust so that fewer and fewer people are turning to the government doctors for medical services" [<xref ref-type="bibr" rid="B58">58</xref>].</p><p>The Bangabundhu Sheikh Mujib Medical University, the only medical university and the highest level of health facilities in the country, has mostly paying beds. It is difficult for the general population to get services there. There are other specialized and medical college hospitals which do have a good number of free beds. But shocking fact is that the general population get difficulty in getting admission to these facilities. So the influential and the affluent have greater access to the services of these hospitals, whether in the public or the private sector. The poorest and the rural residents receive the least service or benefit from the government facilities.</p></sec><sec><title>Priority on curative care</title><p>There have been specific services for preventive and curative care within the public health care system. But the public health services in Bangladesh focus on curative care rather than preventive care. The plans of the government reveal priority on preventive care, but in practice, all efforts are directed towards curative care. In the early 1970s, preventive uni-purpose programmes were launched to control malaria, small pox and other epidemic diseases. In the late 1970s, all the uni-purpose (vertical) programmes were integrated and the preventive efforts were shifted to childhood diseases only through an expanded programme of immunization and health education. In the 1980s, the preventive health programmes became a part of the development programmes for health.</p><p>The revenue expenditure (which provide recurring cost) on preventive care was around 8 per cent in the early 1980s and only 0.14 per cent in early the 1990s. But the shares of allocation increased in Annual Development Programmes (ADP) from 13 per cent in the early 1980s to 25 per cent in the early 1990s. In spite of this increasing trend, the total revenue expenses and development allocations together is not more than 8 per cent of the health sector allocation for preventive care [<xref ref-type="bibr" rid="B52">52</xref>]. However, three-quarters of morbidity originated from infectious and parasitic diseases. If the government takes appropriate measures to control the above mentioned, only then it is possible to prevent them but lack of appropriate measures like resource allocation for preventive care indicates a relative indifference about preventive care. Hence we see biases for curative care. It indicates that curative care is more admired than preventive care.</p></sec><sec><title>Expenditure pattern</title><p>The allocation pattern in the health sector indicates the nature and extent of the government's commitment to health sector's development. The allocation pattern shows that from the first five year plan to the fourth five year plan, the share of allocation to health of the total plan allocation was respectively 3.32 per cent, 3.72 per cent, 2.20 per cent, and 3.05 per cent, and at the same time utilization was 90.47 per cent, 90.79 per cent, 79.63 per cent, and 99 per cent [<xref ref-type="bibr" rid="B59">59</xref>]. During the two-year plan holiday period (1995–1997), the allocation was 4.61 per cent of national outlay and utilization was 77.44 per cent [<xref ref-type="bibr" rid="B46">46</xref>]. The allocation and utilization pattern shows poor commitment of government for the development of the health sector.</p><p>The expenditure pattern also reveals the bias of the government towards the urban residents. The per capita expenditure in the public sector in the urban areas is TK. 118 for in-patient service and TK. 79 for out-patient services. But the corresponding share of the rural areas is TK. 41 and TK. 37, respectively (see Table <xref ref-type="table" rid="T4">4</xref>). The amount of total expenditure on the medical personnel also indicates the same bias. The share of the urban areas is TK. 230 and TK. 110 is of the rural areas. It indicates that rural population is neglected comparing to urban population. The share for men in out-patient care is TK. 49.1 and in-patient care is TK. 56.1 and the corresponding figures for women are TK. 43.7 and TK. 60.9 [<xref ref-type="bibr" rid="B60">60</xref>]. Overall 17 per cent of the total government health subsidies benefit the poorest quintile of the population, while 25 per cent benefits the richest quintile of the population. The per capita public expenditure for the richest (income quintiles) is TK. 90 (31 per cent) and for the poorest is TK. 39 (13 per cent) for in-patient services. The share for out-patient services is TK. 53 (23 per cent) and Tk. 43 (18 per cent) respectively for these two groups [<xref ref-type="bibr" rid="B60">60</xref>]. It indicates that government expenditure pattern is towards urban areas, non-poor and males. The public health services are meant for the economically backward strata of the society but it is pro-rich in practice.</p><table-wrap position="float" id="T4"><label>Table 4</label><caption><p>Expenditure Pattern (Figures in Taka)</p></caption><table frame="hsides" rules="groups"><thead><tr><td align="left"><bold>Pattern of expenditure</bold></td><td align="left"><bold>Types</bold></td><td align="left"><bold>Out-patient care </bold></td><td align="left"><bold>In-patient care</bold></td><td align="left"><bold>Total</bold></td></tr></thead><tbody><tr><td align="left">Per capita subsidy by location</td><td align="left">Urban</td><td align="left">79.1</td><td align="left">117.8</td><td align="left">196.9</td></tr><tr><td></td><td align="left">Rural</td><td align="left">36.7</td><td align="left">40.7</td><td align="left">77.4</td></tr><tr><td align="left">Per capita subsidy by income quintiles</td><td align="left">Rich</td><td align="left">52.5</td><td align="left">90.1</td><td align="left">142.6</td></tr><tr><td></td><td align="left">poorest</td><td align="left">42.8</td><td align="left">38.9</td><td align="left">81.6</td></tr><tr><td align="left">Per capita health expenditure</td><td align="left">Male</td><td align="left">49.1</td><td align="left">56.1</td><td align="left">105.2</td></tr><tr><td></td><td align="left">female</td><td align="left">43.7</td><td align="left">60.9</td><td align="left">104.6</td></tr></tbody></table><table-wrap-foot><p>Source: [60]</p></table-wrap-foot></table-wrap></sec><sec><title>Manpower shortages and level of corruption</title><p>The public health system is facing a problem with manpower shortage. There are only 12,000 physicians working in the public sector and more than 2,000 physicians posts laying vacant in the same sector. It is difficult to provide services with the existing manpower. On the other hand, corruption is rampant in the public health care system of Bangladesh. A study confirms the widespread collection of unofficial fees at various level health facilities is a "common form of rent seeking behaviour in Bangladesh" [<xref ref-type="bibr" rid="B61">61</xref>]. The Transparency International found that the health sector is the second most corrupt sector after the police sector [<xref ref-type="bibr" rid="B62">62</xref>]. The survey found that 48 per cent admitted to government hospital by alternative methods including 56 per cent paid money, 22 per cent used influence, and 18 per cent sought help from hospital staff [<xref ref-type="bibr" rid="B62">62</xref>]. The study shows that doctors are most corrupted followed by hospital staff [<xref ref-type="bibr" rid="B62">62</xref>]. An editorial of an English daily commented that Bangladesh experiences show "more than their number, corruption and lack of integrity of the doctors are perhaps, more important factors that explain the poor quality of services at the government run hospitals" [<xref ref-type="bibr" rid="B63">63</xref>]. So the shortage of manpower as well as corruption makes the public sector in a bad shape. In addition to the above, Gruen et al., maintains that the public sector has incurred problems of equipment, essential supplies, inadequate facilities, lack of cleanliness, long waiting time, absence or lack of doctors and nurses, inappropriate behaviour by doctors, and lack of confidence in public facilities and staff [<xref ref-type="bibr" rid="B64">64</xref>]. Moreover poor management, planning and lack of control make the public sector a defunct system. The public health care system has lost its credibility and people have limited confidence in it. This has resulted in the proliferation of private for-profit oriented health care system.</p></sec></sec><sec><title>Summary</title><sec><title>Health system, human rights and under-development</title><p>Though the health system is expanding in terms of health care facilities and manpower, it is far from a comprehensive and integrated health service. The health system has tried to improve the quality of care both at domiciliary and institution level. But the result is far from a satisfactory level and a comprehensive quality development plan has not been made so far. The health system in Bangladesh is biased in favour of the rich and the urban elite in the provision of health care though policies stress to the poor. The government emphasizes the construction of buildings of hospitals rather than providing essential medical facilities, and aims more at curative than preventive aspects. There were little quantitative developments; qualitatively it remained in a dismal state.</p><p>Though the prime concern of the government commitment and desire is for better provision of health service for the masses, the practical scenario is not good enough. It is disheartening and gloomy. The present health system has not succeeded to materialize and to realize the vision enunciated in government plans, policies and programmes. It is only able to provide basic services for about forty per cent of the population [<xref ref-type="bibr" rid="B37">37</xref>,<xref ref-type="bibr" rid="B51">51</xref>], which indicates that the rest are unable to access the health system, and this may be regarded as a violation of human rights. The state has largely failed to respect, fulfil, protect and to promote basic human rights. According to UDHR, ICESCR and the Bangladesh Constitution, it is the obligation of the state to institute policies to address human development, which realise human dignity and thus help to promote human rights. But one will find a dismal picture if he or she tries to contextualize the health system and the development level of the health situation and its relation to human rights. The government has not achieved the goal to facilitate progressive realization of minimum entitlements to the population. The political process and social dynamics allow the poorer, weaker and the disadvantaged groups to access their rights in a very limited way. The State has limited capacity in implementing universally accepted social rights as the state has limited economic and human resources. Moreover, poor governance also jeopardizes government capacity in enhancing social rights. On the other hand, victims are unable to fight against the powerful. There are few organizations which speak peoples' rights but they are still in formative stage and could not succeed in realizing demands. In addition to problems of the health system, there are a lot more problems with allied sectors which can greatly contribute to the improvement of the health status of the population [<xref ref-type="bibr" rid="B65">65</xref>]. The poor inputs like the shortage of food, drugs, and health facilities and health personnel are the major causes of the poor health status of the Bangladeshis. There are problems with housing, sanitation, safe water, income, employment, education, and accessibility to services that aggravate the poor health status of the population. However, the constraints could be overcome at least moderately if the government would have a strong political will and determination to solve the problems. The government should remember that the ratifying states have obligations under article 12 of ICSECR regardless of their economic development. The article further states that each party should undertake steps for progressive realization of the rights.</p><p>So the responsibility lies with the government for the genuine implementation of the right to health. The government does not have ability to facilitate for advancing the fulfilment of minimum provision of health care to the masses. The <italic>Limburg </italic>principles also implies effective use of available resources. The government also has failed to do so in resource allocation and also proper utilization of the resources. The health sector policy, planning, and action have been poorly implemented. Moreover budgetary allocations, manpower distribution, accessibility, availability, spatial distribution of the health institutions are not homogenous and/or equitable which creates problems to promote the health system as a universal character. The lack of political will, corrupt practices of the officials; unwillingness of the public sector doctors to provide services have made the health system unacceptable. Therefore, the health system has not succeeded to establish as a step forward to development, rather it is merely a case of underdevelopment. The right to development has not been constituted and institutionalized in the provision of health care. So the basics of human rights have not been materialized in the health system of Bangladesh.</p></sec><sec><title>What should be government priority?</title><p>In order to provide better health care facilities that the country urgently needs for the majority population, is to implement the policies and programmes enumerated in various government documents. Moreover, the government should take a comprehensive approach to address the human rights and health issues to pursue as a path of development which will help to proceed as a right to development. Necessary initiatives should also be taken to create employment opportunity, income generation, and more production of food and services. Active measures should be taken to reduce disparity between the rural and urban areas, the male and female gap, rich and poor and equity in access to food, health care facilities and other ancillary services. There is a further need to bring out explicit actions of various inter-sectoral cooperation on different development practices, as well as various intra-sectoral efforts. These could be made in the field of health, family planning, drugs, pharmaceuticals, medical education and research, agriculture, food, water supply sanitation, drainage, housing, education rural development, social welfare, women's development to put human rights and health agenda on a path of right to development.</p></sec></sec><sec><title>Competing interests</title><p>The author(s) declare that they have no competing interests.</p></sec><sec><title>Pre-publication history</title><p>The pre-publication history for this paper can be accessed here:</p><p><ext-link ext-link-type="uri" xlink:href="http://www.biomedcentral.com/1472-698X/6/4/prepub"/></p></sec>
Classification of human breathing sounds by the common vampire bat, <italic>Desmodus rotundus</italic>	<sec><title>Background</title><p>The common vampire bat <italic>Desmodus rotundus </italic>is one of three bat species that feed exclusively on the blood of mammals often more than 1000 times its size. Vampire bats even feed on human blood. Moreover, they tend to feed on the same individual over consecutive nights.</p></sec><sec><title>Results</title><p>Using psychoacoustical methods, we show that vampire bats can recognize individual humans by their breathing sounds. Accompanying psychoacoustical experiments using the same stimuli and procedure but with human listeners show that even these trained and instructed listeners were unable to achieve the vampire bats' performance under the most difficult conditions, where the breathing sounds had been recorded under physical strain.</p></sec><sec><title>Conclusion</title><p>It is suggested that vampire bats can make use of an individual acoustic signature imposed on breathing sounds in a way similar to that in which we identify humans by their vocalizations.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Gröger</surname><given-names>Udo</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" corresp="yes" contrib-type="author"><name><surname>Wiegrebe</surname><given-names>Lutz</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Biology	<sec><title>Background</title><p>Vampire bats are the only mammals that feed exclusively on blood. The common vampire bat, <italic>Desmodus rotundus</italic>, is one of three vampire species (Fig. <xref ref-type="fig" rid="F1">1a</xref>). Typically, <italic>D. rotundus </italic>only feeds on a single prey animal on any one night. The blood meal lasts from 10 min up to 1 hour during which time the vampire bat drinks between 0.5 times and 1.4 times its body weight in blood. Despite the abundance of prey animals since the start of livestock farming, vampire bats repetitively feed on the same individual whereas other individuals are ignored [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B3">3</xref>]. The perceptual features that guide the vampire bats' selection of prey animals and enable them to recognize an animal that they have fed on the night before remain elusive.</p><fig position="float" id="F1"><label>Figure 1</label><caption><p><bold>Animal portrait (a) and experimental setup (b) for the behavioral experiments on breathing-sound classification and absolute thresholds in the common vampire bat, <italic>Desmodus rotundus</italic></bold>. For the breathing-sound classification, animals were trained to associate each of three breathing sounds, presented from a speaker above the starting area, with one of the three feeders. For the absolute-threshold measurements, stimuli were presented from one of three speakers mounted above the feeders.</p></caption><graphic xlink:href="1741-7007-4-18-1"/></fig><p>Earlier studies revealed sensory specializations that support the extraordinary food and feeding strategies of <italic>D. rotundus</italic>: Kurten and Schmidt [<xref ref-type="bibr" rid="B4">4</xref>] found pit organs in the nose of <italic>D. rotundus </italic>that are sensitive to the infrared radiation emitted by the blood-rich skin surfaces of homeothermic vertebrates. However, the detection range for this infrared emission is only 8 to 12 cm [<xref ref-type="bibr" rid="B5">5</xref>]. Thus, infrared sensitivity cannot help to locate or select the prey animal but will identify a promising place to bite it. <italic>D. rotundus </italic>has a very well developed olfactory system. Both anatomical [<xref ref-type="bibr" rid="B6">6</xref>] and behavioral [<xref ref-type="bibr" rid="B7">7</xref>] studies indicate that olfaction may play an important role in both the long-distance orientation towards potential prey and possibly the selection of individual prey animals.</p><p>The importance of passive hearing, as opposed to echolocation, for the common vampire is supported by the very low thresholds of midbrain neurons in the frequency range between 10 and 25 kHz, i.e. considerably below the echolocation frequency range (~40 to 100 kHz). Moreover, these recordings have revealed neurons that are stimulated exclusively by breathing sounds [<xref ref-type="bibr" rid="B8">8</xref>].</p><p>This study was designed to investigate the passive-hearing capabilities of the common vampire bat behaviorally; specifically, the extent to which auditory sensitivity to breathing sounds may support prey selection. Vampire bats were trained to discriminate three sequences of breathing sounds recorded from three different subjects. A spectrogram of a recorded breathing sound is shown in Fig. <xref ref-type="fig" rid="F2">2</xref>. Once the vampire bats had learned this task, additional breathing sounds, recorded under different experimental conditions from the same three subjects, were randomly interspersed. The spontaneous association of these test sounds with the learned training sounds was assessed. These psychophysical data from the vampire bats were compared to the performances of human listeners using the same experimental paradigm and stimuli. The experimental data, together with simulations of the breathing-sound task based on different sound parameters, allow the relevance of these parameters to the perceptual association of breathing sounds in both vampire bats and humans to be evaluated.</p><fig position="float" id="F2"><label>Figure 2</label><caption><p><bold>Spectrogram of a recorded breathing sound</bold>. The spectrogram shows the spectro-temporal features of an air intake (left) and output (right) in a single breathing-sound cycle. Frequency-modulated, ultrasonic tonal components can be observed during air intake. These components result from turbulences in the nasal cavity during depression. Note the good signal-to-noise ratio, which was realizable with the recording equipment described in the methods.</p></caption><graphic xlink:href="1741-7007-4-18-2"/></fig></sec><sec><title>Results</title><sec><title>Vampire behavior</title><p>The percentage spontaneous associations of nine unknown test sounds with the correct subject are shown in the two panels of Fig. <xref ref-type="fig" rid="F3">3</xref> for the two vampire bats, respectively. The data show that the vampire bats reliably associated these test sounds with the correct subject (p < 0.05 [<xref ref-type="bibr" rid="B9">9</xref>], horizontal dotted lines). The vampire bats were even able to recognize the recordings under physical strain, although performance dropped markedly for this test-sound condition. The importance of passive hearing (as opposed to echolocation) is reflected in the very low auditory thresholds in the frequency range of breathing sounds. The absolute thresholds for long-duration narrow-band noise stimuli are shown as a function of the band-pass center frequency by the fine lines and circles in Fig. <xref ref-type="fig" rid="F4">4</xref>. Data are averaged across three vampire bats; error bars represent standard errors. The bold line in Fig. <xref ref-type="fig" rid="F4">4</xref> shows the averaged breathing-sound spectra. The gray lines represent standard errors. These data show that the high auditory sensitivity of the vampire bat in the frequency range between 10 and 25 kHz is well suited to the detection and analysis of the breathing sounds we recorded. Owing to the noisy, broadband nature of breathing sounds, it is not to be expected that breathing sounds from other mammals of comparable size differ markedly from human breathing sounds in their magnitude spectra.</p><fig position="float" id="F3"><label>Figure 3</label><caption><p><bold>Spontaneous classification of unknown breathing sounds by <italic>D. rotundus</italic></bold>. The two panels show results from two individuals. The correct recognition of a breathing sound recorded from one of three human subjects is shown as a function of the condition of breathing-sound recording. The different bar colors represent the three different subjects. The bold horizontal line at 33 % correct indicates chance level, the gray line at 47 % correct indicates significant deviation from chance with p < 0.05. The data show that both animals recognized the breathing sounds correctly under all experimental conditions. Only under the most difficult condition, where the sounds were recorded under physical strain (unlike the training sounds), the bats' performance dropped markedly but still remained significant.</p></caption><graphic xlink:href="1741-7007-4-18-3"/></fig><fig position="float" id="F4"><label>Figure 4</label><caption><p><bold>Vampire hearing and breathing-sound spectra</bold>. Absolute auditory thresholds averaged across three individuals of <italic>D. rotundus </italic>are shown as symbols and the fine line; error bars represent standard errors of the mean. Averaged breathing-sound magnitude spectra are shown by the strong black line; the strong grey lines again represent standard errors. This comparisons show that in <italic>D. rotundus</italic>, auditory sensitivity is well suited to detecting and analyzing breathing sounds.</p></caption><graphic xlink:href="1741-7007-4-18-4"/></fig></sec><sec><title>Human behavior</title><p>Breathing-sound classification by human listeners is shown in Fig. <xref ref-type="fig" rid="F5">5</xref> in the same format as Fig. <xref ref-type="fig" rid="F3">3</xref>. The human listeners were well capable of correctly associating the test sounds, recorded either in the same recording session or in another session under a similar degree of physical strain of the subjects. However, unlike the vampire bats, the naïve listeners (left panel) completely failed to classify the breathing sounds recorded under physical strain correctly. Instead, all sounds recorded under physical strain were consistently associated with subject #3. This subject used the highest breathing frequency in the training set. Breathing frequency is simply defined as the number of breathing cycles per second. Note that the human listeners went through the same training procedure as the vampire bats and scored at least as well as the vampire bats in the training condition and in the first two test conditions.</p><fig position="float" id="F5"><label>Figure 5</label><caption><p><bold>Breathing-sound classification by human listeners with exactly the same stimuli and experimental procedure as used in the vampire experiment</bold>. Results are shown in the same format as in Fig. 3. Error bars represent across-listeners standard errors. While the human listeners were well capable of recognizing breathing sounds recorded in the same session or in a different session in the same physical state, the naïve human listeners failed to recognize the breathing sounds recorded under physical strain (a). Even after the listeners were instructed to ignore breathing-frequency information (b), overall performance did not improve for this most difficult experimental condition.</p></caption><graphic xlink:href="1741-7007-4-18-5"/></fig><p>After completion of this data set, listeners were instructed to ignore breathing-frequency information but to try instead to exploit all other information. This was done because both the listeners' reports and the numerical simulations (see below) suggested that the listeners had been trying to use breathing-frequency information to classify the breathing sounds. This instruction had obviously not been given to the vampire bats. The results obtained under this experimental condition are shown in the right panel of Fig. <xref ref-type="fig" rid="F5">5</xref>. While the pattern of results changed in comparison to the naïve condition, listeners still failed to associate breathing sounds recorded under physical strain correctly.</p></sec><sec><title>Simulations</title><p>The numerical simulations of the breathing-sound association were implemented to assess whether the vampire bats' or humans' performances can be linked to a physical sound parameter. Simulation results for the four extracted sound parameters are shown in the four panels of Fig. <xref ref-type="fig" rid="F6">6</xref>. Overall, the simulation based on breathing frequency provides the best results while the power-spectrum simulation provides the worst. However, none of the simulations can correctly classify all test sounds recorded under physical strain. Note the high similarity between the breathing-frequency simulation (Fig. <xref ref-type="fig" rid="F6">6b</xref>) and the performance of the naïve human listeners (Fig. <xref ref-type="fig" rid="F5">5</xref>, left panel). This similarity supports the hypothesis that human listeners relied on breathing-frequency analysis.</p><fig position="float" id="F6"><label>Figure 6</label><caption><p><bold>Numerical simulation of the breathing-sound recognition</bold>. The simulations are based on four different breathing-sound parameters, namely (a) the sound level, (b) the breathing frequency, (c) the power spectrum of the breathing sounds and (d) the breathing-sound roughness. None of the simulations matches the behavioral performance of the vampire bats (Fig. 3). The breathing-sound simulation, however, matches the performance of the naïve human listeners (Fig. 5a), confirming the listeners' reports that they had tried to rely on breathing-frequency information.</p></caption><graphic xlink:href="1741-7007-4-18-6"/></fig><p>These analyses were performed on the sounds as they were recorded. The audiogram of a vampire bat, however, is quite different from that of humans. To capture the possible effects of the different frequency ranges in which the bats and the humans processed the sounds, the simulations were repeated with the sounds filtered to match either the human or the vampire-bat audiogram. These simulations were only carried out for the three parameters that are potentially influenced by the filtering, namely sound-pressure level, differences in the magnitude spectrum and stimulus roughness. However, none of these parameters alone yielded significantly better predictions with the filtered sounds (not shown).</p><p>It is conceivable that the bats did not rely on a single sound parameter but on different parameters depending on which parameter yielded the strongest predictions for a given comparison of a test sound with the three training sounds. Although the magnitude spectrum did not provide strong predictions for a given test sound (the mean-squared difference between the magnitude spectrum of the test sound and each training sound was similar), the stimulus roughness provided strong predictions e.g. in favour of the training-sound of subject one.</p><p>Consequently, a simulation was implemented in which the model was free to choose the parameter for a given test sound that provided the strongest predictions. Again, test sounds were either filtered to match the human or the vampire-bat audiogram. Simulation results are shown in Fig. <xref ref-type="fig" rid="F7">7</xref>. While the predictions of this model are still poor when the sounds were filtered to match the human audiogram (Fig. <xref ref-type="fig" rid="F7">7a</xref>), they were quite good in predicting the bats' performance when the sounds were filtered to match the vampire-bat audiogram.</p><fig position="float" id="F7"><label>Figure 7</label><caption><p><bold>Simulation of the breathing-sound recognition with dynamically changing analysis parameters</bold>. Predicted breathing-sound classification when the model was free to choose the breathing-sound parameter (sound level, power spectrum or roughness) that provides the strongest predictions (strongest deviations from chance performance) for a given comparison of a test sound with each of the three training sounds. All sounds were filtered to match either the human audiogram (a) or the vampire-bat audiogram (b). While the simulation based on the human audiogram does not yield improved predictions of the classification of the test sounds recorded under physical strain, this simulation approach when combined with the sounds as they are weighted by the vampire-bat audiogram results in qualitatively correct classifications even of the test sounds recorded under physical strain.</p></caption><graphic xlink:href="1741-7007-4-18-7"/></fig><p>In summary, the current functional simulations suggest that the vampire bats spontaneously recruited a rather sophisticated analysis of the sounds based on multiple parameters; and they appeared to base decisions for each test sound on the sound parameter that provided the strongest discriminative capacity for comparing this test sound with the three training sounds. It appears that this analysis is more reliable when based on ultrasonic components of the sounds.</p></sec></sec><sec><title>Discussion</title><p>The current data show that for vampire bats, prey-generated breathing sounds could provide a reliable cue for recognizing prey individuals: during the relatively long time a vampire feeds on a prey animal, it can memorize the prey's breathing sounds and use this information to find the same prey on the following night.</p><p>Breathing sounds are typically faint. The sounds we recorded from human subjects ranged between 25 and 35 dB SPL. This gives rise to the question: over what distance could breathing sounds be perceived and analyzed by vampire bats? Considering the absolute thresholds (cf. Fig. <xref ref-type="fig" rid="F4">4</xref>), the frequency region most likely to be used is around 15 kHz, where thresholds are as low as 0 dB SPL. In this frequency region, atmospheric attenuation is around 0.5 dB/m. Thus, in the (unlikely) absence of any masking sounds, detection of breathing sounds could work over several tens of meters. In the presence of natural masking sounds, however, the effective detection distance will depend on the level and spatial distribution of the masking sound sources.</p><p>While it is unlikely that prey recognition relies exclusively on breathing sounds [<xref ref-type="bibr" rid="B5">5</xref>,<xref ref-type="bibr" rid="B7">7</xref>], these sounds potentially have high individual significance: vocalizations are generated by the vocal cords and filtered through the vocal tract. Both the pattern of vocal-cord vibrations and the filtering are highly individual and this supports our recognition of individual voices. While breathing sounds are unvoiced, and thus do not excite the vocal cords, they will also pass the same vocal and nasal tract and may thus also mediate individually specific information. However, the sounds used in this study were emitted through the nose. It remains to be investigated to what extent the nasal acoustic tract filters breathing sounds in a similarly characteristic way. An early study confirmed that, at least for partially voiced sounds such as consonant-vowel combinations, speaker recognition is feasible on the basis of nasal co-articulation [<xref ref-type="bibr" rid="B10">10</xref>]. However, the current simulations based on the breathing-sound power spectra, and the human-psychophysical experiments, suggest that speaker recognition is difficult with purely unvoiced sounds. If nasal-tract filtering were individually specific, the resulting spectral features would result in a correct breathing-sound association in the power-spectrum simulation. The simulation results in Fig. <xref ref-type="fig" rid="F6">6c</xref> show that this is not the case. Also, the failure of the instructed human listeners to associate the breathing sounds recorded under physical strain argues against the use of power-spectrum information for breathing-sound recognition, at least in the audio frequency range below 20 kHz. Note that human listeners are very sensitive to changes in the spectral composition of broadband stimuli [<xref ref-type="bibr" rid="B11">11</xref>].</p><p>Qualitatively correct predictions of the vampire-bats' performance, even under the most difficult experimental condition where the test sounds had been recorded under physical strain, could be obtained with a refined simulation approach: first, the sounds were filtered to match the vampire-bat audiogram; and second, a simulation paradigm was designed that allows whichever sound parameter yields the strongest predictions to be exploited.</p></sec><sec><title>Conclusion</title><p>The current behavioral study shows that the common vampire bat, <italic>Desmodus rotundus</italic>, is very sensitive to breathing sounds. In the three-alternative, forced-choice setup, it spontaneously associates unknown breathing sounds with the subject who emitted them. This exceptional performance is underlined by the inability of human listeners to match the vampire bats' accomplishment under the most difficult experimental condition where the sounds had been recorded under physical strain. Numerical simulations show that while the human listeners relied on breathing-frequency information, the vampire bats appeared to recruit different acoustic parameters and to choose amongst these parameters depending on which provided the highest discriminative power.</p><p>On the basis of these findings, it is suggested that vampire bats can memorize and classify complex acoustic features of prey-generated breathing sounds to facilitate the identification of prey animals that they have successfully fed on before.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>Breathing-sound recordings</title><p>Breathing sounds were recorded from three human subjects (two female, one male) aged between 27 and 30 and breathing through the nose. The microphone (Bruel&Kjaer 4189 with B&K 2671 preamplifier, Naerum, Denmark) was positioned at a horizontal distance of 10 cm and 2 cm below the nose tip. This position was found after extensive tests to ensure maximum acoustic sensitivity to the breathing sound while, at the same time, minimizing the risk of the air stream hitting the microphone directly. Care was taken to exclude recordings where the air stream hit the microphone. The microphone was connected to a B&K 2525 measuring amplifier; the output was high-pass filtered at 1 kHz (Krohn Hite 3550, Brockton, MA) and analog-digital converted (Tucker Davis Technologies RP2.1, Alachua, FL) at a sampling rate of 100 kHz. This recording system had a flat frequency response up to 35 kHz followed by a shallow decay (about 12 dB/oct.), caused by the 1/2 inch microphone. This microphone was chosen because, unlike a ¼ inch microphone, the background level was low enough to obtain a reasonable signal-to-noise ratio of 20–25 dB with the breathing sounds. Three 40-s sessions were recorded from each subject. One session was recorded after the subject was subjected to physical strain (20 knee bends).</p><p>The breathing sounds used to train the vampire bats consisted of a sequence of full breathing cycles lasting between 7 and 10 s extracted from the first recording session. The test sounds also consisted of full cycles lasting between 7 and 10 s extracted (1) from a different time period within the same 40-s session, (2) from a second 40-s session recorded on a different day and (3) from the session recorded under physical strain. Thus, the test program consisted of nine test sounds (three subjects times three test conditions). Downsampled (to 44.1 kHz) and compressed versions (mp3, 128 kB/s) of all training and test sounds are provided in the supplementary information.</p></sec><sec><title>Stimulus presentation</title><p>The breathing sounds were digital-analog converted at 100 kHz (TDT RP2.1), amplified (Rotel RB 976 MkII, Worthing, England) and presented through a speaker (Technics Matsushita EAS10TH800D, Osaka, Japan) mounted vertically above the starting area of the vampire. Emission sound levels were set to the recorded levels (25 to 35 dB SPL) including their natural intra- and inter-individual variability. The frequency response of the playback system was flat (± 2 dB) between 3 and 48 kHz.</p></sec><sec><title>Procedure</title><p>In a three-alternative, forced-choice paradigm, the vampire bats (Fig. <xref ref-type="fig" rid="F1">1a</xref>) were trained to associate each of the three training sounds with a corresponding reward feeder located at the ends of the three arms of the behavioral setup (Fig. <xref ref-type="fig" rid="F1">1b</xref>). When the vampire arrived at the correct reward feeder before the end of the stimulus presentation, it was rewarded with 0.25 ml of cattle blood provided by an automated syringe system under computer control. A video clip of a vampire approaching a feeder and feeding is provided in the supplementary information [see Additional files <xref ref-type="supplementary-material" rid="S2">2</xref> and <xref ref-type="supplementary-material" rid="S3">3</xref>]. Two of the four animals trained on this task learned to associate each training sound with a specific feeder with more than 70 % correct performance after about 6 months of training. Test trials were then randomly interspersed between the training trials with a probability of 25%. In these test trials, one of the nine test sounds was presented and the vampire bats were rewarded independently of their choice of reward feeder. Whether a trial was a training trial or a test trial, and which feeder would be the correct feeder for the training trials, were determined exclusively by random generators in the software. Control of the automated syringes was done over the IO port of the TDT RP2.1 and this switching only occurred after the animals had made a decision. The results shown are based on at least 30 presentations of each of the nine test sounds. Thus, the data result from at least 270 test trials interspersed between 810 training-sound presentations. Numerical simulation based on random performance shows that the p < 0.05 threshold in this 3-AFC task with 30 trials per condition is 47% correct [<xref ref-type="bibr" rid="B9">9</xref>]. The data-acquisition period lasted about one year.</p></sec><sec><title>Absolute threshold measurements</title><sec><title>Stimuli</title><p>Absolute auditory thresholds were determined for nine center frequencies between 3 and 80 kHz equally spaced on a logarithmic frequency axis. The stimuli were narrow-band noises with a -3 dB bandwidth of the center frequency ± 10 %; the noises were regenerated for each trial. Each noise had a duration of 500 ms including 10 ms raised-cosine ramps. Stimuli were presented through the TDT RP2.1, a TDT PA5 programmable attenuator, the Rotel RB 976 MkII, and a 40 dB passive end attenuation. The end attenuator consisted of a 100 Ω resistor in series and a 1 Ω resistor in parallel to each speaker. The speakers were Technics Matsushita (EAS10TH800D) and they were mounted at the ends of the three arms. The setup was calibrated with a ¼ inch microphone (B&K 4135) connected to a B&K 2670 preamplifier and a B&K 2636 measuring amplifier. Stimuli were presented at a rate of 1 Hz for 15 s or until the vampire had reached one of the feeders. Correct choices were rewarded in the same way as in the breathing-sounds experiment.</p></sec><sec><title>Procedure</title><p>Psychometric functions were obtained over an attenuation range of 35 dB in steps of 5 dB. The overall position of this attenuation range was set according to the individual animal's performance in preliminary trials. Within this 35-dB range, the attenuation for each trial was selected randomly. Each point on the psychometric functions is based on at least 30 trials; the 47 % correct point (corresponding to p < 0.05 in the three-alternative, forced-choice task) on a sigmoidal function fitted to the psychometric function was taken as threshold.</p><p>The shape of the psychometric functions, the frequency dependence of the absolute thresholds and the different associations of the test sounds are incompatible with the hypothesis that the animals relied on other than auditory cues to perform the task. In this context all trials in the absolute-threshold measurements with high attenuations, where the animals' performance was at chance level, can be regarded as blank trials. As the animals typically responded to 20 to 30 trials per day, the data for a single psychometric function required about 8 to 12 training days.</p></sec></sec><sec><title>Animals</title><p>Four adult <italic>D. rotundus </italic>were trained on the breathing sounds. Two animals failed to learn the association of the breathing sounds with the corresponding feeders and were removed from the experiment after 6 months of training. Three animals, two of them identical to those having taken part in the breathing-sound experiment, were subsequently trained on the absolute-threshold measurements. The animals were born in captivity in a colony located at the Zoologisches Institut, Universität Bonn, and they were kindly provided by Prof. Schmidt.</p></sec><sec><title>Human psychophysical experiments</title><p>The breathing-sound experiment was repeated with four human listeners using exactly the same experimental stimuli and paradigm. The listeners were two males and two females (22 to 24 years old) and had normal hearing as determined by preceding measurements of absolute thresholds. The breathing sounds were delivered diotically through AKG K240DF headphones (Wien, Austria) at the recorded sound level. Instead of a blood reward, the listeners received visual feedback from a graphical user interface shown on a touch screen in a sound-proof booth. The touch screen also served as a response interface.</p></sec><sec><title>Simulations</title><p>The simulations were performed to investigate whether the vampire bats' or the humans' performances could be linked to a simple physical sound parameter. Four such parameters (the sound-pressure level, the base-ten logarithm of the breathing frequency, the power spectrum and the breathing-sound roughness) were extracted from the three training sounds and the nine test sounds. The breathing-sound roughness was calculated as the 4<sup>th </sup>moment of the waveform [<xref ref-type="bibr" rid="B12">12</xref>]. The fourth moment is the waveform raised to the power of four divided by the squared waveform raised to the power of two. As the divisor corresponds to the squared variance of the waveform, the division makes the 4<sup>th </sup>moment independent of the sound-pressure level. For each of the four extracted breathing-sound parameters, the similarity between each training sound and each test sound was calculated as the reciprocal of the squared difference between the test sound and a training sound. For the power spectrum, the squared difference was averaged over all frequencies (0 to 50 kHz). Percentage correct identification was calculated by dividing the similarity between the test sound and the correct training sound (from the correct subject) by the sum of the similarities between the test sound and the training sounds from all three subjects.</p><p>To test the influence of the different hearing ranges of vampire bats and humans, the sounds were subsequently filtered with infinite-impulse response filters designed either to match the vampire audiogram (cf. Fig. <xref ref-type="fig" rid="F4">4</xref>) or the human audiogram [<xref ref-type="bibr" rid="B13">13</xref>]. Then the simulations for the three parameters that are potentially affected by this filtering (sound-pressure level, power spectrum and roughness) were repeated. However, the predictions of the breathing-sound classification by the vampire bats did not improve (not shown).</p><p>These simulations only evaluate the contribution of each the extracted parameters but not combinations of parameters. In the first simulations, the breathing-frequency parameter was identified as a distractor for the correct classification of breathing sounds recorded under physical strain. It is conceivable that while none of the other parameters alone can predict the vampire bats' performance, a combination of the remaining parameters, namely sound-pressure level, power spectrum and roughness, may be used by the vampire bats to classify breathing sounds.</p><p>To test this hypothesis, the model was allowed to choose the parameter that produced the strongest predictions, i.e. the strongest deviations from chance level, for the comparison of a given test stimulus with each of the training stimuli. Again, this version of the model was evaluated for both the human and the vampire hearing ranges. Simulation results are shown in Fig. <xref ref-type="fig" rid="F7">7</xref>. While the best-parameter simulation was still unable to achieve a significantly correct performance under the 'physical strain' condition when the sounds were filtered with the human audiogram, filtering with the vampire-bat audiogram resulted in qualitatively correct predictions even for those breathing sounds that had been recorded under physical strain.</p></sec></sec><sec><title>Authors' contributions</title><p>UG and LW designed the vampire experiments. UG carried out the animal training and data acquisition. LW implemented and carried out the human psychophysical experiments and simulations. UG and LW wrote the manuscript and designed the figures.</p></sec><sec sec-type="supplementary-material"><title>Supplementary Material</title><supplementary-material content-type="local-data" id="S2"><caption><title>Additional File 2</title><p>Videoclip of a vampire bat approaching an automated feeder</p></caption><media xlink:href="1741-7007-4-18-S2.avi" mimetype="video" mime-subtype="x-msvideo"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S3"><caption><title>Additional File 3</title><p>XviD is an ISO MPEG-4 compliant video codec, designed to compress/decompress digital video. This needs to be installed to view the .avi.</p></caption><media xlink:href="1741-7007-4-18-S3.exe" mimetype="application" mime-subtype="x-msdownload"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S1"><caption><title>Additional File 1</title><p>Sound files (mp3 format, 44.1 kHz, 128 kB/s) of all sounds used in the breathing-sound experiments.</p></caption><media xlink:href="1741-7007-4-18-S1.zip" mimetype="application" mime-subtype="x-zip-compressed"><caption><p>Click here for file</p></caption></media></supplementary-material></sec>
Kernel-based distance metric learning for microarray data classification	<sec><title>Background</title><p>The most fundamental task using gene expression data in clinical oncology is to classify tissue samples according to their gene expression levels. Compared with traditional pattern classifications, gene expression-based data classification is typically characterized by high dimensionality and small sample size, which make the task quite challenging.</p></sec><sec><title>Results</title><p>In this paper, we present a modified K-nearest-neighbor (KNN) scheme, which is based on learning an adaptive distance metric in the data space, for cancer classification using microarray data. The distance metric, derived from the procedure of a data-dependent kernel optimization, can substantially increase the class separability of the data and, consequently, lead to a significant improvement in the performance of the KNN classifier. Intensive experiments show that the performance of the proposed kernel-based KNN scheme is competitive to those of some sophisticated classifiers such as support vector machines (SVMs) and the uncorrelated linear discriminant analysis (ULDA) in classifying the gene expression data.</p></sec><sec><title>Conclusion</title><p>A novel distance metric is developed and incorporated into the KNN scheme for cancer classification. This metric can substantially increase the class separability of the data in the feature space and, hence, lead to a significant improvement in the performance of the KNN classifier.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Xiong</surname><given-names>Huilin</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" corresp="yes" contrib-type="author"><name><surname>Chen</surname><given-names>Xue-wen</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib>	BMC Bioinformatics	<sec><title>Background</title><p>DNA microarray technology is designed to measure the expression levels of tens of thousands of genes simultaneously. As an important application of this novel technology, the gene expression data are used to determine and predict the state of tissue samples, which has shown to be very helpful in clinical oncology. The most fundamental task using gene expression data in clinical oncology is to classify tissue samples according to their gene expression levels. In combination with pattern classification techniques, gene expression data can provide more reliable means to diagnose and predict various types of cancers than the traditional clinical methods.</p><p>Compared with traditional pattern classifications, gene expression-based data classification is typically characterized by high dimensionality and small sample size, which make the task quite challenging. In the literature, a number of methods have been applied or developed to classify microarray data [<xref ref-type="bibr" rid="B1">1</xref>-<xref ref-type="bibr" rid="B6">6</xref>]. These methods include K-nearest-neighbor (KNN), boosting, linear discriminant analysis (LDA), and support vector machines (SVM), etc. we herein briefly review some of the approaches.</p><sec><title>K-Nearest-Neighbor (KNN)</title><p>The KNN method is a simple, yet useful approach to data classification. The error rate of the KNN has been proven to be asymptotically at most twice that of the Bayessian error rate [<xref ref-type="bibr" rid="B7">7</xref>]. However, its performance deteriorates dramatically when the input data set has a relatively low local relevance [<xref ref-type="bibr" rid="B8">8</xref>]. The most important factor impacting the performance of KNN is the distance metric. It is desirable to adopt an appropriate distance metric for the KNN algorithm. In practice, the Euclidean distance is usually used as the distance metric.</p></sec><sec><title>Diagonal Linear Discriminant Analysis (DLDA)</title><p>DLDA is the simplest case of the maximum likelihood discriminant rule, in which the class densities are supposed to have the same diagonal covariance matrix. In the special case of binary classification, the DLDA scheme can be viewed as the "weighted voting scheme" proposed by Golub <italic>et al</italic>. in [<xref ref-type="bibr" rid="B3">3</xref>]. The major advantage of the DLDA algorithm lies in its computational efficiency.</p></sec><sec><title>Linear Discriminant Analysis (LDA)</title><p>The classical LDA method aims to find the most discriminatory projection directions of the input data and classifies the data in the projected space. A major problem in employing the classical LDA algorithm for classifying gene expression data is that the so called scatter matrices are always singular, due to the nature of high dimensionality and relatively small sample size. The singularity makes the classical LDA algorithm inapplicable. In the areas such as face recognition and text classification, the principal component analysis (PCA) technique is introduced as a preprocessing procedure in order to reduce the dimensionality of the input data. However, since the projection criterion of PCA is essentially different from that of LDA, losing discriminatory information in the PCA step becomes inevitable. A recent development in LDA is the generalized discriminant analysis [<xref ref-type="bibr" rid="B9">9</xref>,<xref ref-type="bibr" rid="B10">10</xref>], in which a more delicate matrix technique, namely, the generalized singular value decomposition (GSVD), is used to modify the classical LDA into a more general version.</p></sec><sec><title>Support Vector Machines (SVM)</title><p>SVM has been recognized as the most powerful classifier in various applications of pattern classification. For binary classification, SVM searches for a hyperplane that separates the two classes of data with the maximum margin. It has been shown that support vector machines perform well in many areas of computational biology [<xref ref-type="bibr" rid="B11">11</xref>-<xref ref-type="bibr" rid="B13">13</xref>]. In the experimental part of this paper, we follow the way in [<xref ref-type="bibr" rid="B14">14</xref>] to implement the SVM algorithm.</p><p>Generally speaking, due to the high dimensionality and small sample size, linear classifiers such as the linear discrimiant analysis (LDA), and the support vector machines (SVM) with linear kernels are used favorably. However, based on some benchmark tests, researchers have shown that nonlinear classfiers are capable of exploring the nonlinear discriminatory information in the microarray data, and usually produce more precise classification results [<xref ref-type="bibr" rid="B15">15</xref>,<xref ref-type="bibr" rid="B16">16</xref>]. This is especially true when more patients' samples are available or the data dimension is substantially reduced, since, in these cases, the linear separability of the microarray data could be considerably degraded.</p><p>Among the general algorithms of pattern classification, K-nearest-neighbor (KNN) is a simple yet useful one. However, in practice, the performance of KNN algorithm is often inferior to those of the sophisticated approaches such as SVM and generalized linear discriminant analysis (GLDA) [<xref ref-type="bibr" rid="B9">9</xref>,<xref ref-type="bibr" rid="B10">10</xref>]. Since the distance metric is of great importance for the KNN scheme, an attractive way to improve the performance of KNN is to adopt a more adaptive distance metric to the input data than the Euclidean diatnce. In this paper, we propose to learn the adaptive distance metric via optimizing a data-dependent kernel. Experimental results show that, compared with the ordinary Euclidean distance-based KNN scheme, our kernel-based KNN algorithm, denoted KerNN, always achieves significant improvement in the performance of classifying gene expression data. Moreover, the performance of the KerNN classifier is shown to be competitive, if not better, to those of the sophisticated classifiers, e.g., SVM and the uncorrelated linear discriminant analysis (ULDA) [<xref ref-type="bibr" rid="B10">10</xref>], in classifying microarray data.</p></sec></sec><sec><title>Results</title><p>We conducted intensive experiments to compare the performances of our KerNN scheme to the commonly-used classification algorithms, i.e., KNN, DLDA [<xref ref-type="bibr" rid="B3">3</xref>], ULDA [<xref ref-type="bibr" rid="B10">10</xref>], and SVM. Ten publicly available microarray data sets were chosen to test our algorithms. The basic information about these data sets is summarized below. Each data set is first normalized to a distribution with zero mean and unity variance in every feature direction, and then, randomly partitioned into two disjoint subsets with equal number of samples, one is used as the training data, and the other the test data. We only consider Gaussian kernel function in the proposed and SVM algorithms.</p><p>1. <italic>ALL-AML Leukemia Data</italic>: This data set, taken from the website [<xref ref-type="bibr" rid="B17">17</xref>], contains 72 samples of human acute leukemia. 47 samples belong to acute lymphoblastic leukemia (ALL), and the other acute myeloid leukemia (AML). Each sample presents the expression levels of 7129 genes. For the detailed information, one can refer to [<xref ref-type="bibr" rid="B3">3</xref>].</p><p>2. <italic>ALL-MLL-AML Leukemia Data</italic>: This leukemia microarray data set is available on the website [<xref ref-type="bibr" rid="B17">17</xref>]. It includes 72 human leukemia samples, 24 of them belong to acute lymphoblastic leukemia (ALL), 20 of them to mixed lineage leukemia (MLL), a subset of human acute leukemia with a chromosomal translocation, and 28 of the samples are acute myelogenous leukemia (AML). Each sample gives the expression levels of 12582 genes. Further information about this data set can be found in [<xref ref-type="bibr" rid="B21">21</xref>].</p><p>3. <italic>Embryonal Tumors of the Central Nervous System (CNS)</italic>: This data set, available at the website [<xref ref-type="bibr" rid="B17">17</xref>], contains 60 patient samples, 21 are survivors of a treatment, and 39 are failures. There are 7129 genes in the data set. One can refer to [<xref ref-type="bibr" rid="B22">22</xref>] to find more information about this data set.</p><p>4. <italic>Breast Cancer Data</italic>: The data are available on the website [<xref ref-type="bibr" rid="B18">18</xref>]. The expression matrix monitors 7129 genes in 49 breast tumor samples. There are two response variables respectively describing the status of the estrogen receptor (ER) and the lymph nodal (LN) status. For the ER status, 25 samples are ER+, whereas the remaining 24 samples are ER-. For the LN variable, there are 25 positive sample and 24 negative samples. The detailed information about this data set can be found in [<xref ref-type="bibr" rid="B6">6</xref>].</p><p>5. <italic>Colon Tumor Data</italic>: This data set is adopted from the website [<xref ref-type="bibr" rid="B17">17</xref>]. The data contain 62 samples collected from colon-cancer patients. Among them, 40 samples are from tumors, and 22 normal biopsies are from healthy parts of the colons of the same patients. 2000 genes were selected to measure their expression levels. One can refer to [<xref ref-type="bibr" rid="B23">23</xref>].</p><p>6. <italic>Lung Cancer Data</italic>: This data set is taken from the website [<xref ref-type="bibr" rid="B17">17</xref>]. It contains 181 tissue samples, which are classified into two classes: malignant pleural mesothelioma (MPM) and adenocarcinoma (ADCA). Each sample is described by 12533 genes. More information about this data set can be found in [<xref ref-type="bibr" rid="B24">24</xref>].</p><p>7. <italic>Lymphoma Data</italic>: The data are available on the website [<xref ref-type="bibr" rid="B19">19</xref>]. This data set contains 77 tissue samples, 58 are diffuse large B-cell lymphomas (DLBCL) and the remaining 19 samples are follicular lymphomas (FL). Each sample is represented by the expression levels of 7129 genes. The detailed information about this data set can be found in [<xref ref-type="bibr" rid="B25">25</xref>].</p><p>8. <italic>Ovarian Cancer Data</italic>: This data set, available on the website [<xref ref-type="bibr" rid="B17">17</xref>], is to distinguish ovarian cancer from non-cancer. It contains 253 samples, and each sample has 15154 features. More details can be found in [<xref ref-type="bibr" rid="B26">26</xref>].</p><p>9. <italic>Prostate Cancer Data</italic>: This data set, adopted from the website [<xref ref-type="bibr" rid="B19">19</xref>], contains the gene expression levels of 12600 genes for 52 prostate tumor samples and 50 normal prostate samples. One can refer [<xref ref-type="bibr" rid="B4">4</xref>] for the details about this data set.</p><p>10. <italic>Subtypes of Acute Lymphoblastic Leukemia</italic>: This data set, available on the website [<xref ref-type="bibr" rid="B20">20</xref>], contains 6 subtypes of pediatric acute lymphoblastic leukemia, corresponding to six diagnostic groups: BCR-ABL, E2A-PBX1, MLL, T-ALL, TEL-AML1, Hyperdiloid>50. Each sample contains 12625 genes.</p><sec><title>Comparisons in terms of the best results</title><p>For each data set, we chose the <italic>N</italic><sub><italic>f </italic></sub>most discriminatory genes, where <italic>N</italic><sub><italic>f </italic></sub>= 10, 20, 40, 60, 80, 100, 200, 400, 600, 800, 1000, 1200, 1400, 1600, 1800, 2000, respectively; repeated the experiment 100 times at each value of <italic>N</italic><sub><italic>f</italic></sub>; and then, calculated the average test error rates and their standard deviations over the 100 experiments. Table <xref ref-type="table" rid="T1">1</xref> lists the best results, i.e., the smallest average test error rate, of different algorithms. It can be seen that the proposed KerNN algorithm reaches the best, which are in bold face, on four data sets. On the other data sets, the performance of the KerNN algorithm is still competitive, if not better, to those of the SVM and ULDA schemes.</p><p>In Table <xref ref-type="table" rid="T1">1</xref>, if we assign a score 1 to the best result, 2 to the next best result, ..., and so on, then, the global performance of a classifier can be roughly evaluated in terms of the average score. We show the average scores of the five classifiers in Table <xref ref-type="table" rid="T1">1</xref>. It can be seen that the proposed KerNN scheme achieves the lowest score among the five classifiers.</p></sec><sec><title>Comparisons under different gene numbers</title><p>To investigate the stability of the 5 classification algorithms, we compared their performances when different number of genes were selected. The experimental results are shown in Fig. <xref ref-type="fig" rid="F1">1</xref>, for the <italic>ALL-AML </italic>data, Fig. <xref ref-type="fig" rid="F2">2</xref>, for the <italic>Colon </italic>data, and Fig. <xref ref-type="fig" rid="F3">3</xref>, for the <italic>Prostate </italic>data, where the horizontal axis is the number of the selected genes and the vertical axis is the average test error rates of the classifiers over 100 experiments. While Fig. <xref ref-type="fig" rid="F1">1 (a)</xref>, Fig. <xref ref-type="fig" rid="F2">2 (a)</xref>, and Fig. <xref ref-type="fig" rid="F3">3 (a)</xref> illustrate the results in the case of choosing a relatively small number of features (from 10 to 100), Fig. <xref ref-type="fig" rid="F1">1 (b)</xref>, Fig. <xref ref-type="fig" rid="F2">2 (b)</xref>, and Fig. <xref ref-type="fig" rid="F3">3 (b)</xref> demonstrate the corresponding results when more genes are chosen (from 200 to 2000). It can be seen that the proposed KerNN scheme performs favorably in most cases. Compared with the ULDA scheme, which always performs poorly in the case of small feature size, and the DLDA algorithm, whose performances usually degrade for relatively large feature size, our KerNN algorithm works with more stability with different feature numbers. More importantly, compared with the ordinary KNN classifier, the kernel optimization-based KNN classifier always gains significant improvements, which implies that the procedure of kernel optimization induces a distance metric that adapts better than the Euclidean metric to the gene expression data in the data space.</p></sec></sec><sec><title>Discussion</title><sec><title>Parameter tuning</title><p>In the experiments, for KNN, ULDA, and the proposed algorithm, the final classification is done via the K-nearest-neighbor algorithm with K = 3. For KNN, ULDA, and DLDA algorithms, the only parameter is the number of selected genes <italic>N</italic><sub><italic>f</italic></sub>. For SVM, in addition to the gene number, two parameters, the <italic>γ </italic>in the Gaussian kernel function and the regulation constant <italic>C</italic>, need to be set in advance. As for the KerNN algorithm, there are more parameters. To avoid the intensive computation in parameter tuning using the cross validation, we respectively chose the <italic>N</italic><sub><italic>f </italic></sub>most discriminatory genes, where <italic>N</italic><sub><italic>f </italic></sub>= 10, 20, 40, 60, 80, 100, 200, 400, 600, 800, 1000, 1200, 1400, 1600, 1800, 2000. The best performance for each method is reported in Table <xref ref-type="table" rid="T1">1</xref>. For our kernel optimization method, the initial learning rate <italic>η</italic><sub>0 </sub>and the total iteration number <italic>N </italic>are always set to 0.01 and 1000 respectively. Furthermore, for the sake of computational simplicity, we empirically set the two Gaussian parameters in the proposed method as <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M1" name="1471-2105-7-299-i1" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mi>γ</mml:mi><mml:mrow><mml:mn>0</mml:mn><mml:mo>=</mml:mo></mml:mrow></mml:msub><mml:mfrac><mml:mrow><mml:msup><mml:mrow><mml:mn>10</mml:mn></mml:mrow><mml:mrow><mml:mo>−</mml:mo><mml:mn>5</mml:mn></mml:mrow></mml:msup></mml:mrow><mml:mrow><mml:msqrt><mml:mrow><mml:msub><mml:mi>N</mml:mi><mml:mi>f</mml:mi></mml:msub></mml:mrow></mml:msqrt></mml:mrow></mml:mfrac></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFZoWzdaWgaaWcbaGaeGimaaJaeyypa0dabeaakmaalaaabaGaeGymaeJaeGimaaZaaWbaaSqabeaacqGHsislcqaI1aqnaaaakeaadaGcaaqaaiabd6eaonaaBaaaleaacqWGMbGzaeqaaaqabaaaaaaa@3744@</mml:annotation></mml:semantics></mml:math></inline-formula> and <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M2" name="1471-2105-7-299-i2" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mi>γ</mml:mi><mml:mn>1</mml:mn></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:msup><mml:mrow><mml:mn>10</mml:mn></mml:mrow><mml:mrow><mml:mo>−</mml:mo><mml:mn>2</mml:mn></mml:mrow></mml:msup></mml:mrow><mml:mrow><mml:msqrt><mml:mrow><mml:msub><mml:mi>N</mml:mi><mml:mi>f</mml:mi></mml:msub></mml:mrow></mml:msqrt></mml:mrow></mml:mfrac></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFZoWzdaWgaaWcbaGaeGymaedabeaakiabg2da9maalaaabaGaeGymaeJaeGimaaZaaWbaaSqabeaacqGHsislcqaIYaGmaaaakeaadaGcaaqaaiabd6eaonaaBaaaleaacqWGMbGzaeqaaaqabaaaaaaa@3740@</mml:annotation></mml:semantics></mml:math></inline-formula> , rather than tune them by the cross validation. This may not be the optimal settings for the parameters <italic>γ</italic><sub>0 </sub>and <italic>γ</italic><sub>1</sub>. However, high computational complexity can be avoided. It is expected that even better results could be obtained if we were to choose them by the cross validation. Therefore, for the KerNN method, there is only one parameter <italic>σ</italic><sub><italic>ε</italic></sub>, the standard variance of the disturbance added to the data in Eq. (10), that need to be tuned. As to the SVM, two parameters are tuned by the cross validation.</p><p>In the experiments, we employed the leave-one-out technique on the training data to choose these parameters. We followed [<xref ref-type="bibr" rid="B14">14</xref>] to implement the SVM algorithm, in which the parameter <italic>C </italic>is chosen from {l.0e+00, l.0e+01, l.0e+02, l.0e+03, l.0e+04, l.0e+05, l.0e+06, l.0e+07} and <italic>γ </italic>from {l.0e-07, 5.0e-07, l.0e-06, 5.0e-06, l.0e-05, 5.0e-05, l.0e-04, 5.0e-04, l.0e-03, 5.0e-03, l.0e-02} using the leave-one-out cross validation. For our KerNN algorithm, the parameter <italic>σ</italic><sub><italic>ε </italic></sub>is selected from {0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0}. Note that only the training samples were used for setting parameters. Test samples are independent of this process.</p></sec><sec><title>Gene selection</title><p>In this paper, we employ the BW ratio used in [<xref ref-type="bibr" rid="B2">2</xref>,<xref ref-type="bibr" rid="B10">10</xref>] to select genes. This ratio is essentially a Fisher discriminant measure. Given a gene <italic>j</italic>, the ratio on gene <italic>j </italic>is calculated as</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M3" name="1471-2105-7-299-i3" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>g</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>j</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo>=</mml:mo> <mml:mfrac> <mml:mrow> <mml:mstyle displaystyle="true"> <mml:msubsup> <mml:mo>∑</mml:mo> <mml:mrow> <mml:mi>k</mml:mi> <mml:mo>=</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> <mml:mi>p</mml:mi> </mml:msubsup> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>k</mml:mi> </mml:msub> <mml:msup> <mml:mrow> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mover accent="true"> <mml:mi>x</mml:mi> <mml:mo>¯</mml:mo> </mml:mover> <mml:mi>k</mml:mi> </mml:msub> <mml:mo stretchy="false">(</mml:mo> <mml:mi>j</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo>−</mml:mo> <mml:mover accent="true"> <mml:mi>x</mml:mi> <mml:mo>¯</mml:mo> </mml:mover> <mml:mo stretchy="false">(</mml:mo> <mml:mi>j</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> <mml:mn>2</mml:mn> </mml:msup> </mml:mrow> </mml:mstyle> </mml:mrow> <mml:mrow> <mml:mstyle displaystyle="true"> <mml:msubsup> <mml:mo>∑</mml:mo> <mml:mrow> <mml:mi>k</mml:mi> <mml:mo>=</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> <mml:mi>p</mml:mi> </mml:msubsup> <mml:mrow> <mml:mstyle displaystyle="true"> <mml:msub> <mml:mo>∑</mml:mo> <mml:mrow> <mml:mi>i</mml:mi> <mml:mo>∈</mml:mo> <mml:msub> <mml:mi>C</mml:mi> <mml:mi>k</mml:mi> </mml:msub> </mml:mrow> </mml:msub> <mml:mrow> <mml:msup> <mml:mrow> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>x</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo stretchy="false">(</mml:mo> <mml:mi>j</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo>−</mml:mo> <mml:msub> <mml:mover accent="true"> <mml:mi>x</mml:mi> <mml:mo>¯</mml:mo> </mml:mover> <mml:mi>k</mml:mi> </mml:msub> <mml:mo stretchy="false">(</mml:mo> <mml:mi>j</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> <mml:mn>2</mml:mn> </mml:msup> </mml:mrow> </mml:mstyle> </mml:mrow> </mml:mstyle> </mml:mrow> </mml:mfrac> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@689B@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>where <italic>C</italic><sub><italic>k </italic></sub>denotes the index set of the <italic>k</italic>-th class (<italic>k </italic>= l,2,...,<italic>p</italic>), <italic>m</italic><sub><italic>k </italic></sub>is the number of samples in <italic>C</italic><sub><italic>k </italic></sub>(<inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M4" name="1471-2105-7-299-i4" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mstyle displaystyle="true"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>p</mml:mi></mml:msubsup><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>k</mml:mi></mml:msub></mml:mrow></mml:mstyle><mml:mo>=</mml:mo><mml:mi>m</mml:mi></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaadaaeWaqaaiabd2gaTnaaBaaaleaacqWGRbWAaeqaaaqaaiabdUgaRjabg2da9iabigdaXaqaaiabdchaWbqdcqGHris5aOGaeyypa0JaemyBa0gaaa@38C2@</mml:annotation></mml:semantics></mml:math></inline-formula>), <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M5" name="1471-2105-7-299-i5" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>x</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWG4baEgaqeaaaa@2E3D@</mml:annotation></mml:semantics></mml:math></inline-formula><sub><italic>k</italic></sub>(<italic>j</italic>) and <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M6" name="1471-2105-7-299-i5" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>x</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWG4baEgaqeaaaa@2E3D@</mml:annotation></mml:semantics></mml:math></inline-formula>(<italic>j</italic>) represent the average expression levels cross the <italic>k</italic>-th class and whole training samples on gene <italic>j</italic>, respectively.</p><p>Gene selection usually has a strong impact on the performances of various classifiers, due to the effect of correlation between genes. Our experiments show that the impact can be considered in two aspects: l)with different numbers of genes, the performance of a classifier could be remarkably different. For example, the ULDA method usually works quite well as a large number of genes is used, but performs poorly in the case of small gene number. Contrarily, the DLDA classifier often reaches its best performance at small number of features. 2) with different numbers of genes, the model parameters, especially for the nonlinear methods, need to be set differently to achieve better result.</p></sec><sec><title>The effect of the disturbed resampling</title><p>Due to the lack of enough training samples, the scheme of the kernel optimization-based classification may lead to an overfitting result in classifying gene expression data. To alleviate the possible overfitting, a strategy of disturbed resampling, as shown in Eq. (10), was adopted. In this section, we demonstrate that using this strategy, the overfiting could be effectively reduced.</p><p>In the case that there are relatively large number of samples, the kernel optimization-based KNN classifier without using the strategy of disturbed resampling, denoted by KerNN0, usually works well on both the training and test data. Fig. <xref ref-type="fig" rid="F4">4</xref> illustrates the performances of KNN, KerNN0, and KerNN on both the training and test data of the <italic>Prostate </italic>data set, which includes 102 samples. It can be seen that, compared with the KNN algorithm, both the KerNN0 and KerNN methods gain significant improvements, not only on the training data, but also on the test data. However, when the sample size is relatively small, the KerNNO algorithm may lead to serious overfitting. We choose the <italic>Breast-ER </italic>data set, which contains only 49 samples, to demonstrate our argument. Fig. <xref ref-type="fig" rid="F5">5 (a)</xref> shows the average error rates of KNN, KerNN0, and KerNN algorithms on the training data, and Fig. <xref ref-type="fig" rid="F5">5 (b)</xref> presents the corresponding results on the test data. It can be seen that, although KerNN0 works quite well on the training data, its performance degrades remarkably on the test data. On the contrary, for the KerNN scheme, no overfitting occurred.</p></sec></sec><sec><title>Conclusion</title><p>In this paper, a novel distance metric is developed and incorporated into a KNN scheme for cancer classification. This metric, derived from the procedure of a data-dependent kernel optimization, can substantially increase the class separability of the data in the feature space, and hence, lead to a significant improvement in the performance of the KNN classifier. Furthermore, in combination with a disturbed resampling strategy, the kernel optimization-based K-nearest-neighbor scheme can achieve competitive performance to the fine tuned SVM and the uncorrelated linear discriminant analysis (ULDA) scheme in classifying gene expression data. Experimental results show that the proposed scheme performs with more stability than the ULDA scheme, which works poorly in the case of small feature size, and the DLDA scheme, whose performance usually degrades in the case of a relatively large feature size.</p></sec><sec sec-type="methods"><title>Methods</title><sec><title>0.1 Data-dependent kernel model</title><p>In this paper, we employ a special kernel function model, which is called date-dependent kernel model, as the objective kernel to be optimized. Apparently, there is no benefit at all if we simply use the common kernel such as the Gaussian kernel or the polynomial kernel in the KNN scheme, since the distance ranking in the Hilbert space derived from the kernel function is the same as that in the input data space. However, when we adopt the data-dependent kernel, especially after the kernel is optimized, the distance metric could be appropriately modified so that the local relevance of the data is significantly improved.</p><p>Let {<italic>x</italic><sub><italic>i</italic></sub>, <italic>ζ</italic><sub><italic>i</italic></sub>} (<italic>i </italic>= 1,2, ..., <italic>m</italic>) be <italic>m </italic><italic>d</italic>-dimensional training samples of the given gene expression data, where <italic>ζ</italic><sub><italic>i </italic></sub>represent the class labels of the samples. We refer the data-dependent kernel as,</p><p><italic>k</italic>(<italic>x</italic>, <italic>y</italic>) = <italic>q</italic>(<italic>x</italic>)<italic>q</italic>(<italic>y</italic>)<italic>k</italic><sub>0</sub>(<italic>x</italic>, <italic>y</italic>)       (1)</p><p>where <italic>x</italic>, <italic>y </italic>∈ <bold>R</bold><sup><italic>d</italic></sup>, <italic>k</italic><sub>0</sub>(<italic>x</italic>, <italic>y</italic>), called the basic kernel, is an ordinary kernel such as a Gaussian or a polynomial kernel function, and <italic>q</italic>(.), the factor function, takes the form as</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M7" name="1471-2105-7-299-i6" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>q</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>x</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo>=</mml:mo> <mml:msub> <mml:mi>α</mml:mi> <mml:mn>0</mml:mn> </mml:msub> <mml:mo>+</mml:mo> <mml:mstyle displaystyle="true"> <mml:munderover> <mml:mo>∑</mml:mo> <mml:mrow> <mml:mi>i</mml:mi> <mml:mo>=</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> <mml:mi>l</mml:mi> </mml:munderover> <mml:mrow> <mml:msub> <mml:mi>α</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:msub> <mml:mi>k</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo stretchy="false">(</mml:mo> <mml:mi>x</mml:mi> <mml:mo>,</mml:mo> <mml:msub> <mml:mi>a</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:mstyle> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>2</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGXbqCcqGGOaakcqWG4baEcqGGPaqkcqGH9aqpiiGacqWFXoqydaWgaaWcbaGaeGimaadabeaakiabgUcaRmaaqahabaGae8xSde2aaSbaaSqaaiabdMgaPbqabaGccqWGRbWAdaWgaaWcbaGaeGymaedabeaakiabcIcaOiabdIha4jabcYcaSiabdggaHnaaBaaaleaacqWGPbqAaeqaaOGaeiykaKcaleaacqWGPbqAcqGH9aqpcqaIXaqmaeaacqWGSbaBa0GaeyyeIuoakiaaxMaacaWLjaWaaeWaaeaacqaIYaGmaiaawIcacaGLPaaaaaa@4D48@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>in which <italic>k</italic><sub>1</sub>(<italic>x</italic>, <italic>a</italic><sub><italic>i</italic></sub>) = <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M8" name="1471-2105-7-299-i7" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msup><mml:mi>e</mml:mi><mml:mrow><mml:mo>−</mml:mo><mml:msub><mml:mi>γ</mml:mi><mml:mn>1</mml:mn></mml:msub><mml:mo>\|</mml:mo><mml:mo>\|</mml:mo><mml:mi>x</mml:mi><mml:mo>−</mml:mo><mml:msub><mml:mi>a</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>\|</mml:mo><mml:msup><mml:mo>\|</mml:mo><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:msup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGLbqzdaahaaWcbeqaaiabgkHiTGGaciab=n7aNnaaBaaameaacqaIXaqmaeqaaSGaeiiFaWNaeiiFaWNaemiEaGNaeyOeI0Iaemyyae2aaSbaaWqaaiabdMgaPbqabaWccqGG8baFcqGG8baFdaahaaadbeqaaiabikdaYaaaaaaaaa@3E53@</mml:annotation></mml:semantics></mml:math></inline-formula>, <italic>α</italic><sub><italic>i</italic></sub>'s are the combination coefficients, and <italic>a</italic><sub><italic>i</italic></sub>'s denote the local centers of the training data.</p><p>Let the kernel matrices corresponding to <italic>k</italic>(<italic>x</italic>, <italic>y</italic>) and <italic>k</italic><sub>0</sub>(<italic>x</italic>, <italic>y</italic>) be <italic>K </italic>and <italic>k</italic><sub>0</sub>. Obviously, <italic>K </italic>= [<italic>q</italic>(<italic>x</italic><sub><italic>i</italic></sub>)<italic>q</italic>(<italic>x</italic><sub><italic>j</italic></sub>)<italic>k</italic><sub>0</sub>(<italic>x</italic><sub><italic>i</italic></sub>, <italic>x</italic><sub><italic>j</italic></sub>)]<sub><italic>m </italic>× <italic>m </italic></sub>= <italic>QK</italic><sub>0</sub><italic>Q</italic>, where <italic>Q </italic>is a diagonal matrix whose diagonal elements are <italic>q</italic>(<italic>x</italic><sub>1</sub>), <italic>q</italic>(<italic>x</italic><sub>2</sub>),...,<italic>q</italic>(<italic>x</italic><sub><italic>m</italic></sub>). Let us denote the vector (<italic>q</italic>(<italic>x</italic><sub>1</sub>), <italic>q</italic>(<italic>x</italic><sub>2</sub>),..., <italic>q</italic>(<italic>x</italic><sub><italic>m</italic></sub>))<sup><italic>T </italic></sup>and (<italic>α</italic><sub>0</sub>, <italic>α</italic><sub>1</sub>, <italic>α</italic><sub>2</sub>,...,<italic>α</italic><sub><italic>n</italic></sub>)<sup><italic>T </italic></sup>by <italic>q </italic>and <italic>α </italic>respectively, we have <italic>q </italic>= <italic>K</italic><sub>1</sub><italic>α</italic>, where <italic>K</italic><sub>1 </sub>is an <italic>m </italic>× (<italic>l </italic>+ 1) matrix</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M9" name="1471-2105-7-299-i8" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo>=</mml:mo> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mtable> <mml:mtr> <mml:mtd> <mml:mn>1</mml:mn> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>k</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>x</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo>,</mml:mo> <mml:msub> <mml:mi>a</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mo>⋯</mml:mo> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>k</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>x</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo>,</mml:mo> <mml:msub> <mml:mi>a</mml:mi> <mml:mi>l</mml:mi> </mml:msub> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:mtd> </mml:mtr> <mml:mtr> <mml:mtd> <mml:mn>1</mml:mn> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>k</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>x</mml:mi> <mml:mn>2</mml:mn> </mml:msub> <mml:mo>,</mml:mo> <mml:msub> <mml:mi>a</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mo>⋯</mml:mo> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>k</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>x</mml:mi> <mml:mn>2</mml:mn> </mml:msub> <mml:mo>,</mml:mo> <mml:msub> <mml:mi>a</mml:mi> <mml:mi>l</mml:mi> </mml:msub> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:mtd> </mml:mtr> <mml:mtr> <mml:mtd> <mml:mo>⋮</mml:mo> </mml:mtd> <mml:mtd> <mml:mo>⋮</mml:mo> </mml:mtd> <mml:mtd> <mml:mo>⋱</mml:mo> </mml:mtd> <mml:mtd> <mml:mo>⋮</mml:mo> </mml:mtd> </mml:mtr> <mml:mtr> <mml:mtd> <mml:mn>1</mml:mn> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>k</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>x</mml:mi> <mml:mi>m</mml:mi> </mml:msub> <mml:mo>,</mml:mo> <mml:msub> <mml:mi>a</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mo>⋯</mml:mo> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>k</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>x</mml:mi> <mml:mi>m</mml:mi> </mml:msub> <mml:mo>,</mml:mo> <mml:msub> <mml:mi>a</mml:mi> <mml:mi>l</mml:mi> </mml:msub> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:mtd> </mml:mtr> </mml:mtable> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>3</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@84DF@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p></sec><sec><title>0.2 Kernel optimization for binary-class data</title><p>We optimized the data-dependent kernel in Eq.(l). This requires optimizing the combination coefficient vector <italic>α</italic>, aiming to increase the class separability of the data in the feature space. A Fisher scalar measuring the class separability of the training data in the feature space is adopted as a criterion for our kernel optimization</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M10" name="1471-2105-7-299-i9" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>J</mml:mi> <mml:mo>=</mml:mo> <mml:mfrac> <mml:mrow> <mml:mtext>tr</mml:mtext> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>S</mml:mi> <mml:mi>b</mml:mi> </mml:msub> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> <mml:mrow> <mml:mtext>tr</mml:mtext> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>S</mml:mi> <mml:mi>w</mml:mi> </mml:msub> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:mfrac> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>4</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGkbGscqGH9aqpdaWcaaqaaiabbsha0jabbkhaYjabcIcaOiabdofatnaaBaaaleaacqWGIbGyaeqaaOGaeiykaKcabaGaeeiDaqNaeeOCaiNaeiikaGIaem4uam1aaSbaaSqaaiabdEha3bqabaGccqGGPaqkaaGaaCzcaiaaxMaadaqadaqaaiabisda0aGaayjkaiaawMcaaaaa@4148@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>where <italic>S</italic><sub><italic>b </italic></sub>represents the "between-class scatter matrix", and <italic>S</italic><sub><italic>w </italic></sub>"within-class scatter matrix".</p><p>Suppose that the training data are grouped according to their class labels, i.e., the first <italic>m</italic><sub>1 </sub>data belong to one class, and the remaining <italic>m</italic><sub>2 </sub>data belong to the other class (<italic>m</italic><sub>1 </sub>+ <italic>m</italic><sub>2 </sub>= <italic>m</italic>). Then the basic kernel matrix <italic>k</italic><sub>0 </sub>can be partitioned as</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M11" name="1471-2105-7-299-i10" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mn>0</mml:mn> </mml:msub> <mml:mo>=</mml:mo> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mtable> <mml:mtr> <mml:mtd> <mml:mrow> <mml:msubsup> <mml:mi>K</mml:mi> <mml:mrow> <mml:mn>11</mml:mn> </mml:mrow> <mml:mn>0</mml:mn> </mml:msubsup> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msubsup> <mml:mi>K</mml:mi> <mml:mrow> <mml:mn>12</mml:mn> </mml:mrow> <mml:mn>0</mml:mn> </mml:msubsup> </mml:mrow> </mml:mtd> </mml:mtr> <mml:mtr> <mml:mtd> <mml:mrow> <mml:msubsup> <mml:mi>K</mml:mi> <mml:mrow> <mml:mn>21</mml:mn> </mml:mrow> <mml:mn>0</mml:mn> </mml:msubsup> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msubsup> <mml:mi>K</mml:mi> <mml:mrow> <mml:mn>22</mml:mn> </mml:mrow> <mml:mn>0</mml:mn> </mml:msubsup> </mml:mrow> </mml:mtd> </mml:mtr> </mml:mtable> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>5</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGlbWsdaWgaaWcbaGaeGimaadabeaakiabg2da9maabmaabaqbaeqabiGaaaqaaiabdUealnaaDaaaleaacqaIXaqmcqaIXaqmaeaacqaIWaamaaaakeaacqWGlbWsdaqhaaWcbaGaeGymaeJaeGOmaidabaGaeGimaadaaaGcbaGaem4saS0aa0baaSqaaiabikdaYiabigdaXaqaaiabicdaWaaaaOqaaiabdUealnaaDaaaleaacqaIYaGmcqaIYaGmaeaacqaIWaamaaaaaaGccaGLOaGaayzkaaGaaCzcaiaaxMaadaqadaqaaiabiwda1aGaayjkaiaawMcaaaaa@45EB@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>where the sizes of the submatrices <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M12" name="1471-2105-7-299-i11" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>K</mml:mi><mml:mrow><mml:mn>11</mml:mn></mml:mrow><mml:mn>0</mml:mn></mml:msubsup><mml:mo>,</mml:mo><mml:msubsup><mml:mi>K</mml:mi><mml:mrow><mml:mn>12</mml:mn></mml:mrow><mml:mn>0</mml:mn></mml:msubsup><mml:mo>,</mml:mo><mml:msubsup><mml:mi>K</mml:mi><mml:mrow><mml:mn>21</mml:mn></mml:mrow><mml:mn>0</mml:mn></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGlbWsdaqhaaWcbaGaeGymaeJaeGymaedabaGaeGimaadaaOGaeiilaWIaem4saS0aa0baaSqaaiabigdaXiabikdaYaqaaiabicdaWaaakiabcYcaSiabdUealnaaDaaaleaacqaIYaGmcqaIXaqmaeaacqaIWaamaaaaaa@3AD2@</mml:annotation></mml:semantics></mml:math></inline-formula>, and <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M13" name="1471-2105-7-299-i12" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>K</mml:mi><mml:mrow><mml:mn>22</mml:mn></mml:mrow><mml:mn>0</mml:mn></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGlbWsdaqhaaWcbaGaeGOmaiJaeGOmaidabaGaeGimaadaaaaa@30CA@</mml:annotation></mml:semantics></mml:math></inline-formula> respectively are <italic>m</italic><sub>1 </sub>× <italic>m</italic><sub>1</sub>, <italic>m</italic><sub>1 </sub>× <italic>m</italic><sub>2</sub>, <italic>m</italic><sub>2 </sub>× <italic>m</italic><sub>1</sub>, and <italic>m</italic><sub>2 </sub>× <italic>m</italic><sub>2</sub>. A close relation between the class separability measure <italic>J </italic>and the kernel matrices can be established [<xref ref-type="bibr" rid="B27">27</xref>].</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M14" name="1471-2105-7-299-i13" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>J</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>α</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo>=</mml:mo> <mml:mfrac> <mml:mrow> <mml:msup> <mml:mi>α</mml:mi> <mml:mi>T</mml:mi> </mml:msup> <mml:msub> <mml:mi>M</mml:mi> <mml:mn>0</mml:mn> </mml:msub> <mml:mi>α</mml:mi> </mml:mrow> <mml:mrow> <mml:msup> <mml:mi>α</mml:mi> <mml:mi>T</mml:mi> </mml:msup> <mml:msub> <mml:mi>N</mml:mi> <mml:mn>0</mml:mn> </mml:msub> <mml:mi>α</mml:mi> </mml:mrow> </mml:mfrac> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>6</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGkbGscqGGOaakiiGacqWFXoqycqGGPaqkcqGH9aqpdaWcaaqaaiab=f7aHnaaCaaaleqabaGaemivaqfaaOGaemyta00aaSbaaSqaaiabicdaWaqabaGccqWFXoqyaeaacqWFXoqydaahaaWcbeqaaiabdsfaubaakiabd6eaonaaBaaaleaacqaIWaamaeqaaOGae8xSdegaaiaaxMaacaWLjaWaaeWaaeaacqaI2aGnaiaawIcacaGLPaaaaaa@43C6@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>where <italic>M</italic><sub>0 </sub>= <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M15" name="1471-2105-7-299-i14" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>K</mml:mi><mml:mn>1</mml:mn><mml:mi>T</mml:mi></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGlbWsdaqhaaWcbaGaeGymaedabaGaemivaqfaaaaa@3019@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>B</italic><sub>0</sub><italic>K</italic><sub>1</sub>, <italic>N</italic><sub>0 </sub>= <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M16" name="1471-2105-7-299-i14" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>K</mml:mi><mml:mn>1</mml:mn><mml:mi>T</mml:mi></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGlbWsdaqhaaWcbaGaeGymaedabaGaemivaqfaaaaa@3019@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>W</italic><sub>0</sub><italic>K</italic><sub>1</sub>, in which</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M17" name="1471-2105-7-299-i15" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:msub> <mml:mi>B</mml:mi> <mml:mn>0</mml:mn> </mml:msub> <mml:mo>=</mml:mo> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mtable> <mml:mtr> <mml:mtd> <mml:mrow> <mml:mfrac> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mn>1</mml:mn> </mml:msub> </mml:mrow> </mml:mfrac> <mml:msubsup> <mml:mi>K</mml:mi> <mml:mrow> <mml:mn>11</mml:mn> </mml:mrow> <mml:mn>0</mml:mn> </mml:msubsup> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mn>0</mml:mn> </mml:mtd> </mml:mtr> <mml:mtr> <mml:mtd> <mml:mn>0</mml:mn> </mml:mtd> <mml:mtd> <mml:mrow> <mml:mfrac> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mn>2</mml:mn> </mml:msub> </mml:mrow> </mml:mfrac> <mml:msubsup> <mml:mi>K</mml:mi> <mml:mrow> <mml:mn>22</mml:mn> </mml:mrow> <mml:mn>0</mml:mn> </mml:msubsup> </mml:mrow> </mml:mtd> </mml:mtr> </mml:mtable> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>−</mml:mo> <mml:mfrac> <mml:mn>1</mml:mn> <mml:mi>m</mml:mi> </mml:mfrac> <mml:msub> <mml:mi>K</mml:mi> <mml:mn>0</mml:mn> </mml:msub> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGcbGqdaWgaaWcbaGaeGimaadabeaakiabg2da9maabmaabaqbaeqabiGaaaqaamaalaaabaGaeGymaedabaGaemyBa02aaSbaaSqaaiabigdaXaqabaaaaOGaem4saS0aa0baaSqaaiabigdaXiabigdaXaqaaiabicdaWaaaaOqaaiabicdaWaqaaiabicdaWaqaamaalaaabaGaeGymaedabaGaemyBa02aaSbaaSqaaiabikdaYaqabaaaaOGaem4saS0aa0baaSqaaiabikdaYiabikdaYaqaaiabicdaWaaaaaaakiaawIcacaGLPaaacqGHsisldaWcaaqaaiabigdaXaqaaiabd2gaTbaacqWGlbWsdaWgaaWcbaGaeGimaadabeaaaaa@4841@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M18" name="1471-2105-7-299-i16" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:msub> <mml:mi>W</mml:mi> <mml:mn>0</mml:mn> </mml:msub> <mml:mo>=</mml:mo> <mml:mtext>diag</mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:msubsup> <mml:mi>k</mml:mi> <mml:mrow> <mml:mn>11</mml:mn> </mml:mrow> <mml:mn>0</mml:mn> </mml:msubsup> <mml:mo>,</mml:mo> <mml:msubsup> <mml:mi>k</mml:mi> <mml:mrow> <mml:mn>22</mml:mn> </mml:mrow> <mml:mn>0</mml:mn> </mml:msubsup> <mml:mo>,</mml:mo> <mml:mo>…</mml:mo> <mml:mo>,</mml:mo> <mml:msubsup> <mml:mi>k</mml:mi> <mml:mrow> <mml:mi>m</mml:mi> <mml:mi>m</mml:mi> </mml:mrow> <mml:mn>0</mml:mn> </mml:msubsup> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>−</mml:mo> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mtable> <mml:mtr> <mml:mtd> <mml:mrow> <mml:mfrac> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mn>1</mml:mn> </mml:msub> </mml:mrow> </mml:mfrac> <mml:msubsup> <mml:mi>K</mml:mi> <mml:mrow> <mml:mn>11</mml:mn> </mml:mrow> <mml:mn>0</mml:mn> </mml:msubsup> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mn>0</mml:mn> </mml:mtd> </mml:mtr> <mml:mtr> <mml:mtd> <mml:mn>0</mml:mn> </mml:mtd> <mml:mtd> <mml:mrow> <mml:mfrac> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mn>2</mml:mn> </mml:msub> </mml:mrow> </mml:mfrac> <mml:msubsup> <mml:mi>K</mml:mi> <mml:mrow> <mml:mn>22</mml:mn> </mml:mrow> <mml:mn>0</mml:mn> </mml:msubsup> </mml:mrow> </mml:mtd> </mml:mtr> </mml:mtable> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@5C76@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>To avoid using the eigenvector solution, an updating algorithm based on the standard gradient approach is developed. This algorithm is summarized below, in which the learning rate <italic>η</italic>(<italic>n</italic>) is adopted in a gradually decreasing form as</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M19" name="1471-2105-7-299-i17" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>η</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo>=</mml:mo> <mml:msub> <mml:mi>η</mml:mi> <mml:mn>0</mml:mn> </mml:msub> <mml:mo stretchy="false">(</mml:mo> <mml:mn>1</mml:mn> <mml:mo>−</mml:mo> <mml:mfrac> <mml:mi>n</mml:mi> <mml:mi>N</mml:mi> </mml:mfrac> <mml:mo stretchy="false">)</mml:mo> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>7</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWF3oaAcqGGOaakcqWGUbGBcqGGPaqkcqGH9aqpcqWF3oaAdaWgaaWcbaGaeGimaadabeaakiabcIcaOiabigdaXiabgkHiTmaalaaabaGaemOBa4gabaGaemOta4eaaiabcMcaPiaaxMaacaWLjaWaaeWaaeaacqaI3aWnaiaawIcacaGLPaaaaaa@3F39@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>where <italic>η</italic><sub>0 </sub>represents an initial learning rate.</p><p>1. Group the data according to their class labels. Calculate <italic>K</italic><sub>0 </sub>and <italic>K</italic><sub>1 </sub>first, then <italic>B</italic><sub>0 </sub>and <italic>W</italic><sub>0</sub>, and then <italic>M</italic><sub>0</sub>, <italic>N</italic><sub>0</sub>.</p><p>2. Initialize <italic>α</italic><sup>(0) </sup>by a vector (1,0,..., 0)<sup><italic>T</italic></sup>, and set <italic>n </italic>= 0.</p><p>3. Calculate <italic>q </italic>= <italic>K</italic><sub>1</sub><italic>α</italic><sup>(<italic>n</italic>)</sup>, and <italic>J</italic><sub>1 </sub>= <italic>q</italic><sup><italic>T </italic></sup><italic>B</italic><sub>0</sub><italic>q</italic>, <italic>J</italic><sub>2 </sub>= <italic>q</italic><sup><italic>T </italic></sup><italic>W</italic><sub>0</sub><italic>q</italic>, and <italic>J</italic>.</p><p>4. Update <italic>α</italic><sup>(<italic>n</italic>)</sup>:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M20" name="1471-2105-7-299-i18" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:msup> <mml:mi>α</mml:mi> <mml:mrow> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo>+</mml:mo> <mml:mn>1</mml:mn> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:msup> <mml:mo>=</mml:mo> <mml:msup> <mml:mi>α</mml:mi> <mml:mrow> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:msup> <mml:mo>+</mml:mo> <mml:mi>η</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo stretchy="false">(</mml:mo> <mml:mfrac> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>J</mml:mi> <mml:mn>2</mml:mn> </mml:msub> </mml:mrow> </mml:mfrac> <mml:msub> <mml:mi>M</mml:mi> <mml:mn>0</mml:mn> </mml:msub> <mml:mo>−</mml:mo> <mml:mfrac> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>J</mml:mi> <mml:mn>2</mml:mn> </mml:msub> </mml:mrow> </mml:mfrac> <mml:msub> <mml:mi>N</mml:mi> <mml:mn>0</mml:mn> </mml:msub> <mml:mo stretchy="false">)</mml:mo> <mml:msup> <mml:mi>α</mml:mi> <mml:mrow> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:msup> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFXoqydaahaaWcbeqaaiabcIcaOiabd6gaUjabgUcaRiabigdaXiabcMcaPaaakiabg2da9iab=f7aHnaaCaaaleqabaGaeiikaGIaemOBa4MaeiykaKcaaOGaey4kaSIae83TdGMaeiikaGIaemOBa4MaeiykaKIaeiikaGYaaSaaaeaacqaIXaqmaeaacqWGkbGsdaWgaaWcbaGaeGOmaidabeaaaaGccqWGnbqtdaWgaaWcbaGaeGimaadabeaakiabgkHiTmaalaaabaGaeGymaedabaGaemOsaO0aaSbaaSqaaiabikdaYaqabaaaaOGaemOta40aaSbaaSqaaiabicdaWaqabaGccqGGPaqkcqWFXoqydaahaaWcbeqaaiabcIcaOiabd6gaUjabcMcaPaaaaaa@5197@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>and normalize <italic>α</italic><sup>(<italic>n</italic>+1) </sup>so that \|\|<italic>α</italic><sup>(<italic>n</italic>+1)</sup>\|\| = 1.</p><p>5. If <italic>n </italic>reaches a pre-specified number <italic>N</italic>, stop. Otherwise, set <italic>n </italic>= <italic>n </italic>+ 1, go to 3.</p></sec><sec><title>0.3 Kernel optimization for multi-class data</title><p>In the case of multi-class data, we decompose the problem of kernel optimization into a series of binary-class kernel optimizations.</p><p>Let (<italic>x</italic><sub><italic>i</italic></sub>, <italic>ζ</italic><sub><italic>i</italic></sub>) ∈ <bold>R</bold><sup><italic>d </italic></sup>× <italic>ζ </italic>(<italic>i </italic>= 1, 2,..., <italic>m</italic>) be the training data set containing <italic>p </italic>classes, that is, <italic>ζ </italic>= {1,2,...,<italic>p</italic>}. We assume the data to be grouped in order, that is, the first <italic>m</italic><sub>1 </sub>data belong to the first class, the next <italic>m</italic><sub>2 </sub>data belong to the second class, and so on, where <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M21" name="1471-2105-7-299-i19" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mstyle displaystyle="true"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>p</mml:mi></mml:msubsup><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:mstyle><mml:mo>=</mml:mo><mml:mi>m</mml:mi></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaadaaeWaqaaiabd2gaTnaaBaaaleaacqWGPbqAaeqaaaqaaiabdMgaPjabg2da9iabigdaXaqaaiabdchaWbqdcqGHris5aOGaeyypa0JaemyBa0gaaa@38BA@</mml:annotation></mml:semantics></mml:math></inline-formula>. Then, the kernel matrix can be written as</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M22" name="1471-2105-7-299-i20" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>K</mml:mi> <mml:mo>=</mml:mo> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mtable> <mml:mtr> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mn>11</mml:mn> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mn>12</mml:mn> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mo>⋯</mml:mo> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mn>1</mml:mn> <mml:mi>p</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> </mml:mtr> <mml:mtr> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mn>21</mml:mn> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mn>22</mml:mn> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mo>⋯</mml:mo> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mn>2</mml:mn> <mml:mi>p</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> </mml:mtr> <mml:mtr> <mml:mtd> <mml:mo>⋮</mml:mo> </mml:mtd> <mml:mtd> <mml:mo>⋮</mml:mo> </mml:mtd> <mml:mtd> <mml:mo>⋱</mml:mo> </mml:mtd> <mml:mtd> <mml:mo>⋮</mml:mo> </mml:mtd> </mml:mtr> <mml:mtr> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mi>p</mml:mi> <mml:mn>1</mml:mn> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mi>p</mml:mi> <mml:mn>2</mml:mn> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mo>⋯</mml:mo> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mi>p</mml:mi> <mml:mi>p</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> </mml:mtr> </mml:mtable> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>8</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@618B@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>where the submatrix <italic>k</italic><sub><italic>ij </italic></sub>is of size <italic>m</italic><sub><italic>i </italic></sub>× <italic>m</italic><sub><italic>j</italic></sub>, and <italic>K</italic><sub><italic>ii </italic></sub>represents the kernel matrix corresponding to the data in the <italic>i</italic>-th class. The class separability of the <italic>i</italic>-th and <italic>j</italic>-th class, denoted by <italic>J</italic><sup><italic>ij </italic></sup>(<italic>i,j </italic>= 1, 2,...,<italic>p</italic>, <italic>i </italic>≠ <italic>j</italic>), is calculated as</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M23" name="1471-2105-7-299-i21" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:msup> <mml:mi>J</mml:mi> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>j</mml:mi> </mml:mrow> </mml:msup> <mml:mo stretchy="false">(</mml:mo> <mml:mi>α</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo>=</mml:mo> <mml:mfrac> <mml:mrow> <mml:msubsup> <mml:mi>J</mml:mi> <mml:mi>i</mml:mi> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>j</mml:mi> </mml:mrow> </mml:msubsup> </mml:mrow> <mml:mrow> <mml:msubsup> <mml:mi>J</mml:mi> <mml:mn>2</mml:mn> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>j</mml:mi> </mml:mrow> </mml:msubsup> </mml:mrow> </mml:mfrac> <mml:mo>=</mml:mo> <mml:mfrac> <mml:mrow> <mml:msubsup> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo>+</mml:mo> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:mrow> <mml:mi>T</mml:mi> </mml:msubsup> <mml:msup> <mml:mi>B</mml:mi> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>j</mml:mi> </mml:mrow> </mml:msup> <mml:msub> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo>+</mml:mo> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:mrow> </mml:msub> </mml:mrow> <mml:mrow> <mml:msubsup> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo>+</mml:mo> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:mrow> <mml:mi>T</mml:mi> </mml:msubsup> <mml:msup> <mml:mi>W</mml:mi> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>j</mml:mi> </mml:mrow> </mml:msup> <mml:msub> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo>+</mml:mo> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:mrow> </mml:msub> </mml:mrow> </mml:mfrac> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>9</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGkbGsdaahaaWcbeqaaiabdMgaPjabdQgaQbaakiabcIcaOGGaciab=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@6ECC@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>where the between-class and within-class kernel scatter matrices <italic>B</italic><sup><italic>ij </italic></sup>and <italic>W</italic><sup><italic>ij </italic></sup>are defined as</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M24" name="1471-2105-7-299-i22" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:msup> <mml:mi>B</mml:mi> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>j</mml:mi> </mml:mrow> </mml:msup> <mml:mo>=</mml:mo> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mtable> <mml:mtr> <mml:mtd> <mml:mrow> <mml:mfrac> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> </mml:mrow> </mml:mfrac> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>i</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mn>0</mml:mn> </mml:mtd> </mml:mtr> <mml:mtr> <mml:mtd> <mml:mn>0</mml:mn> </mml:mtd> <mml:mtd> <mml:mrow> <mml:mfrac> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:mrow> </mml:mfrac> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mi>j</mml:mi> <mml:mi>j</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> </mml:mtr> </mml:mtable> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>−</mml:mo> <mml:mfrac> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo>+</mml:mo> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:mrow> </mml:mfrac> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mtable> <mml:mtr> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>i</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>j</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> </mml:mtr> <mml:mtr> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mi>j</mml:mi> <mml:mi>i</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mrow> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mi>j</mml:mi> <mml:mi>j</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> </mml:mtr> </mml:mtable> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@5FAD@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M25" name="1471-2105-7-299-i23" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:msup> <mml:mi>W</mml:mi> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>j</mml:mi> </mml:mrow> </mml:msup> <mml:mo>=</mml:mo> <mml:msup> <mml:mi>D</mml:mi> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>j</mml:mi> </mml:mrow> </mml:msup> <mml:mo>−</mml:mo> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mtable> <mml:mtr> <mml:mtd> <mml:mrow> <mml:mfrac> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> </mml:mrow> </mml:mfrac> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>i</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> <mml:mtd> <mml:mn>0</mml:mn> </mml:mtd> </mml:mtr> <mml:mtr> <mml:mtd> <mml:mn>0</mml:mn> </mml:mtd> <mml:mtd> <mml:mrow> <mml:mfrac> <mml:mn>1</mml:mn> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:mrow> </mml:mfrac> <mml:msub> <mml:mi>K</mml:mi> <mml:mrow> <mml:mi>j</mml:mi> <mml:mi>j</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:mtd> </mml:mtr> </mml:mtable> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGxbWvdaahaaWcbeqaaiabdMgaPjabdQgaQbaakiabg2da9iabdseaenaaCaaaleqabaGaemyAaKMaemOAaOgaaOGaeyOeI0YaaeWaaeaafaqabeGacaaabaWaaSaaaeaacqaIXaqmaeaacqWGTbqBdaWgaaWcbaGaemyAaKgabeaaaaGccqWGlbWsdaWgaaWcbaGaemyAaKMaemyAaKgabeaaaOqaaiabicdaWaqaaiabicdaWaqaamaalaaabaGaeGymaedabaGaemyBa02aaSbaaSqaaiabdQgaQbqabaaaaOGaem4saS0aaSbaaSqaaiabdQgaQjabdQgaQbqabaaaaaGccaGLOaGaayzkaaaaaa@4A3E@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>in which <italic>D</italic><sup><italic>ij </italic></sup>denotes a diagonal matrix whose diagonal elements are composed of the diagonal entries of the matrix <italic>K</italic><sub><italic>ii </italic></sub>and <italic>K</italic><sub><italic>jj</italic></sub>. We also denote the between-class and within-class kernel matrices corresponding to the basic kernel by <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M26" name="1471-2105-7-299-i24" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>B</mml:mi><mml:mn>0</mml:mn><mml:mrow><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGcbGqdaqhaaWcbaGaeGimaadabaGaemyAaKMaemOAaOgaaaaa@318C@</mml:annotation></mml:semantics></mml:math></inline-formula> and <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M27" name="1471-2105-7-299-i25" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>W</mml:mi><mml:mn>0</mml:mn><mml:mrow><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGxbWvdaqhaaWcbaGaeGimaadabaGaemyAaKMaemOAaOgaaaaa@31B6@</mml:annotation></mml:semantics></mml:math></inline-formula> respectively.</p><p>In each iteration of the updating algorithm, we first find the class index (<italic>u, v</italic>) that corresponds to the minimum <italic>J</italic><sup><italic>ij </italic></sup>in current step, then the value of <italic>α </italic>is updated in such a way that the class separability of the <italic>u</italic>-th and <italic>v</italic>-th class <italic>J</italic><sup><italic>uv </italic></sup>will be maximized. In other words, the objective of the kernel optimization becomes</p><p><inline-graphic xlink:href="1471-2105-7-299-i26.gif"/></p><p>It is easy to modify the kernel optimization algorithm from the case of binary class data to the case of multi-class data. The detailed kernel optimization algorithm for multi-class data set is summarized below, where Γ<sub><italic>ij </italic></sub>denotes the union of the data index sets of the <italic>i</italic>-th and <italic>j</italic>-th class, and <italic>q</italic>(Γ<sub><italic>ij</italic></sub>) and <italic>K</italic><sub>1</sub>(Γ<sub><italic>ij</italic></sub>,:) represent the submatrix extraction as in MATLAB.</p><p>1. Group the data according to their class labels. Calculate <italic>k</italic><sub>0 </sub>and <italic>K</italic><sub>1</sub>.</p><p>2. Initialize <italic>α</italic><sup>(0) </sup>by a vector (1,0,..., 0)<sup><italic>T</italic></sup>, and set <italic>n </italic>= 0.</p><p>3. Calculate <italic>q </italic>= <italic>K</italic><sub>1</sub><italic>α</italic><sup>(<italic>n</italic>)</sup>, <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M28" name="1471-2105-7-299-i27" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>J</mml:mi><mml:mn>1</mml:mn><mml:mrow><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGkbGsdaqhaaWcbaGaeGymaedabaGaemyAaKMaemOAaOgaaaaa@319E@</mml:annotation></mml:semantics></mml:math></inline-formula> = <italic>q</italic>(Γ<sub><italic>ij</italic></sub>)<sup><italic>T </italic></sup><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M29" name="1471-2105-7-299-i24" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>B</mml:mi><mml:mn>0</mml:mn><mml:mrow><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGcbGqdaqhaaWcbaGaeGimaadabaGaemyAaKMaemOAaOgaaaaa@318C@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>q</italic>(Γ<sub><italic>ij</italic></sub>), <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M30" name="1471-2105-7-299-i28" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>J</mml:mi><mml:mn>2</mml:mn><mml:mrow><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGkbGsdaqhaaWcbaGaeGOmaidabaGaemyAaKMaemOAaOgaaaaa@31A0@</mml:annotation></mml:semantics></mml:math></inline-formula> = <italic>q</italic>(Γ<sub><italic>ij</italic></sub>)<sup><italic>T </italic></sup><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M31" name="1471-2105-7-299-i25" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>W</mml:mi><mml:mn>0</mml:mn><mml:mrow><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGxbWvdaqhaaWcbaGaeGimaadabaGaemyAaKMaemOAaOgaaaaa@31B6@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>q</italic>(Γ<sub><italic>ij</italic></sub>), and <italic>J</italic><sup><italic>ij</italic></sup>, where <italic>i</italic>, <italic>j </italic>= l,2,...,<italic>p</italic>, and <italic>i </italic>≠ <italic>j</italic>.</p><p>4. Find <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M32" name="1471-2105-7-299-i29" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mo stretchy="false">(</mml:mo><mml:mi>u</mml:mi><mml:mo>,</mml:mo><mml:mi>v</mml:mi><mml:mo stretchy="false">)</mml:mo><mml:mo>=</mml:mo><mml:mi>arg</mml:mi><mml:mo>⁡</mml:mo><mml:munder><mml:mrow><mml:mi>min</mml:mi><mml:mo>⁡</mml:mo></mml:mrow><mml:mrow><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow></mml:munder></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqGGOaakcqWG1bqDcqGGSaalcqWG2bGDcqGGPaqkcqGH9aqpcyGGHbqycqGGYbGCcqGGNbWzdaWfqaqaaiGbc2gaTjabcMgaPjabc6gaUbWcbaGaemyAaKMaemOAaOgabeaaaaa@3E4D@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>J</italic><sup><italic>ij</italic></sup>(<italic>α</italic>), and calculate <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M33" name="1471-2105-7-299-i30" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>M</mml:mi><mml:mn>0</mml:mn><mml:mrow><mml:mi>u</mml:mi><mml:mi>v</mml:mi></mml:mrow></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGnbqtdaqhaaWcbaGaeGimaadabaGaemyDauNaemODayhaaaaa@31D2@</mml:annotation></mml:semantics></mml:math></inline-formula> = <italic>K</italic><sub>1</sub>(Γ<sub><italic>uv</italic></sub>,:)<sup><italic>T </italic></sup><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M34" name="1471-2105-7-299-i31" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>B</mml:mi><mml:mn>0</mml:mn><mml:mrow><mml:mi>u</mml:mi><mml:mi>v</mml:mi></mml:mrow></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGcbGqdaqhaaWcbaGaeGimaadabaGaemyDauNaemODayhaaaaa@31BC@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>K</italic><sub>1</sub>(Γ<sub><italic>uv</italic></sub>,:), and <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M35" name="1471-2105-7-299-i32" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>N</mml:mi><mml:mn>0</mml:mn><mml:mrow><mml:mi>u</mml:mi><mml:mi>v</mml:mi></mml:mrow></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGobGtdaqhaaWcbaGaeGimaadabaGaemyDauNaemODayhaaaaa@31D4@</mml:annotation></mml:semantics></mml:math></inline-formula> = <italic>K</italic><sub>1</sub>(Γ<sub><italic>uv</italic></sub>,:)<sup><italic>T </italic></sup><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M36" name="1471-2105-7-299-i33" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>W</mml:mi><mml:mn>0</mml:mn><mml:mrow><mml:mi>u</mml:mi><mml:mi>v</mml:mi></mml:mrow></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGxbWvdaqhaaWcbaGaeGimaadabaGaemyDauNaemODayhaaaaa@31E6@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>K</italic><sub>1</sub>(Γ<sub><italic>uv</italic></sub>,:).</p><p>5. Update <italic>α</italic><sup>(<italic>n</italic>)</sup></p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M37" name="1471-2105-7-299-i34" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:msup> <mml:mi>α</mml:mi> <mml:mrow> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo>+</mml:mo> <mml:mn>1</mml:mn> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:msup> <mml:mo>=</mml:mo> <mml:msup> <mml:mi>α</mml:mi> <mml:mrow> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:msup> <mml:mo>+</mml:mo> <mml:mi>η</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo stretchy="false">(</mml:mo> <mml:mfrac> <mml:mn>1</mml:mn> <mml:mrow> <mml:msubsup> <mml:mi>J</mml:mi> <mml:mn>2</mml:mn> <mml:mrow> <mml:mi>u</mml:mi> <mml:mi>v</mml:mi> </mml:mrow> </mml:msubsup> </mml:mrow> </mml:mfrac> <mml:msubsup> <mml:mi>M</mml:mi> <mml:mn>0</mml:mn> <mml:mrow> <mml:mi>u</mml:mi> <mml:mi>v</mml:mi> </mml:mrow> </mml:msubsup> <mml:mo>−</mml:mo> <mml:mfrac> <mml:mrow> <mml:msup> <mml:mi>J</mml:mi> <mml:mrow> <mml:mi>u</mml:mi> <mml:mi>v</mml:mi> </mml:mrow> </mml:msup> </mml:mrow> <mml:mrow> <mml:msubsup> <mml:mi>J</mml:mi> <mml:mn>2</mml:mn> <mml:mrow> <mml:mi>u</mml:mi> <mml:mi>v</mml:mi> </mml:mrow> </mml:msubsup> </mml:mrow> </mml:mfrac> <mml:msubsup> <mml:mi>N</mml:mi> <mml:mn>0</mml:mn> <mml:mrow> <mml:mi>u</mml:mi> <mml:mi>v</mml:mi> </mml:mrow> </mml:msubsup> <mml:mo stretchy="false">)</mml:mo> <mml:msup> <mml:mi>α</mml:mi> <mml:mrow> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:msup> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFXoqydaahaaWcbeqaaiabcIcaOiabd6gaUjabgUcaRiabigdaXiabcMcaPaaakiabg2da9iab=f7aHnaaCaaaleqabaGaeiikaGIaemOBa4MaeiykaKcaaOGaey4kaSIae83TdGMaeiikaGIaemOBa4MaeiykaKIaeiikaGYaaSaaaeaacqaIXaqmaeaacqWGkbGsdaqhaaWcbaGaeGOmaidabaGaemyDauNaemODayhaaaaakiabd2eannaaDaaaleaacqaIWaamaeaacqWG1bqDcqWG2bGDaaGccqGHsisldaWcaaqaaiabdQeaknaaCaaaleqabaGaemyDauNaemODayhaaaGcbaGaemOsaO0aa0baaSqaaiabikdaYaqaaiabdwha1jabdAha2baaaaGccqWGobGtdaqhaaWcbaGaeGimaadabaGaemyDauNaemODayhaaOGaeiykaKIae8xSde2aaWbaaSqabeaacqGGOaakcqWGUbGBcqGGPaqkaaaaaa@6087@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>and normalize <italic>α</italic><sup>(<italic>n</italic>+1) </sup>so that \|\|<italic>α</italic><sup>(<italic>n</italic>+1) </sup>\|\| = 1.</p><p>6. If <italic>n </italic>reaches a prespecified number <italic>N</italic>, stop. Otherwise, set <italic>n </italic>= <italic>n </italic>+ 1, go to step 3.</p></sec><sec><title>0.4 KNN classification using the optimized kernel distance metric</title><p>Given two samples <italic>x,y </italic>∈ <bold>R</bold><sup><italic>d</italic></sup>, the inner product is defined as: <italic>x</italic>·<italic>y </italic>= <<italic>x</italic>, <italic>y </italic>> = <italic>k</italic>(<italic>x</italic>, <italic>y</italic>); therefore, their derived distance can be calculated</p><p><italic>d</italic>(<italic>x</italic>, <italic>y</italic>) = <<italic>x</italic>, <italic>x </italic>> + <<italic>y</italic>, <italic>y </italic>> -2 <<italic>x, y </italic>> = <italic>k</italic>(<italic>x</italic>, <italic>x</italic>) + <italic>k</italic>(<italic>y</italic>, <italic>y</italic>) - 2<italic>k</italic>(<italic>x</italic>, <italic>y</italic>).</p><p>Using our data-dependent kernel model, the distance can be expressed as</p><p><italic>d</italic>(<italic>x</italic>, <italic>y</italic>) = <italic>q</italic><sup>2</sup>(<italic>x</italic>) + <italic>q</italic><sup>2</sup>(<italic>y</italic>) - 2<italic>q</italic>(<italic>x</italic>)<italic>q</italic>(<italic>y</italic>)<italic>k</italic><sub>0</sub>(<italic>x</italic>, <italic>y</italic>) = [<italic>q</italic>(<italic>x</italic>) - <italic>q</italic>(<italic>y</italic>)]<sup>2 </sup>+ 2<italic>q</italic>(<italic>x</italic>)<italic>q</italic>(<italic>y</italic>)(1 - <italic>k</italic><sub>0</sub>(<italic>x</italic>, <italic>y</italic>))</p><p>where we assume that the basic kernel function satisfy: <italic>k</italic><sub>0</sub>(<italic>x,x</italic>) = 1, just like the Gaussian function.</p><p>Since the kernel optimization scheme increases the class separability of the data in the feature space, the performances of kernel machines should be improved. However, for the classification of gene expression data, due to the small size of training samples, the kernel optimization, which performs on training data, may cause overfitting, which means the algorithm may work very well on the training data, but worse on the test data. To handle this problem, we adopted a disturbed resampling strategy to increase the sample size of the training data.</p><p>Suppose that {<italic>x</italic><sub><italic>i</italic></sub>, <italic>ζ</italic><sub><italic>i</italic></sub>} (<italic>i </italic>= 1,2, ... <italic>m</italic>) are the training data, we construct a new set of training data {<italic>y</italic><sub><italic>i</italic></sub><italic>,ξ</italic><sub><italic>i</italic></sub>}(<italic>i </italic>= 1,2,...,3<italic>m</italic>), where</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M38" name="1471-2105-7-299-i35" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:msub> <mml:mi>y</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo>=</mml:mo> <mml:mrow> <mml:mo>{</mml:mo> <mml:mrow> <mml:mtable columnalign="left"> <mml:mtr columnalign="left"> <mml:mtd columnalign="left"> <mml:mrow> <mml:msub> <mml:mi>x</mml:mi> <mml:mi>i</mml:mi> </mml:msub> </mml:mrow> </mml:mtd> <mml:mtd columnalign="left"> <mml:mrow> <mml:mtext>if </mml:mtext> <mml:mn>1</mml:mn> <mml:mo>≤</mml:mo> <mml:mi>i</mml:mi> <mml:mo>≤</mml:mo> <mml:mi>m</mml:mi> </mml:mrow> </mml:mtd> </mml:mtr> <mml:mtr columnalign="left"> <mml:mtd columnalign="left"> <mml:mrow> <mml:msub> <mml:mi>x</mml:mi> <mml:mi>r</mml:mi> </mml:msub> <mml:mo>+</mml:mo> <mml:mi>ε</mml:mi> </mml:mrow> </mml:mtd> <mml:mtd columnalign="left"> <mml:mrow> <mml:mtext>if </mml:mtext> <mml:mi>i</mml:mi> <mml:mo>></mml:mo> <mml:mi>m</mml:mi> </mml:mrow> </mml:mtd> </mml:mtr> </mml:mtable> </mml:mrow> </mml:mrow> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mn>10</mml:mn> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWG5bqEdaWgaaWcbaGaemyAaKgabeaakiabg2da9maaceaabaqbaeaabiGaaaqaaiabdIha4naaBaaaleaacqWGPbqAaeqaaaGcbaGaeeyAaKMaeeOzayMaeeiiaaIaeGymaeJaeyizImQaemyAaKMaeyizImQaemyBa0gabaGaemiEaG3aaSbaaSqaaiabdkhaYbqabaGccqGHRaWkiiGacqWF1oqzaeaacqqGPbqAcqqGMbGzcqqGGaaicqWGPbqAcqGH+aGpcqWGTbqBaaaacaGL7baacaWLjaGaaCzcamaabmaabaGaeGymaeJaeGimaadacaGLOaGaayzkaaaaaa@510C@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>in which <italic>x</italic><sub><italic>r </italic></sub>is a sample randomly selected form {<italic>x</italic><sub><italic>i</italic></sub>} with replacement and <italic>ε </italic>denotes a normal random disturb, that is, <italic>ε </italic>~<italic>N</italic>(0, <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M39" name="1471-2105-7-299-i36" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msubsup><mml:mi>σ</mml:mi><mml:mi>ε</mml:mi><mml:mn>2</mml:mn></mml:msubsup></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWFdpWCdaqhaaWcbaGae8xTdugabaGaeGOmaidaaaaa@3137@</mml:annotation></mml:semantics></mml:math></inline-formula>). The class labels are determined as</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M40" name="1471-2105-7-299-i37" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:msub> <mml:mi>ξ</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo>=</mml:mo> <mml:mrow> <mml:mo>{</mml:mo> <mml:mrow> <mml:mtable columnalign="left"> <mml:mtr columnalign="left"> <mml:mtd columnalign="left"> <mml:mrow> <mml:msub> <mml:mi>ζ</mml:mi> <mml:mi>i</mml:mi> </mml:msub> </mml:mrow> </mml:mtd> <mml:mtd columnalign="left"> <mml:mrow> <mml:mtext>if </mml:mtext> <mml:mn>1</mml:mn> <mml:mo>≤</mml:mo> <mml:mi>i</mml:mi> <mml:mo>≤</mml:mo> <mml:mi>m</mml:mi> </mml:mrow> </mml:mtd> </mml:mtr> <mml:mtr columnalign="left"> <mml:mtd columnalign="left"> <mml:mrow> <mml:msub> <mml:mi>ζ</mml:mi> <mml:mi>r</mml:mi> </mml:msub> </mml:mrow> </mml:mtd> <mml:mtd columnalign="left"> <mml:mrow> <mml:mtext>if </mml:mtext> <mml:mi>i</mml:mi> <mml:mo>></mml:mo> <mml:mi>m</mml:mi> </mml:mrow> </mml:mtd> </mml:mtr> </mml:mtable> </mml:mrow> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaiiGacqWF+oaEdaWgaaWcbaGaemyAaKgabeaakiabg2da9maaceaabaqbaeaabiGaaaqaaiab=z7a6naaBaaaleaacqWGPbqAaeqaaaGcbaGaeeyAaKMaeeOzayMaeeiiaaIaeGymaeJaeyizImQaemyAaKMaeyizImQaemyBa0gabaGae8NTdO3aaSbaaSqaaiabdkhaYbqabaaakeaacqqGPbqAcqqGMbGzcqqGGaaicqWGPbqAcqGH+aGpcqWGTbqBaaaacaGL7baaaaa@4A9E@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>Due to the very high dimensionality and small number of the patient samples, the training data are sparsely distributed in the high dimensional Euclidean space. It is reasonable to assume that the near points of a training datum have the same class characteristic as that of the training datum. Experimentally, using the technique of disturbed resampling (Eq.(l0)), we can effectively reduce the possible overfitting and computational instability, which are mainly caused by the lack of enough training samples for the gene expression data.</p></sec></sec><sec><title>Abbreviations</title><p>KNN: K-nearest-Neighbor</p><p>SVM: support vector machine</p><p>DLDA: diagonal linear discriminant analysis</p><p>ULDA: uncorrelated linear discriminant analysis</p><p>KerNN: kernel optimization-based KNN</p><p>ALL: acute lymphoblastic leukemia</p><p>AML: acute myeloid leukemia</p><p>MLL: mixed lineage leukemia</p><p>CNS: embryonal tumor of central nervous system</p></sec><sec><title>Authors' contributions</title><p>HX and XWC conceived the study. HX designed and implemented the algorithms, and drafted the manuscript. XWC coordinated the study, participated in the algorithm design, and helped draft the manuscript.</p></sec><sec><title>Availability</title><p>The core source codes of our algorithms are available at <ext-link ext-link-type="uri" xlink:href="http://www.ittc.ku.edu/~xwchen/BMCbioinformatics/kernel/"/></p></sec>
Array2BIO: from microarray expression data to functional annotation of co-regulated genes	<sec><title>Background</title><p>There are several isolated tools for partial analysis of microarray expression data. To provide an integrative, easy-to-use and automated toolkit for the analysis of Affymetrix microarray expression data we have developed Array2BIO, an application that couples several analytical methods into a single web based utility.</p></sec><sec><title>Results</title><p>Array2BIO converts raw intensities into probe expression values, automatically maps those to genes, and subsequently identifies groups of co-expressed genes using two complementary approaches: (1) comparative analysis of signal versus control and (2) clustering analysis of gene expression across different conditions. The identified genes are assigned to functional categories based on Gene Ontology classification and KEGG protein interaction pathways. Array2BIO reliably handles low-expressor genes and provides a set of statistical methods for quantifying expression levels, including Benjamini-Hochberg and Bonferroni multiple testing corrections. An automated interface with the ECR Browser provides evolutionary conservation analysis for the identified gene loci while the interconnection with Crème allows prediction of gene regulatory elements that underlie observed expression patterns.</p></sec><sec><title>Conclusion</title><p>We have developed Array2BIO – a web based tool for rapid comprehensive analysis of Affymetrix microarray expression data, which also allows users to link expression data to Dcode.org comparative genomics tools and integrates a system for translating co-expression data into mechanisms of gene co-regulation. Array2BIO is publicly available at <ext-link ext-link-type="uri" xlink:href="http://array2bio.dcode.org."/></p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Loots</surname><given-names>Gabriela G</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Chain</surname><given-names>Patrick SG</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I3">3</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Mabery</surname><given-names>Shalini</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Rasley</surname><given-names>Amy</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Garcia</surname><given-names>Emilio</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A6" corresp="yes" contrib-type="author"><name><surname>Ovcharenko</surname><given-names>Ivan</given-names></name><xref ref-type="aff" rid="I1">1</xref><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib>	BMC Bioinformatics	<sec><title>Background</title><p>Microarray experiments provide a rapid method for directly profiling the expression pattern of an entire gene repertoire in a genome. This experimental approach has become routine for the <italic>en masse </italic>identification of genes associated with different biological processes. We have developed a multifunctional, user-friendly, web-interactive microarray analysis tool, Array2BIO, that identifies and functionally characterizes co-expressed genes. In addition, it integrates other genomic, transcriptional and gene regulatory tools (Loots and Ovcharenko 2005) to allow scientists to explore mechanisms of gene co-regulation specific to co-functional groups of genes. Array2BIO permits users to functionally characterize clusters of co-expressed genes, identify putative biological activities, study interaction networks, as well as predict modules of transcription factors regulating eukaryotic gene expression in different tissues and under different conditions.</p></sec><sec><title>Implementation</title><sec><title>Microarray data analysis</title><sec><title>Background correction</title><p>Array2BIO follows the original Affymetrix procedure of background correction. An array of probes is separated into 16 zones (4 × 4 grid). Raw intensities for each zone are ranked and the background level is defined as the 2% lowest intensity for each zone. The distance from each probe to the zone center is used to estimate the background level at each probe location, which is then subtracted from the raw probe intensity.</p></sec><sec><title>Filtering out non-specific hybridization</title><p>Each probe intensity is measured in duplicates – a perfect match (PM) intensity and mismatch (MM) intensity, where the MM intensity estimates the cross-reactivity with other genes. Array2BIO excludes all probes with a PM intensity less than 1.25MM. It also calculates the ratio of probes with specific hybridization that pass through this filtering. MM intensity is subtracted from the PM intensity for the remaining probes, such that the raw intensity is measured as the relative (PM-MM) intensity.</p></sec><sec><title>Normalization and Log<sub>2 </sub>transformation</title><p>Median (PM-MM) array intensity <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M1" name="1471-2105-7-307-i1" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>I</mml:mi><mml:mo>˜</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGjbqsgaacaaaa@2DD6@</mml:annotation></mml:semantics></mml:math></inline-formula> is calculated for the remaining probes after the filtering step. Individual (PM-MM) probe intensities <italic>I</italic><sub><italic>i </italic></sub>undergo normalization and a base 2 logarithmic transformation:</p><p><italic>EP</italic><sub><italic>i </italic></sub>= log<sub><italic>s</italic></sub>(<italic>I</italic><sub><italic>i</italic></sub>/<inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M2" name="1471-2105-7-307-i1" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>I</mml:mi><mml:mo>˜</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGjbqsgaacaaaa@2DD6@</mml:annotation></mml:semantics></mml:math></inline-formula>).</p></sec><sec><title>Probe to tag mapping</title><p>Affymetrix <italic>.CDF </italic>files are used to map individual probe intensities <italic>EP</italic><sub><italic>i </italic></sub>onto Affymetrix gene tags <italic>GP</italic><sub><italic>j</italic></sub>. Usually each tag accumulates ~ 10 good probes that span the corresponding gene transcript.</p></sec><sec><title>Averaging experiment replicas</title><p>Several experimental replicas can be averaged in comparative analysis to reliably estimate signal and background gene expression levels.</p></sec><sec><title>Filtering out the outliers</title><p>It is common to observe that the expression level of several gene probes differs significantly from the median level of transcript expression <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M3" name="1471-2105-7-307-i2" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>G</mml:mi><mml:mo>˜</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGhbWrgaacaaaa@2DD2@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>P</italic><sub><italic>j </italic></sub>. To filter out the outliers, Array2BIO excludes transcript probes with expression values that differ from <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M4" name="1471-2105-7-307-i2" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mover accent="true"><mml:mi>G</mml:mi><mml:mo>˜</mml:mo></mml:mover><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacuWGhbWrgaacaaaa@2DD2@</mml:annotation></mml:semantics></mml:math></inline-formula><italic>P</italic><sub><italic>j </italic></sub>by an <italic>x </italic>number of standard deviations <italic>σ</italic><sub><italic>j </italic></sub>(thresholds defined by the user). A strict filtering (1 <italic>σ</italic><sub><italic>j</italic></sub>) and a medium stringency filtering (2* <italic>σ</italic><sub><italic>j</italic></sub>) are set as defaults for the comparative and clustering analyses, correspondingly.</p></sec></sec><sec><title>Statistical methods (comparative analysis)</title><sec><title>Handling low-expressors</title><p>The significance of fold-difference in intensity values (ie. expression) varies dramatically for low- vs. high-expressor genes. This occurs because dividing a small number by another small number (in case of low-expressors) can result in a large fold-difference simply by chance. Array2BIO utilizes local mean normalization and local variance correction across intensities to differentially handle low- and high-expressors and to define separate fold-difference thresholds for different intensity levels. Array2BIO employs an approach highly similar to the previously described SNOMAD method (Colantuoni et al. 2002) and represents a 'pooled local variance' approach with 100 bins of gene tags. First, fold-expression levels of Affymetrix tags are ordered by their average expression level across signal and control data. Then gene tags are binned into 100 groups by the average expression level and local variation of fold-expressions is calculated for each group. This allows one to compute the local standard deviation (<italic>σ</italic><sup><italic>i</italic></sup>) and subsequently local z-score (<italic>z</italic><sub><italic>j</italic></sub>) of fold-difference for each individual gene tag in each <italic>i</italic>-th group that <italic>j</italic>-th gene tag belongs to:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M5" name="1471-2105-7-307-i3" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mi>z</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:msub><mml:mi>X</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mover accent="true"><mml:mrow><mml:msup><mml:mi>X</mml:mi><mml:mi>i</mml:mi></mml:msup></mml:mrow><mml:mo stretchy="true">¯</mml:mo></mml:mover></mml:mrow><mml:mrow><mml:msup><mml:mi>σ</mml:mi><mml:mi>i</mml:mi></mml:msup></mml:mrow></mml:mfrac></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWG6bGEdaWgaaWcbaGaemOAaOgabeaakiabg2da9maalaaabaGaemiwaG1aaSbaaSqaaiabdQgaQbqabaGccqGHsisldaqdaaqaaiabdIfaynaaCaaaleqabaGaemyAaKgaaaaaaOqaaGGaciab=n8aZnaaCaaaleqabaacbiGae4xAaKgaaaaaaaa@3ABF@</mml:annotation></mml:semantics></mml:math></inline-formula>, where <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M6" name="1471-2105-7-307-i4" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mover accent="true"><mml:mrow><mml:msup><mml:mi>X</mml:mi><mml:mi>i</mml:mi></mml:msup></mml:mrow><mml:mo stretchy="true">¯</mml:mo></mml:mover></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaadaqdaaqaaiabdIfaynaaCaaaleqabaGaemyAaKgaaaaaaaa@2F7E@</mml:annotation></mml:semantics></mml:math></inline-formula> is the average fold-difference in expression of the <italic>i</italic>-th group. Differentially expressed tags identified by Z-score greater than 2.0 are selected for further analysis (Figure <xref ref-type="fig" rid="F3">3</xref>).</p></sec><sec><title>Welch's t-test of differential expression significance</title><p>Signal and control tags that survive the balance analysis of low- and high-expressors are next subjected to statistical testing using the Welch's t-test method. Statistical testing is performed on the average signal and control tag expression using standard deviations of their probe expression distribution. A p-value is assigned to every differentially expressed tag and tags with p-values less than 0.05 are selected for multiple testing correction analyses.</p></sec><sec><title>Mapping Affymetrix tags onto UCSC known genes</title><p>Array2BIO first identifies a set of unique (non-overlapping) genes in a genome matching the original.<italic>CEL </italic>file by using the 'known genes' annotation provided by the UCSC Genome Browser database (Karolchik et al. 2003). Next, Affymetrix tags are mapped onto (and are grouped by) UCSC 'known genes'. Accession numbers for the corresponding mRNA sequences and their genomic locations are retrieved for each gene during the mapping process. This information is next used to dynamically link genes to the NCBI database and to the ECR Browser.</p></sec><sec><title>Gene Ontology (GO) and KEGG analyses of biological functions and gene interactions</title><p>Array2BIO utilizes a locally installed version of the Gene Ontology (GO) (Harris et al. 2004) and KEGG (Ogata et al. 1999) databases to contrast the distribution of differentially expressed functional categories of genes to the average distribution in the corresponding genome. Observed and expected category population values are compared and the statistical 'enrichment' (or 'depletion') of a category is quantified by using hypergeometric distribution statistics. Functional categories with p-values smaller than 0.05 are selected for subsequent multiple testing correction analyses. The GO database provides biological classification of gene function through membership to functional categories that relate to certain biological processes, molecular functions, or to cellular components. The KEGG database combines information on gene interactions that are grouped into (1) metabolism, (2) genetic information processing, (3) environmental information processing, (4) cellular processes, and (5) human diseases categories.</p></sec><sec><title>Correction for multiple testing</title><p>Array2BIO performs correction for multiple testing to exclude false positive predictions associated with the statistical testing of differential tag expression or enrichment/depletion in GO and KEGG categories that is performed multiple times. Array2BIO provides two statistical methods to correct for multiple testing and also allows omitting multiple testing if the user does not want to apply this function. The default method used by Array2BIO is the medium stringency Benjamini-Hochberg correction (Benjamini and Hochberg 1995). Benjamini-Hochberg correction is based on controlling the false discovery rate (FDR) – the expected proportion of false discoveries amongst the rejected hypothesis. In general it provides a good balance between discovery of statistically significant differences and limitation of false positive occurrences. Alternatively, the Bonferroni correction method can be applied. The latter is one of the most stringent multiple testing correction methods and can be used to select for the most outstanding overexpressor genes or enriched/depleted functional categories.</p></sec></sec><sec><title>Clustering analysis</title><sec><title>Microarray data clustering</title><p>Array2BIO utilizes the Unix version of the Cluster tool (Eisen et al. 1998). Cluster's hierarchical analysis is implemented into Array2BIO, which allows clustering of genes and/or conditions; provides 9 distance measures and 4 methods. Due to Cluster limitations, Array2BIO restricts the maximum number of clustered transcripts to less than 2500 genes. Genes are ranked by their standard deviation in expression across different conditions. Genes with the largest variation from their average expression across all conditions are selected for clustering.</p></sec><sec><title>Interactive tree visualization</title><p>Array2BIO provides an interactive web utility for visualizing clustering results, which is similar in graphical display and operation to Java TreeView (Saldanha 2004). Clustered gene expression across multiple conditions is visualized in a matrix format. The tree of clustering relationships is given to the left of the gene expression image (Figure <xref ref-type="fig" rid="F4">4A</xref>). A mouse click on a tree branch generates a 'zoom in' image of that branch and gives a detailed description of related genes (including gene names, accession numbers, corresponding Affymetrix tags, and genomic locations) (Figure <xref ref-type="fig" rid="F4">4B</xref>).</p></sec></sec><sec><title>Interconnection with external tools</title><sec><title>ECR Browser – evolutionary conservation analysis</title><p>The ECR Browser (Ovcharenko et al. 2004) is a dynamic whole-genome navigation tool for visualizing and studying evolutionary relationships among genomes. Evolutionary Conserved Regions (ECRs) are extracted from genome alignments, mapped to genomes, and graphically visualized in relation to the genes that have been annotated in the reference genome.</p></sec><sec><title>Creme 2.0 – identification of clusters of transcription factor binding sites in promoters</title><p>Crème 2.0 (Sharan et al. 2004) relies on a database of putative transcription factor binding sites that have been carefully annotated across the human genome using evolutionary conservation with the mouse and rat genomes. An efficient search algorithm is applied to this data set to identify combinations of transcription factors whose binding sites tend to co-occur in close proximity to the start site of the input gene set. These combinations are statistically evaluated, and significant combinations are reported and visualized.</p></sec><sec><title>NCBI – detailed sequence information</title><p>Detailed mRNA transcript information including: nucleotide and protein sequences, related publications, gene annotation, etc. are provided through the dynamic interconnection to the NCBI database.</p></sec></sec></sec><sec><title>Results and discussion</title><p>Figure <xref ref-type="fig" rid="F1">1</xref> summarizes the schematics behind Array2BIO analysis. Users are required to submit input textual. CEL files – (i.e. the standard output data derived from Affymetrix microarray experiments). Array2BIO performs multi-step data analysis and filtering, including background correction, exclusion of non-specific hybridizing probes, normalization and logarithmic transformation of raw intensities. Individual probes are automatically mapped to Affymetrix tags and subsequently to UCSC 'known genes' (Karolchik et al. 2003). In contrast to other available microarray analysis software, Array2BIO analysis also incorporates a balanced analysis of low- and high-expressor genes thus providing a reliable method for handling low-expressors that would otherwise lead to false positive predictions.</p><p>Two complementary methods of microarray data analysis are incorporated into the Array2BIO software: 1) comparative and 2) clustering analyses. Comparative analysis identifies genes that are differentially regulated in reference to a control sample (for example gene expression in transgenic animals compared to non-transgenic, wild-type littermates). Clustering analysis identifies groups of genes that are co-expressed under different experimental conditions (e.g. when analyzing time-course experiments).</p><p>The automated functional classification of co-expressed genes is based on the Gene Ontology (Harris et al. 2004) database and allows the identification of 'enriched' or 'depleted' categories in assigned biological processes, molecular functions, and cellular components. Integrated KEGG (Ogata et al. 1999) classification of gene interactions identifies major biochemical processes that underlie observed differences in gene expression and groups genes into five main categories – (1) metabolism, (2) genetic information processing, (3) environmental information processing, (4) cellular processes, and (5) human diseases.</p><p>Every group of differentially expressed genes identified using Array2BIO is dynamically linked to the Evolutionary Conserved Region (ECR) Browser (Ovcharenko et al. 2004) and to the Cis-REgulatory Module Explorer tool (Sharan et al. 2004), as well as to the NCBI database. The ECR Browser provides multi-species evolutionary conservation information for individual genes, and the NCBI database provides detailed information about mRNA sequences and related proteins. The Crème 2.0 tool allows the user to perform an additional step to functionally annotate groups of human genes through the analysis of their promoter elements. In this process the tool will identify shared clusters of evolutionary conserved transcription factor binding sites within promoters of co-expressed genes. Combined, these tools provide a wealth of information regarding the gene(s) in question, its conservation, its transcripts, as well as candidate regulatory mechanisms underlying the observed transcriptional response from the microarray data.</p><sec><title>Application to the analysis of host-pathogen interactions</title><p>To illustrate the different levels of information that can be obtained from Array2Bio analysis we have processed microarray expression data generated in a time-course experiment of human cells infected with <italic>Yersinia pestis</italic>. The plague (commonly known as the Black Death) is an infectious disease that has devastated much of the known world in the 14<sup>th </sup>century, and killed more than 200 million people during three major pandemics. It is primarily a disease in rodents caused by an infection with the bacterium <italic>Yersinia pestis</italic>, but can be transmitted to humans through the bite of infected fleas.</p><p>To address host-pathogen interactions and elucidate the molecular mechanisms underlying the virulence of this pathogen during human infection, human dendritic cells were exposed to <italic>Y. pestis </italic>infection, and RNA samples were collected at different time points and gene expression was analyzed by microarrays. Using Array2Bio we compared HG-U133A microarray expression data of human dendritic cells at 4 hours after exposureto <italic>Y. pestis </italic>to mock-exposed cells. We observed significant increases and decreases in expression (as measured using the Welch's t-test analysis with Benjamini-Hochberg correction for multiple testing) for 139 and 81 human genes, respectively. Gene Ontology (GO) analysis identified 31 'enriched' biological processes and 5 molecular functions corresponding to up-regulated genes; while none were found for down-regulated genes. As expected, the majority of these categories were related to the human immune response, including the "response to pest, pathogen or parasite" (Table <xref ref-type="table" rid="T1">1</xref>). The chemokine (cytokines with chemotactic activities) category was ~ 20-fold 'enriched' when compared to the expected values due to chance alone. Eighteen percent of all human chemokines (primarily CXC chemokines) are activated in response to <italic>Y. pestis </italic>invasion. KEGG analysis of the corresponding gene interactions identified a family of up-regulated CXC cytokines acting upstream of the IL8RB receptor, and several other receptor genes (Figure <xref ref-type="fig" rid="F2">2</xref>). These pathways are likely to reflect the core response of human dendritic cells to this infection. KEGG analysis of enriched cellular processes highlighted two related subcategories: (1) apoptosis (p < 0.001) and (2) cell growth and death (p < 0.002). Six genes are shared between these two subcategories and may be key players in the etiology of this infectious disease.</p><p>We performed Crème 2.0 analysis on 25 genes identified in this study that are related to the "response to pest, pathogen or parasite" GO category. Crème 2.0 predicted transcription factors that potentially act as key regulators of these genes and are likely to up-regulate their expression during <italic>Y. pestis </italic>infection. Several transcription factors binding sites conserved between human and rodents were significantly enriched in the promoters of these genes, including several members of the STAT and NFKB families, as well as TATA transcription factors. While the TATA transcription factor plays a basal role in the TATA-box recognition, the two other identified transcription factor families are known to be involved in regulating the immune system. STAT and NFKB proteins respond to cytokines, are associated with inflammatory disease and can lead to inappropriate immune cell development. (Hirayama et al. 2005; O'Shea et al. 2005).</p></sec></sec><sec><title>Conclusion</title><p>Array2BIO is an addition to the Dcode.org collection of tools (Loots and Ovcharenko 2005) that permits the efficient and unique integration of comparative and transcriptional regulatory genomic utilities with a multi-functional framework for analyzing gene expression data. Most importantly, Array2BIO represents a web-based tool/utility for integrative analysis of microarray expression data that permits experimental biologists with limited background in statistics to perform detailed, highly informative analysis comparable to sophisticated software packages catered to the expert statistician. A "single-click" implementation of the variety of biological characterizations into a single tool permits the standardized, prompt identification of co-expressed genes, their functional annotation, the identification of related interaction pathways, and prediction of key transcription factors underlying observed gene expression responses. Currently our server provides 200 Mb of disk space per account. All the input CEL files are compressed allowing users to store over one hundred CEL files per account. We anticipate additional disk space to be made available per account, with each new release of the tool.</p></sec><sec><title>Availability and requirements</title><p>Project name: Array2BIO;</p><p>Project home page: <ext-link ext-link-type="uri" xlink:href="http://array2bio.dcode.org"/>;</p><p>Operating system(s): Web-based, platform independent;</p><p>Programming language: PHP;</p><p>License: There are no access restrictions and no need for a license for both academic and private entities to use this research tool.</p></sec><sec><title>Authors' contributions</title><p>GGL participated in designing the scheme of the tool and writing the manuscript. PSGC, SM, AR, and EG carried out experimental studies. IO coordinated the developments, created the tool and drafted the manuscript.</p></sec>
An analysis of the Sargasso Sea resource and the consequences for database composition	<sec><title>Background</title><p>The environmental sequencing of the Sargasso Sea has introduced a huge new resource of genomic information. Unlike the protein sequences held in the current searchable databases, the Sargasso Sea sequences originate from a single marine environment and have been sequenced from species that are not easily obtainable by laboratory cultivation. The resource also contains very many fragments of whole protein sequences, a side effect of the shotgun sequencing method.</p><p>These sequences form a significant addendum to the current searchable databases but also present us with some intrinsic difficulties. While it is important to know whether it is possible to assign function to these sequences with the current methods and whether they will increase our capacity to explore sequence space, it is also interesting to know how current bioinformatics techniques will deal with the new sequences in the resource.</p></sec><sec><title>Results</title><p>The Sargasso Sea sequences seem to introduce a bias that decreases the potential of current methods to propose structure and function for new proteins. In particular the high proportion of sequence fragments in the resource seems to result in poor quality multiple alignments.</p></sec><sec><title>Conclusion</title><p>These observations suggest that the new sequences should be used with care, especially if the information is to be used in large scale analyses. On a positive note, the results may just spark improvements in computational and experimental methods to take into account the fragments generated by environmental sequencing techniques.</p></sec>	<contrib id="A1" corresp="yes" contrib-type="author"><name><surname>Tress</surname><given-names>Michael L</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Cozzetto</surname><given-names>Domenico</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Tramontano</surname><given-names>Anna</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Valencia</surname><given-names>Alfonso</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Bioinformatics	<sec><title>Background</title><p>The environmental sequencing of the Sargasso Sea [<xref ref-type="bibr" rid="B1">1</xref>] has raised expectations in fields as diverse as marine ecology and energy conservation. As a result of the work a huge new resource of genomic information, comprising more than one million distinct protein sequences from an estimated 1,800 new species, has been made available to the public sequence databases.</p><p>The quantity of new protein sequences made available by this project is remarkable. At the time of their sequencing there were almost as many protein sequences in the Sargasso Sea resource as were held in the current public databases. At 90% redundancy the combination of SWISSPROT, TREMBL and TREMBLnew databases [<xref ref-type="bibr" rid="B2">2</xref>], for example, contained 783,110 protein sequences as of April 2004, while the Sargasso Sea resource has 643,044 sequences. The environmental genomics community has plans to gather bacteria from more of the world's oceans and from other environments, which makes the released sequences only a taste of what is to come.</p><p>While the protein sequences in the public online databases were derived from organisms from a wide range of ecosystems, the sequences from the Sargasso Sea are from a clearly differentiated marine environment. In addition, the species sequenced from the Sargasso Sea, and those that will be sequenced in similar projects in the future, are non-cultivated species, something else that sets them apart from the species whose sequences have traditionally made up the protein databanks. The details of the process for identifying genes (alignments with bacterial protein sequences were used to determine the most likely coding frames and the stop and start codons) is also likely to play a role in the relative distribution of sequences in the database.</p><p>One further difference from the sequences in the current databases is the technology used to sequence them. The Sargasso Sea sequences come from a pull of the entire DNA present in the Sargasso Sea and were sequenced using shotgun technology with low coverage. Hence there are no complete genomes present in the resource and for most of the annotated genes the species is unknown. Many reads are unassembled or partially assembled DNA fragments.</p><p>Initial analyses [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B4">4</xref>] have compared the functional and base composition of the sequences in the Sargasso Sea database with several other environmental resources. Here we hoped to answer a different question, how the Sargasso Sea sequences are distributed in the context of known protein families. We were interested in how the composition and structure of the new sequences influences their classification into previously known protein families, whether the new sequences complemented the sequences in the existing public databases or whether they formed distinct groups separated by discontinuities. If the new sequences were distinct, to what extent were they different to the sequences from the cultivated species in the current databases and what can they tell us about the limitations of the analyses based on those sequences?</p><p>We find however that the high proportion of sequence fragments in the resource means that it is impossible to reach any conclusions about the sequence distribution and that studies carried out with the new resource may be unduly biased by these sequence fragments. Since it is beyond doubt that environmental sequencing projects will push the numbers of protein sequences far beyond today levels, it is important to understand the effects of such large numbers of sequences from such radically different origins will have on our understanding of sequence space and what effect will this have on analysis of protein structure and function.</p></sec><sec><title>Results and discussion</title><sec><title>Make up of Sargasso Sea protein sequence resource</title><p>The sequences have a low GC-content and consequently a high isoleucine, asparagine and lysine content (40% over average) coupled with decreases in the content of certain other amino acids (see Figure <xref ref-type="fig" rid="F1">1</xref>). The isoleucine, asparagine and lysine content and the lower GC-contents of the Sargasso Sea is only comparable to a small number of other bacterial genomes, such as <italic>Staphylococcus aureas</italic>, <italic>Borrelia burgdorferi</italic>, and <italic>Campylobacter jejuni</italic>[<xref ref-type="bibr" rid="B5">5</xref>]. The relative figures for isoleucine, lysine and asparagine are 8.58, 7.56 and 5.68% for S. aureus and 8.18, 7.74 and 5.75% for the Sargasso Sea sequences. While it is surprising to find that an entire environment can have such a distinct GC-content, a recent study by Foerster et al. [<xref ref-type="bibr" rid="B4">4</xref>] has confirmed our findings and suggests that environment may actually shape GC-content.</p><p>Another very important observation is that the Sargasso Sea sequences are shorter, on average 205 residues compared to the typical 342 of the Curr-nr database. In fact many Sargasso Sea sequences are fragments of whole protein sequences and this fact has been explicitly mentioned before [<xref ref-type="bibr" rid="B6">6</xref>]. The influence of sequence fragments can be seen graphically when the distribution of Sargasso Sea sequence lengths is compared to those from a non-redundant database made up of SWISSPROT, TREMBL and TREMBLnew sequences (the Curr-nr database) and those from a database made up of the sequences from all completed bacterial genomes (Figure <xref ref-type="fig" rid="F2">2a</xref>).</p><p>The protein sequences in the Sargasso Sea resource are split into 11 sections by name. Of these 11 sections, 9 contain 99,999 sequences each. The first section (which has identifiers beginning with the triplet "eak") contains over 80,000 sequences and a smaller 11th section (with identifiers beginning with "eaa") has the remaining sequences. The distribution of sequence lengths within each of these sections is not identical, as can be seen in figure <xref ref-type="fig" rid="F2">2b</xref>. The figure shows that eight of the 11 sections compared ("eaa" to "eah") are composed entirely of sequences with fewer than 400 residues. While the other three sections ("eak", "eai" and "eaj") do contain sequences of greater than 400 residues, there are relative fewer full length sequences than would be expected from the sequence length distributions of the whole prokaryotic genomes, which have a significant tail of longer sequences.</p></sec><sec><title>The relationship between the Sargasso Sea sequences and known protein families</title><p>BLAST [<xref ref-type="bibr" rid="B7">7</xref>] searches of the Sargasso Sea database failed to find a single similar sequence for 47 of the 237 query sequences. For 14 of the 47 cases, remotely related Sargasso Sea sequences could be found using PSI-BLAST when the Sargasso Sea sequences were combined with the Curr-nr sequences (the Combined-nr database), something which indicates that the standard sequences are able to occupy an intermediate position between the query sequence and Sargasso Sea sequences in a small number of cases.</p><p>However, combining the Sargasso Sea sequences with the current non-redundant databases did not appear to help in the search for remotely homologous proteins. As part of the investigation into the effects on alignments for structure prediction (see below), PSI-BLAST searches were carried out with a set of 51 query proteins (63 domains) from the homology modelling section of the CASP 4 and CASP 5 (the Critical Assessment of Techniques for Protein Structure Prediction) experiments [<xref ref-type="bibr" rid="B8">8</xref>]. The searches were supposed to determine whether the sequences from the Sargasso Sea (the SSea-nr database sequences) would help to detect templates that could be used in model building. PSI-BLAST profiles were first generated as per the methods section and then the Protein Data Bank (PDB) [<xref ref-type="bibr" rid="B9">9</xref>] was searched with the profiles. Candidate sequences that could be used as templates were discovered for 48 of the 63 domains in searches of the standard database (Curr-nr), while when the Combined-nr database was used (the database to which the SSea-nr sequences had been added) only 42 of the 48 templates could be identified. The addition of the Sargasso Sea sequences actually decreased the capacity to detect templates.</p><p>The same effect was apparent when searching for sequences belonging to known protein families. We compared those 181 query proteins for which PSI-BLAST had detected at least 5 homologous sequences from searches of each of the three sequence databases (SSea-nr, Curr-nr, and Combined-nr) and found that searches of SSea-nr database turned up fewer sequences on average (618 sequences) than searches of the Curr-nr database (807 sequences). PSI-BLAST searches of the Combined-nr database turned up fewer sequences than the two databases separately (1099 sequences), of which almost half were from SSea-nr database. The number of sequences found from the Curr-nr database, the database that included the sequences with functional and structural information, dropped to just 552.</p><p>While some small variations might be expected due to differences in E-values related to the different size of the databases being used, case by case investigations documented below made it clear that what was actually happening had little to do with database size.</p><p>In many cases searches of the Sargasso Sea and the Combined-nr databases reached a point where fewer sequences were found with successive iterations. For example, a search of the Sargasso Sea with the query 1qorA found 2740 sequences on round 2 and only 1559 on round 3. In other words the profile used to search for sequences in the 3rd round finds 1,181 fewer sequences before converging. In this case the profile has lost 40% of the sequence information. This did not happen with the corresponding searches of the Curr-nr database with this target. The same thing happens with query 2dkb (2122 sequences on round 3, 1766 on round 4) and with a number of others.</p></sec><sec><title>Profiles and optimal sequences</title><p>PSI-BLAST creates multiple sequence alignments using the sequences it finds in each search iteration. The program then constructs a profile based on the frequency of amino acids at each residue position in the multiple sequence alignments and by taking into account substitution matrices. PSI-BLAST uses these profiles to search the databases on subsequent search iterations, so the information contained in the profiles is directly linked to the sequences found with each iteration. In order to investigate the odd effects of addition of the new sequences to the databases, and in particular how searches based on known families are affected we concentrated on these profiles.</p><p>Profiles are effectively a matrix formed by the 20 amino acids and the number of residue positions in the query sequence. Each residue in each position in the matrix has a probability score associated to it, a probability score that is calculated from the frequency of each residue in that position in the multiple alignment and from the replacement probabilities that come from the substitution matrix that is used.</p><p>The profile can be used to derive the so-called optimal sequence, defined as the sequence that can be obtained from the highest scoring residues in each position in the profile. The sequence reflects the conservation of each residue position and also the similarity score of the residue pairs. The profile generated from a correctly aligned set of homologous proteins should be enriched in high-scoring residues in those positions that have the most conserved amino acids in the family. In these cases the optimal sequence will be similar to the homologous proteins that have gone to make up the profile.</p><p>However, if the conserved positions are not properly aligned, the optimal sequence will reflect random matching residues and, by virtue of its definition, will be dominated by residues with a high average similarity score, i.e. by rare residues such as tryptophans and, to a certain extent, cysteines (Figure <xref ref-type="fig" rid="F3">3</xref>). This is what can be observed in the optimal sequences generated from Sargasso sequences. The effect has nothing to do with the residue make up of the Sargasso Sea – the Sargasso Sea sequences have the same percentage of tryptophans and cysteines as the sequences in the current databases and actually have less low complexity regions than the current databases. On closer inspection these rare residue repeats were also found in the optimal sequences of PSI-BLAST profiles generated from other databases – however, they were found much more frequently in profiles generated from searches of the SSea-nr and Combined-nr databases.</p><p>That the optimal sequences extracted from the PSI-BLAST matrices contain high proportions of tryptophans and cysteines can be seen clearly in Figure <xref ref-type="fig" rid="F4">4</xref>. Here we compare the proportions of each residue in the optimal sequences generated from PSI-BLAST searches of the Combined-nr database with the background levels of each residue in the sequences of the Sargasso Sea and Current databases that make up the Combined-nr database. There are ten times as many tryptophan residues in the optimal sequences as in the database sequences and cysteine is represented five times more in the optimal sequences than in the databases.</p></sec><sec><title>Alignment conservation (entropy)</title><p>The optimal sequences are generated from the PSI-BLAST profiles for each target, while the profiles are calculated directly from the PSI-BLAST alignments. It is possible to measure the conservation of residue positions in an alignment using residue entropy. We calculated the residue entropy for all 237 of the PSI-BLAST multiple alignments generated from searches of the Combined-nr database (which contains both Sargasso Sea and current database sequences) and plotted entropy directly against each of the optimal sequences for each of the 237 target sequences (we show an example in Figure <xref ref-type="fig" rid="F5">5</xref>). We also plotted entropy against the residue type of the optimal sequences drawn from the profiles (Figure <xref ref-type="fig" rid="F6">6</xref>).</p><p>The plot of entropy versus optimal sequence residue (Figure <xref ref-type="fig" rid="F6">6</xref>) clearly correlates tryptophans in the optimal sequences with low entropy and therefore with poor residue conservation. Tryptophan is the most frequent residue in the optimal sequences; it has the lowest entropy and the lowest variance around the mean of all the residues. Not only that, but repeated tryptophan residues have even less entropy and very little variation in entropy score, showing that repeated tryptophans always mark residues with little or no evolutionary information.</p><p>One more example of the relationship between entropy and optimal sequence is shown in the plot of entropy against the optimal sequence of target T0171 in Figure <xref ref-type="fig" rid="F5">5</xref>. While much of the optimal sequence is characterized by a series of jagged peaks and troughs, representing the variable levels of conservation at each position, the part of the optimal sequence that is a long string of tryptophans essentially flat-lines, showing that all conservation has disappeared from these residues.</p><p>The regions of repeated rare residues in the optimal sequences are clearly symptomatic of those regions of low entropy and low conservation that are devoid of all evolutionary information. These repeated residues are most often tryptophan. We chose to use a scoring scheme based on the repeated rare residues (Profile Discriminatory Quality, see below) in order to make comparisons, because this score better highlighted the clear differences between the different databases used in the study.</p></sec><sec><title>Profile discriminatory quality</title><p>We calculated the discriminatory quality of the three databases used for the query proteins searches as per the methods section. Discriminatory quality was the percentage of the optimal sequence that was not made up of tryptophan or cysteine repeats. If the optimal sequences are free of cysteines and tryptophans, the profile discriminatory quality will be 100. The discriminatory quality of the current databases (Curr-nr) is considerably higher than the Sargasso Sea database. The discriminatory quality score of profiles generated from the Curr-nr database is 93.69 over all query sequences, compared to 88.49 for the SSea-nr database. Therefore searches against the SSea-nr database turn up optimal sequences with almost twice as many tryptophans and cysteines as searches against the current databases.</p><p>The effect of combining the two sequence databases is to make the discriminatory quality of the profiles substantially worse – the Combined-nr database has a discriminatory quality score of just 85.22 over all query sequences. Profile Discriminatory Quality was calculated from profiles generated for all 237 target sequences so the SSea-nr and Combined-nr scores include those PSI-BLAST searches which found no sequences and therefore will have had discriminatory quality scores approaching 100.</p><p>Given the strange composition of the Sargasso Sea sequences in terms of fragments (fig <xref ref-type="fig" rid="F2">2a</xref>) it is quite possible that the presence of fragments is behind the results obtained with the database searches.</p><p>In order to find out if this is the case and what other reasons might be causing the odd behaviour of the Sargasso Sea sequences, we created three more databases that we could use for comparison. Two of the databases were created in order to eliminate as many fragments as possible. First to investigate the effects of fragments on the profiles we created a combined 90% redundant database from the Curr-nr sequences and sections eak, eaj and eai of the Sargasso Sea (Combined_itok). These three sections have a length distribution that is much more similar to that of the Curr-nr and the combined prokaryote databases (see Figure <xref ref-type="fig" rid="F2">2</xref>). This version of the database contained 1,025,174 sequences</p><p>While many of the fragments from the Sargasso Sea were eliminated while creating the Combined_itok database, it was clear from Fig <xref ref-type="fig" rid="F2">2b</xref> that there are still a number of fragments in sections eak, eaj and eai of the Sargasso Sea resource. We attempted to remove as many fragments as possible, this time by creating a Sargasso Sea resource with a minimum sequence size of 250 residues. Although not all the fragments will have been removed, the smallest fragments will have been taken out. The non-redundant database created in this way (Combined_GT250) had 1,053,952 sequences, approximately the same size as the Combined_itok database (1,025,174 sequences).</p><p>As a comparison and in order to eliminate the effect of database size we generated an updated version of the Curr-nr database, this time with sequences from the April 10, 2005 version of the combined SWISSPROT, TREMBL and TREMBLnew databases. This database contained no sequences at all from the Sargasso Sea resource and had 1,005,858 sequences, almost the same size as the two databases created above.</p><p>The profile discriminatory quality for these two new databases was measured as in the methods section. The results are shown in Table <xref ref-type="table" rid="T1">1</xref>. It seems that increasing size of the search database makes little difference to discriminatory quality. What does make a difference to the discriminatory quality is the fragment content.</p><p>The mean discriminatory quality of the profiles generated from the Curr-nr (April 2005) and Curr-nr (April 2004) databases are almost identical despite the difference in database size. Despite the fact that the Curr-nr (April 2005) and Combined_itok are practically the same size, the mean discriminatory quality of the profiles generated from the latter are considerably worse. This confirms that the strange effects that are being seen in the Combined_itok and Combined-nr databases are not simply the result of adding new sequences to the existing databases.</p><p>While reducing the number of fragments by using the better quality sequences from sections eaj, eai and eak does have the effect of improving the mean discriminatory quality of the profiles, the improvement is by less than one point. However, removing those fragments with fewer than 250 residues improves the discriminatory quality score by 6.5 points compared to the Combined-nr database and by more than 5.5 points combined to the Combined_itok database.</p><p>One other difference between the results from these two databases was that there were more Curr-nr sequences found from searches of the fragment-poor Combined_GT250 database (778 sequences on average) than from the fragment-rich Combined_itok database (631 sequences). The fact that more Curr.-nr sequences were found with the higher quality Combined_GT250 database suggests that removing most of the fragments increases the searching power of PSI-BLAST.</p><p>The profile discriminatory score for the Combined_GT250 database is still not as good as that of the Curr-nr (April 2005) database, but this difference is almost certainly due to the fact that the Combined_GT250 database still contained fragments, fragments that were greater than 250 residues in length and that affected profiles generated for the longest of the query sequences.</p><p>As a further test we also created our own fragmentised database with the April 2005 version of the Curr-nr database. All sequences were split randomly once and the largest part of the sequence retained for the database. Searches with the 237 query sequences of this database gave profile discriminatory scores of 89.1, compared to the 93.36 of the non-fragmented database, confirmation that a simple fragmentation of the current databases was enough to recreate much of the deleterious effects of the Sargasso Sea sequences.</p></sec><sec><title>An example: the Ftsa family</title><p>We looked at one family in particular, the Ftsa family. Ftsa is essential for bacterial cell division. We took the <italic>Thermotoga Maritima </italic>Ftsa sequence from the solved PDB structure 1ef4A and used it as the query in a PSI-BLAST search of the Combined-nr database. 1ef4A has 419 residues. There are three clear ATP-binding motifs, one in the N-terminal, one in the centre of the sequence and one at the C-terminal end of the sequence [<xref ref-type="bibr" rid="B10">10</xref>].</p><p>The sequences found in each round were aligned using CLUSTALW [<xref ref-type="bibr" rid="B11">11</xref>]. The results were instructive even in the first iteration (effectively just a BLAST search). BLAST found 165 sequences from the combined database, 74 from the Curr-nr database and 91 from the SSea-nr database. The alignment showed that all the sequences bar one from sections "eaa" to "eah" of the Sargasso Sea database were fragments that were not long enough to align all three binding motifs. In fact even 20 of the 42 sequences detected from sections "eak" to "eai" of the Sargasso Sea database also turned out to be fragments and not long enough to align all three motifs. None of the Curr-nr sequences were fragments.</p><p>Interestingly one of the 22 whole protein sequences from the Sargasso Sea found by the BLAST search had mutations in each of the three motifs that were not apparent in the sequences from the Curr-nr database, a second whole sequence had mutations in two of the motifs and a number of the fragments from the SSea-nr database had unique and distinct mutations at the C-terminal motif. This suggested that there might indeed be sequence variations in the families present in the Sargasso Sea resource, although clearly this effect was from a single example and on a large scale any such effects were being drowned out by the poor alignments and the high fragment content.</p><p>In the second round CLUSTALW was no longer able to produce a good alignment of the 776 sequences found. While the first motif is relatively well aligned, the central motif is not at all aligned. The alignment from MUSCLE [<xref ref-type="bibr" rid="B12">12</xref>] is somewhat better, though far from perfect. The central motif is well conserved in this alignment. The N-terminal and C-terminal motifs are not well conserved, though more than half the sequences found in the second round are too short to have both the N-terminal motif and the C-terminal motif. One fragment has only 32 residues.</p><p>The results from the second round show how the fragments invade the profile and begin to destroy the discriminatory quality. After 4 rounds the N-terminal and C-terminal motifs are still recognisable in the optimal sequence calculated from the PSI-BLAST profile, but the central motif has disappeared.</p></sec><sec><title>Additional features of the Sargasso Sea resource</title><p>During the sequence analysis we detected a number of interesting differences with the standard behaviour of sequence families from the current databases. Here we describe some of them, with particular emphasis on the influence of the high proportion of fragments in the resource. Differences may be related to differences in family distributions or may simply be due to the influence of the anomalous sequence size distribution.</p></sec><sec><title>Regions of low complexity</title><p>We ran the low complexity detection program SEG [<xref ref-type="bibr" rid="B13">13</xref>] for all the sequences in both the Curr-nr and SSea-nr databases in order to detect regions of low complexity in the sequences that might be biasing the PSI-BLAST searches. SEG finds that in fact the Sargasso Sea sequences have proportionally more complexity than the sequences in the current databases (Curr-nr). 5.7% of the Sargasso Sea sequences are masked by SEG, compared to almost 8% of the sequences from Curr-nr (see Figure <xref ref-type="fig" rid="F1">1</xref>). A database composed of all prokaryotic sequences from complete genomes had just 6.2% of residues in SEG-defined low complexity regions, suggesting that the complexity of the Sargasso Sea sequences was in line with what would be expected. PSI-BLAST searches with the query sequences were carried out both with SEG on and off. It made little difference to final result.</p></sec><sec><title>Sequence clustering</title><p>To assess the distribution of the Sargasso Sea sequences in relation to the rest of the known sequences, we collected all sequences found from the BLAST searches of the Combined-nr database (as above). Fig. <xref ref-type="fig" rid="F7">7</xref> shows the results of the distribution of the E-values from the BLAST searches. These scores are a measure of the similarity between the detected sequence and the query.</p><p>Even though the Combined-nr database contained substantially more Curr-nr sequences than SSea-nr sequences, BLAST detected as many Sargasso Sea sequences as Curr-nr sequences. However, the Sargasso Sea sequences were found with substantially higher E-values. While the shapes of the two distributions are similar, the Sargasso Sea sequence distribution is shifted relative to the Curr-nr sequence distribution and it has a higher mean E-value (lower level of sequence similarity with the target sequence). If there were no length bias in the Sargasso Sea database, this behaviour would indicate that the Sargasso Sea sequences were more divergent. However, the higher E-values of the Sargasso Sea sequences is also likely to be due to the amount of sequence fragments in the databases since in BLAST the shorter the alignment, the higher the E-values in general. This shows too that even the results of BLAST searches with Sargasso Sea resource should be treated with extreme caution.</p></sec><sec><title>Redundancy</title><p>To investigate further the different sequence distribution we created a sequence database from all the bacterial and archaea sequences in the SWISSPROT and TREMBL databases. As a comparison we created a second database of a similar size from sequences from groups "ead" to "eak" of the Sargasso Sea. Both databases contained approximately 780,000 sequences. We used cd-hit [<xref ref-type="bibr" rid="B14">14</xref>] to create non-redundant databases for the two at 90, 80, 70, 60 and 50%. The results are shown in Table <xref ref-type="table" rid="T2">2</xref>.</p><p>It is clear that the Sargasso Sea sequences have more redundancy. This might perhaps not be surprising given their dependence on a very unique ecosystem and might it would be easy to leap to wrong biological conclusions, but again it might be wrong to interpret the results this way since this pattern would also be typical of a database composed of fragments of sequences.</p></sec><sec><title>Sequences in homology modelling</title><p>The capacity to build models by homology is one of the techniques that have improved over recent years, in part due to the expansion of the sequence databases. The different organisation of the Sargasso Sea sequences with respect to the previously known databases might affect this capacity. To assess this question we compared the accuracy of the alignments that could be obtained with and without the Sargasso Sea sequences.</p><p>The addition of the new sequences adversely affected the quality of the pair-wise target-template alignments in 12 of the 32 cases we tested. In these cases the difference in the number of correctly aligned residues in the pairwise alignments implied by the multiple sequence alignment was an average of 17%. In 11 cases the differences were small (no more than 1%), and in just seven cases the alignment shows a modest improvement (an average of 10%), demonstrating again that far from improving the quality of the models, the Sargasso Sea sequences have a tendency to decrease model quality. Again this could be a consequence of sequence fragments.</p></sec></sec><sec><title>Conclusion</title><p>The sequences from the Sargasso Sea differ markedly from those currently in the databases, only 11,700 sequences (under 2%) of the 90% redundant Sargasso Sea database overlapped at 90% identity with the equivalent sequences from the current databases. In addition the new sequences have a much higher isoleucine, asparagine and lysine content and are considerably shorter on average than the sequences currently in the databases. This last observation is due to the sequence fragments from incomplete ORFs that are found in all sections of the Sargasso Sea resource.</p><p>The Sargasso Sea sequences form the first large set of environmental sequences released to the databases and it is therefore interesting to investigate the consequences of adding a great number of sequences from a radically different environment to the protein families in the current databases. For example, some of these new environmental sequences may well occupy distinct and differentiated regions of sequence space at the periphery of the previously known protein families, or may be effective at populating sparsely-populated protein sequence space.</p><p>From a practical point of view, more sequences in the databases ought to lead to more powerful automatic tools for sequence searching, creating multiple alignments and predicting function by linking clusters in sequence space. In particular it is a commonly held belief that the growth of sequence databases has increased and will continue to increase our capacity to define protein families [<xref ref-type="bibr" rid="B15">15</xref>], propose new functions [<xref ref-type="bibr" rid="B16">16</xref>,<xref ref-type="bibr" rid="B17">17</xref>], predict binding sites [<xref ref-type="bibr" rid="B18">18</xref>], predict secondary structure [<xref ref-type="bibr" rid="B19">19</xref>] and derive models by homology [<xref ref-type="bibr" rid="B8">8</xref>].</p><p>We analysed whether the new sequences fulfilled their promise and to what extent they could be assigned to known families from the standard databases. However, we found that the high proportion of sequence fragments in the resource made it impossible to reach any conclusions about the sequence distribution. In addition the new sequences result in more profile drift, a decrease in the quality of pairwise and multiple alignments, more difficulty in detecting homologues and defining families and conserved functional regions.</p><p>Our results show that PSI-BLAST multiple alignments built from these sequences tend to have large, poorly aligned regions with little conservation and low entropy. These "dead" zones of poor conservation and low entropy are characterised by repeated rare residues in the optimal sequences drawn from the profiles. The dead zones indicate where profiles have lost evolutionary information and search power – in fact those profiles that contained large dead zones also often found fewer sequences with successive PSI-BLAST iterations.</p><p>PSI-BLAST has many well documented flaws [<xref ref-type="bibr" rid="B20">20</xref>], none of which were found to have had any bearing on the overall results. The strange results are almost certainly an example of severe database contamination. The poor quality of the alignments generated from the Sargasso Sea sequences were universal, the other two multiple methods used in this study, CLUSTALW and MUSCLE, also had great difficulty in aligning related Sargasso Sea sequences. Nor are hidden Markov model methods much more successful at generating profiles with Sargasso Sea sequences (A. Rojas, personal communication).</p><p>We have shown conclusively that the peculiar behaviour of the Sargasso Sea sequences in this study is caused by the high proportion of sequence fragments. The fragments adversely affect the building of multiple sequence alignments and profiles. The results show that even where sequences can be clustered into sequence families recognisable from PSI-BLAST searches, the fragmentary nature of the new sequences often distorts the multiple alignments to such an extent that family characteristics are lost.</p><p>Chen and Pachter [<xref ref-type="bibr" rid="B21">21</xref>] have recently highlighted the problems of including partial, fragmented sequences from environmental sequence projects in phylogenetic analyses and in multiple sequence alignments. They describe the problem as an extreme case of the missing data problem [<xref ref-type="bibr" rid="B22">22</xref>]. This is almost certainly what is happening here as well. Since almost all multiple sequence alignment methods penalise terminal gaps, they are not good at aligning sequences if there is a high proportion of partial, fragmented sequences in the sequences to be aligned.</p><sec><title>The practical consequence of the Sargasso Sea sequences for bioinformatics tools</title><p>As we have shown here, the quality of the sequences in the Sargasso Sea resource means that it is difficult to carry out large scale investigations into whether these sequences represent a discontinuity in the previously known protein sequence space, or whether our knowledge is biased towards the small corner of the ecosphere we know about.</p><p>When first released these environmental sequences were included in many of the public searchable databases, and for a time results from the main publicly available BLAST servers were tainted by the sequences. They have since been removed from all the main web-based BLAST servers [<xref ref-type="bibr" rid="B23">23</xref>]. These results justify the decision to remove them on the grounds that the fragments were distorting the searches and the profiles.</p><p>The expansion represented by these environmental sequences exposes certain limitations in the current techniques. If researchers are to make use of the new wealth of environmental sequences, how will they deal with the problems caused by the high proportion of sequence fragments if the new sequences are of such poor quality? This is an emerging problem, not only because of the number of environmental sequencing projects currently underway, but also because sequence fragments are being deposited directly into Uniprot by gene annotation projects. Even though they are in smaller number, these sequence fragments are not benign and a number of them have already appeared in expert databases such as Pfam [<xref ref-type="bibr" rid="B24">24</xref>].</p><p>The hope is that these new sequences will push us to improve bioinformatics tools, possibly by developing methods better suited to deal with large numbers of incomplete sequences. Simple, makeshift solutions include filtering databases prior to their use or allowing users to put a length filter on the sequences included in multiple alignments. Meanwhile environmental sequences should be treated with care.</p></sec><sec><title>The quality of the sequences and possible biological conclusion</title><p>The Sargasso Sea sequences are from a range of species subject to the same environmental pressure. This has led researchers to investigate whether there are differences from the current databases. For example, Meyer [<xref ref-type="bibr" rid="B6">6</xref>] used iron-sulphur proteins to suggest that the Sargasso Sea resource showed that microbial diversity has been underestimated by an order of magnitude. While the distribution of the sequences in the Sargasso Sea resource and those in the current databases may indeed be different, the results from this study suggest that additional work may have to be considered before any secure conclusions can be made.</p><p>Indeed the same is true about any study where Sargasso Sea sequences are used in database searches or multiple alignment methods. For example, while one interpretation of the E-value distribution we found in Figure <xref ref-type="fig" rid="F7">7</xref> might be that there is true biological divergence of the sequences, the most likely explanation is that the fragment content of the Sargasso Sea resource is the cause of the higher E-values. In general, the shorter the alignments in BLAST the higher the E-values, so the E-values for the Sargasso Sea sequences must be greater simply because they are fragments. If the E-values of the partial sequences are higher than for the full sequences, then BLAST will automatically find less homologues, and so even BLAST results should be treated with care when the Sargasso Sea resource is used.</p><p>Recently there have been a rash of studies that have used Sargasso Sea sequences in comparisons using BLAST or phylogenetic profiles based on alignments [for example [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B25">25</xref>-<xref ref-type="bibr" rid="B27">27</xref>]]. While the results of these studies are of great interest, the fact that the Sargasso Sea fragments introduce a bias into such studies may need to be taken into account.</p><p>Despite these reservations we did observe interesting deviations from of the behaviour of normal families in isolated examples that might indicate that there are differences in the distribution of sequence families in the Sargasso Sea resource, small differences that are being masked by the poor quality of the Sargasso Sea sequences. Given the masking effect of the sequence fragments, it is difficult to tell to what extent these small changes are a result of the unique evolutionary pressures on the Sargasso Sea sequences and to what extent they might be due to errors resulting from the low coverage depth of the shotgun sequencing techniques used to sequence the Sargasso Sea sequences.</p><p>In the future the sequences from new environmental genomics initiatives may still provide us with invaluable insights into some of the key issues in evolution. In particular, the flooding of the databases with sequences from environmental sequencing projects may impact on key predictions for the total number of families and folds [<xref ref-type="bibr" rid="B28">28</xref>-<xref ref-type="bibr" rid="B30">30</xref>] and the number of structures needed to cover the sequence space by structural genomics efforts [<xref ref-type="bibr" rid="B31">31</xref>,<xref ref-type="bibr" rid="B32">32</xref>]. Revisiting these predictions in the light of the sequences from the environmental sequences may make us more aware of where we have reached in our efforts to describe global protein sequence space.</p></sec></sec><sec sec-type="methods"><title>Methods</title><sec><title>Search databases</title><p>The Sargasso Sea database was built from sequences culled from the whole genome shotgun sequencing of the Sargasso Sea from the GenBank database [<xref ref-type="bibr" rid="B33">33</xref>]. A 90% redundant database was created from these sequences with the clustering program cd-hit. There were just over a million sequences in the original resource deposited in GenBank and the non-redundant Sargasso Sea database (SSea-nr) contained 643,044 sequences.</p><p>There is a small fraction of the Sargasso Sea sequences, fewer than 100 sequences, that contain a non-standard amino acid (marked as X in the sequence), in every case as a result of a translation from the base "N" (any). All these 100 sequences appear in section 6 of the 17 separate environmental sequence files in GenBank, clustered in 4 close groups.</p><p>A local non-redundant database was built from the sequences stored in the SWISSPROT, TREMBL, and TREMBLnew databases as of April 2004, the date of publication of the new sequences. This database was also clustered at 90% redundancy. The resulting non-redundant database (Curr-nr) contained 783,110 protein sequences.</p><p>A third non-redundant database was built by amalgamating the two non-redundant databases. This combined database (Combined-nr) had 1,414,454 sequences at 90% identity after clustering with cd-hit. Only 11,700 sequences (0.82% of the whole database) were removed by cd-hit during the process.</p></sec><sec><title>Query sequences</title><p>For the experiments involving sequence search methods we needed target sequences. We took 87 query sequences from CASP experiments 5 [<xref ref-type="bibr" rid="B34">34</xref>] and 6 [<xref ref-type="bibr" rid="B35">35</xref>] and selected 150 more query sequences from the PDB. These PDB sequences had been used in a previous study [<xref ref-type="bibr" rid="B18">18</xref>] because each of them had a remotely homologous PDB template that contained site residue information. A total set of 237 query sequences were used in the study.</p></sec><sec><title>Creating the profiles for scoring the alignments</title><p>PSI-BLAST sequence profiles were generated from all three non-redundant databases for each of the 237 query sequences. The profiles were generated by running PSI-BLAST for four iterations and with the default options. The profiles were used to deduce the optimal sequences for each of the target sequences and each of the databases.</p></sec><sec><title>Profile discriminatory quality</title><p>In a number of cases PSI-BLAST actually started to find fewer sequences with successive iterations of the databases. We assessed the profiles generated by PSI-BLAST for these sequences and the optimal sequences extracted from the PSI-BLAST multiple alignments. We found that the optimal sequences in these cases were characterised by their low complexity and by very high proportions of tryptophans and cysteines. High proportions of tryptophans and cysteines in the optimal sequences are a side effect of the loss of discriminatory quality in sequence profiles.</p><p>We used the quantity of cysteines and tryptophans in the optimal sequences from profile generated by PSI-BLAST in order to generate a measure of the discriminatory quality of each profile. Profile discriminatory quality here is defined as:</p><p>100 - (W + C-wb-cb)</p><p>where W is the percentage of tryptophans in the optimal sequences, C the percentage of cysteines in the optimal sequences and wb and wc the background percentages of cysteines and tryptophans in the sequence database that PSI-BLAST used to build the profile. If the discriminatory quality of the profiles were perfect there would be no more cysteines and tryptophans in the optimal sequences than in the sequence databases and the profile discriminatory quality would be 100.</p></sec><sec><title>Sargasso Sea sequences in comparative modelling</title><p>The Sargasso Sea resource was also used to create alignments for the purposes of building 3D structural models. 31 domains from 27 CASP 4 [<xref ref-type="bibr" rid="B36">36</xref>] and CASP 5 [<xref ref-type="bibr" rid="B8">8</xref>] comparative modelling targets from a previous study [<xref ref-type="bibr" rid="B37">37</xref>] were used for the comparison. The targets were chosen because they were targets for which PSI-BLAST had been able to identify a structural template at the time of the CASP experiments. The best template for each of the considered CASP 4 and CASP 5 targets was defined as the protein with the highest structural similarity with the target structure according to the LGA structural alignment method [<xref ref-type="bibr" rid="B38">38</xref>].</p><p>Sequences were collected from both the SSea-nr database and the non-redundant databases frozen at the time of the two CASP experiments using the default options of PSI-BLAST and five iterations.</p><p>Sequences were collected that had a percentage sequence identity with the target sequence that was equal or higher to that of the best template and CLUSTALW was used to create multiple alignments, first with the sequences found from the search of the NR databases frozen at the time (the CASP-set of sequences) and then with the CASP-set sequences added to those sequences found from the search of SSea-nr (this set of sequences was called the CASP-SS-set). The pair-wise alignments between target and template implied by the multiple sequence alignment were extracted and compared with the structural alignment. Correctness of the target-template sequence alignments was computed with respect to the LGA structural alignment of the pair.</p></sec></sec><sec><title>Authors' contributions</title><p>MLT conceived of the study, designed the work, carried out the analysis and interpretation of the data and drafted the manuscript. AT and DC designed and carried out the homology modelling section. AT also helped draft the manuscript. AV participated in the design and coordination of the project, was involved in the interpretation of the data and helped to draft the manuscript.</p></sec>
Predicting population coverage of T-cell epitope-based diagnostics and vaccines	<sec><title>Background</title><p>T cells recognize a complex between a specific major histocompatibility complex (MHC) molecule and a particular pathogen-derived epitope. A given epitope will elicit a response only in individuals that express an MHC molecule capable of binding that particular epitope. MHC molecules are extremely polymorphic and over a thousand different human MHC (HLA) alleles are known. A disproportionate amount of MHC polymorphism occurs in positions constituting the peptide-binding region, and as a result, MHC molecules exhibit a widely varying binding specificity. In the design of peptide-based vaccines and diagnostics, the issue of population coverage in relation to MHC polymorphism is further complicated by the fact that different HLA types are expressed at dramatically different frequencies in different ethnicities. Thus, without careful consideration, a vaccine or diagnostic with ethnically biased population coverage could result.</p></sec><sec><title>Results</title><p>To address this issue, an algorithm was developed to calculate, on the basis of HLA genotypic frequencies, the fraction of individuals expected to respond to a given epitope set, diagnostic or vaccine. The population coverage estimates are based on MHC binding and/or T cell restriction data, although the tool can be utilized in a more general fashion. The algorithm was implemented as a web-application available at <ext-link ext-link-type="uri" xlink:href="http://epitope.liai.org:8080/tools/population"/>.</p></sec><sec><title>Conclusion</title><p>We have developed a web-based tool to predict population coverage of T-cell epitope-based diagnostics and vaccines based on MHC binding and/or T cell restriction data. Accordingly, epitope-based vaccines or diagnostics can be designed to maximize population coverage, while minimizing complexity (that is, the number of different epitopes included in the diagnostic or vaccine), and also minimizing the variability of coverage obtained or projected in different ethnic groups.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Bui</surname><given-names>Huynh-Hoa</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Sidney</surname><given-names>John</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A3" contrib-type="author"><name><surname>Dinh</surname><given-names>Kenny</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Southwood</surname><given-names>Scott</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Newman</surname><given-names>Mark J</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A6" corresp="yes" contrib-type="author"><name><surname>Sette</surname><given-names>Alessandro</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Bioinformatics	<sec><title>Background</title><p>T lymphocytes recognize a complex between a specific major histocompatibility complex (MHC) molecule and a particular pathogen-derived epitope. Thus, a given epitope will elicit a response only in individuals that express an MHC molecule capable of binding that particular epitope, explaining to a large extent the phenomenon known as "MHC restriction" [<xref ref-type="bibr" rid="B1">1</xref>]. In humans, MHC molecules are known as human leukocyte antigen (HLA) molecules and two different types exist: class I and class II. HLA class I molecules mostly bind peptides derived from the endogenous processing pathway, and their recognition is primarily associated with cytotoxic T lymphocytes (CTL), which are most important for antiviral and anticancer immunity responses. By contrast, HLA class II molecules bind peptides typically derived from the extracellular milieu, and they are important for helper T lymphocyte (HTL) responses, which regulate antibody and cytotoxic responses.</p><p>HLA molecules are extremely polymorphic. Over a thousand different HLA allelic variants have been defined to date [<xref ref-type="bibr" rid="B2">2</xref>]. Specific HLA alleles are expressed at dramatically different frequencies in different ethnicities [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B4">4</xref>]. Therefore, in the design and development of T-cell epitope-based diagnostics or vaccines, selecting multiple epitopes with different HLA binding specificities will afford increased coverage of the patient population. A pertinent goal, in this context, might be to identify optimal sets of HLA alleles with maximal coverages for different populations [<xref ref-type="bibr" rid="B5">5</xref>,<xref ref-type="bibr" rid="B6">6</xref>]. Extensive analyses by Longmate and coworkers [<xref ref-type="bibr" rid="B7">7</xref>] suggested that 90% population coverage of several ethnic groups can be achieved by targeting eleven different HLA molecules. However, 90% coverage of African and Asian ethnicities required four or more additional molecules. Dawson et al. also analyzed the problem [<xref ref-type="bibr" rid="B8">8</xref>] and concluded that to reach 80% coverage, 3 to 5 HLA molecules were required in a given ethnicity, but the actual HLA specificities required were different in different ethnic groups.</p><p>An important consideration in the process of epitope selection for a T-cell epitope-based diagnostic or vaccine is that the patient population coverage afforded by a given epitope set does not simply correspond to the sum of the coverage of the individual components. To calculate the coverage afforded by a given set of epitopes with multiple and/or overlapped HLA binding specificities, a more comprehensive approach, taking into account MHC binding and T cell recognition patterns, is required for this purpose. A suitable algorithm was previously utilized [<xref ref-type="bibr" rid="B9">9</xref>-<xref ref-type="bibr" rid="B11">11</xref>] but not described in detail. This method calculates the fraction of individuals predicted to respond to a given epitope or epitope set on the basis of HLA genotypic frequencies and on the basis of MHC binding and/or T cell restriction data. In this paper, we describe the algorithm and its implementation as a web application available to the public. We believe this is a useful tool to aid in the design and development of T-cell epitope-based diagnostics and vaccines intended to be effective across diverse populations.</p></sec><sec><title>Implementation</title><p>For a given HLA gene locus, let {<italic>m</italic><sub>1</sub>, <italic>m</italic><sub>2</sub>, ..., <italic>m</italic><sub><italic>N</italic></sub>} denote a set of MHC alleles, with each allele associated with a genotypic frequency <italic>G</italic>(<italic>m</italic><sub><italic>i</italic></sub>) for a population or ethnic group. To account for 100% of alleles of a given locus, the total genotypic frequency (∑<italic>G</italic>(<italic>m</italic><sub><italic>i</italic></sub>)) should add up to 1. If ∑<italic>G</italic>(<italic>m</italic><sub><italic>i</italic></sub>) is less than 1, an unidentified HLA allele with a genotypic frequency equal to the residual (1 - ∑<italic>G</italic>(<italic>m</italic><sub><italic>i</italic></sub>)) is added to the locus. If ∑<italic>G</italic>(<italic>m</italic><sub><italic>i</italic></sub>) is greater than 1, the genotypic frequency of each <italic>m</italic><sub><italic>i </italic></sub>allele of the locus is scaled down proportionately by dividing the frequency by ∑<italic>G</italic>(<italic>m</italic><sub><italic>i</italic></sub>). Next, let {<italic>e</italic><sub>1</sub>, <italic>e</italic><sub>2</sub>, ..., <italic>e</italic><sub><italic>K</italic></sub>} denote a set of epitopes with known MHC binding or restriction data. For each epitope <italic>e</italic><sub><italic>k</italic></sub>, its restriction to an MHC allele <italic>m</italic><sub><italic>i</italic></sub>, <italic>e</italic><sub><italic>k</italic></sub>(<italic>m</italic><sub><italic>i</italic></sub>), is defined as followed:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M1" name="1471-2105-7-153-i1" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:msub> <mml:mi>e</mml:mi> <mml:mi>k</mml:mi> </mml:msub> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo stretchy="false">)</mml:mo> <mml:mo>=</mml:mo> <mml:mrow> <mml:mo>{</mml:mo> <mml:mrow> <mml:mtable columnalign="left"> <mml:mtr columnalign="left"> <mml:mtd columnalign="left"> <mml:mn>0</mml:mn> </mml:mtd> <mml:mtd columnalign="left"> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>f</mml:mi> <mml:mi/> <mml:msub> <mml:mi>e</mml:mi> <mml:mi>k</mml:mi> </mml:msub> <mml:mtext> is not restricted to </mml:mtext> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> </mml:mrow> </mml:mtd> </mml:mtr> <mml:mtr columnalign="left"> <mml:mtd columnalign="left"> <mml:mn>1</mml:mn> </mml:mtd> <mml:mtd columnalign="left"> <mml:mrow> <mml:mtext>if </mml:mtext> <mml:msub> <mml:mi>e</mml:mi> <mml:mi>k</mml:mi> </mml:msub> <mml:mtext> is restricted to </mml:mtext> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> </mml:mrow> </mml:mtd> </mml:mtr> </mml:mtable> </mml:mrow> </mml:mrow> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>1</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>.</mml:mo> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGLbqzdaWgaaWcbaGaem4AaSgabeaakiabcIcaOiabd2gaTnaaBaaaleaacqWGPbqAaeqaaOGaeiykaKIaeyypa0ZaaiqaaeaafaqaaeGacaaabaGaeGimaadabaWexLMBbXgBcf2CPn2qVrwzqf2zLnharyGvLjhzH5wyaGabaiaa=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@8AEA@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>First, for each MHC allele (<italic>m</italic><sub><italic>i</italic></sub>), a total number of epitope "hits", <italic>H</italic>(<italic>m</italic><sub><italic>i</italic></sub>), was tabulated by adding the number of epitopes that are restricted to (or bound by) <italic>m</italic><sub><italic>i</italic></sub>:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M2" name="1471-2105-7-153-i2" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>H</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo stretchy="false">)</mml:mo> <mml:mo>=</mml:mo> <mml:mstyle displaystyle="true"> <mml:munderover> <mml:mo>∑</mml:mo> <mml:mrow> <mml:mi>k</mml:mi> <mml:mo>=</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> <mml:mi>K</mml:mi> </mml:munderover> <mml:mrow> <mml:msub> <mml:mi>e</mml:mi> <mml:mi>k</mml:mi> </mml:msub> </mml:mrow> </mml:mstyle> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo stretchy="false">)</mml:mo> <mml:mtext>  </mml:mtext> <mml:mo stretchy="false">(</mml:mo> <mml:mi>i</mml:mi> <mml:mo>=</mml:mo> <mml:mn>1</mml:mn> <mml:mo>,</mml:mo> <mml:mo>⋯</mml:mo> <mml:mo>,</mml:mo> <mml:mi>N</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>2</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>.</mml:mo> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGibascqGGOaakcqWGTbqBdaWgaaWcbaGaemyAaKgabeaakiabcMcaPiabg2da9maaqahabaGaemyzau2aaSbaaSqaaiabdUgaRbqabaaabaGaem4AaSMaeyypa0JaeGymaedabaGaem4saSeaniabggHiLdGccqGGOaakcqWGTbqBdaWgaaWcbaGaemyAaKgabeaakiabcMcaPiabbccaGiabbccaGiabcIcaOiabdMgaPjabg2da9iabigdaXiabcYcaSiabl+UimjabcYcaSiabd6eaojabcMcaPiaaxMaacaWLjaWaaeWaaeaacqaIYaGmaiaawIcacaGLPaaacqGGUaGlaaa@51B1@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>Next, for each possible diploid MHC combination (<italic>m</italic><sub><italic>i</italic></sub>, <italic>m</italic><sub><italic>j</italic></sub>), a phenotypic frequency <italic>F</italic>(<italic>m</italic><sub><italic>i</italic></sub>, <italic>m</italic><sub><italic>j</italic></sub>) was calculated based on individual allele genotypic frequency:</p><p><italic>F</italic>(<italic>m</italic><sub><italic>i</italic></sub>, <italic>m</italic><sub><italic>j</italic></sub>) = <italic>G</italic>(<italic>m</italic><sub><italic>i</italic></sub>) × <italic>G</italic>(<italic>m</italic><sub><italic>j</italic></sub>)     (3)</p><p>For <italic>n </italic>MHC types, this corresponds to an <italic>n </italic>× <italic>n </italic>tabulation of the phenotypic frequency at which each specific pair of MHCs will be found in the population from which the MHC frequencies were derived. A similar table was also generated to contain the number of epitope hits per each of the MHC combinations <italic>H</italic>(<italic>m</italic><sub><italic>i</italic></sub>, <italic>m</italic><sub><italic>j</italic></sub>). In the case of heterozygous combinations, <italic>H</italic>(<italic>m</italic><sub><italic>i</italic></sub>, <italic>m</italic><sub><italic>j</italic></sub>) was calculated as the sum of the number of epitope hits associated with each of the two alleles, <italic>H</italic>(<italic>m</italic><sub><italic>i</italic></sub>) + <italic>H</italic>(<italic>m</italic><sub><italic>j</italic></sub>). This is because <italic>m</italic><sub><italic>i </italic></sub>and <italic>m</italic><sub><italic>j </italic></sub>are two different alleles, and therefore the number of epitope hits recognized by each allele in the combination is independent of each other. However, in the case of homozygous combinations which contain two identical alleles, the number of epitope hits was the same as the number of epitope hits of the given allele:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M3" name="1471-2105-7-153-i3" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>H</mml:mi> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo>,</mml:mo> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>=</mml:mo> <mml:mrow> <mml:mo>{</mml:mo> <mml:mrow> <mml:mtable columnalign="left"> <mml:mtr columnalign="left"> <mml:mtd columnalign="left"> <mml:mrow> <mml:mi>H</mml:mi> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>+</mml:mo> <mml:mi>H</mml:mi> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>j</mml:mi> </mml:msub> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> </mml:mtd> <mml:mtd columnalign="left"> <mml:mrow> <mml:mtext>if </mml:mtext> <mml:mi>i</mml:mi> <mml:mo>≠</mml:mo> <mml:mi>j</mml:mi> </mml:mrow> </mml:mtd> </mml:mtr> <mml:mtr columnalign="left"> <mml:mtd columnalign="left"> <mml:mrow> <mml:mi>H</mml:mi> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> </mml:mtd> <mml:mtd columnalign="left"> <mml:mrow> <mml:mtext>if </mml:mtext> <mml:mi>i</mml:mi> <mml:mo>=</mml:mo> <mml:mi>j</mml:mi> </mml:mrow> </mml:mtd> </mml:mtr> </mml:mtable> </mml:mrow> </mml:mrow> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>4</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>.</mml:mo> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@5DB7@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>Based on the calculated <italic>F</italic>(<italic>m</italic><sub><italic>i</italic></sub>, <italic>m</italic><sub><italic>j</italic></sub>) and <italic>H</italic>(<italic>m</italic><sub><italic>i</italic></sub>, <italic>m</italic><sub><italic>j</italic></sub>) tables, a frequency distribution was assembled by tabulating the phenotypic frequencies of all MHC combinations associated with a certain number of epitope/HLA combination hits (<italic>h</italic>):</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M4" name="1471-2105-7-153-i4" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>F</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>h</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo>=</mml:mo> <mml:mstyle displaystyle="true"> <mml:munderover> <mml:mo>∑</mml:mo> <mml:mrow> <mml:mi>i</mml:mi> <mml:mo>=</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> <mml:mi>N</mml:mi> </mml:munderover> <mml:mrow> <mml:mstyle displaystyle="true"> <mml:munderover> <mml:mo>∑</mml:mo> <mml:mrow> <mml:mi>j</mml:mi> <mml:mo>=</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> <mml:mi>N</mml:mi> </mml:munderover> <mml:mrow> <mml:mi>F</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo>,</mml:mo> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>j</mml:mi> </mml:msub> <mml:mo stretchy="false">)</mml:mo> <mml:msub> <mml:mi>I</mml:mi> <mml:mrow> <mml:mo>{</mml:mo> <mml:mi>H</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo>,</mml:mo> <mml:msub> <mml:mi>m</mml:mi> <mml:mi>j</mml:mi> </mml:msub> <mml:mo stretchy="false">)</mml:mo> <mml:mo>=</mml:mo> <mml:mi>h</mml:mi> <mml:mo>}</mml:mo> </mml:mrow> </mml:msub> </mml:mrow> </mml:mstyle> </mml:mrow> </mml:mstyle> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>5</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>,</mml:mo> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGgbGrcqGGOaakcqWGObaAcqGGPaqkcqGH9aqpdaaeWbqaamaaqahabaGaemOrayKaeiikaGIaemyBa02aaSbaaSqaaiabdMgaPbqabaGccqGGSaalcqWGTbqBdaWgaaWcbaGaemOAaOgabeaakiabcMcaPiabdMeajnaaBaaaleaacqGG7bWEcqWGibascqGGOaakcqWGTbqBdaWgaaadbaGaemyAaKgabeaaliabcYcaSiabd2gaTnaaBaaameaacqWGQbGAaeqaaSGaeiykaKIaeyypa0JaemiAaGMaeiyFa0habeaaaeaacqWGQbGAcqGH9aqpcqaIXaqmaeaacqWGobGta0GaeyyeIuoaaSqaaiabdMgaPjabg2da9iabigdaXaqaaiabd6eaobqdcqGHris5aOGaaCzcaiaaxMaadaqadaqaaiabiwda1aGaayjkaiaawMcaaiabcYcaSaaa@5DB8@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>where <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M5" name="1471-2105-7-153-i5" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mi>I</mml:mi><mml:mrow><mml:mo>{</mml:mo><mml:mi>H</mml:mi><mml:mo stretchy="false">(</mml:mo><mml:msub><mml:mi>m</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>,</mml:mo><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy="false">)</mml:mo><mml:mo>=</mml:mo><mml:mi>h</mml:mi><mml:mo>}</mml:mo></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mrow><mml:mo>{</mml:mo><mml:mrow><mml:mtable><mml:mtr><mml:mtd><mml:mn>1</mml:mn></mml:mtd><mml:mtd><mml:mrow><mml:mtext>if </mml:mtext><mml:mi>H</mml:mi><mml:mo stretchy="false">(</mml:mo><mml:msub><mml:mi>m</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>,</mml:mo><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy="false">)</mml:mo><mml:mo>=</mml:mo><mml:mi>h</mml:mi></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mn>0</mml:mn></mml:mtd><mml:mtd><mml:mrow><mml:mtext>if </mml:mtext><mml:mi>H</mml:mi><mml:mo stretchy="false">(</mml:mo><mml:msub><mml:mi>m</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>,</mml:mo><mml:msub><mml:mi>m</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy="false">)</mml:mo><mml:mo>≠</mml:mo><mml:mi>h</mml:mi></mml:mrow></mml:mtd></mml:mtr></mml:mtable></mml:mrow></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@6092@</mml:annotation></mml:semantics></mml:math></inline-formula> is an indicator function.</p><p>For calculation of coverage by epitope sets restricted to MHC alleles of multiple <italic>k </italic>different loci, a combined frequency distribution (<italic>P</italic>) as a function of epitope/HLA combination hits (<italic>n</italic>) was generated by merging <italic>k </italic>separate frequency distributions. This merging procedure is based on the assumption that linkages between MHC loci are in equilibrium, and was done as follows:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M6" name="1471-2105-7-153-i6" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>P</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo>=</mml:mo> <mml:mstyle displaystyle="true"> <mml:munder> <mml:mo>∑</mml:mo> <mml:mrow> <mml:msub> <mml:mi>h</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo>≥</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> </mml:munder> <mml:mrow> <mml:mo>⋯</mml:mo> <mml:mstyle displaystyle="true"> <mml:munder> <mml:mo>∑</mml:mo> <mml:mrow> <mml:msub> <mml:mi>h</mml:mi> <mml:mi>k</mml:mi> </mml:msub> <mml:mo>≥</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> </mml:munder> <mml:mrow> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mstyle displaystyle="true"> <mml:munderover> <mml:mo>∏</mml:mo> <mml:mrow> <mml:mi>i</mml:mi> <mml:mo>=</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> <mml:mi>k</mml:mi> </mml:munderover> <mml:mrow> <mml:msub> <mml:mi>F</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo stretchy="false">(</mml:mo> <mml:msub> <mml:mi>h</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo stretchy="false">)</mml:mo> <mml:msub> <mml:mi>I</mml:mi> <mml:mrow> <mml:mrow> <mml:mo>{</mml:mo> <mml:mrow> <mml:mstyle displaystyle="true"> <mml:munderover> <mml:mo>∑</mml:mo> <mml:mrow> <mml:mi>i</mml:mi> <mml:mo>=</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> <mml:mi>k</mml:mi> </mml:munderover> <mml:mrow> <mml:msub> <mml:mi>h</mml:mi> <mml:mi>i</mml:mi> </mml:msub> <mml:mo>=</mml:mo> <mml:mi>n</mml:mi> </mml:mrow> </mml:mstyle> </mml:mrow> <mml:mo>}</mml:mo> </mml:mrow> </mml:mrow> </mml:msub> </mml:mrow> </mml:mstyle> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> </mml:mstyle> </mml:mrow> </mml:mstyle> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>6</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>,</mml:mo> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@672D@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>where <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M7" name="1471-2105-7-153-i7" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:msub><mml:mi>I</mml:mi><mml:mrow><mml:mrow><mml:mo>{</mml:mo><mml:mrow><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>k</mml:mi></mml:munderover><mml:mrow><mml:msub><mml:mi>h</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mi>n</mml:mi></mml:mrow></mml:mstyle></mml:mrow><mml:mo>}</mml:mo></mml:mrow></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mrow><mml:mo>{</mml:mo><mml:mrow><mml:mtable><mml:mtr><mml:mtd><mml:mn>1</mml:mn></mml:mtd><mml:mtd><mml:mrow><mml:mtext>if </mml:mtext><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>k</mml:mi></mml:munderover><mml:mrow><mml:msub><mml:mi>h</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mi>n</mml:mi></mml:mrow></mml:mstyle></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mn>0</mml:mn></mml:mtd><mml:mtd><mml:mrow><mml:mtext>if </mml:mtext><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>k</mml:mi></mml:munderover><mml:mrow><mml:msub><mml:mi>h</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>≠</mml:mo><mml:mi>n</mml:mi></mml:mrow></mml:mstyle></mml:mrow></mml:mtd></mml:mtr></mml:mtable></mml:mrow></mml:mrow></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@60DB@</mml:annotation></mml:semantics></mml:math></inline-formula> is an indicator function, and <italic>F</italic><sub><italic>i</italic></sub>(<italic>h</italic><sub><italic>i</italic></sub>) is a phenotypic frequency associated with <italic>h</italic><sub><italic>i </italic></sub>epitope/HLA combination hits of locus <italic>i </italic>calculated from equation 5.</p><p>The population coverage (<italic>C</italic>) or fraction of individuals projected to respond to the epitope set was then calculated as the sum of the combined phenotypic frequencies associated with at least one epitope hit/HLA combination:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M8" name="1471-2105-7-153-i8" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>C</mml:mi> <mml:mo>=</mml:mo> <mml:mstyle displaystyle="true"> <mml:munder> <mml:mo>∑</mml:mo> <mml:mrow> <mml:mi>n</mml:mi> <mml:mo>≥</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> </mml:munder> <mml:mrow> <mml:mi>P</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:mstyle> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>7</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>.</mml:mo> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGdbWqcqGH9aqpdaaeqbqaaiabdcfaqjabcIcaOiabd6gaUjabcMcaPaWcbaGaemOBa4MaeyyzImRaeGymaedabeqdcqGHris5aOGaaCzcaiaaxMaadaqadaqaaiabiEda3aGaayjkaiaawMcaaiabc6caUaaa@3DF6@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>Based on equation 6, a histogram was generated to summarize the fraction of population coverage (<italic>P</italic>) as a function of the number of HLA/epitope combinations (<italic>n</italic>) recognized. A cumulative population coverage distribution frequency (<italic>Y</italic>) as a function of the number of HLA/epitope combinations (<italic>n</italic>) was also calculated:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M9" name="1471-2105-7-153-i9" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>Y</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo>=</mml:mo> <mml:mstyle displaystyle="true"> <mml:munder> <mml:mo>∑</mml:mo> <mml:mrow> <mml:mi>x</mml:mi> <mml:mo>≥</mml:mo> <mml:mi>n</mml:mi> </mml:mrow> </mml:munder> <mml:mrow> <mml:mi>P</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>x</mml:mi> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:mstyle> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>8</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>.</mml:mo> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGzbqwcqGGOaakcqWGUbGBcqGGPaqkcqGH9aqpdaaeqbqaaiabdcfaqjabcIcaOiabdIha4jabcMcaPaWcbaGaemiEaGNaeyyzImRaemOBa4gabeqdcqGHris5aOGaaCzcaiaaxMaadaqadaqaaiabiIda4aGaayjkaiaawMcaaiabc6caUaaa@41D8@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>From this cumulative population coverage distribution of the whole epitope set, <italic>PC</italic>90, defined as the minimum number of epitope/HLA combination hits (<italic>n</italic>) recognized by 90% of the population, was determined as follow:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M10" name="1471-2105-7-153-i10" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>P</mml:mi> <mml:mi>C</mml:mi> <mml:mn>90</mml:mn> <mml:mo>=</mml:mo> <mml:mi>n</mml:mi> <mml:mo>+</mml:mo> <mml:mfrac> <mml:mrow> <mml:mi>Y</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo>−</mml:mo> <mml:mn>0.9</mml:mn> </mml:mrow> <mml:mrow> <mml:mi>Y</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo>−</mml:mo> <mml:mi>Y</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo>+</mml:mo> <mml:mn>1</mml:mn> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:mfrac> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>9</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>,</mml:mo> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGqbaucqWGdbWqcqaI5aqocqaIWaamcqGH9aqpcqWGUbGBcqGHRaWkdaWcaaqaaiabdMfazjabcIcaOiabd6gaUjabcMcaPiabgkHiTiabicdaWiabc6caUiabiMda5aqaaiabdMfazjabcIcaOiabd6gaUjabcMcaPiabgkHiTiabdMfazjabcIcaOiabd6gaUjabgUcaRiabigdaXiabcMcaPaaacaWLjaGaaCzcamaabmaabaGaeGyoaKdacaGLOaGaayzkaaGaeiilaWcaaa@4C50@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>where <italic>Y</italic>(<italic>n</italic>) ≥ 0.9 > <italic>Y</italic>(<italic>n </italic>+ 1). Because) <italic>PC</italic>90 was determined by data interpolation, it can be of any positive decimal value. Based on equation 9, if the population coverage is less than 90% or <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M11" name="1471-2105-7-153-i11" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>C</mml:mi><mml:mo>=</mml:mo><mml:mstyle displaystyle="true"><mml:munder><mml:mo>∑</mml:mo><mml:mrow><mml:mi>n</mml:mi><mml:mo>≥</mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:munder><mml:mrow><mml:mi>P</mml:mi><mml:mo stretchy="false">(</mml:mo><mml:mi>n</mml:mi><mml:mo stretchy="false">)</mml:mo></mml:mrow></mml:mstyle><mml:mo>≡</mml:mo><mml:mi>Y</mml:mi><mml:mo stretchy="false">(</mml:mo><mml:mn>1</mml:mn><mml:mo stretchy="false">)</mml:mo><mml:mo><</mml:mo><mml:mn>0.9</mml:mn></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGdbWqcqGH9aqpdaaeqbqaaiabdcfaqjabcIcaOiabd6gaUjabcMcaPaWcbaGaemOBa4MaeyyzImRaeGymaedabeqdcqGHris5aOGaeyyyIORaemywaKLaeiikaGIaeGymaeJaeiykaKIaeyipaWJaeGimaaJaeiOla4IaeGyoaKdaaa@42C5@</mml:annotation></mml:semantics></mml:math></inline-formula>, <italic>PC</italic>90 will be less than 1.</p><p>Additionally, the average number of epitope/HLA combination hits (A) recognized by the population is a weighted average and was calculated as follow:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M12" name="1471-2105-7-153-i12" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>A</mml:mi> <mml:mo>=</mml:mo> <mml:mstyle displaystyle="true"> <mml:munder> <mml:mo>∑</mml:mo> <mml:mrow> <mml:mi>n</mml:mi> <mml:mo>≥</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> </mml:munder> <mml:mrow> <mml:mi>n</mml:mi> <mml:mo>×</mml:mo> <mml:mi>P</mml:mi> <mml:mo stretchy="false">(</mml:mo> <mml:mi>n</mml:mi> <mml:mo stretchy="false">)</mml:mo> </mml:mrow> </mml:mstyle> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mn>10</mml:mn> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>.</mml:mo> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGbbqqcqGH9aqpdaaeqbqaaiabd6gaUjabgEna0kabdcfaqjabcIcaOiabd6gaUjabcMcaPaWcbaGaemOBa4MaeyyzImRaeGymaedabeqdcqGHris5aOGaaCzcaiaaxMaadaqadaqaaiabigdaXiabicdaWaGaayjkaiaawMcaaiabc6caUaaa@4250@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p></sec><sec><title>Results and discussions</title><p>The Population Coverage Calculation program was implemented as a Java servlet public web-application (see Availability and Requirements section). HLA allele (genotypic) frequencies were obtained from dbMHC database [<xref ref-type="bibr" rid="B12">12</xref>]. At present, dbMHC database provides allele frequencies for 78 populations grouped into 11 different geographical areas. In addition to the allele frequencies obtained from the dbMHC database, the Population Coverage Calculation program also accepts custom populations with allele frequencies defined by users. Multiple population coverages can be simultaneously calculated and an average population coverage is generated. Since MHC class I and MHC class II restricted T cell epitopes elicit immune responses from two different T cell populations (CTL and HTL, respectively), the program provides three calculation options to accommodate different coverage modes – (1) class I separate, (2) class II separate, and (3) class I and class II combined. For each population coverage, a histogram is generated to summarize the percentage distribution of individuals as a function of the number of epitope/HLA combinations recognized. A cumulative coverage distribution plot is also generated to determine the minimum number of epitope/HLA combinations recognized by 90% of the population (PC90). Finally, the average number of epitope/HLA combinations recognized by the population and coverages of individual epitope are also calculated.</p><p>It should be noted that when population coverages are projected from an epitope set restricted to alleles from multiple HLA loci, linkages between loci are taken into account. The overall population (phenotypic frequency), (<italic>P</italic><sub><italic>total</italic></sub>), is mathematically derived as the sum of the individual locus' coverage corrected for the overlaps: <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M13" name="1471-2105-7-153-i13" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>S</mml:mi><mml:mo>=</mml:mo><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>n</mml:mi></mml:munderover><mml:mrow><mml:mfrac><mml:mrow><mml:mi>n</mml:mi><mml:mo>!</mml:mo></mml:mrow><mml:mrow><mml:mi>k</mml:mi><mml:mo>!</mml:mo><mml:mo stretchy="false">(</mml:mo><mml:mi>n</mml:mi><mml:mo>−</mml:mo><mml:mi>k</mml:mi><mml:mo stretchy="false">)</mml:mo><mml:mo>!</mml:mo></mml:mrow></mml:mfrac></mml:mrow></mml:mstyle></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGtbWucqGH9aqpdaaeWbqaamaalaaabaGaemOBa4MaeiyiaecabaGaem4AaSMaeiyiaeIaeiikaGIaemOBa4MaeyOeI0Iaem4AaSMaeiykaKIaeiyiaecaaaWcbaGaem4AaSMaeyypa0JaeGymaedabaGaemOBa4ganiabggHiLdaaaa@4072@</mml:annotation></mml:semantics></mml:math></inline-formula>, where <italic>P</italic><sub><italic>ij </italic></sub>is the frequency of the <italic>ij </italic>haplotype, <italic>P</italic><sub><italic>ijk </italic></sub>is the frequency of the <italic>ijk </italic>haplotype, etc... If gene linkage equilibrium is assumed, <italic>P</italic><sub><italic>ij </italic></sub>can be calculated as the product of the individual allele phenotypic frequencies (<italic>P</italic><sub><italic>i </italic></sub>× <italic>P</italic><sub><italic>j</italic></sub>), and <italic>P</italic><sub><italic>ijk </italic></sub>= <italic>P</italic><sub><italic>i </italic></sub>× <italic>P</italic><sub><italic>j </italic></sub>× <italic>P</italic><sub><italic>k</italic></sub>, etc... This calculation is implicitly incorporated in our current algorithm (equation 6). However, if gene linkage is in disequilibrium, the frequency of a given haplotype is usually not equal to the product of their individual allele phenotypic frequencies, (<italic>P</italic><sub><italic>ij </italic></sub>≠ <italic>P</italic><sub><italic>i </italic></sub>× <italic>P</italic><sub><italic>j</italic></sub>, <italic>P</italic><sub><italic>ijk </italic></sub>≠ <italic>P</italic><sub><italic>i </italic></sub>× <italic>P</italic><sub><italic>j </italic></sub>× <italic>P</italic><sub><italic>k</italic></sub>, ...). As a result, to account for linkage disequilibrium between HLA loci, complete data on haplotype frequencies must be known. Therefore, it would be difficult to factor in linkage disequilibrium at this time because linkage disequilibrium is known to be different in different ethnicities, and data regarding the specific disequilibrium in different ethnicities in general is not available or incomplete. As more comprehensive MHC linkage disequilibrium data becomes available, our method can be modified to incorporate this type of calculation.</p><p>Although the present program assumes linkage equilibrium between HLA loci, the impact of linkage disequilibrium, which is known to occur in the MHC region, on the calculated coverage is expected, in most contexts, to be minimal. For example, in the North American Caucasian population, the A1 and B8 antigens of HLA-A and -B loci, respectively, are known to be the strongest linked antigen pair with an observed haplotype frequency of 7.95% [<xref ref-type="bibr" rid="B13">13</xref>]. The genotypic frequencies of the A1 and B8 antigens are 15.18% and 9.41%, respectively [<xref ref-type="bibr" rid="B13">13</xref>]. Assuming the linkage between A1 and B8 antigens is in equilibrium, the overall population coverage calculated by the present program is 40.97%, and the individual population coverages by A1 and B8 antigens are 28.06% and 17.93%, respectively. The expected equilibrium frequency for the A1/B8 haplotype, in this case, is 5.03% (28.06% × 17.93%) which is 2.92% less than the observed frequency of 7.95%. Therefore, if linkage disequilibrium is considered, the overall population coverage will be 38.04% (28.06% + 17.93% - 7.95%). Thus, even for the most tightly linked A1/B8 haplotype in the Caucasian population, linkage disequilibrium, in this specific example, only accounted for less than 3% difference in the population coverage calculated by the present program. Furthermore, we have also investigated the deviations between the observed and expected equilibrium frequencies of 1012 HLA-A/-B haplotypes in the North American Caucasian population, based on available antigen- and haplotype-frequencies published by Mori <italic>et al</italic>. [<xref ref-type="bibr" rid="B14">14</xref>,<xref ref-type="bibr" rid="B15">15</xref>]. On average, the observed haplotype frequencies deviated from the expected equilibrium frequencies by approximately 0.58%. As a result, linkage disequilibrium is expected to impact the calculated population coverage, but the degree of the impact is expected to be negligible.</p><p>It should be pointed out that the calculations described herein can also be performed on data spreadsheets, but the process is laborious, error prone and also requires extensive immunological expertise. In our experience, a single calculation without the aid of this tool requires several hours to complete. To the best of our knowledge, at this time, there is no existing program that is publicly accessible as a web-resource that can offer the flexibility and range of utility similar to the Population Coverage Calculation program that we have developed. The present application represents a significant enhancement of the dbMHC database's utility by incorporating its compiled data of world-wide ethnic population frequencies to calculate HLA coverage for user-defined population subsets. The program is flexible by allowing the user to specify groups of related or unrelated ethnicities as well as specify the HLA alleles under consideration. Additional flexibility features include the implementation of separate calculations for both MHC Class I and Class II demarcated recognitions as they involve immune responses from two different populations of T cells – CTL and HTL, respectively. The output of the program was also specifically designed to be accessible to both specialists and neophytes in the field of MHC research. Therefore, having this tool publicly available is highly desirable. Additionally, in our future works, we plan to incorporate in the tool the ability to search for minimal epitope subset(s) within the given epitope set that will afford a specified population coverage level. This is not a trivial task due to a large number of possible epitope subsets (S) that has to be considered, <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M14" name="1471-2105-7-153-i14" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mrow><mml:mi>S</mml:mi><mml:mo>=</mml:mo><mml:mstyle displaystyle="true"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>n</mml:mi></mml:munderover><mml:mrow><mml:mfrac><mml:mrow><mml:mi>n</mml:mi><mml:mo>!</mml:mo></mml:mrow><mml:mrow><mml:mi>k</mml:mi><mml:mo>!</mml:mo><mml:mo stretchy="false">(</mml:mo><mml:mi>n</mml:mi><mml:mo>−</mml:mo><mml:mi>k</mml:mi><mml:mo stretchy="false">)</mml:mo><mml:mo>!</mml:mo></mml:mrow></mml:mfrac></mml:mrow></mml:mstyle></mml:mrow><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGtbWucqGH9aqpdaaeWbqaamaalaaabaGaemOBa4MaeiyiaecabaGaem4AaSMaeiyiaeIaeiikaGIaemOBa4MaeyOeI0Iaem4AaSMaeiykaKIaeiyiaecaaaWcbaGaem4AaSMaeyypa0JaeGymaedabaGaemOBa4ganiabggHiLdaaaa@4072@</mml:annotation></mml:semantics></mml:math></inline-formula> where <italic>n </italic>is the total number of epitopes and <italic>k </italic>is the number of epitopes in a subset. For example, for a set of 20 epitopes, there will be a total of 1,048,575 combinations of epitope subsets that needs to be evaluated. Therefore, a strategic searching approach must be devised to computationally accomplish this task. In summary, with the help of this Population Coverage Calculation program, epitope-based vaccines or diagnostics can be designed to maximize population coverage while minimizing complexity (that is, the number of different epitopes included in the diagnostic or vaccine), and also minimizing the variability of coverage obtained or projected in different ethnic groups.</p></sec><sec><title>Conclusion</title><p>Herein, we have implemented a method to calculate projected population coverage of a T-cell epitope-based diagnostic or vaccine using MHC binding or T cell restriction data and HLA gene frequencies. The Population Coverage Calculation program was designed to be user friendly and flexible. Besides the compiled HLA gene frequencies currently provided, users can also supply their own tabulated HLA gene frequencies for calculation. Therefore, researchers can use this tool to perform coverage analyses on their specific patient populations. We plan to continuously update the compiled HLA gene frequencies as more data are available, and thus to provide researchers with a useful tool to aid in the design and development of effective T-cell epitope-based diagnostics and vaccines.</p></sec><sec><title>Availability and requirements</title><p>Project name: Population Coverage Calculation</p><p>Project home page: <ext-link ext-link-type="uri" xlink:href="http://epitope.liai.org:8080/tools/population"/></p><p>Programming language: Java</p><p>Operating system: Fedora Linux</p><p>Other requirements: Apache Tomcat 5.5.12, MySQL 4.1</p><p>Web browser: Population Coverage Calculation program has been tested and shown to work with the following browsers: Firefox version 1.5 (PC and Mac OS X), Netscape version 8.0.4 (PC), Netscape version 7.2 (Mac OS X), Internet Explorer version 6.0 (PC), Internet Explorer version 5.2 for Mac (Mac OS X). Default security settings were used.</p></sec><sec><title>Authors' contributions</title><p>HHB developed the computer algorithm and designed the web-resource. AS and JS contributed the calculation approaches. KD helped with programming and collecting HLA frequency data. SS and MN were involved in conceptualizing the calculation approaches. HHB wrote the manuscript, AS and JS edited the final version. All authors read and approved the manuscript.</p></sec>
XcisClique: analysis of regulatory bicliques	<sec><title>Background</title><p>Modeling of <italic>cis</italic>-elements or regulatory motifs in promoter (upstream) regions of genes is a challenging computational problem. In this work, set of regulatory motifs simultaneously present in the promoters of a set of genes is modeled as a biclique in a suitably defined bipartite graph. A biologically meaningful co-occurrence of multiple <italic>cis</italic>-elements in a gene promoter is assessed by the combined analysis of genomic and gene expression data. Greater statistical significance is associated with a set of genes that shares a common set of regulatory motifs, while simultaneously exhibiting highly correlated gene expression under given experimental conditions.</p></sec><sec sec-type="methods"><title>Methods</title><p>XcisClique, the system developed in this work, is a comprehensive infrastructure that associates annotated genome and gene expression data, models known <italic>cis</italic>-elements as regular expressions, identifies maximal bicliques in a bipartite gene-motif graph; and ranks bicliques based on their computed statistical significance. Significance is a function of the probability of occurrence of those motifs in a biclique (a hypergeometric distribution), and on the new sum of absolute values statistic (SAV) that uses Spearman correlations of gene expression vectors. SAV is a statistic well-suited for this purpose as described in the discussion.</p></sec><sec><title>Results</title><p>XcisClique identifies new motif and gene combinations that might indicate as yet unidentified involvement of sets of genes in biological functions and processes. It currently supports <italic>Arabidopsis thaliana </italic>and can be adapted to other organisms, assuming the existence of annotated genomic sequences, suitable gene expression data, and identified regulatory motifs. A subset of Xcis Clique functionalities, including the motif visualization component MotifSee, source code, and supplementary material are available at <ext-link ext-link-type="uri" xlink:href="https://bioinformatics.cs.vt.edu/xcisclique/"/>.</p></sec>	<contrib id="A1" contrib-type="author"><name><surname>Pati</surname><given-names>Amrita</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A2" contrib-type="author"><name><surname>Vasquez-Robinet</surname><given-names>Cecilia</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A3" corresp="yes" contrib-type="author"><name><surname>Heath</surname><given-names>Lenwood S</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib><contrib id="A4" contrib-type="author"><name><surname>Grene</surname><given-names>Ruth</given-names></name><xref ref-type="aff" rid="I2">2</xref><email>[email protected]</email></contrib><contrib id="A5" contrib-type="author"><name><surname>Murali</surname><given-names>TM</given-names></name><xref ref-type="aff" rid="I1">1</xref><email>[email protected]</email></contrib>	BMC Bioinformatics	<sec><title>Background</title><p>Gene regulation is an intricate, dynamic phenomenon essential for all biological functions including cell metabolism, development, and response to environmental stress and pathogen attack. Primary actors include transcription factors (TFs), which recognize and bind to specific DNA sequences in gene promoters. These DNA sequences are known variously as <italic>cis</italic>-elements, transcription factor binding sites (TFBSs), or regulatory motifs. Here, these terms are used interchangeably.</p><p>The binding strength of a TF for a given <italic>cis</italic>-element depends on the precise DNA sequence, while each <italic>cis</italic>-element has binding affinity for a particular subset of all TFs. The details determining differential receptivity of a transcription factor for different sequences is not yet known, but sequence specificity and conformational changes are likely to be involved [<xref ref-type="bibr" rid="B1">1</xref>]. Regulation of transcription depends on the binding of one or more TFs to corresponding <italic>cis</italic>-elements in a gene promoter, which may initiate, terminate, enhance, or repress transcription. TFs are often large proteins or protein-complexes, and this imposes geometric and spatial constraints on the separation between and arrangement of <italic>cis</italic>-elements [<xref ref-type="bibr" rid="B2">2</xref>]. The rate of transcription, and hence, gene expression depends on the combination of TFs currently bound to the regulatory regions of genes [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B4">4</xref>]. Often, the sequence distance from the TATA box to the <italic>cis</italic>-element binding a TF influences the amount of control that the TF has on gene transcription [<xref ref-type="bibr" rid="B5">5</xref>]. In summary, transcriptional regulation of a gene depends on a number of factors, including these: the <italic>cis</italic>-elements present in the gene promoter; the distances between <italic>cis</italic>-elements; the order of <italic>cis</italic>-elements; and the distance from a <italic>cis</italic>-element to the transcription start site.</p><p>In the past decade, a number of computational tools have been developed to analyze the promoters of various organisms. These tools fall into three broad categories. Tools in the first category discover or identify gene promoters from nucleotide sequences [<xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B6">6</xref>]. Tools in the second category predict putative <italic>cis</italic>-elements in the promoters of a family of genes using pattern discovery and pattern matching techniques. [<xref ref-type="bibr" rid="B7">7</xref>-<xref ref-type="bibr" rid="B11">11</xref>], and [<xref ref-type="bibr" rid="B12">12</xref>], describe and compare such tools that use both enumerative and probabilistic approaches. Tools in the third category model and analyze the presence of combinations of <italic>cis</italic>-elements in gene promoters and the effect of these combinations on the regulation of transcription. Examples of tools in this category are found in [<xref ref-type="bibr" rid="B4">4</xref>,<xref ref-type="bibr" rid="B13">13</xref>,<xref ref-type="bibr" rid="B14">14</xref>], and [<xref ref-type="bibr" rid="B15">15</xref>].</p><p>The XcisClique system has been developed to incorporate genomic and <italic>cis</italic>-element data for <italic>Arabidopsis thaliana </italic>(AT). Pilpel, et al. [<xref ref-type="bibr" rid="B4">4</xref>] identify regulatory networks in yeast by building a database of known and putative yeast TFBSs and identifying synergistic motif combinations based on the expression coherence score of each gene set having a pair of motifs. Their motif-association maps are highly connected, indicating that transcription factors work in combinations to render different expression patterns and that motif co-occurrence is essential for transcriptional regulation. However, they use position weight matrices (PWMs) for modeling binding sites; PWMs for <italic>Arabidopsis thaliana </italic>are rare with TRANSFAC containing PWMs for just 10 AT binding sites. Kato, et al. [<xref ref-type="bibr" rid="B14">14</xref>] integrate chromatin-immunoprecipitation (ChIP) data available for yeast with combinatorial motif analysis to identify over-represented motif combinations. Genome-wide ChIP data are rarely available for other organisms and are not available for AT. Chiang, et al. [<xref ref-type="bibr" rid="B13">13</xref>] identify regulatory templates consisting of pairs of hexamers identified in yeast genomes as conserved in co-occurrence and spatial separation. A common drawback of the probabilistic methods such as those used in [<xref ref-type="bibr" rid="B13">13</xref>] and [<xref ref-type="bibr" rid="B15">15</xref>] is that they consider <italic>n</italic>-mers only (typically hexamers). These methods discover regulatory templates and not the actual motifs. Furthermore, since, in most cases, the specific <italic>cis</italic>-element regions for each of the members of a given transcription factor family have not yet been determined, what is currently available is a consensus sequence serving as a motif rather than a specific sequence. TFBSs in AT vary widely in length; for instance, the heat shock element in <italic>Arabidopsis thaliana (AT) </italic>is 13 nucleotides long, while the ACGTATERD1 (PLACE identifier) motif is 4 nucleotides long.</p><p>Integrating motif discovery into identification of motif combinations leads to the discovery of a very large number of combinations (exponential) of regulatory templates. Also, probabilistic approaches require that the predictive models be trained in an organism-specific manner. The optimal model parameters to use are dependent upon the organism, the tissue type, the regulatory process, and the particular TFBSs. Most probabilistic models use yeast as their model organism. Yeast is a much more widely studied organism as compared with <italic>Arabidopsis </italic>and data for yeast is available on a much larger scale. Because of the above reasons, XcisClique excludes motif discovery from the system and uses only known motifs to identify over-represented motif combinations. A preliminary analysis of spatial conservation of motif-pairs in AT promoters was done to determine the inclusion of spatial conservation of regulatory elements in combinatorial analysis. We did not find any patterns that suggest conservation of spacing between pairs of <italic>cis</italic>-elements in AT. This may be due to limitations of the current known <italic>cis</italic>-elements for AT. So, XcisClique uses the presence of combinations of <italic>cis</italic>-elements to derive regulatory bicliques. <italic>cis</italic>-regulatory motifs are properly represented as strings over the alphabet {<italic>A</italic>, <italic>B</italic>, <italic>C</italic>, <italic>D </italic><italic>G</italic>, <italic>H</italic>, <italic>K</italic>, <italic>N</italic>, <italic>M</italic>, <italic>R</italic>, <italic>S</italic>,<italic>T</italic>, <italic>V</italic>, <italic>W</italic>, <italic>Y</italic>}, the IUPAC recommended alphabet for bases in nucleic acid sequences IN the case of AT, many of these motifs are consensus sequences. [<xref ref-type="bibr" rid="B16">16</xref>]. A <italic>motifset </italic>is any set of regulatory motifs. The presence of the members of a motifset in the promoters of two or more genes have biological significance in that those motifs may participate in the co-regulation of those genes. The number of distinct motifsets present in the promoters of genes in any genome is quite large, typically exponential in the number of motifs considered. Hence, exhaustively analyzing all motifsets is too expensive computationally.</p><p>More naturally, a biologist starts with a <italic>geneset</italic>, a set of genes of interest. Typically, a geneset is small and consists of genes that are co-regulated under some treatments, and the biologist wishes to identify motifsets common to some of the genes that have biological significance. The number of motifsets identified as co-occurring in subsets of the geneset of interest is still, typically, quite large. The computational setting is best expressed as a bipartite graph with vertices that are either for example, <italic>Arabidopsis </italic>genes or motifs and with edges that connect a gene and a motif if the motif occurs in the promoter of the gene. Then, each subgraph of interest is a <italic>regulatory biclique</italic>, a geneset and a motifset for which every gene in the geneset is adjacent to every motif in the motifset. The statistical significance of a motifset can be assessed using the hypergeometric distribution applied to the occurrence of the motifset in the entire <italic>Arabidopsis </italic>genome. The statistical significance of a geneset <italic>(vis-à-vis </italic>co-expression) can be assessed using correlation of gene expression from microarray experiments. The statistical significance of a biclique is then a combination of the significance for the geneset and the motifset. Biclique significance allows for the identification of the most important motifsets in a particular biological context. For example, some AT <italic>cis</italic>-elements, such as those related to water stress, are present in the promoters of a large fraction (> 89%) of genes in the genome. Consequently, water stress elements appear in many significant bicliques and their presence contributes little to the statistical significance of a biclique. Hence, the biclique obtained by deleting water stress elements remains statistically significant.</p><p>Here, we present the XcisClique system, which integrates the <italic>Arabidopsis </italic>genome with gene expression data to identify statistically significant regulatory bicliques for genesets of interest [<xref ref-type="bibr" rid="B17">17</xref>]. In particular, XcisClique uses the Apriori algorithm [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B19">19</xref>] to identify <italic>maximal regulatory bicliques</italic>, which are bicliques that cannot be made larger by the addition of any gene in the geneset of interest or by the addition of any motif in the known set of regulatory motifs. Due to the lack of reliable tools to predict <italic>Arabidopsis </italic>regulatory motifs and to reduce the search space to include only known regulatory motifs, XcisClique has no motif discovery component. Rather, XcisClique employs known motifs from the PLACE database [<xref ref-type="bibr" rid="B20">20</xref>]. With XcisClique, it is computationally feasible to identify maximal bicliques and to assess their statistical significance for genesets consisting of a few hundred genes and our current set of several hundred regulatory motifs.</p></sec><sec><title>Implementation</title><sec><title>Annotated genome data</title><p>Using Perl scripts and the Entrez Programming Utilities [<xref ref-type="bibr" rid="B21">21</xref>], we populated a PostgreSQL database of <italic>Arabidopsis </italic>genes, proteins, and promoters.</p></sec><sec><title>Microarray expression data</title><p>Expression data for the AT transcriptome was retrieved from NASC arrays in the Nottingham database ([<xref ref-type="bibr" rid="B22">22</xref>]). The slides are Affymetrix ATH1 AT Genome Arrays having 22, 814 genes. These data were generated as part of the AtGenExpress project funded by Das von der DFG geforderte AFGN (Arabidopsis Functional Genomics), which aims to provide the AT community with access to a large set of Affymetrix microarray data. This project generated expression data from 80 biologically different samples and analyzed the data using the Affymetrix Microarray Analysis Suite 5.0 with the Affymetrix MAS 5.0 Scaling Protocol. We selected 272 slides organized as follows. There are 9 abiotic stress experiments, with these stress treatments: Salt, Drought, Genotoxic, Oxidative, UV-B, Wounding, Heat, Cold, and Osmotic. Expression data for each of these is available over a series of time points (some of 0.25 h, 0.5 h, 1 h, 3 h, 6 h, 12 h, 24 h) with two biological replicates per time-point. Control slides also exist for each of these time points. We retained the following five time points, which are common to all 9 treatments: 0.5 h, 1 h, 3 h, 6 h, and 12 h. All expression data are intensity values. Half (136) of the 272 slides contain experiments involving shoots and, the other half (136) slides contain experiments involving roots.</p></sec><sec><title>cis-element data</title><p>PLACE, a database of plant <italic>cis</italic>-acting regulatory elements [<xref ref-type="bibr" rid="B20">20</xref>], is our primary source for <italic>cis</italic>-regulatory element data. These have been compiled from previously published reports and cover vascular plants only. Their variations in other genes or in plant species are also reported along with literature references. XcisClique uses the subset of AT <italic>cis</italic>-elements present in the POPS database. Additional analysis-specific AT motifs have also been curated from various sources in literature. The POPS database contains 276 <italic>Arabidopsis </italic>motifs in all; 9 of these have been curated from [<xref ref-type="bibr" rid="B23">23</xref>], 47 of these are heat shock elements.</p></sec><sec><title>Graph theoretic setting</title><p>For a geneset <italic>G </italic>and a motifset <italic>P </italic>(given as regular expressions), the <italic>occurrence graph </italic><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M1" name="1471-2105-7-218-i1" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mi mathvariant="script">O</mml:mi><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFoe=taaa@383D@</mml:annotation></mml:semantics></mml:math></inline-formula> = (G, P, E) of <italic>G </italic>and <italic>P </italic>is the bipartite graph that has <italic>g ∈ G </italic>adjacent to <italic>p ∈ P </italic>if <italic>p </italic>occurs in the promoter of <italic>g </italic>.A <italic>cis</italic>-element is modeled as a Perl regular expression by manually consolidating all its available forms from PLACE and/or literature, and manually synthesizing a regular expression that matches all the forms. Available forms of <italic>cis</italic>-elements were taken from PLACE and literature. For instance, the metal responsive element (MRE) was specified to have a consensus sequence of TGCRCNC in PLACE and sequences TGCGCAAC and TGCAGAC in literature. So the Perl regular expression for an MRE is (TGCRCNC)\|(TGCGCAAC)\|(TGCAGAC). The database has 9 <italic>cis</italic>-elements whose regular expressions have been synthesized using the above process. Its location in the promoters of genes is determined by exact pattern matching. A <italic>biclique </italic>for <italic>G </italic>and <italic>P </italic>is a complete bipartite graph in <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M2" name="1471-2105-7-218-i1" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mi mathvariant="script">O</mml:mi><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFoe=taaa@383D@</mml:annotation></mml:semantics></mml:math></inline-formula>, which is a geneset <italic>G' </italic>⊂ <italic>G </italic>and a motifset <italic>P' ⊂ P </italic>such that every gene in <italic>G' </italic>is adjacent to every pattern in <italic>P' </italic>.We write ⟨<italic>G</italic>', <italic>P' </italic>⟩ for the biclique. Biclique ⟨<italic>G'</italic>, <italic>P' </italic>⟩ is <italic>maximal </italic>if there is no gene <italic>g ∈ G – G' </italic>such that ⟨<italic>G' </italic>∪ {<italic>g</italic>}, <italic>P'</italic>⟩ is a biclique and there is no pattern <italic>p ∈ P – P' </italic>such that ⟨<italic>G'</italic>, <italic>P' </italic>∪{<italic>p'</italic>}⟩ is a biclique.</p><p>The following is an example of a biclique from an analysis done by XcisClique. The input set of genes is set of 17 genes involved in stress, pathogenicity, and secondary metabolism in AT [<xref ref-type="bibr" rid="B24">24</xref>].</p><p>Expression data for 15 of the 17 genes is available in the POPS database. Promoters of length 1200 for the input geneset were scanned for the set of all AT <italic>cis</italic>-elements. Expression data were correlated over a set of 7 treatments (Cold, Heat, Drought, Osmotic, Oxidative, Salt, UVB) in shoots. The 32<sup><italic>nd </italic></sup>biclique <italic>I</italic><sub>32 </sub>identified by the Apriori algorithm has a <italic>p</italic>-value of 3.955 × 10<sup>-03 </sup>from sequence analysis and a <italic>p-</italic>value of 1.236 × 10<sup>-02 </sup>from expression data analysis.</p><p><italic>I</italic><sub>32 </sub>= ⟨ <italic>G</italic><sub>32</sub>, <italic>M</italic><sub>32 </sub>⟩ where <italic>G</italic><sub>32 </sub>consists of genes {Atlg09500, Atlg73160, At4g37990, At4g39090, At5g59310} and <italic>M</italic><sub>32 </sub>consists of motifs {ARFAT, DPBFCOREDCDC3, GT1CONSENSUS, MYCCONSENSUSAT, RAV1AAT, ZAT12-down}. Figure <xref ref-type="fig" rid="F1">1</xref> illustrates the biclique that models <bold><italic>I</italic></bold><sub>32</sub>.</p></sec><sec><title>XcisClique overview</title><p>Figure <xref ref-type="fig" rid="F2">2</xref> depicts the process flow in XcisClique. XcisClique is an integrated suite of programs in Perl, Matlab, and C++, much of which is directly accessible through its web site. There are three kinds of user input: a set of AGI numbers <italic>G</italic>, corresponding to a geneset of interest; the set <italic>P </italic>of patterns, corresponding to <italic>cis</italic>-elements of interest, typically selected from the database of regulatory elements in XcisClique; and the treatment set <italic>T </italic>of interest, from which expression vectors are correlated.</p><p>The output from XcisClique feeds into the visualization tool MotifSee. This is a web-based tool, implemented in PHP, that accepts input tuples in the format ⟨ Gene, Motif, Sequence, Start_Position, End_Position, TATA_start, TATA_end ⟩. Besides visualizing <italic>cis</italic>-elements exactly as they occur on the promoters, this tool allows viewing subsets of genes and <italic>cis</italic>-elements as well as subsequences of promoters. XcisClique also has a viewer for gene expression vectors integrated into it to visualize expression patterns of genes in a biclique. The web site for XcisClique is hosted at [<xref ref-type="bibr" rid="B25">25</xref>].</p></sec><sec><title>Gene expression vectors</title><p>XcisClique processed gene expression data for 22,814 genes to extract tissue-specific time series data vectors. There are 9 treatments and 5 time points per treatment in the POPS database. Let <italic>e</italic><sub><italic>g</italic>,<italic>k</italic>,<italic>t </italic></sub>be the ratio of treated and control expression for gene <italic>g</italic>, a particular treatment <italic>k</italic>, and time <italic>t </italic>. While XcisClique can process any user-specified subset of the 9 treatments, for convenience of exposition, we assume that all 9 treatments are used. Let <italic>g </italic>be any gene. Define the expression vector for <italic>v </italic>to be the 45-component vector</p><p><italic>v</italic><sub><italic>g </italic></sub>= (<italic>e</italic><sub><italic>g</italic>,1,1</sub>, <italic>e</italic><sub><italic>g</italic>1,2</sub>, <italic>e</italic><sub><italic>g</italic>,1,3</sub>, <italic>e</italic><sub><italic>g</italic>,1,4</sub>, <italic>e</italic><sub><italic>g</italic>,1,5</sub>, <italic>e</italic><sub><italic>g</italic>,2,1</sub>,...,<italic>e</italic><sub><italic>g</italic>,9,4</sub>, <italic>e</italic><sub><italic>g</italic>,9,5</sub>).</p><p>(More generally, if <italic>z </italic>treatments are used, then <italic>v</italic><sub><italic>g </italic></sub>is a 5<italic>z</italic>-component vector.) XcisClique uses correlation between gene expression vectors to assess potential co-regulation of genes. XcisClique computes the Spearman correlation coefficient ρ<italic>(v</italic><sub><italic>g</italic>1</sub>, <italic>v</italic><sub><italic>g</italic>2</sub><italic>) </italic>each gene <italic>g</italic><sub>1 </sub>among the 22,814 genes and between each gene <italic>g</italic><sub>2 </sub>in the geneset of interest. The distribution of ρ-values for the correlation of a gene with all other genes of the genome is approximately normal as illustrated in Supplementary Figure 1 [See <xref ref-type="supplementary-material" rid="S1">Additional file 1</xref>]. This distribution can be used to compute an estimated p-value for the correlation of each gene pair. Many tools correlate gene expression data using Pearson correlation. However, the Pearson correlation coefficient assesses significance based on an assumption of normality, while gene expression data does not fit a normal distribution. This motivates our choice of Spearman correlation.</p></sec><sec><title>Identification of bicliques with Apriori</title><p>Combinations of <italic>cis</italic>-elements that are significantly over-represented in a geneset are identified using the Apriori data mining algorithm. XcisClique encodes the presence of <italic>cis</italic>-elements in gene promoters with a binary matrix whose rows represent genes and whose columns represent <italic>cis</italic>-elements. The Apriori algorithm finds all maximal sub-matrices of all 1s in this binary matrix [<xref ref-type="bibr" rid="B18">18</xref>,<xref ref-type="bibr" rid="B19">19</xref>]. A set of cell values is called <italic>maximal </italic>when no more rows can be added without removing columns and vice versa. Each combination of a set of genes and a set of motifs output by the algorithm is called a <italic>biclique </italic>. The <italic>k</italic><sup><italic>th </italic></sup>biclique <italic>I<sub>k </sub>= ⟨G</italic><sub><italic>k </italic></sub>, <italic>M</italic><sub><italic>k</italic></sub>⟩ is defined as a biclique with a set of \|<italic>M</italic><sub><italic>k</italic></sub>\| motifs, <italic>M</italic><sub><italic>k </italic></sub>in one clique and a set of \|<italic>G<sub>k</sub></italic>\| genes, <italic>G</italic><sub><italic>k </italic></sub>in the other. Edges connect members of one clique with all members of the other and are representative of the presence of every motif in <italic>M</italic><sub><italic>k </italic></sub>in every gene in <italic>G</italic><sub><italic>k</italic></sub>. Table <xref ref-type="table" rid="T3">3</xref> illustrates the working of this algorithm with respect to genes and motifs. Figure <xref ref-type="fig" rid="F3">3</xref> illustrates the concept of a biclique of genes and patterns, using the MotifSee visualization tool. A biclique does not imply any particular ordered arrangement of patterns. It only indicates the presence of a set of patterns in a set of genes.</p></sec><sec><title>Identification of significant bicliques</title><p>The occurrence of a random biclique <italic>M </italic>among all genes of the <italic>Arabidopsis </italic>genome should follow the hypergeometric distribution. A <italic>p</italic>-value is generated for each biclique by calculating the tail probability corresponding to the presence of more than <italic>c </italic>gene promoters with <italic>M </italic>from <italic>n </italic>promoters drawn from the genome set of <italic>N </italic>promoters having <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M3" name="1471-2105-7-218-i3" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mi mathvariant="script">C</mml:mi><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFce=qaaa@3825@</mml:annotation></mml:semantics></mml:math></inline-formula><sub><italic>M </italic></sub>promoters with <italic>M </italic>is given by this equation:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M4" name="1471-2105-7-218-i4" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:msub> <mml:mi>ℋ</mml:mi> <mml:mrow> <mml:mi>t</mml:mi> <mml:mi>a</mml:mi> <mml:mi>i</mml:mi> <mml:mi>l</mml:mi> </mml:mrow> </mml:msub> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mi>N</mml:mi> <mml:mo>,</mml:mo> <mml:msub> <mml:mi mathvariant="script">C</mml:mi> <mml:mi>M</mml:mi> </mml:msub> <mml:mo>,</mml:mo> <mml:mi>n</mml:mi> <mml:mo>,</mml:mo> <mml:mi>c</mml:mi> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>=</mml:mo> <mml:mn>1</mml:mn> <mml:mo>−</mml:mo> <mml:mstyle displaystyle="true"> <mml:msubsup> <mml:mo>∑</mml:mo> <mml:mrow> <mml:mi>i</mml:mi> <mml:mo>=</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> <mml:mi>c</mml:mi> </mml:msubsup> <mml:mrow> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mfrac> <mml:mrow> <mml:mi>C</mml:mi> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mi>n</mml:mi> <mml:mo>,</mml:mo> <mml:mi>c</mml:mi> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>⋅</mml:mo> <mml:mi>C</mml:mi> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mi>N</mml:mi> <mml:mo>−</mml:mo> <mml:mi>n</mml:mi> <mml:mo>,</mml:mo> <mml:msub> <mml:mi mathvariant="script">C</mml:mi> <mml:mi>M</mml:mi> </mml:msub> <mml:mo>−</mml:mo> <mml:mi>c</mml:mi> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:mrow> <mml:mi>C</mml:mi> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:mi>N</mml:mi> <mml:mo>,</mml:mo> <mml:msub> <mml:mi mathvariant="script">C</mml:mi> <mml:mi>M</mml:mi> </mml:msub> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> </mml:mfrac> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> </mml:mstyle> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>1</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFlecsdaWgaaWcbaGaemiDaqNaemyyaeMaemyAaKMaemiBaWgabeaakmaabmaabaGaemOta4KaeiilaWIae8NaXp0aaSbaaSqaaiabd2eanbqabaGccqGGSaalcqWGUbGBcqGGSaalcqWGJbWyaiaawIcacaGLPaaacqGH9aqpcqaIXaqmcqGHsisldaaeWaqaamaabmaabaWaaSaaaeaacqWGdbWqdaqadaqaaiabd6gaUjabcYcaSiabdogaJbGaayjkaiaawMcaaiabgwSixlabdoeadnaabmaabaGaemOta4KaeyOeI0IaemOBa4MaeiilaWIae8NaXp0aaSbaaSqaaiabd2eanbqabaGccqGHsislcqWGJbWyaiaawIcacaGLPaaaaeaacqWGdbWqdaqadaqaaiabd6eaojabcYcaSiab=jq8dnaaBaaaleaacqWGnbqtaeqaaaGccaGLOaGaayzkaaaaaaGaayjkaiaawMcaaaWcbaGaemyAaKMaeyypa0JaeGymaedabaGaem4yamganiabggHiLdGccaWLjaGaaCzcamaabmaabaGaeGymaedacaGLOaGaayzkaaaaaa@7372@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>Bicliques from the output of the Apriori algorithm are filtered using <italic>False Discovery Rate (FDR) </italic>[<xref ref-type="bibr" rid="B26">26</xref>], applied to the <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M5" name="1471-2105-7-218-i2" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mi>ℋ</mml:mi><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFlecsaaa@3763@</mml:annotation></mml:semantics></mml:math></inline-formula><sub><italic>tail</italic></sub>(<italic>N</italic>, <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M6" name="1471-2105-7-218-i3" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mi mathvariant="script">C</mml:mi><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFce=qaaa@3825@</mml:annotation></mml:semantics></mml:math></inline-formula><sub><italic>M</italic></sub>, <italic>n</italic>, <italic>c</italic>) values. The default FDR parameter in XcisClique is 0.05. Ranks are assigned to bicliques in increasing order of their <italic>p-</italic>values.</p></sec><sec><title>Evaluation of genesets using gene expression data</title><p>For any geneset <italic>G </italic>of <italic>Arabidopsis </italic>genes, we compute the Spearman correlation coefficients <italic>ρ</italic>(<italic>v</italic><sub><italic>g</italic>1</sub>, <italic>v</italic><sub><italic>g</italic>2</sub>), as described earlier. Each ρ <italic>(v</italic><sub><italic>g</italic>1</sub>, <italic>v</italic><sub><italic>g</italic>2 </sub>lies between -1 and 1, with 0 meaning uncorrelated, 1 meaning completely correlated, and -1 meaning completely oppositely correlated. Since a negative correlation may be as biologically significant as a positive correlation, we use the absolute value \|ρ <italic>(v</italic><sub><italic>g</italic>1</sub>,<italic>v</italic><sub><italic>g</italic>2</sub>) to avoid unwanted cancellation of negative and positive correlations. We define the Sum of Absolute Values (SAV) statistic <italic>S(G) </italic>to be the sum of these absolute values, namely:</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M7" name="1471-2105-7-218-i5" overflow="scroll"> <mml:semantics definitionURL="" encoding=""> <mml:mrow> <mml:mi>S</mml:mi> <mml:mrow> <mml:mo>(</mml:mo> <mml:mi>G</mml:mi> <mml:mo>)</mml:mo> </mml:mrow> <mml:mo>=</mml:mo> <mml:mstyle displaystyle="true"> <mml:munder> <mml:mo>∑</mml:mo> <mml:mrow> <mml:msub> <mml:mi>g</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo>,</mml:mo> <mml:msub> <mml:mi>g</mml:mi> <mml:mn>2</mml:mn> </mml:msub> <mml:mo>∈</mml:mo> <mml:mi>G</mml:mi> </mml:mrow> </mml:munder> <mml:mrow> <mml:mrow> <mml:mo>\|</mml:mo> <mml:mrow> <mml:mi>ρ</mml:mi> <mml:mrow> <mml:mo>(</mml:mo> <mml:mrow> <mml:msub> <mml:mi>g</mml:mi> <mml:mn>1</mml:mn> </mml:msub> <mml:mo>,</mml:mo> <mml:msub> <mml:mi>g</mml:mi> <mml:mn>2</mml:mn> </mml:msub> </mml:mrow> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:mo>\|</mml:mo> </mml:mrow> </mml:mrow> </mml:mstyle> <mml:mo>,</mml:mo> <mml:mtext>     </mml:mtext> <mml:mrow> <mml:mo>(</mml:mo> <mml:mn>2</mml:mn> <mml:mo>)</mml:mo> </mml:mrow> </mml:mrow> <mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGtbWudaqadaqaaiabdEeahbGaayjkaiaawMcaaiabg2da9maaqafabaWaaqWaaeaaiiGacqWFbpGCdaqadaqaaiabdEgaNnaaBaaaleaacqaIXaqmaeqaaOGaeiilaWIaem4zaC2aaSbaaSqaaiabikdaYaqabaaakiaawIcacaGLPaaaaiaawEa7caGLiWoaaSqaaiabdEgaNnaaBaaameaacqaIXaqmaeqaaSGaeiilaWIaem4zaC2aaSbaaWqaaiabikdaYaqabaWccqGHiiIZcqWGhbWraeqaniabggHiLdGccqGGSaalcaWLjaGaaCzcamaabmaabaGaeGOmaidacaGLOaGaayzkaaaaaa@4D16@</mml:annotation> </mml:semantics> </mml:math></inline-formula></p><p>where <italic>g</italic><sub>1 </sub>ranges over all of the genes for which gene expression data are available. The probability of observing a SAV greater than or equal to <italic>S(G) </italic>is given by <italic>p</italic><sub><italic>r</italic></sub>(<italic>S</italic>(<italic>G</italic>)). This is calculated by sampling the distribution of <italic>S(G) </italic>in the genome. Binding a class of transcription factors in response to a set of treatments might induce transcription of some genes in the biclique while repressing other genes in the same biclique. The sum of absolute values of ρ is an indication of how tightly (both negatively and positively) correlated the geneset in a biclique is. Supplementary Figures 2 [see <xref ref-type="supplementary-material" rid="S2">Additional file 2</xref>] and 3 [see <xref ref-type="supplementary-material" rid="S3">Additional file 3</xref>] illustrate the probability density function and the cumulative distribution function for <italic>S </italic>respectively, for a geneset of size 6.</p></sec><sec><title>Combined p-value for a biclique</title><p>The final, combined <italic>p</italic>-value of a biclique <italic>I </italic>= ⟨<italic>G</italic>', <italic>P'</italic>⟩ is the product</p><p><inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M8" name="1471-2105-7-218-i2" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mi>ℋ</mml:mi><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFlecsaaa@3763@</mml:annotation></mml:semantics></mml:math></inline-formula><sub><italic>tail</italic></sub>(<italic>N</italic>, <inline-formula><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="M9" name="1471-2105-7-218-i3" overflow="scroll"><mml:semantics definitionURL="" encoding=""><mml:mi mathvariant="script">C</mml:mi><mml:annotation encoding="MathType-MTEF"> MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFce=qaaa@3825@</mml:annotation></mml:semantics></mml:math></inline-formula><sub><italic>M</italic></sub>, <italic>n</italic>, <italic>c</italic>)·<italic>pr</italic>(<italic>S</italic>(<italic>G</italic>))</p><p>of the hypergeometric tail probability from analysis of the biclique <italic>P' </italic>(Equation 1) and the SAV <italic>p-</italic>value from expression analysis of <italic>G' </italic>(Equation 2).</p></sec></sec><sec><title>Results</title><p>To evaluate the effectiveness of the XcisClique system, we performed three case studies that applied XcisClique to different genesets and a common set of known regulatory motifs. Case study 1 employs a geneset of 11 AT genes up-regulated by cold stress. Case study 2 employs a geneset of 14 AT genes down-regulated by cold stress. Case study 3 analyzes 113 AT genes involved in senescence.</p><sec><title>Case study 1: metabolism genes up-regulated after cold stress</title><p>For our first case study, we selected a set of 11 AT genes (identified in Supplementary Table 1 [see <xref ref-type="supplementary-material" rid="S7">Additional file 7</xref>]) that are involved in carbohydrate metabolism and secondary metabolism and that are up-regulated long-term by cold stress [<xref ref-type="bibr" rid="B27">27</xref>]. The Apriori algorithm identified 193 bicliques. After False Discovery Rate (FDR) correction of motifset significance at the 0.05 level, 177 significant bicliques remained. Figure <xref ref-type="fig" rid="F4">4</xref> details five of these bicliques that were identified as statistically over-represented both by the hypergeometric (motifset) and SAV (gene expression, see methods) analyses. The motifs in these bicliques include CRT- or DRE-like elements, where the inducible transcription factors CBF1, CBF2, and CBF3 bind [<xref ref-type="bibr" rid="B28">28</xref>,<xref ref-type="bibr" rid="B29">29</xref>], as well as motifs associated with other abiotic stresses such as water stress (ABRE-like motifs), and, unexpectedly, motifs that have been discovered in pathogen or salicylic acid responsive genes, such as the WBOXATNPR1 [<xref ref-type="bibr" rid="B30">30</xref>] and ASF1MOTIFCAMV elements (See the web site for details about these regulatory motifs). The presence of these biotic stress related motifs shows that these genes might play a role not only in abiotic stresses but also in biotic stresses.</p><p>Another unexpected motif is CCA1ATLHCB1, the binding site of the Circadian Cycle Associated protein (CCA1), a Myb-related transcription factor [<xref ref-type="bibr" rid="B31">31</xref>]. Recent studies on cold-response in AT have shown that CBF transcription factors are regulated by the circadian cycle, with the highest expression observed when plants are transfered to a lower temperature 4 hours after dawn. [<xref ref-type="bibr" rid="B32">32</xref>]. The genes analyzed in this group are not CBF transcription factors, but two of the genes that contain the CCA1ATLHCB1 motif (Figure <xref ref-type="fig" rid="F4">4</xref>) increase their expression within one hour of stress (Supplementary Figure 5 [see <xref ref-type="supplementary-material" rid="S5">Additional file 5</xref>] and [<xref ref-type="bibr" rid="B27">27</xref>]), gradually increasing until they reach a maximum after 12 hours. Therefore, the initial response of these genes might be due to CCA1 induction and the peak reached by CBF induction.</p><p>Biclique 111 (biclique rank in the analysis set found on the web site) is interesting because it contains three genes that are part of the CBF regulon [<xref ref-type="bibr" rid="B33">33</xref>] and the motifs contained in this set follow the particular order MYB1LEPR, WBOX, and CCA1ATLHCB1 (Supplementary Figure [see <xref ref-type="supplementary-material" rid="S4">Additional file 4</xref>]). Two of the genes that belong to this motifset (Atlg62570 and Atlg60470) are putative galactinol synthase genes. These genes are part of the raffinose biosynthesis pathway, which accumulates in plants treated by cold and drought [<xref ref-type="bibr" rid="B34">34</xref>]; raffinose is a sugar that is thought to act as an osmoprotectant under cold and drought.</p><p><italic>Biclique 23 is particularly interesting, because its tight co-regulation is supported not only by the gene expression data but also the results of </italic>[<xref ref-type="bibr" rid="B27">27</xref>]<italic>and </italic>[<xref ref-type="bibr" rid="B33">33</xref>]<italic>where these genes show a peak up-regulation after 24 hours of cold stress in plate and soil experiments</italic>. Four genes (Atlg09350, Atlg62570, At2gl6890, At5g20830) in this biclique belong to the CBF regulon, and four genes (Atlg09350, Atlg62570, At2gl6890, At4g27180) have the DRECRTCOREAT consensus motif [<xref ref-type="bibr" rid="B27">27</xref>,<xref ref-type="bibr" rid="B33">33</xref>], which explains their similar expression under cold stress (Supplementary Figure 5 [see <xref ref-type="supplementary-material" rid="S5">Additional file 5</xref>]). Biclique 23 also show up-regulation under salt and osmotic stress in shoots. Water stress related motifs Fed-ABRE-like and Fed-AtMyb4 [<xref ref-type="bibr" rid="B23">23</xref>] are located within 600 bps of the transcription start site in the upstream regions of these genes, so this particular arrangement of motifs might be responsible for their expression under other comparable stresses (cold, osmotic, and salt) (Supplementary Figure 5 [see <xref ref-type="supplementary-material" rid="S5">Additional file 5</xref>]).</p></sec><sec><title>Case study 2: metabolism genes down-regulated after cold stress</title><p>The promoters of 14 metabolism genes (identified in Supplementary Table 2 [see <xref ref-type="supplementary-material" rid="S7">Additional file 7</xref>]) shown by [<xref ref-type="bibr" rid="B27">27</xref>] to be down-regulated after cold stress were analyzed using XcisClique. The Apriori algorithm identified 336 bicliques, which, after a correction with an FDR of 0.05 resulted in 270 significant bicliques. Table <xref ref-type="table" rid="T1">1</xref> shows selected significant motif combinations found for these genes in shoots and roots. [<xref ref-type="bibr" rid="B33">33</xref>] have shown that many of the genes that were down-regulated by cold were also down-regulated by over-expression of the CBF or ZAT12 transcription factors. They found putative motifs responsible for down-regulation, but none of the genes that we have studied were shown to be down-regulated by over-expression of CBF and ZAT12 transcription factors [<xref ref-type="bibr" rid="B33">33</xref>]. This explains why we did not find significant bicliques containing these motifs in promoters of these genes, and even if they might be present individually, their presence was not associated with a significant motif combination. Genes in biclique 203 show down-regulation under cold stress, but also up-regulation under salt stress (Supplementary Figure 6 [see <xref ref-type="supplementary-material" rid="S5">Additional file 5</xref>]), which is a novel observation. This response could be explained by the presence of the combination of ABRELATERD1, and MYCATRD22 which are binding sites of transcription factors responsive to ABA and drought [<xref ref-type="bibr" rid="B35">35</xref>], respectively, but have also been found in salt stress induced genes [<xref ref-type="bibr" rid="B35">35</xref>].</p><p><italic>A novel heat shock element associated with negative regulation of transcription is included in the motif arrangement in biclique </italic>35 (<italic>Supplementary </italic>Figure 7 [see <xref ref-type="supplementary-material" rid="S6">Additional file 6</xref>]). The heat shock element binding site is formed by alternate repeats of the pentamer 5'-nGAAn-3' (5'-nTTCn-3' on the reverse strand). It includes a mutation in the A/T nucleotides of the pentamer [<xref ref-type="bibr" rid="B36">36</xref>]. The HSE motif found in these genes shows a mutation in an A/T in the 1st and the 3rd pentamer of the element and therefore, represents a <italic>cis</italic>-element distinct from the sequence of the canonical HSE. These genes also show down-regulation under heat stress (Supplementary Figure 7 [see <xref ref-type="supplementary-material" rid="S6">Additional file 6</xref>]). Therefore this mutated HSE motif might be a specific binding site for the class B of heat shock factors, which are negative regulators of transcription [<xref ref-type="bibr" rid="B37">37</xref>].</p></sec><sec><title>Case study 3: senescence genes</title><p>An input set of 113 senescence responsive genes in AT (identified in Supplementary Table 3 [see <xref ref-type="supplementary-material" rid="S7">Additional file 7</xref>]) were analyzed using XcisClique. These genes are taken from [<xref ref-type="bibr" rid="B24">24</xref>], and show up-regulation during leaf senescence. These genes are involved in various processes, including protein degradation, oxidation, and detoxification. Expression data for 107 of the 113 genes is available. Promoters of length 1200 for the input geneset were scanned for the set of all AT <italic>cis</italic>-elements. Expression data for the gene set were correlated over a set of 9 treatments (identified as Cold, Heat, Drought, Osmotic, Oxidative, Salt, UVB, Genotoxic, Wounding) in shoots. The complete set of results for this analysis can be viewed at the web site. Table <xref ref-type="table" rid="T2">2</xref> shows a selected set of 2 bicliques that have low <italic>p</italic>-values both from sequence and expression data analysis. Regulation of expression of genes related to senescence involves proteolytic degradation [<xref ref-type="bibr" rid="B38">38</xref>]. Biclique 31 contains two proteases (Atlg47128, At4g39090), the ubiquitin-conjugating enzyme 1 (Atlgl4400), and a putative membrane protein (Atlg68820) that by electronic annotation of the GO consortium has putative ubiquitin protein kinase activity, which take part in the degradation process that occurs during senescence. The other genes in biclique 31 encode an ABC transporter (Atlg59870), two putative ethylene synthesis regulators (At2g25450, 2-oxoglutarate-dependent dioxygenase similar to tomato ethylene synthesis regulatory protein E8, and At5gl0860 a CBS domain protein that binds to ATP, ADP, and SAM), and metal binding proteins (At2g26560, patatin like protein with oxidoreductase activity, acting on iron-sulfur proteins as donors, and At3g09390, metallothionein protein). These genes share the ELRECOREPCRP1 motif (Elicitor Responsive Element core of parsley PR1), where WRKY1 transcription factors binds [<xref ref-type="bibr" rid="B39">39</xref>,<xref ref-type="bibr" rid="B40">40</xref>]. Programmed cell death is observed in plants not only during senescence but also during the hypersensitive response after pathogen attack; therefore, the presence of the ELRECOREPCRP1 motif in these genes suggests up-regulation of these genes after pathogen attack.</p><p>Regulation of UV-B responses is associated with specific variations on a consensus sequence. The Fed-HBF or H-box motif has also been shown to be involved in response to oxidative stress (ozone in particular) and/or pathogen attack and is therefore related to cell death [<xref ref-type="bibr" rid="B23">23</xref>]. The transcription factor that binds to this <italic>cis</italic>-element belongs to the bZIP transcription family and binds also to a G-box motif [<xref ref-type="bibr" rid="B41">41</xref>]. The G-box motif is a palindromic sequence (CACGTG) that is a specific example of the partially defined DPBFCOREDCDC3 consensus sequence (ACACNNG) whose transcription factors also belong to the bZIP family [<xref ref-type="bibr" rid="B42">42</xref>]. Genes in biclique 3854 contain the Fed-HBF and DPBFCOREDCDC3 motifs. Four of these genes are involved in proteolysis or protein catabolism (Atlg21670, Atlg47128, Atlg53750, At5g60360). Five genes have precise matches for the G-box motif: Atlg21670, Atlg53750, Atlg78080, At3gl2120, and At5g60360. Four of these genes also show up-regulation under UV-B stress, while genes that match the DPBFCOREDCDC3 motif but do not match the specific G-box motif show down-regulation under UV-B stress (Supplementary Figure 8 [see <xref ref-type="supplementary-material" rid="S6">Additional file 6</xref>]).</p><p>The H-box and the G-box are also present in the promoter of the chalcone synthase gene (CHS), which catalyzes the first step for the synthesis of flavonoids [<xref ref-type="bibr" rid="B43">43</xref>]. CHS is also up-regulated under UV-B stress, since flavonoid molecules can absorb UV-B radiation [<xref ref-type="bibr" rid="B44">44</xref>]. Atlg21670 and Atlg53750 are related to protein degradation, while Atlg78080 is a transcription factor (RAP2.4) and At3gl2120 is a fatty acid desaturase (FAD2). These genes are not related to flavonoid synthesis but they might protect the plant against UV-B stress in other ways such as catabolism of damaged proteins by UV-B (Atlg21670, Atlg53750) or signaling/activation of other protective pathways (At3gl2120/Atlg53750).</p></sec></sec><sec><title>Discussion</title><p>Several programs have been developed for discovering <italic>cis</italic>-regulatory modules in yeast. The transcriptional mechanisms in yeast are somewhat understood, and there is enough biological data about yeast upon which to base computational findings on. In a higher eukaryote such as <italic>Arabidopsis</italic>, the gene-abundance is much higher (approximately 28,000). While there are databases that identify all TFBSs in yeast, not all TFBSs in AT are known and documented. Only a fraction of TFBSs in AT have documented consensus sequences. Position weight matrices are even rarer with TRANSFAC containing 10 position weight matrices for binding sites in AT. The lack of sufficient biological data in the case of AT makes the validation of promoter discovery tools problematic.</p><p>XcisClique provides a novel platform for investigating regulatory motifs in <italic>Arabidopsis </italic>via an integrated infrastructure combining annotated genome data, annotated <italic>cis</italic>-element data, and gene expression data. XcisClique identifies statistically overrepresented bicliques and evaluates each biclique with respect to gene expression data. This gives an indication of the importance of co-occurrence of a set of regulatory elements in a geneset with respect to transcriptional response. The <italic>p</italic>-value of each biclique is a determinant of biological significance as well. To measure the tightness of correlation of genes in a biclique, we needed a statistic. Initially, we considered the simple sum of the Spearman correlation coefficients of all pairs of genes in a biclique, but negative correlations balanced the positive correlations and the simple sum was not a good indicator of how tightly correlated genes in a biclique were. For instance, a set of correlations {0.1, -0.1,0.3} yields the same statistic as the set {0.8, -0.8,0.3}. Obviously, the latter set of genes is more tightly correlated. The sum of absolute values statistic considers the individual contributions of all correlations and is a sharper test of the tightness of correlations in a biclique. Hence, we used this statistic to measure the co-expression of a biclique.</p><p>Most transcription factors families in plants are large, therefore there is a possibility that some of their members might be activators, and others, repressors. Since, in most cases, the specific <italic>cis</italic>-elements regions for each of the members of a given transcription factor family have not yet been determined, what is currently available is a consensus sequence serving as a motif rather than a specific sequence. Application of SAV results in the grouping of genes that share a set of these, often partially defined, motifs, with the result that some gene groups that share consensus motifs might be down-regulated compared with other groups under the same experimental conditions. In these cases, we have found that the actual sequence of the motifs in the down-regulated gene group is different from those in the up-regulated group; for example in the case of the DPBFCOREDCDC3 motif in the analysis of the senescence genes that we made (Case study 3). The DPBFCOREDCDC3 consensus sequence (ACACNNG) also subsumes the defined G-box motif sequence (CACGTG). Genes in biclique 3854 contain the DPBFCOREDCDC3 motif. Five genes have matches for the DPBFCOREDCDC3 motif that also correspond to the G-box motif: Atlg21670, Atlg53750, Atlg78080, At3gl2120, and At5g60360. Four of these genes show up-regulation under UV-B stress, while genes that have matches to the DPBFCOREDCDC3 motif but do not to the G-box motif show down-regulation under UV-B stress (Supplementary Figure 8 [see <xref ref-type="supplementary-material" rid="S6">Additional file 6</xref>]). The different response/regulation of genes in the biclique can be explained by the different sequences of matches, all of which match the regular expression for the DPBFCOREDCDC3 motif.</p><p>Another example is the heat shock factor (HSF) family. Class B HSFs inhibit transcription (Czarneka-Verner et al 2004). These HSFs do not bind to the canonical heat shock element (alternate repeats of 5' -nGAAn- 3'), and therefore, Class B HSFs must bind to another <italic>cis</italic>-sequence in target genes. Our study allowed us to identify <italic>cis</italic>-sequences that are possible candidates for binding of this class of HSFs. The HSE motif in genes of Biclique 35 (Case Study 2) show a mutation in an A/T in the 1st and the 3rd pentamer of the element. These genes also show down-regulation under heat stress (Supplementary Figure 7 [see <xref ref-type="supplementary-material" rid="S5">Additional file 5</xref>]). Therefore this HSE motif might be a specific binding site for class B heat shock factors. XcisClique uses only known AT motifs curated from various sources. This ensures that the search space for patterns is limited, not confined by motif lengths, and consists of well-defined, annotated motifs. The biologist has the choice of selecting a subset of relevant motifs, and this makes one of the three inputs (<italic>cis</italic>-elements) biologically directed. The second and third inputs which are the genes being analyzed and the treatment sets over which expression data is to be considered, respectively, are also specified by the biologist. The integration of biological knowledge within XcisClique greatly reduces the final search space and yields more biologically relevant results. XcisClique is scalable to more numerous motifs and treatments. The system has been verified with biological data from <italic>Arabidopsis </italic>. Given adequate sequence and gene expression data, the system is sufficiently generic to accommodate any organism. MotifSee, a visualization component of XcisClique, supports viewing combinations of motifs in gene promoters. A viewer for visualizing gene expression patterns of a set of genes is also integrated into the system.</p></sec><sec><title>Conclusion</title><p>Using both motifset significance, assessed using the hypergeometric distribution, and gene expression correlation, assessed using the SAV statistic, ensures that the biological context is present in the final significance value calculated. Consider a set of genes such that every gene in the set is highly correlated to every other gene. The set can be expanded by correlating each gene in the set to every gene in the AT genome. Only those genes are added to the original set, whose correlation coefficient with one of the members of the original set is above a given threshold. This process is not available with the XcisClique web-interface. The enriched set can be input to XcisClique to produce more significant bicliques. Also, conserved arrangements of motifs were observed in significant bicliques. A formalization of the process to identify conserved arrangements is one of the future directions we are pursuing.</p></sec><sec><title>Availability and requirements</title><p>The web-based interface for XcisClique is available at <ext-link ext-link-type="uri" xlink:href="https://bioinformatics.cs.vt.edu/XcisClique"/>. The source code for XcisClique is freely available under the GNU Public License at the following location: <ext-link ext-link-type="uri" xlink:href="https://bioinformatics.cs.vt.edu/XcisClique/XcisClique.tar.gz"/>.</p><p>The following software components are required to install and run the command line version of XcisClique.</p><p>1. Perl 5.8.5 or higher.</p><p>2. Perl Modules.</p><p>(a) <monospace>LWP::Simple</monospace> This Perl module provides a simple, procedural interface to LWP, which is the World-Wide Web library for Perl, a set of Perl modules which provides a sample and consistent application programming interface (API) to the World-Wide Web. <italic>CPAN</italic></p><p>(b) <monospace>Shell</monospace> Perl module to run shell commands transparently within Perl. <italic>CPAN</italic></p><p>(c) <monospace>DBI</monospace> Perl module for database access. It defines a set of methods, variables, and conventions that provide a consistent database interface, independent of the actual database being used. <italic>CPAN</italic></p><p>(d) <monospace>DBD::Pg</monospace> This is the PostgreSQL database driver for the DBI module. <italic>CPAN</italic></p><p>(e) <monospace>Test::Simple</monospace> Pre-requisite for <monospace>DBD::Pg.</monospace><italic> CPAN</italic></p><p>(f) <monospace>Time::localtime</monospace> Perl module with interfaces to Perl's built-in <monospace>localtime()</monospace> function. <italic>CPAN</italic></p><p>(g) <monospace>Math::Matrix</monospace> Perl module with functions for multiplication, inversion, and other common matrix operations. <italic>CPAN</italic></p><p>(h) <monospace>Statistics::Distributions</monospace> Perl module for calculating critical values and upper probabilities of common statistical distributions such as the Normal distribution, the χ<sup>2 </sup>distribution, the t distribution, and the F distribution. <italic>CPAN</italic></p><p>(i) <monospace>PDF</monospace> Perl module with functions for calculating critical values and probabilities of various statistical distributions, such as the Binomial distribution, the Hypergeometric distribution, and the Gaussian distribution. <italic>Packaged with XcisClique</italic></p><p>(j) <monospace>Vector</monospace> Perl module for common vector operations and calculation of Pearson and Spearman correlation coefficients between vectors. <italic>Packaged with XcisClique</italic></p><p>(k) <monospace>Utilities</monospace> Perl module with common text processing utility functions such as removing white space from a string. <italic>Packaged with XcisClique</italic></p><p>3. PostgreSQL 7.4.7 or higher.</p><p>4. MATLAB 7.0.4 with Statistics toolbox.</p></sec><sec><title>Authors' contributions</title><p>AP and LSH conceived of the study, participated in its design and implementation, contributed to the choice of case studies, and drafted the manuscript. The source code for XcisClique has been developed by AP. TMM participated in the design of the system and provided the implementation of the Apriori algorithm. CVR and RG conceived of the biological case studies, ensured biological validity of all methods used in this work, and tested the system.</p></sec><sec sec-type="supplementary-material"><title>Supplementary Material</title><supplementary-material content-type="local-data" id="S1"><caption><title>Additional File 1</title><p>Supplementary Figure 1 : This figure illustrates the distribution ρ value for Spearman correlations of the rd29a gene expression vector with all genes of the AT genome.</p></caption><media xlink:href="1471-2105-7-218-S1.pdf" mimetype="application" mime-subtype="pdf"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S2"><caption><title>Additional File 2</title><p>Supplementary Figure 2 : This figure illustrates the probability density function of the SAV statistic for a geneset of size 6.</p></caption><media xlink:href="1471-2105-7-218-S2.pdf" mimetype="application" mime-subtype="pdf"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S3"><caption><title>Additional File 3</title><p>Supplementary Figure 3 : This figure illustrates the cumulative distribution function of the SAV statistic for a geneset of size 6.</p></caption><media xlink:href="1471-2105-7-218-S3.pdf" mimetype="application" mime-subtype="pdf"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S4"><caption><title>Additional File 4</title><p>Supplementary Figure 4 : This figure illustrates motif arrangements in the biclique ranked 111 in analysis 5.</p></caption><media xlink:href="1471-2105-7-218-S4.pdf" mimetype="application" mime-subtype="pdf"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S5"><caption><title>Additional File 5</title><p>Supplementary Figures 5 and 6 : This is a set of two figures illustrating expression vectors for genes in biclique 23 in Case study 1 and biclique 203 in Case study 2.</p></caption><media xlink:href="1471-2105-7-218-S5.pdf" mimetype="application" mime-subtype="pdf"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S6"><caption><title>Additional File 6</title><p>Supplementary Figures 7 and 8 : This is a set of two figures illustrating the expression vectors for genes in biclique 35 in Case study 2 and biclique 3854 in Case study 3.</p></caption><media xlink:href="1471-2105-7-218-S6.pdf" mimetype="application" mime-subtype="pdf"><caption><p>Click here for file</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="S7"><caption><title>Additional File 7</title><p>Supplementary Tables : Supplementary tables 1, 2, and 3 list the input genes for Case Studies 1,2, and 3 respectively.</p></caption><media xlink:href="1471-2105-7-218-S7.pdf" mimetype="application" mime-subtype="pdf"><caption><p>Click here for file</p></caption></media></supplementary-material></sec>
Molecular Determinants of Ebola Virus Virulence in Mice	<p> <italic>Zaire ebolavirus</italic> (ZEBOV) causes severe hemorrhagic fever in humans and nonhuman primates, with fatality rates in humans of up to 90%. The molecular basis for the extreme virulence of ZEBOV remains elusive. While adult mice resist ZEBOV infection, the Mayinga strain of the virus has been adapted to cause lethal infection in these animals. To understand the pathogenesis underlying the extreme virulence of Ebola virus (EBOV), here we identified the mutations responsible for the acquisition of the high virulence of the adapted Mayinga strain in mice, by using reverse genetics. We found that mutations in viral protein 24 and in the nucleoprotein were primarily responsible for the acquisition of high virulence. Moreover, the role of these proteins in virulence correlated with their ability to evade type I interferon-stimulated antiviral responses. These findings suggest a critical role for overcoming the interferon-induced antiviral state in the pathogenicity of EBOV and offer new insights into the pathogenesis of EBOV infection.</p>	<contrib contrib-type="author"><name><surname>Ebihara</surname><given-names>Hideki</given-names></name><xref ref-type="aff" rid="aff1">1</xref><xref ref-type="aff" rid="aff2">2</xref><xref ref-type="aff" rid="aff3">3</xref></contrib><contrib contrib-type="author"><name><surname>Takada</surname><given-names>Ayato</given-names></name><xref ref-type="aff" rid="aff2">2</xref><xref ref-type="aff" rid="aff4">4</xref></contrib><contrib contrib-type="author"><name><surname>Kobasa</surname><given-names>Darwyn</given-names></name><xref ref-type="aff" rid="aff5">5</xref></contrib><contrib contrib-type="author"><name><surname>Jones</surname><given-names>Steven</given-names></name><xref ref-type="aff" rid="aff3">3</xref><xref ref-type="aff" rid="aff6">6</xref></contrib><contrib contrib-type="author"><name><surname>Neumann</surname><given-names>Gabriele</given-names></name><xref ref-type="aff" rid="aff7">7</xref></contrib><contrib contrib-type="author"><name><surname>Theriault</surname><given-names>Steven</given-names></name><xref ref-type="aff" rid="aff3">3</xref><xref ref-type="aff" rid="aff8">8</xref></contrib><contrib contrib-type="author"><name><surname>Bray</surname><given-names>Mike</given-names></name><xref ref-type="aff" rid="aff9">9</xref></contrib><contrib contrib-type="author"><name><surname>Feldmann</surname><given-names>Heinz</given-names></name><xref ref-type="aff" rid="aff3">3</xref><xref ref-type="aff" rid="aff8">8</xref></contrib><contrib contrib-type="author"><name><surname>Kawaoka</surname><given-names>Yoshihiro</given-names></name><xref ref-type="aff" rid="aff1">1</xref><xref ref-type="aff" rid="aff2">2</xref><xref ref-type="aff" rid="aff7">7</xref><xref ref-type="aff" rid="aff10">10</xref><xref ref-type="corresp" rid="cor1">*</xref></contrib>	PLoS Pathogens	<sec id="s1"><title>Introduction</title><p> <italic>Zaire ebolavirus</italic> (ZEBOV), a member of the family <italic>Filoviridae</italic>, genus <italic>Ebolavirus,</italic> causes severe hemorrhagic fever in humans and nonhuman primates (NHPs). Case-fatality rates for ZEBOV infection in humans are the highest among known viral hemorrhagic fevers, ranging from 70% to 90% [<xref rid="ppat-0020073-b001" ref-type="bibr">1</xref>–<xref rid="ppat-0020073-b003" ref-type="bibr">3</xref>]. On the basis of in vitro data, three Ebola virus (EBOV) proteins, the glycoprotein (GP) [<xref rid="ppat-0020073-b004" ref-type="bibr">4</xref>–<xref rid="ppat-0020073-b006" ref-type="bibr">6</xref>], the membrane-associated viral protein (VP) 24 [<xref rid="ppat-0020073-b007" ref-type="bibr">7</xref>,<xref rid="ppat-0020073-b008" ref-type="bibr">8</xref>], and VP35 [<xref rid="ppat-0020073-b009" ref-type="bibr">9</xref>,<xref rid="ppat-0020073-b010" ref-type="bibr">10</xref>], a component of the replication complex, are thought to play key roles in EBOV pathogenicity. The GP, which mediates viral entry, is a major determinant of viral tropism and may be cytotoxic, although a recent report has challenged the notion of GP's cytotoxicity [<xref rid="ppat-0020073-b004" ref-type="bibr">4</xref>–<xref rid="ppat-0020073-b006" ref-type="bibr">6</xref>,<xref rid="ppat-0020073-b011" ref-type="bibr">11</xref>]. VP24 and VP35 are known as type I interferon (IFN) antagonists and interfere with the type I IFN-mediated antiviral response in vitro [<xref rid="ppat-0020073-b007" ref-type="bibr">7</xref>,<xref rid="ppat-0020073-b009" ref-type="bibr">9</xref>,<xref rid="ppat-0020073-b010" ref-type="bibr">10</xref>]. However, the role of these proteins in viral pathogenicity has not been determined in vivo.</p><p>Three animal models, NHPs, guinea pigs, and mice, have been used to study EBOV pathogenesis [<xref rid="ppat-0020073-b012" ref-type="bibr">12</xref>–<xref rid="ppat-0020073-b014" ref-type="bibr">14</xref>]. Generally, filoviruses do not kill adult immunocompetent rodents, although some strains have been shown to cause lethal infections in newborn mice [<xref rid="ppat-0020073-b014" ref-type="bibr">14</xref>]. Bray et al. [<xref rid="ppat-0020073-b014" ref-type="bibr">14</xref>] adapted ZEBOV to progressively older BALB/c mice and thereby established a lethal model in adult immunocompetent mice. Infection of mice with mouse-adapted virus (MA-ZEBOV) involves primary target cells of the mononuclear phagocytic system, namely monocytes, macrophages, and dendritic cells, as well as target organs (spleen, lymph nodes, and liver), as seen in humans and NHPs, resulting in a disease comparable to that observed in the latter animals [<xref rid="ppat-0020073-b002" ref-type="bibr">2</xref>,<xref rid="ppat-0020073-b015" ref-type="bibr">15</xref>–<xref rid="ppat-0020073-b017" ref-type="bibr">17</xref>]. Although MA-ZEBOV–infected mice do not exhibit coagulation abnormalities, a hallmark of EBOV infection in humans and NHPs, this is understandable given that coagulopathy is not generally seen in mouse models for acute viral infections [<xref rid="ppat-0020073-b015" ref-type="bibr">15</xref>,<xref rid="ppat-0020073-b018" ref-type="bibr">18</xref>]. Thus, this mouse model may not exactly mirror all aspects of human Ebola hemorrhagic fever; however, it does provide a relevant and convenient animal model with which to study aspects of pathogenicity and host immune response in vivo [<xref rid="ppat-0020073-b019" ref-type="bibr">19</xref>–<xref rid="ppat-0020073-b021" ref-type="bibr">21</xref>].</p><p>The adaptation of ZEBOV to adult mice resulted in a number of nucleotide changes in both the coding and noncoding regions (NCRs) of the virus genome [<xref rid="ppat-0020073-b022" ref-type="bibr">22</xref>]. To identify the molecular features that determine EBOV virulence in mice, here, we exploited a reverse genetics system to generate infectious ZEBOV entirely from cloned cDNA [<xref rid="ppat-0020073-b023" ref-type="bibr">23</xref>] and artificially generate recombinant viruses possessing various combinations of wild-type and mouse-adapted genes. The virulence of these recombinant viruses was then tested in adult immunocompetent mice.</p></sec><sec id="s2"><title>Results</title><sec id="s2a"><title>Construction and Generation of Recombinant MA-ZEBOV Mutants from cDNAs</title><p>The ZEBOV variant that served as the starting point for adaptation in mice (referred to as precursor mouse-adapted virus [pre–MA-ZEBOV]) differed from the published sequence of the wild-type ZEBOV (WT-ZEBOV), strain Mayinga, in four nucleotide positions. These mutations may have been acquired during three consecutive passages in the brains of newborn mice and/or two passages in Vero E6 cells [<xref rid="ppat-0020073-b014" ref-type="bibr">14</xref>]. Three of these mutations resulted in amino acid changes in the glycoprotein (GP), while the fourth created a silent mutation in the open reading frame (ORF) encoding VP40 (<xref ref-type="fig" rid="ppat-0020073-g001">Figure 1</xref>A and <xref ref-type="fig" rid="ppat-0020073-g001">1</xref>B). Serial passage of pre–MA-ZEBOV in progressively older suckling mice yielded the fully mouse-adapted variant (MA-ZEBOV), which possessed nine additional nucleotide changes. Nucleotide substitutions leading to amino acid changes were found in VP35, VP24, nucleoprotein (NP) (one amino acid change each), and the polymerase protein L (two amino acid changes) (<xref ref-type="fig" rid="ppat-0020073-g001">Figure 1</xref>A and <xref ref-type="fig" rid="ppat-0020073-g001">1</xref>B). An additional two nucleotide changes were found in NCRs (<xref ref-type="fig" rid="ppat-0020073-g001">Figure 1</xref>A and <xref ref-type="fig" rid="ppat-0020073-g001">1</xref>B), which may affect replication and/or transcription efficiencies. The remaining two nucleotide modifications, in the NP and L genes, were silent.</p><fig id="ppat-0020073-g001" position="float"><label>Figure 1</label><caption><title>Molecular Differences among ZEBOV Mouse-Adapted Variants</title><p>(A) Comparison of ZEBOV variants. Pre–MA-ZEBOV differs from wild-type ZEBOV (WT-ZEBOV) by three amino acids in the GP (red triangles) and a silent mutation in the ORF of the VP40 gene (gray triangle). Serial passage of pre–MA-ZEBOV in progressively older mice yielded MA-ZEBOV [<xref rid="ppat-0020073-b014" ref-type="bibr">14</xref>], which contains coding changes in the NP, VP35, VP24, and polymerase (L) (all shown in red triangles). Two nucleotide changes localize to the NCRs at the 3′ end of the VP30 and the 5′ end of the VP24 gene (red triangles); the remaining three modifications in the NP, VP40, and L ORFs are silent (gray triangles).</p><p>(B) Nucleotide and amino acid differences among WT-ZEBOV, pre–MA-ZEBOV, MA-ZEBOV, and MA-RG. The nucleotide changes in GP of pre–MA-ZEBOV, compared to WT-ZEBOV, as well as the changes acquired during adaptation of pre–MA-ZEBOV in mice are shown in red.</p><p>−, no change.</p></caption><graphic xlink:href="ppat.0020073.g001"/></fig><p>Using a reverse genetics system to create ZEBOV from plasmid DNAs [<xref rid="ppat-0020073-b023" ref-type="bibr">23</xref>], we generated MA-RG (mouse-adapted reverse genetics virus) by introducing all but three mutations into the backbone of WT-ZEBOV (<xref ref-type="fig" rid="ppat-0020073-g001">Figure 1</xref>B). MA-RG differed from MA-ZEBOV by the three silent mutations in the NP, VP40, and L ORFs that were introduced during adaptation in mice (<xref ref-type="fig" rid="ppat-0020073-g001">Figure 1</xref>B). We also generated a variety of recombinant viruses containing various subsets of the mutations found in MA-ZEBOV (depicted in <xref ref-type="fig" rid="ppat-0020073-g002">Figure 2</xref>A) to identify those mutations responsible for mouse adaptation.</p><fig id="ppat-0020073-g002" position="float"><label>Figure 2</label><caption><title>Genetic Determinants of Virulence in Adult Mice</title><p>(A) Determination of MLD<sub>50</sub> values. Bars indicate the genotype of the viral gene: MA-ZEBOV (red), WT-ZEBOV (blue). The MLD<sub>50</sub> values of recombinant viruses were determined by i.p. inoculation of mice (three to six per group) with serial 10-fold dilutions of virus stock and then monitoring of survival rates. Experiments were carried out in duplicate.</p><p>(B) Determination of the mean time to death and dose ranges causing morbidity/mortality. The mean time to death of mice inoculated with 10 FFU (approximately 1,000 MLD<sub>50</sub> for MA-ZEBOV) are indicated. Differences in the mean time to death for mice infected with various mutants, compared to that of mice infected with MA-ZEBOV, were considered significant when the <italic>p-</italic>value was <0.05. The dose range for morbidity/mortality was determined by inoculating groups of six to nine mice with the indicated amounts of viruses and monitoring the mice for weight loss and time to death. Survival numbers (dead/total) are color-coded to indicate the severity of infection in infected mice: no disease (black), illness without mortality (green), less than or equal to 50% mortality (purple), greater than 50% mortality (red). Dead/total. *<italic>p</italic> < 0.05.</p></caption><graphic xlink:href="ppat.0020073.g002"/></fig></sec><sec id="s2b"><title>The VP24 and NP Genes Determine ZEBOV Virulence in Mice</title><p>To identify the genetic determinants of ZEBOV virulence in mice, we inoculated groups of BALB/c mice (5 to 6 wk old; three to six per group) intraperitoneally (i.p.) with 10 focus-forming units (FFU) of WT-ZEBOV, the original mouse-adapted virus (MA-ZEBOV), its counterpart generated using reverse genetics (MA-RG), or viruses containing subsets of the mutations found in MA-ZEBOV (<xref ref-type="fig" rid="ppat-0020073-g002">Figure 2</xref>A). This dose (10 FFU) corresponded to 1,000 MLD<sub>50</sub> (the amount of virus required to kill 50% of the mice) for MA-ZEBOV. As expected, infection of mice with WT-ZEBOV did not produce a lethal infection or any clinical symptoms, unlike infection with either MA-ZEBOV or MA-RG. Interestingly, viruses containing the wild-type NP or VP24 genes in the background of MA-ZEBOV (MA-NP<sub>WT</sub>, MA-24<sub>WT</sub>) did not cause lethal infection or clinical symptoms in mice, suggesting a critical role for mutations in these proteins with respect to the virulence of MA-ZEBOV in mice. In contrast, substitution of the VP35, GP, and/or L genes of MA-RG with those of WT-ZEBOV did not dramatically reduce the virulence of the virus (<xref ref-type="fig" rid="ppat-0020073-g002">Figure 2</xref>A). Determination of the MLD<sub>50</sub>, by i.p. inoculation of mice with 10-fold serial dilutions of viruses, revealed almost identical MLD<sub>50</sub> values for all of these recombinant viruses with few exceptions (<xref ref-type="fig" rid="ppat-0020073-g002">Figure 2</xref>A). We, therefore, tested the inverse genotypes by introducing mouse-adapted NP and VP24 genes, separately and in combination, into the background of WT-ZEBOV. Only WT-NP/24<sub>MA</sub> resulted in a lethal phenotype; neither of the single gene introductions (WT-NP<sub>MA</sub>, WT-24<sub>MA</sub>) caused disease in mice. These findings identified the specific amino acid changes in NP and VP24 that are the critical determinants of ZEBOV virulence in adult mice.</p></sec><sec id="s2c"><title>Multiple Mutations in MA-ZEBOV Contribute to Enhanced Virulence in Mice</title><p>All mice infected with virulent viruses displayed the same clinical symptoms (ruffled fur, decreased activity, and weight loss), but notable differences were observed in the mean time to death and in the dose ranges that caused morbidity or mortality (<xref ref-type="fig" rid="ppat-0020073-g002">Figure 2</xref>B). Infection with MA-ZEBOV resulted in the shortest mean time to death, with lethal infection induced by a broad range of inoculum doses (10<sup>5</sup> to 10<sup>−3</sup> FFU/mouse). The artificially generated MA-RG virus was slightly attenuated, suggesting that the silent mutations in the NP, VP40, and L genes, by which this mutant differs from MA-ZEBOV, may make a minor contribution to virulence.</p><p>Similarly, the introduction of the mouse-adapted NP and VP24 genes into the background of WT-ZEBOV (WT-NP/24<sub>MA</sub>) did not bestow on WT-ZEBOV the fully pathogenic phenotype, as demonstrated by a prolonged time to death and a narrower dose range causing mortality than that of MA-ZEBOV. Insertion of the mutation into the 5′ NCR of the VP24 gene of WT-NP/24<sub> MA</sub> (i.e., WT-NP/24nc/24<sub>MA</sub>) reduced the time to death but did not appreciably expand the dose range for morbidity/mortality, indicating a rather minor contribution of this mutation to virulence in mice. In contrast, the nucleotide substitution in the 3′ NCR of VP30 caused a more appreciable effect on mouse virulence (compare MA-GP<sub>WT</sub> and MA-GP/30nc<sub>WT</sub>, which differ only in the VP30 NCR).</p><p>Single gene replacements of the mouse-adapted VP35, GP, or L with their wild-type counterparts (MA-VP35<sub>WT</sub>, MA-GP<sub>WT</sub> MA-L<sub>WT</sub>) led to different degrees of attenuation, indicating that all of the mutations contributed to some extent to the virulence of MA-ZEBOV in adult mice. Of these mutant viruses, MA-L<sub>WT</sub> showed the most pronounced attenuation, indicating that the two amino acid changes in the L protein of MA-ZEBOV are more critical for virulence in mice than those in the VP35 or GP proteins.</p></sec><sec id="s2d"><title>Growth Characteristics in Mouse Organs</title><p>We next compared the growth characteristics of representative mutant viruses in mice infected with 5 FFU of virus by examining viral titers in the serum (<xref ref-type="fig" rid="ppat-0020073-g003">Figure 3</xref>A) and in two target organs of EBOV, the spleen and liver (<xref ref-type="fig" rid="ppat-0020073-g003">Figure 3</xref>B and <xref ref-type="fig" rid="ppat-0020073-g003">3</xref>C). As expected, WT-ZEBOV replicated poorly in mice, whereas the virus titers of MA-RG were the highest in all organs tested, again demonstrating that the full set of mutations is required to establish the fully pathogenic phenotype. Efficient virus replication required both the mouse-adapted NP and VP24 genes (compare WT-NP/24<sub>MA</sub> with WT-NP<sub>MA</sub> and WT-VP24<sub>MA</sub>), a finding that correlated well with virulence, as measured by lethality (<xref ref-type="fig" rid="ppat-0020073-g002">Figure 2</xref>A and <xref ref-type="fig" rid="ppat-0020073-g002">2</xref>B). Likewise, the introduction of either the wild-type NP or VP24 gene into the background of MA-ZEBOV attenuated the resulting recombinant viruses (MA-NP<sub>WT</sub>, MA-24<sub>WT</sub>), further emphasizing the critical role of these mutations for viral replication in mice.</p><fig id="ppat-0020073-g003" position="float"><label>Figure 3</label><caption><title>Growth Characteristics of Recombinant Viruses in Mice</title><p>Groups of 12 mice were inoculated i.p. with 5 FFU (approximately 500 LD<sub>50</sub> values for MA-ZEBOV) of representative viruses. On days 1, 2, 3, and 5 postinoculation, selected organs were removed from three infected animals per group. Virus titers in serum (A), spleen (B) and liver (C) were determined in Vero E6 cells by using a focus-forming assay [<xref rid="ppat-0020073-b036" ref-type="bibr">36</xref>].</p></caption><graphic xlink:href="ppat.0020073.g003"/></fig></sec><sec id="s2e"><title>Do Mutations in VP24 and NP Affect Virus Sensitivity to IFN?</title><p>The suppression of viral multiplication by type I IFN-mediated antiviral responses has been demonstrated in many viruses [<xref rid="ppat-0020073-b024" ref-type="bibr">24</xref>], including EBOV [<xref rid="ppat-0020073-b025" ref-type="bibr">25</xref>]. In fact, it is abrogation of the type I IFN (IFN-α/β) system that makes adult mice susceptible to EBOV infection [<xref rid="ppat-0020073-b025" ref-type="bibr">25</xref>]. We and others have identified EBOV VP24 as being instrumental in this function (i.e., in inhibiting IFN signaling) [<xref rid="ppat-0020073-b007" ref-type="bibr">7</xref>,<xref rid="ppat-0020073-b026" ref-type="bibr">26</xref>]. To determine whether the amino acid changes in the VP24 and NP proteins of MA-ZEBOV facilitate evasion of the type I IFN-induced antiviral response, we compared virus growth of selected recombinant viruses in a mouse peritoneal macrophage cell line (RAW 264.7 cells) in the absence or presence of murine IFN-α/β (<xref ref-type="fig" rid="ppat-0020073-g004">Figure 4</xref>). WT-ZEBOV grew in nonstimulated cells, albeit to lower titers compared to the growth of the other viruses tested (<xref ref-type="fig" rid="ppat-0020073-g004">Figure 4</xref>A); however, in cells stimulated with IFN at 2 h postinfection (<xref ref-type="fig" rid="ppat-0020073-g004">Figure 4</xref>B) or 12 h prior to and 2 h after infection (<xref ref-type="fig" rid="ppat-0020073-g004">Figure 4</xref>C), virus replication was severely reduced. In contrast, MA-RG grew to much higher titers even in IFN-stimulated cells, suggesting that its virulence in mice is linked to its ability to counteract IFN-induced antiviral responses. This ability clearly correlated with the mutations in the mouse-adapted VP24 and NP genes. In fact, WT-NP/VP24<sub>MA</sub> grew more efficiently in IFN-stimulated cells than did MA-RG, for unknown reasons. In cells that were stimulated with IFN postinfection, WT-NP<sub>MA</sub> or WT-VP24<sub>MA</sub> grew better than WT-ZEBOV. In contrast, WT-VP24<sub>MA</sub>, but not WT-NP<sub>MA</sub>, failed to replicate efficiently in cells treated with IFN prior to and after infection. These data indicate that the ability to counteract IFN-induced antiviral responses is responsible for the high virulence of MA-ZEBOV in this animal model and that IFN evasion may be critical for EBOV virulence in other animal models, as well as in humans. In addition, our findings show that while both mouse-adapted VP24 and NP provide the virus with the ability to counteract the IFN-induced antiviral responses, NP may be more critical for this effect in vitro. The concerted actions of both mutations, however, appear necessary for evading IFN-induced antiviral responses in vivo.</p><fig id="ppat-0020073-g004" position="float"><label>Figure 4</label><caption><title>Effect of Murine Type I IFNs on Recombinant Virus Replication in Mouse Macrophages</title><p>RAW 264.7 cells (mouse peritoneal macrophage-derived cell line) were infected with a multiplicity of infection of 0.05. Cells were untreated (A), treated with murine IFN-α/β (500 units/ml) 2 h postinfection (B), or treated with murine IFN-α/β (500 units/ml) 12 h prior to and again 2 h after virus adsorption (C). Supernatants were collected on days 0, 1, 2, 3, and 4 postinfection and titrated by use of a focus-forming unit assay in Vero E6 cells [<xref rid="ppat-0020073-b036" ref-type="bibr">36</xref>].</p></caption><graphic xlink:href="ppat.0020073.g004"/></fig></sec></sec><sec id="s3"><title>Discussion</title><p>In this study, we have identified molecular determinants of ZEBOV virulence in mice by using reverse genetics. We found VP24 and NP to be the primary determinants for adaptation of WT-ZEBOV in mice and found a correlation between virulence and the ability of the virus to evade the type I IFN-induced antiviral response. The ability to overcome the IFN-induced antiviral response is, therefore, a critical event in the pathogenesis of ZEBOV infection in mice and possibly in humans. Moreover, mutations in other viral proteins and NCRs also contribute to the virulent phenotype, indicating that virulence is a multifactor trait.</p><p>Recent in vitro studies suggest that VP24 functions as a type I IFN antagonist [<xref rid="ppat-0020073-b007" ref-type="bibr">7</xref>,<xref rid="ppat-0020073-b026" ref-type="bibr">26</xref>], but the significance of this finding had not been addressed in vivo until now. Here, we have demonstrated that single amino acid modifications in VP24 and NP of MA-ZEBOV are critical for virus evasion of IFN-induced antiviral responses. NP has not previously been considered as a potential IFN antagonist. However, given that it interacts with VP24 in the formation of nucleocapsids [<xref rid="ppat-0020073-b027" ref-type="bibr">27</xref>,<xref rid="ppat-0020073-b028" ref-type="bibr">28</xref>], it is not unreasonable to imagine that it may act as an enhancer and/or stabilizer of VP24 functions. Nevertheless, since the NP mutation acquired during mouse adaptation of WT-ZEBOV alone allowed the recombinant virus to replicate in IFN-treated cells (see <xref ref-type="fig" rid="ppat-0020073-g004">Figure 4</xref>C), NP likely plays a more direct role in the evasion of the IFN-induced antiviral response. Interestingly, WT-NP/VP24<sub>MA</sub> grew more efficiently than MA-RG in IFN-stimulated cells, in contrast to our in vivo results. For this reason, it seems likely that the mutations in VP24 and NP are critical for resistance to IFN-induced antiviral responses, while the remaining mutations acquired during adaptation of WT-ZEBOV facilitate efficient virus replication and/or spread in mice despite their attenuated phenotype in cell culture systems.</p><p>VP35 and GP have previously been linked to EBOV pathogenicity [<xref rid="ppat-0020073-b004" ref-type="bibr">4</xref>–<xref rid="ppat-0020073-b006" ref-type="bibr">6</xref>,<xref rid="ppat-0020073-b009" ref-type="bibr">9</xref>,<xref rid="ppat-0020073-b010" ref-type="bibr">10</xref>]. VP35 is an IFN antagonist that interferes with type I IFN synthesis by inhibiting IRF-3 (interferon regulatory factor 3) activation, a necessary step for the transcription of IFN genes [<xref rid="ppat-0020073-b009" ref-type="bibr">9</xref>,<xref rid="ppat-0020073-b010" ref-type="bibr">10</xref>]. Although the mutation found in the mouse-adapted VP35 protein was not responsible for the enhanced virulence of MA-ZEBOV (<xref ref-type="fig" rid="ppat-0020073-g002">Figure 2</xref>B), this does not necessarily indicate that VP35-mediated regulation of IFN levels does not play a part in the pathogenicity of EBOV. This activity may still be important for EBOV to achieve high virulence. Likewise, GP is cytotoxic and is, therefore, thought to contribute to viral pathogenicity [<xref rid="ppat-0020073-b005" ref-type="bibr">5</xref>,<xref rid="ppat-0020073-b029" ref-type="bibr">29</xref>,<xref rid="ppat-0020073-b030" ref-type="bibr">30</xref>]. However, a recent report suggested that this cytotoxicity originated from overexpression of GP in cells [<xref rid="ppat-0020073-b011" ref-type="bibr">11</xref>]. As with VP35, the lack of an adaptive mutation in the GP of MA-ZEBOV does not necessarily diminish the role of GP in viral pathogenicity.</p><p>Interestingly, the adaptation of ZEBOV in guinea pigs also resulted in amino acid changes in NP, VP24, and L and in a nucleotide substitution in the VP30 NCR [<xref rid="ppat-0020073-b008" ref-type="bibr">8</xref>]. Although these mutations differed in their amino acid positions from those in MA-ZEBOV, it seems likely that they serve a similar role in adaptation and also function by counteracting innate antiviral responses [<xref rid="ppat-0020073-b031" ref-type="bibr">31</xref>,<xref rid="ppat-0020073-b032" ref-type="bibr">32</xref>]. As with mice, additional mutations (e.g., in L and the VP30 NCR) likely contribute to virulence by affecting viral transcription/replication.</p><p>The mouse model using MA-ZEBOV does not exactly mirror all aspects of human Ebola hemorrhagic fever [<xref rid="ppat-0020073-b015" ref-type="bibr">15</xref>,<xref rid="ppat-0020073-b018" ref-type="bibr">18</xref>]. Thus, determinants for virulence of ZEBOV may differ between mouse and primate models. However, since ZEBOV is naturally lethal to primates, but not to mice, this model provided an opportunity to decipher the roles of viral proteins in expression of high virulence in a host (i.e., mice). Our studies showed that ZEBOV VP24 and NP are inadequate for the expression of high virulence in mice, but upon mutation, optimally expressed this property. Thus, it is possible that these viral proteins play an important role in expression of high virulence in primates. Likewise, the lack of VP35 mutations correlating with pathogenicity in mice is interesting, indicating that ZEBOV VP35 optimally functions in both primates and mice without additional mutation in an IFN pathway. Since MA-ZEBOV is attenuated in NHPs [<xref rid="ppat-0020073-b018" ref-type="bibr">18</xref>], one or more genes into which mutations were introduced during mouse adaptation of ZEBOV likely play a role in virulence in primates. Therefore, it will be interesting to examine the virulence of selected recombinant mouse-adapted variants in NHPs. Such studies will provide us with valuable information for understanding ZEBOV pathogenesis in humans.</p><p>Of note, we observed resistance in mice to infection with high infectious doses of MA-ZEBOV (<xref ref-type="fig" rid="ppat-0020073-g002">Figure 2</xref>B). A similar finding has been observed in a mouse model for rabies virus [<xref rid="ppat-0020073-b033" ref-type="bibr">33</xref>]. The most likely explanation for this finding is that inoculation with high doses of virus causes a rapid stimulation of the innate immune response before virus replication or spread can occur. This topic presents an attractive research subject that may lead to control measures for EBOV infections through immunologic modulation of host responses to viral infection.</p><p>In conclusion, the combination of reverse genetics technology [<xref rid="ppat-0020073-b034" ref-type="bibr">34</xref>] and a small-animal model has allowed us to gain valuable insights into EBOV pathogenesis. Understanding the molecular basis of mouse adaptation of ZEBOV will likely lead to the identification of viral genetic determinants of EBOV virulence and to the elucidation of the roles of specific viral proteins in the pathogenic process. A more detailed molecular understanding of virulence and the host responses will also be crucial to improving our ability to control EBOV infections in the future.</p></sec><sec id="s4"><title>Materials and Methods</title><sec id="s4a"><title>Cells.</title><p>Vero E6 (African green monkey kidney) cells, 293T (human embryonic kidney) cells, and RAW 264.7 cells (mouse peritoneal macrophages) were grown in DMEM supplemented with 10% FBS, 2 mM <sc>l</sc>-glutamine, and penicillin (100 U/ml)-streptomycin (100 μg/ml). Cells were incubated at 37 °C in 5% CO<sub>2</sub>.</p></sec><sec id="s4b"><title>Generation of mutant Ebola viruses.</title><p>Starting with a cDNA clone encoding WT-ZEBOV, strain Mayinga [<xref rid="ppat-0020073-b023" ref-type="bibr">23</xref>], we introduced stepwise mutations to reproduce MA-ZEBOV, using PCR-based mutagenesis. The resulting plasmids, which are flanked by T7 RNA polymerase promoter and ribozyme sequences [<xref rid="ppat-0020073-b023" ref-type="bibr">23</xref>], were cotransfected into a mixed culture of Vero E6 and 293T cells, together with helper plasmids for the expression of T7 RNA polymerase and EBOV NP, VP35, VP30, and L (required components of the viral replication complex), following established protocols [<xref rid="ppat-0020073-b023" ref-type="bibr">23</xref>,<xref rid="ppat-0020073-b035" ref-type="bibr">35</xref>]. All viruses were amplified once in Vero E6 cells. DMEM supplemented with 2% FBS was used to prepare virus stocks.</p><p>Virus infectivity titers (FFU) were obtained by counting the number of infected cell foci detected by use of an indirect immunofluorescent antibody assay, as previously described [<xref rid="ppat-0020073-b036" ref-type="bibr">36</xref>]. EBOV antigen-positive foci were detected with a rabbit polyclonal anti-VP40 antibody and a goat anti-rabbit IgG-FITC conjugate [<xref rid="ppat-0020073-b037" ref-type="bibr">37</xref>].</p></sec><sec id="s4c"><title>Animal experiments.</title><p>Five- to 6-wk-old female BALB/c mice were obtained from a commercial supplier (Charles River Laboratories, Wilmington, Massachusetts, United States). All mice were housed in microisolator cages and allowed to acclimatize for 5 days prior to use in experiments.</p><p>To assay virulence, groups of three to six mice were each inoculated i.p. at two different sites with 10 FFU of virus in 0.1 ml of DMEM. Following infection, mice were observed daily for clinical symptoms and their weights were recorded for 11 d postinoculation. All surviving animals were observed for at least 21 d (three times the average duration of survival of the control animals).</p><p>The MLD<sub>50</sub> was determined by i.p. inoculation of mice (three to six per group) with serial 10-fold dilutions of virus and monitoring of the survival rates.</p><p>To assess virus growth characteristics in mice, groups of 12 animals were inoculated i.p. with 5 FFU of virus (corresponds to approximately 500 MLD<sub>50</sub> for MA-ZEBOV). On days 1, 2, 3, and 5 postinfection, spleen, liver, and blood were collected from three infected mice, and the spleen and liver samples were homogenized. Viral infectivity titers were determined by use of a focus-forming assay in Vero E6 cells [<xref rid="ppat-0020073-b036" ref-type="bibr">36</xref>].</p><p>All work with live EBOV was performed in the BSL-4 laboratory of the National Microbiology Laboratory of the Public Health Agency of Canada. All animal experiments were performed in accordance with approved animal use documents and according to the guidelines of the Canadian Council on Animal Care.</p></sec><sec id="s4d"><title>Replication kinetics in IFN-stimulated murine cells.</title><p>RAW 264.7 cells were infected with the respective viruses at a multiplicity of infection of 0.05. The cells were either left untreated or treated with 500 units/ml murine IFN-α/β 2 h postinfection, or 12 h prior to infection and again 2 h postinfection. For all samples, virus titers in the supernatants were determined on days 0, 1, 2, 3, and 4 postinfection by use of the focus-forming assay in Vero E6 cells [<xref rid="ppat-0020073-b036" ref-type="bibr">36</xref>]. The first sample (day 0) was collected after the virus had adsorbed and the cell monolayer had been washed three times.</p></sec><sec id="s4e"><title>Statistical analyses.</title><p>All virus titers in the growth kinetics experiments are shown as the mean ± SEM. The <italic>p-</italic>values in <xref ref-type="fig" rid="ppat-0020073-g002">Figure 2</xref>B were calculated by using the Student's <italic>t</italic>-test (two-tailed distribution, two-sample unequal variance). <italic>p</italic> < 0.05 was considered to indicate statistical significance.</p></sec></sec><sec sec-type="supplementary-material" id="s5"><title>Supporting Information</title><sec id="s5a"><title>Accession Numbers</title><p>The GenBank accession numbers for the genes mentioned in this paper are WT-ZEBOV, strain Mayinga (AF086833) and MA-ZEBOV variant (AF499101).</p></sec></sec>
Macrophage Pro-Inflammatory Response to <named-content content-type="genus-species">Francisella novicida</named-content> Infection Is Regulated by SHIP	<p> <italic>Francisella tularensis,</italic> a Gram-negative facultative intracellular pathogen infecting principally macrophages and monocytes, is the etiological agent of tularemia. Macrophage responses to <named-content content-type="genus-species">F. tularensis</named-content> infection include the production of pro-inflammatory cytokines such as interleukin (IL)-12, which is critical for immunity against infection. Molecular mechanisms regulating production of these inflammatory mediators are poorly understood. Herein we report that the SH2 domain-containing inositol phosphatase (SHIP) is phosphorylated upon infection of primary murine macrophages with the genetically related F. novicida, and negatively regulates F. novicida–induced cytokine production. Analyses of the molecular details revealed that in addition to activating the MAP kinases, F. novicida infection also activated the phosphatidylinositol 3-kinase (PI3K)/Akt pathway in these cells. Interestingly, SHIP-deficient macrophages displayed enhanced Akt activation upon <named-content content-type="genus-species">F. novicida</named-content> infection, suggesting elevated PI3K-dependent activation pathways in absence of SHIP. Inhibition of PI3K/Akt resulted in suppression of <named-content content-type="genus-species">F. novicida</named-content>–induced cytokine production through the inhibition of NFκB. Consistently, macrophages lacking SHIP displayed enhanced NFκB-driven gene transcription, whereas overexpression of SHIP led to decreased NFκB activation. Thus, we propose that SHIP negatively regulates <italic>F. novicida–</italic>induced inflammatory cytokine response by antagonizing the PI3K/Akt pathway and suppressing NFκB-mediated gene transcription. A detailed analysis of phosphoinositide signaling may provide valuable clues for better understanding the pathogenesis of tularemia.</p>	<contrib contrib-type="author"><name><surname>Parsa</surname><given-names>Kishore V. L</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Ganesan</surname><given-names>Latha P</given-names></name><xref ref-type="aff" rid="aff2">2</xref></contrib><contrib contrib-type="author"><name><surname>Rajaram</surname><given-names>Murugesan V. S</given-names></name><xref ref-type="aff" rid="aff2">2</xref></contrib><contrib contrib-type="author"><name><surname>Gavrilin</surname><given-names>Mikhail A</given-names></name><xref ref-type="aff" rid="aff2">2</xref></contrib><contrib contrib-type="author"><name><surname>Balagopal</surname><given-names>Ashwin</given-names></name><xref ref-type="aff" rid="aff3">3</xref></contrib><contrib contrib-type="author"><name><surname>Mohapatra</surname><given-names>Nrusingh P</given-names></name><xref ref-type="aff" rid="aff3">3</xref></contrib><contrib contrib-type="author"><name><surname>Wewers</surname><given-names>Mark D</given-names></name><xref ref-type="aff" rid="aff2">2</xref></contrib><contrib contrib-type="author"><name><surname>Schlesinger</surname><given-names>Larry S</given-names></name><xref ref-type="aff" rid="aff2">2</xref><xref ref-type="aff" rid="aff3">3</xref></contrib><contrib contrib-type="author"><name><surname>Gunn</surname><given-names>John S</given-names></name><xref ref-type="aff" rid="aff3">3</xref></contrib><contrib contrib-type="author"><name><surname>Tridandapani</surname><given-names>Susheela</given-names></name><xref ref-type="aff" rid="aff1">1</xref><xref ref-type="aff" rid="aff2">2</xref><xref ref-type="corresp" rid="cor1">*</xref></contrib>	PLoS Pathogens	<sec id="s1"><title>Introduction</title><p> <italic>Francisella tularensis,</italic> causative agent of the zoonotic disease tularemia, is a Gram-negative intracellular pathogen. There are four different subspecies of <named-content content-type="genus-species">F. tularensis</named-content>. The <named-content content-type="genus-species">F. tularensis</named-content> subspecies <italic>tularensis</italic> is the most virulent of the four with a 50% lethal dose (LD<sub>50</sub>) less than 10 colony forming units (CFUs) for humans [<xref rid="ppat-0020071-b001" ref-type="bibr">1</xref>]. Other less virulent subspecies of <italic>F. tularensis</italic> include <italic>novicida, holoarctica,</italic> and <italic>mediasiatica</italic>.</p><p> <named-content content-type="genus-species">F. tularensis</named-content> infects primarily monocytes and macrophages. The survival strategy adopted by <named-content content-type="genus-species">F. tularensis</named-content> in the host cell is to avoid phago-lysosomal fusion. After a few hours of phagocytosis of the organism, the membrane of the phagosome ruptures and <named-content content-type="genus-species">F. tularensis</named-content> is released into the host cell cytosol [<xref rid="ppat-0020071-b001" ref-type="bibr">1</xref>,<xref rid="ppat-0020071-b002" ref-type="bibr">2</xref>]. The release of the pathogen into the cytosol requires expression of bacterial proteins such as IglC and MglA [<xref rid="ppat-0020071-b003" ref-type="bibr">3</xref>–<xref rid="ppat-0020071-b005" ref-type="bibr">5</xref>]. On the other hand, bacterial escape is inhibited by interferon-γ (IFNγ) treatment of the cells infected with <named-content content-type="genus-species">F. tularensis</named-content> [<xref rid="ppat-0020071-b006" ref-type="bibr">6</xref>–<xref rid="ppat-0020071-b008" ref-type="bibr">8</xref>].</p><p>The host cell responses to <named-content content-type="genus-species">F. tularensis</named-content> infection are not clearly understood. Interleukin (IL)-12 and IFNγ have been reported to be critical for immunity against <italic>Francisella</italic> infection [<xref rid="ppat-0020071-b009" ref-type="bibr">9</xref>]. Although natural killer (NK) cells are thought to be the major source of IFNγ, IL-12 is produced by infected macrophages [<xref rid="ppat-0020071-b010" ref-type="bibr">10</xref>]. Further, IL-12 also appears to strongly induce IFNγ production. Attesting to the importance of IL-12 in immunity against <named-content content-type="genus-species">F. tularensis</named-content> LVS infection, IL-12 knockout animals, or animals treated with IL-12 neutralizing antibodies are unable to clear the bacteria [<xref rid="ppat-0020071-b011" ref-type="bibr">11</xref>]. Macrophages also produce several other pro-inflammatory cytokines/chemokines upon infection [<xref rid="ppat-0020071-b008" ref-type="bibr">8</xref>,<xref rid="ppat-0020071-b012" ref-type="bibr">12</xref>,<xref rid="ppat-0020071-b013" ref-type="bibr">13</xref>]. Intracellular signaling molecules involved in macrophage response to <italic>Francisella</italic> are not well defined. Although pathways involving NFκB and the MAPKs p38 and JNK have been reported to be activated during <named-content content-type="genus-species">F. tularensis</named-content> LVS infection, the exact role of these signaling pathways in macrophage responses is not clear [<xref rid="ppat-0020071-b013" ref-type="bibr">13</xref>].</p><p>The Src homology 2 (SH2) domain-containing inositol 5′ phosphatase (SHIP) is a hematopoietic cell-specific phosphatase that negatively regulates phosphatidylinositol 3-kinase (PI3K) pathway by consuming the lipid products of PI3K [<xref rid="ppat-0020071-b014" ref-type="bibr">14</xref>,<xref rid="ppat-0020071-b015" ref-type="bibr">15</xref>]. SHIP is a constitutively active enzyme located in cytosol. Membrane translocation of SHIP is required for access to its substrates. Upon stimulation of immunoreceptors, growth factor/cytokine receptors or toll-like receptors, SHIP translocates to the membrane where it is phosphorylated by membrane-associated Src kinases [<xref rid="ppat-0020071-b016" ref-type="bibr">16</xref>–<xref rid="ppat-0020071-b019" ref-type="bibr">19</xref>]. In addition to the central catalytic domain, SHIP also contains an N-terminal SH2 domain, a C-terminal proline-rich domain, and two NPxY motifs, which can all associate with other cellular signaling proteins. Thus, SHIP mediates its functions by both its catalytic domain and its protein–protein interaction domains. For example, the catalytic domain of SHIP regulates cellular responses by hydrolyzing PI3,4,5P<sub>3</sub> into PI3,4P<sub>2</sub>, thus antagonizing the PI3K pathway [<xref rid="ppat-0020071-b015" ref-type="bibr">15</xref>]). SHIP is also reported to suppress activation of the Ras pathway, in some cases, by virtue of its interaction with Dok, a Ras GAP activator [<xref rid="ppat-0020071-b020" ref-type="bibr">20</xref>], and/or its interactions with the Ras adapter Shc [<xref rid="ppat-0020071-b021" ref-type="bibr">21</xref>–<xref rid="ppat-0020071-b023" ref-type="bibr">23</xref>].</p><p>Herein, we have investigated the role of SHIP in <named-content content-type="genus-species">F. novicida</named-content>–induced cytokine response in murine macrophages. We report that SHIP negatively regulates the production of IL-12, IL-6, and RANTES by <named-content content-type="genus-species">F. novicida</named-content>–infected macrophages. Our current studies indicate that the production of these cytokines requires the activation of the PI3K pathway and involves NFκB activation. These studies also demonstrate that <italic>F. novicida–</italic>induced pro-inflammatory cytokine production is negatively regulated by SHIP by opposing the PI3K pathway and NFκB-driven transcriptional activation. Thus, we conclude that SHIP is a regulator of macrophage innate immune responses to <named-content content-type="genus-species">F. novicida</named-content> infection.</p></sec><sec id="s2"><title>Results</title><sec id="s2a"><title> <italic>F. novicida–</italic>Induced Pro-Inflammatory Cytokine Response Is Down-Regulated by SHIP</title><p>The inositol phosphatase SHIP plays an important role in regulating macrophage innate immune responses to IgG immune complexes and bacterial products [<xref rid="ppat-0020071-b016" ref-type="bibr">16</xref>,<xref rid="ppat-0020071-b024" ref-type="bibr">24</xref>–<xref rid="ppat-0020071-b029" ref-type="bibr">29</xref>]. Recent studies have indicated that while SHIP negatively regulates TLR2 function [<xref rid="ppat-0020071-b030" ref-type="bibr">30</xref>], it promotes TLR4 function [<xref rid="ppat-0020071-b016" ref-type="bibr">16</xref>,<xref rid="ppat-0020071-b027" ref-type="bibr">27</xref>,<xref rid="ppat-0020071-b029" ref-type="bibr">29</xref>]. SHIP is a constitutively active, cytoplasmic enzyme that must translocate to the membrane where it accesses its substrate PI3,4,5P<sub>3</sub>. Membrane translocation of SHIP is accompanied by tyrosine phosphorylation of SHIP by membrane-associated Src kinases. Thus, tyrosine phosphorylation of SHIP is often used as an indicator of SHIP translocation to the membrane [<xref rid="ppat-0020071-b016" ref-type="bibr">16</xref>]. To address the role of SHIP in <named-content content-type="genus-species">F. novicida</named-content>–stimulated host cell response, RAW 264.7 murine macrophage cells were infected with <named-content content-type="genus-species">F. novicida</named-content>. SHIP proteins were immunoprecipitated from uninfected and infected cells, and analyzed by Western blotting with phosphotyrosine antibody (<xref ref-type="fig" rid="ppat-0020071-g001">Figure 1</xref>A, upper panel). The same membrane was reprobed with SHIP antibody (lower panel). Results indicate that SHIP is tyrosine phosphorylated in cells infected with <italic>F. novicida,</italic> and that robust phosphorylation of SHIP occurs at the later time points (30 and 60 min post infection). Having determined the time course of SHIP phosphorylation in RAW 264.7 cells, we next tested whether these findings could be validated in primary cells. Here murine bone marrow-derived macrophages (BMMs) were infected with <named-content content-type="genus-species">F. novicida</named-content> for varying times, and phosphorylation of SHIP was analyzed by Western blotting with a phospho-specific SHIP antibody (<xref ref-type="fig" rid="ppat-0020071-g001">Figure 1</xref>B, upper panel). Results indicated that infection of BMMs by <named-content content-type="genus-species">F. novicida</named-content> induces tyrosine phosphorylation of SHIP (<xref ref-type="fig" rid="ppat-0020071-g001">Figure 1</xref>B). To ensure equal loading of protein in all lanes, the same membrane was reprobed with anti-SHIP antibody. These results suggest a potential involvement of SHIP in <named-content content-type="genus-species">F. novicida</named-content>–induced macrophage signaling.</p><fig id="ppat-0020071-g001" position="float"><label>Figure 1</label><caption><title>SHIP Down-Regulates Macrophage Pro-Inflammatory Response to <named-content content-type="genus-species">F. novicida</named-content> Infection</title><p>(A) RAW 264.7 murine macrophage cells were infected with <named-content content-type="genus-species">F. novicida</named-content> (MOI = 100). SHIP immunoprecipitates from uninfected and infected cells were analyzed by Western blotting with anti-phosphotyrosine antibody (Anti-pY; upper panel). The lower panel is a reprobe of the same membrane with anti-SHIP antibody. The lane marked “Ctrl” represents immunoprecipitate with control antibody.</p><p>(B) BMMs were infected with <named-content content-type="genus-species">F. novicida</named-content> for the times shown in the figure. Protein-matched lysates were separated by SDS/PAGE and analyzed by Western blotting with antibodies specific for phosphorylated SHIP (Anti-pSHIP; upper panel). The lower panel is a reprobe with SHIP antibody (Anti-SHIP).</p><p>(C) BMMs from SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> littermate mice were infected for 8 h with <named-content content-type="genus-species">F. novicida</named-content>. Cell supernatants from uninfected and infected cells were analyzed by ELISA for IL-12, IL-6, and RANTES. The graph represents mean and standard deviation (SD) of values obtained from three independent experiments. Data were analyzed by paired Student <italic>t</italic> test. An asterisk () indicates <italic>p</italic>-value < 0.05.</p><p>(D) Protein-matched lysates from SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> BMMs were analyzed by Western blotting with SHIP antibody (upper panel). The same membrane was reprobed with actin antibody (Anti-Actin; lower panel).</p></caption><graphic xlink:href="ppat.0020071.g001"/></fig><p>In order to probe the functional consequence of SHIP activation, <named-content content-type="genus-species">F. novicida</named-content>–induced cytokine production was compared in BMMs obtained from SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> littermate mice. Thus, SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> BMMs were infected with <italic>F. novicida,</italic> and cell supernatants from uninfected and infected cells were harvested 8 h post infection and analyzed by enzyme-linked immunosorbent assay (ELISA) for IL-12, IL-6, and RANTES. As seen in <xref ref-type="fig" rid="ppat-0020071-g001">Figure 1</xref>C, SHIP<sup>−/−</sup> BMMs produced significantly elevated levels of pro-inflammatory mediators compared to their wild-type counterparts. Similar results were obtained with peritoneal macrophages isolated from SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> littermates, and indicate that SHIP-deficient macrophages make more IL-12 and IL-6 (unpublished data; <italic>p</italic> < 0.05). Protein-matched cell lysates from SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> BMMs were analyzed by Western blotting with SHIP antibody to confirm the genotype of these cells (<xref ref-type="fig" rid="ppat-0020071-g001">Figure 1</xref>D). These data indicate that SHIP is a negative regulator of IL-12, IL-6, and RANTES secretion by <italic>F. novicida–</italic>infected macrophages.</p><p>The regulatory influence of SHIP on <named-content content-type="genus-species">F. novicida</named-content>–induced IL-12, IL-6, and RANTES was also observed at different multiplicities of infection (MOIs; <xref ref-type="fig" rid="ppat-0020071-g002">Figure 2</xref>A–<xref ref-type="fig" rid="ppat-0020071-g002">2</xref>C), suggesting that the negative regulatory effect of SHIP is independent of bacterial numbers infecting the macrophages. In addition, the viability of the bacteria is not essential for the negative influence of SHIP as macrophages infected with heat-killed bacteria displayed similar cytokine responses as those infected with live bacteria (<xref ref-type="fig" rid="ppat-0020071-g002">Figure 2</xref>D–<xref ref-type="fig" rid="ppat-0020071-g002">2</xref>F). Likewise, SHIP also down-regulated <named-content content-type="genus-species">F. tularensis</named-content> LVS–induced IL-12, IL-6, and RANTES production (<xref ref-type="fig" rid="ppat-0020071-g002">Figure 2</xref>D–<xref ref-type="fig" rid="ppat-0020071-g002">2</xref>F).</p><fig id="ppat-0020071-g002" position="float"><label>Figure 2</label><caption><title>Analysis of SHIP Influence on Macrophage Inflammatory Response to <named-content content-type="genus-species">F. novicida</named-content> Infection</title><p>(A–C) BMMs from SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> littermate mice were infected at different MOI (indicated in figure) of <named-content content-type="genus-species">F. novicida</named-content> for 8 h. Cytokine levels were measured in uninfected samples (0 h). Cell supernatants from uninfected and infected cells were assayed for IL-12, IL-6, and RANTES by ELISA.</p><p>(D–F) BMMs from SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> littermate mice were infected with 100 MOI of live or heat-killed <named-content content-type="genus-species">F. novicida</named-content> (FN and HKFN, respectively) or with <named-content content-type="genus-species">F. tularensis</named-content> LVS (LVS) for 8 h. Cyt D represents samples that were treated with cytochalasin D (5 μg/ml) before infection with 100 MOI of live <italic>F. novicida.</italic> Cytokine levels were measured in uninfected samples (0 h). Cell supernatants from uninfected and infected cells were analyzed by ELISA for IL-12, IL-6, and RANTES.</p><p>(G) BMMs from SHIP <sup>+/+</sup> and SHIP<sup>−/−</sup> littermate mice were infected with 100 MOI of <named-content content-type="genus-species">F. novicida</named-content> for 2 h, then treated with gentamicin (50 μg/ml), lysed in 0.1% SDS and appropriate dilutions of the lysates were plated on Chocolate II agar plates for enumeration of CFUs. All graphs represents mean and SD of values obtained from three independent experiments. Data were analyzed by paired Student <italic>t</italic> test. An asterisk () indicates <italic>p</italic>-value < 0.05.</p></caption><graphic xlink:href="ppat.0020071.g002"/></fig><p>We next examined whether internalization of bacteria is a requisite for the down-regulatory influence of SHIP on <named-content content-type="genus-species">F. novicida</named-content>–induced cytokine/chemokine production. For this, cells were treated with cytochalasin D, an actin polymerization inhibitor, prior to infection, and cell supernatants were assayed for the secretion of IL-12, IL-6, and RANTES by ELISA. The results are shown in <xref ref-type="fig" rid="ppat-0020071-g002">Figure 2</xref>D–<xref ref-type="fig" rid="ppat-0020071-g002">2</xref>F. The release of RANTES and the negative influence of SHIP on RANTES production were not influenced by failure of internalization. However, treatment of either SHIP<sup>+/+</sup> or SHIP<sup>−/−</sup> BMM with cytochalasin D significantly suppressed the release of IL-12 and IL-6, suggesting that either internalization of bacteria or actin cytoskeletal rearrangements are essential for the secretion of these two cytokines. Failure of internalization of <named-content content-type="genus-species">F. novicida</named-content> did not influence the activation of either the MAPK or the PI3K/Akt pathways (unpublished data), suggesting that surface contact of <italic>Francisella</italic> bacteria is sufficient to trigger the signaling response.</p><p>We then examined whether the increased production of cytokine mediators by SHIP<sup>−/−</sup> macrophages is due to enhanced uptake of <named-content content-type="genus-species">F. novicida</named-content>. The uptake of bacteria was determined by CFU assays, and the results are shown in <xref ref-type="fig" rid="ppat-0020071-g002">Figure 2</xref>G. The uptake of <named-content content-type="genus-species">F. novicida</named-content> by SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> BMM was equivalent. These findings are also supported by transmission electron microscopy (TEM) analysis (SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> macrophages ingested 2.95 and 3.2 bacteria/cell, respectively).</p></sec><sec id="s2b"><title>PI3K/Akt Pathway Is Activated upon <named-content content-type="genus-species">F. novicida</named-content> Infection of Macrophages</title><p>To examine the mechanism by which SHIP down-regulates pro-inflammatory cytokine production by <italic>F. novicida–</italic>infected macrophages, cell signaling events activated in infected macrophages were first analyzed. Here, murine BMMs were infected with <named-content content-type="genus-species">F. novicida</named-content> for varying time points, and the following analyses were performed with the cell lysates and cell supernatants. First, activation of the MAP kinases Erk, p38, and JNK was analyzed by Western blotting protein-matched lysates with phospho-specific antibodies (<xref ref-type="fig" rid="ppat-0020071-g003">Figure 3</xref>A). All membranes were reprobed with antibody to actin to ensure equal loading of protein in all lanes. Results indicated that all three MAPKs were activated in response to <named-content content-type="genus-species">F. novicida</named-content> infection as previously reported [<xref rid="ppat-0020071-b013" ref-type="bibr">13</xref>]. The phosphorylation levels of MAPKs peaked at 30 min and gradually diminished, but were still persistent even at 8 h post infection. Second, we examined the activation of the PI3K pathway by analyzing phosphorylation status of the PI3K-dependent serine/threonine kinase Akt. The results shown in <xref ref-type="fig" rid="ppat-0020071-g003">Figure 3</xref>B indicate that the PI3K/Akt pathway is also activated during <named-content content-type="genus-species">F. novicida</named-content> infection. Finally, cell supernatants were analyzed for pro-inflammatory cytokines by ELISA. As seen in <xref ref-type="fig" rid="ppat-0020071-g003">Figure 3</xref>C, BMMs infected with <named-content content-type="genus-species">F. novicida</named-content> produced significant amounts of the pro-inflammatory mediators IL-12, IL-6, and RANTES, which became detectable at about 5 h post infection.</p><fig id="ppat-0020071-g003" position="float"><label>Figure 3</label><caption><title>PI3K/Akt Pathway Is Activated upon <named-content content-type="genus-species">F. novicida</named-content> Infection</title><p>BMMs were infected with <named-content content-type="genus-species">F. novicida</named-content> for the times shown in the figure.</p><p>(A) Protein-matched lysates were analyzed by Western blotting with phospho-specific antibodies to Erk (Anti-pErk), p38 (Anti-pp38), and JNK (Anti-pJNK). The membranes were reprobed with actin antibody (Anti-Actin) to ensure equal loading of protein in all lanes.</p><p>(B) Protein-matched lysates were analyzed by Western blotting with phospho-specific antibody to Akt (Anti-pSer Akt; upper panel). The lower panel is a reprobe of the same membrane with total Akt antibody (Anti-Akt).</p><p>(C) Cell supernatants from the same experiments were assayed by ELISA for IL-12, IL-6, and RANTES. The graph represents mean and SD of values obtained from three independent experiments.</p></caption><graphic xlink:href="ppat.0020071.g003"/></fig></sec><sec id="s2c"><title>SHIP Regulates <named-content content-type="genus-species">F. novicida</named-content>–Induced Activation of the PI3K/Akt Pathway</title><p>Having identified cellular signaling events activated by <named-content content-type="genus-species">F. novicida</named-content> infection, we next examined the influence of SHIP on these events. For this, BMMs from SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> littermate mice were infected with <named-content content-type="genus-species">F. novicida</named-content> for the times indicated in <xref ref-type="fig" rid="ppat-0020071-g004">Figure 4</xref>, and activation of MAPKs and Akt was analyzed. Infection by <named-content content-type="genus-species">F. novicida</named-content> induced robust phosphorylation of Erk1/2, JNK, p38, and Akt. However, SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> BMMs displayed no significant differences in the activation of the MAPKs at any of the time points tested. On the other hand, SHIP<sup>−/−</sup> BMMs showed enhanced phosphorylation of Akt compared to SHIP<sup>+/+</sup> BMMs. These results indicate that although SHIP may not regulate <named-content content-type="genus-species">F. novicida</named-content>–induced activation of the MAPKs, SHIP down-regulates activation of Akt. Together, these data suggest that the PI3K/Akt pathway may play a critical role in <named-content content-type="genus-species">F. novicida</named-content>–induced macrophage pro-inflammatory responses.</p><fig id="ppat-0020071-g004" position="float"><label>Figure 4</label><caption><title>SHIP Negatively Regulates <named-content content-type="genus-species">F. novicida</named-content>–Induced Akt Activation</title><p>BMMs from SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> littermate mice were infected for the times indicated in the figure.</p><p>(A) Protein-matched lysates were probed with antibodies specific for phosphorylated Erk (Anti-pErk), p38 (Anti-pp38), and JNK (Anti-pJNK). All membranes were reprobed with actin antibody (Anti-Actin).</p><p>(B) Protein-matched lysates were probed with phospho-specific antibody for Akt (Anti-pSer Akt; upper panel). The membranes were reprobed with total Akt antibody (Anti-Akt; middle panel), and with SHIP antibody (Anti-SHIP; lower panel). These data are representative of three independent experiments.</p></caption><graphic xlink:href="ppat.0020071.g004"/></fig></sec><sec id="s2d"><title>Influence of PI3K Pathway on <named-content content-type="genus-species">F. novicida</named-content>–Induced Macrophage Inflammatory Response</title><p>A role for the PI3K/Akt pathway in macrophage response to <named-content content-type="genus-species">F. novicida</named-content> infection has thus far not been described. The above observations of enhanced cytokine production and enhanced activation of Akt in SHIP<sup>−/−</sup> BMMs suggest a potential role for the PI3K/Akt pathway in <named-content content-type="genus-species">F. novicida</named-content>–induced inflammatory response. To test this notion, we next treated <named-content content-type="genus-species">F. novicida</named-content> infected cells with the pharmacologic inhibitor of PI3K LY294002, and monitored the release of IL-12, IL-6, and RANTES by ELISA. The results are displayed in <xref ref-type="fig" rid="ppat-0020071-g005">Figure 5</xref>. Inhibition of the PI3K pathway significantly decreased the production of IL-12 (<italic>p</italic> < 0.009), IL-6 (<italic>p</italic> < 0.003), and RANTES (<italic>p</italic> < 0.03) by <named-content content-type="genus-species">F. novicida</named-content>–infected BMMs, indicating that the PI3K pathway is involved in the production of pro-inflammatory mediators in response to <named-content content-type="genus-species">F. novicida</named-content> infection.</p><fig id="ppat-0020071-g005" position="float"><label>Figure 5</label><caption><title>PI3K Activation Promotes <named-content content-type="genus-species">F. novicida</named-content>–Induced Macrophage Inflammatory Response</title><p>BMMs were pretreated for 30 min with PI3K inhibitor LY294002 (LY; 20 μM), or with vehicle control (DMSO), and subsequently infected with <named-content content-type="genus-species">F. novicida</named-content>. Cell supernatants were harvested 8 h post infection and assayed by ELISA for IL-12, IL-6, and RANTES. Cytokine levels were measured in uninfected samples (0 h). The graphs represent mean and SD of values obtained from three independent experiments. Data were analyzed by paired Student <italic>t</italic> test. An asterisk () indicates <italic>p</italic>-value < 0.05.</p></caption><graphic xlink:href="ppat.0020071.g005"/></fig></sec><sec id="s2e"><title>The PI3K/Akt Pathway Promotes Macrophage Inflammatory Response to <named-content content-type="genus-species">F. novicida</named-content> through Its Influence on Downstream NFκB Activation</title><p>We next examined the molecular mechanism by which PI3K/Akt influences <named-content content-type="genus-species">F. novicida</named-content>–induced cytokine response. It has been previously reported that <named-content content-type="genus-species">F. novicida</named-content> induces the activation of NFκB, and that activation of this transcription factor may play an important role in the secretion of pro-inflammatory cytokines. Hence we wanted to test the involvement of NFκB in the secretion of IL-12, IL-6, and RANTES in response to <named-content content-type="genus-species">F. novicida</named-content> infection, and the influence of PI3K on NFκB activation. For this, we first transiently transfected RAW 264.7 cells with a construct encoding luciferase reporter gene under the influence of NFκB binding (NFκB-luciferase construct). Twelve hours post transfection, cells were infected with <named-content content-type="genus-species">F. novicida</named-content> for the times indicated in <xref ref-type="fig" rid="ppat-0020071-g006">Figure 6</xref>A, and luciferase activity in the cell lysate was determined. Results indicated that NFκB activation is induced in response to <named-content content-type="genus-species">F. novicida</named-content> infection.</p><fig id="ppat-0020071-g006" position="float"><label>Figure 6</label><caption><title>PI3K Promotes <named-content content-type="genus-species">F. novicida</named-content>–Induced Inflammatory Response through Its Influence on NFκB</title><p>(A) RAW 264.7 cells transfected with plasmid encoding the <italic>luciferase</italic> gene driven by an NFκB binding element (NFκB-luc). Transfectants were infected with <named-content content-type="genus-species">F. novicida</named-content> for varying times and analyzed for the luciferase activity as a measure of NFκB activation. RLUs, relative light units.</p><p>(B) RAW 264.7 cells transfected with NFκB-luc plasmid were pretreated with either a PI3K inhibitor LY294002 (LY), a NFκB inhibitor SN50 (SN; 75 μg/ml), or with vehicle control (DMSO), and subsequently infected with <named-content content-type="genus-species">F. novicida</named-content>. At 5 h post infection, cells were lysed and assayed for luciferase activity. RLUs, relative light units.</p><p>(C) BMMs were pretreated for 30 min with either DMSO (middle bar) or the NFκB inhibitor SN50 (SN), and subsequently infected with <named-content content-type="genus-species">F. novicida</named-content>. Cell supernatants were harvested 8 h post infection and assayed for IL-12, IL-6, and RANTES by ELISA. Cytokine levels were measured in uninfected samples (0 h). The graph represents mean and SD of values obtained from three experiments. Data were analyzed by paired Student <italic>t</italic> test. An asterisk () indicates <italic>p</italic>-value < 0.05.</p><p>(D) Protein-matched lysates from BMMs pretreated with either DMSO or with inhibitors of PI3K (LY) or NFκB (SN) and infected with <named-content content-type="genus-species">F. novicida</named-content> were analyzed by Western blotting with phospho-specific antibody to Akt (Anti-pSer-Akt; upper panel). The lower panel is a reprobe of the same membrane with Akt antibody (Anti-Akt).</p><p>(E) RAW 264.7 cells transfected with NFκB-luc plasmid were pretreated with either cytochalasin D (Cyt D), SN50 (SN), or with vehicle control (DMSO), and subsequently infected with <named-content content-type="genus-species">F. novicida</named-content>. At 5 h post infection, cells were lysed and assayed for luciferase activity. The data are representative of three independent experiments.</p></caption><graphic xlink:href="ppat.0020071.g006"/></fig><p>Second, to test whether PI3K influences <named-content content-type="genus-species">F. novicida</named-content>–induced activation of NFκB, transfected cells were treated with LY294002, or an inhibitor of NFκB SN50, and expression of luciferase enzyme in response to <named-content content-type="genus-species">F. novicida</named-content> infection was measured. The results are shown in <xref ref-type="fig" rid="ppat-0020071-g006">Figure 6</xref>B. Inhibition of PI3K with LY294002 significantly decreased expression of the NFκB-driven reporter gene in response to <named-content content-type="genus-species">F. novicida</named-content> infection (<italic>p</italic> < 0.0001), suggesting that the activation of PI3K pathway is necessary for induction of NFκB activation.</p><p>Finally, inhibition of NFκB, using the pharmacologic inhibitor SN50, in murine BMMs infected with <named-content content-type="genus-species">F. novicida</named-content> resulted in significant inhibition of IL-12 (<italic>p</italic> < 0.009), IL-6 (<italic>p</italic> < 0.007), and RANTES (<italic>p</italic> < 0.02) production (<xref ref-type="fig" rid="ppat-0020071-g006">Figure 6</xref>C). The Western blots shown in <xref ref-type="fig" rid="ppat-0020071-g006">Figure 6</xref>D demonstrate that although treatment of cells with the PI3K inhibitor completely abrogated downstream Akt phosphorylation, Akt phosphorylation was not influenced by the NFκB inhibitor SN50, indicating that SN50 inhibition has occurred downstream of Akt. Collectively, these data provide evidence that PI3K influences <named-content content-type="genus-species">F. novicida</named-content>–induced macrophage pro-inflammatory response through its influence on NFκB activation. Of note, the activation of NFκB is unaffected in macrophages pre-treated with cytochalasin D, suggesting that internalization of <named-content content-type="genus-species">F. novicida</named-content> is not required (<xref ref-type="fig" rid="ppat-0020071-g006">Figure 6</xref>E).</p></sec><sec id="s2f"><title>SHIP Negatively Regulates <named-content content-type="genus-species">F. novicida</named-content>–Induced NFκB Activation</title><p>Requirement of PI3K activity for <named-content content-type="genus-species">F. novicida</named-content>–induced NFκB activation suggested that SHIP may regulate <named-content content-type="genus-species">F. novicida</named-content>–stimulated NFκB activation. To test this prediction, RAW 264.7 cells were transfected with NFκB–luciferase construct alone along with either a construct encoding SHIP or the corresponding empty vector. Transfectants were then infected with <italic>F. novicida,</italic> and luciferase activity in the cell lysates was measured as an indicator of NFκB activation. As expected, overexpression of SHIP significantly (<italic>p</italic> < 0.04) suppressed NFκB-dependent reporter gene expression (<xref ref-type="fig" rid="ppat-0020071-g007">Figure 7</xref>A). Overexpression of SHIP in the transfected cells was confirmed by Western blotting SHIP immunoprecipitates with SHIP antibody (<xref ref-type="fig" rid="ppat-0020071-g007">Figure 7</xref>B).</p><fig id="ppat-0020071-g007" position="float"><label>Figure 7</label><caption><title>SHIP Negatively Regulates <named-content content-type="genus-species">F. novicida</named-content>–Induced NFκB Activation</title><p>(A) RAW 264.7 cells were transfected with NFκB-luc plasmid along with either empty vector or plasmid encoding SHIP. Transfectants were infected with <named-content content-type="genus-species">F. novicida</named-content> and assayed for luciferase activity 5 h post infection. Data were analyzed by Student <italic>t</italic> test. An asterisk () indicates <italic>p</italic>-value < 0.05.</p><p>(B) SHIP immunoprecipitates from the RAW 264.7 transfectants described in <xref ref-type="fig" rid="ppat-0020071-g007">Figure 7</xref>A were analyzed by Western blotting for overexpression of SHIP. Ctrl indicates immunoprecipitation with normal rabbit serum.</p><p>(C) SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> BMMs were transfected with NFκB-luc plasmid. Transfectants were infected with <named-content content-type="genus-species">F. novicida</named-content> and assayed for luciferase activity. Graphs represent values from three independent experiments. Data were analyzed by Student <italic>t</italic> test. An asterisk () indicates <italic>p</italic>-value < 0.05.</p></caption><graphic xlink:href="ppat.0020071.g007"/></fig><p>As an additional approach, murine BMMs obtained from SHIP <sup>+/+</sup> and SHIP <sup>−/−</sup> mice were transiently transfected with the NFκB–luciferase construct. Transfection of BMMs using the amaxa Nucleofector (Solution T, program T-20; amaxa biosystems, Cologne, Germany) yielded comparable transfection efficiency for both SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> cells, as determined with plasmid encoding EGFP (enhanced green fluorescent protein; unpublished data). Transfected BMMs were subsequently infected with <named-content content-type="genus-species">F. novicida</named-content> and assayed for luciferase activity. BMMs lacking SHIP displayed significantly higher levels (<italic>p</italic> < 0.05) of luciferase activity than the SHIP<sup>+/+</sup> BMMs, indicating that SHIP negatively regulates <named-content content-type="genus-species">F. novicida</named-content>–induced activation of NFκB (<xref ref-type="fig" rid="ppat-0020071-g007">Figure 7</xref>C).</p></sec><sec id="s2g"><title>SHIP Positively Regulates <named-content content-type="genus-species">F. novicida</named-content>–Induced IL-10 Production</title><p>To determine whether SHIP regulates the production of anti-inflammatory cytokines by infected macrophages, we next measured IL-10 production. Here SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> BMMs were infected with <named-content content-type="genus-species">F. novicida</named-content> for 8 h. Cell supernatants from uninfected and infected cells were analyzed by ELISA for IL-10. The results shown in <xref ref-type="fig" rid="ppat-0020071-g008">Figure 8</xref> demonstrate that SHIP<sup>+/+</sup> BMMs produce significantly higher levels of IL-10 than the SHIP<sup>−/−</sup> cells.</p><fig id="ppat-0020071-g008" position="float"><label>Figure 8</label><caption><title>SHIP Influence on Macrophage IL-10 Response to <named-content content-type="genus-species">F. novicida</named-content> </title><p>BMMs from SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> littermate mice were infected for 8 h with <italic>F. novicida.</italic> Cell supernatants from uninfected and infected cells were analyzed by ELISA for IL-10. The graph represents mean and SD of values obtained from three independent experiments. Data were analyzed by paired Student <italic>t</italic> test. An asterisk (*) indicates <italic>p</italic>-value < 0.05.</p></caption><graphic xlink:href="ppat.0020071.g008"/></fig></sec></sec><sec id="s3"><title>Discussion</title><p>The molecular details of host response to <named-content content-type="genus-species">F. tularensis</named-content> infection are not clearly understood. Both innate and adaptive immune responses provide protection against this infection. Early protection against <named-content content-type="genus-species">F. tularensis</named-content> is dependent upon the production of IFNγ, TNF-α, and IL-12, all of which are produced within a day after infection [<xref rid="ppat-0020071-b007" ref-type="bibr">7</xref>,<xref rid="ppat-0020071-b009" ref-type="bibr">9</xref>,<xref rid="ppat-0020071-b031" ref-type="bibr">31</xref>,<xref rid="ppat-0020071-b032" ref-type="bibr">32</xref>]. To date, regulatory mechanisms controlling the production of pro-inflammatory molecules have not been established. SHIP is a critical regulator of hematopoietic cell functions; hence we have investigated the role of SHIP in <named-content content-type="genus-species">F. novicida</named-content>–induced inflammatory response. Our data indicate that SHIP is a negative regulator of <italic>Francisella</italic>-induced IL-12, IL-6, and RANTES production by BMMs.</p><p>SHIP has been shown to influence hematopoietic cell functions through its inhibitory effect on the PI3K pathway as well as the MAPK pathways [<xref rid="ppat-0020071-b015" ref-type="bibr">15</xref>]. Since previous studies and our current studies have demonstrated the activation of MAPKs in <italic>Francisella</italic>-infected macrophages, we examined whether SHIP influenced activation of the MAPKs. However, our studies indicate that the molecular mechanism by which SHIP influences macrophage response to <italic>Francisella</italic> does not involve inhibition of the MAPKs. Thus, we did not observe any significant differences in the activation levels of MAPKs at any of the time points tested, ranging from 5 min to 8 h (<xref ref-type="fig" rid="ppat-0020071-g004">Figure 4</xref>A and unpublished data). However, we observed significantly elevated Akt activation in SHIP-deficient macrophages infected with <italic>Francisella</italic>. The hyperactivation of Akt in SHIP-deficient macrophages suggested that the PI3K pathway may play a role in macrophage response to <italic>Francisella</italic>. Consistently, inhibition of PI3K attenuated <named-content content-type="genus-species">F. novicida</named-content>–induced activation of NFκB and the subsequent production of pro-inflammatory cytokines.</p><p>Activation of the PI3K/Akt pathway by <italic>Francisella</italic> has not been previously reported. However, other intracellular pathogens such as <named-content content-type="genus-species">Salmonella enterica</named-content> have been shown to modulate host cell phosphoinositide pathways and downstream Akt activation [<xref rid="ppat-0020071-b033" ref-type="bibr">33</xref>]. In the latter case, the functional consequence of phosphoinositide signaling is unclear. Interestingly, pharmacologic inhibition of PI3K failed to prevent invasion of <italic>Salmonella,</italic> suggesting that PI3K may not play a role in internalization of some intracellular pathogens, which use a type III secretion-mediated mechanism to gain entry into the host cell [<xref rid="ppat-0020071-b033" ref-type="bibr">33</xref>,<xref rid="ppat-0020071-b034" ref-type="bibr">34</xref>]. This is in contrast to receptor-mediated phagocytosis in which PI3K activation is critical.</p><p>Macrophage receptors that sense <italic>Francisella</italic> and mediate phagocytosis are just being defined [<xref rid="ppat-0020071-b035" ref-type="bibr">35</xref>], but the linkages between receptors and intracellular signaling pathways are essentially unknown. Earlier studies suggested that, since there is a lack of macrophage inflammatory response to the LPS of <italic>Francisella,</italic> it is unlikely that TLR4 is involved [<xref rid="ppat-0020071-b036" ref-type="bibr">36</xref>–<xref rid="ppat-0020071-b038" ref-type="bibr">38</xref>]. Other studies propose that TLR2 may be important [<xref rid="ppat-0020071-b039" ref-type="bibr">39</xref>]. Recent work also suggests a role for intracellular recognition of the pathogen [<xref rid="ppat-0020071-b039" ref-type="bibr">39</xref>]. The role of SHIP in pro-inflammatory cytokine response elicited by TLR4 or TLR2 engagement has been studied. Thus, studies by Strassheim et al. demonstrated that SHIP negatively regulates TLR2 signaling [<xref rid="ppat-0020071-b030" ref-type="bibr">30</xref>]. In contrast, the presence of SHIP appears to enhance TLR4-induced inflammatory response. In an earlier study, we have demonstrated that SHIP-deficient macrophages are hyporesponsive to TLR4 engagement compared to their wild-type counterparts [<xref rid="ppat-0020071-b016" ref-type="bibr">16</xref>]. These findings are supported by recent reports by Rauh et al. [<xref rid="ppat-0020071-b027" ref-type="bibr">27</xref>,<xref rid="ppat-0020071-b029" ref-type="bibr">29</xref>]. Although the latter group earlier reported a negative regulatory role for SHIP [<xref rid="ppat-0020071-b040" ref-type="bibr">40</xref>], these findings were later attributed by the same group to in vitro culture conditions of macrophages in their study [<xref rid="ppat-0020071-b029" ref-type="bibr">29</xref>]. Our current findings that SHIP negatively regulates <italic>Francisella</italic>-induced inflammatory response are consistent with a role for TLR2 in <italic>Francisella</italic>-induced signaling. Additional studies are required to test this notion.</p><p>SHIP is a cytosolic enzyme that must undergo membrane translocation to access its lipid substrates. Previous reports indicate that the N-terminal SH2 domain of SHIP is critical for this translocation under certain stimulation conditions, whereas the C-terminal region may be more important under other conditions [<xref rid="ppat-0020071-b015" ref-type="bibr">15</xref>]. Indeed, the C-terminal proline-rich region of SHIP has been shown to be required for stabilization of SHIP at the membrane [<xref rid="ppat-0020071-b041" ref-type="bibr">41</xref>,<xref rid="ppat-0020071-b042" ref-type="bibr">42</xref>]. Additional studies are required to understand the mechanism by which SHIP translocates to the membrane in response to <italic>Francisella</italic> infection. A thorough understanding of the role of SHIP in <italic>Francisella-</italic>induced signaling events may shed light on regulatory mechanisms controlling the production of inflammatory mediators that are essential for protection against <italic>Francisella</italic> infection.</p><p>The production of pro-inflammatory cytokines such as IL-12 by monocytes/macrophages contributes further to immunity against <italic>Francisella</italic> infection by augmenting NK cell IFNγ production. In a recent report, we have demonstrated that production of IFNγ by NK cells in response to stimulation by monokines (IL-12, IL-15, and IL-18) is augmented in the absence of SHIP [<xref rid="ppat-0020071-b043" ref-type="bibr">43</xref>,<xref rid="ppat-0020071-b044" ref-type="bibr">44</xref>]. Thus, it is conceivable that SHIP not only regulates macrophage responses to <italic>Francisella</italic> infection, but also the subsequent NK cell responses, thereby regulating overall defense against intracellular pathogens.</p><p>In conclusion, this study unravels novel roles for PI3K and SHIP in regulating intracellular signaling events involved in macrophage innate immune response to <italic>Francisella</italic> infection.</p></sec><sec id="s4"><title>Materials and Methods</title><sec id="s4a"><title>Cells, antibodies, and reagents.</title><p>RAW 264.7 murine macrophage cells were obtained from ATCC (Washington, D. C., United States) and maintained in RPMI with 3.5% heat-inactivated fetal bovine serum (FBS). Antibodies specific for phospho-Erk, phospho-JNK, phospho-SHIP, phospho-Akt, and phospho-p38 were purchased from Cell Signaling Technology (Beverly, Massachusetts, United States). Actin, phosphotyrosine, and Akt antibodies were from Santa Cruz Biotechnology (Santa Cruz, California, United States). Rabbit polyclonal SHIP antibody was a generous gift from Dr. K. Mark Coggeshall (Oklahoma Medical Research Foundation, Oklahoma City, Oklahoma, United States). <named-content content-type="genus-species">F. novicida</named-content> U112 (JSG1819) was used in all experiments. Bacteria were grown on Chocolate II agar plates at 37 °C.</p></sec><sec id="s4b"><title>Culture of murine bone marrow macrophages.</title><p>SHIP<sup>+/−</sup> animals were generously provided by Dr. G. Krystal (BC Cancer Agency, Vancouver, British Columbia, Canada). Heterozygotes were bred to obtain SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> mice. BMMs were derived from these animals as previously described [<xref rid="ppat-0020071-b016" ref-type="bibr">16</xref>]. Briefly, bone marrow cells were cultured in RPMI containing 10% fetal bovine serum plus 10 μg/ml polymixin B and supplemented with 20 ng/ml CSF-1 for 7 d before they were used in experiments. BMMs derived in this manner were more than 99% positive for Mac-1, as determined by flow cytometry.</p></sec><sec id="s4c"><title>Cell stimulation, lysis, and Western blotting.</title><p>BMMs were plated in six-well culture dishes. Macrophages were infected either with <named-content content-type="genus-species">F. novicida</named-content> (at 1. 10, or 100 MOI) or with <named-content content-type="genus-species">F. tularensis</named-content> LVS (100 MOI) that were scraped from Chocolate II agar plates, resuspended, and diluted in RPMI. For some experiments, bacteria were killed by heat at 98 °C for 10 min prior to adding to the macrophages. To prevent internalization of bacteria, cells were treated with 5 μg/ml of cytochalasin D for 30 min at 37 °C and 5% C0<sub>2</sub> prior to infection. Uninfected and infected cells were lysed in TN1 buffer (50 mM Tris [pH 8.0], 10 mM EDTA, 10m M Na<sub>4</sub>P<sub>2</sub>O<sub>7</sub>, 10 mM NaF, 1% Triton-X 100, 125 mM NaCl, 10 mM Na<sub>3</sub>VO<sub>4</sub>, 10 μg/ml each aprotinin and leupeptin). Postnuclear lysates were boiled in Laemmli Sample Buffer and were separated by SDS/PAGE, transferred to nitrocellulose filters, probed with the antibody of interest, and developed by enhanced chemiluminescence (ECL).</p></sec><sec id="s4d"><title>CFU assays.</title><p>SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> BMMs were infected with <named-content content-type="genus-species">F. novicida</named-content> (100 MOI). Two hours post infection, cells were washed two times and incubated with 50 μg/ml of gentamicin for 30 min at 37 °C and 5% CO<sub>2</sub>. The cells were subsequently washed twice and lysed in 0.1% SDS for 5 min. Immediately, 10-fold serial dilutions were made, and appropriate dilutions were plated on Chocolate II agar plates. Assays were performed in triplicate for each test group.</p><p>Transmission electron microscopy (TEM): SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> BMMs were infected with <italic>F. novicida,</italic> and 2 h post infection, the cells were washed, fixed, and prepared for TEM analysis as described previously. Bacterial count in 20 SHIP<sup>+/+</sup> and SHIP<sup>−/−</sup> BMMs was assessed and averaged [<xref rid="ppat-0020071-b045" ref-type="bibr">45</xref>].</p></sec><sec id="s4e"><title>ELISA measurement of cytokine production.</title><p>BMMs were infected with <named-content content-type="genus-species">F. novicida</named-content> for varying time points. Cell supernatants were harvested, centrifuged to remove dead cells, and analyzed by ELISA using cytokine-specific kits from R & D Systems (Minneapolis, Minnesota, United States). Data were analyzed using a paired Student <italic>t</italic>-test. A <italic>p</italic>-value <underline><</underline> 0.05 was considered as significant.</p></sec><sec id="s4f"><title>Transfection and luciferase assays.</title><p>BMMs and RAW 264.7 cells were transfected with the appropriate plasmid DNA using the Amaxa Nucleofector apparatus (amaxa biosystems) as previously described [<xref rid="ppat-0020071-b046" ref-type="bibr">46</xref>,<xref rid="ppat-0020071-b047" ref-type="bibr">47</xref>]. Briefly, 5 × 10<sup>6</sup> cells were resuspended in 100 μl Nucelofector Solution (T for BMMs and V for RAW 264.7 cells), and were nucelofected with 1 μg of NFκB-luc alone or with 5 μg of empty vector or plasmid encoding WT-SHIP. Immediately post nucleofection, 500 μl of pre-warmed RPMI was added to the transfection mix before transferring to 12-well plates containing 1.5-ml pre-warmed RPMI per well. Plates were incubated for 12 h at 37 °C. Transfected cells were either left uninfected or were infected for 5 h with <named-content content-type="genus-species">F. novicida</named-content>. Cells were lysed in 100 μl of Luciferase Cell Culture Lysis Reagent (Promega, Madison, Wisconsin, United States). Luciferase activity was then measured using Luciferase Assay Reagent (Promega), as previously described [<xref rid="ppat-0020071-b024" ref-type="bibr">24</xref>].</p></sec></sec>
Transcriptional Profiling of Aging in Human Muscle Reveals a Common Aging Signature	<p>We analyzed expression of 81 normal muscle samples from humans of varying ages, and have identified a molecular profile for aging consisting of 250 age-regulated genes. This molecular profile correlates not only with chronological age but also with a measure of physiological age. We compared the transcriptional profile of muscle aging to previous transcriptional profiles of aging in the kidney and the brain, and found a common signature for aging in these diverse human tissues. The common aging signature consists of six genetic pathways; four pathways increase expression with age (genes in the extracellular matrix, genes involved in cell growth, genes encoding factors involved in complement activation, and genes encoding components of the cytosolic ribosome), while two pathways decrease expression with age (genes involved in chloride transport and genes encoding subunits of the mitochondrial electron transport chain). We also compared transcriptional profiles of aging in humans to those of the mouse and fly, and found that the electron transport chain pathway decreases expression with age in all three organisms, suggesting that this may be a public marker for aging across species.</p>	<contrib contrib-type="author"><name><surname>Zahn</surname><given-names>Jacob M</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Sonu</surname><given-names>Rebecca</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Vogel</surname><given-names>Hannes</given-names></name><xref ref-type="aff" rid="aff2">2</xref></contrib><contrib contrib-type="author"><name><surname>Crane</surname><given-names>Emily</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Mazan-Mamczarz</surname><given-names>Krystyna</given-names></name><xref ref-type="aff" rid="aff3">3</xref></contrib><contrib contrib-type="author"><name><surname>Rabkin</surname><given-names>Ralph</given-names></name><xref ref-type="aff" rid="aff4">4</xref><xref ref-type="aff" rid="aff5">5</xref></contrib><contrib contrib-type="author"><name><surname>Davis</surname><given-names>Ronald W</given-names></name><xref ref-type="aff" rid="aff6">6</xref><xref ref-type="aff" rid="aff7">7</xref></contrib><contrib contrib-type="author"><name><surname>Becker</surname><given-names>Kevin G</given-names></name><xref ref-type="aff" rid="aff8">8</xref></contrib><contrib contrib-type="author"><name><surname>Owen</surname><given-names>Art B</given-names></name><xref ref-type="aff" rid="aff9">9</xref></contrib><contrib contrib-type="author"><name><surname>Kim</surname><given-names>Stuart K</given-names></name><xref ref-type="aff" rid="aff1">1</xref><xref ref-type="aff" rid="aff6">6</xref><xref ref-type="corresp" rid="cor1">*</xref></contrib>	PLoS Genetics	<sec id="s1"><title>Introduction</title><p>Aging is marked by the gradual decline of a multitude of physiological functions leading to an increasing probability of death. Some aging-related changes affect one's appearance, such as wrinkled skin, whereas others affect organ function, such as decreased kidney filtration rate and decreased muscular strength. At the molecular level, we are just beginning to assemble protein and gene expression changes that can be used as markers for aging. Rather than search for molecular aging markers by focusing on only one gene or pathway at a time, an attractive approach is to screen all genetic pathways in parallel for age-related changes by using full-genome oligonucleotide chips to search for gene expression changes in the elderly. A genome-wide transcriptional profile of aging may identify molecular markers of the aging process, and would provide insight into the molecular mechanisms that ultimately limit human lifespan.</p><p>Molecular markers of aging must reflect physiological function rather than simple chronological age because individuals age at different rates [<xref rid="pgen-0020115-b001" ref-type="bibr">1</xref>]. In the mouse, changes in the levels of CD4 immunocytes and changes in the expression of cell-cycle genes such as <italic>p16<sup>INK4a</sup></italic> are molecular markers of aging, as they predict both the remaining lifespan and the physiological age of the mouse [<xref rid="pgen-0020115-b002" ref-type="bibr">2</xref>–<xref rid="pgen-0020115-b004" ref-type="bibr">4</xref>]. In the human, gene expression profiling experiments identified 447 age-regulated genes that could predict the physiological age of the kidney [<xref rid="pgen-0020115-b005" ref-type="bibr">5</xref>]. Whole-genome expression profiling has also been used to identify genes that change expression with chronological age in the brain [<xref rid="pgen-0020115-b006" ref-type="bibr">6</xref>], skeletal muscle [<xref rid="pgen-0020115-b007" ref-type="bibr">7</xref>,<xref rid="pgen-0020115-b008" ref-type="bibr">8</xref>], and dermal fibroblasts [<xref rid="pgen-0020115-b009" ref-type="bibr">9</xref>], but changes in expression of these marker genes have not yet been shown to correlate with physiological aging.</p><p>In this paper, we have performed a genome wide analysis of gene expression changes in the human skeletal muscle. As age increases, skeletal muscle degenerates, loses mass, loses total aerobic capacity, and becomes markedly weaker [<xref rid="pgen-0020115-b010" ref-type="bibr">10</xref>]. One measure of muscle physiology is the ratio of the diameters of the type I and type II muscle fibers. A decrease in the size of type II muscle fibers (fast twitch) has been found to be correlated with decline in muscle function in both human [<xref rid="pgen-0020115-b011" ref-type="bibr">11</xref>] and rat [<xref rid="pgen-0020115-b012" ref-type="bibr">12</xref>]. Type II muscle fibers are known to atrophy and become smaller with age in the human, partially accounting for decreased muscle strength and flexibility in old age. As type II muscle fibers become smaller with age, the ratio of the diameters of type II fibers to type I fibers becomes smaller.</p><p>The extent to which age regulation of genetic pathways is specific to a particular tissue or common across many tissues is unknown. Age regulation of gene expression between the cortex and medulla regions of the human kidney was found to be highly correlated [<xref rid="pgen-0020115-b005" ref-type="bibr">5</xref>]. There was a high correlation in gene expression changes with age in different regions of the brain cortex, but no similarity was found between the cortex and the cerebellum [<xref rid="pgen-0020115-b013" ref-type="bibr">13</xref>]. Thus, there are similarities in patterns of age regulation between different areas of the kidney and between different areas of the brain cortex, but a common signature for aging across many diverse tissues has not been found.</p><p>Another key issue is whether there are genetic pathways that are commonly age regulated in different species with vastly different lifespans, such as human, mouse, fly, and worm. Transcriptional profiles of aging have been performed on both skeletal muscle and brains in the mouse [<xref rid="pgen-0020115-b014" ref-type="bibr">14</xref>,<xref rid="pgen-0020115-b015" ref-type="bibr">15</xref>], in <named-content content-type="genus-species">Drosophila melanogaster</named-content> [<xref rid="pgen-0020115-b016" ref-type="bibr">16</xref>,<xref rid="pgen-0020115-b017" ref-type="bibr">17</xref>], and in <named-content content-type="genus-species">Caenorhabditis elegans</named-content> [<xref rid="pgen-0020115-b018" ref-type="bibr">18</xref>]. A comparison of the patterns of gene expression changes during aging in the fly and the worm concluded that genes encoding mitochondrial components decreased expression with age in both species [<xref rid="pgen-0020115-b019" ref-type="bibr">19</xref>].</p><p>In this work, we present a transcriptional expression profile of 81 human skeletal muscle samples as a function of age. The symporter activity, sialyltransferase activity, and chloride transport pathways all decrease expression with age in human muscle. The age-regulated genes were found to be markers of physiological age, not just chronological age. By comparing our results on aging in muscle to previous transcriptional profiles of aging in the kidney and the brain, we found a common signature for aging across different human tissues consisting of six genetic pathways that showed common patterns of age regulation in all three tissues. Finally, by comparing the signature for aging in humans to transcriptional profiles of aging in mice, flies, and worms, we found that expression of the electron transport chain decreases with age in humans, mice, and flies, constituting a public signature for aging across species with extremely different lifespans.</p></sec><sec id="s2"><title>Results</title><sec id="s2a"><title>A Global Gene Expression Profile for Aging in Human Muscle</title><p>In order to study the effects of aging in human muscle, we obtained 81 samples of human skeletal muscle from individuals spanning 16 to 89 y of age (<xref ref-type="table" rid="pgen-0020115-t001">Table 1</xref>). Sixty-three samples were obtained from the abdomen, 5 were obtained from the arm, 2 were obtained from the deltoid muscle, 2 were obtained from the inner thigh, and 9 were obtained from the quadriceps (<xref ref-type="supplementary-material" rid="pgen-0020115-st001">Table S1</xref>). We used Affymetrix DNA arrays to generate a transcriptional profile of aging in human muscle. We isolated total RNA from each muscle sample, and synthesized biotinylated cRNA from total RNA. We then hybridized the cRNA to Affymetrix 133 2.0 Plus oligonucleotide arrays, representing nearly the entire human genome (54,675 individual probe sets corresponding to 31,948 individual human genes). We plotted the expression of each gene as a function of age, resulting in a dataset that shows the expression of nearly every gene in the genome as a function of age in human muscle (data are publicly available on the Gene Expression Omnibus at <ext-link ext-link-type="uri" xlink:href="http://www.ncbi.nlm.nih.gov/geo">http://www.ncbi.nlm.nih.gov/geo</ext-link>).</p><table-wrap id="pgen-0020115-t001" content-type="1col" position="float"><label>Table 1</label><caption><p>Patients Recruited by Age Group</p></caption><graphic xlink:href="pgen.0020115.t001"/></table-wrap><p>We used a multiple regression technique on each gene to determine how its expression changes with age, as had been done previously for age regulation in the kidney (Materials and Methods) [<xref rid="pgen-0020115-b005" ref-type="bibr">5</xref>]. We analyzed age regulation in skeletal muscle in two ways. In the first way, we found individual genes that met a stringent statistical significance threshold for correlation with age. In the second way, we found groups of genes (defined by the Gene Ontology consortium) in which there is subtle but consistent age regulation.</p><p>To identify individual genes showing strong age regulation, we examined the slope with respect to age for each gene, and identified 250 genes in which the slope was significantly positive or negative (<italic>p</italic> < 0.001) (<xref ref-type="fig" rid="pgen-0020115-g001">Figure 1</xref>, <xref ref-type="supplementary-material" rid="pgen-0020115-st002">Table S2</xref>, and Materials and Methods). At this statistical threshold, we would expect only 32 genes by chance, suggesting a false discovery rate of 13% or less. Furthermore, we randomly permuted the ages of the muscle samples, keeping the gene expression, sex, and anatomy variables fixed, and counted the number of genes that were significantly age regulated, again at <italic>p</italic> < 0.001. In 1,000 such permutations we found fewer than 107 significant genes 95% of the time. Thus, we are confident that most of the 250 age-regulated genes are not sampling artifacts. Of the 250 age-regulated genes, 125 genes increase expression, and 125 genes decrease expression with age.</p><fig id="pgen-0020115-g001" position="float"><label>Figure 1</label><caption><title>Expression of 250 Age-Regulated Genes in Muscle</title><p>Rows correspond to individual genes, arranged in order from greatest increase in expression with age at top to greatest decrease in expression with age at bottom. Columns represent individual patients, from youngest at left to oldest at right. Ages of certain individuals are marked for reference. Scale represents log<sub>2</sub> expression level (Exp). Genes discussed in the text are marked for reference. A navigable version of this figure showing identities of specific genes can be found at <ext-link ext-link-type="uri" xlink:href="http://cmgm.stanford.edu/~kimlab/aging_muscle">http://cmgm.stanford.edu/~kimlab/aging_muscle</ext-link>.</p></caption><graphic xlink:href="pgen.0020115.g001"/></fig><p>We considered the possibility that some of the 250 genes might not be age regulated per se, but rather might appear to be age regulated because they are associated with a pathological condition that increases with age. For example, the incidence of diabetes is known to increase with age in the general human population [<xref rid="pgen-0020115-b020" ref-type="bibr">20</xref>]. Our selection of patients might show a bias of diabetes in the elderly, in which case genes that change expression in response to diabetes might appear to be age regulated in our study. In addition to diabetes, we considered thirteen other factors that might also confound our study on aging, such as whether the patient was male or female, the anatomical origin of the muscle sample, the type of pathology associated with the patient, and types of medication taken by the patient (<xref ref-type="supplementary-material" rid="pgen-0020115-st001">Table S1</xref>).</p><p>With the exception of hypothyroidism, none of the medical factors showed a strong association with age, and so it is unlikely that these confounding factors would cause genes to appear to be age regulated (<xref ref-type="supplementary-material" rid="pgen-0020115-sg001">Figure S1</xref>). Hypothyroidism was absent in the young and present in about half of the elderly.</p><p>We used two methods to test whether any of the factors affected the slope of gene expression with respect to age of the 250 age-regulated genes. First, we used a multiple regression model that included a fourth term representing the medical factor (such as hypothyroidism) in addition to age, sex, and anatomy. We then compared the aging coefficient using this new model with the one from the original model that did not include the term. If any of the 250 genes were regulated by the medical factor and not by age per se, we would expect marked differences in the aging coefficients generated by the two multiple regression models. None of the fourteen medical factors, including hypothyroidism, had a significant effect on age regulation (<xref ref-type="supplementary-material" rid="pgen-0020115-sg002">Figure S2</xref>). Second, we performed an unsupervised hierarchical cluster analysis of the 250 age-regulated genes. If our analysis of age regulation were confounded by a medical factor, we would expect that the presence of the medical factor would be clustered when we sorted the 81 patients according to their patterns of gene expression. None of the pathological or pharmaceutical factors showed clustering (<xref ref-type="supplementary-material" rid="pgen-0020115-sg003">Figure S3</xref>). Most of the nonabdominal samples were from young patients, and there was some clustering of the muscle samples according to their anatomical origin as expected. This clustering does not affect our analysis of age regulation because anatomical origin was included as a term in the multiple regression model. Thus, these two methods showed no evidence that anatomical, pathological, or pharmaceutical factors confound the results of our aging study.</p><p>In summary, we have generated a global profile of changes in gene expression during aging in human muscle (<xref ref-type="fig" rid="pgen-0020115-g001">Figure 1</xref>). It is well established that aging has many effects on muscle, such as decrease in physiological performance, changes in morphology, and increased susceptibility to disease. The data from <xref ref-type="fig" rid="pgen-0020115-g001">Figure 1</xref> extend our understanding of muscle aging to the level of specific genes and genetic pathways, providing insight into possible mechanisms underlying overall decline of muscle function in old age. Overall, the difference in gene expression between young and old muscle tissue is relatively small. Specifically, only 250 genes show significant changes in expression with age (<italic>p</italic> < 0.001), and the large majority of these age-regulated genes change expression less than two-fold in 50 y. These results are consistent with a model in which age-related decline in cellular functions is caused by the accumulation of multiple, minute changes in the regulation of genes and pathways.</p><p>The genetic functions of many of the 250 genes shown in <xref ref-type="fig" rid="pgen-0020115-g001">Figure 1</xref> are known, and some suggest biological mechanisms that could cause age-related decline in muscle physiology. For example, <italic>CYP26B1</italic> shows an average increase in expression of 90% in 50 y. <italic>CYP26B1</italic> is a member of the cytochrome P450 family, which are monoxygenases used to metabolize toxic substances. Increased expression of <italic>CYP26B1</italic> in old age could help eliminate toxins that accumulate with age.</p><p> <italic>LASS5</italic> decreases expression approximately 25% in 50 y. <italic>LASS5</italic> is the human ortholog of the yeast <italic>lag1</italic> longevity assurance gene. In yeast, <italic>lag1</italic> expression decreases in older yeast cells [<xref rid="pgen-0020115-b021" ref-type="bibr">21</xref>] similar to our results showing decreased expression in old age in human muscle. <italic>LASS5</italic> is involved in the ceramide signaling pathway, which plays important roles on several lifespan-associated processes, such as stress resistance and apoptosis [<xref rid="pgen-0020115-b022" ref-type="bibr">22</xref>]. Reduced expression of <italic>LASS5</italic> in old age could impair cell function by reducing ceramide signaling.</p><p>In addition to searching for age regulation one gene at a time, we also screened known genetic pathways for those showing an overall change with age. With this approach, age regulation for every gene in a pathway is combined to determine whether there is an overall regulation of the entire pathway. Screening for coordinated age regulation of genetic pathways increases the sensitivity of our analysis, as the combined effects of small regulation of many genes in a pathway can be significant. For example, in a previous study of type 2 diabetes, screening genetic pathways for changes in expression provided key insights that were not possible from analyzing genes individually [<xref rid="pgen-0020115-b023" ref-type="bibr">23</xref>].</p><p>We developed a variant of gene set enrichment analysis (GSEA) to determine whether a genetic pathway shows evidence for age regulation [<xref rid="pgen-0020115-b023" ref-type="bibr">23</xref>]. We assayed 624 gene sets defined by the Gene Ontology consortium [<xref rid="pgen-0020115-b024" ref-type="bibr">24</xref>] (<xref ref-type="supplementary-material" rid="pgen-0020115-st003">Table S3</xref>). We modified the original GSEA paradigm because it was intended for datasets with two categories of sample, and we were instead fitting regression models to continuously varying independent and dependent variables. Accordingly, we replaced the two-sample test statistic in GSEA with an estimated regression slope for age. We also replaced the Kolmogorov-Smirnov statistic with a van der Waerden statistic because we prefer the type of dependence that the van der Waerden statistic captures. Finally, we replaced the permutation strategy with a bootstrap in order to better handle covariates (Materials and Methods).</p><p>Our version of the GSEA algorithm scores a gene set according to how the genes in it show coordinated increase (or decrease) on average in response to increasing age. The increase is measured by a van der Waerden statistic. To judge whether a specific van der Waerden statistic is significant, we used bootstrap resampling. Each bootstrap sample was drawn by resampling the arrays and keeping the gene expression measurements linked with the age, sex, and anatomy variables. The 624 van der Waerden scores for the gene groups were recomputed for each of the 1,000 bootstrap samples. Six gene sets were found to have statistically significant van der Waerden scores (<italic>p</italic> < 0.001) in this resampling. At this <italic>p</italic> value threshold, we would only expect to find 0.6 gene sets by chance alone. We also required the gene groups to show some practical significance by rejecting groups with a van der Waerden score smaller than 3.1 in absolute value. We found three pathways that passed both criteria: symporter genes, sialyltransferases, and chloride transporters showed decreasing expression with age (<xref ref-type="fig" rid="pgen-0020115-g002">Figure 2</xref> and <xref ref-type="table" rid="pgen-0020115-t002">Table 2</xref>). Aging coefficients for all genes in these pathways are listed in <xref ref-type="supplementary-material" rid="pgen-0020115-st004">Table S4</xref>.</p><fig id="pgen-0020115-g002" position="float"><label>Figure 2</label><caption><title>Three Gene Sets Are Regulated with Age in Muscle</title><p>Rows represent the symporter activity, sialyltransferase activity, and chloride transport gene sets. Columns correspond to individual genes within a given gene set. Scale represents the slope of the change in log<sub>2</sub> expression level with age (<italic>β<sub>1j</sub></italic>). A navigable version of this figure showing identities of specific genes can be found at <ext-link ext-link-type="uri" xlink:href="http://cmgm.stanford.edu/~kimlab/aging_muscle">http://cmgm.stanford.edu/~kimlab/aging_muscle</ext-link>.</p></caption><graphic xlink:href="pgen.0020115.g002"/></fig><table-wrap id="pgen-0020115-t002" content-type="1col" position="float"><label>Table 2</label><caption><p>Age-Regulated Gene Sets in Muscle</p></caption><graphic xlink:href="pgen.0020115.t002"/></table-wrap><p>Symporter genes (63 genes) and chloride transporters (35 genes) are necessary for transporting solutes during muscle contraction [<xref rid="pgen-0020115-b025" ref-type="bibr">25</xref>]; the decreased expression levels of these transporters may be associated with weakness of old muscle. Genes with sialyltransferase activity (19 genes) mediate glycosylation by transferring sialic acid groups to secreted molecules. Decreases in sialyltransferase activity have been previously detected in aging human serum [<xref rid="pgen-0020115-b026" ref-type="bibr">26</xref>], neurons [<xref rid="pgen-0020115-b027" ref-type="bibr">27</xref>], and lymphocytes [<xref rid="pgen-0020115-b028" ref-type="bibr">28</xref>].</p></sec><sec id="s2b"><title>Molecular Markers of Physiological Aging</title><p>Some people age slowly and remain strong and fit in their 70s, whereas others age rapidly, becoming frail and susceptible to age-related disease. We wanted to determine whether the expression profile for the 250 aging-regulated genes correlated with physiological in addition to chronological aging. For example, patient V17 was 41 y old but expressed his age-regulated genes similarly to patients who were 10 to 20 y older, and we would like to determine whether this patient had poor muscle physiology for his age (<xref ref-type="fig" rid="pgen-0020115-g001">Figure 1</xref>). Conversely, patient M73 was 64 y old but had a molecular profile similar to other patients that were 30 y younger, and we wanted to determine whether this patient had relatively good muscle physiology for his age. Our list consists of 250 genes that correlate significantly with chronological age. We sought to determine whether they also correlate with physiological age, as measured by the type II/type I diameter ratio. We prepared histological sections for all 81 skeletal muscle samples, and were able to reliably measure the diameters of the type I and type II muscle fibers for 32 samples (<xref ref-type="fig" rid="pgen-0020115-g003">Figure 3</xref>A and <xref ref-type="fig" rid="pgen-0020115-g003">3</xref>B; <xref ref-type="supplementary-material" rid="pgen-0020115-st005">Table S5</xref>).</p><fig id="pgen-0020115-g003" position="float"><label>Figure 3</label><caption><title>Gene Expression Predicts Physiology of Aging</title><p>(A) Cross-section of histologically unremarkable deltoid muscle from a 48-y-old woman demonstrating relatively equivalent sizes of types I and II muscle fibers. Arrows denote fibers types as distinguished by enzyme histochemistry (cryosection, 200×, myosin ATPase at pH 9.4).</p><p>(B) Cross-section of deltoid muscle from an 88-y-old woman demonstrating selective atrophy of type II muscle fibers that stain darkly by ATPase enzyme histochemistry (cryosection, 200×, myosin ATPase at pH 9.4).</p><p>(C) Histograms showing a correlation between muscle physiology and gene expression for age-regulated genes. Top panel: for each of the 250 age-regulated genes, we calculated the partial correlation coefficients between the type II/type I muscle fiber diameter ratio and gene expression excluding age variation (<italic>x</italic>-axis). Bottom panel: same as top panel, except that correlation coefficients were calculated for all 31,948 genes. The squared partial correlation coefficient denotes the amount that changes in gene expression account for variance in type II/type I muscle fiber diameter ratios while excluding the effects of age.</p><p>(D) Histogram showing the likelihood of finding 92 genes with \|<italic>r</italic>\| > 0.2 from a set of random genes. We performed a Monte Carlo experiment by randomly selecting sets of 250 genes from the genome, and calculating how many genes in the set had \|<italic>r</italic>\| > 0.2 as in (C). The procedure was repeated 1,000 times and the histogram shows the number of genes from each random selection that have \|<italic>r</italic>\| > 0.2. The arrow shows the number of genes exceeding this threshold (92) from the set of 250 age-regulated genes (<italic>p</italic> < 0.001). We also determined the total number of genes in the genome with \|<italic>r</italic>\| > 0.2, and then showed that 92 genes from a set of 250 is significant (hypergeometric distribution; <italic>p</italic> < 1 × 10<sup>−4</sup>).</p></caption><graphic xlink:href="pgen.0020115.g003"/></fig><p>A simple correlation of gene expression with muscle type ratio would not be sufficient for our purposes. Such a correlation could arise simply because the gene expression and muscle type ratio are both correlated with age. Accordingly, we employed partial correlations of gene expression with muscle type ratios after adjusting for the effect of chronological age. To do this, we regressed type II/type I muscle fiber diameter ratio on age, regressed gene expression on age, and finally correlated residuals from both regressions to obtain partial correlation coefficients. The partial correlations for the 250 age-related genes are shown in <xref ref-type="fig" rid="pgen-0020115-g003">Figure 3</xref>C.</p><p>If a gene correlates with muscle diameter ratio only because both it and muscle diameter are correlated with age, then the partial correlation described above should be close to zero. We found that a large number of the genes in our list had a statistically significant relationship with type II/type I ratio after adjusting for age. However, many of the genes not on our list were also related to type II/type I ratio adjusted for age. We were able to show that genes with large partial correlations were significantly overrepresented in our list of 250 age-regulated genes. We counted 92 of 250 age-related genes for which the (absolute) partial correlation was more than 0.2 (<xref ref-type="supplementary-material" rid="pgen-0020115-st006">Table S6</xref>). There were only 7,768 of 31,948 genes not in the list with a partial correlation this large. Using a hypergeometric distribution, we found a <italic>p</italic> value below 0.0001 and concluded that the age-related genes are more likely than other genes to have some partial correlation with muscle diameter ratio. To illustrate this effect, we also sampled 250 genes from the genome 1,000 different times, each time counting how many had a partial correlation larger than 0.2 in absolute value. None of the samples had a count larger than 92 (<xref ref-type="fig" rid="pgen-0020115-g003">Figure 3</xref>D).</p><p>Our result indicating that the 250 age-regulated genes are enriched for genes regulated by type II/type I muscle fiber diameter ratio is valid even when we use other selection thresholds for muscle physiology (i.e., other than the absolute of <italic>r</italic> > 0.2). We compared the distribution of partial correlations of the 250 age-regulated genes with type II/type I ratios to the distribution of partial correlations of the rest of the genes in the genome using nonparametric methods (<xref ref-type="fig" rid="pgen-0020115-g003">Figure 3</xref>C). Using a Kolmogorov-Smirnov goodness-of-fit test, we found that the distribution of the 250 age-regulated genes is wider than the total distribution in a two-sided test (<italic>p</italic> < 1 × 10<sup>−15</sup>, with D = 0.27). This result indicates that the apparent physiological basis of our gene set is not a consequence of our having chosen 0.2 as a threshold.</p><p>In summary, these statistical tests show that the set of age-regulated genes are markers of the relative level of muscle function, even among patients that are similar in age. Our findings are further supported by two additional statistical tests described in Materials and Methods (<xref ref-type="supplementary-material" rid="pgen-0020115-st007">Tables S7</xref> and <xref ref-type="supplementary-material" rid="pgen-0020115-st008">S8</xref>). Thus, the age-regulated genes are enriched for those that predict physiological, not just chronological, age. The correlation between gene expression profile and physiological age can be seen in patients V17 and M73 in <xref ref-type="fig" rid="pgen-0020115-g001">Figure 1</xref>. Although patient V17 is relatively young (41 y old), the gene expression profile for the 250 age-regulated genes is most similar to older individuals, and the type II/type I muscle fiber diameter ratio is low for his age. Conversely, although patient M73 is relatively old (64 y old), the gene expression pattern is similar to younger individuals, and the type II/type I muscle fiber diameter ratio is high for his age (<xref ref-type="fig" rid="pgen-0020115-g001">Figure 1</xref>).</p></sec><sec id="s2c"><title>A Common Signature for Aging in Muscle, the Kidney, and the Brain</title><p>Some aspects of aging affect only specific tissues; examples include progressive weakness of muscle, declining synaptic function in the brain, or decreased filtration rate in the kidney. Other aspects of aging occur in all cells regardless of their tissue type, such as the accumulation of oxidative damage from the mitochondria, DNA damage, and protein damage. Our genome-wide search for gene expression changes during aging would include both types of expression changes, and it would be interesting to discern which expression changes are muscle specific and which are common to all tissues. Expression profiles that are common to aging in all tissues would provide insight into the core mechanisms that underlie cellular aging. Therefore, we compared the DNA chip expression data from our studies on muscle aging to previous DNA chip expression studies on aging in the brain and the kidney. Rodwell et al. have characterized gene expression changes with age in the cortex and the medulla of the kidney from 74 patients, and Lu et al. have examined gene expression changes in the frontal cortex of the brain from 30 patients [<xref rid="pgen-0020115-b005" ref-type="bibr">5</xref>,<xref rid="pgen-0020115-b006" ref-type="bibr">6</xref>].</p><p>Our initial attempt to compare transcriptional changes between tissues relied on a Venn analysis, in which we directly compared the overlap in the lists of the age-regulated genes from the three tissues. Next, we searched for a common aging signature by comparing the Pearson correlation of age regulation between two tissues. Both of these straightforward methods showed only borderline statistical evidence for similarities in aging between the three tissues (Materials and Methods), but neither is expected to be powerful. Ultimately, we compared tissues using a grouped gene analysis. Grouping genes can be more powerful if there are small but consistent effects in each of a number of genes. Furthermore, the specific biological processes associated with each genetic pathway provide insights into mechanisms of aging. We used the modified GSEA described above to analyze previously published data on age regulation in the kidney and the brain [<xref rid="pgen-0020115-b005" ref-type="bibr">5</xref>,<xref rid="pgen-0020115-b006" ref-type="bibr">6</xref>]. As before, we considered the possibility that the observed correlations could be due to the fact that there might be random sampling differences in the different tissues that coordinately affect the expression levels of genes in an entire gene set. To control for this possibility, we resampled the microarray data 1,000 times (with replacement) and repeated the analysis of 624 gene sets on every resample. We selected only gene sets that were statistically significant in all three tissues at <italic>p</italic> < 0.05. We then removed any gene set that did not attain a van der Waerden score of 1.65 or more in absolute value in all three tissues. From a total of 624 sets of genes, we found that extracellular matrix genes, cell growth genes, and complement activation genes significantly increase expression with age on average in all three human tissues, whereas chloride transport genes and electron transport genes significantly decrease expression on average with age in those same tissues (<xref ref-type="table" rid="pgen-0020115-t003">Table 3</xref>). The cytosolic ribosomal pathway showed increased expression that was significant in the muscle and kidney, and almost significant in the brain (bootstrap <italic>p</italic> < 0.06). Aging coefficients for all genes in each of these pathways are listed in <xref ref-type="supplementary-material" rid="pgen-0020115-st009">Table S9</xref>. We would expect 0.08 (essentially none) of the 624 pathways to appear commonly age regulated by chance (<italic>p</italic> < 0.05 in all three tissues, and hence a combined <italic>p</italic> < 1.25 × 10<sup>−4</sup>).</p><table-wrap id="pgen-0020115-t003" content-type="1col" position="float"><label>Table 3</label><caption><p>Age Regulation of Gene Sets in Three Human Tissues</p></caption><graphic xlink:href="pgen.0020115.t003"/></table-wrap><p>Increased overall expression of the extracellular matrix gene set (152 genes) with advancing age may contribute to widespread fibrosis in the elderly (<xref ref-type="fig" rid="pgen-0020115-g004">Figure 4</xref>). Fibrosis is a process by which fibrous connective tissue proliferates throughout organs and impairs function of many tissues. <italic>TIMP1,</italic> which encodes tissue inhibitor of metalloproteinase 1, shows the largest increase in expression with age (average of 236% in 50 y).</p><fig id="pgen-0020115-g004" position="float"><label>Figure 4</label><caption><title>A Common Signature for Aging in Muscle, the Kidney, and the Brain</title><p>Shown are expression data from sets of extracellular matrix genes, cell growth genes, complement activation genes, cytosolic ribosomal genes, chloride transport genes, and electron transport chain genes. Rows are human tissues (M, muscle; K, kidney; B, brain). Columns correspond to individual genes in each gene set. Scale represents the slope of the change in log<sub>2</sub> expression level with age <italic>(β<sub>1j</sub>).</italic> Gray indicates genes were not present in the dataset. A navigable version showing identities of specific genes can be found at <ext-link ext-link-type="uri" xlink:href="http://cmgm.stanford.edu/~kimlab/aging_muscle">http://cmgm.stanford.edu/~kimlab/aging_muscle</ext-link>.</p></caption><graphic xlink:href="pgen.0020115.g004"/></fig><p>The cell growth gene set (29 genes) includes genes coding for growth factors, such as <italic>TGFB1</italic> and <italic>FGFR1.</italic> Induction of genes in this gene set may reflect an attempt to repair tissue damage that accumulates over lifespan.</p><p>Although complement activation genes (22 genes) are induced in muscle, the kidney, and the brain, they are expressed primarily in liver [<xref rid="pgen-0020115-b029" ref-type="bibr">29</xref>]. Therefore, unless complement genes are also age regulated in the liver, the physiological relevance of age regulation of complement genes in muscle, the kidney, and the brain is currently unclear.</p><p>Cytosolic ribosomal genes include 85 genes that show a general increase in expression with age in all three tissues. This result is interesting because the rate of protein synthesis is known to decrease in old age [<xref rid="pgen-0020115-b030" ref-type="bibr">30</xref>], and yet our expression results show an increase in the expression of ribosomal genes. One possibility is that decreased protein synthesis in old cells induces expression of ribosomal genes as part of a homeostatic feedback loop to partially compensate for loss of translational efficiency.</p><p>The chloride transport pathway is composed of 35 genes that show an overall decrease in expression with age in all three tissues. Ion transport of many types is important not only in the contraction of muscle [<xref rid="pgen-0020115-b025" ref-type="bibr">25</xref>], but also for maintenance of salt balance in the kidney [<xref rid="pgen-0020115-b031" ref-type="bibr">31</xref>] and neuron function in the brain through GABA-mediated receptors [<xref rid="pgen-0020115-b032" ref-type="bibr">32</xref>]. Decreased transport of chloride with age could lead to many types of physiological decline linked to ion transport deficiency.</p><p>The mitochondrial electron transport chain was found to show an overall decrease in expression with age. This group contains 95 genes, including genes associated with the NADH dehydrogenase family (complex I), succinate-coenzyme Q reductase (complex II), ubiquinone-cytochrome c reductase (complex III), cytochrome c oxidase (complex IV), H<sup>+</sup>-ATP synthase (complex V), and the uncoupling proteins. The finding that expression of genes involved in the electron transport chain decreases in old age supports the mitochondrial free-radical theory of aging [<xref rid="pgen-0020115-b033" ref-type="bibr">33</xref>], as free-radical generation by mitochondria would preferentially damage the electron transport chain protein complex. Decreased expression of the electron transport genes (encoded in the nucleus) might be caused by feedback regulation from damage to the electron transport chain protein complex. Other protein complexes in the mitochondria (such as mitochondrial ribosomal genes) do not decrease expression with age. Thus, aging does not have a general effect on genes encoding mitochondrial components, but rather specifically affects expression of genes that are part of the electron transport chain.</p><p>The above results show that there is common age regulation for these six genetic pathways in the kidney, muscle, and the brain. Next, we determined that there was little statistical evidence for the correlation of age regulation of individual genes in a pathway in one tissue with their age regulation in another tissue (Materials and Methods). Thus, it is unclear whether or not the same genes or different genes within a pathway show age regulation between different tissues. For example, certain genes in the electron transport pathway might be age regulated in the kidney, whereas other electron transport genes might be age regulated in the muscle.</p></sec><sec id="s2d"><title>A Public Age-Regulated Pathway in Humans, Mice, and Flies</title><p>Having identified genetic pathways that are commonly age regulated in different human tissues, we next determined whether their age regulation is specific for humans (private) or whether these groups are also age regulated in other species (public). Genetic pathways that are age regulated in different species would be of particular interest because they would identify mechanisms that are inextricably related to aging, even in animals that have vastly different lifespans.</p><p>We compared age regulation in humans to previously published studies of age regulation in <named-content content-type="genus-species">D. melanogaster</named-content> [<xref rid="pgen-0020115-b016" ref-type="bibr">16</xref>] and <named-content content-type="genus-species">C. elegans</named-content> [<xref rid="pgen-0020115-b018" ref-type="bibr">18</xref>]<italic>.</italic> To examine age regulation in aging mouse kidneys, we collected a kidney sample from ten C57BL/6 mice at 1, 6, 16, and 24 mo of age for a total of 40 mouse kidney samples. RNA from each kidney was extracted, labeled with P<sup>33</sup>-dCTP, and hybridized to cDNA filter membranes comprising 16,896 cDNA clones corresponding to 11,512 unique genes. We normalized expression values using the Z-score method [<xref rid="pgen-0020115-b034" ref-type="bibr">34</xref>], and analyzed age regulation of each gene using a multiple regression model taking into account age and sex of each mouse donor. <xref ref-type="supplementary-material" rid="pgen-0020115-st010">Table S10</xref> shows the slope of expression with respect to age for each gene.</p><p>We first identified orthologs of human genes in each of the other three species. Next, we determined the change in expression with respect to age for each gene in each species, using multiple regression techniques similar to the ones used for our studies of aging in human muscle (Material and Methods). We took the six gene sets shown to be aging-regulated in diverse human tissues, and then asked whether they also showed age regulation in any of the other three species. We analyzed the expression of each of the gene sets using modified GSEA to determine whether they showed an overall bias in expression with age in each species. Extracellular matrix genes, cell growth genes, complement activation genes, cytosolic ribosomal genes, and chloride transport genes did not show age regulation in other species.</p><p>The electron transport chain genes showed a consistent overall decrease in expression with age in humans, mice, and <italic>Drosophila,</italic> but did not show significant age regulation in <named-content content-type="genus-species">C. elegans</named-content> (<xref ref-type="fig" rid="pgen-0020115-g005">Figure 5</xref> and <xref ref-type="table" rid="pgen-0020115-t004">Table 4</xref>). To show that age regulation is not likely to be due to random biological sampling error, we resampled the electron transport data set in each species with replacement and found that the electron transport chain genes showed significant age regulation in mice (<italic>p</italic> < 0.02) and flies (<italic>p</italic> < 0.001) but not <italic>C. elegans.</italic> The electron transport chain gene set also shows a large van der Waerden score in mice and flies (less than −3.7). In summary, humans, mice, and flies show decreased expression of the electron transport chain during aging, defining a public pathway for aging across species with very different lifespans. In <italic>C. elegans,</italic> it is unclear whether the lack of support for age regulation of the electron transport chain pathway is because the pathway is not age regulated or because the DNA microarray experiments lack statistical power to detect age regulation.</p><fig id="pgen-0020115-g005" position="float"><label>Figure 5</label><caption><title>The Electron Transport Chain Decreases Expression with Age in Humans, Mice, and Flies</title><p>Rows represent either human tissues or model organisms. Columns correspond to individual human genes and homologs to human genes defined by reciprocal best BLAST hits in other species. Scale represents the normalized slope of the change in log<sub>2</sub> expression level with age (<italic>β<sub>1j</sub></italic>). Data from different species were normalized by dividing the slope of expression with age by the standard deviation of all similar slopes in the dataset. Gray indicates genes were not present in that species. A navigable version of this figure showing identities of specific genes can be found at <ext-link ext-link-type="uri" xlink:href="http://cmgm.stanford.edu/~kimlab/aging_muscle">http://cmgm.stanford.edu/~kimlab/aging_muscle</ext-link>.</p></caption><graphic xlink:href="pgen.0020115.g005"/></fig><table-wrap id="pgen-0020115-t004" content-type="1col" position="float"><label>Table 4</label><caption><p>Age Regulation of the Electron Transport Chain in Three Species</p></caption><graphic xlink:href="pgen.0020115.t004"/></table-wrap></sec></sec><sec id="s3"><title>Discussion</title><p>In this study, we have generated a high-resolution transcriptional profile of aging in the human muscle. Welle et al. have previously used DNA chips to profile expression changes during aging for the human muscle [<xref rid="pgen-0020115-b007" ref-type="bibr">7</xref>,<xref rid="pgen-0020115-b008" ref-type="bibr">8</xref>], and this work extends their previous studies because we used a significantly larger sample size that enabled much higher statistical resolution.</p><p>People age at different rates, especially with regard to muscular aging. Some remain fit and strong, whereas other become frail and weak when they are old. The transcriptional profile for aging in this study reflects the physiological age of the subjects, as measured by muscle diameter ratio, after making an adjustment for their chronological ages. Previous work on age regulation in the kidney also identified molecular markers that could predict the physiological age of the kidney [<xref rid="pgen-0020115-b005" ref-type="bibr">5</xref>].</p><p>Our results provide the some of the first evidence for a common signature of changes of gene expression in different human tissues. Specifically, we found similar patterns of age regulation for six biological pathways in the muscle, the kidney, and the brain. Previous studies found similar patterns of aging between different parts of the same tissue, but not between entirely different organs (i.e., age regulation was found to be similar between the cortex and medulla of the kidney [<xref rid="pgen-0020115-b005" ref-type="bibr">5</xref>] and between the frontal pole and the prefrontal cortex in the brain [<xref rid="pgen-0020115-b013" ref-type="bibr">13</xref>]).</p><p>Except for the complement activation gene set, the pathways that show common age regulation in diverse tissues also function in all cells. Changes in expression of these pathways in old age may lead to degeneration of not only core cellular functions (such as ion transport and energy production) but also to degeneration of tissue-specific functions (such as kidney filtration and synaptic signaling) that rely on housekeeping pathways. By identifying a common aging signature across tissues, we can now focus on aging pathways that are general instead of tissue-specific. The common aging signature reflects the age of diverse organs, whereas genes that are age regulated in just one tissue reflect the age of that tissue. Finally, treatments or therapies that alter expression of the four common age-regulated pathways might be expected to affect diverse tissues instead of a specific tissue, and may therefore have an overall effect on longevity.</p><p>Although some patterns of aging are similar between different human tissues, much of aging is tissue-specific. Decreases in expression of the sialyltransferases and symporter genes are changes specific to muscle, and do not appear to occur in either the kidney or the brain.</p><p>Nearly all of the age regulation that we found is specific to humans, and does not seem to occur in old mice, flies, or worms. Thus, much of age regulation in humans is species-specific (private) rather than universal for all animals (public). This result emphasizes the importance of studying aging in humans rather than model organisms with short lifespans in order to understand how people grow old.</p><p>Nevertheless, we did find one pathway that was age regulated in humans, mice, and flies. The electron transport chain gene pathway decreases expression with age in all three species. Previous studies found little or no similarity in age regulation between humans and mice [<xref rid="pgen-0020115-b005" ref-type="bibr">5</xref>] or primates [<xref rid="pgen-0020115-b013" ref-type="bibr">13</xref>]. These studies might have overlooked public patterns of age regulation in different species because they searched for similarities in age regulation at the level of individual genes rather than of entire genetic pathways (too little sensitivity) or because the aging experiments involved only a few individuals (too much experimental noise). Another previous study compared aging in flies and worms, and reported that there was a common decrease in expression, seen in young adulthood, of genes that encode mitochondrial proteins [<xref rid="pgen-0020115-b019" ref-type="bibr">19</xref>].</p><p>In mammals, direct genetic tests of the functional relevance of reduced expression of the electron transport chain pathway on lifespan have not yet been reported. However, in <italic>C. elegans,</italic> reducing the activity of eight genes involved in the electron transport chain using RNAi increased lifespan significantly [<xref rid="pgen-0020115-b035" ref-type="bibr">35</xref>,<xref rid="pgen-0020115-b036" ref-type="bibr">36</xref>]. A gene encoding a subunit of NADH dehydrogenase <italic>(NDUFA10)</italic> is one of the genes showing the largest decrease in expression with age in humans, and its ortholog in worms, <italic>K04G7.4,</italic> was also found to cause one of the largest increases in lifespan using RNAi in <named-content content-type="genus-species">C. elegans</named-content> [<xref rid="pgen-0020115-b036" ref-type="bibr">36</xref>]. Indeed, in these global RNAi screens, the electron transport chain pathway stands out as the pathway showing the largest and most consistent effect on extending lifespan in worms [<xref rid="pgen-0020115-b035" ref-type="bibr">35</xref>]. The genetic results from worms suggest that diminished expression of the electron transport chain pathway in old age in humans may be beneficial, contributing toward extending lifespan.</p><p>What types of upstream events might cause a decrease in expression of the electron transport chain pathway with age? Other mitochondrial pathways, such as the mitochondrial ribosome, do not show age regulation similar to the electron transport chain pathway. One potential cause of decreased expression of the electron transport chain pathway is that metabolism may slow in old age, resulting in reduced expression of the energy producing machinery of the cell. Another possibility is that oxidative damage to the proteins in the electron transport chain in the mitochondria may lead to reduced expression of the corresponding genes in the nucleus. The electron transport chain creates free radicals in the process of generating energy that would preferentially damage protein components of the electron transport chain [<xref rid="pgen-0020115-b033" ref-type="bibr">33</xref>].</p><p>It seems unlikely that common age regulation of the electron transport chain pathway is directly due to evolutionary conservation. Events in old age are unlikely to have a significant effect on fitness of a population because old animals (such as 3-y-old mice and 80-y-old people) are a small fraction of natural populations (except in recent human history). It could be that the electron transport chain is regulated during aging as an indirect consequence of regulation during development (antagonistic pleiotropy) [<xref rid="pgen-0020115-b037" ref-type="bibr">37</xref>]. Alternatively, age regulation of this pathway may be an unavoidable consequence of aging (e.g., oxidative damage to the electron transport chain in old age may occur in all animals) [<xref rid="pgen-0020115-b033" ref-type="bibr">33</xref>].</p><p>It is interesting that the level of age regulation of the electron transport chain is nearly the same in each species, whereas lifespan varies greatly. Compared to humans, mice age 20- to 30-fold and flies age 400-fold more rapidly. Thus, the kinetics of the changes in gene expression for the electron transport chain genes precisely matches the difference in lifespan between species. This suggests that decreased expression of the electron transport chain pathway with age may be particularly informative as a marker of physiological aging.</p></sec><sec id="s4"><title>Materials and Methods</title><sec id="s4a"><title>Sample collection.</title><p>The muscle samples were obtained from patient biopsies collected either during surgery or in an outpatient procedure, and the medical conditions associated with each biopsy are listed in <xref ref-type="supplementary-material" rid="pgen-0020115-st001">Table S1</xref>. For example, the abdominal muscle samples were harvested during surgeries to treat gastrointestinal pathologies. There was no known pathology associated with the abdominal muscle samples themselves, except that they were obtained from patients with various gastrointestinal disorders. In the case of patients with gastrointestinal cancer, the abdominal muscle samples were harvested from regions of the abdomen that were not affected by the cancer. Each muscle sample was immediately frozen in liquid nitrogen and subsequently stored at −80 °C. Finally, we checked each sample by histological staining, and excluded any samples that appeared abnormal or diseased.</p></sec><sec id="s4b"><title>RNA isolation.</title><p>Frozen muscle samples were weighed (50–100 mg), cut into small pieces on dry ice, and then placed in 1 ml of TRIzol Reagent (Invitrogen, Carlsbad, California, United States). The tissue was homogenized using a PowerGen700 homogenizer (Fisher Scientific, Pittsburgh, Pennsylvania, United States), and the total RNA was isolated according to the TRIzol Reagent protocol.</p></sec><sec id="s4c"><title>DNA gene chip hybridization.</title><p>A standard protocol designed by Affymetrix (Santa Clara, California, United States) for their HG-U133 2.0 Plus high-density oligonucleotide arrays was slightly modified by the Stanford Genome Technology Center (Stanford, California, United States), and all samples were processed in their facility (see Protocol S1). Eight micrograms of total RNA was used to synthesize cRNA for each sample, and 15 μg of cRNA was hybridized to each DNA chip. The samples were processed in random order with respect to age.</p></sec><sec id="s4d"><title>Microarray data normalization and analysis.</title><p>We used the DChip program [<xref rid="pgen-0020115-b038" ref-type="bibr">38</xref>] to normalize the data and to generate expression levels for each individual probe set by a perfect-match–only model. All expression data will be publicly available on the Gene Expression Omnibus website upon acceptance. When different probe sets corresponded to the same gene, we averaged the expression levels together. After averaging, we used log<sub>2</sub>-transformed expression values for all subsequent analyses.</p></sec><sec id="s4e"><title>Muscle fiber diameter measurement.</title><p>Cross-sections of muscle cryosections were photographed at 200×, and the pictures were either measured digitally (diagnostic muscle biopsy samples, ATPase preparations) or printed (abdominal muscle samples, combined SDH-cytochrome <italic>c</italic> oxidase preparations) and measured by hand. All of the diagnostic muscle biopsies were considered, and 32 of the 81 muscle samples were sufficiently intact for measurement, the remainder being inadequately oriented for cross-sections or too small for meaningful data. Digital analysis consisted of measuring the shortest width through the approximate center of the cell. After calibration with a known length, the diameters were measured and converted to microns using SigmaScan Pro 5.0 software (SPSS Software, Chicago, Illinois, United States). Diameters were tabulated by type I and type II cell types. The counts ranged from approximately 30 cells per type to more than 100 depending on the sample size. Print analysis was by similar methodology. Raw measurements in millimeters were used to calculate the ratio of type II to type I diameters without converting to microns.</p></sec><sec id="s4f"><title>Multiple regression analysis.</title><p>To determine the change in expression with age, we used a multiple regression model in which the change in expression with age takes into account the possibility that expression levels might differ in men versus women, or in abdominal muscle versus peripheral muscle. Specifically, we used the following multiple regression model:</p><disp-formula id="pgen-0020115-e001"><graphic xlink:href="pgen.0020115.e001.jpg" position="anchor" mimetype="image"/></disp-formula><p>where <italic>Y<sub>ij</sub></italic> is the expression level of the <italic>j</italic>th probe set for the <italic>i</italic>th sample, <italic>Age<sub>i</sub></italic> is the age in <italic>y</italic> of the <italic>i</italic>th sample, <italic>Sex<sub>i</sub></italic> corresponds to the sex of the <italic>i</italic>th sample (0 for male, or 1 for female), <italic>Anatomy<sub>i</sub></italic> is the anatomic location from which the muscle sample was harvested (0 for abdominal or 1 for peripheral muscle), <italic>ɛ<sub>ij</sub></italic> represents an error term, <italic>β<sub>1j</sub></italic> is the change of expression with age, <italic>β<sub>2j</sub></italic> is the change of expression with sex, <italic>β<sub>3j</sub></italic> is the change of expression with anatomical origin of sample, and <italic>β<sub>0j</sub></italic> is the regression intercept. For each gene <italic>j,</italic> we used least-squares to determine all of its coefficients, with our primary interest in the one with respect to age <italic>(β<sub>1j</sub>).</italic> We were interested in genes that show either a positive or negative value for <italic>β<sub>1j</sub>,</italic> indicating either increasing or decreasing expression in old age, respectively.</p><p>For human brain, mouse kidney, and <italic>D. melanogaster,</italic> we determined the change in expression with age for each gene using the following multiple regression model:</p><disp-formula id="pgen-0020115-e002"><graphic xlink:href="pgen.0020115.e002.jpg" position="anchor" mimetype="image"/></disp-formula><p>For human kidney, we used the multiple regression model:</p><disp-formula id="pgen-0020115-e003"><graphic xlink:href="pgen.0020115.e003.jpg" position="anchor" mimetype="image"/></disp-formula><p>In <xref ref-type="disp-formula" rid="pgen-0020115-e003">Equation 3</xref>, the tissue term is a binary term scored 0 for cortex and 1 for medulla. For <named-content content-type="genus-species">C. elegans</named-content> data, we used a simple linear regression with age:</p><disp-formula id="pgen-0020115-e004"><graphic xlink:href="pgen.0020115.e004.jpg" position="anchor" mimetype="image"/></disp-formula><p>The reviewers suggested two additional methods to show that the age-regulated genes could serve as markers for physiological age. First, we showed that genes regulated by muscle physiology can also predict chronological age. We found genes that were significantly regulated by type II/type I muscle fiber diameter ratio using the multiple regression model:</p><disp-formula id="pgen-0020115-e005"><graphic xlink:href="pgen.0020115.e005.jpg" position="anchor" mimetype="image"/></disp-formula><p>Here, <italic>TypeRatio</italic> is the ratio of type II to type I muscle fiber diameters. We found 585 genes with a statistically significant coefficient for <italic>TypeRatio</italic> using the threshold <italic>p</italic> < 0.01. Of these 585 genes, 114 showed partial correlation with age (absolute value of <italic>r</italic> > 0.2), indicating a significant overlap (<italic>p</italic> < 0.02; hypergeometric distribution) (<xref ref-type="supplementary-material" rid="pgen-0020115-st007">Table S7</xref>). The 92 genes found in the analysis shown in <xref ref-type="fig" rid="pgen-0020115-g003">Figure 3</xref> and the 114 genes found in this analysis share a common set of 7 genes, indicating a statistically significant overlap (<italic>p</italic> < 1 × 10<sup>−8</sup>; hypergeometric distribution).</p><p>Second, we repeated our age analysis taking into consideration the effect of type II/type I muscle fiber diameter ratio on age regulation. To do this, we used a four-term multiple regression model that includes terms for both age and type II/type I ratio:</p><disp-formula id="pgen-0020115-e006"><graphic xlink:href="pgen.0020115.e006.jpg" position="anchor" mimetype="image"/></disp-formula><p>Using <xref ref-type="disp-formula" rid="pgen-0020115-e006">Equation 6</xref>, we found 543 genes that were regulated by age (<italic>p</italic> < 0.01) and 12,786 genes regulated by type II/type I ratio (<italic>p</italic> < 0.01; <xref ref-type="supplementary-material" rid="pgen-0020115-st008">Table S8</xref>). There are 271 genes shared in common between these two sets of genes, which is a significantly larger number than would be expected by chance (hypergeometric <italic>p</italic> < 1 × 10<sup>−5</sup>; <xref ref-type="supplementary-material" rid="pgen-0020115-st008">Table S8</xref>). We repeated this experiment using a threshold of <italic>p</italic> < 0.001 and found similar enrichment, confirming our results. This analysis shows that the set of genes that are regulated by age is enriched for those that mark the physiology of aging muscle.</p></sec><sec id="s4g"><title>False discovery rate determined by permutation analysis.</title><p>We used a permutation analysis to simulate the number of genes that would pass our cutoff by chance (<italic>p</italic> < 0.001). We randomized the age variables of muscle samples 1,000 times while maintaining the sex and anatomy variables with the sample. <xref ref-type="disp-formula" rid="pgen-0020115-e001">Equation 1</xref> was used to recalculate regression coefficients and <italic>p</italic> values in every randomization. Theory predicts, and our simulation verifies, that on average about 32 genes pass our threshold (<italic>p</italic> < 0.001) by chance. This result suggests that there are about 13% false positives in our set of 250 age-regulated genes in the muscle. In 95% of the permuted datasets, 107 or fewer genes were significant at the 0.001 level.</p></sec><sec id="s4h"><title>Cluster analysis of pathological and pharmaceutical factors.</title><p>To examine whether pathological or pharmaceutical factors were confounding the analysis of age regulation in muscle, we performed unsupervised, average-linkage hierarchical clustering of the 81 muscle samples using the Cluster software [<xref rid="pgen-0020115-b039" ref-type="bibr">39</xref>]. The 81 muscle samples were clustered on the basis of the 250 genes previously determined to be age regulated in human muscle.</p></sec><sec id="s4i"><title>Modified gene set enrichment analysis.</title><p>GSEA [<xref rid="pgen-0020115-b040" ref-type="bibr">40</xref>] uses a nonparametric test to decide when the n genes in a group <italic>G</italic> have age coefficients that differ significantly from the N-n genes that are not in <italic>G.</italic> The model is that the n age coefficients in <italic>G</italic> are sampled from a distribution G, while the N-n coefficients not in <italic>G</italic> are sampled from a distribution F. We then test the null hypothesis that F = G. The Kolmogorov-Smirnov test is based on counting how many genes from <italic>G</italic> are in the top K genes of the combined list of age coefficients and comparing it to the number expected when F = G. By letting K vary from 1 to N, the test is sensitive to any alternative F ≠ G. GSEA employs a weighted Kolmogorov-Smirnov test obtained by using a weighted count of genes (with more weight on the extreme ones). In our analysis, we have replaced the weighted Kolmogorov-Smirnov test by a weighted sum, the van der Waerden normal scores test.</p><p>The van der Waerden test conforms more closely to our interpretation of what it means for a group <italic>G</italic> of genes to be age related than does the weighted Kolmogorov-Smirnov test. When N is large, then any small group that contains the single most age-related gene is significantly age related by the weighted Kolmogorov-Smirnov test. Such a group displays a genuine statistical significance and comprises strong evidence that F ≠ G, but isn't necessarily biologically increasing or decreasing expression as a mechanistic unit with age. For example, a group of 30 genes with two of the most age-increasing genes and 2 of the most age-decreasing genes could be found to be both an age-increasing group and also an age-decreasing group with significance, even when the other 26 genes are not particularly age related. Here it is clear that F ≠ G, but perhaps it is simply because G has higher variance than F.</p><p>To compute the van der Waerden test, we first find the rank r(j) for every gene j ∈ <italic>G</italic>. This rank is the number of the original N genes with an age coefficient smaller than that of gene j. The raw van der Waerden score is</p><disp-formula id="pgen-0020115-e007"><graphic xlink:href="pgen.0020115.e007.jpg" position="anchor" mimetype="image"/></disp-formula><p>where <italic>Φ</italic> is the standard normal cumulative distribution function. When the N age coefficients are independent with a common continuous distribution F = G, then the distribution of Y is very nearly normally distributed with mean 0 and a variance V<italic>(Y)</italic> close to N-n. We replaced the GSEA enrichment score by the van der Waerden statistic, Z = Y/<inline-formula id="pgen-0020115-ex001"><inline-graphic xlink:href="pgen.0020115.ex001.jpg" mimetype="image"/></inline-formula> , which is very nearly N<italic>(0,1)</italic> under the null hypothesis. When distribution G is shifted left or right relative to F, then the value of <italic>Z</italic> tends to increase beyond what we would expect from the N<italic>(0,1)</italic> distribution. </p></sec><sec id="s4j"><title>Bootstrap test for significance of GSEA.</title><p>It is better to use resampling methods instead of the N<italic>(0,1)</italic> null distribution to assess the significance of the enrichment score Z. The reason is that there are ordinarily correlations among the expression levels of the genes in <italic>G.</italic> When the expression levels of two genes in <italic>G</italic> are correlated, the age coefficients for those genes are correlated as well [<xref rid="pgen-0020115-b041" ref-type="bibr">41</xref>]. It then follows that their ranks are correlated, and this typically increases the variance of Y so that ultimately Z is no longer N<italic>(0,1).</italic> The value of Z can become large either because the genes are age related or because they are correlated with each other. Both may be biologically real, but the second is not an interesting finding, except possibly as confirmation that the group <italic>G</italic> is well constructed.</p><p>The original GSEA [<xref rid="pgen-0020115-b040" ref-type="bibr">40</xref>] randomly permutes the labels of two groups being tested while keeping the gene expression data intact. This preserves correlations within the groups so that any significant findings are relative to a null simulation that includes correlations among genes. In many random permutations, one gets a histogram of enrichment scores for age that is centered around zero. If the sample value is far outside the histogram then that enrichment score is statistically significant.</p><p>We adopted instead a bootstrap approach. We resampled the data and recomputed enrichment scores, obtaining a histogram roughly centered over the observed enrichment score. If the null value (zero) is far outside the resampled histogram, then the enrichment score is statistically significant. The bootstrap approach also preserves correlations among genes as well as correlations between genes and covariates.</p><p>The primary motivation for bootstrapping is the presence of covariates in our problems. Consider for example data with age, sex, and expression variables. If we permute the ages with respect to the expression data and repeat the regression, we have to decide whether the sex variable should be attached to the ages or to the expressions in the random permutation. Attaching sex to the age variables will leave us with simulated data sets in which females express Y chromosome genes as much as males. Because of such artifacts, this is not a suitable null distribution. Attaching a covariate to the expression variables is also problematic. Suppose that one of the covariates is somewhat correlated with age. The effect will be to increase the variance of the originally sampled age coefficient. In permutation samples where the covariate is attached to the expression data, it is resampled independently of age. Such independence reduces the variance of the age coefficient in the permutation data. The consequence is that the permutation-based histogram of age coefficients is then too narrow and false discoveries will result.</p><p>In the bootstrap approach we generated 1,000 sample datasets. In each sample dataset we mimicked the sampling process that gave rise to the data by resampling 81 subjects from the population of 81 subjects. The resampling keeps age, expression, and all covariates of any given subject together. Bootstrap sampling mimics the random process that generated the data.</p><p>We remark that both bootstrap and permutation sampling of the van der Waerden scores gave rise to Z scores that were nearly normally distributed, but not necessarily N(0,1) (unpublished data). In permutation sampling, the histogram of enrichment scores tended to have means near zero, but several groups had variances larger than 1.0. In bootstrap sampling, the variances often differed from 1 and the means were usually between zero and the original enrichment score.</p></sec><sec id="s4k"><title>Venn and correlation analysis of human muscle, the kidney, and the brain.</title><p>The most direct way to compare aging in muscle, the kidney, and the brain is via a Venn analysis: we find which genes attain a stringent significance level for each tissue and judge whether the overlap is statistically significant according to a hypergeometric distribution. We did a pairwise comparison between each tissue to find genes that are aging-regulated in both sets. There are six aging-regulated genes in both the muscle and the kidney (<italic>p</italic> < 0.09, hypergeometric distribution), five aging-regulated genes in both muscle and the brain (<italic>p</italic> < 0.07), and 13 aging-regulated genes in common between the kidney and the brain (<italic>p</italic> < 0.29). There were no genes that were strongly age regulated in all three datasets. The Venn analysis approach is very interpretable but lacks power because it replaces actual measured correlations by a less informative notion of whether they are over a threshold.</p><p>A more sensitive comparison can be based on correlating the age coefficients of genes in two tissues. We selected all genes that are age regulated in either of two tissues, plotted the age coefficient of each gene in one tissue versus that gene's coefficient in the other tissue, and computed the Pearson correlation <italic>(r)</italic> of the resulting points (<xref ref-type="supplementary-material" rid="pgen-0020115-st011">Table S11</xref>). We found the strongest overlap in aging between the kidney and the brain (<italic>r</italic> = 0.219), and smaller but positive overlaps in aging between the muscle and the kidney (<italic>r</italic> = 0.103) or the muscle and the brain (<italic>r</italic> = 0.078).</p><p>Because the genes are correlated we cannot use textbook formulas to judge the statistical significance of these Pearson scores. To get a <italic>p</italic> value for a Pearson correlation between kidney and muscle, we used 1,000 sets of random genes. The number of genes in each set was the same as the number we used to compute the correlation in <xref ref-type="supplementary-material" rid="pgen-0020115-st011">Table S11</xref>. For each random gene group we computed the Pearson correlation between age coefficients in kidney and muscle. Of the 1,000 samples, there were six in which the random gene group gave rise to a larger Pearson correlation than the one we saw in the real data. This corresponds to a <italic>p</italic> value of 0.006 for kidney–muscle. We similarly found a <italic>p</italic> value of 0.001 for kidney–brain but only 0.058 for muscle–brain. With the possible exception of the kidney–brain pair, the age-related genes have more consistent age coefficients across tissues than randomly selected genes do.</p><p>We also ran a bootstrap test of the tissue comparisons. In this test we resampled the microarray data with replacement 1,000 times. Each time we recomputed the correlations between age coefficients for genes in the kidney and muscle. In 1,000 trials we saw 39 in which the sample correlation was less than or equal to zero. After converting to a two-tailed test, this corresponds to a <italic>p</italic> value of 0.078 for kidney–muscle. To save computation, we used the same set of genes in each bootstrap sample instead of making the age-related gene set vary with the sample separately. The <italic>p</italic> value for muscle–brain was 0.07 while that for kidney–brain was 0.001. Based on these individual gene-level analyses, the age-related genes in the kidney and brain tended to be very similar. The muscle–kidney and muscle–brain comparisons were weaker.</p></sec><sec id="s4l"><title>Tests for correlation of tissues within commonly age-regulated gene sets.</title><p>To test for the correlation of gene ranks between tissues within those gene sets found to be commonly age regulated in the human, we used a two-tailed Spearman correlation method to first calculate a correlation coefficient for every pairwise combination of tissues (i.e., muscle–kidney, kidney–brain, muscle–brain) for that age-regulated gene set (e.g., extracellular matrix genes). In order to test for the significance of the calculated correlations, we used a permutation-based Monte Carlo method, randomizing the ranks for each gene and tissue in the gene set and recalculating Spearman correlations 1,000 times. We found that most of the correlations between tissues were not significant (<xref ref-type="supplementary-material" rid="pgen-0020115-st012">Table S12</xref>).</p></sec></sec><sec sec-type="supplementary-material" id="s5"><title>Supporting Information</title><supplementary-material content-type="local-data" id="pgen-0020115-sg001"><label>Figure S1</label><caption><title>Age Distribution of Anatomical, Medical, and Pharmaceutical Factors</title><p>Each row denotes a medical or pharmaceutical factor. Age of patients is shown on the <italic>x</italic>-axis. Sex, biopsy location, and 12 medical factors are shown in the legend. Only hypothyroidism shows any overt association with age.</p><p>(253 KB TIF)</p></caption><media xlink:href="pgen.0020115.sg001.tif"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020115-sg002"><label>Figure S2</label><caption><title>Medical and Pharmaceutical Factors do not Affect Age Regulation</title><p>(A) Coronary artery disease was included as an additional term in <xref ref-type="disp-formula" rid="pgen-0020115-e001">Equation 1</xref>, and the model was recalculated for the 250 genes that significantly change expression with age. The slope of expression with age (age coefficient) from models with (<italic>y</italic>-axis) and without (<italic>x</italic>-axis) the coronary artery disease term was plotted. If coronary artery disease affected expression, we would expect a large deviation in age coefficient. No significant deviation was seen for any of the 250 age-regulated genes, indicating that coronary artery disease does not adversely affect our study of age regulation.</p><p>(B–L) Similar to (A) for 11 other medical factors. (B) Coronary artery disease. (C) Colorectal cancer. (D) End-stage renal disease. (E) Hyperlipidemia. (F) Hypertension. (G) Hypothyroidism. (H) Pancreatic cancer. (I) Prostate cancer. (J) Radiotherapy. (K) Statins. (L) Villous adenoma.</p><p>(319 KB TIF)</p></caption><media xlink:href="pgen.0020115.sg002.tif"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020115-sg003"><label>Figure S3</label><caption><title>Cluster Analysis of Medical and Pharmaceutical Factors</title><p>Samples are clustered on the basis of 250 age-regulated genes in muscle, shown by the top dendrogram. Columns are individual muscle samples, marked by age of the patient. Top seven rows correspond to the expression of the first seven age-regulated genes. The diagram shows anatomical, medical, and pharmaceutical factors for each patient. Each row corresponds to one medical or pharmaceutical factor.</p><p>(1.2 MB TIF)</p></caption><media xlink:href="pgen.0020115.sg003.tif"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020115-st001"><label>Table S1</label><caption><title>Clinical Data</title><p>(2.1 MB XLS)</p></caption><media xlink:href="pgen.0020115.st001.xls"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020115-st002"><label>Table S2</label><caption><title>250 Age-Regulated Genes (<italic>p</italic> < 0.001), Arranged by Slope with Age</title><p>(59 KB XLS)</p></caption><media xlink:href="pgen.0020115.st002.xls"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020115-st003"><label>Table S3</label><caption><title>624 Gene Sets Assayed for Age Regulation</title><p>(63 KB XLS)</p></caption><media xlink:href="pgen.0020115.st003.xls"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020115-st004"><label>Table S4</label><caption><title>Contributing Genes in Gene Sets Age Regulated in Muscle</title><p>(45 KB XLS)</p></caption><media xlink:href="pgen.0020115.st004.xls"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020115-st005"><label>Table S5</label><caption><title>Type II/Type I Muscle Fiber Diameter Ratios of 32 Patients</title><p>(16 KB XLS)</p></caption><media xlink:href="pgen.0020115.st005.xls"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020115-st006"><label>Table S6</label><caption><title>92 Age-Regulated Genes that Predict Type II/Type I Ratio</title><p>(27 KB XLS)</p></caption><media xlink:href="pgen.0020115.st006.xls"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020115-st007"><label>Table S7</label><caption><title>114 Type II/Type I–Regulated Genes that Predict Age</title><p>(32 KB XLS)</p></caption><media xlink:href="pgen.0020115.st007.xls"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020115-st008"><label>Table S8</label><caption><title>Significant Overlap between Sets of Age-Regulated and Type II/Type I–Regulated Genes</title><p>(14 KB XLS)</p></caption><media xlink:href="pgen.0020115.st008.xls"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020115-st009"><label>Table S9</label><caption><title>Contributing Genes in Gene Sets Commonly Age Regulated in Muscle, the Kidney, and the Brain</title><p>(101 KB XLS)</p></caption><media xlink:href="pgen.0020115.st009.xls"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020115-st010"><label>Table S10</label><caption><title>Aging Coefficients of 11,512 Genes in Mouse Kidneys</title><p>(982 KB XLS)</p></caption><media xlink:href="pgen.0020115.st010.xls"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020115-st011"><label>Table S11</label><caption><title>Correlation of Age-Regulated Genes in Three Tissues</title><p>(15 KB XLS)</p></caption><media xlink:href="pgen.0020115.st011.xls"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020115-st012"><label>Table S12</label><caption><title>Spearman Correlations between Tissues for Age-Regulated Gene Sets</title><p>(15 KB XLS)</p></caption><media xlink:href="pgen.0020115.st012.xls"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material></sec>
Structural Model Analysis of Multiple Quantitative Traits	<p>We introduce a method for the analysis of multilocus, multitrait genetic data that provides an intuitive and precise characterization of genetic architecture. We show that it is possible to infer the magnitude and direction of causal relationships among multiple correlated phenotypes and illustrate the technique using body composition and bone density data from mouse intercross populations. Using these techniques we are able to distinguish genetic loci that affect adiposity from those that affect overall body size and thus reveal a shortcoming of standardized measures such as body mass index that are widely used in obesity research. The identification of causal networks sheds light on the nature of genetic heterogeneity and pleiotropy in complex genetic systems.</p>	<contrib contrib-type="author"><name><surname>Li</surname><given-names>Renhua</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Tsaih</surname><given-names>Shirng-Wern</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Shockley</surname><given-names>Keith</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Stylianou</surname><given-names>Ioannis M</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Wergedal</surname><given-names>Jon</given-names></name><xref ref-type="aff" rid="aff2">2</xref><xref ref-type="aff" rid="aff3">3</xref><xref ref-type="aff" rid="aff4">4</xref></contrib><contrib contrib-type="author"><name><surname>Paigen</surname><given-names>Beverly</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Churchill</surname><given-names>Gary A</given-names></name><xref ref-type="aff" rid="aff1">1</xref><xref ref-type="corresp" rid="cor1">*</xref></contrib>	PLoS Genetics	<sec id="s1"><title>Introduction</title><p>The most common and pervasive human health problems, including heart disease, osteoporosis, diabetes, and cancer, result from the complex interaction of multiple genetic and environmental factors. Disease states are often associated with multiple, correlated traits, referred to as subphenotypes. The ability to assay subphenotypes of a disease state presents a unique opportunity to investigate the mechanisms underlying disease susceptibility and progression [<xref rid="pgen-0020114-b001" ref-type="bibr">1</xref>]. Observed associations among these traits may be driven by common genetic factors or may result from physiological interactions [<xref rid="pgen-0020114-b001" ref-type="bibr">1</xref>,<xref rid="pgen-0020114-b002" ref-type="bibr">2</xref>]. For example, low- and high-density lipoprotein cholesterol, triglycerides, blood pressure, plasma insulin levels, and C-reactive proteins are all measurable phenotypes associated with cardiovascular disease. We would like to know whether these phenotypes share common genetic determinants, which genetic factors are specific to different subphenotypes, and the nature of the nongenetic interactions among these phenotypes.</p><p>The genetic analysis of complex traits is facilitated by the study of inbred line crosses using animal models, typically rodents. The genetically varied progeny from a cross can be reared in a controlled environment, and multiple quantitative phenotypes that are relevant to a disease outcome can be measured in individual animals [<xref rid="pgen-0020114-b003" ref-type="bibr">3</xref>]. Genetically randomized experimental populations that segregate naturally occurring allelic variants can provide a basis for the inference of networks of causal associations among genetic loci, physiological phenotypes, and disease states. The inbred cross experimental design provides a setting in which the direction of causality, from genes to phenotypes, can be inferred unambiguously. The randomization of genetic variants that occurs during meiosis provides a setting that is analogous to a randomized experimental design and thus admits causal inferences [<xref rid="pgen-0020114-b004" ref-type="bibr">4</xref>], consistent with our intuition that variation in genetic factors causes phenotypic variation.</p><p>Multivariate analysis of quantitative traits can be used to investigate the structure of a genetic system that includes allelic variation at multiple loci, intermediate phenotypes, and disease states [<xref rid="pgen-0020114-b005" ref-type="bibr">5</xref>]. Jiang and Zeng [<xref rid="pgen-0020114-b006" ref-type="bibr">6</xref>] proposed a method for quantitative trait locus (QTL) detection based on a multivariate normal model with unconstrained covariance structure. Alternatively, dimension reduction techniques, such as principal component analysis, can be applied to sets of correlated traits [<xref rid="pgen-0020114-b007" ref-type="bibr">7</xref>]. Multivariate QTL analyses can provide enhanced power and resolution in QTL mapping when traits are highly correlated and share common genetic determinants [<xref rid="pgen-0020114-b008" ref-type="bibr">8</xref>]. However, neither QTL analysis nor dimension reduction techniques provide insight into the relationships among the phenotypes or the differential effects of the genetic loci. Mapping studies that investigate clusters of related phenotypes often reveal a network of genetic effects, in which each phenotype is influenced by multiple loci (heterogeneity) and different phenotypes share one or more loci in common (pleiotropy) [<xref rid="pgen-0020114-b001" ref-type="bibr">1</xref>,<xref rid="pgen-0020114-b005" ref-type="bibr">5</xref>]. The complexity of observed QTL networks will vary depending on the traits and the power of the study design. It is also likely that physiological interactions independent of genetic factors may result in correlated phenotypic responses [<xref rid="pgen-0020114-b002" ref-type="bibr">2</xref>]. For example, heart rate and blood pressure will covary when measured simultaneously on the same animal. The methods described here represent a next step in the analysis of the genetic architecture of multiple traits and QTLs that have been detected using either univariate or multivariate genome scans.</p><p>Structural equation models (SEMs), also known as path models [<xref rid="pgen-0020114-b009" ref-type="bibr">9</xref>], are related to Bayesian networks [<xref rid="pgen-0020114-b010" ref-type="bibr">10</xref>–<xref rid="pgen-0020114-b012" ref-type="bibr">12</xref>]. In each of these approaches, we can represent the model structure as a directed graph in which measured variables are represented as nodes and causal relationships are represented as directed edges between the nodes. Multivariate probability distributions are defined by the conditional dependencies among variables represented in the graphical model [<xref rid="pgen-0020114-b012" ref-type="bibr">12</xref>]. Bayesian networks emphasize probabilistic relationships among discrete variables, whereas SEMs emphasize the correlation structure of continuously variable data. SEMs are an extension of standard multiple regression techniques. They impose structure on the expected correlation through a system of linear equations that define the causal relationships among measured variables in a system. Covariates—factors such as sex, batch, or litter that are external to direct genetic causality but introduce variations in phenotypes of interest—can be incorporated into SEM analysis [<xref rid="pgen-0020114-b013" ref-type="bibr">13</xref>]. SEM is essentially a hierarchical system of regression relationships in which any given variable may be both a response and a predictor.</p><p>Herein, we propose a SEM approach to analyze complex genetic systems using mouse inbred crosses. SEM methodology has been applied in several studies of human inheritance with the aim of improving QTL detection [<xref rid="pgen-0020114-b014" ref-type="bibr">14</xref>–<xref rid="pgen-0020114-b019" ref-type="bibr">19</xref>]. The approach presented here does not explicitly use structural modeling for detection. Instead, we focus on SEM as a descriptive and inferential tool to investigate the simultaneous effects of QTLs on multiple phenotypes and interactions among those phenotypes. The relationships among QTLs and phenotypes can be tested and quantified to establish the nature of genetic heterogeneity, pleiotropy, and the role of physiological pathways in mediating genetic effects [<xref rid="pgen-0020114-b020" ref-type="bibr">20</xref>,<xref rid="pgen-0020114-b021" ref-type="bibr">21</xref>]. We illustrate the approach with an analysis of two phenotypes related to obesity: one, an SM/J × NZB/B1NJ intercross [<xref rid="pgen-0020114-b022" ref-type="bibr">22</xref>], and two, bone phenotypes in a NZB/B1NJ × RF/J intercross [<xref rid="pgen-0020114-b023" ref-type="bibr">23</xref>]. For brevity, these strains will be denoted SM, NZB, and RF.</p></sec><sec id="s2"><title>Materials and Methods</title><sec id="s2a"><title>Mouse Intercross Populations</title><p>The SM × NZB intercross population of 260 female and 253 male mice was raised on an atherogenic diet for 16 wk starting at 8 wk of age. At 24 wk we obtained total body weight and weights of the inguinal, gonadal, peritoneal, and mesenteric fat pads. Lean body weight was computed by subtracting the total of the fat pad weights from the body weight. In the analyses described here, all traits were square-root transformed to obtain the best linear relationships. Additional information on husbandry, phenotyping, and genotyping of this cross can be found in Stylianou et al. [<xref rid="pgen-0020114-b022" ref-type="bibr">22</xref>].</p><p>The NZB × RF intercross population consists of 661 female mice raised on standard (4% fat) diet. At 10 wk of age, femurs were isolated and their geometric properties were determined by peripheral quantitative computed tomography. We consider body weight and two bone geometry traits, femur length and the periosteal circumference of the femur (PCIR). All traits were log transformed to obtain the best linear relationships. Additional information on this cross can be found in Wergedal et al. [<xref rid="pgen-0020114-b023" ref-type="bibr">23</xref>].</p><p>The Institutional Animal Care and Use Committee of The Jackson Laboratory approved all experimental protocols. All data used in this study are available at <ext-link ext-link-type="uri" xlink:href="http://www.jax.org/staff/churchill/labsite/datasets/qtl/qtlarchive">http://www.jax.org/staff/churchill/labsite/datasets/qtl/qtlarchive</ext-link>.</p></sec><sec id="s2b"><title>Structural Equation Models</title><p>In structural equation modeling, variables are standardized by centering on sample means. Thus the variances and covariances are the parameters of interest. The key idea behind SEM is that causal relationships among the variables determine the expected pattern of correlations [<xref rid="pgen-0020114-b024" ref-type="bibr">24</xref>]. A SEM represents causal relationships among measured and latent variables both graphically and as a set of linear equations, the <italic>structural equations</italic>, that define the interactions among the variables. In the graphical representation of a SEM, a variable with an arrow pointing to it is termed an <italic>endogenous</italic> variable, which is similar to a <italic>response</italic> or <italic>dependent</italic> variable in regression terminology. An endogenous variable is causally affected by the state of at least one other variable in the model. A variable without an arrow pointing to it is termed an <italic>exogenous</italic> variable, which is similar to an <italic>independent</italic> variable or <italic>predictor</italic> in regression terminology. Exogenous variables are upstream of all causal effects in the model. Variables that are connected through a single edge in the graphical model are said to have <italic>direct path</italic>. An effect that is mediated through other measured variables is represented by an <italic>indirect path</italic> with more than one edge. Variables may be connected by more than one path. The sign and magnitude of the direct effects in the graphical model are represented by <italic>path coefficients</italic> in the structural equations.</p><p>The endogenous variables in a SEM are assumed to follow a multivariate normal distribution, while exogenous variables can be either continuous or categorical. The maximized log likelihood function takes the form [<xref rid="pgen-0020114-b025" ref-type="bibr">25</xref>,<xref rid="pgen-0020114-b026" ref-type="bibr">26</xref>]: <disp-formula id="pgen-0020114-e001"><graphic xlink:href="pgen.0020114.e001.jpg" position="anchor" mimetype="image"/></disp-formula>where <italic>θ</italic> is a vector of model parameters that includes path coefficients, variances of all exogenous variables, and covariances for all pairs of exogenous variables; <italic>p</italic> is the number of variables included in the model; Σ is the predicted covariance matrix and is implicitly a function of the model parameters; <italic>S</italic> is the observed covariance matrix; and \|<italic>X</italic>\| denotes the determinant of a matrix <italic>X</italic>. The number of unique elements in a covariance matrix is <italic>p</italic>(<italic>p</italic> + 1)/2 due to symmetry, and the number of model parameters is denoted by <italic>q</italic>. Intuitively, the parameter values are chosen to give a predicted covariance matrix that is as similar as possible to the observed covariance matrix, subject to the constraints imposed by the structural equations. A critical quantity in determining our ability to fit and assess a model is the residual degrees of freedom (df), where df = <italic>p</italic>(<italic>p</italic> + 1)/2 − <italic>q</italic>. If the residual df are less than 1, the model has as many or more parameters than data points and thus the goodness-of-fit cannot be assessed. </p></sec><sec id="s2c"><title>Causal Inference</title><p>To illustrate causal inference in the context of QTL mapping, consider the possible relationships between a genetic locus, Q, and two correlated traits, A and B, as in <xref ref-type="fig" rid="pgen-0020114-g001">Figure 1</xref>. In each of the models M1–M9, there is a direct causal connection between A and B that represents a physiological interaction. In model M10 the correlation between A and B is indirect—a result of the shared QTL. Directed paths represent causal effects, and bidirectional paths represent undirected associations. The latter may indicate the presence of an unobserved factor that influences both traits. The QTL may be pleiotropic, with direct effects on both traits, as in models M1–M3 and M10. Alternatively, the QTL may have a direct effect on only one trait.</p><fig id="pgen-0020114-g001" position="float"><label>Figure 1</label><caption><title>Causal Relationships among a QTL and Two Phenotypes</title><p>Single-headed arrows indicate causal effects and doubled-headed arrows indicate unresolved associations between the two phenotypes. Phenotypes are indicated by A and B; QTL by Q.</p></caption><graphic xlink:href="pgen.0020114.g001"/></fig><p>Causal effects from a QTL to a phenotype have a defined direction. The causal inference follows as a consequence of meiotic randomization of genetic factors in the inbred line cross. The causal inference can be extended to include the relationships among phenotypes by considering both the unadjusted and partial correlation relationships among the variables. For example, in model M7, Q and A are unconditionally independent. However, conditioning on B will result in a nonzero partial correlation. By contrast, in M8, Q and A will be unconditionally correlated. Conditioning on B breaks the causal chain from Q to A, and their partial correlation will be zero. When all three variables are causally connected, as in models M1–M3, the raw and partial correlations will all be nonzero, but they will change in magnitude depending on the signs of the path coefficients. The relationships among the various raw and partial correlations are subject to statistical fluctuations but they can be captured in the log of odds ratio (LOD) scores, as described below, and these form the basis for constructing multilocus SEMs.</p></sec><sec id="s2d"><title>Development of a Structural Equation Model</title><p>The process of developing a SEM for genetic mapping data is described in five steps, below. Steps 1 and 2 involve QTL detection and they employ standard QTL detection methods and serve the purpose of identifying the variables that will be used in the structural modeling. Once the QTL have been identified, their genotypes are imputed [<xref rid="pgen-0020114-b027" ref-type="bibr">27</xref>] and the covariance matrix of all traits and QTLs are computed by averaging over imputations. This estimated covariance matrix is analyzed in steps 3 through 5 to build, assess, and revise the SEM. Some iteration among these steps may be required to arrive at a suitable model. However, extensive model refinement may lead to a model that fits the existing data but does not generalize well (i.e., overfitting) and is not recommended.</p><sec id="s2d1"><title>Step 1. Identify QTLs for individual phenotypes.</title><p>We use genome scans to identify the genetic loci that will be included in the SEM [<xref rid="pgen-0020114-b027" ref-type="bibr">27</xref>]. A single locus genome scan is based on the linear model <disp-formula id="pgen-0020114-e002"><graphic xlink:href="pgen.0020114.e002.jpg" position="anchor" mimetype="image"/></disp-formula>where <italic>Y</italic> is a vector of trait values, <italic>β</italic> <sub>0</sub> is the population mean, <italic>Q</italic> is a vector of QTL genotypes, <italic>β</italic> <sub>1</sub> is the QTL effect, and ɛ is the residual vector. The location of the QTL is scanned over the entire genome and a LOD score (or equivalently, likelihood ratio) is used to determine if a QTL is present. If covariates, such as sex, are important predictors of the trait values, these should be included in the linear model underlying the genome scans [<xref rid="pgen-0020114-b013" ref-type="bibr">13</xref>]. In addition, the traits may be scanned using a pairwise genome scan to identify epistatic QTLs [<xref rid="pgen-0020114-b027" ref-type="bibr">27</xref>]. </p></sec><sec id="s2d2"><title>Step 2. Identify pleiotropic QTLs.</title><p>In this step we perform conditional genome scans using one trait as a covariate in the analysis of another trait. The choice of which trait(s) to use as covariates may be dictated by the known biological relationships among the traits, or it may be carried out systematically. In this setting, factors that are believed to be upstream in causal pathways should be employed as conditioning variables. The linear model used in the conditional genome scans is <disp-formula id="pgen-0020114-e003"><graphic xlink:href="pgen.0020114.e003.jpg" position="anchor" mimetype="image"/></disp-formula>where <italic>X</italic> is the conditioning variable and <italic>β</italic> <sub>2</sub> is the effect of <italic>X</italic> on the response <italic>Y,</italic> adjusted for the <italic>Q</italic> effect. </p><p>Comparison of the unconditioned and conditioned scans may reveal a substantial change in the LOD score at a locus. If the change (ΔLOD) is large in absolute value, this suggests that the variable <italic>X</italic> is causally connected to <italic>Q</italic> and to <italic>Y</italic>. We use a critical value of 2.0 as a guideline, corresponding to a 0.05 type I error rate based on simulations in which the conditioning variable is unrelated to the response. The significance of these edges will be evaluated further in subsequent steps.</p><p>The direction of change in the LOD score upon adjustment for a covariate depends on the signs of QTL effects on the trait and on the covariate, as well as the sign of the path coefficient connecting the trait and the covariate. When both the direct and indirect paths share the same sign, the LOD score in the conditional scan will be smaller than the unconditional LOD score. When the direct path and the indirect path have opposite effects, the conditional LOD score will be greater than the unconditional LOD score. In this case, the direct and indirect effects cancel one another and the QTL effect is reduced in the unconditional scan. It follows that changes in the LOD upon conditioning can be used to infer the sign of QTL effects along different paths.</p></sec><sec id="s2d3"><title>Step 3. Define an initial path model.</title><p>In the graphical SEM, each measured trait is represented as a node; QTLs identified in steps 1 and 2 are also included as nodes. Edges should be directed from the QTL nodes to the corresponding traits. When a significant ΔLOD value is observed, edges should be directed from the QTL to each of the traits, and an edge from the conditioning trait to the response should also be added. The significance and causal direction of these edges should be examined in model refinement (step 5).</p><p>In many instances, a SEM that includes only measured variables will be sufficient. However, latent variables [<xref rid="pgen-0020114-b028" ref-type="bibr">28</xref>] representing unmeasured or hypothetical quantities that can be inferred from other measured variables, may be incorporated in a SEM. Addition of a latent variable can effectively reduce the dimensionality of the data when several highly correlated variables are being influenced by an underlying quantity that is not directly observed.</p><sec id="s2d4"><title>Step 4. Assessment of the model.</title><p>This step involves a comparison of the predicted and observed covariance matrices, t-tests for individual path coefficients, and consideration of other model diagnostics. The goodness-of-fit test statistic is (<italic>N</italic> − 1)<italic>F(θ),</italic> where <italic>N</italic> is the sample size and <italic>F(θ)</italic> is defined in Equation 1. It follows a χ<sup>2</sup> distribution with df <italic>p</italic>(<italic>p</italic> + 1)/2 − <italic>q</italic>. Significant values (<italic>p</italic> < 0.05) indicate that the model does not provide a good fit to the data. Each path coefficient should be individually significant (<italic>p</italic> < 0.05) using a t-test (Wald's test). The maximum standardized residual difference between the observed and predicted covariance matrices should be small. The root mean square error of approximation measures the lack of fit of the model to the theoretic population covariance matrix and values of 0.05 or less indicate an acceptable fit. A model that does not provide an accurate prediction of the observed covariance should be refined.</p></sec><sec id="s2d5"><title>Step 5. Refine the model.</title><p>Model refinement involves the proposal and assessment of a new model. The procedures used to suggest a new model include adding a new path to the initial model, removing a path, or reversing the causal direction of a path [<xref rid="pgen-0020114-b009" ref-type="bibr">9</xref>]. When a new path is added to an existing model it should result in a significant change in the goodness of fit statistic (> 3.84 for 1 df). This is the likelihood ratio test. When an existing path is removed from the model, the change in the goodness of fit statistic should be nonsignificant. Particular attention should be paid to the edges between phenotypes. Whereas the direction of causality from QTL to a trait is clear, the direction of causal connections between traits should be carefully examined by exploring all possible alternatives, as illustrated below.</p><p>Additionally, one may consider model selection methods. For example, Akaike's information criterion (AIC) [<xref rid="pgen-0020114-b029" ref-type="bibr">29</xref>], provides a penalized likelihood statistic for model comparison. CAIC is an adjustment of AIC to correct for bias in small samples and is recommended when the ratio of the sample size to the number of estimable parameters is less than 40 [<xref rid="pgen-0020114-b030" ref-type="bibr">30</xref>]. In our examples this ratio is around 10. A model with the smallest value of AIC or CAIC among several candidate models is preferred. The expected value of the cross-validation index (ECVI) estimates the overall error and predictive validity of a model [<xref rid="pgen-0020114-b031" ref-type="bibr">31</xref>]. ECVI values may be compared among several models. An interval estimate is helpful to avoid paying too much attention to small differences.</p><p>Model refinement and assessment (steps 4 and 5) are often carried out iteratively. A final model should meet several standards [<xref rid="pgen-0020114-b009" ref-type="bibr">9</xref>]: (1) it should be identified or overidentified with at least 1 residual df; (2) the <italic>p</italic>-value associated with the goodness-of-fit test should be greater than 0.05; (3) the largest standardized residual should not exceed 2.0 in absolute value; (4) individual path coefficients should be significantly different from zero based on the t-test; (5) standardized path coefficients should not be trivial (absolute values exceed 0.05); and (6) a substantial proportion of phenotypic variance of the endogenous variables should be explained by the model. If all of these standards are met, we may conclude that the model provides a reasonable description of the data.</p></sec></sec></sec><sec id="s2e"><title>Modeling Genetic Effects</title><p>In the context of an intercross (F2) population, each QTL has three possible genotypic states. We can assume intralocus additivity by encoding genotypes as 0, 1, or 2, and treating these scores as a continuous variable in the SEM. Alternatively, the genetic effect is represented as a pair of 0,1 (dummy) variables. The most widely used parameterization, following Cockerham [<xref rid="pgen-0020114-b032" ref-type="bibr">32</xref>], partitions the genetic effects into additive (A) and dominant (D) components. The A and D components are orthogonal to one another and thus we can fix their covariances in the SEM to be zero. In the model-fitting and refinement steps, we always retain or drop both components as a unit, even if only one component is significant. Epistatic interactions for pairs of QTLs can be partitioned into four components [<xref rid="pgen-0020114-b032" ref-type="bibr">32</xref>], additive × additive (AA), additive × dominant (AD), dominant × additive (DA) and dominant × dominant (DD). These are also orthogonal and are treated as a single unit in the model. If an epistatic term is included in the model, we retain the main effect terms, regardless of their marginal significance, to ensure that the model is interpretable. Although we considered epistatic terms in the examples below, we found that the added explanatory power was not sufficient to justify the additional free parameters in the model. With larger sample sizes it may become practical to include epistatic effects in a SEM.</p></sec><sec id="s2f"><title>Path Analysis</title><p>One important feature of SEM is that direct and indirect effects of a QTL on a trait can be distinguished, and the relative strengths of effects along different paths can be calculated and compared [<xref rid="pgen-0020114-b020" ref-type="bibr">20</xref>,<xref rid="pgen-0020114-b021" ref-type="bibr">21</xref>]. The effect of a direct path from a QTL to a trait is represented by the path coefficient. Path coefficients are typically standardized relative to the residual variance of the endogenous variable to which they are directed. Thus all error terms in a standardized model have variance 1, and path coefficients are expressed in standard deviation units. In this way all path coefficients are directly comparable and are independent of the original measurement scale. The effect of an indirect path is the product of all the standardized path coefficients (including + and − signs) along this path. The total effect of the QTL on a trait is the sum of the effects along all the direct and indirect paths connecting the two variables.</p><p>We note that the sign of a path coefficient from a QTL to a trait is determined by the choice of the reference genotype (encoded as zero). If the effect of an allelic substitution away from the reference is to increase the mean trait value, the sign of the path coefficient is positive. If allelic substitution away from the reference causes the trait value to decrease, the sign is negative. In the examples below, the homozygous NZB genotype is our reference. The same interpretation applies to path coefficients for categorical covariates such as sex. In the models below we have used female as the reference.</p></sec><sec id="s2g"><title>Computing</title><p>Genome scans were carried out in the MATLAB software environment (MathWorks, Natick, Massachusetts, United States) using the pseudomarker package (v2.02) [<xref rid="pgen-0020114-b027" ref-type="bibr">27</xref>]. Scans were conducted at 2 cM resolution using the imputation method with 256 imputations. Significance of QTLs was assessed using permutation analysis [<xref rid="pgen-0020114-b033" ref-type="bibr">33</xref>] with 1,000 permutations. Genome-wide significance is defined as the 95th percentile of the maximum LOD score and suggestive QTLs exceed the 37th percentile. Structural equation modeling analysis was carried out using the PROC CALIS procedure in the SAS software package (SAS, Cary, North Carolina, United States) or in AMOS 6.0 (SPSS, Chicago, Illinois, United States). QTLs were represented by the marker nearest to the LOD peak, and missing genotypes (< 5%) were inferred from flanking marker data.</p></sec></sec><sec id="s3"><title>Results</title><sec id="s3a"><title>Measuring Adiposity</title><p>We first consider the problem of defining a measure of adiposity. Fat pad weights and lean body weight are positively correlated in the SM × NZB cross population. Fat pad weight by itself is inadequate as a measure of obesity. Two animals with similar fat pad weights may have different body size such that one animal is considered obese and the other not. Thus it is important to consider the fat pad weight relative to the size of the animal. In our analysis we used the lean body weight as surrogate for size. The lean body weight is computed by subtracting the weights of the four major fat pads from the total body weight and thus will include a small contribution from other adipose tissues.</p><p>There are two common approaches to defining a relative measure. First, one could compute an <italic>adiposity index</italic> as the ratio of fat pad weight to lean body weight. Second, one could regress fat pad weight on lean body weight and consider the deviation from the regression line (the residual) as a measure of adiposity. Although the regression approach is clearly preferred, ratio standardization is still widely used [<xref rid="pgen-0020114-b034" ref-type="bibr">34</xref>,<xref rid="pgen-0020114-b035" ref-type="bibr">35</xref>]. To see why this is so, consider the implications of using each measure. The ratio standard assumes a proportional relationship <italic>(y</italic> = <italic>bx)</italic> between the variables, which implies that the regression line relating these traits should pass through the origin. One might argue that a hypothetical animal with zero lean body mass should also have a fat pad mass of zero. However, the argument is fallacious, as it extrapolates beyond the range of the data over which the linear approximation is valid. Regression standardization allows one additional df in the form of an intercept term <italic>(y</italic> = <italic>a</italic> + <italic>bx)</italic>. The relationship of mesenteric fat pad weight to lean body weight is shown graphically in <xref ref-type="fig" rid="pgen-0020114-g002">Figure 2</xref>. The regression line and the ratio line meet at the mean of each trait but the two diverge away from the mean point. The regression line fits the data well over its entire range. For either standard, adiposity is measured as the deviation from the fitted line. In this case, the ratio standard is severely biased. An animal with higher than average lean body weight and normal adiposity according to the regression standard would be considered obese by the ratio standard. The converse would be true for an animal with lower than average lean body weight. Therefore we conclude that the regression method provides a more appropriate adjustment of fat pad weight for lean body weight and is to be preferred as a measure of adiposity. SEMs generalize the regression model and thus provide a regression standardization.</p><fig id="pgen-0020114-g002" position="float"><label>Figure 2</label><caption><title>Relationship of Mesenteric Fat Pad Weight to Lean Body Weight for Female and Male Animals</title><p>Both phenotypes are square root transformed. The dotted line indicates the ratio standard of constant adiposity index. The solid line is the regression of mesenteric fat pad weight on lean body weight.</p></caption><graphic xlink:href="pgen.0020114.g002"/></fig></sec><sec id="s3b"><title>Body Size and Fat Pad QTL</title><p>A genome scan of the mesenteric fat pad weight (with sex as an additive covariate) reveals significant QTLs on Chromosomes 1, 17, and 19 (<xref ref-type="fig" rid="pgen-0020114-g003">Figure 3</xref>A). A second genome scan for mesenteric fat pad weight, using both sex and lean body weight as additive covariates, shows suggestive QTL peaks on Chromosomes 2, 4, and 10 (<xref ref-type="fig" rid="pgen-0020114-g003">Figure 3</xref>B). The differences between these two scans are dramatic, with both upward and downward changes in the LOD scores (<xref ref-type="fig" rid="pgen-0020114-g003">Figure 3</xref>C). The peak LOD score decreases after adjustment on Chromosomes 1, 17, and 19, and it increases after adjustment on Chromosomes 2 and 10. A genome scan of the lean body weight (adjusted for sex) reveals QTLs on Chromosomes 1, 2, 5, 6, 12, 15, 17, and 19 (<xref ref-type="fig" rid="pgen-0020114-g003">Figure 3</xref>D). The substantial overlap with QTLs for fat pad weight suggests that pleiotropic effects are an important component of the genetic architecture. One goal of structural equation modeling is to translate the observed changes in LOD scores following adjustment for a covariate into an interpretation of the nature of these pleiotropic effects.</p><fig id="pgen-0020114-g003" position="float"><label>Figure 3</label><caption><title>Genome-Wide Scans for Mesenteric Fat Pad Weight at 2 cM Resolution</title><p>Genome scans shown are (A) mesenteric fat pad weight with sex as an additive covariate; (B) mesenteric fat pad weight with sex and lean body weight as additive covariates; (C) difference in LOD scores between scans in (A) and (B); and (D) lean body weight with sex as an additive covariate. LBWT, lean body weight; MES, mesenteric fat pad weight.</p></caption><graphic xlink:href="pgen.0020114.g003"/></fig><p>We note here that a genome scan of the adiposity index (unpublished data) reveals only two significant QTLs, one on Chromosome 19 and the other on Chromosome 1, and there are no suggestive QTLs. As indicated below, these two loci primarily affect the overall body size and are not specifically impacting adiposity. Genome scans for all four fat pad traits are provided in <xref ref-type="supplementary-material" rid="pgen-0020114-sg001">Figure S1</xref>.</p></sec><sec id="s3c"><title>A SEM of Mesenteric Fat Pad Weight</title><p>As a starting point for the development of a SEM we consider all of the QTLs that are significant (<italic>p</italic> < 0.05, genome-wide adjusted) in at least one of the genome scans and any suggestive QTLs that have significant ΔLOD values. Directed edges from each QTL to the corresponding trait are included in the graphical SEM. QTLs with a ΔLOD exceeding 2 in absolute value are connected with directed edge to both traits. In addition, a directed edge is tentatively connected from lean body weight to the fat pad weight. We used an additive model of the genetic effects. This initial model is assessed and refined, using the steps described in Materials and Methods, to arrive at the model shown in <xref ref-type="fig" rid="pgen-0020114-g004">Figure 4</xref>. In this case, no refinements were made to the initial model. The path coefficients are all significantly different from zero (<xref ref-type="table" rid="pgen-0020114-t001">Table 1</xref>), the proportion of variance explained by the model is 46.5% for mesenteric fat pad mass and 54.4% for lean body weight, and the maximized standard residual is 0.93. Several goodness-of-fit statistics for model M1 are shown in <xref ref-type="table" rid="pgen-0020114-t002">Table 2</xref>. These measures all indicate that fit of the model is acceptable.</p><fig id="pgen-0020114-g004" position="float"><label>Figure 4</label><caption><title>Graphical Representation of the SEM for Mesenteric Fat Pad Weight</title><p>Genetic loci are indicated by Q followed by the chromosome number, and @ followed by the cM position of the LOD peak. Single-headed arrows indicate causal paths, and the thickness of each arrow is proportion to the effect size (path coefficient). A negative sign from a QTL to a trait indicates that the NZB allele is associated with high trait values. E1 and E2 denote unobserved residual error.</p></caption><graphic xlink:href="pgen.0020114.g004"/></fig><table-wrap id="pgen-0020114-t001" content-type="1col" position="float"><label>Table 1</label><caption><p>Structural Equations of the Mesenteric Fat Pad Model</p></caption><graphic xlink:href="pgen.0020114.t001"/></table-wrap><table-wrap id="pgen-0020114-t002" content-type="1col" position="float"><label>Table 2</label><caption><p>Model Comparison for Mesenteric Fat Pad Mass and Lean Body Weight</p></caption><graphic xlink:href="pgen.0020114.t002"/></table-wrap><p>In order to assess the direction of causality between lean body weight and mesenteric fat pad weight, we consider the models listed in <xref ref-type="table" rid="pgen-0020114-t002">Table 2</xref>. These models are derived from the initial model by varying the direction of the causal relationship between the traits. In general it may be necessary to allow for model refinement but none was required here. Goodness-of-fit statistics indicate that the model, M2, in which mesenteric fat weight is upstream does not fit. The overall fit of the bidirectional model, M3, is acceptable, but a t-test (<italic>t</italic> = 1.48) indicates that the path coefficient from mesenteric fat pad weight to lean body weight is nonsignificant. M1 also has the smallest CAIC value (<xref ref-type="table" rid="pgen-0020114-t002">Table 2</xref>) and is preferred over the other models. We conclude that lean body weight is causal to fat pad weight. This is consistent with our intuition that a mouse with a larger body size will tend to have larger fat pads. Caution is recommended in this interpretation, because both traits represent a composite of multiple causal effects, some of which are mediated by unobserved factors. The SEM indicates that the net effect is unidirectional.</p><p>A unique feature of SEMs is the ability to resolve the sign and magnitude of effects along multiple direct and indirect paths. Path analysis provides a detailed description of the pleiotropic effects in the system that takes into account multiple genetic and nongenetic sources of variation. For those QTLs that are associated with both mesenteric fat pad weight and lean body weight, path analysis establishes the relative contributions of the direct and indirect effects. Decomposition of QTL effects along direct and indirect paths is illustrated in <xref ref-type="table" rid="pgen-0020114-t003">Table 3</xref>, using two QTLs as examples (see <xref ref-type="supplementary-material" rid="pgen-0020114-st001">Tables S1</xref>–<xref ref-type="supplementary-material" rid="pgen-0020114-st004">S4</xref> for a complete set). The locus Q19@52 shows negative effects on mesenteric fat pad weight along both the direct and indirect paths, indicating that this locus affects mesenteric fat pad weight in the same direction through two physiological pathways. The peak LOD score in an unadjusted genome scan is a reflection of the net effect over all paths. The contribution from an indirect path can be blocked by conditioning on the intermediate variable. In this case, the result is a downward change in LOD score in the conditional scan. In contrast, the Q10@52 shows opposite effects on mesenteric fat pad weight along direct and indirect paths. Conditioning on lean body weight blocks the indirect path, resulting in an increased LOD score. In this way, changes in LOD score upon conditioning provide information about the signs of path coefficients in the SEM.</p><table-wrap id="pgen-0020114-t003" content-type="1col" position="float"><label>Table 3</label><caption><p>Path Analysis of QTL Effects</p></caption><graphic xlink:href="pgen.0020114.t003"/></table-wrap><p>In the SEM for mesenteric fat pad weight (<xref ref-type="fig" rid="pgen-0020114-g004">Figure 4</xref> and <xref ref-type="table" rid="pgen-0020114-t001">Table 1</xref>) we can identify three distinct classes of QTL effects. The QTLs on Chromosomes 5, 6, and 12 have direct effects on lean body weight and only indirect effects on the fat pad weight. The QTLs on Chromosomes 1, 17, and 19 have pleiotropic “body size” effects. NZB alleles at these loci are associated with increased lean body weight and increased fat pad weight. The QTLs on Chromosomes 2 and 10 may be interpreted as “adiposity” loci. At these loci, SM alleles are associated with smaller lean body weight and larger fat pad weights. The largest influence on mesenteric fat pad weight is through its positive correlation with lean body weight. Females have relatively larger fat pad weights. The largest genetics effects are on Chromosomes 2 and 10, where SM alleles contribute to increased adiposity. The largest single effect on lean body weight is sex, but the total effect of multiple genetic loci (with all high alleles contributed by NZB) is greater still.</p></sec><sec id="s3d"><title>Modeling Adiposity and Body Weight</title><p>Univariate and multivariate genome scans [<xref rid="pgen-0020114-b006" ref-type="bibr">6</xref>] were performed for the four fat pad traits. The multivariate scan (<xref ref-type="fig" rid="pgen-0020114-g005">Figure 5</xref>A) detected all of the QTLs that were detected by scanning each fat pad trait one at a time (<xref ref-type="supplementary-material" rid="pgen-0020114-sg001">Figure S1</xref>), and a new QTL was detected on Chromosome 14. The QTLs on Chromosomes 1, 2, 5, 6, 12, 17, and 19 show significant changes in LOD scores (ΔLOD > 2), when conditioning on lean body weight (<xref ref-type="fig" rid="pgen-0020114-g005">Figure 5</xref>C). In our initial SEM, we included all QTLs that were significant in any of the genome scans and all suggestive QTLs that showed a significant (> 2) change in LOD score between these two multivariate genome scans.</p><fig id="pgen-0020114-g005" position="float"><label>Figure 5</label><caption><title>Genome-Wide Scans for Multiple Fat Pad Traits at 2 cM Resolution</title><p>Genome scans shown are (A) the four fat pad traits (inguinal, gonadal, peritoneal, and mesenteric fat pad weight) with sex as an additive covariate; (B) the four fat pad traits with sex and lean body weight as additive covariates; and (C) the difference in LOD scores between scans in (A) and (B). FP, the four fat pad traits; LBWT, lean body weight.</p></caption><graphic xlink:href="pgen.0020114.g005"/></fig><p>Correlations among the fat pad weights were all about 0.7, thus multicollinearity is a concern for developing a SEM. To address this, we introduced a latent variable to capture the shared variation among the fat pad weights and refer to this latent variable as “adiposity.” Variance in any of the fat pad traits that is unrelated to adiposity is captured by the residual error variances. We formulated an initial model using only main effect terms and additive genetic effects as described in Materials and Methods. The population size of 531 limits the size of the models that we can consider. A final model, obtained after a few iterations of model refinement, is shown in <xref ref-type="fig" rid="pgen-0020114-g006">Figure 6</xref>. The model selection and goodness of fit statistics are summarized in <xref ref-type="table" rid="pgen-0020114-t004">Table 4</xref>.</p><fig id="pgen-0020114-g006" position="float"><label>Figure 6</label><caption><title>Structural Equation Model for Adiposity and Lean Body weight</title><p>The four fat pad traits are gonadal (GON); inguinal (ING); mesenteric (MES); and peritoneal (PERI). Single-headed arrows indicate causal paths, and the thickness of each arrow is proportional to the effect sizes. Doubled-headed arrows denote unresolved covariance. The boxes indicate measured traits or QTL and the oval denotes a latent variable. E1, E2, E3, E4, and E5 denote unobserved residual error. The negative sign from a QTL to a trait indicates that the NZB allele is associated with high trait values.</p></caption><graphic xlink:href="pgen.0020114.g006"/></fig><table-wrap id="pgen-0020114-t004" content-type="2col" position="float"><label>Table 4</label><caption><p>Model Assessment and Path Coefficients for the Adiposity SEM</p></caption><graphic xlink:href="pgen.0020114.t004"/></table-wrap><p>Several conclusions are available from inspection of the graphical model in <xref ref-type="fig" rid="pgen-0020114-g006">Figure 6</xref>. For example, we see that female mice have lower body weight, higher adiposity, larger gonadal fat pads, and smaller peritoneal fat pads compared to males. SM alleles on Chromosomes 17 and 19 are associated with lower adiposity and lower body weight compared to NZB alleles. The Chromosome 2 QTL is associated with higher adiposity and lower lean body weight. QTLs specific to adiposity or fat pad traits include Q4@46, Q5@24, Q12@14, and Q14@4. QTLs specific to lean body weight include Q5@46 and Q12@54. Each QTL contributes to overall body composition with a unique pattern of effects.</p></sec><sec id="s3e"><title>Modeling Bone Geometry and Body Weight</title><p>Body weight and bone geometry phenotypes are known to be associated but the causal relationships among them are largely unknown [<xref rid="pgen-0020114-b035" ref-type="bibr">35</xref>]. Here we look at bone geometry data from a NZB × RF intercross population [<xref rid="pgen-0020114-b023" ref-type="bibr">23</xref>] and consider the relationships among femur length, midshaft PCIR, and body weight. QTLs associated with these traits have been described previously [<xref rid="pgen-0020114-b023" ref-type="bibr">23</xref>]. We employed an additive genetic model in the SEM because all QTL effects showed intralocus additivity. We developed an initial SEM (<xref ref-type="fig" rid="pgen-0020114-g007">Figure 7</xref> and <xref ref-type="table" rid="pgen-0020114-t005">Table 5</xref>) following the steps of model formulation, assessment, and refinement described in Materials and Methods. In order to resolve the causal relationships among the three phenotypes, we examined the complete set of 11 models listed in <xref ref-type="fig" rid="pgen-0020114-g008">Figure 8</xref>. Models M1, M9, M10, and M11 each provide a close fit to the data, but t-tests for the extra path coefficients in models M9, M10, and M11 that are not found in model M1 are all nonsignificant (<italic>t</italic> = 1.68, −0.43, and 0.65, respectively). In addition, M1 has the smallest CAIC value relative to other models. We conclude that model M1 provides the best description of these data.</p><fig id="pgen-0020114-g007" position="float"><label>Figure 7</label><caption><title>Structural Equation Model for Bone Geometry</title><p>Genetic effects have been grouped. Sign and magnitude of path coefficients can be found in <xref ref-type="table" rid="pgen-0020114-t005">Table 5</xref>. Group Q1 includes loci with effects that are specific to PCIR (Q4@66, Q5@84, Q6@32, Q7@50, and Q11@68). Group Q2 includes loci have pleiotropic effects on PCIR and BWT (Q19@50, Q1@20, and Q8@0). Group Q3 includes loci with pleiotropic effects on PCIR and FLEN (Q10@64, Q5@52, Q15@10, and Q12@56). Group Q4 loci are pleiotropic loci that affect all three traits (Q12@2, Q2@66, and Q3@30). E1, E2, and E3 denote <italic>N</italic>(0,1) residual error.</p></caption><graphic xlink:href="pgen.0020114.g007"/></fig><table-wrap id="pgen-0020114-t005" content-type="1col" position="float"><label>Table 5</label><caption><p>Structural Equations for the Bone Geometry Model</p></caption><graphic xlink:href="pgen.0020114.t005"/></table-wrap><fig id="pgen-0020114-g008" position="float"><label>Figure 8</label><caption><title>Model Comparison for the Bone Geometry Data</title><p>Model comparisons for the bone geometry data were derived from the model in <xref ref-type="fig" rid="pgen-0020114-g007">Figure 7</xref> by varying the relationships among body weight, femur length, and periostial circumference.</p></caption><graphic xlink:href="pgen.0020114.g008"/></fig><p>The graphical SEM is shown in <xref ref-type="fig" rid="pgen-0020114-g007">Figure 7</xref>. Path coefficients and t-statistics are summarized in <xref ref-type="table" rid="pgen-0020114-t005">Table 5</xref>. The model explains 67.3% of the variance in PCIR, 29.5% of the variance in body weight, and 11.9% of the variance in femur length. The QTLs in group Q1 are specific to PCIR. The contributions are balanced in that Q5@84 and Q11@68 contribute high alleles from NZB, and Q7@50 and Q4@66 contribute low alleles from NZB. The QTLs in group Q2 affect both PCIR and body weight. Path coefficients indicate that Q12@2 and Q19@50 are primarily body weight QTLs. Other members of Q2 (Q2@66, Q3@30, and Q8@0) show opposing effects on body weight and PCIR. Causal connections among the traits are consistent with our prior expectations for this study. Muscle mass, a major determinant of body weight, is associated with the length of the femur, and a mouse with greater body weight will tend to have thicker bones. Again, we should use caution in these interpretations as there is certainly some cross-talk among these traits as well as unobserved factors that could influence the relationships among the measured variables. The SEM describes the net effects of many factors.</p></sec></sec><sec id="s4"><title>Discussion</title><p>In experimental crosses, meiosis serves as a randomization mechanism that distributes naturally occurring genetic variation in a combinatorial fashion among a set of cross progeny. Genetically randomized populations share the properties of statistically designed experiments that provide a basis for causal inference. This is consistent with the notion that causation flows from genes to phenotypes. We propose that the inference of causal direction can be extended to include relationships among phenotypes and demonstrate that causal hypotheses can be tested. When models are nested, likelihood ratio tests can be applied. For non-nested sets of models, penalized likelihood methods such as AIC or CAIC can be used to identify good models. However, the credibility of any causal hypothesis must be judged by biological standards and not solely on statistical evidence. From this perspective, we view SEM as a device for generating causal hypotheses to be tested by subsequent experimentation.</p><p>As descriptive models, SEMs are useful for identifying the nature of pleiotropic QTL effects. The phenotypes that we choose to measure and the QTLs that we detect will rarely, if ever, occur in one-to-one correspondence. Allelic variation at a locus will often impact multiple traits, and these effects may be mediated through multiple physiological pathways. SEMs provide a quantitative description of the entire system of QTLs and phenotypes. In our analysis of the SM × NZB fat pad data we identified distinct classes of QTLs that affect body size and adiposity. The SEM analysis reveals a shortcoming of standardized variables, such body mass index and adiposity index, that are widely used in obesity research. This suggests that a reexamination of obesity QTLs in the literature may be worthwhile. The SEM is able to distinguish QTLs that affect adiposity from those that affect body size, whereas standardized measures could confound these two distinct types of effects.</p><p>Specification of a good initial model is perhaps the most important step in structural equation modeling of genetic data. We have developed a strategy to specify the initial model based on genome-wide QTL scans with conditioning on intermediate phenotypes. For the data examined here, these initial models fit reasonably well and final models were obtained with minor or no refinements. Our model development strategy incorporates elements of both exploratory and confirmatory SEM, because the initial model structure has been shaped by considering the same data that are used to assess and refine the SEM. Goodness-of-fit criteria are most informative regarding a lack of fit, and are subject to the error of overfitting. In the models presented here, most of the residual df derive from the known structure of linkage among QTLs. We used additive genetic models, but also described approaches to include dominance and epistasis (<xref ref-type="supplementary-material" rid="pgen-0020114-sg002">Figure S2</xref>). Although these are important features of realistic genetic models, they may artificially inflate the df and further degrade the utility of an overall goodness of fit statistic. Therefore, model selection statistics and likelihood ratio tests are the most appropriate methods for assessing the structure of the phenotype component of the model [<xref rid="pgen-0020114-b030" ref-type="bibr">30</xref>].</p><p>Sample size is an important consideration for both QTL mapping and for structural modeling. We recommend using at least 200 mice in an intercross mapping design to ensure that all large to moderate effect QTL are detected. Structural modeling relies on estimated variances and covariances and sample size will determine the precision of these estimates. Guidelines developed for SEM suggest that the sample size should be at least five, and preferably ten times the number of variables in the model [<xref rid="pgen-0020114-b004" ref-type="bibr">4</xref>]. Realistic sample sizes can impose tight limitations on the number of variables that can be studied simultaneously using SEM methods. We have applied SEM methods to small numbers of phenotypes and thus we were able to consider all possible structures for the trait component of the SEM. This may not be possible when larger sets of phenotypes are considered. Extensions of this method to high dimensional data, such as gene expression microarrays, may present additional challenges. However, the same principles should apply, which suggests that applications of causal inference in high dimensional data [<xref rid="pgen-0020114-b036" ref-type="bibr">36</xref>] deserve careful consideration.</p><p>Awareness of the limitations of any model is essential to its proper interpretation. Genetic mapping in line crosses provides only coarse resolution, and any apparently singular QTL may represent the composite effects of several tightly linked polymorphic loci. Tight linkage may be misinterpreted as pleiotropy. Important QTLs or epistatic interactions may not be detected and critical physiological parameters may not have been measured. When important components of a system are missing, correlations or spurious effects may be induced that would otherwise not be inferred. For example, significant covariances among residual errors for all of the fat pad traits suggest that nongenetic factors may be involved. Food intake is a plausible source of the correlated variation. Latent variables are useful when multiple traits are highly correlated and are presumed to share a common causal factor that is not directly measured, or one that is not measurable.</p><p>Structural equation modeling provides a powerful descriptive approach to the genetic analysis of multiple traits. SEMs allow characterization of pleiotropic and heterogeneous genetic effects of multiple loci on multiple traits as well as the physiological interactions among traits. With both graphical and algebraic representations, SEMs provide an intuitive and precise description of the genetic architecture of a complex system.</p></sec><sec sec-type="supplementary-material" id="s5"><title>Supporting Information</title><supplementary-material content-type="local-data" id="pgen-0020114-sg001"><label>Figure S1</label><caption><title>Genome Scans for Each of the Fat Pad Weights</title><p>Genome-wide scans for inguinal (A), gonadal (B), peritoneal (C), and mesenteric (D) fat pad traits, based on a single QTL model. Through (A) to (D), the top scan is unconditioned for lean body weight; the middle scan is conditioned for lean body weight; and the bottom scan is the difference in LOD scores between conditioning and unconditioning. ING, inguinal fat pad weight; GON, gonadal fat pad weight; LBWT, lean body weight; MES, mesenteric fat pad weight and PERI, peritoneal fat pad weight. Phenotypes are square root transformed.</p><p>(585 KB PDF)</p></caption><media xlink:href="pgen.0020114.sg001.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020114-sg002"><label>Figure S2</label><caption><title>Structural Equation Models for Individual Fat Pad traits</title><p>Structural equation models for inguinal (A), gonadal (B), peritoneal (C), and mesenteric (D) fat pad traits. See <xref ref-type="fig" rid="pgen-0020114-g004">Figure 4</xref> legend for details. For QTL-dominant effect, a positive sign (+) is assigned when the effect of heterozygous genotype is not different from that of homozygous SM allele relative to homozygous NZB allele. Genetic loci are parameterized using the Cockerham model.</p><p>(521 KB PDF)</p></caption><media xlink:href="pgen.0020114.sg002.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020114-st001"><label>Table S1</label><caption><title>Path Analysis of the Inguinal Fat Pad Model</title><p>(29 KB DOC)</p></caption><media xlink:href="pgen.0020114.st001.doc"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020114-st002"><label>Table S2</label><caption><title>Path Analysis of the Gonadal Fat Pad Model</title><p>(28 KB DOC)</p></caption><media xlink:href="pgen.0020114.st002.doc"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020114-st003"><label>Table S3</label><caption><title>Path Analysis of the Peritoneal Fat Pad Model</title><p>(28 KB DOC)</p></caption><media xlink:href="pgen.0020114.st003.doc"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020114-st004"><label>Table S4</label><caption><title>Path Analysis of the Mesenteric Fat Pad Model</title><p>(29 KB DOC)</p></caption><media xlink:href="pgen.0020114.st004.doc"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material></sec>
Lysine 63-Polyubiquitination Guards against Translesion Synthesis–Induced Mutations	<p>Eukaryotic cells possess several mechanisms to protect the integrity of their DNA against damage. These include cell-cycle checkpoints, DNA-repair pathways, and also a distinct DNA damage–tolerance system that allows recovery of replication forks blocked at sites of DNA damage. In both humans and yeast, lesion bypass and restart of DNA synthesis can occur through an error-prone pathway activated following mono-ubiquitination of proliferating cell nuclear antigen (PCNA), a protein found at sites of replication, and recruitment of specialized translesion synthesis polymerases. In yeast, there is evidence for a second, error-free, pathway that requires modification of PCNA with non-proteolytic lysine 63-linked polyubiquitin (K63-polyUb) chains. Here we demonstrate that formation of K63-polyUb chains protects human cells against translesion synthesis–induced mutations by promoting recovery of blocked replication forks through an alternative error-free mechanism. Furthermore, we show that polyubiquitination of PCNA occurs in UV-irradiated human cells. Our findings indicate that K63-polyubiquitination guards against environmental carcinogenesis and contributes to genomic stability.</p>	<contrib contrib-type="author" equal-contrib="yes"><name><surname>Chiu</surname><given-names>Roland K</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author" equal-contrib="yes"><name><surname>Brun</surname><given-names>Jan</given-names></name><xref ref-type="aff" rid="aff2">2</xref></contrib><contrib contrib-type="author"><name><surname>Ramaekers</surname><given-names>Chantal</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Theys</surname><given-names>Jan</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Weng</surname><given-names>Lin</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Lambin</surname><given-names>Philippe</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Gray</surname><given-names>Douglas A</given-names></name><xref ref-type="aff" rid="aff2">2</xref><xref ref-type="corresp" rid="cor1"></xref></contrib><contrib contrib-type="author"><name><surname>Wouters</surname><given-names>Bradly G</given-names></name><xref ref-type="aff" rid="aff1">1</xref><xref ref-type="corresp" rid="cor1"></xref></contrib>	PLoS Genetics	<sec id="s1"><title>Introduction</title><p>In contrast to DNA-repair pathways, DNA damage tolerance (DDT) is characterized by bypass of DNA lesions rather than their direct removal or repair. The DDT pathway is likely responsible for the ability of cells to continue to proliferate with tremendous amounts of damage in their genomes [<xref rid="pgen-0020116-b001" ref-type="bibr">1</xref>]. The genetic and mechanistic basis of DDT is best understood in yeast, where it is known to be an extremely important determinant of the toxicity and mutagenicity of many DNA-damaging agents [<xref rid="pgen-0020116-b002" ref-type="bibr">2</xref>,<xref rid="pgen-0020116-b003" ref-type="bibr">3</xref>]. Often referred to as RAD6-dependent repair or post-replication repair, DDT requires interaction of the E2 ubiquitin (Ub) conjugase RAD6 and the E3 Ub ligase RAD18 at sites of DNA damage [<xref rid="pgen-0020116-b004" ref-type="bibr">4</xref>]. Here they mediate mono-ubiquitination of proliferating cell nuclear antigen (PCNA) at K164 and subsequent recruitment of a specialized translesion synthesis (TLS) polymerase capable of replication past the lesion [<xref rid="pgen-0020116-b005" ref-type="bibr">5</xref>,<xref rid="pgen-0020116-b006" ref-type="bibr">6</xref>]. Several yeast and mammalian TLS polymerases have been identified, including POLη (RAD30A), POLι (RAD30b), REV1, REV3, and POLκ [<xref rid="pgen-0020116-b007" ref-type="bibr">7</xref>]. These are highly error-prone polymerases that allow replication past a variety of DNA lesions [<xref rid="pgen-0020116-b007" ref-type="bibr">7</xref>]. POLη plays a uniquely important role in the repair of UV damage as it mediates error-free bypass of thymine–thymine dimers, the most common UV-induced lesion [<xref rid="pgen-0020116-b008" ref-type="bibr">8</xref>]. <italic>Saccharomyces cerevisiae RAD6</italic> and <italic>RAD18</italic> mutants that are unable to carry out DDT are highly sensitive to various genotoxic agents including UV irradiation and methyl methane sulfonate (MMS) [<xref rid="pgen-0020116-b009" ref-type="bibr">9</xref>]. These mutants also show a reduction in UV-induced mutations [<xref rid="pgen-0020116-b010" ref-type="bibr">10</xref>] that arises due to the inability to recruit the error-prone TLS polymerases [<xref rid="pgen-0020116-b011" ref-type="bibr">11</xref>].</p><p>Genetic epistasis studies in yeast have established a second arm of the DDT pathway that is distinct from TLS and is referred to as damage avoidance [<xref rid="pgen-0020116-b005" ref-type="bibr">5</xref>,<xref rid="pgen-0020116-b012" ref-type="bibr">12</xref>–<xref rid="pgen-0020116-b014" ref-type="bibr">14</xref>]. This pathway is also downstream of RAD6/RAD18, but in contrast to the error-prone TLS pathway resolves blocked replication forks through an error-free manner. Its mechanism is not fully understood, but may involve fork reversal and recombination with the undamaged replicated sister chromatid [<xref rid="pgen-0020116-b005" ref-type="bibr">5</xref>]. This damage-avoidance pathway requires a second ubiquitination complex composed of RAD5 and the UBC13/MMS2 heterodimer [<xref rid="pgen-0020116-b005" ref-type="bibr">5</xref>]. UBC13/MMS2 is a unique Ub conjugase that synthesizes polyUb chains linked through K63–G76 bonds rather than through the typical K48–G76 bonds [<xref rid="pgen-0020116-b013" ref-type="bibr">13</xref>]. Although lysine 63-linked polyubiquitin (K63-polyUb) chains can serve as competent proteolytic signals, they are less efficient at targeting substrates to the proteasome than K48-linked chains [<xref rid="pgen-0020116-b015" ref-type="bibr">15</xref>], and the proteolytic activity of the proteasome may not be required for error-free repair [<xref rid="pgen-0020116-b013" ref-type="bibr">13</xref>]. In yeast, a model has emerged in which error-free damage avoidance occurs when mono-ubiquitinated PCNA becomes further modified by K63-polyUb via RAD5 and MMS2/UBC13. Interestingly, modification of K164 in PCNA by sumoylation rather than by ubiquitination reduces homologous recombination [<xref rid="pgen-0020116-b016" ref-type="bibr">16</xref>,<xref rid="pgen-0020116-b017" ref-type="bibr">17</xref>].</p><p>There is convincing evidence that the DDT pathway, and particularly the TLS arm, is also important in higher eukaryotes including humans. Mouse and human homologs of <italic>RAD6, RAD18, PCNA,</italic> and many of the TLS polymerases have been identified [<xref rid="pgen-0020116-b018" ref-type="bibr">18</xref>]. The TLS polymerases form foci at sites of DNA damage following UV irradiation and are associated with other proteins in the replication machinery [<xref rid="pgen-0020116-b019" ref-type="bibr">19</xref>]. As in yeast, RAD6 and RAD18 mediate mono-ubiquitination of PCNA at K164 in UV-irradiated mammalian cells in a dose- and time-dependent manner [<xref rid="pgen-0020116-b011" ref-type="bibr">11</xref>]. Mono-ubiquitination of human PCNA has been suggested to provide a signal for polymerase switching since it leads to its increased association with POLη via its ubiquitin-binding domain (UBD) or the UBZ (ubiquitin-binding zinc-finger) in this TLS polymerase [<xref rid="pgen-0020116-b020" ref-type="bibr">20</xref>]. In vitro studies have also demonstrated that mono-ubiquitination of PCNA in yeast can stimulate the activities of both POLη and REV1 [<xref rid="pgen-0020116-b021" ref-type="bibr">21</xref>]. Recently, the deubiquitinating (DUB) enzyme USP1 was shown to directly remove the monoUb from PCNA, leading to the suggestion that USP1 is required to suppress the error-prone activity of TLS [<xref rid="pgen-0020116-b022" ref-type="bibr">22</xref>]. The functional importance of TLS is exemplified by the fact that mutations in <italic>POLη</italic> are responsible for the variant form of Xeroderma Pigmentosum (XP), a disease characterized by a 2,000-fold increased risk of developing skin cancer [<xref rid="pgen-0020116-b008" ref-type="bibr">8</xref>]. In contrast to other XP patients, those with the variant form (XPV) of Xeroderma Pigmentosum have no defect in excision repair [<xref rid="pgen-0020116-b008" ref-type="bibr">8</xref>], but are deficient in post-replication repair [<xref rid="pgen-0020116-b023" ref-type="bibr">23</xref>]. Furthermore, they display enhanced mutation at T–T sites, owing to usage of an alternative error-prone TLS polymerase [<xref rid="pgen-0020116-b024" ref-type="bibr">24</xref>].</p><p>In contrast to TLS, the importance of the damage-avoidance arm of DDT in mammalian cells is not yet firmly established. Perhaps the strongest evidence supporting a role for this pathway comes from Li et al., who showed that antisense inhibition of hMMS2 resulted in an increase in mutation frequency [<xref rid="pgen-0020116-b025" ref-type="bibr">25</xref>]. Nonetheless, several open questions remain to be resolved. First, a human homolog of <italic>RAD5</italic> has not yet been identified. This may be due to the fact that yeast RAD5 contains a helicase activity required for its function in DNA double-strand break repair, but is unimportant for DDT [<xref rid="pgen-0020116-b026" ref-type="bibr">26</xref>]. These authors speculated that <italic>RAD5</italic> in higher organisms may have evolved to lose this domain. Second, although homologs of <italic>MMS2</italic> exist <italic>(hMMS2</italic> and <italic>hCroc1)</italic> and are able to functionally complement loss of yeast MMS2 [<xref rid="pgen-0020116-b027" ref-type="bibr">27</xref>], they are additionally required for polyubiquitination of proteins in pathways unrelated to DDT [<xref rid="pgen-0020116-b028" ref-type="bibr">28</xref>]. Third, although evidence for human PCNA mono-ubiquitination is strong [<xref rid="pgen-0020116-b011" ref-type="bibr">11</xref>,<xref rid="pgen-0020116-b029" ref-type="bibr">29</xref>], there is less evidence for its polyubiquitination. High molecular weight bands in PCNA Western blots were noted in mouse fibroblasts following UV irradiation [<xref rid="pgen-0020116-b011" ref-type="bibr">11</xref>]. However, Kannouche and colleagues found no evidence for polyubiquitination in human fibroblasts [<xref rid="pgen-0020116-b029" ref-type="bibr">29</xref>]. They concluded that polyUb forms of PCNA were either insignificant, occurred only at low levels, or were rapidly turned over [<xref rid="pgen-0020116-b029" ref-type="bibr">29</xref>]. Thus, whether polyubiquitination of PCNA and subsequent activation of an error-free damage-avoidance pathway is evolutionarily conserved in humans is a source of uncertainty that we sought to resolve.</p><p>Here, we provide evidence that the ability to create K63-based polyUb chains is required for an error-free damage-avoidance pathway in human cells. We implicate this ubiquitination step in a pathway that contributes to genomic stability by suppressing translesion polymerase–mediated mutagenesis. Moreover, we show that DNA damage–induced PCNA polyubiquitination is indeed conserved in human cells, suggesting that this Ub-based molecular switch plays a decision role in directing repair in either an error-free or error-prone manner.</p></sec><sec id="s2"><title>Results</title><sec id="s2a"><title>Dominant Negative Approach to Disrupt K63-PolyUb Chain Assembly</title><p>In order to directly investigate the functional importance of K63-linked polyUb chains in DDT, we employed a strategy similar to that first described in yeast, to specifically inhibit assembly of these chains. In yeast, replacement of Ub with a mutant in which lysine 63 is mutated to arginine <italic>(K63R)</italic> disrupts the error-free arm of DDT and results in a phenotype equivalent to loss of UBC13 or MMS2 [<xref rid="pgen-0020116-b014" ref-type="bibr">14</xref>]. The <italic>K63R</italic> mutation disrupts K63-polyUb chain assembly, but has no effect on K48-linked chains that mediate proteasomal-based protein turnover [<xref rid="pgen-0020116-b014" ref-type="bibr">14</xref>]. In human cells, an equivalent knock-in approach is not feasible because Ub is expressed from multiple genes. The <italic>UBA52</italic> and <italic>UBA80</italic> genes encode a Ub monomer fused in frame with ribosomal subunits, while the <italic>UBB</italic> and <italic>UBC</italic> genes encode variable-length linear polymers of (typically three to four Ub and nine Ub proteins, respectively) [<xref rid="pgen-0020116-b030" ref-type="bibr">30</xref>]. The fusion proteins are cleaved by DUB enzymes to release individual Ub monomers.</p><p>Our approach was to express the <italic>K63R-Ub</italic> mutant in <italic>trans</italic> so that it competed with wild-type (WT) Ub for inclusion into polyUb chains. Its incorporation blocks further ubiquitination through K63 and thus acts in a dominant way. In a previous study, we validated and used this approach to specifically suppress K48-linked Ub chains by expressing a <italic>K48R-Ub</italic> mutant [<xref rid="pgen-0020116-b031" ref-type="bibr">31</xref>]. This same construct has also been used to inhibit K48 polyubiquitination in transgenic mice [<xref rid="pgen-0020116-b032" ref-type="bibr">32</xref>]. Here, we expressed a six-his-tagged <italic>K63R-Ub</italic> or <italic>WT-Ub</italic> fused in frame with <italic>GFP</italic> from the <italic>UbC</italic> promoter (<xref ref-type="fig" rid="pgen-0020116-g001">Figure 1</xref>). Expression yields a fusion protein that is cleaved, releasing a six-his-tagged Ub monomer and free GFP (<xref ref-type="fig" rid="pgen-0020116-g001">Figure 1</xref>B). GFP was used to sort pools of cells with stable high expression of the transgene. Both WT-Ub and K63R-Ub monomers were efficiently incorporated into polyUb chains as evidenced by their detection in high molecular weight smears characteristic of the heterogeneity of ubiquitinated proteins (<xref ref-type="fig" rid="pgen-0020116-g001">Figure 1</xref>B). The <italic>K63R-Ub</italic> mutant did not affect normal cell proliferation as demonstrated by the identical growth rates in the sorted stable high <italic>K63R-Ub-GFP–</italic>expressing pools and in the similarly sorted <italic>WT-Ub-GFP</italic>–expressing and the untransfected cells (<xref ref-type="fig" rid="pgen-0020116-g001">Figure 1</xref>C). Furthermore, disrupting K63-polyUb chain formation did not alter normal proteasome-mediated protein degradation of p53 or HIF1α (unpublished data). These data indicate that the K63R-Ub fusion protein is properly processed into K63R-Ub monomers, incorporates normally into chains, and does not alter the ability of the proteasome to recognize polyubiquitinated substrates targeted for degradation.</p><fig id="pgen-0020116-g001" position="float"><label>Figure 1</label><caption><title>Disruption of K63-PolyUb Chain Assembly</title><p>(A) Cartoon depicting dominant negative <italic>K63R-Ub-GFP</italic> construct. The expressed fusion protein is processed by endogenous Ub proteases generating free GFP used for detection on a flow cytometer and mono-K63R-Ub. Incorporation of this mutant will terminate K63-polyUb chains while not affecting canonical K48-polyUb chain assembly.</p><p>(B) Whole-cell lysates were isolated from untransfected cells, and from cells stably expressing either <italic>WT-Ub</italic> or <italic>K63R-Ub,</italic> followed by immunoblot analysis with antibodies directed against Ub, His, and GFP.</p><p>(C) The growth of untransfected, WT-Ub, or K63R-Ub cells was followed by cell counting over the course of 7 d.</p></caption><graphic xlink:href="pgen.0020116.g001"/></fig></sec><sec id="s2b"><title>Disruption of K63-PolyUb Chain Assembly Sensitizes Cells to Cisplatin—but Not UV—Induced Cell Death</title><p>Creation of stable cell lines expressing <italic>WT-Ub</italic> or <italic>K63R-Ub</italic> allowed us to examine the role of K63-polyUb chain assembly during recovery from DNA damage. We first investigated whether inhibition of K63-polyubiquitination would sensitize cells to agents known to sensitize yeast mutants in the error-free damage-avoidance arm of DDT [<xref rid="pgen-0020116-b002" ref-type="bibr">2</xref>,<xref rid="pgen-0020116-b003" ref-type="bibr">3</xref>]. We found that cisplatin, a chemotherapeutic agent highly toxic to yeast mutants in this pathway [<xref rid="pgen-0020116-b002" ref-type="bibr">2</xref>,<xref rid="pgen-0020116-b003" ref-type="bibr">3</xref>], is also significantly more toxic to A549 cells expressing <italic>K63R-Ub</italic> (<xref ref-type="fig" rid="pgen-0020116-g002">Figure 2</xref>). This sensitivity is specific to expression of <italic>K63R-Ub</italic> since the response of cells expressing either <italic>WT-Ub</italic> or <italic>K33R-Ub</italic> is identical to that of untransfected controls (<xref ref-type="fig" rid="pgen-0020116-g002">Figure 2</xref>A and <xref ref-type="fig" rid="pgen-0020116-g002">2</xref>B). This effect was not mediated by a general inhibition of ubiquitination since A549 cells expressing the <italic>K48R-Ub</italic> mutant are not sensitized (unpublished data). Furthermore, a K63R-Ub clone that lost expression of the transgene (as evidenced by a low GFP signal) returned to normal sensitivity (<xref ref-type="fig" rid="pgen-0020116-g002">Figure 2</xref>B). These data imply that K63-polyUb chain assembly is essential for recovery from at least a subset of cisplatin-induced lesions.</p><fig id="pgen-0020116-g002" position="float"><label>Figure 2</label><caption><title>Cells Deficient in K63-Ub Chain Formation Are Sensitized to Cisplatin Treatment while UV Sensitivity Is Revealed only upon POLη Knockdown</title><p>(A and B) Clonogenic survival assays were used to determine sensitivity to 1 h acute treatment with cisplatin in untransfected A549 cells or in A549 cells stably expressing <italic>WT-Ub</italic> or <italic>K63R-Ub.</italic> The mean values of three independent experiments are shown with standard error of the mean (error bars). Cells expressing <italic>K33R-Ub</italic> or cells that lost <italic>K63R-Ub</italic> expression revert to WT-Ub cisplatin sensitivity.</p><p>(C) Cells were treated for 24 h with 100 μM cisplatin followed by Hoechst staining to detect apoptosis. The mean values of three independent experiments are shown with standard deviation.</p><p>(D) Clonogenic survival assays were used to determine sensitivity to UV irradiation in untransfected A549 cells or in A549 cells stably expressing <italic>WT-Ub</italic> or <italic>K63R-Ub.</italic> </p><p>(E) Clonogenic survival of A549 cells stably expressing <italic>WT-Ub</italic> or <italic>K63R-Ub</italic> with or without POLη RNAi following 10 J/m<sup>2</sup> UV treatment.</p></caption><graphic xlink:href="pgen.0020116.g002"/></fig><p>We also examined the functional importance of K63-polyubiquitination in the recovery from UV-induced damage. In contrast to the data with cisplatin, the cell line with stable expression of <italic>K63R-Ub</italic> exhibited a dose response to UV irradiation that was identical to the parental cells or to cells expressing <italic>WT-Ub</italic> (<xref ref-type="fig" rid="pgen-0020116-g002">Figure 2</xref>D). Thus, despite evidence that K63-polyUb chains are required for cisplatin tolerance, we found no evidence that disruption of K63-polyUb chain assembly on its own influences UV toxicity. A possible explanation for this lack of sensitivity to UV irradiation is that cells can compensate for loss of K63-polyUb–dependent repair through increased utilization of the error-prone TLS arm of the pathway. A similar situation occurs in yeast where inhibition of the error-free damage-avoidance arm of DDT results in a much milder UV sensitivity than mutations in <italic>RAD6</italic> or <italic>RAD18</italic> which additionally prevent TLS [<xref rid="pgen-0020116-b033" ref-type="bibr">33</xref>]. Using siRNA, we were able to knock down expression of POLη by ~13-fold (<xref ref-type="supplementary-material" rid="pgen-0020116-sg001">Figure S1</xref>). Similar to inhibition of K63-polyubiquitination, knockdown of POLη had no effect on UV sensitivity on its own. This observation is not unexpected since XPV cells (defective in POLη) are not sensitive to killing by UV irradiation. In contrast, knockdown of POLη in cells also expressing <italic>K63R-Ub</italic> did cause increased cell kill after UV treatment (<xref ref-type="fig" rid="pgen-0020116-g002">Figure 2</xref>E). This increase in UV sensitivity suggests that K63-polyUb and POLη function in distinct, complementary pathways that mediate recovery from UV-induced damage.</p></sec><sec id="s2c"><title>Disruption of K63-PolyUb Chain Assembly Increases UV-Induced Mutations</title><p>Disruption of the error-free arm in yeast is also known to result in a dramatic increase in UV-induced mutations that is synergistic with the TLS mutant, REV3 [<xref rid="pgen-0020116-b034" ref-type="bibr">34</xref>]. If playing a similar role in mammalian cells, inhibition of K63-polyubiquitination should also increase UV-induced mutations. We thus analyzed mutation induction at the <italic>HPRT</italic> locus after UV irradiation and cisplatin exposure in these same cell lines as well as in normal human fibroblasts expressing <italic>WT-Ub</italic> or <italic>K63R-Ub</italic> (<xref ref-type="fig" rid="pgen-0020116-g003">Figure 3</xref>). Consistent with this hypothesis, A549 cells expressing <italic>K63R-Ub</italic> show a 2.5-fold increase in UV-induced mutations compared to cells expressing <italic>WT-Ub</italic> (<xref ref-type="fig" rid="pgen-0020116-g003">Figure 3</xref>B), and a similar increase (2.2-fold) is observed in normal fibroblasts (<xref ref-type="fig" rid="pgen-0020116-g003">Figure 3</xref>C). Untransfected and <italic>WT-Ub</italic>–expressing cells have similar mutation frequencies (unpublished data). The increase in mutations upon inhibition of K63-polyubiquitination is consistent with a recent report that used antisense to suppress the expression of MMS2 in human cells [<xref rid="pgen-0020116-b025" ref-type="bibr">25</xref>]. Similar to the cells expressing <italic>Ub-K63R,</italic> loss of MMS2 led to an ~2-fold increase in UV-induced mutations without increasing UV-induced cell death [<xref rid="pgen-0020116-b025" ref-type="bibr">25</xref>]. Thus, both the enzyme that is implicated in the synthesis of K63-polyUb chains, and the chains themselves, are required for recovery from UV damage through a pathway that prevents mutations.</p><fig id="pgen-0020116-g003" position="float"><label>Figure 3</label><caption><title>Cells Deficient in K63-Ub Chain Formation Are Mutagenic in Response to UV Treatment</title><p>(A and B) Cells were treated with cisplatin for 1 h or UV irradiation and subcultured for 7 d. Cells were then plated and grown in 6-TG to select for <italic>HPRT</italic> mutants. The mean values of three independent experiments are shown with standard deviation.</p><p>(C) Normal fibroblasts stably expressing <italic>WT-Ub</italic> or <italic>K63R-Ub</italic> were UV-irradiated (10 J/m<sup>2</sup>) and cultured for 5 d. Cells were then plated and grown in 6-TG to select for <italic>HPRT</italic> mutants.</p><p>(D) The number of <italic>HPRT</italic> mutants was quantitated for A549 cells stably expressing <italic>WT-Ub</italic> or <italic>K63R-Ub</italic> with or without POLη RNAi. Cells were treated as described in <xref ref-type="fig" rid="pgen-0020116-g003">Figure 3</xref>C.</p><p>(E) Cells were UV-irradiated and plated in the absence or presence of 0.4 mM caffeine. The mean values of three independent experiments are shown with standard error of the mean (error bars).</p></caption><graphic xlink:href="pgen.0020116.g003"/></fig></sec><sec id="s2d"><title>Increases in UV-Induced Mutations Are Due to Increased Utilization of TLS</title><p>Many of the TLS polymerases are known to be important contributors to UV-induced mutagenesis as is illustrated by a reduction in mutation frequency when inactivated in yeast [<xref rid="pgen-0020116-b035" ref-type="bibr">35</xref>–<xref rid="pgen-0020116-b038" ref-type="bibr">38</xref>]. The data presented thus far are consistent with a model in which inhibition of K63-polyubiquitination increases UV-induced mutations owing to increased use of the error-prone branch of the TLS pathway. However, the possibility that <italic>K63R-Ub</italic> expression in some way increases mutations by affecting the function of one or more TLS polymerases cannot be ruled out. In fact, the phenotype of cells expressing <italic>K63R-Ub</italic> is similar to that described for XPV cells. Both cell types display an increase in UV-induced mutations with no significant change in UV-induced cell death. In XPV cells, this is due to loss of POLη which replicates past T–T dimers in an error-free manner [<xref rid="pgen-0020116-b039" ref-type="bibr">39</xref>]. Defects in POLη can be revealed by a significant increase in UV sensitivity when irradiated in the presence of caffeine, an assay used to establish the XPV phenotype [<xref rid="pgen-0020116-b040" ref-type="bibr">40</xref>]. However, we found that cells expressing <italic>K63R-Ub</italic> are not similarly hypersensitive to this combined treatment (<xref ref-type="fig" rid="pgen-0020116-g003">Figure 3</xref>E), suggesting no overt defect in POLη function in these cells.</p><p>In contrast, our data suggest that POLη and K63-polyUb chains participate in separate, alternative pathways for recovery from UV-induced DNA damage. Consistent with this idea, knockdown of POLη in combination with the inhibition of K63-polyUb chain assembly resulted in both an increased toxicity to UV irradiation (<xref ref-type="fig" rid="pgen-0020116-g002">Figure 2</xref>E) and in a further increase in UV-induced mutations (<xref ref-type="fig" rid="pgen-0020116-g003">Figure 3</xref>D). Interestingly, the number of mutations in cells following knockdown of POLη in combination with inhibition of K63-polyUb chain assembly were far greater than additive. As expected, loss of POLη, which replicates past T–T dimers with high fidelity, resulted in a large induction in UV-induced mutations in <italic>WT-Ub</italic>–expressing cells (<xref ref-type="fig" rid="pgen-0020116-g003">Figure 3</xref>D). These mutations are likely due to the activity of alternative TLS polymerases that can substitute for POLη, but which are error-prone across T–T dimers [<xref rid="pgen-0020116-b041" ref-type="bibr">41</xref>]. Additional suppression of K63-polyUb chain assembly increased the number of UV-induced mutations by 3.5-fold. This synergistic increase in mutations strongly suggests that the inability to form K63-polyUb chains places a greater requirement on the TLS pathway, and thus POLη; it is also likely that there will be a greater requirement for other lesion bypass polymerases such as POLζ [<xref rid="pgen-0020116-b042" ref-type="bibr">42</xref>,<xref rid="pgen-0020116-b043" ref-type="bibr">43</xref>] for recovery from UV damage. Moreover, the synergistic increase in mutations suggests that a significant proportion of the repair is normally carried out by the error-free component of the damage-avoidance pathway.</p><p>To further investigate the relationship between inhibition of K63R-polyUb chain assembly and TLS, we examined the spatial dynamics of the TLS polymerase POLη. This polymerase is recruited to sites of damage and can be visualized in discrete foci that co-localize with PCNA [<xref rid="pgen-0020116-b044" ref-type="bibr">44</xref>]. We analyzed the effects of <italic>K63R-Ub</italic> expression on POLη foci formation in live cells using a <italic>POLη-GFP</italic> fusion construct [<xref rid="pgen-0020116-b044" ref-type="bibr">44</xref>] (<xref ref-type="fig" rid="pgen-0020116-g004">Figure 4</xref>). Since our original cells co-expressed <italic>GFP,</italic> we generated new stable lines from both A549 and HeLa cells expressing <italic>WT-Ub</italic> or <italic>K63R-Ub</italic> fused with the puromycin-resistance gene. These cell lines are phenotypically equivalent to the original <italic>GFP</italic>-expressing cells (an ~3-fold increase in <italic>HPRT</italic> mutants in cells expressing <italic>K63R-Ub</italic> compared to <italic>WT-Ub</italic>). Similar to previous observations [<xref rid="pgen-0020116-b044" ref-type="bibr">44</xref>], the majority of nonirradiated cells show homogenous nuclear distribution of the tagged polymerases (<xref ref-type="fig" rid="pgen-0020116-g004">Figures 4</xref> and S2). Foci were observed in ~11%–12% of cells and likely represent sites of ongoing replication [<xref rid="pgen-0020116-b044" ref-type="bibr">44</xref>]. When treated with 10 J/m<sup>2</sup> UV irradiation, the percentage of cells with foci increased to 30% in cells expressing <italic>WT-Ub</italic> and to 49% in cells expressing <italic>K63R-Ub</italic> 6 h posttreatment (<xref ref-type="fig" rid="pgen-0020116-g004">Figure 4</xref>B and <xref ref-type="fig" rid="pgen-0020116-g004">4</xref>D). This corresponds to an ~2-fold increase in UV-induced foci as a consequence of inhibition of K63-polyubiquitination (2.4-fold increase over background for <italic>WT-Ub</italic> versus 4.6-fold for <italic>K63R-Ub, p</italic> < 0.007).</p><fig id="pgen-0020116-g004" position="float"><label>Figure 4</label><caption><title>Disrupting K63-PolyUb Chain Formation Increases Reliance of Cells on the Error-Prone TLS Pathway</title><p>(A) HeLa cells stably expressing <italic>WT-Ub-puro</italic> or <italic>K63R-Ub-puro</italic> were transiently transfected with a plasmid expressing a <italic>POLη-GFP</italic> fusion. Twenty-four hours post-transfection, cells were UV-irradiated (10 J/m<sup>2</sup>). POLη (green) and PCNA (red) were detected using antibodies. Shown are representative confocal photographs of cells 6 h post-UV treatment.</p><p>(B) Kinetics of POLη foci formation in <italic>WT-Ub–</italic> and <italic>K63R-Ub–</italic>expressing HeLa cell lines.</p><p>(C) <italic>HPRT</italic> mutation spectra. RNA was isolated from 6-TG resistant 10 J/m<sup>2</sup> UV-treated clones followed by RT-PCR and sequence analysis of the <italic>HPRT</italic> locus. The UV-induced mutations are shown in the upper table. Most of the point mutations were G→A or C→T transitions indicated as G/C→A/T. The lower table in (C) shows the same mutants in sequence context.</p><p>(D) Foci were quantitated 6 h post-UV treatment using a live-cell imaging fluorescent microscope.</p><p>(E) The number of <italic>HPRT</italic> mutants was quantitated for A549 cells stably expressing <italic>K63R-Ub</italic> with or without RAD18 RNAi. Cells were treated as described in <xref ref-type="fig" rid="pgen-0020116-g004">Figure 4</xref>B.</p></caption><graphic xlink:href="pgen.0020116.g004"/></fig><p>We also analyzed the co-localization of these foci with sites of DNA replication as revealed by positive PCNA foci. We found that in both <italic>WT-Ub</italic> and <italic>K63R-Ub</italic>–expressing cells, 100% of the UV-induced POLη foci co-localized with PCNA foci (<xref ref-type="fig" rid="pgen-0020116-g004">Figure 4</xref>A). This suggests that the foci produced in the <italic>K63R-Ub</italic>–expressing cells are typical of those previously reported to occur at sites of blocked replication [<xref rid="pgen-0020116-b044" ref-type="bibr">44</xref>]. To rule out the possibility that UV differentially affects the cell cycle in the two cell lines (and thus the number of cells in the S phase), we measured cell-cycle distributions before and after UV treatment and found no significant differences (<xref ref-type="supplementary-material" rid="pgen-0020116-sg003">Figure S3</xref>). In both cell lines, the percentage of cells with foci after UV treatment increased rapidly during the first 30 min and then reached a plateau after 3–4 h (<xref ref-type="fig" rid="pgen-0020116-g004">Figure 4</xref>B). Thus, although the percentage of positive cells was consistently higher in cells expressing <italic>K63R-Ub,</italic> the kinetics of foci formation are similar. This supports the argument that the K63R-Ub mutant is not interfering in some way with TLS polymerase recruitment dynamics. Interestingly, the magnitude of the increase in foci formation in the <italic>K63R-Ub</italic> mutant cells is similar to the increase in UV-induced mutation frequency in these cells (<xref ref-type="fig" rid="pgen-0020116-g003">Figures 3</xref>B and <xref ref-type="fig" rid="pgen-0020116-g004">4</xref>B).</p><p>We also looked for possible changes in the types of mutations induced by UV irradiation after inhibition of K63-polyubiquitination. The two predominant UV-induced lesions are the <italic>cis</italic>-syn cyclobutane pyrimidine dimer (CPD) and the pyrimidine-6/4-pyrimidone (6–4PP) photoproduct [<xref rid="pgen-0020116-b045" ref-type="bibr">45</xref>,<xref rid="pgen-0020116-b046" ref-type="bibr">46</xref>]. The most common lesion is the thymine–thymine CPD (represented by T–T) followed by T–C and the thymine–cytosine 6–4PP (represented by T(6,4)C) [<xref rid="pgen-0020116-b047" ref-type="bibr">47</xref>]. Levels of T(6,4)T, C–T, and C–C lesions are comparatively much lower. However, the normal spectrum of UV-induced mutations does not match this pattern of damage induction. Mutations are primarily C to T transitions that arise at T–C and C–C sites due to mis-incorporation of adenine opposite the 3′C [<xref rid="pgen-0020116-b048" ref-type="bibr">48</xref>,<xref rid="pgen-0020116-b049" ref-type="bibr">49</xref>]. The weak contribution of the T–T lesion to mutation may be explained by the activities of POLη and POLι, which accurately bypass T–T and T(6,4)T lesions, respectively [<xref rid="pgen-0020116-b050" ref-type="bibr">50</xref>,<xref rid="pgen-0020116-b051" ref-type="bibr">51</xref>].</p><p>To further probe for possible changes in the function of these polymerases upon inhibition of K63-polyubiquitination, we examined the spectrum of UV-induced mutations in cells expressing <italic>K63R-Ub.</italic> By sequence analysis of the expressed <italic>HPRT</italic> transcript, we found that the increase in mutations noted in <xref ref-type="fig" rid="pgen-0020116-g003">Figure 3</xref> can be accounted for entirely by additional point mutations. We sequenced 20 of these mutations and found that they were all located at dipyrimidine sites (<xref ref-type="table" rid="pgen-0020116-t001">Table 1</xref>). The majority of mutations were C to T transitions (80%), with most of these being TC to TT (55%) (<xref ref-type="fig" rid="pgen-0020116-g004">Figure 4</xref>C). These data are consistent with the mutation spectrum of other normal cell lines and contrast with data reported for cells with disruptions in TLS polymerases [<xref rid="pgen-0020116-b048" ref-type="bibr">48</xref>–<xref rid="pgen-0020116-b050" ref-type="bibr">50</xref>,<xref rid="pgen-0020116-b052" ref-type="bibr">52</xref>]. Importantly, inhibition of K63-polyubiquitination did not cause any mutations at T–T sites, suggesting normal function of both POLη and POLι in these cells.</p><table-wrap id="pgen-0020116-t001" content-type="2col" position="float"><label>Table 1</label><caption><p>Disrupting K63-PolyUb Chain Assembly Induces a Characteristic UV Mutation Signature</p></caption><graphic xlink:href="pgen.0020116.t001"/></table-wrap><p>Collectively, these data suggest that inhibition of K63-polyUb chain assembly results in an increased requirement for TLS after UV irradiation and consequently increased numbers of visible TLS foci, and an increase in TLS-associated mutations. To further support this assertion, we examined the dependence of the observed phenotype on RAD18 function. We transfected our <italic>WT-Ub</italic> and <italic>K63R-Ub</italic>–expressing stable cell lines with siRNA directed against RAD18 using conditions which consistently showed >10-fold reduction in expression (<xref ref-type="supplementary-material" rid="pgen-0020116-sg001">Figure S1</xref>). In both cell lines, UV-induced POLη foci formation was abrogated by RAD18 knockdown, implying that the recruitment of TLS polymerases to sites of damage are RAD18-dependent (<xref ref-type="fig" rid="pgen-0020116-g004">Figure 4</xref>D). This is similar to previous reports showing the requirement of RAD18 for POLη foci formation [<xref rid="pgen-0020116-b011" ref-type="bibr">11</xref>]. Significantly, the UV-induced foci formation in <italic>K63R-Ub</italic>–expressing cells was also reduced to nonirradiated levels, suggesting that the increased number of foci that are found in cells expressing the <italic>K63R</italic> mutant is also downstream of RAD18. RAD18 has been previously shown to be important for recombinational repair, and RAD18-knockout mouse embryonic stem cells exhibit more sister chromatid exchanges in response to DNA damage [<xref rid="pgen-0020116-b053" ref-type="bibr">53</xref>]. The combination of disrupting K63R-polyUb chain formation and RAD18 knockdown did not show an increase in mutations; in fact a modest, but non-significant, decrease was observed (<xref ref-type="fig" rid="pgen-0020116-g004">Figure 4</xref>E). The lack of an additive mutation effect supports the foci data implicating a role for K63-polyUb chain formation downstream of RAD18.</p></sec><sec id="s2e"><title>PCNA Is Polyubiquitinated</title><p>Our data support a role for the formation of K63-polyUb chains in promoting the recovery of human cells from DNA damage through an error-free pathway that is distinct from TLS. A likely target of this polyubiquitination is PCNA, which in yeast is modified by K63-polyUb via the RAD5/MMS2 complex. However, similar modification of PCNA has not been observed in UV-irradiated human cells [<xref rid="pgen-0020116-b029" ref-type="bibr">29</xref>]. We investigated PCNA modification after UV irradiation in three separate human cell lines: A549 lung cancer cells; 293T embryonic kidney cells; and Hela cervical cancer cells. Six hours following a dose of 30 J/m<sup>2</sup>, we observed the appearance of a prominent band consistent with mono-ubiquitinated PCNA, and overexposure of this blot revealed additional PCNA-immunoreactive bands of higher molecular weight consistent with PCNA modified with 2, 3, and 4 Ub molecules (<xref ref-type="fig" rid="pgen-0020116-g005">Figure 5</xref>).</p><fig id="pgen-0020116-g005" position="float"><label>Figure 5</label><caption><title>Modification of PCNA by Polyubiquitin in Human Cells after DNA Damage</title><p>(A) A549, 293T, and Hela cells were irradiated with 0 or 30 J/m<sup>2</sup> UV and lysed 6 h posttreatment followed by immunoblotting for PCNA.</p><p>(B) 293T cells were transfected with 100 nM of either control siRNA, siRNA Ubc13, or siRNA RAD18. Seventy-two hours post-transfection, cells were treated as described in <xref ref-type="fig" rid="pgen-0020116-g001">Figure 1</xref>A. A darker and lighter exposure of the PCNA immunoblot is shown.</p><p>(C) A549, 293T, and Hela cells were irradiated with 30 J/m<sup>2</sup> UV and lysed in boiling SDS, diluted in lysis buffer and subjected to immunoprecipitation with a PCNA antibody and detected with PCNA or Ub antibodies. The controls in the immunoprecipitations were “no 1”, in which lysates were incubated with beads but no PCNA antibody, and “1 B” in which PCNA antibody was incubated with beads alone.</p><p>(D) 293T cells were transfected as described in <xref ref-type="fig" rid="pgen-0020116-g005">Figure 5</xref>B. Seventy-two hours post-transfection, cells were irradiated with 30 J/m<sup>2</sup> of UV and lysed 6 h later in boiling SDS, diluted in lysis buffer, and subjected to immunoprecipitation with a PCNA antibody and immunoblotted for PCNA (upper panel) and Ub (lower panel). A lighter exposure of the PCNA IP immunoblotted for Ub is also shown. A PCNA immunoblot with darker and lighter exposure performed on protein lysates from the same samples used in the immunoprecipitations is also shown. Asterisks denote immunoglobulin heavy and light chains as detected on the immunoprecipitations.</p></caption><graphic xlink:href="pgen.0020116.g005"/></fig><p>As it has been previously demonstrated that RAD18 is required for the mono-ubiquitination of PCNA in human cells, and that this monoUb PCNA species is required as a substrate for UBC13-mediated K63-polyubiquitination in yeast, we sought to determine whether the observed higher molecular weight bands are dependent on either RAD18 or UBC13. To this end, the expression of RAD18 or UBC13 was knocked down using the appropriate siRNAs (<xref ref-type="supplementary-material" rid="pgen-0020116-sg001">Figure S1</xref>). As expected, the band corresponding with monoUb PCNA was substantially reduced in lysates from RAD18 siRNA-transfected cells, and this also resulted in suppression of the higher molecular weight (polyUb) forms of PCNA presumably modified with 2, 3, or 4 Ub molecules (<xref ref-type="fig" rid="pgen-0020116-g005">Figures 5</xref>B and <xref ref-type="supplementary-material" rid="pgen-0020116-sg004">S4</xref>A). In contrast, knockdown of the E2 Ub ligase responsible for K63-polyubiquitination had no effect on the formation of monoUb PCNA after UV irradiation, but did effectively reduce the di, tri, and quad polyUb PCNA bands to levels similar to those in the RAD18 knockdowns (<xref ref-type="fig" rid="pgen-0020116-g005">Figures 5</xref>B and <xref ref-type="supplementary-material" rid="pgen-0020116-sg004">S4</xref>A). Together these data suggest that in both cell lines tested, UV induces modification of PCNA by both monoUb (in a RAD18-dependent manner) and by K63-polyUb chains of length 2, 3, and 4 (in a RAD18 and UBC13-dependent manner).</p><p>To further demonstrate that these higher molecular weight species are indeed ubiquitinated forms of PCNA, we immunoprecipitated PCNA from A549, 293T, and Hela cells, and probed using an antibody directed against Ub or PCNA (<xref ref-type="fig" rid="pgen-0020116-g005">Figure 5</xref>C). In addition, we excluded the possibility that an ubiquitinated protein was co-immunoprecipitated with PCNA by lysing cells in boiling 0.5% sodium dodecyl sulfate (SDS) to ensure dissociation of PCNA complexes. Following a 5-fold dilution (0.1% SDS), the lysates were immunoprecipitated. Under these conditions, we reproducibly observed several higher molecular weight bands consistent with polyubiquitination of PCNA in each of the three cell lines (<xref ref-type="fig" rid="pgen-0020116-g005">Figure 5</xref>C and <xref ref-type="fig" rid="pgen-0020116-g005">5</xref>D, and <xref ref-type="supplementary-material" rid="pgen-0020116-sg004">Figure S4</xref>B and <xref ref-type="supplementary-material" rid="pgen-0020116-sg004">S4</xref>C). These Ub-immunoreactive bands correspond well with the predicted molecular weights for di-, tri-, and quad-ubiquitinated PCNA. The antibody against Ub reproducibly demonstrated less affinity for the mono-ubiquitinated form of PCNA, although this was clearly the most abundant form as shown by PCNA immunoblots (<xref ref-type="fig" rid="pgen-0020116-g005">Figure 5</xref>C and <xref ref-type="fig" rid="pgen-0020116-g005">5</xref>D, and <xref ref-type="supplementary-material" rid="pgen-0020116-sg004">Figure S4</xref>B and <xref ref-type="supplementary-material" rid="pgen-0020116-sg004">S4</xref>C). This appears to be a characteristic of the antibody, as we have seen this reproducibly for other ubiquitinated proteins (unpublished data).</p><p>Interestingly, each of the cell lines, particularly 293T cells, also show low levels of PCNA polyubiquitination in the absence of UV irradiation. However, in all cases the Ub-immunoreactive bands are significantly increased upon irradiation in a manner consistent with the increase in mono-ubiquitinated PCNA (<xref ref-type="fig" rid="pgen-0020116-g005">Figure 5</xref>C and <xref ref-type="fig" rid="pgen-0020116-g005">5</xref>D, and <xref ref-type="supplementary-material" rid="pgen-0020116-sg004">Figure S4</xref>B and <xref ref-type="supplementary-material" rid="pgen-0020116-sg004">S4</xref>C). Similar to previous reports [<xref rid="pgen-0020116-b029" ref-type="bibr">29</xref>], mono-ubiquitinated PCNA was readily visible 1.5 h after UV treatment and remained present for up to 24 h as detected by the PCNA antibody (<xref ref-type="supplementary-material" rid="pgen-0020116-sg004">Figure S4</xref>B). Similarly, bands consistent with di, tri, and quad polyUb forms of PCNA became visible within 1.5 h following UV irradiation, and remained present up to 24 h after exposure (<xref ref-type="supplementary-material" rid="pgen-0020116-sg004">Figure S4</xref>B). Importantly, consistent with the PCNA Western blots (<xref ref-type="fig" rid="pgen-0020116-g005">Figures 5</xref>B and <xref ref-type="supplementary-material" rid="pgen-0020116-sg004">S4</xref>), the Ub-immunoreactive bands following PCNA IP in both Hela (<xref ref-type="supplementary-material" rid="pgen-0020116-sg004">Figure S4</xref>C) and 293T (<xref ref-type="fig" rid="pgen-0020116-g005">Figure 5</xref>D) cells were substantially reduced following knockdown of either RAD18 or UBC13. As expected, RAD18 knockdown blocked both mono-ubiquitination and polyubiquitination of PCNA, whereas UBC13 knockdown inhibited only the di, tri, and quad polyUb forms (<xref ref-type="fig" rid="pgen-0020116-g005">Figure 5</xref>B and <xref ref-type="fig" rid="pgen-0020116-g005">5</xref>D, and <xref ref-type="supplementary-material" rid="pgen-0020116-sg004">Figure S4</xref>A and <xref ref-type="supplementary-material" rid="pgen-0020116-sg004">S4</xref>C).</p><p>Collectively, these data show that PCNA is indeed modified by polyUb chains in human cell lines. Similar modification was observed in primary skin and lung fibroblasts (unpublished data) and in response to other forms of damage such as cisplatin (<xref ref-type="supplementary-material" rid="pgen-0020116-sg005">Figure S5</xref>). We speculate that the lack of PCNA polyubiquitination reported earlier [<xref rid="pgen-0020116-b029" ref-type="bibr">29</xref>] may be explained by technical difficulties in detecting Ub owing to its strong tertiary structure [<xref rid="pgen-0020116-b054" ref-type="bibr">54</xref>], the low abundance of polyubiquitinated PCNA, or perhaps differences in cell types. In our studies, Ub blots were autoclaved to overcome detection problems associated with its strong tertiary structure [<xref rid="pgen-0020116-b054" ref-type="bibr">54</xref>]. We also excluded the possibility that DUB enzymes in cell lysates may have activity against ubiquitinated PCNA by repeating the immunoprecipitation in the presence of <italic>N</italic>-ethylmaleimide (NEM), a nonspecific inhibitor of DUBs (<xref ref-type="supplementary-material" rid="pgen-0020116-sg006">Figure S6</xref>). Under these conditions, no change in PCNA polyubiquitination was observed.</p></sec></sec><sec id="s3"><title>Discussion</title><p>The highly conserved Ub protein serves as a pleiotropic covalent tag for many cellular proteins. It has essential proteolytic and nonproteolytic functions that are based on the length and topology of the chain formed. The pathway in which Ub is most commonly associated is the proteasome pathway, a system for targeting protein substrates via K48-linked polyUb chains for degradation in the 26S proteasome [<xref rid="pgen-0020116-b055" ref-type="bibr">55</xref>]. However, there is increasing evidence that Ub plays an important role in a number of nonproteolytic pathways including receptor internalization [<xref rid="pgen-0020116-b056" ref-type="bibr">56</xref>], translation [<xref rid="pgen-0020116-b057" ref-type="bibr">57</xref>], signal transduction [<xref rid="pgen-0020116-b028" ref-type="bibr">28</xref>], gene regulation [<xref rid="pgen-0020116-b058" ref-type="bibr">58</xref>], and DNA repair [5,6,14,25,29,59]. These roles appear to be mediated in part by the non-canonical polyUb chains. Much less is known about this aspect of ubiquitination compared with the role of K48-polyUb in protein degradation. Of particular interest are chains linked through K63, as genetic studies in <named-content content-type="genus-species">S. cerevisiae</named-content> have shown that the enzymatic complex (RAD5, UBC13/MMS2) that assembles these chains is required to protect cells from the harmful effects of genotoxic agents by allowing the replication machinery to bypass DNA lesions in a faithful manner [<xref rid="pgen-0020116-b014" ref-type="bibr">14</xref>]. In fact, ubiquitination of the DNA polymerase processivity factor PCNA is emerging as a key “molecular switch” for DDT [<xref rid="pgen-0020116-b005" ref-type="bibr">5</xref>,<xref rid="pgen-0020116-b006" ref-type="bibr">6</xref>,<xref rid="pgen-0020116-b029" ref-type="bibr">29</xref>]. Mono-ubiquitination of PCNA promotes error-prone TLS, while K63-polyUb activates error-free damage avoidance. The body of evidence supporting the requirement of PCNA post-translational modifications for DDT in mammalian cells is only now emerging.</p><p>In this report, we provide evidence to support a model (<xref ref-type="fig" rid="pgen-0020116-g006">Figure 6</xref>) in which ubiquitination of PCNA acts at a central decision point to direct the recovery of blocked replication forks towards one of two alternative pathways in mammalian cells. Recent reports have confirmed that RAD6-dependent mono-ubiquitination of PCNA also stimulates TLS in human cells. This stimulation appears to result through direct binding of the TLS polymerases to mono-ubiquitinated PCNA [<xref rid="pgen-0020116-b011" ref-type="bibr">11</xref>,<xref rid="pgen-0020116-b029" ref-type="bibr">29</xref>].</p><fig id="pgen-0020116-g006" position="float"><label>Figure 6</label><caption><title>Model of the DDT Pathway in Mammalian Cells</title><p>Recovery from a stalled replication fork at sites of DNA damage can occur by one of two alternative pathways. Previous work has shown that PCNA mono-ubiquitination by the RAD6/RAD18 complex stimulates lesion bypass through recruitment of the error-prone TLS polymerases. Here we show that an alternative error-free pathway requires formation of K63-polyUb chains. Blockade of this error-free pathway results in increased use of the TLS polymerases after DNA damage and a corresponding increase in mutations. As the TLS polymerases POLη and POLι both bind directly and avidly to polyUb chains [<xref rid="pgen-0020116-b020" ref-type="bibr">20</xref>], it is hypothesized that the interaction with K63-polyUb causes a disengagement of the polymerase from the DNA, allowing other proteins to migrate to the site of damage to perform error-free repair. This model predicts that K63-polyubiquitination acts to suppress environmental carcinogenesis by preventing genomic instability that would otherwise be introduced by the TLS polymerases.</p></caption><graphic xlink:href="pgen.0020116.g006"/></fig><p>Our data indicate that formation of K63-polyUb chains is required to utilize an error-free pathway distinct from TLS. This pathway is required for cell survival from at least some types of DNA damage, as its inhibition cannot be compensated for by the alternative TLS pathway in the case of cisplatin-damaged cells. For UV-induced damage, inhibition of K63-polyubiquitination does not affect overall cell survival, but instead causes an increase in mutations arising from an apparent increased requirement for the error-prone branch of TLS. This is supported by several lines of evidence. First, blockade of K63-polyUb chain formation led to a 2.4-fold increase in RAD18-dependent TLS foci after UV irradiation. Second, we found that the number of UV-induced mutations increased by a similar factor in these cells, and that the spectra of these mutations are consistent with that produced normally by error-prone TLS polymerases after UV treatment. Third, POLη knockdown in combination with blockade of K63-polyUb chain formation led to increased toxicity to UV irradiation, although no change was seen with either individually. Fourth, an increased reliance on the TLS arm upon blockade of K63-polyUb chain assembly was revealed by a synergistic increase in UV-induced mutations when expressed in POLη knockdown cells. POLη knockdown cells showed a high mutation rate as expected, but this rate increased by a factor of 3.5 when K63-polyUb chain assembly was inhibited.</p><p>Together, these data imply that formation of K63-polyUb chains can activate an error-free mechanism to protect cells against mutations that would otherwise be induced by the error-prone TLS polymerases. It will be of interest to determine whether K63-polyUb chain formation also plays a role in protection against sunlight-induced skin cancer.</p><p>An obvious question that emerges is how formation of K63-polyUb acts to suppress TLS. Recent reports have demonstrated that the TLS polymerases POLη and POLι both bind directly and avidly to polyUb chains through newly discovered binding domains [<xref rid="pgen-0020116-b020" ref-type="bibr">20</xref>,<xref rid="pgen-0020116-b060" ref-type="bibr">60</xref>]. A C-terminal zinc finger domain of POLη and the proline residue at position 692 of POLι are required for the respective interaction with Ub [<xref rid="pgen-0020116-b020" ref-type="bibr">20</xref>]. Together with our data, this suggests a possible mechanism whereby differential ubiquitination of PCNA could act as a switch between TLS and an alternative error-free pathway (<xref ref-type="fig" rid="pgen-0020116-g006">Figure 6</xref>). In this model, the TLS polymerases are recruited to the sites of replication through interaction with mono-ubiquitinated PCNA and subsequently mediate TLS across DNA lesions. Extension of the Ub chain through K63-linked polyubiquitination in some way suppresses TLS activity and promotes recovery through an alternative error-free pathway. This suppression may be mediated through the recently discovered ability of POLη and POLι to directly bind K63-polyUb chains. An intriguing possibility is that K63-polyUb chains are cleaved upon binding to TLS polymerases, thereby functionally removing them from the site of the lesion. This possibility is supported by the low detectable levels of polyubiquitinated PCNA as well as by the observed increase in POLη foci in <italic>K63R-Ub</italic>–expressing cells.</p><p>Although our data suggest that PCNA is indeed a target for K63-polyubiquitination, they do not exclude the possibility that other key proteins in this pathway are also important substrates for these chains. Indeed, K63-polyubiquitination occurs on at least three proteins (RIP, NEMO, and TRAF6) in an unrelated pathway that activates NF-κB [<xref rid="pgen-0020116-b028" ref-type="bibr">28</xref>,<xref rid="pgen-0020116-b061" ref-type="bibr">61</xref>,<xref rid="pgen-0020116-b062" ref-type="bibr">62</xref>]. In this pathway, K63-polyUb chains on multiple proteins may facilitate their assembly into an active complex [<xref rid="pgen-0020116-b062" ref-type="bibr">62</xref>]. It is therefore intriguing to speculate that K63-polyUb chains may not only uncouple the TLS polymerases from the site of damage, but may also provide a mechanism for recruitment of other proteins required for error-free repair.</p><p>Non-proteolytic roles for Ub have also been implicated in other DNA-repair pathways that may interact with DDT, most notably that involving Fanconi's anemia (FA) gene products [<xref rid="pgen-0020116-b018" ref-type="bibr">18</xref>]. FANCD2 becomes mono-ubiquitinated after DNA damage and localizes to nuclear foci [<xref rid="pgen-0020116-b063" ref-type="bibr">63</xref>]. FANCC has been associated with the TLS polymerases REV1 and REV3 [<xref rid="pgen-0020116-b064" ref-type="bibr">64</xref>] and may also interact with the BLM helicase [<xref rid="pgen-0020116-b065" ref-type="bibr">65</xref>], a candidate for promoting fork reversal in the error-free damage-avoidance pathway [<xref rid="pgen-0020116-b066" ref-type="bibr">66</xref>]. A challenge for future investigations will be to understand how K63-polyUb chain assembly is regulated and how these chains promote interaction with other pathways to mediate error-free recovery from DNA damage.</p></sec><sec id="s4"><title>Materials and Methods</title><sec id="s4a"><title>Cell culture and treatments.</title><p>The construction of the Ub-expressing plasmids has been described elsewhere [<xref rid="pgen-0020116-b031" ref-type="bibr">31</xref>]. The <italic>POLη-GFP</italic> plasmid was a gift of Dr. Alan R. Lehmann, (Genome Damage and Stability Centre, University of Sussex, Falmer, Brighton, United Kingdom). All cell lines were cultured in DMEM (Sigma, St. Louis, Missouri, United States) supplemented with 10% FBS (Sigma). A549 cells were co-transfected with <italic>WT-Ub-GFP</italic> or <italic>K63R-Ub-GFP</italic> plasmids and the <italic>pBabePuro</italic> plasmid (for selection) using FuGene 6 (Roche, Basel, Switzerland). HeLa cells were transfected with <italic>WT-Ub-puro</italic> or <italic>K63R-Ub-puro</italic> constructs using lipofectamine (Invitrogen, Carlsbad, California, United States). Stable transfectants were selected in 1 μg/ml puromycin (Sigma) and/or by flow cytometry (FACSAria, BD Biosciences Pharmingen, San Diego, California, United States).</p><p>The sensitivity to UV irradiation alone, UV combined with caffeine, and cisplatin alone was evaluated using clonogenic survival assays. UV irradiation was performed on 80% confluent cells in 6-cm dishes using a UVC (254-nm) germicidal lamp at a dose rate of 1 J/m<sup>2</sup>/s. UV and caffeine combination studies were carried out as above, but cells were plated in 0.4 mM caffeine immediately after UV irradiation. Cells were treated for 1 h in cisplatin diluted in culture media. Cells were plated in 6-cm dishes in triplicate and incubated for 2 wk to obtain colony formation. Colonies were fixed, stained with 2% bromophenol blue in 70% ethanol, and colonies containing ≥50 cells were counted. All experiments were normalized for plating efficiency.</p><p>The sensitivity to UV irradiation in POLη knockdown cells was performed as above with the exception that cells were transfected twice with SiGenome Smartpool reagent specific for human POLη (Dharmacon Research, Lafayette, Colorada, United States) using oligofectamine (Invitrogen). The transfections were carried out 72 and 24 h before UV treatment to achieve optimal long-term knockdown as determined by quantitative RT-PCR.</p><p>Quantitation of gene expression was performed using an Applied Biosystems (Foster City, California, United States) 7500 Real-Time PCR system using their “assay on demand” technology. <italic>RAD18</italic> expression was determined with the Hs00220119_m1 probe, <italic>POLη</italic> with the Hs00197814 probe, and <italic>18S</italic> with the Hs99999901_s1 probe. Reactions were performed using Taqman Universal PCR Master Mix from Applied Biosystems.</p></sec><sec id="s4b"><title>Immunoblotting.</title><p>Following the indicated treatments with either UV irradiation, cisplatin, and/or SiGenome Smartpool reagent specific for human UBC13 or human RAD18 (Dharmacon), cells were harvested in lysis buffer (20 mM Tris-HCl [pH 7.5], 150 mM NaCl, 1% Triton-X-100, 2 mM EDTA, and 5% glycerol with 200 μg/ml phenylmethylsulfonyl fluoride, 2 mM NaVO<sub>4</sub>, 2 mM NaF, and 2 mM NaPPi protease-inhibitor cocktail). Samples were sonicated, soluble fractions were recovered, and proteins were quantified using the Bradford protein assay (Bio-Rad). Proteins were resolved on either a single-phase (10%) or two-phase SDS-polyacrylamide gel (10% and 15%) and electroblotted onto a Hybond C nitrocellulose membrane (Amersham Pharmacia Biotech, Piscataway, New Jersey, United States). The membrane was stained with Ponceau S (Sigma) prior to Western blotting with the indicated primary antibody. The following antibodies were used: rabbit polyclonal Ub (Dako, Glostrup, Denmark), mouse monoclonal RGS-His (Qiagen, Valencia, California, United States), mouse monoclonal PCNA PC10 (Chemicon, http://www.chemicon.com), rabbit polyclonal GFP (Santa Cruz Biotechnology, Santa Cruz, California, United States), and mouse monoclonal actin (Sigma). Proteins were visualized by a horseradish peroxidase method using ECL (Kirkegaard and Perry Laboratories, <ext-link ext-link-type="uri" xlink:href="http://www.kpl.com">http://www.kpl.com</ext-link>).</p></sec><sec id="s4c"><title>Immunoprecipitation.</title><p>Cells were UV-irradiated with 30 J/m<sup>2</sup> as described above and either left untreated or transfected with SiGenome Smartpool reagent specific for human UBC13 or human RAD18 (Dharmacon). Cells were lysed (6 h after irradiation) in lysis buffer supplemented with 0.5% SDS. Lysates were sonicated and boiled for 5 min followed by dilution to 0.1% SDS. After protein quantitation, 500 μg of protein was incubated overnight at 4 °C with anti-PCNA (1/200). The following day, lysates were incubated for 48 h at 4 °C with 100 μl of Gamma-Bound Sepharose Beads (Amersham Pharmacia Biotech). Beads were washed extensively in lysis buffer, and proteins were eluted by boiling in Laemmli's SDS sample buffer. Immunoblotting was performed as described above except that the membranes were autoclaved for 20 min in ddH<sub>2</sub>O after protein transfer, and proteins were visualized using SuperSignal West Pico Chemiluminescent Substrate (Pierce Biotechnology, Rockford, Illinois, United States).</p></sec><sec id="s4d"><title>Mutation spectrum.</title><p>To eliminate background <italic>HPRT</italic> mutations, cells were cultured in hypoxanthine, aminopterin, and thymidine (HAT)–supplemented culture medium for 1 wk. UV-induced <italic>HPRT</italic> mutants were obtained by seeding 1.5 × 10<sup>4</sup> cells in 24-well plates, followed by 10 J/m<sup>2</sup> UV irradiation 24 h later. Cells were subcultured for 7 d, and re-seeded at 5.0 × 10<sup>4</sup> cells on 35-mm dishes in medium containing 30 μM 6-thioguanine (6-TG). Individual colonies were picked and grown until enough cells were obtained for RNA isolation using RNA-aqueous kit (Ambion, Austin, Texas, United States). The <italic>HPRT</italic> gene was subjected to RT-PCR, followed by sequencing using the following overlapping primers: HPRT1–5′CTTCCTCCTCCTGAGCAGTC3′; HPRT2–5′AAGCAGATGGCCACAGAACT3′; HPRT3–5′CCTGGCGTCGTGATTAGTG3′; HPRT4–5′TTTACTGGCGATGTCAATAGGA3′; HPRT5–5′GACCAGTCAACAGGGGACAT3′; and HPRT6 5′ATGTCCCCTGTTGACTGGTC3′.</p></sec><sec id="s4e"><title>Mutation frequency.</title><p> <italic>HPRT</italic> mutant–free cells (1.0 × 10<sup>6</sup>) were seeded and treated the following day with either UV irradiation (0, 5, and 10 J/m<sup>2</sup>) or cisplatin (0, 5, and 10 μM for 1 h). After subculturing the treated cells for 1 wk, 4.0 × 10<sup>5</sup> cells were seeded in selective medium containing 6-TG (as above) and incubated until colonies were formed. Colonies were counted and <italic>HPRT</italic> mutation frequency was defined after correcting for plating efficiency.</p><p>Mutation frequency in response to UV treatment in POLη and RAD18 knockdown cells was performed as above with the exception that cells were transfected twice with SiGenome Smartpool reagent specific for human POLη or human RAD18 (Dharmacon) using oligofectamine. The transfections were performed 72 and 24 h before UV treatment to achieve optimal long-term knockdown as determined by quantitative PCR.</p></sec><sec id="s4f"><title>Foci.</title><p>A549 and Hela cells stably expressing <italic>WT-Ub-puro</italic> and <italic>K63R-Ub-puro</italic> were transiently transfected with a <italic>POLη-GFP</italic> plasmid. Twenty-four hours post-transfection, cells were UV-irradiated at a dose of 10 J/m<sup>2</sup>. To observe living cells, cells were cultured in 35-mm glass-bottomed dishes (MatTek, <ext-link ext-link-type="uri" xlink:href="http://www.mattek.com">http://www.mattek.com</ext-link>) with coverslips. Real-time excitation measurements to monitor fluorescent signals in transfected cells were subsequently performed using a live-cell microscopy unit mounted on a Leica DR IRBE inverted microscope (Wetzlar, Germany), equipped with a polychromator that allows generation of light of the required wavelength, using a 63× objective. Both the polychromator and filterwheel were controlled via the PC using specialized Openlab software from Improvision (<ext-link ext-link-type="uri" xlink:href="http://www.improvision.com/products/openlab">http://www.improvision.com/products/openlab</ext-link>). At least 100 cells were counted for each cell line at each time point per experiment by a blinded independent observer.</p><p>The recruitment of POLη to foci was determined in response to UV irradiation in RAD18 knockdown cells performed as above with the exception that cells were transfected twice with SiGenome Smartpool reagent specific for human RAD18 using oligofectamine. The transfections were performed 72 and 24 h before UV treatment to achieve optimal long-term knockdown as determined by quantitative PCR.</p><p>For colocalization studies, Hela cells stably expressing <italic>WT-Ub-puro</italic> and <italic>K63R-Ub-puro</italic> were transiently transfected with a <italic>POLη-GFP</italic> plasmid in a chamber slide (BD Biosciences Pharmingen). Twenty-four hours post-transfection, cells were UV-irradiated at a dose of 10 J/m<sup>2</sup>. For detection of PCNA and POLη, cells were fixed in cold methanol for 20 min at −20 °C followed by 30 sec in cold acetone. Cells were washed twice with PBS and then incubated at room temperature with both anti-PCNA and anti-POLη. After 1 h, cells were washed with PBS and then incubated with FITC-conjugated goat antimouse IgG (Invitrogen) and Texas red–conjugated goat antirabbit (Invitrogen), for 45 min. After washing in PBS, cells were dehydrated for 1 min in 70% ethanol followed by two 1-min incubations in 100% ethanol. Cells were then mounted with Fluorescent Mounting Media (Dako) and visualized by confocal microscopy.</p></sec></sec><sec sec-type="supplementary-material" id="s5"><title>Supporting Information</title><supplementary-material content-type="local-data" id="pgen-0020116-sg001"><label>Figure S1</label><caption><title>Knockdown of UBC13, RAD18, and POLη</title><p>(A) Hela cells were transfected with siRNA against UBC13 and analyzed by Western blot.</p><p>(B) A549 cells were transfected twice (48 h apart) with siRNA against RAD18 or POLη. Knockdown was analyzed 24 and 72 h post–second transfection for mRNA expression relative to 18S rRNA using quantitative RT-PCR.</p><p>(28 KB PDF)</p></caption><media xlink:href="pgen.0020116.sg001.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020116-sg002"><label>Figure S2</label><caption><title>Increase in POLη Foci Is Also Observed in A549 Cells</title><p>A549 cells stably expressing <italic>WT-Ub</italic> or <italic>K63R-Ub</italic> were treated as described in <xref ref-type="fig" rid="pgen-0020116-g004">Figure 4</xref>. Two independent experiments were performed.</p><p>(16 KB PDF)</p></caption><media xlink:href="pgen.0020116.sg002.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020116-sg003"><label>Figure S3</label><caption><title>Cell-Cycle Profile following UV Treatment</title><p>(A) A549 cells expressing <italic>WT-Ub-GFP</italic> or <italic>K63R-Ub-GFP</italic> were treated with the indicated dose of UV irradiation.</p><p>(B) Hela cells expressing <italic>WT-Ub-puro</italic> or <italic>K63R-Ub-puro</italic> were treated with 10 J/m<sup>2</sup> UV irradiation and fixed either immediately or 6 h posttreatment. Following propidium iodide staining, cells were analyzed for DNA content using a FACSAria flow cytometer (BD Biosciences Pharmingen).</p><p>(33 KB PDF)</p></caption><media xlink:href="pgen.0020116.sg003.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020116-sg004"><label>Figure S4</label><caption><title>Modification of PCNA by Polyubiquitin in Human Cells after DNA Damage</title><p>(A) Hela cells were subjected to the same procedure as carrried out in <xref ref-type="fig" rid="pgen-0020116-g005">Figure 5</xref>B<bold>.</bold> A darker (upper panel) and lighter exposure (lower panel) of the PCNA immunoblot is shown.</p><p>(B) A549 cells were UV-irradiated as described in <xref ref-type="fig" rid="pgen-0020116-g005">Figure 5</xref>A and lysed at the indicated times posttreatment. Whole-cell lysates were subjected to immunoprecipitation with an anti-PCNA antibody followed by immunoblotting for PCNA (upper panel) and Ub (lower panel). The controls in the immunoprecipitations are the same as carried out in <xref ref-type="fig" rid="pgen-0020116-g005">Figure 5</xref>C. −UV indicates no UV treatment.</p><p>(C) Hela cells were subjected to the same procedure as performed in <xref ref-type="fig" rid="pgen-0020116-g005">Figure 5</xref>D. A lighter exposure of the PCNA IP immunoblotted for Ub is shown. A PCNA immunoblot with darker and lighter exposure performed on protein lysates from the same samples used in the immunoprecipitations is also shown.</p><p>(67 KB PDF)</p></caption><media xlink:href="pgen.0020116.sg004.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020116-sg005"><label>Figure S5</label><caption><title>Cisplatin Treatment also Induces Modification of PCNA by Polyubiquitin in Human Cells</title><p>Untreated, 30 J/m<sup>2</sup> UV-irradiated, and 160 μM cisplatin-treated A549 and Hela cells were lysed 6 h posttreatment followed by immunoblotting for PCNA.</p><p>(33 KB PDF)</p></caption><media xlink:href="pgen.0020116.sg005.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pgen-0020116-sg006"><label>Figure S6</label><caption><title>Inhibition of DUB Enzymes Does Not Affect Appearance of PolyUb-PCNA</title><p>A549, 293T, and Hela cells were treated with 30 J/m<sup>2</sup> UV irradiation and lysed in the presence or absence of the general thiol protease-inhibitor NEM. Immunoprecipitation and Western blots were carried out as described in Materials and Methods. The controls in the immunoprecipitations were “no 1”, in which lysates were incubated with beads but no PCNA antibody, and “1 B”, in which PCNA antibody was incubated with beads alone.</p><p>(30 KB PDF)</p></caption><media xlink:href="pgen.0020116.sg006.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><sec id="s5a"><title>Accession Numbers</title><p>The Entrez Gene (<ext-link ext-link-type="uri" xlink:href="http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?CMD=search&DB=gene">http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?CMD=search&DB=gene</ext-link>) accession numbers for the gene and gene products discussed in this paper are BLM (641), CROC1 (7335), FANCC (2176), FANCD2 (2177), HPRT (3251), MMS2 (7336), NEMO (8517), NF-kappaB (4790), PCNA (5111), POLH (5429), POLI (11201), RAD18 (56852), RAD5 (850719), REV3 (5980), TRAF6 (7189), UBA52 (7311), UBA80 (6233), UBB (7314), UBC (7316), and UBC13 (7334).</p></sec></sec>
The Ancestral Eutherian Karyotype Is Present in Xenarthra	<p>Molecular studies have led recently to the proposal of a new super-ordinal arrangement of the 18 extant Eutherian orders. From the four proposed super-orders, Afrotheria and Xenarthra were considered the most basal. Chromosome-painting studies with human probes in these two mammalian groups are thus key in the quest to establish the ancestral Eutherian karyotype. Although a reasonable amount of chromosome-painting data with human probes have already been obtained for Afrotheria, no Xenarthra species has been thoroughly analyzed with this approach. We hybridized human chromosome probes to metaphases of species <italic>(Dasypus novemcinctus, Tamandua tetradactyla,</italic> and <italic>Choloepus hoffmanii)</italic> representing three of the four Xenarthra families<italic>.</italic> Our data allowed us to review the current hypotheses for the ancestral Eutherian karyotype, which range from 2<italic>n</italic> = 44 to 2<italic>n</italic> = 48. One of the species studied, the two-toed sloth <named-content content-type="genus-species">C. hoffmanii</named-content> (2<italic>n</italic> = 50), showed a chromosome complement strikingly similar to the proposed 2<italic>n</italic> = 48 ancestral Eutherian karyotype, strongly reinforcing it.</p>	<contrib contrib-type="author"><name><surname>Svartman</surname><given-names>Marta</given-names></name><xref ref-type="corresp" rid="cor1">*</xref><xref ref-type="aff" rid="aff1"/></contrib><contrib contrib-type="author"><name><surname>Stone</surname><given-names>Gary</given-names></name><xref ref-type="aff" rid="aff1"/></contrib><contrib contrib-type="author"><name><surname>Stanyon</surname><given-names>Roscoe</given-names></name><xref ref-type="aff" rid="aff1"/></contrib>	PLoS Genetics	<sec id="s1"><title>Introduction</title><p>Extensive molecular data on mammalian genomes have led recently to the proposal of a new phylogenetic tree for Eutherians, which encompasses four super-orders: Afrotheria (elephants, manatees, hyraxes, tenrecs, aardvark, and elephant shrews); Xenarthra (sloths, anteaters, and armadillos); Euarchontoglires (rodents and lagomorphs as a sister taxon to primates, flying lemurs, and tree shrews); and Laurasiatheria (cetaceans, artiodactyls, perissodactyls, carnivores, pangolins, bats, and insectivores) [<xref rid="pgen-0020109-b001" ref-type="bibr">1</xref>,<xref rid="pgen-0020109-b002" ref-type="bibr">2</xref>]. At first very debated, the grouping of the 18 living placental mammalian orders into these four clades is rapidly approaching a consensus [<xref rid="pgen-0020109-b003" ref-type="bibr">3</xref>]. One of the pivotal questions still remaining relates to the root of the placental tree. Either Xenarthra or Afrotheria, or the combination Xenarthra plus Afrotheria, could represent the first split from a common Eutherian ancestor (reviewed in [<xref rid="pgen-0020109-b003" ref-type="bibr">3</xref>]). Although Afrotheria has been favored as the first offshoot in the Eutherian tree [<xref rid="pgen-0020109-b002" ref-type="bibr">2</xref>], both Xenarthra and Afrotheria emerged around 100 million years ago (mya), within a short interval from each other [<xref rid="pgen-0020109-b004" ref-type="bibr">4</xref>,<xref rid="pgen-0020109-b005" ref-type="bibr">5</xref>]. This is also the time of the separation between South America, where Xenarthrans are endemic, and Africa, home to the Afrotheria. Therefore, a causal connection between the plate-tectonic events and the diversification of mammals has been proposed [<xref rid="pgen-0020109-b001" ref-type="bibr">1</xref>,<xref rid="pgen-0020109-b003" ref-type="bibr">3</xref>].</p><p>Extant Xenarthra (sloths, anteaters, and armadillos) are restricted to the Americas, mostly Central and South America. The 30 known living species are relicts of the once highly diversified Xenarthra that dominated the South American fauna from the late Cretaceous (80–65 mya) until the late Tertiary (3–2.5 mya), while South America was an isolated landmass. Xenarthrans were abundant until the Pleistocene (10,000 years ago), and more than 200 fossil genera have been reported [<xref rid="pgen-0020109-b005" ref-type="bibr">5</xref>].</p><p>Morphological and molecular studies support the monophyly of the group, which also extends to each of the four recognized extant families: Myrmecophagidae (four species of anteater), Bradypodidae (four species of three-toed sloth), Megalonychidae (two species of two-toed sloth), and Dasypodidae (about 20 species of armadillo) [<xref rid="pgen-0020109-b006" ref-type="bibr">6</xref>–<xref rid="pgen-0020109-b009" ref-type="bibr">9</xref>]. According to molecular data estimates, the radiation of Xenarthra occurred around 65 mya, during the Cretaceous/Tertiary boundary. Armadillos, the most basal group of Xenarthra, diverged around 65 mya, followed by the divergence between anteaters and sloths around 55 mya [<xref rid="pgen-0020109-b005" ref-type="bibr">5</xref>].</p><p>Molecular data on Xenarthra are rapidly accumulating in the literature [<xref rid="pgen-0020109-b005" ref-type="bibr">5</xref>–<xref rid="pgen-0020109-b009" ref-type="bibr">9</xref>], but only scarce information is available on their chromosomes. Nineteen species had their karyotypes described [<xref rid="pgen-0020109-b010" ref-type="bibr">10</xref>], but only a few with banding techniques [<xref rid="pgen-0020109-b011" ref-type="bibr">11</xref>–<xref rid="pgen-0020109-b013" ref-type="bibr">13</xref>]. The reported diploid numbers range from 2<italic>n</italic> = 38 in <named-content content-type="genus-species">Tolypeutes matacus</named-content> to 2<italic>n</italic> = 64 in <italic>Dasypus novemcinctus, D</italic>. <italic>hybridus,</italic> and <named-content content-type="genus-species">Cyclopes didactyla</named-content> [<xref rid="pgen-0020109-b011" ref-type="bibr">11</xref>]. The first comparative study between karyotypes of different Xenarthra species using molecular cytogenetics has recently been published, and the rate of chromosome repatterning in Xenarthra was considered low [<xref rid="pgen-0020109-b013" ref-type="bibr">13</xref>].</p><p>Chromosome painting to establish the karyotype of a common ancestral Eutherian mammal has already been applied in 12 out of the 18 orders in the group. The proposed diploid numbers range from 2<italic>n</italic> = 44 to 2<italic>n</italic> = 50 [<xref rid="pgen-0020109-b014" ref-type="bibr">14</xref>–<xref rid="pgen-0020109-b021" ref-type="bibr">21</xref>]. The most recent studies suggest 2<italic>n</italic> = 44 [<xref rid="pgen-0020109-b019" ref-type="bibr">19</xref>], 2<italic>n</italic> = 46 [<xref rid="pgen-0020109-b020" ref-type="bibr">20</xref>], and 2<italic>n</italic> = 48 [<xref rid="pgen-0020109-b017" ref-type="bibr">17</xref>,<xref rid="pgen-0020109-b021" ref-type="bibr">21</xref>]. Chromosome-painting studies in Afrotheria and Xenarthra, the most basal of the living Eutherians, are essential to establish the putative ancestral Eutherian complement. Some Afrotheria have already had their genomes analyzed by chromosome painting with human probes [<xref rid="pgen-0020109-b019" ref-type="bibr">19</xref>–<xref rid="pgen-0020109-b022" ref-type="bibr">22</xref>]. Data of comparative molecular cytogenetics in Xenarthra using human chromosomes as painting probes are restricted to very limited data for two species [<xref rid="pgen-0020109-b017" ref-type="bibr">17</xref>,<xref rid="pgen-0020109-b018" ref-type="bibr">18</xref>,<xref rid="pgen-0020109-b023" ref-type="bibr">23</xref>].</p><p>In this study, we hybridized paints of all human chromosomes in species representing three of the four families of Xenarthra, and reevaluated the current hypotheses for the ancestral Eutherian karyotype based on our results.</p></sec><sec id="s2"><title>Results</title><p>Some examples of human chromosome paints hybridized to the three Xenarthra species are shown in <xref ref-type="fig" rid="pgen-0020109-g001">Figure 1</xref>. The results of the chromosome painting with human probes in the three species are summarized in <xref ref-type="table" rid="pgen-0020109-t001">Table 1</xref>. The G-banded karyotype of the three species studied were mounted according to Jorge et al. [<xref rid="pgen-0020109-b011" ref-type="bibr">11</xref>] and are presented with the corresponding human chromosome segments in <xref ref-type="fig" rid="pgen-0020109-g002">Figure 2</xref> <italic>(D. novemcinctus</italic> and <italic>Tamandua tetradactyla)</italic> and <xref ref-type="fig" rid="pgen-0020109-g003">Figure 3</xref>A <italic>(Choloepus hoffmanii).</italic> </p><fig id="pgen-0020109-g001" position="float"><label>Figure 1</label><caption><title>Partial Metaphases of <italic>D. novemcinctus, T. tetradactyla,</italic> and <named-content content-type="genus-species">C. hoffmanii</named-content> after In Situ Hybridization with Human Chromosome–Specific Probes</title><p>(A–C) <italic>D. novemcinctus,</italic> (D–F) <italic>T. tetradactyla,</italic> and (G–I) <named-content content-type="genus-species">C. hoffmanii</named-content>. Two probes were used in each experiment, and the hybridizations were detected with avidin–FITC (green) and antidigoxigenin–rhodamine (red). The chromosomes were counterstained with DAPI, and the human chromosome probes used are indicated.</p></caption><graphic xlink:href="pgen.0020109.g001"/></fig><table-wrap id="pgen-0020109-t001" content-type="2col" position="float"><label>Table 1</label><caption><p>Results of Chromosome Painting with Human Probes in Xenarthra</p></caption><graphic xlink:href="pgen.0020109.t001"/></table-wrap><p>The X chromosome of the three Xenarthra species hybridized with human chromosome X and showed a highly conserved G-banding pattern when compared to the human X chromosome, reinforcing its known conservation. No signals were obtained with human chromosome Y.</p><p>In the nine-banded armadillo <named-content content-type="genus-species">D. novemcinctus</named-content> (2<italic>n</italic> = 64; fundamental number [<italic>FN</italic>] = 78), chromosome painting resulted in 41 segments, and seven associations, or combinations of segments from two human chromosomes, were observed (<xref ref-type="fig" rid="pgen-0020109-g002">Figure 2</xref>A).</p><fig id="pgen-0020109-g002" position="float"><label>Figure 2</label><caption><title>G-Banded Karyotypes of Female Nine-Banded Armadillo and Female Lesser Anteater</title><p>(A) Female nine-banded armadillo <named-content content-type="genus-species">D. novemcinctus</named-content> (2<italic>n</italic> = 64) and (B) female lesser anteater <named-content content-type="genus-species">T. tetradactyla</named-content> (2<italic>n</italic> = 54).</p><p>The correspondence to human chromosomes as revealed by chromosome painting is shown on the left of each chromosome. The question mark indicates questionable results. Some regions were not painted by any human probe and are probably composed of repetitive sequences (see text).</p></caption><graphic xlink:href="pgen.0020109.g002"/></fig><p>The karyotype of the lesser anteater <named-content content-type="genus-species">T. tetradactyla</named-content> (2<italic>n</italic> = 54; <italic>FN</italic> = 104) consists exclusively of biarmed chromosomes. Painting with human probes resulted in a total of 44 segments, and 15 associations were observed (<xref ref-type="fig" rid="pgen-0020109-g002">Figure 2</xref>B).</p><p>The two-toed sloth <named-content content-type="genus-species">C. hoffmanii</named-content> is the species with the lowest diploid number (2<italic>n</italic> = 50; <italic>FN</italic> = 61). Owing to a heterozygous translocation between pairs 2 and 24, pair 2 was heteromorphic in the cells analyzed; this individual is thus trisomic for Chromosome 24. A total of 32 segments and six associations were detected with the human probes (<xref ref-type="fig" rid="pgen-0020109-g003">Figure 3</xref>A).</p><fig id="pgen-0020109-g003" position="float"><label>Figure 3</label><caption><title>The Karyotype of Hoffmann's Two-Toed Sloth (2<italic>n</italic> = 50) Closely Resembles the Proposed Eutherian Ancestral Karyotype (2<italic>n</italic> = 48)</title><p>(A) G-banded karyotype of a male <named-content content-type="genus-species">C. hoffmannii</named-content> (2<italic>n</italic> = 50) with the corresponding human chromosomes indicated to the left of each chromosome. (B) Diagram of <named-content content-type="genus-species">C. hoffmannii</named-content> chromosomes. (C) The proposed ancestral Eutherian karyotype (2<italic>n</italic> = 48) [<xref rid="pgen-0020109-b017" ref-type="bibr">17</xref>,<xref rid="pgen-0020109-b021" ref-type="bibr">21</xref>]. In (B) and (C), the corresponding human chromosomes are each represented by a different color, and their numbers are indicated at the bottom of each chromosome. The chromosomes are drawn only roughly to scale. In (C), the corresponding <named-content content-type="genus-species">C. hoffmannii</named-content> chromosomes are indicated to the left of each ancestral chromosome. The asterisks and lines show the chromosomes that differ between the karyotypes of <named-content content-type="genus-species">C. hoffmanii</named-content> (CHO) and those proposed as ancestral (ANC) for Eutheria (CHO 14 and ANC 10q and 10p [black asterisks]; CHO 13 and 19 and ANC 8q [green asterisks]; and CHO 20 and 21 and ANC 7b/16p [red asterisks]).</p></caption><graphic xlink:href="pgen.0020109.g003"/></fig><p>It is interesting to note that some chromosome regions of the three Xenarthra species were not labeled by any of the human probes (<xref ref-type="fig" rid="pgen-0020109-g002">Figures 2</xref> and <xref ref-type="fig" rid="pgen-0020109-g003">3</xref>A). These regions, which were either proximal (for instance <named-content content-type="genus-species">D. novemcinctus</named-content> Chromosomes 25 and 30), distal (<named-content content-type="genus-species">T. tetradactyla</named-content> Chromosome 1 and <named-content content-type="genus-species">C. hoffmanii</named-content> Chromosome 7), or encompassed whole short arms (as with <named-content content-type="genus-species">D. novemcinctus</named-content> Chromosomes 2 and 4, <named-content content-type="genus-species">T. tetradactyla</named-content> Chromosome 12, and <named-content content-type="genus-species">C. hoffmanii</named-content> Chromosomes 3, 4, and 5), are very likely to represent repetitive sequences that are absent from the human genome.</p></sec><sec id="s3"><title>Discussion</title><p>The effort to establish the ancestral Eutherian karyotype relies on the concept of parsimony, according to which likely ancestral chromosomes are present in species of divergent Eutherian orders. There is a conspicuous lack of data from Xenarthra, one of the most basal mammalian groups, in the quest to establish the ancestral karyotype. Aiming to fill this gap, we conducted painting experiments with human chromosome–specific probes in species representing three of the four major Xenarthra groups.</p><p>Richard et al. [<xref rid="pgen-0020109-b018" ref-type="bibr">18</xref>,<xref rid="pgen-0020109-b023" ref-type="bibr">23</xref>] previously reported results for ten human chromosomes (3, 7, 12, 14, 15, 16, 18, 21, 22, and the X chromosome) in <named-content content-type="genus-species">D. novemcinctus</named-content> metaphases. The armadillo chromosomes labeled were not specified, and three of the probes (human Chromosomes 7, 12, and 22) gave results that were questioned by Richard et al. In the three cases, only one signal was observed, while the same probes labeled two (human Chromosomes 7 and 22) or three (human Chromosome 12) autosomes in our sample. We obtained the same number of signals already reported for human Chromosomes 3, 14, 15, 16, 18, and the X chromosome [<xref rid="pgen-0020109-b018" ref-type="bibr">18</xref>], but we detected two signals for human Chromosome 21, instead of the single labeling described by Richard et al. [<xref rid="pgen-0020109-b018" ref-type="bibr">18</xref>]. Human Chromosome 1 was previously shown to be conserved in <named-content content-type="genus-species">C. hoffmanii</named-content> [<xref rid="pgen-0020109-b017" ref-type="bibr">17</xref>], and this was confirmed in our experiments.</p><p>Six human chromosome associations (human Chromosomes 3/21, 4/8, 7/16, 12/22 twice, 14/15, and 16/19) are agreed, by most authors, as having been present in the ancestral Eutherian karyotype. All of these syntenies were found in the three species of Xenarthra that we studied (<xref ref-type="table" rid="pgen-0020109-t001">Table 1</xref>). The associations human Chromosomes 3/21 and 14/15 were each disrupted in, respectively, <named-content content-type="genus-species">D. novemcinctus</named-content> and <named-content content-type="genus-species">T. tetradactyla</named-content> (<xref ref-type="fig" rid="pgen-0020109-g001">Figure 1</xref>F), suggesting that they underwent rearrangements in these species after the split from a common ancestor. In <italic>C. hoffmanii,</italic> no other syntenies besides the six already mentioned were found, and in <italic>D. novemcinctus,</italic> a single extra combination (human Chromosome 10/12) was identified. Human Chromosomes 9, 13, 17, 18, and 20 were found to be conserved in toto in the three Xenarthra species.</p><p>Although Xenarthra is widely accepted as a monophyletic group on morphological and molecular grounds, no common chromosome syntenies besides those considered present in the common Eutherian ancestral karyotype were found in the three species studied. Thus, there is an unusual absence of chromosome synapomorphic traits linking the three families that we studied. The only exception might be human Chromosome 8, believed to be disrupted into two blocks in a common Eutherian ancestral and found as three segments in both <named-content content-type="genus-species">D. novemcinctus</named-content> and <named-content content-type="genus-species">T. tetradactyla</named-content> and possibly also in <named-content content-type="genus-species">C. hoffmannii</named-content> (<xref ref-type="table" rid="pgen-0020109-t001">Table 1</xref>). Additional experiments are needed to confirm this observation.</p><p>On the other hand, we observed a striking resemblance of Xenarthran complements to the proposed 2<italic>n</italic> = 48 ancestral Eutherian karyotype [<xref rid="pgen-0020109-b017" ref-type="bibr">17</xref>,<xref rid="pgen-0020109-b021" ref-type="bibr">21</xref>], particularly in the case of <named-content content-type="genus-species">C. hoffmannii</named-content> (2<italic>n</italic> = 50) (<xref ref-type="fig" rid="pgen-0020109-g003">Figure 3</xref>B). In this species, 20 autosome pairs appear to be the same as the most recent hypothesis for the putative Eutherian ancestral karyotype with 2<italic>n</italic> = 48 (human Chromosomes 1, 2p, 2q, 3/21, 4/8p, 5, 6, 7a, 7b/16p, 9, 11, 12/22 twice, 13, 14/15, 16q/19q, 17, 18, 19p, and 20). The only differences are with regard to human Chromosome 10, considered as two blocks in the ancestral karyotype and kept intact in this species; human Chromosome 8q, which might be further disrupted in <named-content content-type="genus-species">C. hoffmannii</named-content> resulting in a third signal for this chromosome; and an additional signal for human Chromosome 16, represented by <named-content content-type="genus-species">C. hoffmanii</named-content> Chromosome 20 (<xref ref-type="fig" rid="pgen-0020109-g003">Figure 3</xref>).</p><p>In <italic>D. novemcinctus,</italic> some of the putative ancestral Eutherian chromosomes are disrupted into further blocks (for instance, human Chromosomes 1, 3/21, 6, 7, and 11), which would account for its higher diploid number (2<italic>n</italic> = 64) in relation to an ancestral karyotype, but otherwise this chromosome complement is also very close to that considered to be the ancestral one.</p><p> <named-content content-type="genus-species">T. tetradactyla</named-content> presents the most rearranged complement in relation to the ancestral karyotype, and was also reported as such when compared to other Xenarthra taxa [<xref rid="pgen-0020109-b013" ref-type="bibr">13</xref>]. In this case, several rearrangements seem to have occurred, resulting in chromosome disruptions and new associations in relation to a common ancestral complement. This species presented two associations previously reported in Afrotheria (human Chromosomes 1/19 and 2/8). One of them, human Chromosome 1/19, was considered as a synapomorphic trait of Afrotheria [<xref rid="pgen-0020109-b021" ref-type="bibr">21</xref>] and was postulated as ancestral to all Eutheria by Yang et al. [<xref rid="pgen-0020109-b019" ref-type="bibr">19</xref>]. Nevertheless, its occurrence in only one highly rearranged primate species [<xref rid="pgen-0020109-b024" ref-type="bibr">24</xref>] outside Afrotheria was considered insufficient to accept it as ancestral [<xref rid="pgen-0020109-b021" ref-type="bibr">21</xref>]. As human Chromosome 1/19 was found only in one of the Xenarthrans herein analyzed, and it is likely to be a different association regarding the segments involved, we do not consider it as a probable ancestral Eutherian combination.</p><p>Human Chromosome 2/8 was absent in elephants, but was found in four other species of Afrotheria that were analyzed with chromosome painting [<xref rid="pgen-0020109-b019" ref-type="bibr">19</xref>–<xref rid="pgen-0020109-b022" ref-type="bibr">22</xref>]. It was thus considered a phylogenetic link between those species after the divergence of the elephant [<xref rid="pgen-0020109-b021" ref-type="bibr">21</xref>]. Human Chromosome 2/8 was observed only in one species of Xenarthra, and it is not clear whether the segments involved are the same as those detected in Afrotheria, which rules out this association as a common ancestral one.</p><p>An alternative putative ancestral karyotype with 2<italic>n</italic> = 46 was proposed, in which the combination human Chromosome 10/12/22 would be present [<xref rid="pgen-0020109-b020" ref-type="bibr">20</xref>,<xref rid="pgen-0020109-b025" ref-type="bibr">25</xref>]. Although we observed an association human Chromosome 10/12 in <italic>D. novemcinctus,</italic> we could not detect it in the other two species. As previously discussed [<xref rid="pgen-0020109-b021" ref-type="bibr">21</xref>], the ancestral status of human Chromosome 10/12/22, which seems to be widespread in Afrotheria and Carnivores, remains doubtful. Our new data on Xenarthra do not lend support to this association as an ancestral Eutherian karyotype feature, thus reinforcing the 2<italic>n</italic> = 48 ancestral karyotype hypothesis.</p><p>The status of Xenarthra as a basal group in the Eutherian phylogenetic tree is still disputed on molecular grounds [<xref rid="pgen-0020109-b003" ref-type="bibr">3</xref>]. Based on the chromosome-painting results, it is clear that no Afrotheria has a complement so close to the putative ancestral Eutherian karyotype as the one we found herein in Xenarthra. Thus, Xenarthra cannot yet be ruled out as the most basal Eutheria, and more data should be gathered on the two basal groups before a more definitive answer is reached.</p><p>A final conclusion on the ancestral Eutherian karyotype will ultimately depend on comparisons with the most suitable outgroup, represented by marsupials. Technical constraints currently prevent cross-species chromosome painting between such distantly related groups as Eutherians and marsupials.</p><p>It is important to point out that the resolution limit of interspecies chromosome painting is about 5 Mbp. Another constraint of this approach is that it cannot detect intrachromosomal rearrangements, as inversions, that result in gene-order changes. Recent comparative studies encompassing results obtained from chromosome painting, radiation hybrid maps, and whole-genome sequencing of various mammalian groups show the potential of this broad kind of study to reveal conserved genome segments in mammals as well as the breakpoints involved in chromosome repatterning at a much more detailed level [<xref rid="pgen-0020109-b025" ref-type="bibr">25</xref>,<xref rid="pgen-0020109-b026" ref-type="bibr">26</xref>]. This approach promises ultimately to allow the reconstruction of the ancestral Eutherian karyotype, as whole-genome sequencing data become available for more mammalian species, including marsupials and monotremes, in the not too distant future.</p></sec><sec id="s4"><title>Materials and Methods</title><p>Chromosome preparations were obtained from cultured fibroblasts of a female nine-banded armadillo <named-content content-type="genus-species">D. novemcinctus</named-content> (2<italic>n</italic> = 64), a male two-toed sloth <named-content content-type="genus-species">C. hoffmanii</named-content> (2<italic>n</italic> = 50), and a female lesser anteater <named-content content-type="genus-species">T. tetradactyla</named-content> (2<italic>n</italic> = 54). The cells from the two former species were kindly supplied by the Center for Reproduction of Endangered Species (CRES) of the San Diego Zoo, California, United States. The lesser anteater cells were purchased from ATCC (Manassas, Virginia, United States). Cells were cultivated in αDMEM with 10% fetal bovine serum at 35 °C. Routine procedures were used for chromosome preparations. GTG banding was performed as described [<xref rid="pgen-0020109-b027" ref-type="bibr">27</xref>] with modifications.</p><p>Human chromosome–specific probes were obtained by flow sorting, and DOP-PCR as well as the interspecific in situ hybridization experiments were performed as already described [<xref rid="pgen-0020109-b021" ref-type="bibr">21</xref>].</p></sec>
Extracting Gene Networks for Low-Dose Radiation Using Graph Theoretical Algorithms	<p>Genes with common functions often exhibit correlated expression levels, which can be used to identify sets of interacting genes from microarray data. Microarrays typically measure expression across genomic space, creating a massive matrix of co-expression that must be mined to extract only the most relevant gene interactions. We describe a graph theoretical approach to extracting co-expressed sets of genes, based on the computation of cliques. Unlike the results of traditional clustering algorithms, cliques are not disjoint and allow genes to be assigned to multiple sets of interacting partners, consistent with biological reality. A graph is created by thresholding the correlation matrix to include only the correlations most likely to signify functional relationships. Cliques computed from the graph correspond to sets of genes for which significant edges are present between all members of the set, representing potential members of common or interacting pathways. Clique membership can be used to infer function about poorly annotated genes, based on the known functions of better-annotated genes with which they share clique membership (i.e., “guilt-by-association”). We illustrate our method by applying it to microarray data collected from the spleens of mice exposed to low-dose ionizing radiation. Differential analysis is used to identify sets of genes whose interactions are impacted by radiation exposure. The correlation graph is also queried independently of clique to extract edges that are impacted by radiation. We present several examples of multiple gene interactions that are altered by radiation exposure and thus represent potential molecular pathways that mediate the radiation response.</p>	<contrib contrib-type="author"><name><surname>Voy</surname><given-names>Brynn H</given-names></name><xref ref-type="aff" rid="aff1">1</xref><xref ref-type="corresp" rid="cor1">*</xref></contrib><contrib contrib-type="author"><name><surname>Scharff</surname><given-names>Jon A</given-names></name><xref ref-type="aff" rid="aff2">2</xref></contrib><contrib contrib-type="author"><name><surname>Perkins</surname><given-names>Andy D</given-names></name><xref ref-type="aff" rid="aff2">2</xref></contrib><contrib contrib-type="author"><name><surname>Saxton</surname><given-names>Arnold M</given-names></name><xref ref-type="aff" rid="aff3">3</xref></contrib><contrib contrib-type="author"><name><surname>Borate</surname><given-names>Bhavesh</given-names></name><xref ref-type="aff" rid="aff2">2</xref></contrib><contrib contrib-type="author"><name><surname>Chesler</surname><given-names>Elissa J</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Branstetter</surname><given-names>Lisa K</given-names></name><xref ref-type="aff" rid="aff1">1</xref></contrib><contrib contrib-type="author"><name><surname>Langston</surname><given-names>Michael A</given-names></name><xref ref-type="aff" rid="aff2">2</xref></contrib>	PLoS Computational Biology	<sec id="s1"><title>Introduction</title><p>“Guilt-by-association,” the assumption that genes with similar expression patterns participate in common cellular functions, drives a growing body of effort to extract cellular pathways from microarray data [<xref rid="pcbi-0020089-b001" ref-type="bibr">1</xref>–<xref rid="pcbi-0020089-b004" ref-type="bibr">4</xref>]. The general tenet is that genes encoding proteins participating in a common pathway will display correlated expression levels when analyzed at sufficient scale, and that the identities and known functions of these genes can be used to highlight existing and assimilate new functional pathways. A number of recent studies validate the concept of guilt-by-association, demonstrating that genes co-expressed across multiple conditions are more likely to represent common functions than would be expected by chance alone [<xref rid="pcbi-0020089-b005" ref-type="bibr">5</xref>,<xref rid="pcbi-0020089-b006" ref-type="bibr">6</xref>]. To date the computational methods to extract such patterns lag far behind the general agreement about their utility.</p><p>The majority of methods to extract pathways of co-regulation from microarray data begin with a measure of similarity—e.g., Euclidean distance, Pearson's correlation coefficient—that describes the degree to which expression levels between pairs of genes are correlated across multiple conditions [<xref rid="pcbi-0020089-b007" ref-type="bibr">7</xref>]. The matrix of correlations across the microarray, typically representing the pairwise similarity of the expression patterns of thousands of genes, is the starting point from which to organize genes into clusters. Clustering includes a wide variety of algorithms for organizing multivariate data into groups with approximately similar expression patterns, and a wealth of clustering approaches has been proposed [<xref rid="pcbi-0020089-b008" ref-type="bibr">8</xref>]. However, there are several important limitations to the vast majority of clustering algorithms that are in contrast to the reality of biology. The first is that they are disjoint, requiring that a gene be assigned to only one cluster. While this simplifies the amount of data to be evaluated, it places an artificial limitation on the biology under study in that many genes play important roles in multiple but distinct pathways. The other main problem is that most measures of similarity used in clustering algorithms do not permit the recognition of negative correlations, which are also common and equally meaningful.</p><p>As an alternative to assigning genes to clusters, the correlation matrix can be thresholded to create a graph comprised only of edges (gene-gene correlation values) whose weights exceed a predefined value. Allocco and colleagues originally described such graphs as relevance networks [<xref rid="pcbi-0020089-b009" ref-type="bibr">9</xref>]. In a relevance network, both positive and negative correlations exceeding a specified threshold are retained and displayed graphically, allowing visual recognition of highly connected subsets of genes. Recent studies have mined relevance networks to extract co-expressed genes in cancer cells [<xref rid="pcbi-0020089-b010" ref-type="bibr">10</xref>,<xref rid="pcbi-0020089-b011" ref-type="bibr">11</xref>] and myopathic muscle biopsies [<xref rid="pcbi-0020089-b012" ref-type="bibr">12</xref>]. While those efforts provided gene subsets of biological relevance to the respective conditions, they were limited to pairwise relationships that could be extracted manually from the graphs. Relevance networks contain many dense sub-graphs of tightly interconnected gene sets that intuitively represent the greatest potential for identifying members of common pathways. Without a systematic means to extract the aggregate relationships between multiple genes, however, many of the most interesting relationships remain embedded within the web of correlations.</p><p>We have developed a computational approach that exploits graph theoretical algorithms to identify comprehensively the tightly connected subsets of genes present in relevance networks. In the most extreme case, in which a sub-graph contains all possible edges between vertices in the sub-graph, this structure is called a clique. In terms of gene expression, clique represents the most trusted potential for identifying a set of interacting genes. Solving clique, however, is a nondeterministic polynomial-complete problem, and a classic graph-theoretic problem in its own right [<xref rid="pcbi-0020089-b013" ref-type="bibr">13</xref>]. We have previously developed novel graph algorithms that employ vertex cover and allow clique to be solved in polynomial time [<xref rid="pcbi-0020089-b014" ref-type="bibr">14</xref>–<xref rid="pcbi-0020089-b017" ref-type="bibr">17</xref>]. Recently we applied these algorithms to identify cliques of co-expressed genes as part of an effort to annotate quantitative trait loci associated with neural function [<xref rid="pcbi-0020089-b018" ref-type="bibr">18</xref>]. Here we extend these algorithms to identify differential gene relationships, i.e., gene-gene interactions that are induced or repressed by a specific treatment. We illustrate our approach using a set of microarray data that was generated from spleen of mice exposed in vivo to low-dose ionizing radiation (IR). Radiation is a well known agent of DNA damage at relatively high but sub-lethal doses[<xref rid="pcbi-0020089-b019" ref-type="bibr">19</xref>]. The response to lower doses, however, such as those received from medical imaging, radiotherapy and occupational exposures, is poorly defined and largely dependent upon genetic background [<xref rid="pcbi-0020089-b020" ref-type="bibr">20</xref>]. The data used herein were derived from a study that explored the role of genetic susceptibility in the response to IR. Six strains of inbred laboratory mice were exposed to 10 cGy X-rays in vivo, after which gene expression changes in spleen were profiled using microarrays. We describe our graph theoretical-based toolchain for identifying overlapping subsets of genes with tightly correlated expression levels and demonstrate the biological insight that this method provides.</p></sec><sec id="s2"><title>Results</title><sec id="s2a"><title>Calculation of the Correlation Matrix</title><p>Microarray data used in this study were collected as part of an effort to examine the effects of genetic background on the response to ionizing radiation in multiple mouse tissues. A panel of six common inbred strains of laboratory mice were exposed to a single acute dose of 10 cGy X-ray, and tissues were harvested 3.5 hr after irradiation for microarray gene expression profiling. Data analyzed in this effort were restricted to spleen, which is a target of the immunological effects of radiation exposure. RNA samples from sham-irradiated (controls) and irradiated mice were randomly paired to form biological replicates (a minimum of three per mouse strain). Each microarray hybridization consisted of one biological replicate, in which the control and irradiated samples were labeled with either Cy3 or Cy5 fluorescent dyes and hybridized to a single array. Each biological replicate was hybridized in duplicate, swapping the dyes to control for potential dye-specific effects. Both genetic variation between the six inbred strains and inter-individual variation within strains drive non-zero correlations in the graphs described below. An overview of the experimental design and microarray hybridizations is depicted in <xref ref-type="supplementary-material" rid="pcbi-0020089-sg001">Figure S1</xref>. Complete results of the biological response and differential expression results will be reported elsewhere (B. Voy, unpublished data), including complete access to the primary microarray data through the Gene Expression Omnibus (<ext-link ext-link-type="uri" xlink:href="http://ncbi.nlm.nih.gov/geo/">ncbi.nlm.nih.gov/geo/</ext-link>).</p><p>Our first step in using graph theory to extract gene networks was to create a correlation matrix across the entire set of microarray data. A total of 21,547 mRNAs were represented on the microarray platform used (Compugen Mouse OligoLibrary, 2.0; <ext-link ext-link-type="uri" xlink:href="http://www.labonweb.com/">http://www.labonweb.com</ext-link>), resulting in a correlation matrix of ~200,000,000 values. Although we describe our approach as applied to data from long oligonucleotide arrays, our method is equally relevant for data from Affymetrix or other array platforms, or for other types of high throughput, quantitative data. To explore the relative utility of Spearman's rank versus Pearson's correlation coefficients for our purposes, we calculated the correlation matrix using each metric and plotted the distributions. Pearson's is preferred as it utilizes information in the data more fully, but Spearman's would be less sensitive to the noise typically seen in array data [<xref rid="pcbi-0020089-b021" ref-type="bibr">21</xref>]. As shown in <xref ref-type="fig" rid="pcbi-0020089-g001">Figure 1</xref>, the relationship between the two coefficients was weak, and for a given Spearman correlation, Pearson coefficients had a wider range of values, producing vertical spikes in <xref ref-type="fig" rid="pcbi-0020089-g001">Figure 1</xref>. These reflect the greater information in Pearson's correlations, and since no strong evidence of deviations from normality were found, Pearson's correlations were used for all further analyses. To confirm the efficacy of between slide normalization procedures we plotted the distribution of Pearson's coefficients for the control data before and after normalization, as shown in <xref ref-type="fig" rid="pcbi-0020089-g002">Figure 2</xref>. It is expected that the majority of genes are not correlated at all, and that the center of the distribution should approximate zero. Normalization effectively shifted the distribution to one centered very close to zero with a slight positive shift, which we predict reflects a slight degree of positive correlation due to technical measures. We then plotted the distribution of correlations for control and IR separately, to determine if they exhibited the same characteristics. As shown in <xref ref-type="fig" rid="pcbi-0020089-g003">Figure 3</xref>, the two distributions were highly similar, as evidenced by the near overlap of the distribution plots.</p><fig id="pcbi-0020089-g001" position="float"><label>Figure 1</label><caption><title>Comparison of Pearson's and Spearman's Correlation Coefficients</title></caption><graphic xlink:href="pcbi.0020089.g001"/></fig><fig id="pcbi-0020089-g002" position="float"><label>Figure 2</label><caption><title>Impact of Normalization on the Correlation Distributions</title><p>Normalization results in a distribution approximately centered around zero.</p></caption><graphic xlink:href="pcbi.0020089.g002"/></fig><fig id="pcbi-0020089-g003" position="float"><label>Figure 3</label><caption><title>The Distributions of Correlations in Control and IR Are Highly Similar</title><p>Control (blue dashed) and IR (red solid) lines overlay each other across the entire distribution. The vertical lines in each tail of the distribution delineate the edges that were included in the graph after applying the threshold of \|0.875\| to the correlation matrix.</p></caption><graphic xlink:href="pcbi.0020089.g003"/></fig></sec><sec id="s2b"><title>Distribution of the Matrix and Threshold Selection</title><p>From the resultant matrix we wanted to create a graph representing the strongest—and most likely to reflect biological meaning—relationships between genes. Weighted graphs produced from this type of data consist of vertices representing genes and edges whose weights are indicative of the correlation between each pair of vertices (genes). Given a suitable threshold, <italic>t</italic>, edges with weights less than <italic>t</italic> are discarded; edges with weights at least <italic>t</italic> are retained. This produces an unweighted graph, <italic>G</italic>, whose structural properties are of interest. Creating the graph requires selecting a threshold correlation value, above which all edges will be included in the graph. Our main criterion was to select a value that would largely represent true expression-based correlations and would not be influenced by non-specific signal from the arrays. To assess this, we recreated the correlation matrix, including the background-subtracted signal values for a series of nonspecific buffer spots distributed across the arrays. These spots tend to give a signal value that is approximately 10% above background, and they are not included in analysis of differential expression. We determined the numbers of edges connecting these spots as vertices across a range of threshold correlation values. Correlation values with salt spots dropped markedly as the threshold value increased, and very few correlations exceeded 0.875 (0.38%, a total of 162 out of 42,194 correlations), indicating that using this as a threshold for creating a graph would largely exclude any nonspecific correlations. From a statistical significance viewpoint, if we choose <italic>p</italic> = 0.01 to indicate biologically real correlations, and correct this for multiple testing by dividing by 21,754 genes on the arrays, a standard normal critical value of 5.042 results. Applying Fisher's z-transformation in reverse, this corresponds to a correlation of 0.85 (<italic>n</italic> = 19 minimum). Therefore our threshold of 0.875 will give edges in the resultant graphs representing only statistically significant correlations (<italic>p</italic> < 0.01).</p><p> <xref ref-type="table" rid="pcbi-0020089-t001">Table 1</xref> displays the statistics of edges included in the graph using the 0.875 threshold value. For control, 0.061% of all possible correlations were included in the graph, with a slightly higher percentage (0.068%) included in the graph for IR. Degree refers to the number of other vertices with which a gene is connected. The average degree for a vertex in controls was 18 other genes, which increased to 20 in IR. The mode for degree equaled 1 in both control and IR, indicating that most genes in the graph had very low connectivity. The vast majority of edges in each graph were positive (91% in control and 91.5% in IR), indicating that expression levels for most pairs of genes changed in parallel rather than inversely. Most genes that did display inverse correlations with other genes had only one or two such negative edges, reflected by a median degree equal to 1 for negative edges in controls and 2 in IR.</p><table-wrap id="pcbi-0020089-t001" content-type="1col" position="float"><label>Table 1</label><caption><p>Summary of Edges Included in the Graphs</p></caption><graphic xlink:href="pcbi.0020089.t001"/></table-wrap></sec><sec id="s2c"><title>Computation of Clique</title><p>The output of thresholding the correlation matrix was an edge-weighted graph comprised of the tightest relationships between genes. Many types of <italic>k</italic>-dense sub-graphs existed within this graph, but we sought to extract cliques as the most intuitively interesting structure in terms of biological relationships. Clique in an undirected graph <italic>G</italic>, is a set of vertices V such that for every two vertices in V, there exists an edge connecting the two. In particular, we solved the maximum clique problem, the goal of which is to find the largest <italic>k</italic> for which <italic>G</italic> contains a clique of size <italic>k</italic>, that is, a sub-graph isomorphic to <italic>K<sub>k</sub></italic>, the complete graph on <italic>k</italic> vertices [<xref rid="pcbi-0020089-b013" ref-type="bibr">13</xref>]. The importance of <italic>K<sub>k</sub></italic> lies in the fact that each and every pair of its vertices is joined by an edge in <italic>G</italic>. <xref ref-type="fig" rid="pcbi-0020089-g004">Figure 4</xref> shows the distributions of cliques by size for both control and for IR. As is apparent from this graph, clique size (number of genes in a clique) tended to be larger in IR than in control (control = 12, IR = 16; mean size). The number of cliques was considerably greater in IR (<italic>n</italic> = 1,079,156) than in control (<italic>n</italic> = 268,611), as detailed in <xref ref-type="table" rid="pcbi-0020089-t002">Table 2</xref>. We predict that this is due to the fact that variance in expression was significantly greater in data from the IR group compared to controls (unpublished data), which served to drive more non-zero correlations in IR. Approximately half of all genes on the arrays were involved in cliques in both control and in IR (11,233, control; 10,846, IR). Like the worldwide web and many other networks, biological networks are predicted to be scale-free, with most vertices connected to few or no others, and with a smaller subset of vertices displaying high connectivity [<xref rid="pcbi-0020089-b022" ref-type="bibr">22</xref>–<xref rid="pcbi-0020089-b025" ref-type="bibr">25</xref>]. This feature is illustrated in <xref ref-type="fig" rid="pcbi-0020089-g005">Figure 5</xref>, depicting the numbers of cliques and the degree according to gene. As shown in <xref ref-type="table" rid="pcbi-0020089-t002">Table 2</xref>, only about 1.7 % of all genes on the arrays were involved in more than 0.5% of cliques for both control and for IR.</p><fig id="pcbi-0020089-g004" position="float"><label>Figure 4</label><caption><title>Distribution of Clique Sizes in Control and IR</title><p>Maximum clique and average clique sizes were larger in IR (red bars) than control (blue bars).</p></caption><graphic xlink:href="pcbi.0020089.g004"/></fig><table-wrap id="pcbi-0020089-t002" content-type="1col" position="float"><label>Table 2</label><caption><p>Clique Summary Statistics</p></caption><graphic xlink:href="pcbi.0020089.t002"/></table-wrap><fig id="pcbi-0020089-g005" position="float"><label>Figure 5</label><caption><title>Scale Free Properties of Gene Connectivity</title><p>Gene lists were sorted in order of abundance for each condition, and the 400 genes most abundant in control (blue bars) and IR (red bars) were plotted against clique membership (A) and vertex degree (B). Although average vertex degree and clique membership were not markedly different between control and IR, the genes most abundant in IR cliques were more highly connected and present in more cliques than in control.</p></caption><graphic xlink:href="pcbi.0020089.g005"/></fig><p>Our overall goal is to develop a method that uses microarray data to identify genes involved in shared cellular pathways as a means to gain additional understanding about a biology of interest. Therefore we next queried both the graphs and the cliques, asking a series of questions about the relationships between genes that were altered by radiation exposure. We illustrate below several ways in which the graph can be queried to identify relationships between genes that respond to a specific condition, in this case exposure to IR. We refer to this overall approach as differential clique analysis.</p></sec><sec id="s2d"><title>Differentially Expressed Genes</title><p>Mixed model analysis of the spleen IR dataset revealed that many differentially expressed genes were involved in the immune response and inflammation, consistent with biological effects of radiation [<xref rid="pcbi-0020089-b026" ref-type="bibr">26</xref>,<xref rid="pcbi-0020089-b027" ref-type="bibr">27</xref>], while others had little or no functional annotation. We applied guilt-by-association by selecting genes from the latter group and then filtering the clique lists to identify those with a high degree of connectivity in radiation but not controls. As an example, latent transforming growth factor beta binding protein 2 <italic>(Ltbp2)</italic> expression was significantly down-regulated by IR. Ltbp2 is a member of a family of proteins so named because of their ability to bind and regulate the availability of transforming growth factor beta (TGF-β), a hormone that orchestrates the cellular response to DNA damage after IR [<xref rid="pcbi-0020089-b028" ref-type="bibr">28</xref>]. Unlike other Ltbp family members, Ltbp2 may not bind TGF-β but rather may play a structural role by integrating with elastin containing microfibrils in the cortex of the spleen [<xref rid="pcbi-0020089-b029" ref-type="bibr">29</xref>,<xref rid="pcbi-0020089-b030" ref-type="bibr">30</xref>]. Virtually nothing is known about its role in the radiation response. <italic>Ltbp2</italic> was represented in 573 times as many cliques in IR than in control. It shared edges with 13 genes in control but with 119 in dose (seven of which were in common between the two conditions), indicating that IR activated connections between <italic>Ltbp2</italic> and many other genes. We then used Gene Ontology (GO) analysis to determine if the set of 112 genes connected to <italic>Ltbp2</italic> only in IR were enriched in any specific biological functions, which would provide additional insight into its role in the IR response. Of the 112 genes, 54 were annotated with a GO term(s) in the category of molecular function and thus amenable to GO analysis. This subset of 54 genes was significantly enriched for the GO categories GTPase activator activity (three genes; <italic>p</italic> = 0.030) and marginally for the category of structural molecule activity (six genes; <italic>p</italic> = 0.055), consistent with a predicted role for Ltbp2 in maintaining the structural integrity of elastin fibers in spleen [<xref rid="pcbi-0020089-b030" ref-type="bibr">30</xref>].</p></sec><sec id="s2e"><title>Disproportionate Abundance</title><p>Genes with disproportionate abundance in cliques from one treatment can also be used as a starting point for gene-centered approaches to extracting biological information. To assess each gene's relative representation in cliques in each condition we assigned a scaled difference score (SDS). SDS was calculated as the difference in clique membership between control and IR, expressed as a percentage to correct for different clique numbers between the two conditions, and scaled between 0 and 1 with higher scores indicating a greater difference. An SDS of 1 represents the most extreme case, in which a gene is present in cliques in one condition but not in the other. Based on this metric we selected several genes of interest. For example, cytochrome P450-family 2, subfamily s, polypeptide 1 <italic>(Cyp2s1)</italic> exhibited an SDS of 1, present in 0.4% of IR cliques formed through significant edges with 146 other genes. <italic>Cyp2s1</italic> encodes a novel member of the cytochrome P450 (Cyp) family and is abundantly expressed in spleen as well as epithelial tissues [<xref rid="pcbi-0020089-b031" ref-type="bibr">31</xref>,<xref rid="pcbi-0020089-b032" ref-type="bibr">32</xref>]. Cyp enzymes are well known for their role in oxidative metabolism of endogenous compounds and xenobiotics, and some Cyp family members may also play roles in basic developmental processes [<xref rid="pcbi-0020089-b033" ref-type="bibr">33</xref>–<xref rid="pcbi-0020089-b035" ref-type="bibr">35</xref>]. GO enrichment analysis indicated that the genes with which <italic>Cyp2s1</italic> shares edges in IR were significantly enriched (<italic>p</italic> = 0.0019) in the functional terms of primary and cellular metabolism, accounting for 33% of <italic>Cyp2s1</italic> partners. Many of the genes within this subgroup (21/43) were annotated with the GO term cellular protein metabolism, consistent with the general function of Cyp enzymes. Therefore, although the functions of <italic>Cyp2s1</italic> in this context are undefined, its tight co-expression with a set of genes only in IR identifies a putative gene network with which it interacts in the response to radiation. Another example of disproportionate abundance is phospholipase C-L2 <italic>(Plcl2),</italic> present in 17.2% of IR cliques but only 0.3% of control cliques. The largest cliques containing <italic>Plcl2</italic> are enriched for genes involved in immune response. <italic>Plcl2</italic> is expressed in hematopoietic cells and encodes a novel phospholipase C-like protein that lacks lipase activity and instead regulates B-cell receptor signaling and immune responses [<xref rid="pcbi-0020089-b036" ref-type="bibr">36</xref>], consistent with the general pathways that were altered in irradiated mice.</p><p>Both individual genes and subsets of genes may show disproportionate abundance in cliques of one treatment compared to another. In other words, a set of genes may consistently appear together in cliques of IR but not control, suggesting that these genes might represent the core of a pathway that is treatment-specific. We identified a set of seven genes that co-appeared in many cliques in IR but not control; four of the seven were differentially expressed after radiation. More specifically, 12,238 (~1.1%) cliques in IR contained at least five of these genes. In controls, no more than two of these genes appeared together, and those pairwise interactions were limited to only two combinations, representing 0.07% of cliques in the graph. The core included TGF-β, inducible form <italic>(Tgfbi),</italic> signal transducer and activator of transcription 1 <italic>(Stat1),</italic> sperm mitochondria-associated cysteine-rich protein <italic>(Smcp),</italic> the variable regions of two antibodies (Ig active kappa-chain mRNA V-region and anti-DNA antibody kappa light chain variable region), transmembrane protein 65 <italic>(Tmem65),</italic> and tubby-like protein 4 <italic>(Tulp4).</italic> Of this set, <italic>Tgfbi, Stat1,</italic> and the anti-DNA antibody have clear links to radiation exposure and its potential consequences. <italic>Tgfbi</italic> encodes a secreted adhesion molecule whose expression is sharply induced by TGF-β [<xref rid="pcbi-0020089-b037" ref-type="bibr">37</xref>], a protein that senses oxidative stress and orchestrates the response to the DNA-damaging effects of ionizing radiation [<xref rid="pcbi-0020089-b028" ref-type="bibr">28</xref>]. Stat1 integrates signal transduction and transcriptional responses to cell stressors, including ultraviolet B radiation, inflammation and infection [<xref rid="pcbi-0020089-b038" ref-type="bibr">38</xref>]. Anti-DNA antibodies are activated by reactive oxygen species, which are byproducts of the hydrolysis of intracellular water by IR. The ensuing free radicals cause oxidative damage to DNA, and DNA modified in this way becomes highly immunogenic, activating the production of antibodies directed against it [<xref rid="pcbi-0020089-b039" ref-type="bibr">39</xref>]. Among those with no direct link to radiation, Mcsp is a structural protein of mitochondria that has been characterized for its role in sperm motility [<xref rid="pcbi-0020089-b040" ref-type="bibr">40</xref>]. However it is also relatively highly expressed in spleen (UCSC Mouse Genome Browser, Aug. 2005 build), where its function is unknown to date. No functional information is available for <italic>Tmem65.</italic> However given its repeated and tight associations with the other genes discussed here, Tmem65 may also have an important role in the response to IR, a possibility that could now be pursued experimentally. Tulp4 is an uncharacterized member of the tubby superfamily of proteins, all of which share the tubby signature motif, nuclear localization signals and suppressor of cytokine signaling domains [<xref rid="pcbi-0020089-b041" ref-type="bibr">41</xref>]. Co-expression between all seven members of this gene set in IR but not control suggests that they may function in the same or intersecting pathways in the radiation response, a possibility worth further exploration.</p><p>In addition to the six genes with which it often shares cliques, <italic>Tulp4</italic> was abundant in cliques enriched for immune response genes. To determine if there was independent evidence from other data linking <italic>Tulp4</italic> with the immune system, we identified genes highly correlated with <italic>Tulp4</italic> in an independent set of gene expression data collected from hematopoietic stem cells. WebQTL (<ext-link ext-link-type="uri" xlink:href="http://www.webqtl.org">http://www.webqtl.org</ext-link>) is an internet resource that serves as a data repository and analysis engine for physiological, microarray and proteomic data collected across several recombinant inbred panels of rodents [<xref rid="pcbi-0020089-b042" ref-type="bibr">42</xref>]. Included within WebQTL is a set of microarray expression data collected from hematopoietic stem cells (HSCs), a cell type enriched in spleen [<xref rid="pcbi-0020089-b043" ref-type="bibr">43</xref>]. We used the analytical tools within WebQTL to identify genes highly correlated with <italic>Tulp4</italic> in the HSC dataset. Most genes significantly correlated (<italic>p</italic> < 0.000001) with <italic>Tulp4</italic> in HSCs were related to immune function. The 11 most highly correlated (r > 0.77) included five immunoglobulin segments, a T-cell receptor and interleukin 1 receptor-like 1 <italic>(Il1rl1);</italic> all are displayed as a network graph in <xref ref-type="fig" rid="pcbi-0020089-g006">Figure 6</xref>. Although unproven experimentally at this point, these data from an external source conceptually validate the hypothesis that <italic>Tulp4</italic> plays an as yet undefined role in immune function. Given that <italic>Tulp4</italic> is upregulated after IR and that it encodes a putative transcription factor, it is possible that is plays a role in orchestrating the immune response to radiation exposure. These data illustrate that relationships highlighted by clique membership can be independently supported using other datasets and tools and highlight how this approach can be used to filter genes worthy of further experimental study.</p><fig id="pcbi-0020089-g006" position="float"><label>Figure 6</label><caption><title>Genes Co-Expressed with Tulp4 in HSCs</title><p>Gene expression data from HSCs [<xref rid="pcbi-0020089-b043" ref-type="bibr">43</xref>] were used in WebQTL (webqtl.org) to identify genes most highly correlated with <italic>Tulp4</italic>. The majority of genes encode proteins involved in immune function (e.g., immunoglobulins).</p></caption><graphic xlink:href="pcbi.0020089.g006"/></fig></sec><sec id="s2f"><title>Edge Level Comparisons—Differential Correlations</title><p>The next level of queries focused on what we term differential correlations, i.e., significant gene-gene relationships (edges) found in one condition but not another. Differential correlation provides a means to identify edges that are dramatically altered by treatment and that would not be recognized by examining cliques alone. We defined a differential correlation as an edge that was present above the specified 0.875 threshold (\| r\| > 0.875) in one condition (control or IR) and for which the corresponding correlation value in the other condition was less than 0.25 (\| r\| < 0.25). <xref ref-type="fig" rid="pcbi-0020089-g007">Figure 7</xref> illustrates the results of differential correlation analysis. An overall representation is depicted in <xref ref-type="fig" rid="pcbi-0020089-g007">Figure 7</xref>A, containing vertices with eight or more differential edges. These graphs represent a way to visualize sets of edges activated or repressed by IR that are centered around a single gene. For example, <xref ref-type="fig" rid="pcbi-0020089-g007">Figure 7</xref>B illustrates the differential edges linked to topoisomerase III alpha <italic>(Top3a)</italic>. Unlike other members of the topoisomerase family, Top3a has poor DNA helicase activity and instead appears to interact with RecQ helicases to maintain genomic stability [<xref rid="pcbi-0020089-b044" ref-type="bibr">44</xref>]. Top3a associates with Bloom protein, a RecQ family member that participates in cell cycle checkpoint control after exposure to IR [<xref rid="pcbi-0020089-b045" ref-type="bibr">45</xref>,<xref rid="pcbi-0020089-b046" ref-type="bibr">46</xref>]. Although there are relatively few negative edges in the graph, all of the differential edges connected with <italic>Top3a</italic> are negative, reflecting a set of inverse relationships appearing only in mice from the IR group. The network of genes around <italic>Top3a</italic> is connected to two other structures in the graph centered around an uncharacterized gene (2210009P08Rik) and <italic>Notch3,</italic> which encodes a signaling protein crucial for T cell development [<xref rid="pcbi-0020089-b047" ref-type="bibr">47</xref>]. Further study will be necessary to determine if these three sets of genes interact in a functional way in the radiation response.</p><fig id="pcbi-0020089-g007" position="float"><label>Figure 7</label><caption><title>Differential Correlation Identifies Edges Impacted by IR</title><p>The graph was filtered to identify edges that exceeded r = \|0.875\| in one condition but were less than \|0.25\| in the other. Vertices with > 8 differential correlations are represented in (A). Red indicates edges that are present only in IR, while blue edges are only found in control. Dark edges for each color represent the subset of edges that are differentially correlated and of opposite direction (+ vs. −) in the two conditions, while bright edges are of the same direction. The portion of the graph containing three connected sub-graphs centered around <italic>Top3a</italic>, <italic>Notch3,</italic> and an unannotated gene is shown in (B).</p></caption><graphic xlink:href="pcbi.0020089.g007"/></fig></sec><sec id="s2g"><title>Multilevel Criteria for Identifying Genes for More Detailed Study</title><p>We used a triplet of criteria to highlight genes that should be prioritized for further study, based on their responses to radiation and on their presence in dense sub-graphs. Specifically, we identified a set of genes that were 1) differentially expressed after irradiation in at least one strain of mice, based on a mixed model analysis and <italic>p</italic> <0.05, 2) differentially correlated, participating in at least one edge that was activated/repressed by radiation (\| r\| > 0.85 / \| r\| < 0.25), and 3) differentially abundant, exhibiting a scaled difference score > 0.65. We refer to this approach as the triple screen. A total of 114 genes met all three criteria. We then used GO analysis to determine if these genes that appear to play significant roles in the radiation response were enriched in any functional categories. GO annotations within the category of biological process existed for 43 (of 114) genes. This subset of 43 was significantly enriched (<italic>p</italic> = 0.020) in genes annotated with the GO term “negative regulation of physiological processes.” Three of the four genes in this category were associated with “regulation of apoptosis,” a known response to radiation. The category “response to stress” was also significantly overrepresented (7/43) among this group of genes, reflective of the stress response induced by radiation exposure [<xref rid="pcbi-0020089-b048" ref-type="bibr">48</xref>]. The triple screen illustrates how graph structures can be combined with differential expression analysis to highlight sets of genes that respond collectively to IR. The genes identified represent an interesting set of targets to mine for further study of the effects of IR in spleen.</p></sec></sec><sec id="s3"><title>Discussion</title><p>Microarrays represent an incredibly powerful tool to identify sets of genes that respond to condition(s) of interest and underlie biological responses. Both exciting and often frustrating is the sheer volume of data that arrays produce. Even as few as 20–40 differentially expressed genes may be difficult to filter through to select a limited number of candidates for further experimental study, particularly when the goal is validation with in vivo models. Another challenge is that, despite complete sequencing of many genomes, many genes highlighted as interesting have little or no functional annotation, limiting the biological insight that results from their changes in expression. The concept of guilt-by-association, annotating functions of unknown genes based on their co-expression with better-characterized partners, forms the basis for a broad range of clustering methods designed to group genes based on similarity in expression level [<xref rid="pcbi-0020089-b002" ref-type="bibr">2</xref>]. A number of recent analyses of large scale expression datasets have validated the concept by mapping correlated gene sets onto GO annotation as a surrogate for gene function. GO represents the best available systematic method for gene functional assignments, although context-specific actions of some proteins prevent the annotations from being comprehensive. Using variations on this strategy, Zhang et al. [<xref rid="pcbi-0020089-b049" ref-type="bibr">49</xref>], Wolfe et al. [<xref rid="pcbi-0020089-b001" ref-type="bibr">1</xref>], Lee et al. [<xref rid="pcbi-0020089-b050" ref-type="bibr">50</xref>], and Stuart et al. [<xref rid="pcbi-0020089-b004" ref-type="bibr">4</xref>] all have reported that gene co-expression is a compelling indicator of gene function. These studies validate the rationale for grouping genes by co-expression as a means to expand biological insight provided by array data.</p><p>Many algorithmic approaches to guilt-by-association have been applied, and all begin with a measure of similarity (e.g., Euclidean distance, Pearson's coefficient, etc.) calculated for all possible pairwise combinations of expression values followed by an algorithm to organize genes into clusters in a supervised or unsupervised fashion [<xref rid="pcbi-0020089-b007" ref-type="bibr">7</xref>,<xref rid="pcbi-0020089-b051" ref-type="bibr">51</xref>]. Although they are used almost universally in analysis of microarray data, most clustering algorithms share several limitations. The main caveat is that they are disjoint, assigning a gene to only one cluster. In reality, many genes (proteins) participate in multiple pathways that may have little functional overlap, and examples of such functional diversity abound in the literature [<xref rid="pcbi-0020089-b052" ref-type="bibr">52</xref>–<xref rid="pcbi-0020089-b054" ref-type="bibr">54</xref>]. Hierarchical clustering methods also do not recognize inverse relationships between genes, which are of equivalent biological interest. Finally, the interpretation is largely visual, based on recognizing patterns displayed in dendograms.</p><p>Relevance networks for microarray data are similar to common clustering methods in that both are based upon some measure of similarity between gene expression [<xref rid="pcbi-0020089-b009" ref-type="bibr">9</xref>]. They differ in that genes are analyzed for co-expression only after restricting the correlation matrix to edges exceeding a threshold selected to represent biologically meaningful co-expression. To date, applications of relevance networks have been limited by the ability to extract embedded relationships from within the graphs. Previous reports of their use have relied on either identifying pairwise interactions between genes or graphically displaying the entire set of nodes and edges that remain in the graph [<xref rid="pcbi-0020089-b004" ref-type="bibr">4</xref>,<xref rid="pcbi-0020089-b010" ref-type="bibr">10</xref>–<xref rid="pcbi-0020089-b012" ref-type="bibr">12</xref>]. For cases in which the matrix was limited to only genes of interest, the resulting graphs were small and tractable using these methods [<xref rid="pcbi-0020089-b011" ref-type="bibr">11</xref>,<xref rid="pcbi-0020089-b012" ref-type="bibr">12</xref>]. When the correlation matrix is created from genome scale expression data, however, additional measures are needed to identify co-expressed genes. Our approach was developed as a means to extend beyond pairwise interactions and to identify all complete sub-graphs (cliques) from within a relevance network graph. Clique is widely known for its application in a variety of combinatorial settings, a great number of which are relevant to computational molecular biology [<xref rid="pcbi-0020089-b055" ref-type="bibr">55</xref>]. It is particularly useful in microarray analysis, because it addresses the previously-noted shortcomings of traditional clustering algorithms. A vertex can be in more than one clique, and negative correlations are included by temporarily taking the absolute value of correlation coefficients just prior to thresholding. Solving clique is a major computational bottleneck, however, and a classic graph-theoretic problem in its own right [<xref rid="pcbi-0020089-b013" ref-type="bibr">13</xref>]. We applied novel graph algorithms that allowed us to compute clique efficiently by employing fixed parameter tractability and focusing on clique's complementary dual, the vertex cover problem [<xref rid="pcbi-0020089-b014" ref-type="bibr">14</xref>–<xref rid="pcbi-0020089-b017" ref-type="bibr">17</xref>]. These algorithms were used to extract all complete sub-graphs present in the graph, which was created by thresholding the correlation matrix at r >\|0.875\| and which included only a very small percentage of all possible edges (0.061% control; 0.068% IR). Again, in contrast to clustering methods, this insured that only gene-gene interactions of a specified strength were identified as co-expressed.</p><p>Selection of an appropriate threshold value is an important issue for this approach, and there is little experimental data on which to base the selection. Allocco et al. [<xref rid="pcbi-0020089-b009" ref-type="bibr">9</xref>] previously analyzed microarray data from several hundred hybridizations across multiple conditions in yeast and related co-expression to the presence of shared transcription factor binding sites as an index of co-regulation. They reported that 50% of gene pairs with r > 0.84 were likely to be co-regulated (not just co-expressed) if a sufficient number of hybridizations were analyzed. We used a value slightly more restrictive than 0.84 to account for the reduced number of arrays analyzed in our example. Other applications of relevance networks arbitrarily have selected a similar but somewhat lower value of 0.8 [<xref rid="pcbi-0020089-b011" ref-type="bibr">11</xref>,<xref rid="pcbi-0020089-b012" ref-type="bibr">12</xref>]. Moriyama et al. [<xref rid="pcbi-0020089-b010" ref-type="bibr">10</xref>] used permutations to identify correlation threshold values with increasing confidence levels (<italic>p</italic> < 0.05, 0.01, 0.001), applying these cutoffs to relevance networks of chemosensitivity and gene expression [<xref rid="pcbi-0020089-b010" ref-type="bibr">10</xref>]. Based on Fisher's z-transformation and Bonferroni correction for multiple testing, our threshold of 0.875 produced an effective <italic>p</italic> = 0.0013, indicating a statistically significant level of confidence in relationships represented in the control and IR graphs.</p><p>Differential clique analysis can be used to ask a variety of questions from the data, several of which we have illustrated using the response to low-dose IR. Our results with <italic>Ltbp2</italic> and <italic>Plcl2</italic> demonstrate using guilt-by-association to learn more about a gene based on its treatment-specific co-expression patterns. Although both of these genes are members of well-studied gene families (TGFβ-binding proteins and phospholipase enzymes, respectively), each has been suggested to be a novel member of its respective family, with atypical functions [<xref rid="pcbi-0020089-b056" ref-type="bibr">56</xref>,<xref rid="pcbi-0020089-b057" ref-type="bibr">57</xref>]. Therefore the co-expression profiles identified in response to IR suggest biological roles for each gene that would not be revealed based on structural similarity to known proteins or on sequence conservation. Differential clique analysis can also used to identify core sets of genes that appear together in a condition-specific manner, as illustrated by the co-abundance of the set of seven genes including <italic>Stat1</italic> and <italic>Tulp4</italic> in IR but not control. This core includes two predicted transcription factors <italic>(Stat1</italic> and <italic>Tulp4),</italic> two members of a class of genes (immune responders) differentially expressed in our model, two genes <italic>(Tmem65, Mcsp)</italic> for which there is little or no information about their role in spleen biology, and <italic>Tgfbi,</italic> a gene induced by a key player in the radiation response (TGFβ). The next step will be to determine if these genes are not just co-expressed but co-regulated, potentially through common regulation by Stat1 and Tulp4. Co-expression of <italic>Tulp4</italic> with immune genes in a completely independent dataset from a comparable biological sample (HSCs) further supports this possibility.</p><p>Many issues can be further developed to improve this approach. Systematic and statistical approaches to compare clique membership between conditions will improve the iterative process we presented here. Although an advantage of clique is that it is not disjoint, this also creates a high degree of overlap between cliques that complicates the analysis. Similarity metrics that merge overlapping cliques into metacliques are under development. Solving clique also relies on an edge meeting the defined threshold. As a result, edges that fall just short of that value (e.g., r = 0.875 in our dataset) are excluded from the graph, even though they may represent correlations of biological significance. Setting thresholds based only on statistical criteria may result in excluding biologically relevant genetic relationships in small experiments, or including irrelevant relationships in large experiments due to statistical power. We are exploring alternatives, but it is likely that a combination of threshold setting methods will be needed to correctly address statistical and biological concerns. Another consideration is that low thresholds produce a highly connected graph with many vertices, and the computational burden can become unmanageable. We recently described an algorithm we refer to as paraclique that begins to address some aspects of the thresholding problem [<xref rid="pcbi-0020089-b058" ref-type="bibr">58</xref>]. Paraclique works by iteratively adding in edges that fall just shy of the original threshold value but still meet user-defined criteria for acceptance. The result are <italic>k</italic>-dense (but not complete) sub-graphs that are more inclusive than those that result from clique alone. We also want to point out that, although the low-dose study was limited to measures of gene expression, almost any type of quantitative data can be included in the correlation matrix. For example, if the current study had included systemic parameters relevant to the immune response, relationships between genes and functional measures (e.g., T-cell numbers) could have been extracted directly, rather than inferred from the data. Future efforts will be directed toward this application.</p><p>The complete schema of our method is summarized in <xref ref-type="fig" rid="pcbi-0020089-g008">Figure 8</xref>. Our approach offers the microarray user at least three potential applications: 1) annotation of poorly described genes based on guilt-by-association in a way that permits multiple functional assignments; 2) prioritization of genes for biological validation based on both differential expression and enriched connectivity; and 3) identification of relationships between genes that are differentially activated in a specific condition, rather than just differential expression of individual genes. The latter use may prove to be especially useful for conditions in which a number of genes change in parallel, but few or none of the changes are marked enough to meet statistical criteria for differential expression. For example, Mootha et al. [<xref rid="pcbi-0020089-b059" ref-type="bibr">59</xref>] associated coordinate changes in expression of a group of functionally related genes with diabetes. None of the genes were significantly altered individually, with only ~20% differences in expression in skeletal muscle of diabetics compared to healthy controls. However when changes in expression were analyzed in concert, a group of genes was identified that not only correlated with the diabetic phenotype but signaled metabolic alterations in the prediabetic state. By comparison, clique extraction represents a potential means to identify such subsets de novo, without a priori knowledge of the genes that might be involved. Ideally, clique extraction would be followed by validation using an independent set of data to determine if the same pathways could be identified in a replicate experiment. In particular, it is important to determine if some or all of the pathways described herein based on GO enrichment are robust to genetic variation or are unique to the six inbred strains used in this study. For example, if we exposed another set of six different inbred strains of mice to low-dose radiation, would cliques be enriched for the same functional pathways described herein? An ideal population in which to validate findings in this manner would be use of recombinant inbred strains of mice, such as the BXD RI strains created from C57BL6/J and DBA/2J parental strains [<xref rid="pcbi-0020089-b060" ref-type="bibr">60</xref>]. The large numbers of strains in RI panels (80 for BXD) create the opportunity to extract pathways across a significant spectrum of genetic variation, for example based on 40 BXD strains, and then validate findings in an equally large subset.</p><fig id="pcbi-0020089-g008" position="float"><label>Figure 8</label><caption><title>Overall Schema of Our Approach</title></caption><graphic xlink:href="pcbi.0020089.g008"/></fig><p>In conclusion, we have described a method to extend the utility of relevance networks by computationally extracting dense sub-graphs of tightly interconnected genes. This furthers the effort to identify gene networks from microarray data and potentially other types of data based on the concept of guilt-by-association. Once created, the graph can be probed in many ways to identify potentially meaningful relationships between genes and sets of genes. Ongoing efforts are directed toward refined methods for selecting a meaningful threshold and incorporating data of multiple types in the graphs.</p></sec><sec id="s4"><title>Materials and Methods</title><sec id="s4a"><title>Animals and tissue collection.</title><p>All mice were bred at Oak Ridge National Laboratory and experiments were conducted under approved Institutional Animal Care and Use Committee protocols. Six standard inbred strains of mice (C57Bl6/J, Balb/C, DBA/2J, A/J, C3H/HeJ and B6.C) 8–10 wk of age were exposed to an acute 10 cGy dose of a broad-spectrum X-ray flux produced by a standard bremsstrahlung source (maximum voltage = 250 kVp, maximum current = 10 mA, filter = 0.2 mm Cu). Only males were used, and each group (control and exposed) consisted of 4–8 mice. Mice were sacrificed 3.5 hr after exposure and tissues were harvested into RNALater (Ambion, The Woodlands, Texas, United States) and stored at −20 °C until RNA isolation.</p></sec><sec id="s4b"><title>RNA and microarrays.</title><p>Microarrays representing ~15,000 unique mouse genes were printed by the Center for Applied Genomics (PHRI, Newark, New Jersey, United States) using the Compugen Mouse OligoLibrary (2.0). After printing, slides were air-dried and the cDNAs irreversibly immobilized by UV-crosslinking. Spot quality was assessed by hybridization with fluorescently-labeled panomers according to manufacturer's protocols (Molecular Probes, Carlsbad, California, United States).</p><p>Total RNA was isolated from spleen using the RNeasy midi RNA isolation system (QIAGEN, Valencia, California, United States), including a DNAse I treatment step to eliminate contaminating genomic DNA. RNA quality was assessed by visualization in denaturing agarose gel electrophoresis and spectrophotometrically by the 260 nm/280 nm ratio of absorbance. Samples were quantified spectrophotometrically based on the absorbance at 260 nm. Only RNA samples of high quality were used for further analysis.</p><p>Total RNA (10 μg) from spleen was fluorescently labeled and hybridized using standard labeling and hybridization protocols [<xref rid="pcbi-0020089-b061" ref-type="bibr">61</xref>]. Dye incorporation and labeled cDNA yield were measured by scanning spectrophotometry and calculated from the absorbance values at 260 nm (cDNA) and at either 550 nm (Cy3) or 650 nm (Cy5). Each hybridization consisted of a pair of RNA samples from control and IR-exposed mice (a biological replicate); animals were paired randomly. A dye swap was performed for each biological replicate to control for dye-specific bias in labeling and to provide a replicate hybridization for each pair of samples. A total of at least three biological replicates were analyzed for each inbred strain. Data were normalized using Lowess to adjust for intensity-dependent dye bias after removing spots of poor quality or low expression and subtracting local background [<xref rid="pcbi-0020089-b062" ref-type="bibr">62</xref>]. Differentially expressed genes were identified using mixed model ANOVA performed in SAS (Cary, North Carolina, United States) as described by Wolfinger [<xref rid="pcbi-0020089-b063" ref-type="bibr">63</xref>] and using a 95% false discovery rate protected confidence interval. Because control and treatment data are treated separately in correlation analysis, data for this application were also normalized between slides by median centering to control for technical variation between hybridizations. Due to the incorporation of a dye swap in the experimental design, two replicate measures of expression existed for each animal, and the normalized values were averaged to produce one measure of expression per animal. Normalized data were used to calculate Pearson and Spearman correlations. Entrez Gene IDs are included parenthetically, as available, for each gene mentioned in text.</p></sec><sec id="s4c"><title>Graph algorithms.</title><p>The matrix of Pearson's correlation coefficients from all data meeting criteria for quality and expression level was converted into an unweighted graph. Only genes with observations in at least 19 hybridizations were retained as vertices; only gene pairs whose correlation coefficients were at least \|0.875\| were included as edges. We employed principles of fixed parameter tractability [<xref rid="pcbi-0020089-b064" ref-type="bibr">64</xref>,<xref rid="pcbi-0020089-b065" ref-type="bibr">65</xref>] to extract vertex covers, and from them cliques. Thus we reduced problem size using kernelization and searched the resultant kernel efficiently with branching. A complete description of our algorithms can be found in [<xref rid="pcbi-0020089-b014" ref-type="bibr">14</xref>,<xref rid="pcbi-0020089-b015" ref-type="bibr">15</xref>]. Source codes are freely available from M. A. Langston or any co-author. Our algorithms have also been installed in Clustal XP, a high-performance, parallel version of the Clustal W package (<ext-link ext-link-type="uri" xlink:href="http://ClustalXP.cgmlab.org">http://ClustalXP.cgmlab.org</ext-link>).</p><p>Differential abundance of genes in cliques was determined based on the SDS, which was calculated for each gene based on the percentage of clique membership for each condition (control and IR) and then scaled between 0 and 1. Per cent IR and %control represent the total percentage of cliques in which each gene has membership in each condition. First the relative difference in clique membership between IR and control was calculated as %difference = \| (%IR-%control)/((%IR+%control)/2)\| * 100. This value was scaled between 0 and 1 by normalizing to the most extreme difference across the entire gene set for which a gene was present in at least one clique from each condition (SDS = (%diff - min %diff) / (max %diff - min %diff)).</p><p>Differential correlation describes marked edge level differences in correlation between a pair of genes in control and IR. Herein, a gene is defined as differentially correlated if the following statement is true: r > \|0.875\| <sub>dose,control</sub> AND r <\|0.25\|<sub>control, dose</sub>. Differential correlation graphs were created using GraphViz (2.6).</p></sec><sec id="s4d"><title>Gene ontology enrichment.</title><p>Analyses of overrepresentation of GO categories within the ontologies of Biological Process and Molecular Function across the set of differentially expressed genes was conducted using the Database for Annotation, Visualization and Integrated Discovery 2.1 (DAVID 2.1, <ext-link ext-link-type="uri" xlink:href="http://apps1.niaid.nih.gov/David">http://apps1.niaid.nih.gov/David</ext-link>) [<xref rid="pcbi-0020089-b066" ref-type="bibr">66</xref>]. The detailed protocols and primary data from this study will be available through the Gene Expression Omnibus database (GEO; <ext-link ext-link-type="uri" xlink:href="http:/www.ncbi.nlm.nih.gov/geo">http:/www.ncbi.nlm.nih.gov/geo</ext-link>).</p></sec></sec><sec sec-type="supplementary-material" id="s5"><title>Supporting Information</title><supplementary-material content-type="local-data" id="pcbi-0020089-sg001"><label>Figure S1</label><caption><title>Diagram of the Experimental Design for the Radiation Exposure and Microarray Hybridizations</title><p>(14 KB GIF)</p></caption><media xlink:href="pcbi.0020089.sg001.gif"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020089-st001"><label>Table S1</label><caption><title>Complete List of Genes Identified using the Triple Screen</title><p>(133 KB DOC)</p></caption><media xlink:href="pcbi.0020089.st001.doc"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><sec id="s5a"><title>Accession Numbers</title><p>The Entrez Gene (<ext-link ext-link-type="uri" xlink:href="http://www.ncbi.nlm.nih.gov/entrez">http://www.ncbi.nlm.nih.gov/entrez</ext-link>) ID numbers for the genes and gene products discussed in this paper are <italic>Cyp2s1</italic> (74134), <italic>Ltbp2</italic> (16997), <italic>Notch3</italic> (18131), <italic>Plcl2</italic> (224860), <italic>Smcp</italic> (17235), <italic>Stat1</italic> (20846), <italic>Tgfbi</italic> (21810), <italic>Tmem65</italic> (74868), <italic>Top3a</italic> (21975), <italic>Tulp4</italic> (68842).</p></sec></sec>
Nonoptimal Component Placement, but Short Processing Paths, due to Long-Distance Projections in Neural Systems	<p>It has been suggested that neural systems across several scales of organization show optimal component placement, in which any spatial rearrangement of the components would lead to an increase of total wiring. Using extensive connectivity datasets for diverse neural networks combined with spatial coordinates for network nodes, we applied an optimization algorithm to the network layouts, in order to search for wire-saving component rearrangements. We found that optimized component rearrangements could substantially reduce total wiring length in all tested neural networks. Specifically, total wiring among 95 primate (Macaque) cortical areas could be decreased by 32%, and wiring of neuronal networks in the nematode <named-content content-type="genus-species">Caenorhabditis elegans</named-content> could be reduced by 48% on the global level, and by 49% for neurons within frontal ganglia. Wiring length reductions were possible due to the existence of long-distance projections in neural networks. We explored the role of these projections by comparing the original networks with minimally rewired networks of the same size, which possessed only the shortest possible connections. In the minimally rewired networks, the number of processing steps along the shortest paths between components was significantly increased compared to the original networks. Additional benchmark comparisons also indicated that neural networks are more similar to network layouts that minimize the length of processing paths, rather than wiring length. These findings suggest that neural systems are not exclusively optimized for minimal global wiring, but for a variety of factors including the minimization of processing steps.</p>	<contrib contrib-type="author"><name><surname>Kaiser</surname><given-names>Marcus</given-names></name><xref ref-type="aff" rid="aff1">1</xref><xref ref-type="aff" rid="aff2">2</xref><xref ref-type="aff" rid="aff3">3</xref><xref ref-type="corresp" rid="cor1">*</xref></contrib><contrib contrib-type="author"><name><surname>Hilgetag</surname><given-names>Claus C</given-names></name><xref ref-type="aff" rid="aff3">3</xref><xref ref-type="aff" rid="aff4">4</xref></contrib>	PLoS Computational Biology	<sec id="s1"><title>Introduction</title><p>The organization of neural systems is shaped by multiple constraints, ranging from limits placed by physical and chemical laws to diverse functional requirements. In particular, it is of interest to identify factors influencing the layout of neural connectivity networks. One prominent idea is that the establishment and maintenance of neural connections carry a significant metabolic cost that should be reduced wherever possible [<xref rid="pcbi-0020095-b001" ref-type="bibr">1</xref>,<xref rid="pcbi-0020095-b002" ref-type="bibr">2</xref>]. As a consequence, wiring length should be globally minimized in neural systems. This requirement places strong constraints on the design of neural networks across different levels of organization [<xref rid="pcbi-0020095-b003" ref-type="bibr">3</xref>–<xref rid="pcbi-0020095-b008" ref-type="bibr">8</xref>].</p><p>A trend toward wiring minimization is apparent in the distributions of projection lengths for various neural systems, which show that most neuronal projections are short [<xref rid="pcbi-0020095-b009" ref-type="bibr">9</xref>–<xref rid="pcbi-0020095-b012" ref-type="bibr">12</xref>]. However, wiring length distributions also indicate a significant number of longer-distance projections, which are not formed between immediate neighbours in the network. Such projections may be required for functional reasons, so that a strictly minimal wiring of networks, using only local projections, is not attainable.</p><p>Alternatively, it has been suggested that wiring length reductions in neural systems are achieved not by minimal rewiring of projections within the networks, but by suitable spatial arrangement of the components. Under these circumstances, the connectivity patterns of neurons or regions remain unchanged, maintaining their structural and functional connectivity, but the layout of components is perfected such that it leads to the most economical wiring. In the sense of this “component placement optimization” (CPO [<xref rid="pcbi-0020095-b003" ref-type="bibr">3</xref>]), any rearrangement of the position of neural components, while keeping their connections unchanged, would lead to an increase of total wiring length in the network. CPO has been reported for neural networks at different levels of organization, such as interconnected ganglia in the nematode <named-content content-type="genus-species">Caenorhabditis elegans</named-content> [<xref rid="pcbi-0020095-b003" ref-type="bibr">3</xref>] and cerebral cortical areas in rat, cat, and monkey brains [<xref rid="pcbi-0020095-b003" ref-type="bibr">3</xref>,<xref rid="pcbi-0020095-b005" ref-type="bibr">5</xref>,<xref rid="pcbi-0020095-b006" ref-type="bibr">6</xref>,<xref rid="pcbi-0020095-b013" ref-type="bibr">13</xref>]. Moreover, optimal component placement was suggested to apply to the arrangement of human cranial nerves [<xref rid="pcbi-0020095-b003" ref-type="bibr">3</xref>] as well as the layout of cortical maps [<xref rid="pcbi-0020095-b014" ref-type="bibr">14</xref>], and mechanisms yielding optimal placement were proposed [<xref rid="pcbi-0020095-b015" ref-type="bibr">15</xref>,<xref rid="pcbi-0020095-b016" ref-type="bibr">16</xref>].</p><p>We revisited the concept of CPO and investigated the layout of two representative neural networks in metric space, analyzing three-dimensional spatial positions of connected cortical areas in the primate (Macaque) brain and two-dimensional positions of individual neurons in the neuronal network of <named-content content-type="genus-species">C. elegans</named-content>. Due to the more limited availability of data in the past, previous analyses of these networks [<xref rid="pcbi-0020095-b003" ref-type="bibr">3</xref>,<xref rid="pcbi-0020095-b006" ref-type="bibr">6</xref>] included less detail and fewer components and were based on neighbourhood (adjacency) relationships, rather than on metric spatial coordinates of the components. We investigated component rearrangements in the networks with the help of a simulated annealing search approach and found that, remarkably, the extended and refined datasets possessed no optimal component placement. Instead, the spatial rearrangement of network components could lead to substantial wire saving, due to a considerable number of long-distance projections in the networks.</p><p>While the role of long-range connections, or network shortcuts, has been previously explored in topological analyses of neural connectivity [<xref rid="pcbi-0020095-b007" ref-type="bibr">7</xref>,<xref rid="pcbi-0020095-b011" ref-type="bibr">11</xref>], the specific role of long-distance projections in the spatial layout of neural systems is still uncertain. In order to address this question and to explore alternative constraints on the organization of neural connectivity, we also compared the biological neural networks with different benchmark networks of the same size in which connections were rewired minimally or randomly, or same-size networks optimized for minimum and maximum wiring length and average path lengths, respectively. The comparisons demonstrated that biological neural networks feature shorter average path lengths than networks lacking long-distance connections. Moreover, the average path lengths of neural networks, corresponding to the average number of processing steps, were close to path lengths in networks optimized for minimal paths.</p></sec><sec id="s2"><title>Results</title><sec id="s2a"><title>Overview</title><p>First, we derived the wiring length distribution of the primate and <named-content content-type="genus-species">C. elegans</named-content> network (<xref ref-type="fig" rid="pcbi-0020095-g001">Figure 1</xref>A–<xref ref-type="fig" rid="pcbi-0020095-g001">1</xref>C). Second, we used an optimization approach based on simulated annealing to search for wire-saving component rearrangements in the two networks (see <xref ref-type="supplementary-material" rid="pcbi-0020095-sg001">Figure S1</xref>). For both networks, alternative component rearrangements existed that resulted in substantially reduced total wiring (<xref ref-type="fig" rid="pcbi-0020095-g001">Figure 1</xref>D–<xref ref-type="fig" rid="pcbi-0020095-g001">1</xref>F). The rearranged networks showed a decrease in the number of long-distance connections (<xref ref-type="fig" rid="pcbi-0020095-g001">Figures 1</xref>G–<xref ref-type="fig" rid="pcbi-0020095-g001">1</xref>I, <xref ref-type="fig" rid="pcbi-0020095-g002">2</xref>, and <xref ref-type="fig" rid="pcbi-0020095-g003">3</xref>). Third, we explored the role of long-distance projections, by comparing the original neural networks with minimally rewired networks not possessing long-distance connections. The relative comparisons, carried out against the background of a variety of benchmark networks (<xref ref-type="fig" rid="pcbi-0020095-g004">Figures 4</xref> and <xref ref-type="fig" rid="pcbi-0020095-g005">5</xref>), demonstrated that long-distance connections may confer adaptive benefits, particularly by reducing the number of intermediate processing steps in neural networks.</p><fig id="pcbi-0020095-g001" position="float"><label>Figure 1</label><caption><title>Projection Length Distribution and Total Wiring Length for Original and Rearranged Neural Networks</title><p>(A–C) Approximated projection length distribution in neural networks. Macaque monkey cortical connectivity network with 95 areas and 2,402 projections (A). Local distribution of connections within rostral ganglia of <named-content content-type="genus-species">C. elegans</named-content> with 131 neurons and 764 projections (B). Global <named-content content-type="genus-species">C. elegans</named-content> neural network with 277 neurons and 2,105 connections (C).</p><p>(D–F) Reduction in total wiring length by rearranged layouts yielded by simulated annealing for Macaque cortical network (D), <named-content content-type="genus-species">C. elegans</named-content> local network (neurons within rostral ganglia) (E), and global <named-content content-type="genus-species">C. elegans</named-content> network (F).</p><p>(G–I) Approximated projection length distribution in neural networks with optimized component placement. Macaque monkey cortical connectivity network (G). Local distribution of connections within rostral ganglia of <named-content content-type="genus-species">C. elegans</named-content> (H). Global <named-content content-type="genus-species">C. elegans</named-content> neural network (I). For all optimized networks, the number of long distance connections is reduced compared to the original length distribution in (A)–(C).</p></caption><graphic xlink:href="pcbi.0020095.g001"/></fig></sec><sec id="s2b"><title>Wiring Length Distribution</title><p>In the primate brain and in the neuronal network of <italic>C. elegans,</italic> the reach of connections among components quickly decays with distance (<xref ref-type="fig" rid="pcbi-0020095-g001">Figure 1</xref>A–<xref ref-type="fig" rid="pcbi-0020095-g001">1</xref>C). Nonetheless, some connections are not formed between immediate neighbours and extend over a considerable distance. For example, more than 10% of the primate cortical projections connect components that are separated by more than 40 mm (more than half of the maximally possible spatial distance between components of 69 mm; <xref ref-type="fig" rid="pcbi-0020095-g001">Figure 1</xref>A). A similar distribution emerged for the local connectivity in <named-content content-type="genus-species">C. elegans</named-content> (<xref ref-type="fig" rid="pcbi-0020095-g001">Figure 1</xref>B). For the global <named-content content-type="genus-species">C. elegans</named-content> network, some connections were almost as long as the entire organism (<xref ref-type="fig" rid="pcbi-0020095-g001">Figure 1</xref>C). Thus, biological neural networks are not strictly minimally wired in the sense that only the shortest possible connections are established ([<xref rid="pcbi-0020095-b011" ref-type="bibr">11</xref>]; also see section “Minimally rewired networks”, below). However, their components may be spatially arranged in such a way that the overall wiring is minimal, given the specific connectivity of the network. This hypothesis is tested in the next section.</p></sec><sec id="s2c"><title>Reduction of Total Wiring Length by Component Rearrangement</title><p>We tested the concept of CPO, which states that any spatial rearrangement of neural network components leads only to an increase, not a decrease, of total network wiring. As an exhaustive search of all possible alternative node arrangements was not feasible for the given large-scale networks, we used a simulated annealing algorithm to specifically search for overall wiring reductions in spatially permutated component layouts (see <xref ref-type="sec" rid="s4">Materials and Methods</xref> for details). Briefly, at each step three nodes were cyclically rearranged, and testing explored whether the new solution led to a shorter total wiring length. The algorithm quickly converged and led to an approximate solution for the optimal wiring layout of the components (<xref ref-type="supplementary-material" rid="pcbi-0020095-sg001">Figure S1</xref>). Note that this approach modified only spatial node positions, whereas the network topology (the specific afferent and efferent connections of each node) remained unchanged.</p><sec id="s2c1"><title>Macaque: Cortical area rearrangements.</title><p>Area rearrangement by simulated annealing reduced the total wiring length of the primate cortical network by up to 32% (<xref ref-type="fig" rid="pcbi-0020095-g001">Figure 1</xref>D). In the wire-saving new arrangement, the overall number of long-distance connections was reduced (<xref ref-type="fig" rid="pcbi-0020095-g001">Figure 1</xref>G). This reduction resulted from placing areas with many projections (e.g., area V1) closer to the areas to which they are mainly connected. Moreover, areas possessing fewer connections were moved to the spatial periphery of the rearranged network (<xref ref-type="fig" rid="pcbi-0020095-g002">Figure 2</xref>).</p><fig id="pcbi-0020095-g002" position="float"><label>Figure 2</label><caption><title>Original and Optimally Rearranged Macaque Cortical Networks</title><p>(A) Original placement of 95 cortical areas.</p><p>(B) Network layout after evolutionary rearrangement of areas to minimize total wiring.</p><p>A larger version of this figure is available at <ext-link ext-link-type="uri" xlink:href="http://www.biological-networks.org">http://www.biological-networks.org</ext-link>.</p></caption><graphic xlink:href="pcbi.0020095.g002"/></fig></sec><sec id="s2c2"><title>Macaque monkey: Influence of cortical area sizes.</title><p>We considered two potential confounds of the spatial rearrangement analysis. First, swapping areas with different area sizes might also affect area positions. For example, exchanging a small cortical area such as the lateral intraparietal area (about 50 mm<sup>2</sup> [<xref rid="pcbi-0020095-b017" ref-type="bibr">17</xref>]) with area V1 or V2 (size about 1,200 mm<sup>2</sup> [<xref rid="pcbi-0020095-b017" ref-type="bibr">17</xref>]) would also result in shifting the positions of other cortical areas, and ultimately in changing total wiring length. Therefore, in a modified approach, we limited spatial permutations to areas whose surface sizes did not differ by more than 5%. This constraint substantially restricted the number of permissible rearrangements, as only 4% of all possible swaps met the same-size criterion. Despite this restriction, total wiring length in the cortical network could still be reduced by 12.5%.</p></sec><sec id="s2c3"><title>Macaque monkey: Role of white matter volume.</title><p>As a second potential confound, the rearrangement analysis investigated metric projection distances between cortical areas, but not white matter volume. However, two fibre tracts of the same metric length might differ in the actual number of axons that form the projections. Therefore, fibre tract diameter or volume would constitute a better measure of total axonal wiring. Unfortunately, no systematically collated metric information is available for the diameters of corticocortical fibres. Therefore, white matter volume of specific projections cannot be calculated in a straightforward way. An approximation used in earlier studies of CPO [<xref rid="pcbi-0020095-b006" ref-type="bibr">6</xref>] was to employ connection strength or density as an estimate of white matter volume. One of the sources for the present cortical network data [<xref rid="pcbi-0020095-b018" ref-type="bibr">18</xref>] reported connection strengths for projections among 18 primate prefrontal areas as ordinal values: 0 (absent or unknown), 1 (light), 2 (moderate), and 3 (heavy). We used this cortical subnetwork of 18 areas to explore the role of connection strength in wire-saving component arrangements. Total wiring volume was calculated by multiplying the distance between connected nodes by the square of the connection strength of the respective projection, since the cross section of a fibre is proportional to the square of its diameter (therefore, cross-section areas were 1, 4, and 9, respectively). Thus, for two connections with the same length, but respective connection strengths 1 and 3, the connection with strength 3 was assigned a white matter volume nine times as large as the connection with strength 1. With these measures in the simulated annealing search algorithm, the total wiring volume of projections among prefrontal areas could still be reduced by 16%. By comparison, the approach of optimizing total wiring length without accounting for connection strength led to a reduction of 10% in the wiring among these 18 prefrontal areas. Therefore, considering total wiring volume instead of total wiring length did not appear to alter the principal conclusion of suboptimal component placement in cortical networks. However, further investigations, using improved information about fibre bundle diameter as well as larger datasets, need to be conducted.</p></sec><sec id="s2c4"><title> <italic>C. elegans:</italic> Neuronal network rearrangements.</title><p>For the global neural network of <italic>C. elegans,</italic> which includes many long-distance connections (<xref ref-type="fig" rid="pcbi-0020095-g001">Figure 1</xref>C), the component rearrangement algorithm produced a maximum wiring reduction of 48%. Meanwhile, rearrangement of neurons in the rostral ganglia alone reduced total wiring length by 49% (<xref ref-type="fig" rid="pcbi-0020095-g001">Figure 1</xref>E and <xref ref-type="fig" rid="pcbi-0020095-g001">1</xref>F). In line with the lowered total wiring length, the number of long-distance connections also decreased (<xref ref-type="fig" rid="pcbi-0020095-g001">Figure 1</xref>H and <xref ref-type="fig" rid="pcbi-0020095-g001">1</xref>I), as described for the primate cortical network above. During optimization by simulated annealing, connections aligned with the longitudinal (horizontal) axis were rearranged along the shorter vertical axis for the local <named-content content-type="genus-species">C. elegans</named-content> network within rostral ganglia (<xref ref-type="fig" rid="pcbi-0020095-g003">Figure 3</xref>A and <xref ref-type="fig" rid="pcbi-0020095-g003">3</xref>B). Similarly, rearranging neuron positions in the global <named-content content-type="genus-species">C. elegans</named-content> network (<xref ref-type="fig" rid="pcbi-0020095-g003">Figure 3</xref>C and <xref ref-type="fig" rid="pcbi-0020095-g003">3</xref>D) reduced the number of long-distance connections running along the longitudinal axis. These findings also remained valid when variations in connection site, of the third spatial coordinate, or in synaptic type were taken into account (see <xref ref-type="sec" rid="s4">Materials and Methods</xref>, “Control calculations for <named-content content-type="genus-species">C. elegans</named-content> network”).</p><fig id="pcbi-0020095-g003" position="float"><label>Figure 3</label><caption><title>Original and Optimally Rearranged Layouts of Local and Global Neural Networks of <named-content content-type="genus-species">C. elegans</named-content> </title><p>(A) Original placement of neurons within rostral ganglia.</p><p>(B) Optimized wire-saving component placement of rostral ganglia neurons.</p><p>(C) Original layout of global <named-content content-type="genus-species">C. elegans</named-content> network (lateral view).</p><p>(D) Global <named-content content-type="genus-species">C. elegans</named-content> neuronal network, rearranged to minimize total network wiring.</p><p>A larger version of this figure is available at <ext-link ext-link-type="uri" xlink:href="http://www.biological-networks.org">http://www.biological-networks.org</ext-link>.</p></caption><graphic xlink:href="pcbi.0020095.g003"/></fig></sec></sec><sec id="s2d"><title>Alternative Constraints on Network Organization</title><sec id="s2d1"><title>Minimally rewired networks.</title><p>For both the primate cortical and the <named-content content-type="genus-species">C. elegans</named-content> network, rearrangement of components yielded shorter total wiring lengths by reducing the number of long-distance connections (<xref ref-type="fig" rid="pcbi-0020095-g001">Figures 1</xref>A–<xref ref-type="fig" rid="pcbi-0020095-g001">1</xref>C versus 1G–1I, 2, and 3). Thus, biological neural networks possess more long-distance connections than do networks with strictly optimized component placement. To explore the potential benefits of long-distance connections, we compared the original neural networks with minimally rewired networks in which component positions and numbers of connections remained unchanged, but in which neighbouring nodes were preferentially connected (see <xref ref-type="sec" rid="s4">Materials and Methods</xref>, “Minimal rewiring of networks”). Thus, very few long-distance connections existed in these networks, which represented the extreme state of adaptation to global minimal wiring.</p><p>We characterized the original and rewired networks with several network indices (<xref ref-type="table" rid="pcbi-0020095-t001">Table 1</xref>). For the minimally wired networks, the clustering coefficient [<xref rid="pcbi-0020095-b019" ref-type="bibr">19</xref>]—which describes the average connectivity between neighbours of a node—was higher than in the original networks (77% versus 64% in the primate cortical network; 43% versus 17% in the global <italic>C. elegans</italic> neuronal network; and 51% versus 14% in the local <named-content content-type="genus-species">C. elegans</named-content> neuronal network). Moreover, minimally rewired networks showed a lower total wiring length, leading to a global wire reduction of up to 90% (<xref ref-type="fig" rid="pcbi-0020095-g004">Figure 4</xref>A). Also, the average metric path length of the shortest path between two components was shorter in minimally rewired networks (<xref ref-type="fig" rid="pcbi-0020095-g004">Figure 4</xref>B). However, minimally rewired networks had one notable structural disadvantage: They possessed significantly longer average path lengths, as measured by the average number of connection segments between any two components, than did the biological networks (<xref ref-type="fig" rid="pcbi-0020095-g004">Figure 4</xref>C). This result is intuitively plausible, since the rewired networks lacked the network shortcuts provided by long-distance projections, so that all paths had to be routed via local neighbourhoods. For example, while direct connections between the occipital and frontal lobe exist in the original primate cortical network, such connections were absent in the minimally rewired network. In this case, the path between both regions involved several short-distance connections.</p><table-wrap id="pcbi-0020095-t001" content-type="2col" position="float"><label>Table 1</label><caption><p>Network Measures for the Original Neural Systems as well as for Minimally Rewired Networks of the Same Size</p></caption><graphic xlink:href="pcbi.0020095.t001"/></table-wrap></sec><sec id="s2d2"><title>Relative comparison of constraints.</title><p>Although the total wiring length in all analyzed networks could be reduced by node rearrangement, it might be the case that the original node configuration is already very close to the optimum compared to alternative arrangements. Therefore, we compared the actual wiring length of the different neural networks on a relative scale, on which the optimized wire-saving spatial arrangement yielded by simulated annealing represented the lower boundary (relative value “0”). The upper boundary (“1”) was provided by an arrangement, also obtained by simulated annealing, in which the components were rearranged for maximum total wiring length. This relative comparison demonstrated that the actual wiring of neural networks was far from the optimal configuration (<xref ref-type="fig" rid="pcbi-0020095-g005">Figure 5</xref>A).</p><p>Moreover, an exhaustive search of all two-node and all three-node permutations in these networks also revealed a substantial proportion of alternative configurations with shorter total wiring: 29% of the two-node (and 17% of the three-node) permutations in the primate network, 32% (26%) of permutated alternative configurations in the local <named-content content-type="genus-species">C. elegans</named-content> network, and 19% (13%) of the permutations in the global <named-content content-type="genus-species">C. elegans</named-content> network displayed shorter wiring than did the original configurations.</p><p>A different pattern emerged when the number of processing steps in the neural networks, as indicated by the average path length, was also compared on a relative scale. Here, the boundaries (relative values “0” or “1”) were set by the path length of networks rewired for minimum or maximum average path lengths, respectively. Such networks were obtained by a simulated annealing search similar to that for optimal spatial rearrangements (see <xref ref-type="sec" rid="s4">Materials and Methods</xref> for details). As shown by <xref ref-type="fig" rid="pcbi-0020095-g005">Figure 5</xref>B, the average path lengths of the cortical and <named-content content-type="genus-species">C. elegans</named-content> networks were close to those of same-size networks optimized for minimum path length.</p><p>As an additional benchmark, we also investigated randomly organized networks. The total wiring length for a random arrangement of nodes had relative values of 0.518 ± 0.030 (Macaque), 0.687 ± 0.026 (<named-content content-type="genus-species">C. elegans</named-content> global), and 0.544 ± 0.028 (<named-content content-type="genus-species">C. elegans</named-content> local). On the other hand, the random arrangement of edges resulted in relative values for average path length of 0.011 ± 9.6 × 10<sup>−17</sup> (Macaque), 0.021 ± 3.5 × 10<sup>−17</sup> (<named-content content-type="genus-species">C. elegans</named-content> global), and 0.018 ± 3.0 × 10<sup>−17</sup> (<named-content content-type="genus-species">C. elegans</named-content> local).</p></sec></sec></sec><sec id="s3"><title>Discussion</title><sec id="s3a"><title>Summary</title><p>The distribution of projection lengths (<xref ref-type="fig" rid="pcbi-0020095-g001">Figure 1</xref>A–<xref ref-type="fig" rid="pcbi-0020095-g001">1</xref>C) and the absence of optimal component placement in diverse neural networks (<xref ref-type="fig" rid="pcbi-0020095-g001">Figure 1</xref>D–<xref ref-type="fig" rid="pcbi-0020095-g001">1</xref>F) suggest that wiring minimization may not be a predominant constraint on the design of neural networks at all levels. Instead, adding long-distance projections, and thereby reducing the number of processing steps across the system, might for some neural networks outweigh the costs of establishing and maintaining additional fibre tracts. In the following sections we discuss how our analysis differed from previous studies that found evidence for CPO in neural networks, whether CPO might arise at some levels of neural organization but not others, and how CPO may be balanced with alternative demands on the organization of neural systems.</p></sec><sec id="s3b"><title>Differences between the Present and Previous Studies</title><sec id="s3b1"><title>Datasets.</title><p>Our findings, which are based on the analysis of extended and refined spatial representations of neural networks, are in apparent contrast to the widely reported optimal component placement in neural systems ([<xref rid="pcbi-0020095-b003" ref-type="bibr">3</xref>,<xref rid="pcbi-0020095-b005" ref-type="bibr">5</xref>,<xref rid="pcbi-0020095-b006" ref-type="bibr">6</xref>,<xref rid="pcbi-0020095-b013" ref-type="bibr">13</xref>], but see [<xref rid="pcbi-0020095-b020" ref-type="bibr">20</xref>,<xref rid="pcbi-0020095-b021" ref-type="bibr">21</xref>]). Previous analyses, however, were more limited in their approach. For example, earlier studies considered only a subset of 11 areas in the Macaque prefrontal cortex employing two-dimensional coordinates and adjacency relations as distance measure [<xref rid="pcbi-0020095-b006" ref-type="bibr">6</xref>], or 17 Macaque and 15 cat visual areas, for which distance was also measured by adjacency relations [<xref rid="pcbi-0020095-b013" ref-type="bibr">13</xref>]. Other investigations examined the adjacency arrangement of 11 entire <named-content content-type="genus-species">C. elegans</named-content> ganglia [<xref rid="pcbi-0020095-b003" ref-type="bibr">3</xref>], but not the actual spatial positions of neurons within the ganglia nor the positions of all individual <named-content content-type="genus-species">C. elegans</named-content> neurons as analyzed here. Because of the smaller number of nodes in the previously analyzed datasets, fewer opportunities existed for wire-saving rearrangements than for the larger sets of 95 (primate) or 277 <italic>(C. elegans)</italic> components considered here. Moreover, adjacency relations provide more coarsely defined distance measures than did the continuous Euclidean distances used in the present study, also resulting in more limited degrees of freedom for wire reductions. Finally, previous datasets were analyzed as one-dimensional <italic>(C. elegans)</italic> or two-dimensional (primate) layouts, and the present consideration of additional dimensions expanded the search space for wire-saving component rearrangements.</p></sec><sec id="s3b2"><title>Analysis approaches.</title><p>The choice of the employed analytical techniques is related to the size of analyzed datasets. While it was possible to exhaustively test all possible component arrangements for the smaller datasets in previous studies, a stochastic optimization approach (simulated annealing) had to be applied for the large networks used in the current study. However, this approach also brought methodological advantages, as only small modifications in component placement were made at each step of the annealing algorithm. Moreover, the algorithm searched specifically for improvements in total wiring length. We note that an alternative approach, of randomly sampling different component rearrangements in the hope of identifying layouts with lower total wiring length, would have been unlikely to succeed. Scanning such random samples of primate area rearrangements, we found that all permutations had longer total wiring length compared with the original network, increasing wiring by at least 14%. This was due to the fact that more than 85 areas were placed at new positions in each of these permutations, and wire-saving effects of more limited exchanges involving only two to five nodes were hidden.</p><p>However, when we exhaustively searched all configurations that could be obtained from the original network arrangements by permutation of just two or three nodes (a large search space with up to 21,024,300 configurations that represents the practical limit of exhaustive analysis), we also found a high proportion (up to 32%) of permutated configurations with shorter wiring length. Therefore, the analysis revealed abundant alternative network arrangements with shorter wiring than in the original neural networks.</p><p>The current study used three-dimensional Euclidean distances between cortical areas as a metric measure of inter-area wiring length. This is an improvement over previous approaches in which adjacency relations on a two-dimensional cortical map (e.g., “neighbour,” “next-neighbour-but-one”) were used as ordinal estimates of wiring length. While the actual length of fibres in the cerebral network is still inaccessible due to practical limitations of current experimental techniques, recent work suggests that only a minority of corticocortical projections are strongly curved, whereas the majority of projections are straight or just mildly curved. For instance, only about 15% of the connections among prefrontal cortices in the Macaque monkey possess strongly curved trajectories, and dense fibres, in particular, tend to be completely straight [<xref rid="pcbi-0020095-b022" ref-type="bibr">22</xref>]. Therefore, Euclidean distances may represent a reasonable approximation for the actual length of most projections.</p></sec><sec id="s3b3"><title>Component placement optimization may differ for different neural networks.</title><p>Given the more restricted focus of earlier studies finding CPO, it also appears possible that optimal placement exists at some levels of neural organization but not others. For instance, connectivity within the prefrontal subnetwork of the primate cortex may be dominated by connections to neighbouring areas; and significant numbers of long-range projections might only arise from the interconnections of prefrontal cortices with structures in other lobes of the brain. Indeed, an area rearrangement analysis of 11 prefrontal areas studied previously [<xref rid="pcbi-0020095-b006" ref-type="bibr">6</xref>] confirmed optimal component placement within this restricted network (unpublished data). In addition, our rearrangement analysis of 18 prefrontal areas defined by three-dimensional positions identified a possible reduction of 10%, much lower than the 32% reduction obtained for the global network. Therefore, optimal component placement of the 11 prefrontal areas could be due to the smaller subset of areas involved as well as the greater number of their local interconnections. For the prefrontal network, for example, edge density was much higher than for the global network (67% compared to 27%). Similarly, optimal placement, based on adjacency of connected components, was found for comparably small datasets of closely linked areas in the cat and Macaque visual cortex [<xref rid="pcbi-0020095-b013" ref-type="bibr">13</xref>].</p><p>Optimal component placement is also more likely to occur when competing design constraints are absent in specific neural populations, for example, when a reduction of the number of processing steps is impossible. One example is the parallel wiring between two directly connected regions where intermediate processing steps do not exist. Indeed, optimal component placement was found for the mapping of fibres between ocular dominance columns [<xref rid="pcbi-0020095-b014" ref-type="bibr">14</xref>], or the vertical integration across cortical layers [<xref rid="pcbi-0020095-b008" ref-type="bibr">8</xref>].</p></sec></sec><sec id="s3c"><title>CPO and Alternative Constraints on Neural Network Organization</title><p>Neural networks contain a substantial proportion of long-distance projections, many more than minimally rewired networks of the same size. Due to these far-reaching spatial shortcuts, minimal rewiring of the biological networks led to a global reduction in the amount of wiring and to a reduced metric length of the average shortest path between components. Moreover, minimally rewired networks showed increased clustering of connections within local neighbourhoods (<xref ref-type="table" rid="pcbi-0020095-t001">Table 1</xref>). Such features are potentially beneficial for the organization of neural networks, resulting in economical wiring as well as greater local integration of network nodes.</p><p>However, minimal network rewiring also resulted in significantly increased average path length (corresponding to the number of connection steps in the shortest pathways) between the components (<xref ref-type="fig" rid="pcbi-0020095-g004">Figure 4</xref>C). Thus, it appears that in biological neural networks economical wiring and tight integration of local components are counterbalanced by the global minimization of processing steps across the network (but compare [<xref rid="pcbi-0020095-b023" ref-type="bibr">23</xref>]). Indeed, when evaluated on a comparative scale for minimal wiring and minimal path lengths, the neural networks considered here were placed farther away from the optimal configuration for minimal wiring than from the best network arrangement for minimizing processing steps (<xref ref-type="fig" rid="pcbi-0020095-g005">Figure 5</xref>).</p><p>The importance of network shortcuts for reducing the number of processing steps has been pointed out before (e.g., [<xref rid="pcbi-0020095-b008" ref-type="bibr">8</xref>]), particularly in the context of small-world network architectures [<xref rid="pcbi-0020095-b019" ref-type="bibr">19</xref>]. The present study adds a spatial perspective to the previous topological investigations, by demonstrating that network shortcuts are formed mainly by long-distance connections. This conclusion, while intuitive, is not trivial, as one could also imagine alternative scenarios in which network shortcuts arise from short-distance connections (see <xref ref-type="supplementary-material" rid="pcbi-0020095-sg003">Figure S3</xref>). The coincidence of long-distance connections with network shortcuts hints at a close match between the spatial layout and topology of neural networks [<xref rid="pcbi-0020095-b020" ref-type="bibr">20</xref>]. It will be an interesting task for future studies to explore more fully the developmental and evolutionary reasons for this coincidence.</p><p>Minimizing average path length—that is, reducing the number of intermediate transmission steps in neural integration pathways—has several functional advantages. First, the number of intermediate nodes that may introduce interfering signals and noise is limited. Second, by reducing transmission delays from intermediate connections, the speed of signal processing and, ultimately, behavioural decisions is increased. Third, long-distance connections enable neighbouring as well as distant regions to receive activation nearly simultaneously [<xref rid="pcbi-0020095-b011" ref-type="bibr">11</xref>,<xref rid="pcbi-0020095-b024" ref-type="bibr">24</xref>] and thus facilitate synchronous information processing in the system (compare [<xref rid="pcbi-0020095-b025" ref-type="bibr">25</xref>,<xref rid="pcbi-0020095-b026" ref-type="bibr">26</xref>]). Fourth, the structural and functional robustness of neural systems increases when processing pathways (chains of nodes) are shorter. Each further node introduces an additional probability that the signal is not transmitted, which may be substantial (e.g., failure rates for transmitter release in individual synapses are between 50% and 90% [<xref rid="pcbi-0020095-b002" ref-type="bibr">2</xref>]). Even when the signal survives, longer chains of transmission may lead to an increased loss of information. A similar conclusion, on computational grounds, was first drawn by John von Neumann [<xref rid="pcbi-0020095-b027" ref-type="bibr">27</xref>] when he compared the organization of computers and brains. He argued that, due to the low precision of individual processing steps in the brain, the number of steps leading to the result of a calculation (“logical depth”) should be reduced and highly parallel computing would be necessary. A loss of long-distance connections might also underlie pathological changes in neural network function. For instance, functional path length is increased in patients with Alzheimer's disease due to the loss of long-distance neural projections [<xref rid="pcbi-0020095-b028" ref-type="bibr">28</xref>].</p><fig id="pcbi-0020095-g004" position="float"><label>Figure 4</label><caption><title>Network Properties of Original Cortical Networks and Minimally Rewired Networks of the Same Size Lacking Long-Distance Connections</title><p>Total wiring length (A) is substantially reduced in minimally rewired networks. Average metric length of the shortest path between any two nodes (B) is also lowered in the rewired networks. However, average path length (C), corresponding to the number of processing steps in the shortest path between components, is considerably smaller in the original than in the minimally rewired networks.</p></caption><graphic xlink:href="pcbi.0020095.g004"/></fig><fig id="pcbi-0020095-g005" position="float"><label>Figure 5</label><caption><title>Wiring Arrangement of Neural Networks Compared to Minimum and Maximum Case Benchmark Networks</title><p>(A) Actual total wiring length relative to the minimum wiring length solution (value “0,” yielded by simulated annealing of component positions) and to networks optimized for maximum total wiring length (value “1,” also yielded by simulated annealing). The wiring of the different neural networks lies close to the middle between minimum and maximum case component arrangements.</p><p>(B) Average shortest path length (characteristic path length) in neural networks relative to networks optimized for minimum path length (value “0,” yielded by simulated annealing of wiring organization) and maximum path length (value “1”). Actual path lengths in the neural networks are close to the lower bound of networks optimized for minimum paths.</p></caption><graphic xlink:href="pcbi.0020095.g005"/></fig><p>The advantages resulting from a short average number of processing steps appear as least as beneficial for the organization and function of neural networks as those commonly cited for wiring minimization. Previous analyses of cortical organization have demonstrated that the convoluted, laminar architecture of the mammalian cerebral cortex and the segregation into gray and white matter reduce total white matter volume and shorten projection lengths [<xref rid="pcbi-0020095-b029" ref-type="bibr">29</xref>,<xref rid="pcbi-0020095-b030" ref-type="bibr">30</xref>], also reducing conduction delays [<xref rid="pcbi-0020095-b031" ref-type="bibr">31</xref>]. Ultimately, these adaptations point in the same direction as reductions in the number of processing steps, toward maximizing information-processing speed. Therefore, it is plausible that neural systems are adapted to more than just one design constraint, and that their observed organization is the outcome of an optimization of multiple parameters, which may be partly opposed to each other. For cortical networks, for example, additional constraints may arise from spatial factors that limit growth [<xref rid="pcbi-0020095-b011" ref-type="bibr">11</xref>,<xref rid="pcbi-0020095-b032" ref-type="bibr">32</xref>] or from critical periods for the establishment of cortical areas and their interconnections [<xref rid="pcbi-0020095-b033" ref-type="bibr">33</xref>]. In addition to current ontogenetic constraints, the evolutionary history of a neural network might also conserve features of its predecessors, some of which may not be optimal for the present system [<xref rid="pcbi-0020095-b008" ref-type="bibr">8</xref>].</p><p>In conclusion, we demonstrated that the organization of neural systems at different levels is not primarily shaped by a drive for minimal wiring. Rather, wiring minimization appears to be just one constraint among a variety of desirable factors [<xref rid="pcbi-0020095-b011" ref-type="bibr">11</xref>,<xref rid="pcbi-0020095-b034" ref-type="bibr">34</xref>] that also include the minimization of processing steps. It remains to be seen how, exactly, different structural, functional, and evolutionary constraints interact and compete to shape the organization of neural systems, and under which circumstances one singular constraint may dominate.</p></sec></sec><sec id="s4"><title>Materials and Methods</title><sec id="s4a"><title>Primate corticocortical network.</title><p>We analyzed the spatial arrangement of 2,402 projections among 95 cortical areas and sub-areas of the primate (Macaque) brain. The connectivity data were retrieved from CoCoMac (<ext-link ext-link-type="uri" xlink:href="http://www.cocomac.org">http://www.cocomac.org</ext-link> [<xref rid="pcbi-0020095-b035" ref-type="bibr">35</xref>]) and are based on three extensive neuroanatomical compilations [<xref rid="pcbi-0020095-b018" ref-type="bibr">18</xref>,<xref rid="pcbi-0020095-b017" ref-type="bibr">17</xref>,<xref rid="pcbi-0020095-b036" ref-type="bibr">36</xref>] that collectively cover large parts of the cerebral cortex. Spatial positions of cortical areas were estimated from surface parcelling using the CARET software (<ext-link ext-link-type="uri" xlink:href="http://brainmap.wustl.edu/caret">http://brainmap.wustl.edu/caret</ext-link>). The spatial positions of areas were calculated as the average surface coordinate (or centre of mass) of the three-dimensional extension of an area (compare [<xref rid="pcbi-0020095-b011" ref-type="bibr">11</xref>]). While the current cortical dataset is more extensive than those used in previous studies, it may still be partially incomplete, particularly for connections of motor, auditory, and somatosensory areas. The restriction arose from the fact that only studies could be used for which a parcellation scheme with spatial coordinates existed in CARET.</p></sec><sec id="s4b"><title> <named-content content-type="genus-species">C. elegans</named-content> neuronal networks.</title><p>We further analyzed two-dimensional spatial representations of the global neuronal network (277 neurons and 2,105 connections) of the nematode <italic>C. elegans,</italic> as well as a local subnetwork of neurons within <named-content content-type="genus-species">C. elegans</named-content> rostral ganglia (anterior, dorsal, lateral, and ring, 131 neurons with 764 unidirectional connections). Spatial two-dimensional positions (in the lateral plane), representing the position of the soma of individual neurons in <italic>C. elegans,</italic> were provided by Y. Choe [<xref rid="pcbi-0020095-b037" ref-type="bibr">37</xref>]. Neuronal connectivity was obtained from [<xref rid="pcbi-0020095-b038" ref-type="bibr">38</xref>]. This compilation is largely based on the dataset of White et al. [<xref rid="pcbi-0020095-b039" ref-type="bibr">39</xref>] in which connections were identified by electron microscope reconstructions. The previously presented connectivity data [<xref rid="pcbi-0020095-b038" ref-type="bibr">38</xref>] were modified in the following way. Neurons of the pharyngeal ring, for which there was no internal connection information [<xref rid="pcbi-0020095-b038" ref-type="bibr">38</xref>], were removed from the network, leaving 280 neurons. In addition, three neurons (AIBL, AIYL, and SMDVL) had to be removed, because their positions were not provided in the set of spatial coordinates. Eventually 277 neurons were included in the analyses. The size of the global and local <named-content content-type="genus-species">C. elegans</named-content> datasets analyzed here was comparable to that used in previous studies. For example, studies of the small-world properties [<xref rid="pcbi-0020095-b019" ref-type="bibr">19</xref>] or characteristic motifs [<xref rid="pcbi-0020095-b040" ref-type="bibr">40</xref>] of <named-content content-type="genus-species">C. elegans</named-content> considered 282 and 187 neurons, respectively. Both chemical and electric synapses (gap junctions) were included as connections in our analysis.</p><p>Wiring length was calculated as direct Euclidean distance between connected components in three dimensions (Macaque, compare [<xref rid="pcbi-0020095-b011" ref-type="bibr">11</xref>]) or two dimensions <italic>(C. elegans).</italic> Both datasets are available at <ext-link ext-link-type="uri" xlink:href="http://www.biological-networks.org">http://www.biological-networks.org</ext-link>.</p></sec><sec id="s4c"><title>Component placement optimization analysis.</title><p>In line with the definition of CPO [<xref rid="pcbi-0020095-b003" ref-type="bibr">3</xref>], we investigated the possibility that spatial permutations of network components would lead to reductions in total wiring length of the neural networks. Because an exhaustive search of all possible 95! = 10<sup>148</sup> (primate) or 139! and 256! <italic>(C. elegans)</italic> component rearrangements was computationally unattainable, we employed a stochastic optimization algorithm to specifically search for wiring length reductions.</p><p>The network layouts were explored by simulated annealing [<xref rid="pcbi-0020095-b041" ref-type="bibr">41</xref>], which is a popular algorithm for combinatorial optimization. We implemented a Matlab version (MathWorks, Natick, Massachusetts, United States) of a standard algorithm [<xref rid="pcbi-0020095-b042" ref-type="bibr">42</xref>]. Briefly, at each step, a cyclic permutation of three randomly chosen node positions was performed, after which the procedure recalculated the total wiring length of the networks. During the initial steps of the algorithm, decreases as well as increases of wiring length were carried through to the next stage of the optimization, while during later steps the selection of solutions was more strongly biased toward decreases in total wiring length. By this mechanism, the procedure could escape local minima and approximated solutions close to or at the global minimum. The wiring length for primate as well as <named-content content-type="genus-species">C. elegans</named-content> networks converged to a minimum after 40–60 and 8–12 steps, respectively (<xref ref-type="supplementary-material" rid="pcbi-0020095-sg001">Figure S1</xref>). The simulated annealing process was performed independently 50 times on the original placement configuration. The minimum wiring length out of all trials was then used as an indicator of the possible reduction in total fibre length.</p><p>In addition, we exhaustively tested all possible two- and three-node permutations of the original network component arrangements, to establish the proportion of alternative component arrangements possessing a lower total wiring length. The total numbers of arrangements tested in the primate, local <italic>C. elegans,</italic> and global <named-content content-type="genus-species">C. elegans</named-content> were 8,930, 17,556, and 76,452 configurations for two-node permutations and 830,490, 2,299,836, and 21,024,300 configurations for three-node permutations, respectively.</p><p>Similar simulating annealing approaches were employed to create benchmark networks in which components were rearranged for maximum total wiring length, or in which connections were rewired to minimize or maximize average path lengths across the network. In each case, 20 runs of simulated annealing were performed and the minimal or maximal values were chosen as lower or upper limit for the benchmark, respectively.</p></sec><sec id="s4d"><title>Minimal rewiring of networks.</title><p>To investigate the role of long-distance connections in neural systems, we also compared the original networks with networks of the same size (that is, with an identical number of nodes and connections) that had been rewired with the shortest possible connections. In order to generate such minimally rewired networks, possessing a greatly reduced proportion of long-distance connections, we employed the following procedure.</p><p>Starting with the spatial configuration of nodes in the original networks, but without edges, a minimum spanning tree was generated, to ensure that the resulting network would be connected. The minimum spanning tree for <italic>N</italic> nodes consists of <italic>N</italic> − 1 edges such that all nodes are part of the network and the total wiring length of all edges is minimal (compare [<xref rid="pcbi-0020095-b043" ref-type="bibr">43</xref>]). The application of this initial step was required, since wiring of only the shortest available distances in <named-content content-type="genus-species">C. elegans</named-content> produced a fragmented network with multiple compartments. For the more densely connected cortical network of the primate, in contrast, the wiring of only the shortest distances already resulted in a connected network. However, for consistency, we started by creating the minimum spanning tree for both <named-content content-type="genus-species">C. elegans</named-content> and primate cortical networks. This constituted a conservative approach with regard to the subsequent computation of average path lengths.</p><p>In the next step, all pairwise distances of nodes were calculated and sorted by length. Starting with the shortest distance between any two nodes, edges between these nodes were generated until the total number of edges matched those of the original <named-content content-type="genus-species">C. elegans</named-content> or cortical networks. Thus, the resulting minimally rewired networks represented a lower bound for the wiring length of a connected network with the same total number of edges and identical node positions as in the original neural networks.</p></sec><sec id="s4e"><title>Network characterization measures.</title><p>The following measures were used to characterize original and rearranged networks. Total wiring length was the sum of the metric length of all individual connections. In addition, total wiring volume included information about the anatomical strength of connections (as a first approximation of fibre diameter), as derived from tract-tracing experiments. The volume of an individual fibre was calculated as its metric length multiplied by its squared anatomical strength or density (as given by ordinal values 1, 2, or 3; 1 being the sparsest), with the total wiring volume as the sum of the volume of all individual fibres.</p><p>The average path length was the average number of connections that had to be passed on the shortest paths between all pairs of network nodes. It was calculated using Floyd's algorithm (compare [<xref rid="pcbi-0020095-b043" ref-type="bibr">43</xref>]). The average metric path length described the metric distance that had to be travelled on average along the edges of the shortest path between any two network nodes. Therefore, it represented the average total wiring length of the shortest paths of the network. The clustering coefficient [<xref rid="pcbi-0020095-b019" ref-type="bibr">19</xref>] was the proportion of actually present connections, out of all possible connections, among network nodes directly connected to a target node (i.e., the index measured the local connectivity among the target node's neighbours). The coefficient was calculated as the average over all individual nodes of the network.</p></sec><sec id="s4f"><title>Control calculations for <named-content content-type="genus-species">C. elegans</named-content> network: Variations in connection site.</title><p>One possible confound of the <named-content content-type="genus-species">C. elegans</named-content> analyses was that exact positions of synapses were not included in the data, so morphological differences between neurons may have influenced the actual length of projections between neurons. Therefore, we tested the effect of variations in the position of synapses, by varying the connection distances between individual neurons through shifts in soma positions. As the animal extends mainly in the horizontal direction, variations in the spatial positions of synapses and cell bodies matter most along the horizontal axis. The position of each neuron along the longitudinal axis was shifted randomly rostrally or caudally, following a normal distribution with the mean around the actual position and a standard deviation of 10% of the total length of the worm (0.12 mm). Twenty different networks with varied longitudinal position were tested for potential reductions of total wiring length. A reduction of total wiring length was found that was almost as large as for the original positions (<xref ref-type="supplementary-material" rid="pcbi-0020095-sg002">Figure S2</xref>). Therefore, random variations in neuronal positions did not alter the main finding.</p><p>However, could systematic errors, rather than random variations, substantially bias the results? The main sources of unknown synaptic positions are synapses along muscles and sensory endings. We noted that only 118 connections onto muscles are present, forming less than 6% of the total connections in <named-content content-type="genus-species">C. elegans</named-content> [<xref rid="pcbi-0020095-b038" ref-type="bibr">38</xref>]. In the worst-case scenario, systematically overestimating the length of such connections might lead to an overestimation of the potential reduction in the real network. However, additional information about the detailed connection patterns of sensory and motor projections is unlikely to reduce the possible wire saving from 49% to 0%.</p></sec><sec id="s4g"><title>Control calculations for <named-content content-type="genus-species">C. elegans</named-content> network: Variations of the third spatial coordinate.</title><p>Due to limitations of the currently available data, the <named-content content-type="genus-species">C. elegans</named-content> analyses involved only two dimensions. However, the third dimension, which forms the transversal axis, may not contribute much variation to the spatial positions of neurons. Because the animal extends mainly along one axis, it has been noted that the “layout problem is roughly one-dimensional” as “the length:diameter ratio of the worm body is about 20:1” [<xref rid="pcbi-0020095-b003" ref-type="bibr">3</xref>]. In any case, we explored the effect of coordinate variations in the third dimension. Based on the used spatial database [<xref rid="pcbi-0020095-b037" ref-type="bibr">37</xref>], neurons were classified as (a) placed on the left or right side of the animal, or (b) unspecified in their lateral position. In the simulation of a worst-case scenario, neurons of type (a) were positioned in the transversal axis as far away from the midline as possible; thus, neurons were placed at a distance of 57.3 μm from the midline, which represents the radius of the roundworm. Neurons of type (b) were placed on the origin of the <italic>z</italic>-coordinate (i.e., at the midline). The simulation of these conservative three-dimensional coordinates created a total wiring length of 573.6 mm (only 7.8% more than for the original data in two dimensions). The application of the simulated annealing approach for reducing total wiring length then led to a total wiring length of 321.6 mm—that is, a reduction of 44%. Therefore, using three instead of two coordinates for <named-content content-type="genus-species">C. elegans</named-content> neuronal positions only slightly increased total wiring length, and did not change the potential for wiring length reduction.</p></sec><sec id="s4h"><title>Control calculations for <named-content content-type="genus-species">C. elegans</named-content> network: Variations of synapse type.</title><p>The analyzed dataset of 277 nodes contained chemical as well as electrical (gap junction) synapses. Potentially, neurons connected by gap junctions may be more closely associated than neurons linked by chemical synapses, and the network layout may be affected by the type of interconnection. Therefore, we tested the effect of rearranging networks where electrical synapses were either present or absent. We used a smaller dataset of 256 neurons that included information on the type of synapse for each connection between neurons. We tested component rearrangement by simulated annealing with either chemical and electrical synapses or only chemical synapses present. In both cases a substantial reduction in total wiring length was found after rearrangement (for chemical and electrical synapses, 58% reduction on the global and 47% reduction on the local level; for chemical synapses only, 64% reduction on the global and 46% on the local level).</p></sec></sec><sec sec-type="supplementary-material" id="s5"><title>Supporting Information</title><supplementary-material content-type="local-data" id="pcbi-0020095-sg001"><label>Figure S1</label><caption><title>Successive Reduction of Total Wiring Length during Simulated Annealing Optimization</title><p>The curves summarize 50 individual trials for the Macaque cortical network (A) and the local <named-content content-type="genus-species">C. elegans</named-content> network (within rostral ganglia) (B). For each network, 50 simulated annealing runs were performed. All runs converged to a solution with shorter total wiring length than the original solution. For subsequent analysis, the single best configuration out of the 50 trials was used as the closest approximate of an optimal component placement solution.</p><p>(258 KB TIF)</p></caption><media xlink:href="pcbi.0020095.sg001.tif"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020095-sg002"><label>Figure S2</label><caption><title>Random Variations in the Longitudinal Connection Site of Neurons in the <named-content content-type="genus-species">C. elegans</named-content> Neuronal Network</title><p>Dots represent the outcome of wiring optimization for 20 networks with randomly varied neuronal connection sites. The upper boundary of the diagram represents the total wiring length of the original network; the lower boundary represents the wiring optimization outcome for the original network.</p><p>(100 KB TIF)</p></caption><media xlink:href="pcbi.0020095.sg002.tif"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020095-sg003"><label>Figure S3</label><caption><title>Linking Clusters by Short- or Long-Distance Connections</title><p>The two clusters are indicated by node colour. Note that both networks contain long-distance connections as well as the same number of connections, yet the link between the clusters (grey line) is alternatively provided by a short-distance connection (A) or a long-distance connection (B). The existence of long-distance connections, together with their effect on reducing the path length, suggests that the latter scenario does occur in neural systems.</p><p>(165 KB TIF)</p></caption><media xlink:href="pcbi.0020095.sg003.tif"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material></sec>
Protein–Protein Interactions More Conserved within Species than across Species	<p>Experimental high-throughput studies of protein–protein interactions are beginning to provide enough data for comprehensive computational studies. Today, about ten large data sets, each with thousands of interacting pairs, coarsely sample the interactions in fly, human, worm, and yeast. Another about 55,000 pairs of interacting proteins have been identified by more careful, detailed biochemical experiments. Most interactions are experimentally observed in prokaryotes and simple eukaryotes; very few interactions are observed in higher eukaryotes such as mammals. It is commonly assumed that pathways in mammals can be inferred through homology to model organisms, e.g. the experimental observation that two yeast proteins interact is transferred to infer that the two corresponding proteins in human also interact. Two pairs for which the interaction is conserved are often described as interologs. The goal of this investigation was a large-scale comprehensive analysis of such inferences, i.e. of the evolutionary conservation of interologs. Here, we introduced a novel score for measuring the overlap between protein–protein interaction data sets. This measure appeared to reflect the overall quality of the data and was the basis for our two surprising results from our large-scale analysis. Firstly, homology-based inferences of physical protein–protein interactions appeared far less successful than expected. In fact, such inferences were accurate only for extremely high levels of sequence similarity. Secondly, and most surprisingly, the identification of interacting partners through sequence similarity was significantly more reliable for protein pairs within the same organism than for pairs between species. Our analysis underlined that the discrepancies between different datasets are large, even when using the same type of experiment on the same organism. This reality considerably constrains the power of homology-based transfer of interactions. In particular, the experimental probing of interactions in distant model organisms has to be undertaken with some caution. More comprehensive images of protein–protein networks will require the combination of many high-throughput methods, including <italic>in silico</italic> inferences and predictions. <ext-link ext-link-type="uri" xlink:href="http://www.rostlab.org/results/2006/ppi_homology/">http://www.rostlab.org/results/2006/ppi_homology/</ext-link> </p>	<contrib contrib-type="author"><name><surname>Mika</surname><given-names>Sven</given-names></name><xref ref-type="aff" rid="aff1">1</xref><xref ref-type="aff" rid="aff2">2</xref><xref ref-type="corresp" rid="cor1">*</xref></contrib><contrib contrib-type="author"><name><surname>Rost</surname><given-names>Burkhard</given-names></name><xref ref-type="aff" rid="aff2">2</xref><xref ref-type="aff" rid="aff3">3</xref><xref ref-type="aff" rid="aff4">4</xref></contrib>	PLoS Computational Biology	<sec id="s1"><title>Introduction</title><sec id="s1a"><title>Experiments Peek at Complete Protein–Protein Networks</title><p>The faster large-scale sequencing projects determine the alphabet of life, the higher the pressure to determine some of the actual processes that make life what it is. The understanding of functional relations among all proteins is essential to understanding how cells work. Recent breakthroughs in experimental high-throughput techniques have begun to peek at complete protein–protein interaction networks of entire organisms (<xref ref-type="supplementary-material" rid="pcbi-0020079-st001">Table S1</xref>). One central method is to use yeast two-hybrid (Y2H) assays [<xref rid="pcbi-0020079-b001" ref-type="bibr">1</xref>] that are based on a genially simple idea: first, separate two domains (activation and DNA-binding) of a transcription factor that activates a reporter gene, then merge each of the two domains to a different protein (A and B) [<xref rid="pcbi-0020079-b002" ref-type="bibr">2</xref>,<xref rid="pcbi-0020079-b003" ref-type="bibr">3</xref>]. If A and B interact, the two transcription domains will merge, and thereby activate the reporter gene that will be detected. The difficulty of using Y2H is in mastering the details of the experimental setup. Other high-throughput methods to detect protein–protein interactions, such as phage-display assays [<xref rid="pcbi-0020079-b004" ref-type="bibr">4</xref>], tandem affinity purifications (TAP) [<xref rid="pcbi-0020079-b005" ref-type="bibr">5</xref>,<xref rid="pcbi-0020079-b006" ref-type="bibr">6</xref>], co-immunoprecipitation, and affinity chromatography [<xref rid="pcbi-0020079-b002" ref-type="bibr">2</xref>,<xref rid="pcbi-0020079-b007" ref-type="bibr">7</xref>–<xref rid="pcbi-0020079-b009" ref-type="bibr">9</xref>], are also commonly used. An important advantage of using Y2H over these other high-throughput techniques is the ability to measure physical interactions between proteins as opposed to pure functional associations. Also, Y2H experiments work with physiological conditions, i.e., conditions that resemble those in eukaryotic cells [<xref rid="pcbi-0020079-b002" ref-type="bibr">2</xref>,<xref rid="pcbi-0020079-b003" ref-type="bibr">3</xref>,<xref rid="pcbi-0020079-b010" ref-type="bibr">10</xref>,<xref rid="pcbi-0020079-b011" ref-type="bibr">11</xref>]. Ito et al. [<xref rid="pcbi-0020079-b012" ref-type="bibr">12</xref>] and Uetz et al. [<xref rid="pcbi-0020079-b013" ref-type="bibr">13</xref>] first scanned large fractions of the yeast proteome for protein–protein interactions. Others added further interactions: Ho et al. [<xref rid="pcbi-0020079-b014" ref-type="bibr">14</xref>] used mass spectrometry and Gavin et al. [<xref rid="pcbi-0020079-b015" ref-type="bibr">15</xref>] used TAP. Protein networks in the fly (<named-content content-type="genus-species">Drosophelia melanogaster</named-content>) have been targeted through three different Y2H studies [<xref rid="pcbi-0020079-b011" ref-type="bibr">11</xref>,<xref rid="pcbi-0020079-b016" ref-type="bibr">16</xref>,<xref rid="pcbi-0020079-b017" ref-type="bibr">17</xref>], in the worm (<named-content content-type="genus-species">Caenorhabditis elegans</named-content>) through one [<xref rid="pcbi-0020079-b018" ref-type="bibr">18</xref>], and a large subset of about 1,500 human protein network relations were detected through TAP [<xref rid="pcbi-0020079-b019" ref-type="bibr">19</xref>]. These data bear deeper insights into cellular processes.</p></sec><sec id="s1b"><title>Today's Data Are Incomplete and Not Fully Reliable</title><p>Y2H systems are not 100% accurate; they, for instance, identify many putative interactions that cannot be confirmed by other studies. One reason for false positives (interactions incorrectly postulated) is that the two proteins A and B may activate the reporter gene directly without having to interact [<xref rid="pcbi-0020079-b003" ref-type="bibr">3</xref>]. The Margalit group has estimated the false positive rate in high-throughput Y2H assays to be about 50% [<xref rid="pcbi-0020079-b020" ref-type="bibr">20</xref>]; the Eisenberg group has arrived at the same estimate through measuring the reliability of interactions in the Database of Interacting Proteins [<xref rid="pcbi-0020079-b021" ref-type="bibr">21</xref>]. Y2H experiments also do not achieve complete coverage, i.e., they miss many interactions. Conversely, false negatives (missed interactions) might result from the particular experimental setup (which may prevent the interaction between A and B) or from problems in the assembly of the two transcriptional domains (activation and DNA-binding) needed for Y2H. These problems do not prevent Y2H from evolving as one of the major experimental probes for interactions; they do, however, imply that today's data sets are neither complete nor fully accurate [<xref rid="pcbi-0020079-b020" ref-type="bibr">20</xref>,<xref rid="pcbi-0020079-b022" ref-type="bibr">22</xref>]. One of the strong arguments in favor of large-scale Y2H experiments is that they are more systematic and much less driven by happenstance than hypothesis-driven, detailed experiments.</p></sec><sec id="s1c"><title>Known Interactions Are Expanded through Homology-Based Inference</title><p>Evolutionary connections help explain the rapid success of molecular biology: we can study a particular protein in a simple bacterium and learn about the function of the same protein in multicellular eukaryotes. This idea enables us to use model organisms to predict protein structure [<xref rid="pcbi-0020079-b023" ref-type="bibr">23</xref>–<xref rid="pcbi-0020079-b025" ref-type="bibr">25</xref>], subcellular localization [<xref rid="pcbi-0020079-b026" ref-type="bibr">26</xref>], enzymatic activity [<xref rid="pcbi-0020079-b027" ref-type="bibr">27</xref>–<xref rid="pcbi-0020079-b029" ref-type="bibr">29</xref>], and other aspects of protein function [<xref rid="pcbi-0020079-b030" ref-type="bibr">30</xref>–<xref rid="pcbi-0020079-b034" ref-type="bibr">34</xref>]. The same principle is frequently applied to the extension of interactions (<xref ref-type="fig" rid="pcbi-0020079-g001">Figure 1</xref>): Assume that two proteins A and B are experimentally observed to bind in organism o, and that alignment methods identify related protein pairs in organism o (A′-B′) and in organism p (A″-B″). Can we infer that the pairs A′-B′ and A″-B″ also interact with each other? The Vidal group [<xref rid="pcbi-0020079-b010" ref-type="bibr">10</xref>] has investigated how yeast interactions detected by Ito [<xref rid="pcbi-0020079-b035" ref-type="bibr">35</xref>] and Uetz [<xref rid="pcbi-0020079-b013" ref-type="bibr">13</xref>] map to interactions in worm. They concluded that at BLAST E-values <10<sup>−10</sup>, only 16%–30% of the yeast interactions are transferable [<xref rid="pcbi-0020079-b036" ref-type="bibr">36</xref>]; similar results were reported by the Gerstein group [<xref rid="pcbi-0020079-b037" ref-type="bibr">37</xref>]. Although homology inference is common practice, no large-scale study has ever estimated levels of accuracy and coverage for physical interactions. A particular aspect of this question relates to paralogs and orthologs. Two proteins are often considered as paralogs when they originate from the same organism and differ in function. Paralogs are assumed to have arisen from gene duplication followed by the specialization and drifting away of one of the copies, while the other copy has maintained its original function. Orthologs, on the other hand, are described as two proteins with largely identical function and a common ancestor that reside in different organisms [<xref rid="pcbi-0020079-b037" ref-type="bibr">37</xref>–<xref rid="pcbi-0020079-b039" ref-type="bibr">39</xref>]. Applied to homology-based inference of interactions, a common assumption is that interactions are more conserved between orthologs than between paralogs [<xref rid="pcbi-0020079-b040" ref-type="bibr">40</xref>–<xref rid="pcbi-0020079-b042" ref-type="bibr">42</xref>], i.e., interactions are more conserved between than within organisms. If true, model organisms would be ideal for the study of interactions.</p><fig id="pcbi-0020079-g001" position="float"><label>Figure 1</label><caption><title>Concept of Homology Inference and Interologs</title><p>Interologs are two pairs of protein interactions that fulfill the following conditions: (A interacts with B) + (A is similar to A′) + (B is similar to B′) → (A′ interacts with B′). All quadruples (A, B, A′, B′) for which this relation is true are referred to as interologs [<xref rid="pcbi-0020079-b037" ref-type="bibr">37</xref>,<xref rid="pcbi-0020079-b079" ref-type="bibr">79</xref>]. To illustrate our analysis, we have to extend this simple relation. Assume that a physical protein–protein interaction (PPI) between proteins A and B is observed in organism o. If A and B are both sequence similar (above a certain threshold) to two other proteins A′ and B′ in the same organism o, we should be able to infer the physical interaction between A′ and B′. Note that both pairs, A/A′ as well as B/B′, have to be above the particular similarity threshold for us to be able to make this inference. Thus, we neither use an average similarity of both pairs (A/A′ and B/B′) nor a minimum similarity for just one pair (A/A′ or B/B′). Now let us assume that we have another pair of proteins A″ and B″ in another organism p, and that both are as similar to A and B as are A′ and B′, respectively. One of our findings was that homology transfers A-B → A′-B′ were more reliable than those from A-B → A″-B″.</p></caption><graphic xlink:href="pcbi.0020079.g001"/></fig></sec><sec id="s1d"><title>Focus on Transient Physical Interactions (PPIs)</title><p>One important difference between Y2H and TAP is that while Y2H aims at the detection of physically interacting proteins, TAP identifies large groups of proteins that are associated, for instance, through a common pathway [<xref rid="pcbi-0020079-b043" ref-type="bibr">43</xref>]. Most high-throughput techniques resemble TAP in the sense that they reveal association rather than physical interaction. To illustrate this difference, assume we hypothesized that co-expressed proteins interact physically, and we wanted to use this hypothesis to predict physical interactions directly from co-expression data. Assume further that six proteins are strung together in a linear pathway (1 binds 2, 2 binds 3, etc.), and that all six are co-expressed. Of the 15 [N*(N − 1)/2] possible interactions, only 5 (N − 1) are physical, i.e., only 33% of the co-expressed proteins interact. Since most pathways involve many more than six interactions this example is likely to significantly underestimate the actual problem. In other words, even if all physically interacting proteins were co-expressed, predictions of interactions based on such association alone would still be more often wrong than right. This significantly constrains the way in which we can use association-type data to analyze physical interactions. In order to emphasize our focus on physical interactions, we used the abbreviation PPI for transient physical protein–protein interactions (as opposed to functional associations as measured by TAP-like data, and as opposed to permanent physical interactions between, e.g., two different domains or two different chains of the same protein [<xref rid="pcbi-0020079-b044" ref-type="bibr">44</xref>]).</p></sec><sec id="s1e"><title>Coping with the Dilemma of Incomplete Data Sets</title><p>How can we evaluate accuracy and coverage of homology transfer (<xref ref-type="fig" rid="pcbi-0020079-g001">Figure 1</xref>) of interactions if the data are incomplete? An extreme stance is to simply not assess the performance at all. The rationale is simple: assume a method inferred that A″ and B″ in <xref ref-type="fig" rid="pcbi-0020079-g001">Figure 1</xref> interacted without any experimental evidence for this interaction. May be the inference was wrong; it also may just have been a new <italic>in silico</italic> discovery not yet identified by experiments. If the set of all interactions were complete, the absence of an observation would imply noninteraction. Although there is currently no such complete set, we challenge that the performance of homology transfer has to be estimated somehow to render a tool that is controllable in the context of genome annotation pipelines. Here, we took the opposite radical stance by treating all interactions that have not been observed as nonexisting. While this is obviously wrong, we assume that today's incompleteness is not systematic. If true, our results will simply underestimate the quantities that we measured, but will correctly capture relative values (such as that homology transfer is half as accurate at ~40% sequence identity as at ~60%, <xref ref-type="fig" rid="pcbi-0020079-g002">Figure 2</xref>). We also did not merge data sets that measure functional association (e.g., TAP) with those that measure physical interaction (e.g., Y2H). Instead, we regarded only physical interactions as positives.</p><fig id="pcbi-0020079-g002" position="float"><label>Figure 2</label><caption><title>Sequence Conservation of PPIs</title><p>The performance of homology transfer was evaluated with the data sets in Experiment 1 (<xref ref-type="table" rid="pcbi-0020079-t004">Table 4</xref>). Each panel plots the conservation (accuracy of homology transfer) using a different measure for sequence similarity: HVAL (<xref ref-type="disp-formula" rid="pcbi-0020079-e001">Equation 1</xref>), PIDE (percentage pairwise sequence identity), and the PSI-BLAST E-value. It is surprising that even at high similarity thresholds (PIDE > 50; HVAL > 30), accuracy remained low and never reached levels of 20%. This behavior was partially explained by our overlap analysis: for low overlap (Equations 2 and 3) between datasets, we expect a low accuracy. Numbers at HVAL = 40 (which equals a PIDE of 68 at an alignment length of 100 residues) were marked with red lines. HVAL = 40 is the point, where the overlap-values (<xref ref-type="disp-formula" rid="pcbi-0020079-e003">Equation 3</xref>) for two identical datasets seem to indicate a zone of > 70% data consistency (see <xref ref-type="table" rid="pcbi-0020079-t003">Table 3</xref>). Error bars for the three plots were calculated by bootstrapping over the PPIs in the source datasets (see Methods section).</p></caption><graphic xlink:href="pcbi.0020079.g002"/></fig><fig id="pcbi-0020079-g003" position="float"><label>Figure 3</label><caption><title>Performance of Homology Transfer</title><p>Plots compiled for experiments 2–7 in <xref ref-type="table" rid="pcbi-0020079-t004">Table 4</xref>. Each of the upper three graphs stands for one particular organism o and shows two plots: (1) Use all known PPIs (large-scale and small-scale) of organism o to find Y2H large-scale detected PPIs in the same organism (but from different experiment, blue line). (2) Use all PPIs (large-scale and small-scale) of all other organisms (not o) to find PPIs detected by Y2H in o (red line). Only organisms with available Y2H datasets in IntAct were chosen in order to be able to create complete interaction matrices for the target datasets (yeast, worm, and fruit fly). All error bars were calculated through bootstrapping over the source PPIs (100 times, Methods). Some lines end at certain thresholds because the counts for true positives and false positives were too low (< 30 true or false positives) to calculate accuracy (Equation 4, see <xref ref-type="sec" rid="s4">Materials and Methods</xref>, often also referred to as specificity or precision). <xref ref-type="supplementary-material" rid="pcbi-0020079-sg001">Figure S1</xref> shows the correlation between the size of the error bars and the counts of true positives at each HSSP-value cutoff. The three bottom plots show ROC-like curves, where accuracy is plotted versus coverage for the exact same data as for the three upper plots. The figures demonstrate that for all levels of similarity, the accuracy of intraspecies predictions of PPIs is significantly higher than for predictions across two organisms.</p></caption><graphic xlink:href="pcbi.0020079.g003"/></fig><fig id="pcbi-0020079-g004" position="float"><label>Figure 4</label><caption><title>Interspecies Failure and Intraspecies Success of Homology Transfer</title><p>(A) Same family, different ancestors, different PPI: Two yeast peroxisomal proteins (<italic>PEX1</italic> and <italic>PEX2</italic>) are closely related through their common ancestor protein and their function as AAA ATPases to the two yeast <italic>26S protease regulatory subunits 6A</italic> and <italic>6B</italic>. In the fruit fly, gene duplication of a second ancestor protein (the <italic>NSF</italic> ancestor) led to two distinct <italic>NSF</italic> proteins (<italic>NSF1</italic> and <italic>2</italic>). Since the ancestors for the NSFs (<italic>NSF1</italic> and <italic>2</italic>) and for the <italic>26S protease subunits</italic> were two different proteins, we conclude that despite their common biochemical function as ATPases, the different cellular functions of NSFs and 26S protease subunits also led to a distinct behavior with respect to protein–protein interactions. Therefore, neither <italic>NSF1</italic> nor <italic>NSF2</italic> were observed to bind to the <italic>26S protease subunit 4</italic>.</p><p>(B) Same pathway, different functions, different binding: Evolutionary plasticity in the <italic>chk2</italic> family led to a diverse range of functions of these proteins while staying in the same pathway. For example <italic>Rad53p</italic> in yeast is a main player in the cell cycle checkpoint during mitosis, whereas <italic>Mek1p</italic> acts in the same position during meiosis. Also, <italic>drosophila chk2</italic> and human <italic>chk2</italic> act at different times during the cell cycle different from <italic>Mek1p</italic> and <italic>Rad53p</italic>. No <italic>drosophila Pp1</italic> homolog in yeast was found to interact with either <italic>Mek1p</italic> or <italic>Rad53p</italic>, even though <italic>drosophila Pp1</italic> was shown to bind to <italic>drosophila</italic> chk2.</p></caption><graphic xlink:href="pcbi.0020079.g004"/></fig><p>Here, we presented the analysis of PPI in, to our knowledge, the largest data set investigated thus far. We defined and measured the overlap between different data sets, and analyzed the expected levels of accuracy and coverage for homology-based inference of PPIs depending on the level of sequence similarity. The most surprising finding originated from differentiating between intraspecies and interspecies inferences (o ≠ p in <xref ref-type="fig" rid="pcbi-0020079-g001">Figure 1</xref>), namely that PPIs are more conserved within than between organisms.</p></sec></sec><sec id="s2"><title>Results/Discussion</title><sec id="s2a"><title>Different Experiments Overlap Very Little</title><p>If we want to homology infer PPIs between organisms, we first have to measure the overlap within organisms and then between organisms. We introduced such a measure (<xref ref-type="disp-formula" rid="pcbi-0020079-e002">Equation 2</xref> and <xref ref-type="disp-formula" rid="pcbi-0020079-e003">Equation 3</xref>, see <xref ref-type="sec" rid="s4">Materials and Methods</xref>) and applied it to assessing the overlap between datasets in IntAct [<xref rid="pcbi-0020079-b045" ref-type="bibr">45</xref>]. A large overlap value implies high agreement between two experimental sets of interactions. Our definition of overlap takes into account that two data sets may not have used the same proteins thereby rendering a score that is, in principle, independent of the size of common subsets (see <xref ref-type="sec" rid="s4">Materials and Methods</xref> section). The scores are straightforward when comparing different datasets within the same organism (<xref ref-type="disp-formula" rid="pcbi-0020079-e002">Equation 2</xref>) because we only have to identify identical pairs of proteins. As noted before [<xref rid="pcbi-0020079-b022" ref-type="bibr">22</xref>,<xref rid="pcbi-0020079-b046" ref-type="bibr">46</xref>–<xref rid="pcbi-0020079-b049" ref-type="bibr">49</xref>], the data sets overlap maximally for about 30% of all PPIs in yeast (<italic>Saccharomyces Cerevisiae</italic>) and much less for PPIs in fly (<italic>Drosophila Melanogaster</italic>, <xref ref-type="table" rid="pcbi-0020079-t001">Table 1</xref>). Interspecies comparisons are trickier because we now have to identify the corresponding homologous pairs in the other organism. <xref ref-type="disp-formula" rid="pcbi-0020079-e003">Equation 3</xref> solves this problem by counting homologous instead of identical pairs of proteins; it is applicable to intraspecies and interspecies comparisons. A consequence of counting homologous rather than identical protein pairs is that the same data set no longer overlaps 100% with itself (<xref ref-type="table" rid="pcbi-0020079-t002">Table 2</xref>), because the interaction between A and B may be detected while that between the homologs A′ and B′ may not be. The application of <xref ref-type="disp-formula" rid="pcbi-0020079-e003">Equation 3</xref> to the intraspecies comparison for yeast and fly datasets yielded similar results as the application of <xref ref-type="disp-formula" rid="pcbi-0020079-e002">Equation 2</xref> to the same datasets (<xref ref-type="table" rid="pcbi-0020079-t001">Table 1</xref>). The overlap between different yeast datasets seems to be generally higher than that between different fly datasets. Finally, we merged datasets of different large-scale experiments for each organism and compared these pseudo-complete PPIs between organisms by using <xref ref-type="disp-formula" rid="pcbi-0020079-e003">Equation 3</xref> (<xref ref-type="table" rid="pcbi-0020079-t003">Table 3</xref>). As expected the overlap between organisms was increased with increasing thresholds in what was considered homologous (<xref ref-type="table" rid="pcbi-0020079-t003">Table 3</xref>; HSSP-value (HVAL)>40 highest, HVAL>0 lowest, <xref ref-type="disp-formula" rid="pcbi-0020079-e001">Equation 1</xref>; note that the HSSP value (homology derived secondary structure of proteins) is an empirical measure for sequence similarity that empirically embeds the simple fact that high levels of sequence similarity are less meaningful for short than they are for long alignments). This increase in overlap was achieved by finding fewer matches (<xref ref-type="table" rid="pcbi-0020079-t003">Table 3</xref>, empty cells). Conversely, the overlap was very low at levels of sequence similarity that mark the twilight zone of sequence-structure inference [<xref rid="pcbi-0020079-b025" ref-type="bibr">25</xref>], i.e., the line above which most pairs of proteins have largely similar structure (HVAL>0, <xref ref-type="table" rid="pcbi-0020079-t003">Table 3</xref>). In other words, overall fold similarity does not suffice to infer similarity in interactions.</p><table-wrap id="pcbi-0020079-t001" content-type="1col" position="float"><label>Table 1</label><caption><p>Identity-Based Overlap (<xref ref-type="disp-formula" rid="pcbi-0020079-e002">Equation 2</xref>) between Original Experimental Y2H Datasets from Fly and Yeast</p></caption><graphic xlink:href="pcbi.0020079.t001"/></table-wrap><table-wrap id="pcbi-0020079-t002" content-type="1col" position="float"><label>Table 2</label><caption><p>Homology-Based Overlap (<xref ref-type="disp-formula" rid="pcbi-0020079-e003">Equation 3</xref>) between Original Experimental Y2H Datasets from Fly and Yeast</p></caption><graphic xlink:href="pcbi.0020079.t002"/></table-wrap></sec><sec id="s2b"><title>Automatic Homology Transfer of PPIs Is Very Limited</title><p>We generated a homology performance plot (see <xref ref-type="sec" rid="s4">Materials and Methods</xref> section) by comparing an unbiased, nonredundant data set (no two pairs of proteins in the set had significant sequence similarity (see <xref ref-type="sec" rid="s4">Materials and Methods</xref> section) against the redundant set with all PPIs (note that we removed identical pairs even in this set, <xref ref-type="table" rid="pcbi-0020079-t004">Table 4</xref>, Experiment 1). When using the observed PPI between two proteins (A-B), we applied the same sequence similarity threshold to identify both homologs (A/A′, B/B′) to infer the PPI between A′-B′. Pairs such as A-B′ or A′-B were not counted because those pairs could only be detected within the same organism and not across two species. Not surprisingly, the accuracy of homology transfer was proportional to sequence similarity (<xref ref-type="fig" rid="pcbi-0020079-g002">Figure 2</xref>). However, accuracy dropped rapidly already at very high levels of sequence similarity (e.g., at ~80% pairwise sequence identity, and below position-specific iterative basic local alignment search tool expectation values [PSI-BLAST E-values] < 10<sup>−150</sup>). Closer inspection of the HSSP formula (<xref ref-type="disp-formula" rid="pcbi-0020079-e001">Equation 1</xref>) reveals that the curves for HSSP values and percentage sequence identity were very similar to each other. The problem with E-values largely originated from including short alignments, i.e., many of the proteins identified at very significant E-values (E < 10<sup>−50</sup>) might have been aligned to only small fractions of the source protein. This is a known limitation of E-values that cannot easily be normalized away because PPI interfaces may be rather short (i.e., even alignments of 20 residues in very long proteins may correctly reflect binding similarity). Although the small overlap between experimental data sets (<xref ref-type="table" rid="pcbi-0020079-t003">Table 3</xref>) suggested that these estimates for accuracy at a given similarity threshold were most likely overpessimistic, the overlap scores also showed that at HVAL > 40, the consistency of the data was above 70% (<xref ref-type="table" rid="pcbi-0020079-t003">Table 3</xref>). Therefore, our estimates at such high thresholds might be approximately correct; if so, the accuracy of homology transfer for high similarity (HVAL > 40, Percentage sequence IDEntity (PIDE) > 70) were just over 10% (<xref ref-type="fig" rid="pcbi-0020079-g002">Figure 2</xref>). Clearly, our findings suggested that automatic homology-based inferences of PPIs have to be taken with extreme caution.</p></sec><sec id="s2c"><title>Homology Transfer Is Better within than between Organisms</title><p>Arguably [<xref rid="pcbi-0020079-b040" ref-type="bibr">40</xref>–<xref rid="pcbi-0020079-b042" ref-type="bibr">42</xref>], homology transfer is expected to be slightly better between organisms than within organisms. Instead, we observed the extreme opposite (<xref ref-type="fig" rid="pcbi-0020079-g003">Figure 3</xref>): at all levels of sequence similarity, and for all organisms with sufficient data, homology-inference was significantly more accurate for pairs of homologs from the same organism (intraspecies) than for pairs of homologs between different organisms (interspecies). In other words, if we experimentally observed the interaction between A and B in yeast, and if we found another pair of similar proteins A′ and B′ in yeast (not A-B′ or A′-B), as well as another pair A″ and B″ in fruit fly, then the interactions between A′ and B′ would be much more likely than those between A″ and B″. Consequently, yeast would be a rather poor model organism for the interaction network in fly.</p><p> <xref ref-type="table" rid="pcbi-0020079-t004">Table 4</xref> and <xref ref-type="fig" rid="pcbi-0020079-g002">Figures 2</xref> and <xref ref-type="fig" rid="pcbi-0020079-g003">3</xref> clearly establish our main messages that intraspecies homology transfer is more accurate than interspecies transfer and that homology transfer is accurate only at unexpectedly high levels of sequence similarity. These results were stable with respect to different ways of processing the data for the experimental interactions. Changes that influenced the outcome insignificantly included the following alternatives.</p></sec><sec id="s2d"><title>Results Were Stable with Respect to Details in Filtering Data</title><p>(1) Different sampling of intraspecies vs. interspecies: We allowed transfers of the type A-B to A′-B or A-B to A-B′ (see <xref ref-type="sec" rid="s4">Materials and Methods</xref> section). The performance became significantly better for intraspecies PPI transfers, thus further widening the gap between intraspecies and interspecies transfers (<xref ref-type="supplementary-material" rid="pcbi-0020079-sg002">Figure S2</xref>A). (2) Inclusion of transfers within the same data set: we included homology transfers within the same experimental dataset (see <xref ref-type="sec" rid="s4">Materials and Methods</xref> section). The effect was very similar to those observed for different sampling (see #1), i.e., the gap was widened between intraspecies and interspecies inferences (<xref ref-type="supplementary-material" rid="pcbi-0020079-sg002">Figure S2</xref>B). (3) We used TAP-like data (<xref ref-type="supplementary-material" rid="pcbi-0020079-st001">Table S1</xref>) as a constraint for the negatives. To illustrate this, assume that TAP pulled down a complex of six proteins. While we cannot infer that all 15 possible interactions are physical, all could be. Therefore, we ignored a false positive prediction (i.e., we did not count it) if we could find the interaction in those 15 TAP protein–protein pairs. The accuracy slightly increased for both yeast versus yeast (intraspecies) comparisons as well as for nonyeast versus yeast (interspecies) comparisons (<xref ref-type="supplementary-material" rid="pcbi-0020079-sg002">Figure S2</xref>C). Note that yeast is the only organism with available TAP-like data. (4) We used a redundant dataset (instead of a nonredundant, bias-reduced set) from organism o (<xref ref-type="fig" rid="pcbi-0020079-g007">Figure 7</xref>) to hunt for interologs in organism p (<xref ref-type="fig" rid="pcbi-0020079-g007">Figure 7</xref>). The main message indicated by the results for this latter experiment stays the same as in our original procedure (see <xref ref-type="sec" rid="s4">Materials and Methods</xref> section): Intraspecies comparisons are more accurate than interspecies comparisons. Because there were more samples in the dataset for organism o (<xref ref-type="fig" rid="pcbi-0020079-g007">Figure 7</xref>) and thus higher counts, the errors slightly decreased (<xref ref-type="supplementary-material" rid="pcbi-0020079-sg002">Figure S2</xref>D).</p></sec><sec id="s2e"><title>Examples</title><p>In the following, we presented a few representative examples that illustrate these points with more details than it is possible through averages over large data sets. Both show how homology transfer fails across species while it succeeds within an organism (Ao-Bo observed, A′o-B′o observed, A″m-B″m not observed).</p><sec id="s2e1"><title>Example 1: same family, different ancestors, different PPI.</title><p>The two peroxins <italic>PEX1</italic> and <italic>PEX6</italic> are known to functionally and physically interact in both human [<xref rid="pcbi-0020079-b050" ref-type="bibr">50</xref>] and yeast [<xref rid="pcbi-0020079-b051" ref-type="bibr">51</xref>–<xref rid="pcbi-0020079-b053" ref-type="bibr">53</xref>] (<xref ref-type="fig" rid="pcbi-0020079-g004">Figure 4</xref>A). A particular mutation in human <italic>PEX1</italic> disrupts the interaction with <italic>PEX6</italic>, and appears directly linked to the Zellweger Syndrome, an autosomal, recessive peroxisome biogenesis disorder, in which the growth of the myelin sheath (the fatty cover of nerve cells in the brain) is strongly affected. Patients usually suffer from visual disturbances, high iron and copper blood levels, and enlarged livers [<xref rid="pcbi-0020079-b053" ref-type="bibr">53</xref>]. Both proteins <italic>PEX1</italic> and <italic>PEX6</italic> belong to the ATPases associated with various cellular activities (AAA) family and are involved in the import of proteins into the peroxisome [<xref rid="pcbi-0020079-b052" ref-type="bibr">52</xref>,<xref rid="pcbi-0020079-b053" ref-type="bibr">53</xref>]. Thereby, the complex of <italic>PEX1</italic> and <italic>PEX6</italic> is associated with the cytoplasmic side of the peroxisomal membrane [<xref rid="pcbi-0020079-b051" ref-type="bibr">51</xref>]. Searching for proteins that are sequence-similar to <italic>PEX1</italic> and <italic>PEX6</italic> within yeast at an HVAL > 20 (<xref ref-type="disp-formula" rid="pcbi-0020079-e001">Equation 1</xref>, see <xref ref-type="sec" rid="s4">Materials and Methods</xref>) brought up two <italic>26S protease regulatory subunits, 6A</italic> and <italic>6B</italic> (proteins A′o and B′o); experts have also classified both these yeast proteins as AAA ATPases (<xref ref-type="fig" rid="pcbi-0020079-g004">Figure 4</xref>A). The interaction between these two yeast proteins was surprisingly found in all Y2H large scale protein–protein interaction scans [<xref rid="pcbi-0020079-b013" ref-type="bibr">13</xref>–<xref rid="pcbi-0020079-b015" ref-type="bibr">15</xref>,<xref rid="pcbi-0020079-b035" ref-type="bibr">35</xref>]. Using the same threshold (HVAL > 20) the closest proteins in fly were the <italic>26S protease subunit 4</italic> and the <italic>NEM-sensitive fusion protein 2</italic> (<italic>NSF2</italic>) (<xref ref-type="fig" rid="pcbi-0020079-g004">Figure 4</xref>A). The latter<italic>—NSF2</italic>— is a special form of the <italic>NEM-sensitive fusion protein 1</italic> (<italic>NSF1</italic>) and is fly-specific in the sense that it does not exist in yeast, worm, or human [<xref rid="pcbi-0020079-b054" ref-type="bibr">54</xref>–<xref rid="pcbi-0020079-b056" ref-type="bibr">56</xref>]. An interaction between <italic>26S protease subunit 4</italic> and <italic>NSF2</italic> was not found in any of our PPI <italic>drosophila</italic> datasets, nor has it been reported in the literature. <italic>NSF2</italic> is, among other things, responsible for exocytose through vesicle fusion by disassembling the postfusion SNARE protein complexes [<xref rid="pcbi-0020079-b054" ref-type="bibr">54</xref>,<xref rid="pcbi-0020079-b057" ref-type="bibr">57</xref>]. Like the other <italic>PEX1</italic> and <italic>PEX6</italic> relatives discussed so far, <italic>NSF2</italic> is also an ATPase [<xref rid="pcbi-0020079-b054" ref-type="bibr">54</xref>]. A detailed phylogenetic analysis of all proteins in the AAA family has suggested three major subfamilies, one with NSF homologs (<italic>NSF1</italic> and <italic>2</italic>), one with the <italic>26S protease subunits</italic>, and a third with <italic>p97/Cdc48p</italic> homologs [<xref rid="pcbi-0020079-b056" ref-type="bibr">56</xref>]. Most importantly these three subfamilies apparently did not arise from a common ancestor but rather, they evolved independently during speciation [<xref rid="pcbi-0020079-b056" ref-type="bibr">56</xref>].</p><table-wrap id="pcbi-0020079-t003" content-type="1col" position="float"><label>Table 3</label><caption><p>Homology-Based Overlap (<xref ref-type="disp-formula" rid="pcbi-0020079-e003">Equation 3</xref>) between Merged Datasets for Different Similarity Thresholds</p></caption><graphic xlink:href="pcbi.0020079.t003"/></table-wrap><table-wrap id="pcbi-0020079-t004" content-type="1col" position="float"><label>Table 4</label><caption><p>Datasets Used for Homology Performance Plots</p></caption><graphic xlink:href="pcbi.0020079.t004"/></table-wrap><p>This particular example illustrated how yeast may generally be a rather poor model organism for more complex species such as fly, worm or vertebrates. Proteins from these higher eukaryotes have to perform many different tasks in often highly specialized cell types (e.g., nerve cells). This might have lead to an evolutionary pressure to build new protein-interaction networks from the available protein building blocks (e.g., ATPase function). Thus, by only slightly altering the existing sequences, new binding properties were added to these proteins, while others were lost. A similar argument could be used to explain a likely poor homology transfer between fly and human or worm and human.</p></sec><sec id="s2e2"><title>Example 2: same pathway, different functions, different binding properties.</title><p>The <italic>drosophila Ser/Thr protein phosphatase 4</italic> (<italic>Pp4</italic>) and the <italic>cyclin dependent kinase 4</italic> (<italic>Cdk4</italic>) were found in our small-scale dataset for <italic>drosophila</italic> PPIs. At HVAL>20, we found two sequence-similar proteins in fly, namely <italic>Ser/Thr protein phosphatase alpha 2</italic> (<italic>Pp1</italic>) similar to <italic>Pp4</italic>, and <italic>chk2</italic> similar to <italic>Cdk4</italic>; both these fly proteins (<italic>Pp1</italic> and <italic>chk2</italic>) have been shown to interact [<xref rid="pcbi-0020079-b016" ref-type="bibr">16</xref>]. Fly <italic>chk2</italic> as well as its sequence relatives in yeast (<italic>Mek1p</italic> and <italic>Rad53p</italic>) and human are involved in cell-cycle checkpoints, which are signal transduction pathways that control the cell cycle and prevent the cell from further replication if the DNA double strand breaks, the DNA is incompletely replicated, or in case of other DNA damages [<xref rid="pcbi-0020079-b058" ref-type="bibr">58</xref>–<xref rid="pcbi-0020079-b060" ref-type="bibr">60</xref>]. A checkpoint can halt an ongoing mitosis or meiosis or even terminate it and induce apoptosis. A phylogenetic analysis of the <italic>chk2</italic> family members found that fly <italic>chk2</italic> and its yeast and human homologs stem from the same ancestor (<xref ref-type="fig" rid="pcbi-0020079-g004">Figure 4</xref>B). Nevertheless, it is also known that this family of proteins has a rather strong evolutionary plasticity in terms of the particular tasks of its members [<xref rid="pcbi-0020079-b060" ref-type="bibr">60</xref>,<xref rid="pcbi-0020079-b061" ref-type="bibr">61</xref>]. For example in yeast, <italic>Mek1p</italic> only controls the meiotic pachytene checkpoint by making sure that only homologous chromosomes recombine with each other [<xref rid="pcbi-0020079-b061" ref-type="bibr">61</xref>], whereas yeast <italic>Rad53p</italic> controls mitotic cell replication and does not seem to be required for meiotic checkpoint control at all [<xref rid="pcbi-0020079-b060" ref-type="bibr">60</xref>]. Also, the timing within the cell cycle is different for yeast <italic>Rad53p</italic> and its <italic>drosophila</italic> ortholog <italic>chk2</italic> [<xref rid="pcbi-0020079-b060" ref-type="bibr">60</xref>]. This plasticity in the chk2 family might explain why many yeast proteins homologous to <italic>drosophila Pp1</italic> were not found to interact with either <italic>Rad53p</italic> or <italic>Mek1p</italic>.</p></sec></sec><sec id="s2f"><title>Sequence-Based Homology Transfer Is Limited Although Binding Sites Are Partially Conserved in Three-Dimensional (3-D) Structure</title><p>Recently, the Sali group analyzed the conservation of protein–protein binding sites on homologous and structurally aligned protein surfaces. They found that the differences in the localization of binding sites between homologous proteins are significantly smaller than the differences expected at random [<xref rid="pcbi-0020079-b062" ref-type="bibr">62</xref>]. On the one hand, this result is similar to what we found for higher levels of similarity (<xref ref-type="fig" rid="pcbi-0020079-g003">Figure 3</xref>). On the other hand of very little similarity the difference between the 3-D–based results and ours lie most likely in the additional constraints implicitly used by the Sali group, namely that we know the 3-D structures and that we can focus in our alignment on all residues in the binding site. Using only sequence information, we cannot do this because binding residues close in 3-D may be separated considerably in sequence, thereby diluting the pattern of conservation picked up by alignment methods. However, for most PPIs from IntAct, we can neither label the binding site, nor do we have 3-D structural information. Therefore, we are limited to having to measure overall sequence similarity. If we were able to predict binding sites [<xref rid="pcbi-0020079-b063" ref-type="bibr">63</xref>–<xref rid="pcbi-0020079-b066" ref-type="bibr">66</xref>], we might improve homology transfer considerably.</p></sec></sec><sec id="s3"><title>Conclusions</title><p>As demonstrated again by our overlap measure, today's datasets of PPIs are still rather inconsistent (<xref ref-type="table" rid="pcbi-0020079-t001">Tables 1</xref>–<xref ref-type="table" rid="pcbi-0020079-t003">3</xref>). The discrepancies were significantly smaller between yeast than between fly datasets (<xref ref-type="table" rid="pcbi-0020079-t001">Tables 1</xref> and <xref ref-type="table" rid="pcbi-0020079-t002">2</xref>). This finding also explains the much higher accuracy for intrayeast as opposed to intrafly or intraworm transfer. Why datasets of yeast appear more consistent than those of fly datasets remains speculation. One reason might be that measurements of protein–protein interactions are performed within yeast (Y2H) and are thus more precise for yeast proteins than for other species′ proteins, since those might behave differently in the unfamiliar yeast cell. Although incomplete and not fully consistent, PPI datasets are finally large enough to validate quantitative analyses. In particular, this enables a large-scale assessment of the performance of automated homology transfer for PPIs. Assuming that today's errors are largely nonsystematic, estimates for the performance of homology transfer will provide correct qualitative pictures, albeit the actual numbers will be overpessimistic. In the extreme regimen of comparing very similar pairs of proteins, we could establish that data sets appeared very consistent (<xref ref-type="fig" rid="pcbi-0020079-g002">Figure 2</xref>). Consequently, our estimates for the performance of homology transfer were likely to be relatively reliable in this regimen. Nevertheless, even for very high similarity, automated homology transfer was often mistaken; it approached random when approaching the sequence-structure twilight zone, i.e. the region in which sequence similarity no longer implies 3-D similarity (<xref ref-type="fig" rid="pcbi-0020079-g003">Figure 3</xref>). Although many interactions observed in one organism were not observed in another, similar interactions in the same organism (at similar levels of sequence similarity) were often observed (<xref ref-type="fig" rid="pcbi-0020079-g003">Figure 3</xref>). Consequently, our results challenge that using homology to transfer a protein–protein interaction from one organism to another is more difficult and less accurate than a transfer within the same species. This implies that distant model organisms have a limited value to unravel protein networks. We showed that these results are stable even when making major changes to the ways in which we analyzed the experimental data. Whether we used high- or low-confidence data, whether we allowed for same-set PPI transfers or not, whether we reduced bias or not, whether or not we filtered the negatives by TAP-like data about putative physical interactions, whether or not we restricted our analysis to limited inferences per family, we always observed the same: PPIs are more conserved within than across species. This discrepancy between intraspecies and interspecies conservation of interologs was valid for all levels of sequence similarity. Finally, we tested the ability of homology transfers to predict another functional annotation and then compared the performances of interspecies versus intraspecies comparisons thereof. We chose subcellular localization as an easily extractable and available protein feature. By using a list of proteins annotated for subcellular localizations from UniProt [<xref rid="pcbi-0020079-b067" ref-type="bibr">67</xref>], we could show that there is no significant difference in performances for interspecies versus intraspecies homology transfers for this particular feature.</p></sec><sec id="s4"><title>Materials and Methods</title><sec id="s4a"><title>Data sets.</title><p>Several publicly available databases such as GRID [<xref rid="pcbi-0020079-b068" ref-type="bibr">68</xref>], BIND [<xref rid="pcbi-0020079-b069" ref-type="bibr">69</xref>], MINT [<xref rid="pcbi-0020079-b070" ref-type="bibr">70</xref>], and DIP [<xref rid="pcbi-0020079-b071" ref-type="bibr">71</xref>,<xref rid="pcbi-0020079-b072" ref-type="bibr">72</xref>] gather information about interacting proteins in different organisms. For our analysis, we used the IntAct database [<xref rid="pcbi-0020079-b045" ref-type="bibr">45</xref>], a protein–protein interaction resource maintained at the European Bioinformaics Institute (EBI) in Cambridge (<ext-link ext-link-type="uri" xlink:href="http://www.ebi.ac.uk/intact/">http://www.ebi.ac.uk/intact/</ext-link>). IntAct uses the PSI format (extended markup language (XML)-tagged) to store data [<xref rid="pcbi-0020079-b073" ref-type="bibr">73</xref>], fly [<xref rid="pcbi-0020079-b012" ref-type="bibr">12</xref>–<xref rid="pcbi-0020079-b015" ref-type="bibr">15</xref>], fly [<xref rid="pcbi-0020079-b011" ref-type="bibr">11</xref>,<xref rid="pcbi-0020079-b016" ref-type="bibr">16</xref>,<xref rid="pcbi-0020079-b017" ref-type="bibr">17</xref>], worm [<xref rid="pcbi-0020079-b018" ref-type="bibr">18</xref>] and human [<xref rid="pcbi-0020079-b019" ref-type="bibr">19</xref>] as well as about 30 so called small-scale datasets, which are collections of results from many detailed experiments for different organisms. The largest small-scale dataset is that of human with about 38,000 interactions. Concerning the high-throughput datasets, IntAct carries detailed information about which proteins were used as baits and which proteins were used as preys, so that a complete interaction matrix can easily be reconstructed from these sets. <xref ref-type="supplementary-material" rid="pcbi-0020079-st001">Table S1</xref> contains all protein–protein interaction datasets deposited in IntAct at the moment along with links to these datasets (small-scale and large-scale). The Giot [<xref rid="pcbi-0020079-b017" ref-type="bibr">17</xref>], Ito [<xref rid="pcbi-0020079-b035" ref-type="bibr">35</xref>], and Li [<xref rid="pcbi-0020079-b018" ref-type="bibr">18</xref>] datasets contain some information about the level of confidence that was assigned to each interaction. For these three sets, we excluded everything from our analysis that either had a confidence-value of less than 0.4 (Giot: values range from 0 to 1) or those that were not in a so called “core” dataset of trusted interactions (Ito and Li divide their sets into core and full or core and noncore subsets, where core means a higher confidence in the measured interaction). Note that for the initial submission of this manuscript we had compiled all results for unfiltered data sets, i.e., we had included all experimental interactions; the results were qualitatively identical to those given here (data not shown).</p></sec><sec id="s4b"><title>True positives and false negatives: focus on Physical Interactions = PPIs.</title><p>Technically, we realized our goal of exclusively focusing on PPIs through the particular way of labeling positives and negatives. We labeled as positives (true PPIs) only those pairs that were identified by experiments that target the detection of physical interactions (only Y2H experiments).</p><p>We then also assumed that these data for each organism was complete, i.e., we labeled all pairs as negatives that were not detected by Y2H.</p></sec><sec id="s4c"><title>Measuring sequence similarity/homology.</title><p>The term homology usually implies an evolutionary relation in the sense of having a common ancestor. Strictly speaking, we cannot measure homology. Instead, alignment methods measure sequence similarity in some way or other. In our work the ranges of similarity were so high that the pairs of proteins were most likely homologous. We used BLAST and PSI-BLAST [<xref rid="pcbi-0020079-b074" ref-type="bibr">74</xref>] to align all protein sequences in IntAct against each other (standard procedure [<xref rid="pcbi-0020079-b075" ref-type="bibr">75</xref>]: 3 iterations at E<10-<sup>10</sup> against filtered database of all proteins to build clean profiles, then one run with frozen profile against unfiltered database at E < 10<sup>−3</sup>, freeze profile again and run against all IntAct proteins). Then we extracted the PSI-BLAST E-values for each alignment, as well as the percentage of sequence identity (PIDE) and the distance to the HSSP curve, i.e. the HSSP-value [<xref rid="pcbi-0020079-b025" ref-type="bibr">25</xref>,<xref rid="pcbi-0020079-b076" ref-type="bibr">76</xref>,<xref rid="pcbi-0020079-b077" ref-type="bibr">77</xref>] (HVAL). The HVAL is defined as: <disp-formula id="pcbi-0020079-e001"><graphic xlink:href="pcbi.0020079.e001.jpg" position="anchor" mimetype="image"/></disp-formula>where L was the number of residues aligned between two proteins, and PIDE the percentage of pairwise identical residues. HSSP values consider both pairwise sequence identity and alignment length: the higher the value the more similar two proteins. Values around 0 typically imply that two proteins have similar 3-D structures and correspond to about 22% pairwise sequence identity at alignment lengths above 250 residues. </p></sec><sec id="s4d"><title>Nonredundant data sets.</title><p>We removed bias from PPI datasets by the following procedure (<xref ref-type="fig" rid="pcbi-0020079-g005">Figure 5</xref>). (1) Move down a list L of PPIs starting with pair A-B. (2) Group all interactions in this list into clusters of similar PPIs. Consider two distinct PPIs as similar only if both partners of the first interaction are homologs to the respective protein in the second interaction. For instance, let A′ be a homolog of A, and B′ be a homolog of B. Then all interactions A′-B, A′-B′, and A-B′ will fall into the same group as the interaction A-B. Note that this also means that any interaction A-C will not end up in this group if C is not a homolog of B. Here, we used a very conservative criterion for homolog, namely HVAL > 0 (<xref ref-type="disp-formula" rid="pcbi-0020079-e001">Equation 1</xref>). This threshold is conservative in the sense that it will also remove nonredundant pairs, i.e., many proteins that are actually not homologs. (3) Reduce each group formed in step 2 to one single representative PPI. (4) Continue working with the final unique (nonredundant) dataset.</p><fig id="pcbi-0020079-g005" position="float"><label>Figure 5</label><caption><title>Creating Sequence-Unique PPI sets</title><p>(1) Starting with a dataset of PPIs, we first cluster the data according to sequence similarity (apply a certain homology threshold) into sequence similar PPIs (2). Note here that the interactions A′-B′ and A′-C′ do not fall into the same cluster because B′ and C′ are unrelated. Thus, for two interactions (e.g., A-B and A′-B′) to be considered similar by our algorithm, both interacting proteins (A and B) have to be homologous to the two proteins of the other interaction (A has to be similar to A′ and B has to be similar to B′). (3) We randomly throw out all redundant interactions in each cluster so that only one PPI remains as a representative of each cluster. (4) Those representatives constitute the final unique dataset of PPIs.</p></caption><graphic xlink:href="pcbi.0020079.g005"/></fig></sec><sec id="s4e"><title>Identity- and homology-based overlap between datasets.</title><p>We defined two procedures resembling the Jaccard correlation to measure the overlap between two different datasets of PPIs in IntAct. <xref ref-type="disp-formula" rid="pcbi-0020079-e002">Equation 2</xref> defines the first measure; for clarity we refer to this measure as the identity-based overlap. This measure can only be applied to two PPI sets from the same organism. <disp-formula id="pcbi-0020079-e002"><graphic xlink:href="pcbi.0020079.e002.jpg" position="anchor" mimetype="image"/></disp-formula>where <italic>PPI</italic>(<italic>MandN</italic>) is the number of PPIs that were detected in both sets (common PPIs) and <italic>PPI</italic>(<italic>MxorN</italic>) is the number of PPIs that were only detected in one of the two datasets (exclusive or). <xref ref-type="fig" rid="pcbi-0020079-g006">Figure 6</xref>A describes this procedure. Note that only those interactions contributed to the count of <italic>PPI</italic>(<italic>MxorN</italic>) that could possibly have been detected in both datasets. For example, if the PPI A-B is detected in dataset 1, but not in dataset 2, we only increase <italic>PPI</italic>(<italic>MxorN</italic>) by one, if A and B were both included in dataset 2. In other words, we completely ignored interactions A-B in one dataset, if either A, or B (or both) were not present in the other dataset. Given this definition (<xref ref-type="disp-formula" rid="pcbi-0020079-e002">Equation 2</xref>), an overlap value of 0.5 means that every second PPI of dataset 1 is not present in dataset 2. Inversely, every second PPI from dataset 2 cannot be found in dataset 1. Furthermore, applying <xref ref-type="disp-formula" rid="pcbi-0020079-e002">Equation 2</xref> to calculate the overlap of one dataset with itself always results in 1 (100% overlap). </p><fig id="pcbi-0020079-g006" position="float"><label>Figure 6</label><caption><title>Ways of Calculating the Overlap between Two Y2H Datasets</title><p>(A) Identity-based overlap between Datasets 1 and 2 according to <xref ref-type="disp-formula" rid="pcbi-0020079-e002">Equation 2</xref>. Note that we can only calculate this score if both datasets are from the same organism. Starting with the observed interaction C-E in Dataset 1, we are trying to find the exact same interaction in Dataset 2. The following situations might occur: (a) C and E are also observed to interact in Dataset 2. (b) C and E are not observed to interact in Dataset 2. (c) It is impossible for C and E to be interacting in Dataset 2 due to either of these two reasons: (i) Either C or E are not part of Dataset 2 or (ii) C and E are either both used as preys or both used as baits in Dataset 2. Repeating the above procedure for all other observed interactions in Datasets 1 and 2, we finally calculate the identity-based overlap by dividing the number of common interactions found in Datasets 1 and 2 by the total number of expected interactions (observed and not-observed).</p><p>(B) The same procedure as described above is applied to the two Datasets 1 and 3, which are now allowed to be from different organisms. The only difference to <xref ref-type="disp-formula" rid="pcbi-0020079-e002">Equation 2</xref> (A) is the usage of homology for comparing two PPIs instead of a binary decision scheme (PPIs identical or not-identical). Thus, starting with the interaction D-E from Dataset 1, we try to find possible homologous interactions (not only the identical PPI) in Dataset 3. The only two options in this example are D-E and D′-E (Dataset 3), which in our example are both observed in Dataset 3. Iterating through all observed interactions of Datasets 1 and 3 and summing up the expected interactions and the overlapping homologous interactions, we can then calculate the homology-based overlap (<xref ref-type="disp-formula" rid="pcbi-0020079-e003">Equation 3</xref>). Note that any results from <xref ref-type="disp-formula" rid="pcbi-0020079-e002">Equation 2</xref> are not comparable to any results from <xref ref-type="disp-formula" rid="pcbi-0020079-e003">Equation 3</xref>.</p></caption><graphic xlink:href="pcbi.0020079.g006"/></fig><p>The second measure capturing an overlap between two interaction datasets was applicable to any two datasets, even if they were from different organisms. We referred to this measure as the homology-based overlap. It was defined as follows (<xref ref-type="fig" rid="pcbi-0020079-g006">Figure 6</xref>B): <disp-formula id="pcbi-0020079-e003"><graphic xlink:href="pcbi.0020079.e003.jpg" position="anchor" mimetype="image"/></disp-formula>where <italic>PPI</italic>(<italic>MandN</italic>)<sup>(<italic>h</italic>)</sup> is the number of homologous PPIs reported in both datasets considering a homology threshold of HVAL > h. Assume again that A is homolog of A′ and B of B′. If the interaction A-B is in dataset 1 and the interaction A′-B′ is in dataset 2, the count for <italic>PPI</italic>(<italic>MandN</italic>)<sup>(<italic>h</italic>)</sup> will increase by one. The quantities <italic>PPI</italic>(<italic>MandN</italic>)<sup>(<italic>h</italic>)</sup> and <italic>PPI</italic>(<italic>MxorN</italic>)<sup>(<italic>h</italic>)</sup> are similar to those in <xref ref-type="disp-formula" rid="pcbi-0020079-e002">Equation 2</xref> with the simple caveat that we substituted identical pairs with homologous pairs, because there are no identical pairs between two different organisms. Unlike for <xref ref-type="disp-formula" rid="pcbi-0020079-e002">Equation 2</xref>, when using <xref ref-type="disp-formula" rid="pcbi-0020079-e003">Equation 3</xref> to measure the overlap between a dataset and itself, the result usually happens to be < 1 (< 100%). For an explanation consider the following example. Assume that our dataset contains the interaction A(bait)-B(prey) along with another protein A′ (bait, homologous to A) that is not found to interact with B. The absence of A′-B will increase the count of <italic>PPI</italic>(<italic>MxorN</italic>)<sup>(<italic>h</italic>)</sup> by one, thereby yielding a self overlap <1. On the one hand, for very high levels of similarity (say A and A′ have 99% pairwise sequence identity), the reduction from 1 can be interpreted as a reflection of the limitation of experimental accuracy. On the other hand, for low levels of similarity, the reduction is related to the fact that PPIs are simply not conserved between distant relatives. Note that we also investigated overlap when replacing HVAL (<xref ref-type="disp-formula" rid="pcbi-0020079-e001">Equation 1</xref>) by PSI-BLAST E-values as a measure for sequence similarity. While the resulting numbers differed slightly, the trends that we reported remained the same (data not shown). </p></sec><sec id="s4f"><title>Homology performance curves.</title><p>For given levels of sequence similarity, we monitored and plotted the accuracy of inferring PPIs through homology from one dataset to another. The procedure is described in <xref ref-type="fig" rid="pcbi-0020079-g007">Figure 7</xref>.</p><fig id="pcbi-0020079-g007" position="float"><label>Figure 7</label><caption><title>Evaluating Homology Inference of PPIs</title><p>Starting with the entirety of observed interactions in any organism o (Y2H plus small scale experiments), we first reduce the sequence redundancy from this dataset as described in <xref ref-type="fig" rid="pcbi-0020079-g003">Figure 3</xref>. Then we try to find homologs in the organism p for each of the unique PPIs of organism o. Since we want to be able to conclude that every nondetected interaction in organism p does actually not exist in real life, we need to have a complete interaction matrix (baits × preys) for organism p. Thus, we are forced to exclude all small-scale data from the organism p dataset and remain with a merger of all (redundant) Y2H interactions for this organism. For each interaction A-B from organism o, we can face any of the following situations: (a) A homologous interaction A′-B′ can be found in organism p, (b) no homologous interaction can be found in p, or (c) It is impossible to detect an interaction of type A′-B′ in p because of one of the following two reasons: (i) either A′ or B′ are missing in the dataset for p or (ii) Both A′ and B′ are either preys or both are baits in the dataset for organism p. The latter case (c.ii) is illustrated by the interaction E-F in organism o, which cannot be detected in organism p only because E′ and F′ are both used as preys in the experiment. No counts for false positives are made for those cases. Adding the numbers of true positives (expected and observed PPIs), false positives (expected but not observed) and false negatives (observed interaction only in organism p) allows us to calculate accuracy and coverage for each homology threshold used to infer interactions (Equation 4). It is important to note that in the case where o = p, comparisons between two identical experimental PPI-sets are ignored (e.g. A-B in o′s set “<italic>yeast-Ito-2001”</italic> is not used to predict A′-B′ in p′s set “<italic>yeast-Ito-2001”</italic>; o = p = yeast).</p></caption><graphic xlink:href="pcbi.0020079.g007"/></fig><p>The resulting curves can be interpreted as the degree to which PPIs are evolutionarily conserved. In a more technical sense, the curves reflect the performance of homology transfer of PPIs (<xref ref-type="fig" rid="pcbi-0020079-g001">Figure 1</xref>). The HVAL (<xref ref-type="disp-formula" rid="pcbi-0020079-e001">Equation 1</xref>) determined the minimal similarity between A and A′, as well as between B and B′. Other ways of considering two pairs of interacting proteins as related, for instance the arithmetic or geometric average of both HVALs (A/A′ and B/B′), led to a slightly worse performance of our homology inferences, i.e. the curves were similar albeit lower overall (data not shown). Note that each large-scale Y2H data set (<xref ref-type="supplementary-material" rid="pcbi-0020079-st001">Table S1</xref>) should, by experimental design, contain a complete interaction matrix (preys × baits) that is, ideally, both fully correct and comprehensive for all the proteins tested in that experiment. Consider an interaction A-B from any dataset (small-scale or large-scale) of an organism o; if we find the homologs A′ and B′ in a large-scale dataset of another organism p, we can transfer the interaction property from A-B to A′-B′. In other words, by looking at the PPI between A and B (A-B), we simply predict that A′ and B′ also interact. Because of the complete interaction matrix that we are looking at for organism p, we can now also say whether this prediction was actually right or wrong. In particular, the prediction is correct, if we find the interaction A′-B′ in p and wrong if we do not find it in p plus A′ and B′ are on different axes of the interaction matrix (A′ = prey, B′ = bait or vice versa). In order to compare the performance of homology transfers across two organisms (o ≠ p) to the one for intraorganism transfers (o = p), we have to allow p and o to be the same. Therefore, in order to be able to compare results from both types of experiments (intraspecies versus interspecies), we have to apply the following restrictions to comparisons within the same species (o = p): Transfers from an interaction A-B to another PPI of the type A-B′ or A′-B (one protein identical, the other homologous) are not allowed since these cases are only observable in intraspecies predictions but not in interspecies transfers. Additionally for intraspecies predictions, we required that A-B and the predicted interaction (A′-B′) stem from different datasets (different Y2H experiments) in order to ignore possible homology-based assumptions about two PPIs within the same dataset. The problem here is that in case a research group found an interaction (e.g., A-B) through a Y2H scan, would they work harder to also find an interaction A′-B′ (A′ = homolog to A, B′ = homolog to B) or A′-B rather than an unrelated interaction (e.g., M-N).</p></sec><sec id="s4g"><title>Accuracy and coverage.</title><p>We measured the accuracy (Acc) and coverage (Cov) for the inference (prediction) of interacting protein pairs by the standard formulas: <disp-formula id="pcbi-0020079-e004"><graphic xlink:href="pcbi.0020079.e004.jpg" position="anchor" mimetype="image"/></disp-formula>where TP are the true positives (i.e., physical interactions that are experimentally observed [e.g., by Y2H, note TAP-like relations are not included here] and that are also correctly inferred by homology). FP are the false positives (i.e., the pairs inferred through homology but not observed by Y2H experiments). Finally, FN are the false negatives (i.e., the physical interactions that have been observed but were not identified). We monitored levels of accuracy and coverage as a function of the sequence similarity between the proteins of known and those of unknown annotations. There is a trade-off between these two: the more restrictive the sequence similarity threshold, the more interactions will be inferred (higher coverage) at the expense of reduced accuracy; and the higher the threshold, the more will be right (high accuracy) at the expense of few inferences (low coverage). </p></sec><sec id="s4h"><title>Error estimate.</title><p>The error in the estimates of accuracy and coverage were determined by bootstrapping [<xref rid="pcbi-0020079-b078" ref-type="bibr">78</xref>] over the protein–protein interactions in the source datasets. In particular, we picked <italic>n</italic> interactions at random from the non-redundant source dataset and compiled the averages over a larger set with possibly many replicas of the same incidence. The levels of accuracy/coverage for different thresholds in sequence similarity were then calculated according to the procedure described above (<xref ref-type="fig" rid="pcbi-0020079-g007">Figure 7</xref>). For the bootstrapping, these two steps had been repeated 100 times before the standard deviation (sigma) for all levels of accuracy were calculated.</p></sec></sec><sec sec-type="supplementary-material" id="s5"><title>Supporting Information</title><supplementary-material content-type="local-data" id="pcbi-0020079-st001"><label>Table S1</label><caption><title>Large-Scale Protein–Protein Interaction Datasets from IntAct</title><p>(74 KB DOC)</p></caption><media xlink:href="pcbi.0020079.st001.doc"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020079-sg001"><label>Figure S1</label><caption><title>Number of true positive counts versus HVAL</title><p>Each curve shows the accuracy (red) as shown in <xref ref-type="fig" rid="pcbi-0020079-g003">Figure 3</xref> and the number of true positives counted at a certain HSSP-value cutoff (green)</p><p>(72 KB DOC)</p></caption><media xlink:href="pcbi.0020079.sg001.doc"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020079-sg002"><label>Figure S2</label><caption><title>Results Are Stable with Respect to Variations in the Experimental Setup</title><p>(A) Different sampling of intra- versus inter-species: we allowed transfers of the type A-B to A'-B or A-B to A-B' (see <xref ref-type="sec" rid="s4">Materials and Methods</xref> section). The performance became significantly better for intra-species PPI-transfers, thus further widening the gap between intra- and inter-species transfers.</p><p>(B) Inclusion of transfers within the same data set: we included homology transfers within the same experimental dataset (see <xref ref-type="sec" rid="s4">Materials and Methods</xref> section). The effect was very similar to those observed for different sampling (#1), i.e. widening the gap between intra- and inter-species inferences.</p><p>(C) Using TAP-like data (<xref ref-type="supplementary-material" rid="pcbi-0020079-st001">Table S1</xref>) as a constraint for the negatives. To illustrate this, assume that TAP pulled down a complex of six proteins. While we cannot infer that all 15 possible interactions are physical, all could be. Therefore, we ignored a false positive prediction (did not count it) if we could find the interaction in those 15 TAP protein-protein pairs. The accuracy slightly increased for both yeast versus yeast (intra-species) comparisons as well as for non-yeast versus yeast (inter-species) comparisons. Note that yeast is the only organism with available TAP-like data.</p><p>(D) We used a redundant dataset (instead of a non-redundant, bias-reduced set) from organism o (<xref ref-type="fig" rid="pcbi-0020079-g007">Figure 7</xref>) to hunt for interologs in organism p (<xref ref-type="fig" rid="pcbi-0020079-g007">Figure 7</xref>). The main message indicated by the results for this latter experiment (#4) stays the same as in our original procedure (see <xref ref-type="sec" rid="s4">Materials and Methods</xref> section): intra species comparisons are more accurate than inter-species comparisons. Due to more samples in the dataset for organism o (<xref ref-type="fig" rid="pcbi-0020079-g007">Figure 7</xref>) and thus higher counts, the errors slightly decreased.</p><p>(153 KB DOC)</p></caption><media xlink:href="pcbi.0020079.sg002.doc"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material></sec>
Toward a Census of Bacteria in Soil	<p>For more than a century, microbiologists have sought to determine the species richness of bacteria in soil, but the extreme complexity and unknown structure of soil microbial communities have obscured the answer. We developed a statistical model that makes the problem of estimating richness statistically accessible by evaluating the characteristics of samples drawn from simulated communities with parametric community distributions. We identified simulated communities with rank-abundance distributions that followed a truncated lognormal distribution whose samples resembled the structure of 16S rRNA gene sequence collections made using Alaskan and Minnesotan soils. The simulated communities constructed based on the distribution of 16S rRNA gene sequences sampled from the Alaskan and Minnesotan soils had a richness of 5,000 and 2,000 operational taxonomic units (OTUs), respectively, where an OTU represents a collection of sequences not more than 3% distant from each other. To sample each of these OTUs in the Alaskan 16S rRNA gene library at least twice, 480,000 sequences would be required; however, to estimate the richness of the simulated communities using nonparametric richness estimators would require only 18,000 sequences. Quantifying the richness of complex environments such as soil is an important step in building an ecological framework. We have shown that generating sufficient sequence data to do so requires less sequencing effort than completely sequencing a bacterial genome.</p>	<contrib contrib-type="author"><name><surname>Schloss</surname><given-names>Patrick D</given-names></name><xref ref-type="author-notes" rid="n105">¤</xref><xref ref-type="aff" rid="aff1"/></contrib><contrib contrib-type="author"><name><surname>Handelsman</surname><given-names>Jo</given-names></name><xref ref-type="corresp" rid="cor1">*</xref><xref ref-type="aff" rid="aff1"/></contrib>	PLoS Computational Biology	<sec id="s1"><title>Introduction</title><p>Enumerating the human population of a country or region through a census is an ancient problem that is complicated by the challenges inherent in accurately representing a large and often inaccessible population. The same issues manifest in censuses of microbial communities, but are intensified by greater complexity and methodological challenges. Although a complete census of a country is theoretically possible, it is currently impractical to survey all 10<sup>9</sup> bacterial cells in a gram of soil [<xref rid="pcbi-0020092-b001" ref-type="bibr">1</xref>], making a sample-based census the best option for estimating richness—the number of bacterial taxa in soil. To do so accurately requires a reliable means to access the bacteria, a reasonable definition of “species,” and a robust description of the frequency distribution of the species. Just as a country's census describes a fundamental property of that country, an environment's richness is the most fundamental descriptor of community structure, and patterns of richness can be correlated with an environment's geography, productivity, extremeness, climate change, and degree of isolation [<xref rid="pcbi-0020092-b002" ref-type="bibr">2</xref>]. Our inability to estimate richness impedes investigation of the effects of soil chemistry, pollution, and land use on the soil microbial community.</p><p>The method used to access the microbial biodiversity assuredly shapes the outcome of a census. Culture-based methods suggest that a gram of soil contains fewer than 100 species [<xref rid="pcbi-0020092-b003" ref-type="bibr">3</xref>], but these are undoubtedly underestimates because multiple lines of evidence indicate that fewer than 1% of the species in soil are presently culturable [<xref rid="pcbi-0020092-b004" ref-type="bibr">4</xref>]. Culture-independent methods include DNA reassociation and 16S rRNA gene sequencing, which have provided conflicting results due to the problems inherent in defining a species and in estimating the frequency distribution of species in soil. Depending on how the data are analyzed, DNA reassociation experiments produce richness estimates ranging from 4,000 to 10,000,000 genome equivalents per 10 or 30 g of soil [<xref rid="pcbi-0020092-b005" ref-type="bibr">5</xref>–<xref rid="pcbi-0020092-b011" ref-type="bibr">11</xref>]. The variability in these estimates stems from application of different assumptions to reassociation curves, and their interpretation is complicated by the lack of controls that account for intergenomic variation. Finally, DNA reassociation kinetics cannot be used to compare the membership of different communities.</p><p>An alternative method relies on analysis of 16S rRNA gene sequences amplified from soil by PCR [<xref rid="pcbi-0020092-b012" ref-type="bibr">12</xref>]. The power of this method lies in its use of the universal tool of bacterial phylogeny and our ability to define operational taxonomic units (OTUs) based on the relatedness of sequences. Estimates of richness have been obtained through parametric or nonparametric empirical models of species frequency distribution to produce richness estimates between 590 and 100,000 species per gram of soil [<xref rid="pcbi-0020092-b013" ref-type="bibr">13</xref>–<xref rid="pcbi-0020092-b015" ref-type="bibr">15</xref>]. Parametric models have assumed that the incidence of different species follows a lognormal [<xref rid="pcbi-0020092-b013" ref-type="bibr">13</xref>], Pareto [<xref rid="pcbi-0020092-b016" ref-type="bibr">16</xref>], or uniform distribution [<xref rid="pcbi-0020092-b014" ref-type="bibr">14</xref>]. Although the lognormal model has been useful as a “null model” [<xref rid="pcbi-0020092-b017" ref-type="bibr">17</xref>], data are insufficient from any soil community to support reliance on a lognormal or Pareto frequency distribution, and we are unaware of any dataset that supports a uniform frequency distribution [<xref rid="pcbi-0020092-b018" ref-type="bibr">18</xref>]. Analyses based on nonparametric models, which do not assume a defined frequency distribution but are based on the frequency of abundant community members [<xref rid="pcbi-0020092-b019" ref-type="bibr">19</xref>–<xref rid="pcbi-0020092-b021" ref-type="bibr">21</xref>], estimate a minimum richness of 590 species based on 16S rRNA gene sequences in a Scottish soil [<xref rid="pcbi-0020092-b015" ref-type="bibr">15</xref>,<xref rid="pcbi-0020092-b022" ref-type="bibr">22</xref>]. Although the extent of the universality of phylum-specific PCR primers and potential toxicity of some fragments is not well understood, these effects would reduce the perceived richness. Finally, use of 16S rRNA gene sequences permits direct comparisons of the membership of different communities.</p><p>Previously, Dunbar et al. [<xref rid="pcbi-0020092-b017" ref-type="bibr">17</xref>] modeled the frequency distribution of 16S rRNA genes in four Arizonan soil communities by fitting a lognormal frequency distribution. Using 200 16S rRNA gene fragments from each soil community, this analysis estimated that 10 g of soil contained between 3,000 and 8,000 16S rRNA gene restriction fragment length polymorphism (RFLP) profiles. Previous analysis using the same four libraries found that the similarity of 16S rRNA gene sequences with the same RFLP profile ranged between 52.2% and 99.9% [<xref rid="pcbi-0020092-b003" ref-type="bibr">3</xref>], which makes interpretation of the analysis difficult. We were interested in developing this approach further by analyzing large 16S rRNA gene sequence collections that had not been initially screened by RFLP profiling. Our approach was to find a simulated community whose samples resembled our sampling of 1,033 16S rRNA genes from a clone library constructed from a single 0.5-g sample of Alaskan soil. For the purposes of comparison, we also analyzed two large 16S rRNA gene sequence collections that were recently published as part of a soil metagenomic sequencing project, but were not characterized beyond their phylogenetic affiliation [<xref rid="pcbi-0020092-b023" ref-type="bibr">23</xref>].</p></sec><sec id="s2"><title>Results</title><sec id="s2a"><title>Estimating the Bacterial Richness in the Alaskan Soil Library</title><p>The aim of this work was to estimate the taxonomic richness in an Alaskan soil sample through a library of 16S rRNA gene sequences derived from the sample. We assigned more than 92% of the 1,033 Alaskan 16S rRNA gene sequences to seven phyla, including the Proteobacteria (48.6%), Acidobacteria (15.3%), Bacteroidetes (9.3%), Actinobacteria (5.8%), Gemmimonas (5.7%), Planctomycetes (4.0%), and Verrucomicrobia (4.0%); the remaining sequences clustered within 12 phylum-level delineations, four of which had no cultured members (<xref ref-type="fig" rid="pcbi-0020092-g001">Figure 1</xref>). Each phylum was sampled at least twice, except for candidate phylum BD Group and the phylum Chlamydiae, which were each observed once.</p><fig id="pcbi-0020092-g001" position="float"><label>Figure 1</label><caption><title>Phylum-Level Delineation of the 16S rRNA Gene Fragments in Alaskan Soil</title><p>Gene fragments (<italic>n</italic> = 1,033) were isolated and sequenced from an Alaskan soil. Candidate phyla WCHB1, OP10, ACE, and BD Group have no sequenced representatives.</p></caption><graphic xlink:href="pcbi.0020092.g001"/></fig><p>We used furthest neighbor clustering to assign sequences to OTUs based on the pairwise genetic distance between sequences. Although controversial [<xref rid="pcbi-0020092-b024" ref-type="bibr">24</xref>], Jukes-Cantor-corrected distances less than 0.03 are considered to correspond to a strain-level delineation, 0.03 to species, 0.05 to genus, 0.15 to class, and 0.30–0.40 to phylum [<xref rid="pcbi-0020092-b025" ref-type="bibr">25</xref>–<xref rid="pcbi-0020092-b028" ref-type="bibr">28</xref>]. Considering potential intragenomic differences between copies of 16S rRNA genes and errors due to sequencing and alignment [<xref rid="pcbi-0020092-b029" ref-type="bibr">29</xref>,<xref rid="pcbi-0020092-b030" ref-type="bibr">30</xref>], the 0.03 cutoff is also a pragmatic choice since it probably represents the most stringent OTU definition that is practically obtainable using 16S rRNA genes. Since the intragenomic distance between 16S rRNA gene sequences is typically less than 0.03, at this distance, replicate 16S rRNA gene sequences from the same genome would form a single OTU.</p><p>To simplify the reporting of our results, OTUs will be designated OTU<sub>x.xx</sub>, where the subscript represents the maximum distance between any two sequences within that OTU (<xref ref-type="fig" rid="pcbi-0020092-g002">Figure 2</xref>). In the Alaskan 16S rRNA gene sequence collection containing 1,033 sequences, we observed 633 OTUs<sub>0.03</sub>. We observed 472 OTUs<sub>0.03</sub> once and 94 OTUs<sub>0.03</sub> twice (<xref ref-type="fig" rid="pcbi-0020092-g002">Figure 2</xref>A). The three most abundant OTUs<sub>0.03</sub> affiliated with members of the phylum Gemmimonas (<italic>n</italic> = 23 sequences in the OTU<sub>0.03</sub> from 19 distinct primary sequences), <named-content content-type="genus-species">Duganella</named-content> sp. (<italic>n</italic> = 17 sequences in the OTU<sub>0.03</sub> from 13 distinct sequences), and <named-content content-type="genus-species">Rhodoferax</named-content> sp. (<italic>n</italic> = 17 sequences in the OTU<sub>0.03</sub> from 15 distinct sequences). These three OTUs<sub>0.03</sub> were not observed in the Minnesotan sequence collection.</p><fig id="pcbi-0020092-g002" position="float"><label>Figure 2</label><caption><title>Rank Abundance Plot of the Alaskan and Minnesotan 16S rRNA Gene Libraries</title><p>Alaskan (<italic>n</italic> = 1,033) (A) and Minnesotan (<italic>n</italic> = 1,633) (B) 16S rRNA gene libraries are plotted and describe the distribution of the 16S rRNA genes among OTUs defined as a group of sequences that are either identical or no more than 3%, 10%, or 20% different.</p></caption><graphic xlink:href="pcbi.0020092.g002"/></fig><p>Since the most abundant OTU<sub>0.03</sub> in the Alaskan 16S rRNA gene library was observed only 23 times, we were unable to obtain meaningful fits of parametric frequency distribution models to the OTU<sub>0.03</sub> frequency distribution [<xref rid="pcbi-0020092-b031" ref-type="bibr">31</xref>]. Attempts to identify parameters that would define simulated communities following either a Pareto or uniform frequency distribution resembling the observed distribution were unsuccessful. The predicted abundance of the most abundant OTUs in the Pareto-distributed communities was too high, and the abundance of the rarest OTUs was too low. We successfully simulated the OTU<sub>0.03</sub> frequency distribution observed in the Alaskan 16S rRNA gene sequence collection by altering the richness and evenness of random frequency distributions using a truncated lognormal frequency distribution (<xref ref-type="table" rid="pcbi-0020092-t001">Table 1</xref>). The relative abundance of each OTU in the simulated communities that followed the truncated lognormal model was</p><table-wrap id="pcbi-0020092-t001" content-type="2col" position="float"><label>Table 1</label><caption><p>Example of Simulation Results for Lognormal and Uniformly Distributed Communities with a Richness of 5,000</p></caption><graphic xlink:href="pcbi.0020092.t001"/></table-wrap><disp-formula id="pcbi-0020092-e001"><graphic xlink:href="pcbi.0020092.e001.jpg" position="anchor" mimetype="image"/></disp-formula><p>where <italic>S</italic> <sub>i</sub> is the i<sup>th</sup> OTU and <italic>N</italic> <sub>i</sub> is the relative abundance of individuals in that OTU. The maximum possible value of i is the total number of OTUs in the community, <italic>S</italic> <sub>T</sub>. <italic>N</italic> <sub>1</sub> is the abundance of the most abundant OTU (<italic>N</italic> <sub>max</sub>), and <italic>N</italic> <sub>T</sub> is the sum of all <italic>N</italic> <sub>i</sub> values.</p><p>Next, we heuristically identified the normal mean (<italic>μ</italic> = 6.000), standard deviation (<italic>σ</italic> = 3.020), and OTU<sub>0.03</sub> richness (<italic>S</italic> <sub>T</sub> = 5,000) for a truncated lognormal distribution in which the distribution of its samples resembled the distribution observed in the Alaskan sequence collection (<xref ref-type="fig" rid="pcbi-0020092-g002">Figure 2</xref>A and <xref ref-type="table" rid="pcbi-0020092-t001">Table 1</xref>). Further confirmation for the plausibility of the simulated community was obtained by comparing the percentage of the total community represented by the most abundant OTU (100 × <italic>N</italic> <sub>max</sub>/<italic>N</italic> <sub>T</sub>) in the simulated community (2.9%) to the value observed from the sequence collection (2.2%). These values are comparable to the range 2.9%–8.3% observed by Dunbar et al. [<xref rid="pcbi-0020092-b017" ref-type="bibr">17</xref>], but are considerably higher than the range 0.1%–1% suggested by Curtis et al. [<xref rid="pcbi-0020092-b013" ref-type="bibr">13</xref>]. We found that the reciprocal of the Simpson's index (1/<italic>D</italic>) for the simulated community was 288, which was similar to the value observed for the sequence collection of 223. The values for 1/<italic>D</italic> are considerably higher than the range 52–107 observed by Dunbar et al. [<xref rid="pcbi-0020092-b017" ref-type="bibr">17</xref>]. To sample every OTU in the Alaskan simulated community twice with 95% confidence would require sequencing 480,000 16S rRNA gene fragments, and to observe 95% of the richness, 71,000 16S rRNA genes would be required (<xref ref-type="fig" rid="pcbi-0020092-g003">Figure 3</xref> and <xref ref-type="table" rid="pcbi-0020092-t001">Table 1</xref>). To obtain an estimate of the true richness using either the ACE or Chao1 nonparametric richness estimator would require sampling 18,000 or 39,000 16S rRNA genes, respectively, which represented sampling 65% and 85% of the true richness (<xref ref-type="fig" rid="pcbi-0020092-g003">Figure 3</xref> and <xref ref-type="table" rid="pcbi-0020092-t001">Table 1</xref>).</p><fig id="pcbi-0020092-g003" position="float"><label>Figure 3</label><caption><title>Estimating the Richness of Taxa in the Simulated Alaskan Soil Community</title><p>(A) Average number of taxa observed and estimated richness for the simulated Alaskan soil community (<italic>S</italic> <sub>T</sub> = 5,000, μ = 6.000, and σ = 3.020) over the course of randomly sampling 480,000 individuals.</p><p>(B–D) The distribution of 16S rRNA sequences obtained from Alaskan soil falls within the 95% CI that would have been obtained for the distribution derived from sampling 1,033 individuals from this simulated community as measured using the observed (B) and estimated—Chao1 (C) and ACE (D)—richness. The thin blue lines in (B), (C), and (D) represent the 95% CIs for each metric using the simulated Alaskan soil community.</p></caption><graphic xlink:href="pcbi.0020092.g003"/></fig><p>Since we were unable to obtain a robust estimate of species richness with our 16S rRNA gene sequence collection without assuming some distribution a priori, we relaxed the OTU definition to obtain a robust nonparametric richness estimate. The OTU<sub>0.20</sub> richness estimate collector's curves began to stabilize late in sampling (<xref ref-type="fig" rid="pcbi-0020092-g003">Figure 3</xref>). Although additional sampling would improve the precision of the OTU<sub>0.20</sub> richness estimate, the Chao1 (188.20, 95% confidence interval [CI] 174–212), ACE (200, CI 181–234), and Jackknife (203, CI 184–222) estimates were similar.</p></sec><sec id="s2b"><title>Comparison of Alaskan and Minnesotan Soils Microbial Communities</title><p>Recently, the microbial community of a Minnesota farm soil was characterized by metagenomic (direct cloning and analysis of DNA from a soil sample) and 16S rRNA gene sequencing analyses [<xref rid="pcbi-0020092-b023" ref-type="bibr">23</xref>]. The authors constructed two separate 16S rRNA gene libraries by using a cell fractionation-based DNA isolation procedure, and sequenced 1,633 overlapping gene fragments from the two libraries [<xref rid="pcbi-0020092-b023" ref-type="bibr">23</xref>,<xref rid="pcbi-0020092-b032" ref-type="bibr">32</xref>]. We reanalyzed their pooled sequence data to determine the richness of the Minnesota farm soil and to determine the degree of OTU membership that was conserved between the Minnesotan and Alaskan soil communities.</p><p>Collector's curves for the number of OTU<sub>0.03</sub> observed and estimated in the Minnesota soil library were flatter than the Alaskan collector's curves (<xref ref-type="fig" rid="pcbi-0020092-g004">Figure 4</xref>). In the Minnesotan collection, the observed OTU<sub>0.03</sub> richness was 767, and we observed 477 OTUs<sub>0.03</sub> once and 128 OTUs<sub>0.03</sub> twice (<xref ref-type="fig" rid="pcbi-0020092-g002">Figure 2</xref>B). The nonparametric richness estimates were 1,647 (Chao1), 1,704 (ACE), and 2,248 (Jackknife); however, each estimate continued to increase with sampling. The three OTUs<sub>0.03</sub> most frequently observed in the Minnesotan sequence collection contained 37, 27, and 26 sequences, and each clustered within the phylum Chloroflexi; no representatives of these OTUs<sub>0.03</sub> were observed in the Alaskan sequence collection.</p><fig id="pcbi-0020092-g004" position="float"><label>Figure 4</label><caption><title>Estimating the Richness of Taxa in the Simulated Minnesotan Soil Community</title><p>(A) Average number of taxa observed and the estimated richness for the simulated Minnesotan soil community (<italic>S</italic> <sub>T</sub> = 2,000, <italic>μ</italic> = 8.000, and <italic>σ</italic> = 3.813) over the course of randomly sampling 100,000 individuals.</p><p>(B–D) The distribution of 16S rRNA gene sequences obtained from the Minnesotan soil falls within the 95% CI that would have been obtained for the distribution derived from sampling 1,633 individuals from this simulated community as measured using the observed (B) and estimated—Chao1 (C) and ACE (D)—richness. The thin blue lines in (B), (C), and (D) represent the 95% CI for each metric using the simulated Minnesotan soil community.</p></caption><graphic xlink:href="pcbi.0020092.g004"/></fig><p>We identified one simulated community with a truncated lognormal distribution that had a richness of 2,000 (<italic>μ</italic> = 8.000, <italic>σ</italic> = 3.813) whose samples resembled the distribution observed in the Minnesotan sequence collection. The percentage of the clones represented by the most abundant OTU<sub>0.03</sub> was 2.3%, and it was 3.9% in the simulated community. The simulated community had a higher 1/<italic>D</italic> value than that observed from the sequence data (311 versus 237). The Minnesotan simulated community had a lower richness and more uniform evenness than we observed for the Alaskan simulated community. If the Minnesotan simulated community is a true reflection of the OTU<sub>0.03</sub> distribution, then we are 95% confident that sequencing of 90,000 16S rRNA genes would result in observing every OTU<sub>0.03</sub> at least twice. Sequencing 16,000 16S rRNA gene fragments would allow us to observe 95% of the true richness. To obtain a nonparametric estimate of richness through the ACE and Chao1 estimators, 2,000 and 5,500 16S rRNA gene sequences, respectively, would need to be sequenced for the estimate's CI to include 2,000; the CI for the Jackknife estimator already includes 2,000. The sequencing of 2,000 and 5,500 16S rRNA genes would result in observing 43.9% and 72.5% of the true richness, respectively.</p><p>It is difficult to determine whether the difference in estimated richness between the Alaskan and Minnesotan simulated communities was due to ecological differences or differences in DNA extraction methods [<xref rid="pcbi-0020092-b033" ref-type="bibr">33</xref>] or both. The collector's curve for the estimated fraction of the Minnesotan library shared with the Alaskan library, 0.18 (standard error [SE] = 0.07), indicated that this value is close to the true value and that the fraction of the Alaskan library shared with the Minnesotan library, 0.17 (SE = 0.06), continued to increase with additional sampling (<xref ref-type="fig" rid="pcbi-0020092-g005">Figure 5</xref>A). Our observation that 18% of the sequences in the Minnesotan library belonged to OTUs<sub>0.03</sub> shared with the Alaskan collection indicates either that a large fraction of these OTUs<sub>0.03</sub> are endemic to different soils or that the different DNA extraction procedures preferentially lysed a subset of the OTU<sub>0.03</sub> membership, or both.</p><fig id="pcbi-0020092-g005" position="float"><label>Figure 5</label><caption><title>Similarity of Alaskan and Minnesotan Soil Microbial Communities</title><p>Collector's curves describing the effect of sampling on the estimated fraction of sequences from the Minnesotan (red lines) and Alaskan (blue lines) libraries belonging to shared OTUs<sub>0.03</sub> and OTUs<sub>0.20</sub>.</p></caption><graphic xlink:href="pcbi.0020092.g005"/></fig><p>Similar to our analysis of the Alaskan 16S rRNA library, when we relaxed the OTU definition to analyze the OTU<sub>0.20</sub> richness of the Minnesotan 16S rRNA membership, we observed richness estimates that were not sensitive to further sampling. The terminal Chao1 (165, CI 151–197), ACE (169, CI 156–196), and Jackknife (174, CI 158–190) estimates were similar; this is approximately 85% of the OTU<sub>0.20</sub> richness observed in the Alaskan 16S rRNA library. The fraction of sequences from the Minnesotan library that belonged to OTUs<sub>0.20</sub> shared between the two libraries was 0.86 (SE = 0.10) and the fraction of sequences from the Alaskan sequences that belonged to OTUs<sub>0.20</sub> shared between the two libraries was 0.88 (SE = 0.07) (<xref ref-type="fig" rid="pcbi-0020092-g005">Figure 5</xref>B). At this point in sampling, it was not possible to conclude with statistical confidence that the OTU<sub>0.20</sub> memberships were significantly different, since both CIs included 1.00; however, we expect further sampling to make the estimates more precise.</p></sec></sec><sec id="s3"><title>Discussion</title><p>In the 20th century, the view of soil microbial ecology shifted from being described by Selman Waksman as a “clear picture” [<xref rid="pcbi-0020092-b034" ref-type="bibr">34</xref>] to E. O. Wilson's pronouncement that its diversity is “beyond practical calculation” [<xref rid="pcbi-0020092-b035" ref-type="bibr">35</xref>]. We have shown that neither view is wholly correct, but that a confident estimate of bacterial richness is attainable using a set of parameters that have a reasonable biological basis. We have shown that it is possible to obtain an OTU<sub>0.03</sub> richness estimate for soil for considerably less effort than is required to shotgun sequence a bacterial genome (assuming ~100,000 sequence reads per genome and one to five reads for each of 17,000 16S rRNA gene fragments). Determining the richness of specific phylogenetic groups using lineage-specific PCR primers would further reduce the required effort.</p><p>Our analysis can also be applied to guide the design of functional and sequence-based metagenomics projects [<xref rid="pcbi-0020092-b036" ref-type="bibr">36</xref>]. Tringe et al. [<xref rid="pcbi-0020092-b023" ref-type="bibr">23</xref>] estimated that more than 2 × 10<sup>9</sup> bp of sequence from 3 × 10<sup>6</sup> sequence reads would be necessary to obtain 8-fold sequence coverage of the most abundant species in their soil sample assuming a genome size of 6 Mbp. To sequence 8-fold coverage of the most abundant OTU<sub>0.03</sub> from the simulated Alaskan soil community, approximately 450 genome equivalents, or 3 × 10<sup>9</sup> bp, would need to be sequenced from the Alaskan soil. To sequence 8-fold coverage of the ten most abundant OTUs<sub>0.03</sub> from the simulated Alaskan soil community, approximately 1,600 genome equivalents, or 10<sup>10</sup> bp, would need to be sequenced. Although this amount of DNA may be beyond our current sequencing capacity, the 10<sup>10</sup> bp is approximately the content of a 275,000-clone fosmid library. Such a library could be easily constructed and would be useful for functional metagenomic approaches. Although not currently feasible, sequencing 8-fold coverage of every OTU<sub>0.03</sub> in the Alaskan soil metagenome would require sequencing 950,000 genome equivalents or 6 × 10<sup>12</sup> bp of DNA. Although PCR bias may affect the true community distribution, these values are a helpful guide when designing metagenomics-based experiments. For some groups of organisms, the 3% cutoff between 16S rRNA gene sequences has been found to correlate with 70% similarity between genome sequences; therefore, it is unclear how many contigs would assemble for the predicted level of sequencing effort given the substantial intragenomic variation that may exist between members of the same OTU<sub>0.03</sub>.</p><p>Estimating richness does not provide the identity of each bacterial type; in the Alaskan soil we studied, identifying every one of the 5,000 different types of bacteria would require sampling more than 480,000 sequences. Furthermore, our analysis assumes an operational species definition of a group of 16S rRNA sequences that are no more than 3% different from one another. Among the members of a single OTU, there is undoubtedly considerable phenotypic and genomic diversity that is not reflected by 16S rRNA sequences [<xref rid="pcbi-0020092-b024" ref-type="bibr">24</xref>]. Our attempt to perform a census of the number of bacteria in a gram of soil provides a guidepost from which we can begin to assess the effects of environmental perturbations on community composition, diversity, evenness, and richness. Moreover, an accurate census would quantify the part of the microbial community that is not accounted for in the current models of community structure and function. In the Alaskan sequence collection, two sequences belonging to the sparsely sampled candidate phylum ACE were found only after sampling 832 sequences. We suspect that members of many poorly sampled candidate phyla are rare members in microbial communities [<xref rid="pcbi-0020092-b037" ref-type="bibr">37</xref>], but may play significant functional roles in the microbial community. Although a reliable estimate of richness will inform the development of a conceptual framework for describing the functional biology of the soil microbial community, we will not know the texture and composition of that richness until we have exhaustively sampled and identified every member of the community.</p></sec><sec id="s4"><title>Materials and Methods</title><sec id="s4a"><title>Clone library construction, sequencing, and analysis.</title><p>We obtained a soil core from the Bonanza Creek Long-Term Ecological Research site approximately 30 km southwest of Fairbanks, Alaska, United States (64° 48′ N, 147° 52′ W) on the site designated BP-1 on an island in the Tanana River [<xref rid="pcbi-0020092-b038" ref-type="bibr">38</xref>]. The L1A 16S rRNA gene library was constructed using a single 0.5-g sample of soil. The Bio101 soil DNA kit (Bio101, Irvine, California, United States) was used to extract and partially purify genomic DNA and the sample was further purified using a silica matrix (ExpressMatrix; Bio101) until it was suitable for PCR amplification.</p><p>16S rRNA genes were amplified in a single reaction by PCR using primers 27f (AGRGTTTGATYMTGGCTCAG) and 1492r (GGYTACCTTGTTACGACTT) and the products were purified by gel extraction (Qiaex II; Qiagen, Valencia, California, United States). Purified PCR products were ligated into the pGEM-T TA cloning vector as described by the manufacturer (Promega, Madison, Wisconsin, United States) and electroporated into <named-content content-type="genus-species">E. coli</named-content> (DH5α). Positive transformants were inoculated overnight into LB with ampicillin (100 μg/ml) and the culture was used as template for PCR using the universal M13f and M13r vector primers. These PCR products were purified using AmpPure (Agencourt Bioscience, Beverly, Massachusetts, United States) and sequenced using the 27f and 787r (CTACCRGGGTATCTAAT) primers. If the 787r primer did not produce quality sequence, we used either the M13f or the M13r primer for sequencing. Sequencing reactions were performed using BigDye version 3.1 (Applied Biosystems, Foster City, California, United States) and were analyzed at the University of Wisconsin-Madison biotechnology center. All clones had 2-fold sequencing coverage for the first approximately 700 bp of the 16S rRNA gene.</p><p>Sequence contigs were constructed using STADEN [<xref rid="pcbi-0020092-b039" ref-type="bibr">39</xref>] and aligned using ARB [<xref rid="pcbi-0020092-b040" ref-type="bibr">40</xref>] with a reference database of more than 16,000 sequences longer than 1 kb. Putative chimeric sequences were identified using Bellerophon [<xref rid="pcbi-0020092-b041" ref-type="bibr">41</xref>] and were further screened using CHIMERA_CHECK [<xref rid="pcbi-0020092-b042" ref-type="bibr">42</xref>], partial treeing, and comparing the sequence alignment to predicted secondary structure to detect changes in helical base pairing and nucleotide signatures [<xref rid="pcbi-0020092-b043" ref-type="bibr">43</xref>]. Phylogenetic placement of the 1,033 sequences was determined by identifying the phylum to which each sequence showed affinity after adding sequences to the database tree using the parsimony algorithm implemented in ARB with a 50% consensus mask.</p><p>We also obtained (from Susannah Green Tringe) two 16S rRNA gene sequence collections (<italic>N</italic> <sub>AKYG</sub> = 875 sequences; <italic>N</italic> <sub>AKYH</sub> = 758 sequences) constructed using a single 0.5-g sample of Minnesotan (Waseca County, Minnesota, United States [<xref rid="pcbi-0020092-b023" ref-type="bibr">23</xref>]) farm soil. The original soil genomic DNA was obtained by cell fractionation followed by enzymatic and chemical extraction of the DNA [<xref rid="pcbi-0020092-b023" ref-type="bibr">23</xref>,<xref rid="pcbi-0020092-b032" ref-type="bibr">32</xref>]. Since our preliminary analysis using a nonparametric estimator of the fraction of shared OTUs [<xref rid="pcbi-0020092-b044" ref-type="bibr">44</xref>] showed that the two Minnesota soil libraries harbored more than 68% of each others' OTU<sub>0.03</sub> membership, and they were made from the same soil sample but different PCR reactions, we pooled the 1,633 sequences into a single dataset. For direct comparison, the Minnesotan and Alaskan sequence collections were realigned using the NAST aligner [<xref rid="pcbi-0020092-b045" ref-type="bibr">45</xref>] at the greengenes Web site (<ext-link ext-link-type="uri" xlink:href="http://greengenes.lbl.gov">http://greengenes.lbl.gov</ext-link>), and the nucleotide sites between positions 150 and 700 (<named-content content-type="genus-species">E. coli</named-content> numbering) were used in subsequent analyses.</p></sec><sec id="s4b"><title>Community analyses.</title><p>To describe the community structure of each soil we used DOTUR's implementation of the furthest neighbor algorithm [<xref rid="pcbi-0020092-b018" ref-type="bibr">18</xref>] to assign sequences to OTUs after exporting a Jukes-Cantor corrected distance matrix constructed in ARB using unmasked sequences. Output files from DOTUR were used to calculate collector's curves for the nonparametric estimators of the fraction of sequences from one library that affiliated with the OTUs shared between the libraries [<xref rid="pcbi-0020092-b044" ref-type="bibr">44</xref>].</p></sec><sec id="s4c"><title>Model community analysis.</title><p>Using a truncated lognormal, Pareto, or uniform distribution, we were able to construct model communities in which the probability of drawing an individual species followed a defined distribution. To identify the most appropriate truncated lognormal distribution that described the observed data, we first selected reasonable values for <italic>μ</italic> between 6.000 and 12.00 and <italic>S</italic> <sub>T</sub> between 1,000 and 10,000. Next, using Equation 1, we identified values of <italic>σ</italic> that would yield <italic>N</italic> <sub>T</sub>/<italic>N</italic> <sub>max</sub> values of 10, 35, 40, 45, 50, 100, and 1,000. <italic>N</italic> <sub>T</sub>/<italic>N</italic> <sub>max</sub> is the reciprocal of the probability observed at the distribution's mode, where <italic>N</italic> <sub>T</sub> represents the total number of individuals in a community and <italic>N</italic> <sub>max</sub> represents the abundance of the most abundant member in the community. The values of <italic>N</italic> <sub>T</sub>/<italic>N</italic> <sub>max</sub> shown in <xref ref-type="table" rid="pcbi-0020092-t001">Table 1</xref> were selected because they fell within the range suggested by Curtis et al. [<xref rid="pcbi-0020092-b013" ref-type="bibr">13</xref>] for microbial communities and because they resembled the frequency data observed from the Alaskan soil collection. For a given value of <italic>N</italic> <sub>T</sub>/<italic>N</italic> <sub>max</sub>, increasing <italic>μ</italic> increased the value of the reciprocal of the Simpson's index (1/<italic>D</italic>). 1/<italic>D</italic> represents the number of uniformly abundant OTUs needed to observe the same level of diversity found in the community. Using these parameters, we drew random values from the desired lognormal distribution by first drawing a random normal variable with mean <italic>μ</italic> and standard deviation <italic>σ</italic>. Random lognormal variables were then obtained by determining the integer value of <italic>e</italic> <sup>X</sup>, where X is the value of the random normal variable. Values larger than <italic>S</italic> <sub>T</sub> were discarded, resulting in random variables drawn from a truncated lognormal distribution. Random variables drawn from either Pareto or uniformly distributed communities were done in an analogous manner.</p><p>As random values were generated, we constructed collector's curves for the observed richness and the full bias-corrected Chao1 [<xref rid="pcbi-0020092-b019" ref-type="bibr">19</xref>], ACE [<xref rid="pcbi-0020092-b020" ref-type="bibr">20</xref>], and interpolated Jackknife [<xref rid="pcbi-0020092-b021" ref-type="bibr">21</xref>] nonparametric estimators. The heuristic search did not include the Jackknife estimator because the estimates were highly variable and uninformative. As measures of diversity, we determined 1/<italic>D</italic> and the longest string where duplicate members of the same taxa were not observed. We determined the CI for each metric as a function of sampling effort by constructing 1,000 model communities for each set of model parameters. The sampling of the truncated lognormal distributions and parameter calculation was performed using a C++ computer program that we wrote. If the collector's curve for the sampling of the Alaskan or Minnesotan 16S rRNA sequence collections crossed the CI for any parameter, the simulated community was rejected.</p></sec></sec><sec sec-type="supplementary-material" id="s5"><title>Supporting Information</title><sec id="s5a"><title>Accession Numbers</title><p>The GenBank (<ext-link ext-link-type="uri" xlink:href="http://www.ncbi.nlm.nih.gov">http://www.ncbi.nlm.nih.gov</ext-link>) accession numbers of the Alaskan 16S rRNA gene sequences are AY988608 through AY989640. All sequence alignments are available from the authors' Web site (<ext-link ext-link-type="uri" xlink:href="http://www.plantpath.wisc.edu/fac/joh/soil_census_data.html">http://www.plantpath.wisc.edu/fac/joh/soil_census_data.html</ext-link>).</p></sec></sec>
Complex Parameter Landscape for a Complex Neuron Model	<p>The electrical activity of a neuron is strongly dependent on the ionic channels present in its membrane. Modifying the maximal conductances from these channels can have a dramatic impact on neuron behavior. But the effect of such modifications can also be cancelled out by compensatory mechanisms among different channels. We used an evolution strategy with a fitness function based on phase-plane analysis to obtain 20 very different computational models of the cerebellar Purkinje cell. All these models produced very similar outputs to current injections, including tiny details of the complex firing pattern. These models were not completely isolated in the parameter space, but neither did they belong to a large continuum of good models that would exist if weak compensations between channels were sufficient. The parameter landscape of good models can best be described as a set of loosely connected hyperplanes. Our method is efficient in finding good models in this complex landscape. Unraveling the landscape is an important step towards the understanding of functional homeostasis of neurons.</p>	<contrib contrib-type="author"><name><surname>Achard</surname><given-names>Pablo</given-names></name><xref ref-type="aff" rid="aff1"/></contrib><contrib contrib-type="author"><name><surname>De Schutter</surname><given-names>Erik</given-names></name><xref ref-type="corresp" rid="cor1">*</xref><xref ref-type="aff" rid="aff1"/></contrib>	PLoS Computational Biology	<sec id="s1"><title>Introduction</title><p>Neuronal electrical activity is governed by ion fluxes. Whereas intracellular currents are primarily determined by the cell morphology and its electrical passive properties, the major components of the electrical activity of a neuron are transmembrane currents driven by gated ionic channels present all over its surface. Small changes in the channel conductances of a neuron can lead to drastically different activities. Nevertheless, robustness of electrical activity to channel alterations, also called functional homeostasis, has recently been observed in several experiments. For example, by overexpressing the <italic>ShaI</italic> gene into lobster stomatogastric ganglion neurons, MacLean et al. [<xref rid="pcbi-0020094-b001" ref-type="bibr">1</xref>] nearly doubled the expression of the transient potassium current (I<sub>A</sub>). This increase was spontaneously compensated by an increase of the hyperpolarization-activated current (I<sub>h</sub>) and the activity of the neurons remained almost unaffected. Swensen and Bean [<xref rid="pcbi-0020094-b002" ref-type="bibr">2</xref>] have shown that similar firing patterns can be obtained in vitro from mouse Purkinje cells (PCs) with dissimilar combinations of sodium and calcium currents. The robustness of PC burst firing was also observed in mice where the expression of the sodium channel Na<sub>v</sub>1.6 was genetically silenced. In this case, homeostasis was maintained by an increase of calcium currents. In a recent set of experiments, Schulz et al. [<xref rid="pcbi-0020094-b003" ref-type="bibr">3</xref>] measured potassium currents and their mRNA expression in stomatogastric crab lateral pyloric neurons and found two- to four-fold interanimal variability. They also demonstrated clear correlations in K<sup>+</sup> channel expression between coupled pyloric dilatator neurons of a single crab, while a larger variation of this expression was found between crabs. Computational models made by Prinz et al. [<xref rid="pcbi-0020094-b004" ref-type="bibr">4</xref>] and Goldman et al. [<xref rid="pcbi-0020094-b005" ref-type="bibr">5</xref>] have demonstrated that identical network or neuron activities can be obtained from disparate modeling parameters. However, these modeling studies were limited in the number of free parameters used and in the complexity and details of the measured electrical activity. This raises the question of whether it is also possible to reproduce in full detail much more complex neuronal electrical activity with models using dissimilar sets of ionic currents.</p><p>The dendritic arborization and electrical activity of PCs are among the most complex of the brain. In this study we used the electrical activity produced by an existing model of PC [<xref rid="pcbi-0020094-b006" ref-type="bibr">6</xref>] as the data to be reproduced by newly generated models. An evolution strategy (ES) algorithm was used to tune 24 different channel densities in these models until they reproduced the electrical activity with enough detail. We then retained 20 sets of conductances that are very different from each other, which reproduce the voltage traces of the original model in a very detailed fashion. An analysis of the parameter landscape was used to understand this diversity of parameters. Finally, a comparison with previous results or methods will demonstrate the exceptional quality of the models we found.</p></sec><sec id="s2"><title>Results</title><p>The original PC model, which we consider in this study as the data to be reproduced, consists of 1,600 compartments and exhibits different modes of activity depending upon the amplitude of the injected current: it can be silent, spiking, bursting with short bursts of half a dozen of spikes followed by a short inter-burst interval, or bursting with long bursts consisting of around 20 spikes and inter-burst intervals larger than 0.2 s (see <xref ref-type="fig" rid="pcbi-0020094-g001">Figure 1</xref>). Our generated models should reproduce not only this general behavior but also the details of electrical activity.</p><fig id="pcbi-0020094-g001" position="float"><label>Figure 1</label><caption><title>The Search Method</title><p>(A) Example of the four firing modes of the PC: silent (top left), tonic (bottom left), small bursts (top right), and long bursts (bottom right) obtained with respectively 0, 0.5, 2, and 3 nA of current injected in the soma.</p><p>(B) The (V,dV/dt) matrix obtained for data when a current of 0.5 nA is injected in the soma. The red points correspond to the first 0.1 s after current injection (transitory period), while the blue ones represent 1.1 s of data recorded 0.9 s after the beginning of current injection (stable period). The black arrow shows the direction followed by successive points in time during a spike.</p><p>(C) The (V,dV/dt) matrix obtained for data when a current of 3 nA is injected in the soma. The black points represent 2.1 s of data recorded 0.9 s after the beginning of current injection. The red lines link successive points.</p><p>(D) Time evolution of the mean fitness of the population (full lines). The nine runs, shown as different colors, have very similar evolution, and were stopped after 415 generations. The time evolution of the fitness of the best individual of runs 1, 3, and 7 is shown as dashed lines.</p><p>(E) Fitness of all individuals of each population when the runs were stopped. Open points represent individuals selected for the rest of the analysis. The full line corresponds to the fitness upper limit for selecting individuals.</p></caption><graphic xlink:href="pcbi.0020094.g001"/></fig><sec id="s2a"><title>Parameter Search</title><p>The cell model possesses ten different voltage or calcium-gated channels. It is subdivided in four different morphological zones with each zone exhibiting constant channel densities (see <xref ref-type="sec" rid="s5">Materials and Methods</xref>). In total, there are 24 conductance parameters in this model, all of which were supposed to be unknown at the beginning of the parameter search.</p><p>Among the wide variety of optimization algorithms available, we have chosen ES. There are good theoretical and practical reasons to prefer ES to other algorithms when the parameters to tune are real numbers and their number is high [<xref rid="pcbi-0020094-b007" ref-type="bibr">7</xref>,<xref rid="pcbi-0020094-b008" ref-type="bibr">8</xref>]. As in other evolutionary algorithms, each set of parameters is called an “individual.” Several individuals form a “population” and populations are evolving, “generation” after “generation,” through “mutation” and “cross-over” of their individuals and “selection” of the best ones. After fine tuning of this optimization algorithm (see <xref ref-type="supplementary-material" rid="pcbi-0020094-sd001">Protocol S1</xref>), nine runs were performed with different random number generator seeds. Each run was stopped after the evaluation of a fixed number of individuals (~8,000).</p><p>To assess the quality of an individual and to allow selection while the ES algorithm runs, a single real number, called “fitness”, measures its distance to the data. Several fitness functions have been used with electrophysiological data [<xref rid="pcbi-0020094-b009" ref-type="bibr">9</xref>–<xref rid="pcbi-0020094-b011" ref-type="bibr">11</xref>]. Our distance measurement is based on LeMasson's phase plane analysis [<xref rid="pcbi-0020094-b011" ref-type="bibr">11</xref>], where the time evolution of the cell voltage, V(t), is summarized in a matrix in which V is associated with dV/dt for every time step (see <xref ref-type="fig" rid="pcbi-0020094-g001">Figure 1</xref>B and C). The fitness function that we have used takes into account different amplitudes of injected current, different periods in time after current injection, and different recording sites (see <xref ref-type="sec" rid="s4">Materials and Methods</xref>, and <xref ref-type="supplementary-material" rid="pcbi-0020094-sd002">Protocol S2</xref>). The fitness measure has arbitrary units and best models have lowest “fitness.” The evolution of the population fitness with each generation is shown in <xref ref-type="fig" rid="pcbi-0020094-g001">Figure 1</xref>D.</p></sec><sec id="s2b"><title>Quality of the 20 Best Models</title><p>The fitnesses of the 57 individuals in each of the final nine populations are shown in <xref ref-type="fig" rid="pcbi-0020094-g001">Figure 1</xref>E. To check the quality of the models obtained, we compared the best models with the data by injecting current at additional amplitudes (see <xref ref-type="supplementary-material" rid="pcbi-0020094-sg001">Figure S1</xref>). After visual examination of the best solutions, we decided to select, for the rest of the analysis, all individuals with a good fitness that exhibit the four modes of activity with somatic current injection (see <xref ref-type="sec" rid="s4">Materials and Methods</xref>). The 20 good individuals that form our final selection are drawn as open circles in <xref ref-type="fig" rid="pcbi-0020094-g001">Figure 1</xref>E. Their fitness values are in the range 2.58 to 3.45.</p><p>To demonstrate the quality of this selection, the electrical activity in the soma, as well as in one dendrite, of the models with lowest and highest fitness value of the selection are shown in <xref ref-type="fig" rid="pcbi-0020094-g002">Figure 2</xref> and <xref ref-type="supplementary-material" rid="pcbi-0020094-sg001">Figure S1</xref>. All the obtained models very precisely match the data. The main macroscopic difference is due to transitions between the four modes (silence, spikes, small bursts, and long bursts) that sometimes occur at slightly different current amplitudes. The shape of the spikes, the dendritic activity, the waveform of the bursts, the activity between bursts, etc., are all reproduced accurately by every selected model (<xref ref-type="supplementary-material" rid="pcbi-0020094-sd002">Protocol S2</xref> and <xref ref-type="supplementary-material" rid="pcbi-0020094-sg002">Figures S2</xref>–<xref ref-type="supplementary-material" rid="pcbi-0020094-sg004">S4</xref>). Interestingly, the models also reproduce burst-to-burst variations that are present in the data. In <xref ref-type="fig" rid="pcbi-0020094-g002">Figure 2</xref>C, for example, the envelope of the bursts, as well as the length of their pre-spiking oscillations, vary both in a similar way in the data and models.</p><fig id="pcbi-0020094-g002" position="float"><label>Figure 2</label><caption><title>Comparison of the Models with the Data for Current Clamp</title><p>(A–C) The membrane potential of the soma is shown for the data (red traces), best (blue), and worst (green) model of our selection for different somatic current amplitudes: 0.5 (A), 2 (B), and 3 nA (C).</p><p>(D) Same as (B) for dendritic membrane potential.</p></caption><graphic xlink:href="pcbi.0020094.g002"/></fig><p>In addition, we have looked at features for which the models were not tuned, i.e. their response to synaptic input [<xref rid="pcbi-0020094-b012" ref-type="bibr">12</xref>]. The somatic and dendritic waveforms of complex spikes generated by climbing fiber excitation are very well reproduced by all models (see <xref ref-type="fig" rid="pcbi-0020094-g003">Figure 3</xref>). The firing rate evoked by combined random excitatory and inhibitory input is also reproduced quite well by the best model (<xref ref-type="fig" rid="pcbi-0020094-g003">Figure 3</xref>B). The worst selected model has a sharper transition between the silent mode and the high frequency firing one, but it occurs at the same balance between excitation and inhibition. The active propagation of excitatory post-synaptic potentials (EPSP) [<xref rid="pcbi-0020094-b013" ref-type="bibr">13</xref>] also remains in all models: in <xref ref-type="fig" rid="pcbi-0020094-g003">Figure 3</xref>C, EPSPs triggered at four different synapse locations are recorded at the soma and compared for the data and models.</p><fig id="pcbi-0020094-g003" position="float"><label>Figure 3</label><caption><title>Comparison of the Models with the Data for Synaptic Responses</title><p>Data (red lines), best (blue lines), and worst (green lines) model of our selection are compared for different synaptic responses.</p><p>(A) Complex spike in the soma (left), main dendrite (middle), and smooth dendrite (right) after activation of the climbing fiber at time 0.2 s.</p><p>(B) Simple spike frequency response to different levels of excitation and inhibition.</p><p>(C) EPSPs generated by a synchronous parallel fiber input plus an asynchronous background excitation and inhibition. EPSPs are generated in four different branchlets and recorded at the soma, which is passive. The traces show the average of 40 EPSPs obtained with different random number generator seeds.</p></caption><graphic xlink:href="pcbi.0020094.g003"/></fig><p>All in all, these results show an exceptionally good agreement between data and models (see also <xref ref-type="supplementary-material" rid="pcbi-0020094-sd002">Protocol S2</xref>).</p></sec><sec id="s2c"><title>Variability of the Conductances</title><p>We evaluated, for each of the 24 parameters, how diverse the 20 best solutions were. As an example, the value of the maximum conductance of the persistent sodium current (g<sub>NaPs</sub>, see <xref ref-type="table" rid="pcbi-0020094-t001">Table 1</xref> for the list of current abbreviations) of each individual is plotted against its fitness value in <xref ref-type="fig" rid="pcbi-0020094-g004">Figure 4</xref>. The mean value, standard deviation (sdv), and total range of the 20 selected individuals, normalized to the data, are shown for each conductance density in <xref ref-type="fig" rid="pcbi-0020094-g004">Figure 4</xref>B and for summed total conductances in <xref ref-type="fig" rid="pcbi-0020094-g004">Figure 4</xref>C. Overall, these results exhibit large differences when compared to the data, showing that the precise reproduction of electrical activity we have shown in <xref ref-type="fig" rid="pcbi-0020094-g002">Figure 2</xref> can be obtained from a wide range of parameter values. On one extreme, parameters like g<sub>CaTm</sub>, g<sub>KAm</sub>, g<sub>KMd</sub>, or g<sub>Khs</sub> took values in the whole allowed range, while parameters like g<sub>NaFs</sub>, g<sub>CaPd</sub>, g<sub>Kdrs,</sub> or g<sub>Kdrm</sub> needed to be more constrained to replicate the desired activity.</p><table-wrap id="pcbi-0020094-t001" content-type="1col" position="float"><label>Table 1</label><caption><p>Parameter Bounds and Data Values</p></caption><graphic xlink:href="pcbi.0020094.t001"/></table-wrap><fig id="pcbi-0020094-g004" position="float"><label>Figure 4</label><caption><title>Conductance Spread</title><p>(A) The fitness of each individual is plotted against its g<sub>NaPs</sub> value. Points of the same color belong to the same population (see <xref ref-type="fig" rid="pcbi-0020094-g001">Figure 1</xref>E). Blue lines give the range in which channel densities were allowed to vary while the black line gives the value of the data. On the right, the red marker shows the mean ± sdv of the 20 values.</p><p>(B) For each conductance, the mean value of the 20 selected individuals is shown, normalized to the data. The full red bars delimit the whole range covered by the 20 models while the horizontal red lines give sdv. Blue bars show the range of variation allowed during the search. Green lines are linear fit, supposing regular spacing on the abscissa.</p><p>(C) Same as (B) for total conductances, obtained by summing conductance densities of the same type, weighted by the surface area of the membrane regions where they apply.</p></caption><graphic xlink:href="pcbi.0020094.g004"/></fig><p>The spatial distribution of the conductance densities is also instructive. The decomposition of the conductances that have three dendritic components shows a clear pattern: for all of them, except g<sub>KM</sub>, the mean values of the 20 selected individuals are below the data in the distal spiny dendrites, while they are above it in the proximal smooth and main dendrites. This shows that using equal densities in the smooth and spiny dendrite, as was done in the original model, is not the easiest way to obtain the desired output.</p><p>What can explain this large diversity of good conductance density values? We will explore four possible hypotheses. First, the model could have too many dimensions; some parameters have a very low influence on the neuron's electrical behavior [<xref rid="pcbi-0020094-b014" ref-type="bibr">14</xref>] and therefore they can take almost any value without prejudice. Second, all the solutions we have found could belong to a continuous region of the parameter space where the models reproduce well the data; there is a large region around the data for which changes of the parameters have small effects. Third, there could be strong compensatory mechanisms between some ionic currents, exactly as demonstrated experimentally by MacLean et al. [<xref rid="pcbi-0020094-b001" ref-type="bibr">1</xref>] or Swensen and Bean [<xref rid="pcbi-0020094-b002" ref-type="bibr">2</xref>]. In this case, if two currents compensate, hyperplanes of good solutions will exist in the parameter space; if more complicated correlations exist between several currents, hyperspaces of good models will be present. Fourth and opposite to the previous hypothesis, all the solutions we have found could belong to small regions of the parameter space that are isolated from each other. This would imply discontinuities such as threshold mechanisms. To disentangle these hypotheses a better knowledge of the parameter landscape is required.</p></sec><sec id="s2d"><title>Analysis of the Parameter Landscape</title><p>We have tested how sensitive the data is to variations of each conductance separately. To do so, all the densities were set to the data value but one for which 500 random points were taken in the whole allowed interval (see <xref ref-type="table" rid="pcbi-0020094-t001">Table 1</xref>). The fitness values of these 24 × 500 points are shown in <xref ref-type="fig" rid="pcbi-0020094-g005">Figure 5</xref>A. Seven of the parameters (g<sub>KAs</sub>, g<sub>KAm</sub>, g<sub>Kdrm</sub>, g<sub>KMs</sub>, g<sub>kMm</sub>, g<sub>KMt</sub>, g<sub>K2m</sub>) had a very small effect on the fitness of the models over the whole tested range. We can therefore suppose that the diversity of their values is explained by the low effect of these parameters. However, a model with these seven parameters set to zero and all other parameters equal to the data value had a poor fitness of 4.25. So, if their individual effect was small, their collective effect was still quite considerable. Additionally, since the allowed values for g<sub>KAs</sub>, g<sub>KMt</sub>, and especially g<sub>Kdrm</sub> are not fully covered by our 20 models (<xref ref-type="fig" rid="pcbi-0020094-g004">Figure 4</xref>B), their influence was probably stronger in other locations of the parameter space.</p><fig id="pcbi-0020094-g005" position="float"><label>Figure 5</label><caption><title>Points of the Parameter Space around the Data and the Selected Individuals</title><p>(A) Fitness of 24 × 500 points for which all the parameters are equal to that of the data but one, labeled on the abscissa, and varied randomly in the full allowed range (see <xref ref-type="table" rid="pcbi-0020094-t001">Table 1</xref>). The mean ± sdv range covered by the 20 selected individuals is shown in blue for each conductance density. The exact data value is never randomly selected so none of the distributions reached the perfect fitness value of 0.</p><p>(B) For each of the 20 selected individuals (blue dots), the fitness of the 48 individuals obtained by changing its parameters by ±1% (red) or ±5% (green). Only one parameter is changed at a time.</p></caption><graphic xlink:href="pcbi.0020094.g005"/></fig><p>These conclusions limit the scope of the first hypothesis, but <xref ref-type="fig" rid="pcbi-0020094-g005">Figure 5</xref>A is also in disfavor of the second hypothesis. As one can see, fitnesses observed in the range of parameters values corresponding to the mean ± sdv of the good solutions (blue points) are often bad. This indicates that the individuals we found can not belong to a large continuum of good solutions around the data.</p><p>Testing the third and fourth hypotheses is a bit harder, as it is impossible to try every point of the parameter space, even with a small sample of points in each dimension. Compensatory effects between two conductance densities should give, in the simplest case, linear correlations between them. Therefore we have calculated, for the 276 possible pairs of parameters, the Pearson's correlation coefficient (see <xref ref-type="sec" rid="s4">Materials and Methods</xref> and <xref ref-type="supplementary-material" rid="pcbi-0020094-sg005">Figure S5</xref>). Only five pairs of conductances had a probability of correlation (<italic>p</italic> < 0.01): (g<sub>K2t</sub>, g<sub>K2d</sub>), (g<sub>Kdrm</sub>, g<sub>K2d</sub>), (g<sub>Kdrm</sub>, g<sub>CaPm</sub>), (g<sub>Kdrm</sub>, g<sub>K2t</sub>), and (g<sub>CaPt</sub>, g<sub>CaPd</sub>), with respective correlation coefficients of r = −0.78, −0.63, −0.59, 0.57, and −0.58. Four of these pairs involve conductance densities that had a limited range around the data (g<sub>Kdrm</sub> and g<sub>CaPd</sub>). Two pairs showed anti-correlation between conductances of the same channel in smooth and spiny dendrites (distinguished by the last letter <italic>t</italic> or <italic>d</italic>). But this was not true for other similar pairs, (g<sub>CaTt</sub>, g<sub>CaTd</sub>), (g<sub>KMt</sub>, g<sub>KMd</sub>) and (g<sub>KCt</sub>, g<sub>KCd</sub>), which were not correlated (<italic>p</italic> > 0.05). So, linear correlations could explain the large range of values found for a few parameters, but they clearly were not a general explanation for the observed variability.</p><p>The presence of the same type of ionic channels on different regions of the PC suggests that compensatory mechanisms could simply be a balance between the same currents in different locations. This would then result in a reduced dispersion of the ten total conductances of the 20 models (<xref ref-type="fig" rid="pcbi-0020094-g004">Figure 4</xref>C), which is indeed less pronounced than the spread of the 24 channel densities (<xref ref-type="fig" rid="pcbi-0020094-g004">Figure 4</xref>B). Nevertheless this dispersion is still significant: the total conductances are far from being constant. Linear correlations between pairs of total conductances were absent for 44 combinations, only the (g<sub>CaT</sub>, g<sub>CaP</sub>) pair was significantly anticorrelated with r = −0.62 (<italic>p</italic> < 0.01).</p><p>To test whether our 20 best models were isolated from each other in the parameter space, we varied every parameter separately by plus or minus 1% or 5% around the values of the 20 individuals. The fitnesses of these close neighbors in the parameter space are shown in <xref ref-type="fig" rid="pcbi-0020094-g005">Figure 5</xref>B. Several conclusions can be drawn. First, all good individuals had some good neighbors; they belonged at least to small volumes of good solutions. Second, some neighbors had a better fitness than the individuals we found, so there is room for improvement of the searching technique by applying a local optimization algorithm after the ES. Such hybrid optimization has been proven to work very efficiently [<xref rid="pcbi-0020094-b014" ref-type="bibr">14</xref>]. Third, the parameter space was not smooth; around every good individual, a 5% difference in only one of its parameters could lead to very bad models.</p><p>To understand the parameter space better, we also investigated how the linear combinations of our best solutions behaved. We calculated the fitness of thousands of points in the hyperplanes defined by several triplets of solutions (see <xref ref-type="sec" rid="s4">Materials and Methods</xref>). Some of these hyperplanes, projected onto the (g<sub>NaPs</sub>, g<sub>NaFs</sub>) plane, are shown in <xref ref-type="fig" rid="pcbi-0020094-g006">Figure 6</xref>A–<xref ref-type="fig" rid="pcbi-0020094-g006">6</xref>D and Supplementary <xref ref-type="fig" rid="pcbi-0020094-g006">Figure 6</xref>. Examples of bad models are shown in <xref ref-type="supplementary-material" rid="pcbi-0020094-sg007">Figure S7</xref>. Regions of good models surround our 20 solutions and often link them together. But this is not always the case: in <xref ref-type="fig" rid="pcbi-0020094-g006">Figure 6</xref>A for example, the middle point between individuals 10 and 20 is not a good model; in <xref ref-type="fig" rid="pcbi-0020094-g006">Figure 6</xref>C, the individual 11 is isolated from individuals 4 and 16. None of the 20 solutions was completely isolated from others in every possible hyperplane. In <xref ref-type="fig" rid="pcbi-0020094-g006">Figure 6</xref>B, one can notice that some of the linear combinations of good solutions have a very good fitness value (below 2) whereas some others have a fitness comparable to that of model with completely random parameters (above 10; compare with the first points of <xref ref-type="fig" rid="pcbi-0020094-g001">Figure 1</xref>D).</p><fig id="pcbi-0020094-g006" position="float"><label>Figure 6</label><caption><title>Two-Dimensional Views of Some Hyperplanes of the Parameter Space</title><p>(A–D) Some typical projections onto the (g<sub>NaPs</sub>, g<sub>NaFs</sub>) plane of hyperplanes defined by triplets of individuals. The fitness values of all points belonging to these hyperplanes are color scaled. The three original individuals of each hyperplane are labeled and highlighted by a red square. Grey lines delimitate iso-fitnesses.</p><p>(E) The hyperplane of (D) is shown in red in projection onto the (g<sub>CaTs</sub>, g<sub>CaTd</sub>) plane. The 20 best individuals are represented by points. A blue hyperplane is drawn parallel to the red one. It is defined by adding to the red hyperplane points 10% of the sdv of all solutions in every dimension. Note that individuals that are in between these hyperplanes in this projection can be very far away in other dimensions.</p><p>(F–H) Parallel hyperplanes of (D), with the same projection. These hyperplanes are obtained by adding respectively −5%, +5%, and +10% of sdv to the points belonging to the hyperplane shown in (D). The red cross mark the region of best fitness in the original hyperplane (D).</p></caption><graphic xlink:href="pcbi.0020094.g006"/></fig><p>To go further, it is interesting to visualize what happens in hyperplanes parallel to the hyperplane of <xref ref-type="fig" rid="pcbi-0020094-g006">Figure 6</xref>D for example. Parallel hyperplanes were defined by adding to each point of the original hyperplane −5, +5, or +10% of the sdv of the model distributions for every parameter. Two parallel hyperplanes are visible in the (g<sub>CaTs</sub>, g<sub>CaTd</sub>) plane in <xref ref-type="fig" rid="pcbi-0020094-g006">Figure 6</xref>E. The fitness values of points belonging to parallel hyperplanes are shown in <xref ref-type="fig" rid="pcbi-0020094-g006">Figure 6</xref>F−6H. With −5% of sdv (<xref ref-type="fig" rid="pcbi-0020094-g006">Figure 6</xref>F), almost none of the points make a good model. With +5% of sdv (<xref ref-type="fig" rid="pcbi-0020094-g006">Figure 6</xref>G), a band of good models still exists, but is not fully covering the best region found in <xref ref-type="fig" rid="pcbi-0020094-g006">Figure 6</xref>D, marked with a red cross. With +10% of sdv (<xref ref-type="fig" rid="pcbi-0020094-g006">Figure 6</xref>H), the region of good models is much reduced. The hypervolume delimitating good points is clearly quite restricted to the hyperplane defined by our original solutions.</p></sec><sec id="s2e"><title>Comparison with other Methods</title><p>In <xref ref-type="fig" rid="pcbi-0020094-g007">Figure 7</xref>A, the fitness of hundreds of individuals obtained during algorithm evolution are shown with respect to their g<sub>CaTs</sub> value. The absence of a clear relation between fitness and conductance again confirms the complex interdependencies between the different parameters of the model. Similar distributions, with as much variation, were obtained for all parameters (unpublished data). The electrical activity of the PC is a complicated combination of its conductances, making hand-tuning of parameters impossible if a very precise output is desired. In this section we investigate the likelihood that other automatic parameter search methods would discover comparable high precision models as the ES did.</p><fig id="pcbi-0020094-g007" position="float"><label>Figure 7</label><caption><title>Comparison with other Methods</title><p>(A) Fitness of hundreds of individuals obtained during the searching algorithm evolution, as a function of their g<sub>CaTs</sub> value and centered around the data value (equal to 5).</p><p>(B) Parameter space simplified to a grid of three black points in two dimensions. All individual falling in a pink region will have just one close neighbor, all the points in yellow area will have two close neighbors, and all the points in a white area will have four close neighbors.</p><p>(C) Depending upon individuals (blue circles), between eight and 4,096 close neighbors are found on a six-points-per-dimension grid. The fitnesses of all these neighbors are shown as red crosses. For some individuals which have neighbors with fitness values above 3.45 but below 4 a grid of ten points per dimension was also tested (green crosses).</p></caption><graphic xlink:href="pcbi.0020094.g007"/></fig><p>Prinz et al. [<xref rid="pcbi-0020094-b015" ref-type="bibr">15</xref>] have proposed to use systematic sampling of the parameter space as a way to tune a model's parameters. To test the performances of grid searching on our model we used a grid made of six points in every dimension. This leads to 6<sup>24</sup> = 4.7 billions of billions of points to be tested which is impossible with current computing power. As we already know 20 good solutions, we looked instead at grid points which are neighbors of these solutions to see whether a grid search would have found them. Depending upon individuals, we found between eight and 4,096 close neighbors on the grid (see <xref ref-type="sec" rid="s4">Materials and Methods</xref> and <xref ref-type="fig" rid="pcbi-0020094-g007">Figure 7</xref>B). The fitness values of these grid neighbors are shown in <xref ref-type="fig" rid="pcbi-0020094-g007">Figure 7</xref>C. For most of the individuals (numbered 2, 3, 7, 8, 9, 11, 12, 18, 19) all their grid neighbors had bad fitness values. It would have been impossible to find them with the grid method. Some individuals (4, 5, 6, 10, 14, 15, 17) had some grid neighbors with comparable or better fitness value and would have been discovered with the grid method. A few individuals (1, 13, 16, 20) had neighbors with fitness values above 3.45 but not too high (below 4) so their discovery might be possible with another grid resolution. To check whether this was the case, we used a grid with ten points per dimension (green markers). All the closest neighbors in a ten points grid of these individuals had very bad fitness values. So their discovery with the grid method was quite unlikely. We conclude that a grid search would have discovered only 35% of the solutions found by the ES. Of course, the ES is a stochastic process and new runs of the algorithm would certainly discover additional good individuals. Therefore the grid search is likely to find solutions that the ES did not. What we want to outline here is that, despite its enormous computing cost, a systematic sampling of the parameter space with a resolution of six or ten points per dimension fails to reveal the complexity of the parameter landscape.</p></sec></sec><sec id="s3"><title>Discussion</title><p>We have shown that, by combining evolution strategies with a fitness function based on the phase-plane analysis, it was possible to replicate with high precision the electrophysiological activity of a neuron with complex firing behavior. Strengths of our search method will be described below. The large variety of conductance values found is made possible by the combined effect of several compensatory mechanisms. We have partly unraveled the very complex parameter landscape of the PC model that these mechanisms induce. The implications of these findings will be discussed.</p><sec id="s3a"><title>Strengths of the Method</title><p>Very precise models have been found in a reasonable amount of time despite the high number of variables to tune and the complexity of the cell activity. Several strong points of our method should be underlined. First of all, the phase plane method was successful in avoiding time shift problems. Indeed traces of <xref ref-type="fig" rid="pcbi-0020094-g002">Figure 2</xref>A were considered as almost equal by our fitness function while, because of the time shift between spikes, a fitness function based on the absolute difference between traces would give a very bad value, even worse than the one obtained by comparing a spiking to a silent neuron. The fitness function is also very simple to use as there was no need to define criteria for spiking, bursting, calcium spikes, etc., or to calculate inter spike intervals, hyperpolarizing potentials, spike widths, burst lengths, and so on, while such measures were still reproduced accurately. With neurons presenting a complex behavior, different periods of activity can be easily separated by using different matrices, as we did to distinguish the transitory period after current injection from a stable one later in time.</p><p>A second strength of the method is brought by the use of ES. As a global optimization technique, it produced results which were not local improvements of each other (compare several dots of the same color in <xref ref-type="fig" rid="pcbi-0020094-g004">Figure 4</xref>A). This is important to avoid falling into the local optima's trap. Unlike genetic algorithms, the parameters are coded as real numbers, so the mutation is performed by adding a random number generated from a Gaussian distribution. The width of this distribution can evolve with generations and even be part of the genotype of the individuals. This kind of evolution is impossible with genetic algorithms, where only the probability of a mutation to occur, but not its strength, can be varied. Also, the standard crossover approaches in genetic algorithms request that correlated parameters are put side-by-side in the genotype, which is difficult to achieve when the correlations between parameters are poorly understood. No such requirement exists with ES.</p><p>A couple of interesting conclusions can be made about the original PC model if we consider it as a solution among others to reproduce experimental data. The behavior of the 20 models with respect to the synaptic responses demonstrated that the original model was quite robust in this regard. This is important since, at the time this model was made, these synaptic responses were predictions or tests of its goodness [<xref rid="pcbi-0020094-b012" ref-type="bibr">12</xref>,<xref rid="pcbi-0020094-b013" ref-type="bibr">13</xref>]. Oppositely, the spatial distribution of conductances in the 20 selected models shows that the original choice, driven by simplicity, of equalizing smooth dendritic and spiny dendritic channel densities was not the best choice for reproducing the desired output.</p><p>The quality of the models we obtained can be compared with similar studies published earlier. On one hand, several groups have tested evolutionary algorithms. However, they were trying to tune much simpler cell models. For example, the neocortical pyramidal neuron used by Keren et al. [<xref rid="pcbi-0020094-b009" ref-type="bibr">9</xref>] was either silent or fired tonically without much adaptation. That was also the case for the three active models tuned by Vanier and Bower [<xref rid="pcbi-0020094-b010" ref-type="bibr">10</xref>].</p><p>On the other hand, the grid method was applied by Goldman et al. [<xref rid="pcbi-0020094-b005" ref-type="bibr">5</xref>], as well as Prinz et al. [<xref rid="pcbi-0020094-b004" ref-type="bibr">4</xref>], for tuning the parameters of a single neuron or a small network. These models showed more complex activity but had only a small number of free parameters, which can hardly be augmented for computing time reasons. Additionally, in certain cases, the high number of models evaluated made it impossible, for memory concerns, to save all the membrane potential values. So, while lobster stomatogastric neuron models studied by Prinz et al. [<xref rid="pcbi-0020094-b015" ref-type="bibr">15</xref>] were correctly reproducing data for the characteristics that were explicitly measured (periodicity, burst duration, duty cycle, phase-response properties), the actual voltage traces were quite different when studied in more detail (see their Figure 9C–9D). The grid method may provide a good description of the parameter landscape of simple models but the parameter landscape of the PC model was too complex for this method (<xref ref-type="fig" rid="pcbi-0020094-g007">Figure 7</xref>C).</p><p>We have to notice however that the goodness of these results depends on some assumptions about the available data. First the morphology of the cell needs to be already known. While technically reconstruction of real cell morphologies poses little problems, great variability relevant for the properties of models has been observed between different laboratories [<xref rid="pcbi-0020094-b016" ref-type="bibr">16</xref>]. Second, the kinetics of the channels were not modified. We assumed, like for the original PC model [<xref rid="pcbi-0020094-b012" ref-type="bibr">12</xref>], that the kinetic parameters were sufficiently constrained to have biologically plausible highly precise models.</p></sec><sec id="s3b"><title>Precise Replication of Neuronal Activity from Different Sets of Conductances</title><p>Several very good solutions were found, each comprising a different set of parameter values. The observed differences between the good models are not insignificant, since variations of the same magnitude, applied to each parameter separately, resulted in bad models (<xref ref-type="fig" rid="pcbi-0020094-g005">Figure 5</xref>A). This observation is in agreement with what was expected from the experimental data. The possibility to have the same behavior for computational models from different sets of conductances was already shown by Goldman et al. [<xref rid="pcbi-0020094-b005" ref-type="bibr">5</xref>], as well as by Prinz et al [<xref rid="pcbi-0020094-b004" ref-type="bibr">4</xref>]. But this is, to our knowledge, the first time that very different solutions were found to reproduce much more complex neuronal activity in such detail.</p><p>The required level of detail influences the ratio of maximal conductances of identically behaving models. Indeed, in this study, this ratio ranges for total conductances between 1.2 for g<sub>NaF</sub> to 6.2 for g<sub>KM</sub> (see <xref ref-type="supplementary-material" rid="pcbi-0020094-sg008">Figure S8</xref>) and 9.7 for g<sub>Kh</sub>, but we have shown that this latter current has a small influence on the PC model. In stomatogastric ganglion neuronal models [<xref rid="pcbi-0020094-b004" ref-type="bibr">4</xref>,<xref rid="pcbi-0020094-b005" ref-type="bibr">5</xref>], this ratio is one order of magnitude higher: maximal conductances can vary by factors up to 40 fold. But experimentally measured variations in crab stomatogastric ganglion neurons [<xref rid="pcbi-0020094-b003" ref-type="bibr">3</xref>] are 2–4 fold, much closer to our findings.</p><p>Altogether, recent experimental and theoretical studies are changing our view on how neurons' electrical activity is shaped. For a long time, ionic channel expression was assumed to be relatively constant, hard coded in the genome of the cell and therefore allowing easy classification of neuron types. We have come to realize that instead, channel expression [<xref rid="pcbi-0020094-b017" ref-type="bibr">17</xref>] is constantly being regulated to obtain the desired electrical activity. The modulation state of existing channels could be regulated as well, for example, by cAMP or intracellular calcium dynamics. The mechanisms of this functional homeostasis have still to be deciphered and we believe that further studies of parameter landscapes of different neuron models will provide important contributions.</p><p>Our analysis of the PC parameter landscape demonstrate that several mechanisms contribute to the large differences between the good models. First, some channels had very small effects on the electrical behavior of the model. Second, a few conductance densities were linearly correlated or anti-correlated. Anti-correlation of calcium channel total conductances probably implies that the total amount of calcium that flows in and out of the cell is severely constrained. Anti-correlation of some ionic channels between different regions of the dendrites was also observed. Third, compensation mechanisms between different ionic currents caused small continuous regions of good models in the parameter pace. These regions were quite limited to planes intersecting some good individuals. They did not fully cover the hyperplanes of averages, indicating that threshold mechanisms or non-linear correlations between currents made some averages of good solutions behave badly, as has been reported previously in simpler models [<xref rid="pcbi-0020094-b018" ref-type="bibr">18</xref>]. Understanding how these different mechanisms act together to maintain a precise output and discovering their molecular basis remain two fascinating challenges.</p></sec></sec><sec id="s4"><title>Materials and Methods</title><sec id="s4a"><title>The cell model.</title><p>The model of PC we use has been fully described as “PM9” [<xref rid="pcbi-0020094-b006" ref-type="bibr">6</xref>]. The ionic channels are distributed in the four different physiological zones as follows: the soma has Na<sub>F</sub>, Na<sub>P</sub>, Ca<sub>T</sub>, K<sub>dr</sub>, K<sub>h</sub>, K<sub>M,</sub> and K<sub>A</sub> conductances, whereas the dendrites have Ca<sub>P</sub>, Ca<sub>T</sub>, K<sub>M</sub>, K<sub>C</sub> and K<sub>2</sub> conductances. In addition, the main dendrite expresses also K<sub>dr</sub> and K<sub>A</sub> channels. This leads to a total of 24 channel density parameters to be tuned. They were allowed to vary within a wide physiological range. The bounds for each parameter, as well as the values of the data, are given in <xref ref-type="table" rid="pcbi-0020094-t001">Table 1</xref>. The synaptic properties are exactly the same for the data and the models and are taken from [<xref rid="pcbi-0020094-b012" ref-type="bibr">12</xref>].</p></sec><sec id="s4b"><title>The search algorithm.</title><p>We have used an evolutionary algorithm belonging to the family of ES. The strategy used in the different runs is called (57+19), meaning that, at each generation, the population is composed of 57 individuals and produces 19 offspring. Members of the next generation are selected among the 57 parents, plus 19 children, according to their fitness value. The mutation parameters are self-adaptive and correlated: the width of Gaussian mutations are included in the chromosome and evolve with generations, as well as their covariance matrix angles (see [<xref rid="pcbi-0020094-b007" ref-type="bibr">7</xref>] for details). The recombination scheme used is called “global intermediary”: two parents are selected for each parameter and the offspring inherits the average values for their parameter.</p><p>The PC model was simulated with Genesis 2.2.1 software [<xref rid="pcbi-0020094-b019" ref-type="bibr">19</xref>] under Mac OS X. A ready-to-use ES C++ implementation, called ESEA, is available within the free Evolving Objects library [<xref rid="pcbi-0020094-b020" ref-type="bibr">20</xref>]. We have used a parallel version of it on a cluster of ten Apple 2.3 GHz G5 dual processor nodes. Each run comprising 415 generations took around 6 d of computing.</p></sec><sec id="s4c"><title>The fitness function.</title><p>In LeMasson's phase plane analysis [<xref rid="pcbi-0020094-b011" ref-type="bibr">11</xref>], the fitness measurement is based on the (V, dV/dt) matrix. The electrophysiological trace V(t) is sampled with a frequency of 50 kHz. Each point has V and a dV/dt value and can therefore be ordered in a two-dimensional matrix. The distance between data and model is then given by the difference in density for every cell of the matrix:</p><disp-formula id="pcbi-0020094-e001"><graphic xlink:href="pcbi.0020094.e001.jpg" position="anchor" mimetype="image"/></disp-formula><p>where data<italic><sub>ij</sub></italic> and model<italic><sub>ij</sub></italic> represent the number of points in the matrix cell (<italic>i,j</italic>) for data and model respectively; <italic>N<sub>data</sub></italic> and <italic>N<sub>model</sub></italic>, the total number of data and model points; and <italic>N<sub>x</sub></italic> and <italic>N<sub>y</sub></italic>, the size of the matrix. We have chosen 100 × 100 matrices, where V can vary between −80 and +120 mV and dV/dt between −1500 and +2500 mV/ms.</p><p>The fitness function <italic>F</italic> that we have used is a weighted sum of such <italic>f</italic>:</p><disp-formula id="pcbi-0020094-e002"><graphic xlink:href="pcbi.0020094.e002.jpg" position="anchor" mimetype="image"/></disp-formula><p>where <italic>k</italic> tags different injected current amplitudes: 1.5 nA injected in one thick dendrite, no current, 0.5, 1, 1.5, 2.5, and 3 nA injected in the soma; <italic>l</italic> corresponds to different recording sites: the soma and two thick dendrites; <italic>m</italic> addresses different time periods: the first 100 ms after current injection (the “transitory period”, giving red points in <xref ref-type="fig" rid="pcbi-0020094-g001">Figure 1</xref>B) or the entire recording after 1s (0.5 s if no current is injected) after the beginning of the experiment (the “stable period” giving blue points in <xref ref-type="fig" rid="pcbi-0020094-g001">Figure 1</xref>B). The weights <italic>w<sub>klm</sub></italic> were equal to 1 when <italic>l</italic> designates soma recordings and 0.5 for dendritic recordings, except for null current injection where they are equal to 0.6 and 0.3 respectively.</p></sec><sec id="s4d"><title>The selection of best individuals.</title><p>To check the quality of the models obtained in nine runs of the optimization algorithm, we have simulated the best models with 15 different injected current amplitudes: 1.5 nA in the dendrite, no current, from 0.25 nA to 3 nA with steps of 0.25 nA in the soma and 4 nA in the soma. We have examined all the best solutions and decided to select, for the rest of the analysis, individuals with a fitness below 3.45. Among the 23 individuals that fulfilled this criterion, three were rejected from our final selection because they were bursting and not spiking at the lowest injected current in the soma (0.25 nA). We are certain that these individuals were able to fire in a spiking mode at lower current amplitudes since they were spiking when current was injected in the dendrite. Nevertheless, we preferred not to include them in the rest of the analysis.</p></sec><sec id="s4e"><title>Visualization of hyperplanes.</title><p>From several triplets of solutions, we defined hyperplanes as follows: every point of it was a weighted sum of the three solutions, the first two weights vary from −1.5 to 2.5 by step of 0.04 and the third weight is such that the sum of the weights is equal to 1. All the points with some negative parameters were rejected, which caused multiple borders to the hyperplanes. In total we had thousands of points for each hyperplane. We ran simulations for every point and computed its fitness. In the <xref ref-type="fig" rid="pcbi-0020094-g006">Figure 6</xref> and <xref ref-type="supplementary-material" rid="pcbi-0020094-sg006">Figure S6</xref>, the “land and sea” color scale allows a clear distinction between models above and below the 3.45 value that we used as a threshold between good and bad models. Grey lines in the figures link points that have the same integer fitness values.</p></sec><sec id="s4f"><title>Grid neighbors selection.</title><p>We have used a grid of six (and later ten) points in 24 dimensions. For each dimension, an individual is located between two grid points. If the distance to the closest point was < 1/3 of the distance between two grid points, we considered only the closest neighboring point in this dimension. If not, we considered both. As an illustrative example in the <xref ref-type="fig" rid="pcbi-0020094-g007">Figure 7</xref>B, the grid is simplified to three (black) points in two dimensions. All individuals falling in a pink region will have just one close neighbor, all the individuals in yellow areas will have two close neighbors, and all the individuals in white areas will have four close neighbors (two in each dimension).</p></sec><sec id="s4g"><title>Data analysis.</title><p>Data analysis was done with Igor Pro 5.04b software. The linear correlations between conductances were tested by a simple Pearson's correlation. To obtain total conductances we summed channel densities, normalized to the membrane area where they apply.</p></sec></sec><sec sec-type="supplementary-material" id="s5"><title>Supporting Information</title><supplementary-material content-type="local-data" id="pcbi-0020094-sg001"><label>Figure S1</label><caption><title>Full Comparison of Models with Data</title><p>(A) Comparison of somatic voltages between data (red), the best (blue), and the worst (green) model with 15 injected currents. The amplitudes of injected current are labeled on the left axis where “rest” means no current injected, “d 1.5 nA” means 1.5 nA injected in the dendrite and “s xx nA” means xx nA injected in the soma. Amplitudes included in the fitness function are in bold fonts. The voltage scale is the same for all traces (see <xref ref-type="fig" rid="pcbi-0020094-g002">Figure 2</xref> for scales).</p><p>(B) Same as (A). During 30 ms (spiking traces), 200 ms (small bursting traces), or 800 ms (long bursting traces) only.</p><p>(C) Same as (B) for dendritic voltage. The voltage scale is larger than in (A) and (B) but the same for all traces in (C).</p><p>(2.9 MB PDF).</p></caption><media xlink:href="pcbi.0020094.sg001.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020094-sg002"><label>Figure S2</label><caption><title>Data/Model Comparison for Spike Parameters for 0.25-nA Current Injected Protocol</title><p>The data (mean ± sdv shown as black horizontal bars) is compared to the 20 selected models (means shown as red circles and sdv as vertical bars) for:</p><p>(A) the spike duration; (B) the time between the start and the peak of the spikes; (C) the duration of after-hyperpolarizing potentials; (D) the full width at half maximum of the spikes; (E) the inter-spike interval; (F) the spike area; (G) the spike amplitude; (H) the after-hyperpolarizing potentials amplitude; (I) the potential at which spikes start; (J) the potential at half maximum of the spikes; (K) the potential of the spike peaks; and, (L) the after-hyperpolarizing potentials. See <xref ref-type="fig" rid="pcbi-0020094-g002">Figure 2</xref>A for typical electrophysiological traces. Variability observed in real electrophysiological data is usually of a couple of mV for the spike height and a few tenths of ms for the spike width.</p><p>(883 KB PDF)</p></caption><media xlink:href="pcbi.0020094.sg002.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020094-sg003"><label>Figure S3</label><caption><title>Data/Model Comparison for Spike Parameters for 3-nA Current Injected Protocol</title><p>Same as <xref ref-type="supplementary-material" rid="pcbi-0020094-sg002">Figure S2</xref> during bursting behavior with 3 nA injected current. See <xref ref-type="fig" rid="pcbi-0020094-g002">Figure 2</xref>C for typical electrophysiological traces.</p><p>(893 KB PDF)</p></caption><media xlink:href="pcbi.0020094.sg003.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020094-sg004"><label>Figure S4</label><caption><title>Data/Model Comparison for Burst Parameters for 3-nA Current Injected Protocol</title><p>The data (mean ± sdv shown as black horizontal bars) is compared to the 20 selected models (means shown as red circles and sdv as vertical bars) for:</p><p>(A) the inter-burst interval; (B) the duration of bursts; (C) the number of spikes per burst; and, (D) the inter-spike interval inside bursts.</p><p>(594 KB PDF)</p></caption><media xlink:href="pcbi.0020094.sg004.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020094-sg005"><label>Figure S5</label><caption><title>Linear Correlation between Conductance Densities</title><p>(A) For each pair of conductance densities a Pearson's correlation coefficient is calculated. five couples have a probability <italic>(p)</italic> to have a null correlation below 1%.</p><p>(B) Distribution of g<sub>K2t</sub> vs. g<sub>K2d</sub> for the 20 selected individuals. The colors of the points are the same as in <xref ref-type="fig" rid="pcbi-0020094-g001">Figure 1</xref>E. The axes extend over the limits in which conductance densities were allowed to vary. The black cross indicates the value of the data.</p><p>(C–F) Same as (B) for other pairs of correlated conductance densities.</p><p>(716 KB PDF)</p></caption><media xlink:href="pcbi.0020094.sg005.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020094-sg006"><label>Figure S6</label><caption><title>Examples of other Hyperplanes</title><p>Same as <xref ref-type="fig" rid="pcbi-0020094-g006">Figure 6</xref>A–<xref ref-type="fig" rid="pcbi-0020094-g006">6</xref>D.</p><p>Red crosses in <xref ref-type="supplementary-material" rid="pcbi-0020094-sg006">Figure S6</xref>B label models that are shown in <xref ref-type="supplementary-material" rid="pcbi-0020094-sg007">Figure S7</xref>.</p><p>(2.9 MB PDF)</p></caption><media xlink:href="pcbi.0020094.sg006.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020094-sg007"><label>Figure S7</label><caption><title>Examples of Badly Behaving Models</title><p>Electrophysiological activity of the three models marked with a red cross in <xref ref-type="supplementary-material" rid="pcbi-0020094-sg006">Figure S6</xref>B. The models have a fitness equal to 3.9997 (red), 5.0003 (green), and 6.0001 (blue). The abscissa and ordinates are the same as in <xref ref-type="supplementary-material" rid="pcbi-0020094-sg001">Figure S1</xref>.</p><p>(3.0 MB PDF)</p></caption><media xlink:href="pcbi.0020094.sg007.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020094-sg008"><label>Figure S8</label><caption><title>Range of Variation of the Total Conductances</title><p>Ratio of maximal over minimal value found for the total conductances of <xref ref-type="fig" rid="pcbi-0020094-g004">Figure 4</xref>C.</p><p>(18 KB EPS)</p></caption><media xlink:href="pcbi.0020094.sg008.eps"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020094-sd001"><label>Protocol S1</label><caption><title>Fine Tuning of the Search Method</title><p>(24 KB DOC)</p></caption><media xlink:href="pcbi.0020094.sd001.doc"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020094-sd002"><label>Protocol S2</label><caption><title>Accuracy of the Models</title><p>(28 KB DOC)</p></caption><media xlink:href="pcbi.0020094.sd002.doc"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material></sec>
Key Role of Local Regulation in Chemosensing Revealed by a New Molecular Interaction-Based Modeling Method	<p>The signaling network underlying eukaryotic chemosensing is a complex combination of receptor-mediated transmembrane signals, lipid modifications, protein translocations, and differential activation/deactivation of membrane-bound and cytosolic components. As such, it provides particularly interesting challenges for a combined computational and experimental analysis. We developed a novel detailed molecular signaling model that, when used to simulate the response to the attractant cyclic adenosine monophosphate (cAMP), made nontrivial predictions about <italic>Dictyostelium</italic> chemosensing. These predictions, including the unexpected existence of spatially asymmetrical, multiphasic, cyclic adenosine monophosphate–induced PTEN translocation and phosphatidylinositol-(3,4,5)P<sub>3</sub> generation, were experimentally verified by quantitative single-cell microscopy leading us to propose significant modifications to the current standard model for chemoattractant-induced biochemical polarization in this organism. Key to this successful modeling effort was the use of “Simmune,” a new software package that supports the facile development and testing of detailed computational representations of cellular behavior. An intuitive interface allows user definition of complex signaling networks based on the definition of specific molecular binding site interactions and the subcellular localization of molecules. It automatically translates such inputs into spatially resolved simulations and dynamic graphical representations of the resulting signaling network that can be explored in a manner that closely parallels wet lab experimental procedures. These features of Simmune were critical to the model development and analysis presented here and are likely to be useful in the computational investigation of many aspects of cell biology.</p>	<contrib contrib-type="author"><name><surname>Meier-Schellersheim</surname><given-names>Martin</given-names></name><xref ref-type="aff" rid="aff1">1</xref><xref ref-type="corresp" rid="cor1"></xref></contrib><contrib contrib-type="author"><name><surname>Xu</surname><given-names>Xuehua</given-names></name><xref ref-type="aff" rid="aff2">2</xref></contrib><contrib contrib-type="author"><name><surname>Angermann</surname><given-names>Bastian</given-names></name><xref ref-type="aff" rid="aff1">1</xref><xref ref-type="aff" rid="aff4">4</xref></contrib><contrib contrib-type="author"><name><surname>Kunkel</surname><given-names>Eric J</given-names></name><xref ref-type="aff" rid="aff3">3</xref></contrib><contrib contrib-type="author"><name><surname>Jin</surname><given-names>Tian</given-names></name><xref ref-type="aff" rid="aff2">2</xref></contrib><contrib contrib-type="author"><name><surname>Germain</surname><given-names>Ronald N</given-names></name><xref ref-type="aff" rid="aff1">1</xref><xref ref-type="corresp" rid="cor1"></xref></contrib>	PLoS Computational Biology	<sec id="s1"><title>Introduction</title><p>The ability of eukaryotic cells to process nonisotropic extracellular stimuli lies at the heart of many aspects of cellular behavior including directed cell growth, movement, and cell-cell communication [<xref rid="pcbi-0020082-b001" ref-type="bibr">1</xref>–<xref rid="pcbi-0020082-b004" ref-type="bibr">4</xref>]. When an external stimulus is localized to a discrete patch of membrane, for example, during contact-dependent cell-cell communication, the physical recruitment of signaling components into receptor-associated multimolecular complexes provides a straightforward mechanism for establishing the appropriate directionality of intracellular responses. Under these circumstances, it is readily apparent how the activation of excitatory components at response onset and of inhibitory components during negative regulation (sometimes called <italic>adaptation</italic>) is constrained to a specific region of the cell.</p><p>In contrast, during chemotaxis (directed movement along a chemical gradient) cells display a very strong intracellular biochemical polarization even though the external stimulus may surround the entire cell and differ by only a few percent in concentration at different points along the membrane. Here, the principle of physical locality invoked in the case of cell-cell contact does not seem to provide an obvious mechanism for the translation of small extracellular directional clues into an almost digital biochemical, cytoskeletal, and morphological polarization of the responding cell. The striking capacity of cells to reliably sense and respond to minute gradients has led to intense experimental analysis at the cell biological level as well as to development of quantitative mathematical models of the chemosensing phenomenon.</p><p>Many of the core components of the underlying signaling network have been elucidated in several cell types, especially neutrophils and the amoeba of <named-content content-type="genus-species">Dictyostelium discoidium</named-content> [<xref rid="pcbi-0020082-b005" ref-type="bibr">5</xref>–<xref rid="pcbi-0020082-b007" ref-type="bibr">7</xref>]. A central role is played by the phospholipid phosphatidylinositol-(3,4,5)P<sub>3</sub> (PIP<sub>3</sub>), the product of the action of the lipid kinase phosphoinositide 3-kinase (PI3K) on the relatively abundant membrane component phosphatidylinositol-(4,5)P<sub>2</sub> (PIP<sub>2</sub>). The local concentration of PIP<sub>3</sub> is determined by the competing actions of PI3K and the phosphatase PTEN (phosphatase and tensin homolog deleted on Chromosome 10) that dephosphorylates PIP<sub>3</sub> at the 3′ position to produce PIP<sub>2</sub> [<xref rid="pcbi-0020082-b008" ref-type="bibr">8</xref>]. PIP<sub>3</sub> functions as a membrane anchor for signaling proteins with pleckstrin-homology (PH) domains [<xref rid="pcbi-0020082-b009" ref-type="bibr">9</xref>]. Such proteins include several that regulate the polymerization of actin, which, in turn, is needed for the extension of pseudopods/lamellipodia that allow cells to crawl [<xref rid="pcbi-0020082-b010" ref-type="bibr">10</xref>]. For this reason, localized PIP<sub>3</sub> accumulation is frequently used as a read-out of polarization in response to chemoattractant gradients. In spite of this progress in defining the main molecular players and their interactions, however, we do not yet fully understand how eukaryotic cells are able to amplify the primary receptor signals induced by shallow extracellular gradients of chemoattractants into steep intracellular gradients of signaling molecules like PIP<sub>3</sub>. Nor is it known precisely how the chemosensing signaling machinery adapts to attractant stimulations or what the connection is between the mechanisms that lead to amplification and those that lead to adaptation.</p><p>The dominant hypothesis embodied in existing conceptual and computational models for how cells achieve polarization in the face of shallow chemoattractant gradients is termed “local excitation – global inhibition” [<xref rid="pcbi-0020082-b006" ref-type="bibr">6</xref>]. Meinhardt and Gierer [<xref rid="pcbi-0020082-b011" ref-type="bibr">11</xref>,<xref rid="pcbi-0020082-b012" ref-type="bibr">12</xref>] in particular showed that combinations of localized positive feedback and long-range inhibition could potentially produce the biochemical inhomogeneity encountered during chemotactic responses, and variations on this theme appear in the other models in this class [<xref rid="pcbi-0020082-b013" ref-type="bibr">13</xref>,<xref rid="pcbi-0020082-b014" ref-type="bibr">14</xref>]. The main idea is that polarization results from a combination of local activation processes and one (or more) globally acting, signal-induced inhibitor(s) tuned to be just strong enough to suppress the local responses everywhere except in those regions of the cell where the extracellular concentration of the stimulus is highest. The globally acting inhibitor, whose level is determined by the overall occupancy of the chemoattractant receptors, indirectly plays the role of a messenger that provides cell-wide information about the average extracellular chemoattractant concentration. This average is the standard used by the cell to determine which side is experiencing “high” and which side “low” relative chemoattractant concentrations. Adaptation, in this context, is the failure of local signals to significantly rise above the average and overcome the induced level of global inhibition.</p><p>While these modeling efforts have had a major influence on the way investigators view the molecular basis for chemosensing, the field still lacks a model that reflects to the greatest extent possible the detailed biochemistry revealed by bench experiments and whose predictions have been widely validated by experimental tests at the single cell level. For example, in <italic>Dictyostelium,</italic> stimulation with homogeneous fields of the chemoattractant cyclic adenosine monophosphate (cAMP) produces a fast rise in PIP<sub>3</sub> around the entire cell circumference, followed by a rapid return to (or below) prestimulus level [<xref rid="pcbi-0020082-b006" ref-type="bibr">6</xref>,<xref rid="pcbi-0020082-b015" ref-type="bibr">15</xref>]. PTEN shows the inverse pattern, leaving the membrane upon exposure to cAMP and then returning during the adaptation phase [<xref rid="pcbi-0020082-b007" ref-type="bibr">7</xref>]. However, the concentration of membrane-bound PTEN is still considerably below its prestimulus level at the time when the PIP<sub>3</sub> concentration peaks and starts to decay. Given the current belief that in these cells, PI3K is the only generator of PIP<sub>3</sub> and PTEN is the most important enzyme metabolizing PIP<sub>3</sub>, the point at which PIP<sub>3</sub>'s concentration in the membrane peaks must mark the moment at which the activities of PI3K and PTEN are temporarily in equilibrium. The cAMP signal must therefore rapidly activate a component that negatively regulates PI3K so that its activity is balanced by the lower than resting level of PTEN present at the membrane at this point in time. Such a negative regulatory component is not explicitly included in existing models of <italic>Dictyostelium</italic> chemosensing.</p><p>In part, the absence of such specific components in previous computational treatments arises from a tendency to use abstract modules in models of complex biological signaling networks whenever the details are incompletely understood. Although such approaches are in part guided by the principle of parsimony, another, more practical reason for the introduction of abstract modules is that the literature mining involved in the creation of complex biological signaling models and their translation into computational representations can be quite difficult.</p><p>Here we report a two-pronged approach to the development of a more complete and explicit model of <italic>Dictyostelium</italic> chemosensing. First, at the biochemical level, we started with a basic chemosensing model containing only the principal molecular players incorporated into the current paradigm in the field. We then tried to fill in “logical” blanks (such as the negative regulator of PI3K discussed above) so as to be able to reproduce the experimentally reported dynamical behaviors of PIP<sub>3</sub>, PTEN, and PI3K upon exposure of a morphologically unpolarized <italic>Dictyostelium</italic> cell to chemoattractant exposure [<xref rid="pcbi-0020082-b007" ref-type="bibr">7</xref>,<xref rid="pcbi-0020082-b016" ref-type="bibr">16</xref>]. Our specific aim was to construct a model that did not contain abstract “black-box” modules but instead used specific molecular components and interactions reported in the experimental literature to provide necessary functions (positive feedback, negative regulation, amplification).</p><p>Second, to help achieve this goal of molecular detail, we utilized a new software approach called Simmune that was created to facilitate the development and simulation of realistic, and therefore frequently quite complex, cell-biological models. Originally developed because of an interest in simulating immune responses, Simmune has no attributes that are unique to immunology and it is applicable to the simulation of any cell-biological system. The software suite allows molecular reactions to be defined directly at the level of interactions between molecular binding sites, using simple graphical representations of molecules and molecular complexes. This permits expert biologists to construct and simulate complex models without direct involvement in the underlying mathematics.</p><p>The signaling network that we developed through literature mining and the application of Simmune shares some properties with the abstract model developed by Meinhardt [<xref rid="pcbi-0020082-b012" ref-type="bibr">12</xref>], but it goes further by providing an explicit molecular definition of the feedback module and the inhibitory regulator postulated in this earlier work. Our model also includes the action of a second, slower inhibitory component that is reminiscent of Meinhard's “poisoning” element. This element was introduced to prevent cells from locking into the direction of given stimulus and becoming blind to subsequent changes in direction and strength of the external signal. However, the present scheme differs in important aspects from this previous proposal and from the “local excitation–global inhibition” models suggested by Devreotes and coworkers [<xref rid="pcbi-0020082-b014" ref-type="bibr">14</xref>] because the locally acting feedback mechanism and the distribution of the inhibitory component (PTEN) are now coupled.</p><p>Simulations performed with our model predicted previously unrecognized biphasic spatiotemporal changes in PIP<sub>3</sub> and PTEN localization and concentration in <italic>Dictyostelium</italic> cells responding to cAMP gradients. These predictions were confirmed across a range of gradient conditions by single-cell imaging studies, leading us to propose a substantial modification to the standard “local excitation–global inhibition” model, with locally acting negative feedback now seen as playing a key role in controlling the development of chemoattractant-induced biochemical polarity.</p></sec><sec id="s2"><title>Results</title><p>In constructing a new detailed model of <italic>Dictyostelium</italic> cAMP chemosensing, we needed to ensure that the major features of the cellular response to chemoattractant exposure would be present in the simulated behavior of a cell. First, when exposed to a uniform concentration of attractant, the model should show a transient membrane response that rapidly reverts to the resting state (global adaptation). Second, when exposed to a shallow gradient of attractant, the simulated cell should generate a much steeper intracellular biochemical gradient (amplification). The starting point for the construction of a model with these features was the well-established biochemical scheme described in [<xref rid="pcbi-0020082-b008" ref-type="bibr">8</xref>] and [<xref rid="pcbi-0020082-b017" ref-type="bibr">17</xref>] (<xref ref-type="fig" rid="pcbi-0020082-g001">Figure 1</xref>). Ligand binding to the cAMP receptor leads to activation of the associated Gαβγ, yielding Gα and Gβγ. These effectors, in turn, promote the activation of Ras and the membrane recruitment and activation of PI3K, along with a G protein–dependent translocation of PTEN from the membrane to the cytosol. The combination of PI3K activation and a loss of membrane-bound PTEN results in a rapid increase of PIP<sub>3</sub> upon stimulation of the cells with cAMP (<xref ref-type="fig" rid="pcbi-0020082-g001">Figure 1</xref>).</p><fig id="pcbi-0020082-g001" position="float"><label>Figure 1</label><caption><title>Standard Model of the Extracellular and Intracellular Distribution of Key Components of <italic>Dictyostelium</italic> Chemotactic Signaling</title><p>(A) In an unstimulated cell PTEN is homogeneously distributed at the membrane. The cell membrane contains very little PIP<sub>3</sub>.</p><p>(B) Stimulation of the cell leads to the membrane recruitment and activation of PI3K, as indicated by the arrows (1) leading from inactive, mainly cytosolic PI3K (yellow) to membrane-proximal, active PI3K (orange). Activated PI3K transforms PIP<sub>2</sub> into PIP<sub>3</sub>. PTEN is deactivated following cAMP stimulation and leaves the membrane. This process is indicated by arrows (2) connecting active PTEN (dark green) and the mainly cytosolic inactive PTEN (light green). Regulatory processes lead to reactivation of PTEN (3). Differences in the speed and degree of cAMP receptor ligation between front and back of the cell lead to preferential accumulation of PTEN at the back of the cell. As a result, the front experiences a higher concentration of PI3K and a lower concentration of PTEN than the back and accumulates PIP<sub>3</sub>. Note: To emphasize the changes in PIP<sub>3</sub> content, the amount of PIP<sub>3</sub> relative to that of PIP<sub>2</sub> has been overstated. Even after cAMP stimulation, the actual amount of PIP<sub>2</sub> will be much higher than that of PIP<sub>3</sub>.</p></caption><graphic xlink:href="pcbi.0020082.g001"/></fig><sec id="s2a"><title>Development of a Refined Signaling Model of Chemosensing Based on Quantitative Live Cell Imaging and Literature Mining</title><p>Fusion molecules containing both a PIP<sub>3</sub>-specific PH domain and a fluorescent protein domain have been produced to monitor the dynamics of the distribution of PIP<sub>3</sub> under various modes of <italic>Dictyostelium</italic> exposure to chemoattractant [<xref rid="pcbi-0020082-b001" ref-type="bibr">1</xref>,<xref rid="pcbi-0020082-b018" ref-type="bibr">18</xref>,<xref rid="pcbi-0020082-b019" ref-type="bibr">19</xref>]. When stimulated with a homogeneous field of cAMP, previously unpolarized cells respond with a rapid, transient increase of PIP<sub>3</sub> around their entire perimeter [<xref rid="pcbi-0020082-b006" ref-type="bibr">6</xref>,<xref rid="pcbi-0020082-b020" ref-type="bibr">20</xref>]. In contrast, upon exposure to gradients of chemoattractant, the plasma membrane concentration of PIP<sub>3</sub> increases throughout the cell initially and then decays everywhere to below the pre-stimulus level, except for the side of the cell exposed to the higher chemoattractant concentration [<xref rid="pcbi-0020082-b006" ref-type="bibr">6</xref>,<xref rid="pcbi-0020082-b015" ref-type="bibr">15</xref>]. Using a fluorescent fusion of PTEN (GFP [green fluorescent protein]-PTEN) [<xref rid="pcbi-0020082-b007" ref-type="bibr">7</xref>,<xref rid="pcbi-0020082-b021" ref-type="bibr">21</xref>], similar imaging studies have been used to show that, following a homogeneous cAMP stimulus, much of the pool of PTEN translocates from the membrane to the cytosol within a few seconds and then slowly returns to the membrane [<xref rid="pcbi-0020082-b007" ref-type="bibr">7</xref>].</p><p>Activation of membrane-recruited PI3K and translocation of PTEN to the cytosol are the two “excitatory” mechanisms that lead to the initial increase in membrane PIP<sub>3</sub> revealed by PH-GFP relocalization (<xref ref-type="fig" rid="pcbi-0020082-g001">Figure 1</xref>) and qualitative diagrams of chemosensory signaling frequently focus on these excitatory signaling events [<xref rid="pcbi-0020082-b017" ref-type="bibr">17</xref>]. Simulations using a model incorporating only these mechanisms would, however, show the concentration of membrane PIP<sub>3</sub> to increase upon introduction of cAMP without being followed by either the experimentally reported rapid decay in the concentration of this membrane phospholipid or the relocalization of PTEN to the membrane. Therefore, the cells must contain inhibitory components that quench PI3K signaling and also reverse those changes that cause PTEN to dissociate from the membrane, allowing it to return to its pre-stimulus distribution (adaptation). As already noted, a comparison of the dynamic changes in the concentrations of PIP<sub>3</sub> and membrane-bound PTEN upon exposure of <italic>Dictyostelium</italic> cells to a homogeneous field of cAMP allowed us to conclude that PI3K activity must be rapidly negatively regulated to account for the observed post-stimulus changes in membrane PIP<sub>3</sub> levels.</p><p>In seeking to account for this regulation of PI3K activity, we took advantage of our previous findings that the α and the βγ subunits of the G proteins remained dissociated as long as chemoattractant was present, establishing that there was ongoing signaling through the excitatory pathways [<xref rid="pcbi-0020082-b015" ref-type="bibr">15</xref>]. The inhibitory components thus do not simply shut off receptor signaling but actively compete with ongoing excitatory mechanisms. For the PI3K signaling branch, we therefore included stimulus-induced activation and membrane recruitment of a phosphatase that deactivates PI3K (<xref ref-type="fig" rid="pcbi-0020082-g002">Figure 2</xref>A, module 1) and of a Ras GAP [<xref rid="pcbi-0020082-b022" ref-type="bibr">22</xref>] that suppresses Ras activity after an initial spike [<xref rid="pcbi-0020082-b016" ref-type="bibr">16</xref>] (<xref ref-type="fig" rid="pcbi-0020082-g002">Figure 2</xref>A, module 2). Additionally, we assumed that the amount of free Gβγ contributing to activation of PI3K is controlled by a blocking element, because the dynamics of the system did not appear to permit re-association with Gα to dampen Gβγ effector function quickly enough (<xref ref-type="fig" rid="pcbi-0020082-g002">Figure 2</xref>A, module 3). Receptor-associated kinase (RAK) was subsequently reported to function in mammalian cells in a manner consistent with this assumption [<xref rid="pcbi-0020082-b023" ref-type="bibr">23</xref>].</p><fig id="pcbi-0020082-g002" position="float"><label>Figure 2</label><caption><title>PI3K and PTEN Regulatory Modules</title><p>(A) Elements controlling the activity of PI3K and upstream components. In addition to the basic, “excitatory,” signaling elements like the cAMP receptor, Gβγ, and Ras, we introduced further elements controlling the activity of PI3K and upstream components. “PI3Ktp” (module 1) stands for a tyrosine phosphatase that deactivates PI3K. This phosphatase becomes enzymatically activated and is recruited to the membrane after interaction with Gβγ. “RasGAP” (module 2) translocates to the membrane and deactivates Ras after activation by Gβγ. RAK blocks Gβγ, Gα, and the receptor, thereby reducing all signals (module 3).</p><p>(B) Elements controlling the activity and localization of PTEN. In our model, PTEN is phosphorylated by a Src-like kinase, here simply called “Src” (module 1). Src is activated by Gα and deactivated by Csk, which in turn is recruited by phosphoPaxillin (“pPaxillin”) (module 2). SHP2, which is membrane recruited by pGab1 bound to PIP<sub>3</sub>, dephosphorylates pPaxillin (module 3), thereby leading to increased activation of PTEN.</p></caption><graphic xlink:href="pcbi.0020082.g002"/></fig><p>We also needed to add to the model a set of specific biochemical processes regulating PTEN localization within the cell, in particular a link from activation of the sensing receptor to the dissociation of PTEN from its resting location at the plasma membrane and a negative regulatory limb that reverses this active induction of membrane dislocation. We hypothesized that the molecular change causing PTEN to dissociate from the membrane is a phosphorylation event that in mammalian cells is mediated by a Src-like kinase [<xref rid="pcbi-0020082-b024" ref-type="bibr">24</xref>] (<xref ref-type="fig" rid="pcbi-0020082-g002">Figure 2</xref>B, module 1). Reversal of the process that leads to release of PTEN from the plasma membrane thus requires control of the activity of this kinase. We adopted the pathway reported by Ren et al. [<xref rid="pcbi-0020082-b025" ref-type="bibr">25</xref>], according to which Src activity is negatively controlled by Csk, which is recruited to the plasma membrane by phosphopaxillin (<xref ref-type="fig" rid="pcbi-0020082-g002">Figure 2</xref>B, module 2). The phosphorylation of paxillin in turn is controlled by the tyrosine phosphatase SHP2, which is brought into proximity of the membrane-bound paxillin through the PIP<sub>3</sub>-binding adaptor Gab1 (<xref ref-type="fig" rid="pcbi-0020082-g002">Figure 2</xref>B, module 3). <xref ref-type="supplementary-material" rid="pcbi-0020082-sg001">Figure S1</xref> shows the full network of enzymatic interactions. <xref ref-type="supplementary-material" rid="pcbi-0020082-sg002">Figures S2</xref> and <xref ref-type="supplementary-material" rid="pcbi-0020082-sg003">S3</xref> focus on those branches of the network that control PI3K and PTEN activity, respectively.</p></sec><sec id="s2b"><title>Simmune Permits Construction of Detailed, Biology-Based Models That Avoid “Black-Box” Abstractions</title><p>In order to construct the model signaling network outlined above, with regulatory modules that account for the time-dependent return of the chemosensing apparatus to baseline after a stimulus (adaptation) and that could be tested for whether the model was also adequate to account for intracellular amplification, we incorporated molecules and mechanisms that are well-documented in the literature whenever possible, though in some cases we had to speculate about components and interactions (see also <xref ref-type="supplementary-material" rid="pcbi-0020082-sd001">Text S1</xref>). <xref ref-type="supplementary-material" rid="pcbi-0020082-sd002">Text S2</xref> provides a detailed analysis of the resultant chemosensing signaling network in terms of modules with specific functionalities (“signal transduction module,” “adaptation module,” and “gradient amplification module”) and a discussion of the behavior of the network upon modifications of these modules.</p><p>This approach of incorporating molecular detail in regulatory pathways that have been incompletely defined in the experimental literature differs from the more frequently encountered abstract methods of filling these gaps that appear in many theoretical treatments of this topic. Often, limitations in experimental data and also in computational tools have led modelers to utilize conceptual signaling modules (“black boxes”) with the right input-output behavior, to achieve the desired overall behavior of a signaling network. While this approach reduces the effort needed to construct the differential equation equivalent of the signaling network and may provide insights about general dynamic properties of the simulated system, it cannot be used to investigate the behavior of specific molecular mechanisms and frequently fails to incorporate essential regulatory dynamics into the resultant model.</p><p>The software suite Simmune was developed to overcome many of the difficulties associated with the creation of detailed, biologically realistic, quantitative models suitable for simulation. It uses the familiar steps of dialog box entry and menu selection to allow the user to readily define molecule types as well as to specify the number and properties of their binding and/or enzymatic sites (<xref ref-type="fig" rid="pcbi-0020082-g003">Figure 3</xref>). This same interface also allows the investigator to define whether a molecule is membrane bound, in which case its movement will be limited to intramembranous diffusion, and to note which part of a transmembrane molecule is in the extracellular environment and which inside the cell, restricting the other molecular species with which these topologically distinct domains can interact during a simulation. Symbolic graphical representations of the defined molecule types appear on the screen and interactions between these entities can be specified in Simmune by simple mouse clicking and dragging to draw connections between the potential binding sites of the reaction partners.</p><fig id="pcbi-0020082-g003" position="float"><label>Figure 3</label><caption><title>Defining Quantitative Interactions between Molecular Binding Sites Using Simmune (Screenshot)</title><p>The screenshot of Simmune's modeling interface shows the graphical representations of PI3K and PIP<sub>2</sub> and the binding sites through which they interact, as well as the sites of membrane attachment. The turquoise circle around the upper binding site of PI3K identifies it as an enzymatically active site. The dotted line represents the possibility of a binding interaction between these two molecules. Selecting the interaction by clicking on the handle on the dotted line allows entry of the relevant binding parameters.</p></caption><graphic xlink:href="pcbi.0020082.g003"/></fig><p>By this simple process of defining binary interactions between the binding sites of reacting species (which closely follows the way biologists traditionally think about signaling networks), the user provides Simmune with the information needed to determine which molecular complexes are possible in the system, for example, a complex consisting of the cAMP receptor and an associated Gαβγ heterotrimer. The program then automatically constructs the complete network of complexes from the component definition input, collects the reactions in which each complex participates, and calculates the contributions of these reactions to the rate of change of the concentration of each complex (which depends on the reaction rates and the—typically changing—concentrations of the reaction partners). Given a set of initial conditions, Simmune can then calculate the time course of changes in the concentrations of all molecular complexes in the model after a stimulus is applied. The software also permits the investigator to both visualize and obtain a dynamic quantitative readout of the subcellular location (<xref ref-type="fig" rid="pcbi-0020082-g004">Figure 4</xref>) and binding states of all the molecular complexes and of the flux through the network during the simulated signaling process (<xref ref-type="supplementary-material" rid="pcbi-0020082-sg001">Figure S1</xref>, <xref ref-type="supplementary-material" rid="pcbi-0020082-sv001">Video S1</xref>).</p><fig id="pcbi-0020082-g004" position="float"><label>Figure 4</label><caption><title>Comparison of the Simulated Activities of PI3K, Membrane-Bound PTEN, and the Resulting Behavior of PIP<sub>3</sub> (Composite Screenshot)</title><p>Stimulation of a cell in a 2:1 cAMP gradient (mean concentration 500 nmol) leads to a rapid 3-fold increase in the membrane proximal activity of PI3K (green) and to a loss of membrane-bound PTEN (blue; tracked as GFP-PTEN in real cells). This results in a rapid accumulation of PIP<sub>3</sub> (red; reported by the location of PH-GFP in real cells). Subsequently, the PI3K activity is strongly quenched by the recruitment of regulatory components to the membrane and falls below its prestimulus level in less than 20 s. PTEN returns more slowly to the membrane. During the phase of downregulation of PI3K activity and reattachment of PTEN to the membrane, the concentration of PIP<sub>3</sub> decays. In the front of the cell (which experiences a high cAMP concentration), membrane-associated PTEN only returns to a fraction of its prestimulus level and then enters a second phase of decline. After approximately 50 s, the low level of membrane-bound PTEN that is reached in the front of the cell allows PIP<sub>3</sub> to increase again, even though the amount of active PI3K in this region is modest. In the back of the cell (low cAMP concentration), membrane-bound PTEN increases beyond its prestimulus level, resulting in a decrease of PIP<sub>3</sub> below its resting state concentration. The circular inset shows a two-dimensional representation of the dynamics of membrane-bound PTEN and PIP<sub>3</sub> in different regions of the three-dimensional simulation of a cell.</p></caption><graphic xlink:href="pcbi.0020082.g004"/></fig></sec><sec id="s2c"><title>Simulation Predictions versus Experimental Findings</title><p>Computational models are best evaluated by examining their ability to make correct predictions how a biological system will behave under conditions for which experimental data are not yet available and that were not used to establish the model's core parameters. In the present case, this meant using responses to homogeneous chemoattractant fields to establish parameters that were then used by Simmune to simulate the response of <italic>Dictyostelium</italic> not only to other (untested) homogeneous concentrations of cAMP but, more important, to gradients of cAMP that were not used for creating the model.</p><p>To measure quantitatively the behavior of PIP<sub>3</sub> and PTEN during the chemosensing response of <italic>Dictyostelium,</italic> experiments were performed by single-cell microscopy of cells in a one-well chamber reacting to microinjector-delivered cAMP stimulation (see [<xref rid="pcbi-0020082-b015" ref-type="bibr">15</xref>] for details). We simulated these experiments using Simmune by placing simulated cells with a given initial composition of signaling molecules into three-dimensional extracellular compartments (representing the culture well). The model was then run for a period of time to allow the simulated cells to achieve a biochemical steady-state in the absence of cAMP stimulation The fluctuations in shape and PIP<sub>3</sub> distribution seen even in the absence of cAMP exposure using amoeboid cells with actin-based morphological prepolarization [<xref rid="pcbi-0020082-b026" ref-type="bibr">26</xref>] were avoided in the present study by using lactrunculin to inhibit actin polymerization, thus allowing the result of this computational equilibration process to reflect the biological situation before stimulation with cAMP. After equilibration the simulated cells were exposed to in silico cAMP gradients equivalent to the actual chemoattractant conditions experienced by the real cells under experimental conditions. Simmune then calculated the spatiotemporal response of the concentrations of components like PIP<sub>3</sub>, activated PI3K, and membrane-bound or cytosolic PTEN. The software furthermore allowed us to save the time-dependent behavior of molecular concentrations and flows through the signaling pathways of the simulated cell to data files for visualization using automatically generated diagrams of the cell's reaction network (<xref ref-type="supplementary-material" rid="pcbi-0020082-sg001">Figure S1</xref>, <xref ref-type="supplementary-material" rid="pcbi-0020082-sv001">Video S1</xref>).</p><p>First, we examined in this manner the responses of simulated cells to varying concentrations of homogeneously applied cAMP. In these simulations, the time T<sub>max</sub> post-cAMP exposure at which PIP<sub>3</sub> reached its peak decreased with increasing concentration of the applied chemoattractant. After having adjusted the model parameters in such a way that T<sub>max</sub> for one concentration of cAMP (25 nmol) in the simulation matched the experimentally observed value, simulations quantitatively predicted the correct concentration dependence of T<sub>max</sub> for 10-fold higher and 10-fold lower concentrations of cAMP in latrunculin-treated (morphologically nonpolarized) <italic>Dictyostelium</italic> cells expressing the PIP<sub>3</sub>-binding chimeric molecule PH<sub>crac</sub>-GFP (<xref ref-type="supplementary-material" rid="pcbi-0020082-sg004">Figure S4</xref>; [<xref rid="pcbi-0020082-b001" ref-type="bibr">1</xref>,<xref rid="pcbi-0020082-b018" ref-type="bibr">18</xref>,<xref rid="pcbi-0020082-b019" ref-type="bibr">19</xref>]). This established that the model was able to accurately predict cellular behavior under conditions not used to establish the specific parameter values employed for the simulations.</p><p>We next examined the model's performance when used to simulate a cell's more complex response following exposure to extracellular cAMP gradients. The simulated cells showed the previously reported transient PIP<sub>3</sub> elevation along their entire perimeter [<xref rid="pcbi-0020082-b006" ref-type="bibr">6</xref>,<xref rid="pcbi-0020082-b020" ref-type="bibr">20</xref>], which then completely decayed everywhere except for the “front” side of the cell that was exposed to the highest concentration of cAMP. Also, simulated PTEN translocated from the cytoplasmic membrane to the cytosol and reassociated with the membrane in the posterior of the cell as expected [<xref rid="pcbi-0020082-b007" ref-type="bibr">7</xref>]. However, the PIP<sub>3</sub> response at the “front” side of the cell showed an unanticipated behavior in these simulations: after an initial peak, it decayed transiently and then recovered to go through a second phase of increase, followed by a second, typically higher, peak. Previous experimental studies had not reported this type of biphasic behavior of PIP<sub>3</sub> at the front of a cell [<xref rid="pcbi-0020082-b006" ref-type="bibr">6</xref>], so we sought to determine whether the results indicated a flaw in the model or whether the model parameters were “incorrect” and could be adjusted to make the dip in between the two peaks shallower, yet preserve the cells' quantitatively correct behavior upon exposure to homogeneous stimuli. Although for gradients with low (less than 10% of the <italic>K</italic> <sub>d</sub> for the binding of cAMP to the receptor) absolute cAMP concentrations we could find a parameter set yielding a monotonic response, this was not possible to achieve with any reasonable parameter set when the concentration of cAMP was increased.</p><p>We then examined in detail the time-dependent changes in PIP<sub>3</sub>, active PI3K, and membrane-bound as well as phosphorylated cytosolic PTEN in the different regions of the cell during a simulated gradient response, using a reasonable parameter set consistent with the accurate simulation of homogeneous field behavior. This analysis indicated that the biphasic PIP<sub>3</sub> kinetics arose from a difference between the speed of signal-induced negative regulation of PI3K activity and that of PTEN membrane reattachment (<xref ref-type="fig" rid="pcbi-0020082-g004">Figure 4</xref>). In accordance with previously published data [<xref rid="pcbi-0020082-b027" ref-type="bibr">27</xref>], the simulation showed PI3K activity to be downregulated very rapidly (within several seconds) after the onset of stimulation (local adaptation), while the translocation phase of PTEN took considerably longer, typically 1 min [<xref rid="pcbi-0020082-b007" ref-type="bibr">7</xref>]. Because the level of PIP<sub>3</sub> is determined by the balance between local PI3K activity and local availability of membrane-bound PTEN, the simulated cell's front experienced two phases of PIP<sub>3</sub> accumulation. During the first phase PI3K is highly activated and PTEN has just begun its dissociation from the membrane. This phase ends with the rapid decrease of PI3K activity due to the operation of the regulatory pathways noted above and a partial return of PTEN to the membrane. During the second phase, the imbalance in Gα-induced Src-like kinase activity between the front and the back of the cell leads to a gradual loss of membrane-bound PTEN in the front and accumulation of PTEN in the back. With a strongly decreased level of PTEN at the front of the cell, the residual activity of PI3K now can induce a second, slower increase of PIP<sub>3</sub> at the cell's anterior. <xref ref-type="supplementary-material" rid="pcbi-0020082-sg005">Figure S5</xref> illustrates this process that translates (even small) differences in receptor occupancy between front and back of the cell into steeper intracellular gradients of PIP<sub>3</sub> and PTEN (amplification).</p><p>Following the development of new experimental techniques for rapid exposure of <italic>Dictyostelium</italic> cells to a defined cAMP gradient [<xref rid="pcbi-0020082-b015" ref-type="bibr">15</xref>], we found that the predicted biphasic PIP<sub>3</sub> response at the front of the cell corresponded to the actual behavior of the gradient sensing system, as did the concentration dependence of the extent of the decline in PIP<sub>3</sub> levels after the initial peak (<xref ref-type="fig" rid="pcbi-0020082-g005">Figure 5</xref>). Furthermore, the simulation predicted that the second increase would happen later and take longer to reach its peak if the absolute concentration of the chemoattractant was increased while maintaining the relative concentration difference between front and back of the cell. We also found this prediction to be correct (<xref ref-type="fig" rid="pcbi-0020082-g005">Figure 5</xref>). Adjusting the model parameters so that the dip between the first and second peak of PIP<sub>3</sub> corresponded quantitatively to the experimentally determined time for a single chemoattractant concentration, the simulation correctly predicted the altered slope for the second increase and for the position in time of the second PIP<sub>3</sub> peak when using 10-fold higher or lower absolute concentrations of cAMP (<xref ref-type="fig" rid="pcbi-0020082-g005">Figure 5</xref>). The model was thus able to predict with great accuracy in both time and space the PIP<sub>3</sub> response of <italic>Dictyostelium</italic> cells over a several-log range of chemoattractant concentration.</p><fig id="pcbi-0020082-g005" position="float"><label>Figure 5</label><caption><title>Correspondence in Time and Space between the Predicted and Measured Changes in PIP<sub>3</sub> at the Front and Back of Cells Exposed cAMP Gradients</title><p>Experimental data from exposure of <italic>Dictyostelium</italic> to a 2:1 gradient with a mean cAMP concentration of 100 nmol were used to adjust model parameters. The other two responses are predictions of the model. (A), (B), and (C) are simulated responses. (D), (E), and (F) are experimental measurements, using PH-GFP to monitor PIP<sub>3</sub> levels in single cells exposed to gradients with a mean cAMP concentration of 1 μmol, 100 nmol, and 10 nmol. See <xref ref-type="supplementary-material" rid="pcbi-0020082-sg006">Figure S6</xref> for details on the full dataset of experimental replicates.</p></caption><graphic xlink:href="pcbi.0020082.g005"/></fig><p>To test further the predictive power of the model, we used cells expressing GFP-PTEN to examine whether the simulated spatiotemporal properties of PTEN relocation corresponded to those measured experimentally. At the cell's front, GFP-PTEN showed the triphasic behavior we expected, involving a decline, a temporary increase, and then a substantial decay. The local peak of PTEN occurred at the predicted time point and the ensuing decay correlated well with the second increase of PH<sub>crac</sub>-GFP at the front of an adjacent cell exposed to the same gradient. As predicted, with progressively lower concentrations of cAMP, the temporary increase of PTEN seen with a high-concentration stimulus first turned into a shallow plateau and then vanished (<xref ref-type="fig" rid="pcbi-0020082-g006">Figures 6</xref>, <xref ref-type="supplementary-material" rid="pcbi-0020082-sg006">S6</xref>, and <xref ref-type="supplementary-material" rid="pcbi-0020082-sg007">S7</xref>; see also <xref ref-type="supplementary-material" rid="pcbi-0020082-sd003">Text S3</xref>).</p><fig id="pcbi-0020082-g006" position="float"><label>Figure 6</label><caption><title>Correspondence in Time and Space between the Predicted and Measured Changes in Membrane-Bound PTEN at the Front and Back of the Cells Exposed to cAMP Gradients</title><p>These dose-dependent dynamics of PTEN were produced by the simulation after the dynamics of PIP<sub>3</sub> for one cAMP concentration (100 nM) had been used to adjust model parameters. (A), (B), and (C) are simulated responses. (D), (E), and (F) are experimental measurements based on GFP-PTEN analysis in single cells exposed to gradients with a mean cAMP concentration of 1 μmol, 100 nmol, and 10 nmol. See <xref ref-type="supplementary-material" rid="pcbi-0020082-sg007">Figure S7</xref> for details on the full dataset of experimental replicates.</p></caption><graphic xlink:href="pcbi.0020082.g006"/></fig></sec></sec><sec id="s3"><title>Discussion</title><p>Here we describe the construction and evaluation of a new detailed model of the chemotactic signaling response of <italic>Dictyostelium</italic>. We tested the model's explanatory and predictive capabilities in a close interplay between computer simulations and laboratory analysis using quantitative single-cell microscopy. The simulations correctly predicted the biochemical behavior of the cells under conditions that had never before been experimentally investigated in detail and led us to propose specific substantial changes to the existing model of chemosensory signaling in this organism, modifications that are needed to reconcile theory with experimental observations.</p><p>The model and the simulations reported here illustrate how signal-induced activation and membrane-recruitment of components controlling the availability and/or activity of Gβγ, Ras, PI3K, and PTEN can induce rapid adaptation of the level of PIP<sub>3</sub> and of the localization of PTEN following homogeneous stimuli with chemoattractant. Interestingly, these same mechanisms lead to a polarized distribution of PIP<sub>3</sub> in gradients of cAMP. In contrast to the predominant view in the field, our studies suggest that the intracellular amplification of the applied chemotactic gradient is the consequence of the local, rather than global, feedback-regulation of excitation and adaptation. Upon stimulation of the cell, the product (PIP<sub>3</sub>) of the activation of an excitatory component (PI3K) reinforces the processes leading to deactivation of an inhibitory component (PTEN), thereby allowing for a rapid, pronounced response. Following this first response, adaptation is achieved through a strong, locally controlled and signal-dependent suppression of the excitatory component which, in turn, decreases the deactivation of the inhibitory component, i.e., allows for reactivation of the inhibition. In a gradient, however, the higher concentration of PIP<sub>3</sub> at the side of the cell facing the higher chemoattractant concentration (“front”) supports the activation of more PTEN-deactivating components than at the opposite side (“back”) of the cell. This leads to a net transfer of PTEN from the front to the back, which in turn decreases the concentration of PIP<sub>3</sub> in the back, accelerating that side's transition into the adaptation phase. In our model, polarization of a cell in a chemoattractant gradient is thus caused by the (diffusive) communication between the sides of the cell with high or low chemoattractant concentration, respectively, interfering with the local adaptation processes.</p><p>The recruitment of Ras GAP through PIP<sub>3</sub> [<xref rid="pcbi-0020082-b022" ref-type="bibr">22</xref>] and the multiple functions of receptor-associated kinase, which in addition to phosphorylating the receptor binds to Gα and blocks Gβγ from acting on Ras and PI3K [<xref rid="pcbi-0020082-b023" ref-type="bibr">23</xref>], are well-documented negative feedback mechanisms in G protein–coupled receptor signal transduction. Our model suggests the action of two additional regulatory components: a protein phosphatase that deactivates PI3K and a lipid phosphatase (SHIP [SH2-containing inositol 5′-phosphatase]) that is recruited to the front of the cell to control the level of PIP<sub>3</sub> when the local concentration of active PTEN is low. These elements act on two different time scales. The phosphatase that deactivates PI3K is rapidly activated to control PIP<sub>3</sub> levels during early adaptation of the network. SHIP is more slowly recruited to those regions of the membrane that have lost PTEN. An abstract slow second inhibitory (“poisoning”) element had been previously suggested by Meinhardt [<xref rid="pcbi-0020082-b012" ref-type="bibr">12</xref>] to prevent the cells from losing their sensitivity to gradient changes and we can speculate that this hypothetical element corresponds to SHIP in our more specific scheme. Biochemical studies are now required to see if the various components we have incorporated into our extended biochemical model from mammalian studies have molecular analogs in <italic>Dictyostelium</italic> with the anticipated functions.</p><p>As noted before (and discussed in detail in <xref ref-type="supplementary-material" rid="pcbi-0020082-sd001">Text S1</xref>), we think that it is helpful to use modeling to speculate about the role of the different components in a signaling network not in terms of abstract modules but in terms of interacting molecules. Instead of postulating, e.g., a “PTEN-translocation-control-module,” we asked which molecules have been observed to play a role in the control of the membrane attachment of PTEN and how they exert this control. Simmune in particular allows the facile construction and simulation of such biochemically specific models because components represent concrete signaling molecules. This allows simulated responses to be analyzed in molecular detail and the predicted behavior of these molecules to be directly compared to results obtained by traditional laboratory methods such as immunoprecipitation, immunoblotting, flow cytometry, and confocal microscopy. As we discuss in <xref ref-type="supplementary-material" rid="pcbi-0020082-sd002">Text S2</xref>, previous, more abstract models with PTEN-control modules neither suggested concrete molecular equivalents of those modules nor were they able to generate the correct behavior of PTEN or PIP<sub>3</sub> in response to chemoreceptor engagement.</p><p>Not all aspects of <italic>Dictyostelium</italic> chemotactic behavior are fully captured in even the extended model we have implemented here. While the simulations arising from the present model were quite accurate in their spatiotemporal details under diverse conditions, it is very likely that details of the regulatory mechanisms included in the present model are incorrect or that important additional components are absent. An indication that this is the case comes from observations made while experimenting with models containing alternative pathway configurations or very different parameter sets (<xref ref-type="supplementary-material" rid="pcbi-0020082-sd002">see Text S2</xref>). In certain situations, residual PI3K inhibitors persisted at the former high-concentration side of the cell membrane once the stimulating gradient was removed. In these simulations, the persistence of these inhibitors then led to a temporarily inverted biochemical response during secondary exposure to a chemotactic gradient, a behavior that has recently been observed experimentally (XX, MMS, and TJ, manuscript in preparation). The specific model we describe here does not permit a single parameter set to give a simulation output that reproduces in detail this inversion behavior, while also being in full quantitative agreement with the primary polarization kinetics reported in this paper across a wide range of cAMP concentrations. These recent findings suggest that additional elements will need to be added to the circuits controlling local PIP<sub>3</sub> levels to generate a more robust molecular model.</p><p>This issue of persistence of inhibitory components at the leading edge after cAMP removal is closely related to the key question of what mechanisms keep a cell responsive to changes in the externally applied gradient. <italic>Dictyostelium</italic> cells that have fully differentiated into the chemotactically competent stage behave like neutrophils and turn their “body” instead of changing the orientation of their internal biochemical polarization when the extracellular gradient changes direction. During earlier differentiation stages, however, <italic>Dictyostelium</italic> reacts to such changes by reorienting its internal PIP<sub>3</sub>-dependent polarity (C. Parent, personal communication). Cells simulated with our current model react to changes in the direction of externally applied cAMP gradient in a manner similar to these immature amoeba, with a reorientation of the internal PIP<sub>3</sub> gradient. However, the simulated reorientation is only partial and occurs more slowly than that observed directly in live cells (XX, unpublished data), again pointing to the need for a reworking of the mechanisms that control the activity of PI3K at the leading edge of a cell exposed to a gradient.</p><p>Many experimental biologists see no immediate need for measuring in a highly quantitative manner either intracellular molecular concentrations, their stimulus-induced changes, or the corresponding reaction rates. One factor contributing to this viewpoint is the limited availability of user-friendly computational tools that would allow investigators to make direct use of such data for detailed simulations that could enhance their biological understanding. Simmune, the software we developed and used here to investigate the mechanisms behind <italic>Dictyostelium</italic>'s chemosensing capabilities, enables biologists lacking advanced mathematical or computer coding expertise to conduct the type of simulations that make the pursuit of quantitative experimental data a more useful undertaking. The software provides a simple interface for model building and parameter entry, provisions for automated construction of complex molecular interaction networks from the input of binary molecular interactions, visual output of the spatial distribution and number of specific molecules/molecular complexes in model cells during a simulation, and the capacity to view at varying levels of resolution the time-dependent changes in component concentrations or higher-order molecular complexes within a signaling pathway. It also permits creating models across varying scales of biological resolution, from intracellular molecular networks to individual cell behavior to the activity of groups of simulated cells.</p><p>Quantitative modeling is just beginning its foray into the cell biology of intracellular signaling, and measurements of the temporal behavior of the concentrations of most of the relevant molecules are still missing. We hope that Simmune will help encourage investigators in diverse fields of biological research to fill these gaps by providing them with an easily mastered yet powerful tool for translating biological knowledge and data into models and simulations that can enhance understanding of a complex biological system, as we show here for chemosensing in <italic>Dictyostelium</italic>.</p></sec><sec id="s4"><title>Materials and Methods</title><sec id="s4a"><title>Cell lines, single-cell microscopy.</title><p>Single-cell imaging of <italic>Dictyostelium</italic> for analysis of the activation of trimeric G proteins, relocation of PH-GFP in response to changing PIP<sub>3</sub> levels, and chemoattractant concentrations at the membrane was performed as described in [<xref rid="pcbi-0020082-b015" ref-type="bibr">15</xref>]. For the experiments with GFP-PTEN, cells from the <named-content content-type="genus-species">D. discoideum</named-content> cell line expressing GFP-PTEN [<xref rid="pcbi-0020082-b021" ref-type="bibr">21</xref>] were developed into the chemotactic stage, then exposed to cAMP and imaged as described in [<xref rid="pcbi-0020082-b015" ref-type="bibr">15</xref>]. The cells were treated with latrunculin prior to the experiments to suppress actin polymerization. This treatment immobilizes the cells, thereby facilitating the quantitative analysis of changes in the local concentration of PH-GFP. Furthermore, it eliminates any morphological prepolarization of the cells that otherwise might cause different sensitivities toward chemotactic stimulation at their different sides, and, finally, it decouples actin dynamics from the purely biochemical network we wished to study.</p></sec><sec id="s4b"><title>Simulation software.</title><p>We used the “Simmune” modeling and simulation software, as described in the text and supplementary material. Simmune runs on Linux, MacOS 10, and Windows XP. Copies of Simmune (executables) can be requested by sending an e-mail to <email>[email protected]</email>. Updated documentation and information about new releases and bug fixes are available through <ext-link ext-link-type="uri" xlink:href="http://www.simmune.org">http://www.simmune.org</ext-link>.</p></sec></sec><sec sec-type="supplementary-material" id="s5"><title>Supporting Information</title><supplementary-material content-type="local-data" id="pcbi-0020082-sg001"><label>Figure S1</label><caption><title>Automatically Constructed Visualization of the Enzymatic Reaction Network of the Chemotactic Signaling Model</title><p>These screenshots of the network browser show the enzymatic reaction network before (A) and after (B) stimulation with cAMP. (B) Shows the response at the “front” (high cAMP concentration) of the cell. Simmune uses the binary interaction possibilities defined by the user to automatically construct the permitted network topology and to generate a graphical view of the network with several user defined outputs available for the display (binding interactions, enzymatic reactions, and so on). Concentrations of enzymes and their (single-) molecular substrates are represented in the color saturation of the circles and ellipsoids representing the network components. The concentration of each component is displayed relative to its maximum concentration during a simulated experiment (darker = greater concentration). Similarly, the (relative) reaction flow for enzymatic transformations is encoded in the saturation level of the blue lines representing the biochemical transformations. A line with a circle represents the action of an enzyme on its substrate. A line with an arrow indicates the transformation of the substrate into the product of the reaction. <xref ref-type="supplementary-material" rid="pcbi-0020082-sv001">Video S1</xref> shows how concentrations and flows in the front of a cell (high cAMP concentration) change during the course of a stimulation of a model cell exposed to a gradient of cAMP. Because concentrations and reaction flows are encoded in the color saturation of the component symbols and their connecting lines, some lines are almost invisible in the snapshots of the network dynamics presented here.</p><p>(101 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sg001.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sg002"><label>Figure S2</label><caption><title>The Branch of the Signaling Network That Controls the Activity of PI3K</title><p>A line with a circle represents the action of an enzyme on its substrate. A line with an arrow indicates the transformation of the substrate into the product of the reaction. The green lines represent those passive binding possibilities that play important roles in regulation of PI3K activity.</p><p>Gβγ is assumed to recruit PI3K to the membrane where it then can interact with Ras and become fully activated [<xref rid="pcbi-0020082-b007" ref-type="bibr">7</xref>]. Ras itself becomes activated through a GEF that in turn also depends on Gβγ for its activation [<xref rid="pcbi-0020082-b037" ref-type="bibr">37</xref>]. In our model, we adopt this activation pathway for PI3K but allow for a direct activation of Ras through Gβγ. In addition, Gβγ activates RasGAP and the tyrosine phosphatase “PI3Ktp”. Active Ras (“Ras_act”) activates PI3K. Activated RasGAP (“RasGAPopen”) can attach itself to PIP<sub>2</sub> and PIP<sub>3</sub>, which places it in close proximity of its membrane-bound substrate, Ras_act. The activated phosphatase PI3Ktp_act deactivates PI3K. It can attach itself to the phosphorylated membrane-bound adaptor pPI3Ktp_anch (not shown) to localize near membrane-associated, active PI3Kact. The receptor-mediated stimulus leads to a local higher concentration of inhibitors of PI3K activity. Note: Unactivated enzymes have a (low) basal affinity for/rate of action on their substrates. Accordingly, even the unactivated enzyme states are linked to their substrates by lines with dots.</p><p>(31 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sg002.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sg003"><label>Figure S3</label><caption><title>The Branch of the Signaling Network That Controls the Activity of PTEN</title><p>A line with a circle represents the action of an enzyme on its substrate. A line with an arrow indicates the transformation of the substrate into the product of the reaction. The green lines represent those passive binding possibilities that play important roles for the regulation of PTEN activity.</p><p>PTEN is an interfacial enzyme, having a significantly higher phosphatase activity when positioned in close proximity to its membrane-embedded substrates like PIP<sub>3</sub> [<xref rid="pcbi-0020082-b038" ref-type="bibr">38</xref>]. Mammalian PTEN contains two domains that contribute to recruitment of the phosphatase to the membrane [<xref rid="pcbi-0020082-b039" ref-type="bibr">39</xref>]. Both domains can lose their membrane-binding ability upon phosphorylation of the phosphatase at two specific sites in the C-terminus [<xref rid="pcbi-0020082-b040" ref-type="bibr">40</xref>]. A recent report on the role and mechanism of membrane recruitment of <italic>Dictyostelium</italic> PTEN has shown that in this organism, the N-terminal region possesses a PIP<sub>2</sub> binding motif that is crucial for the membrane recruitment of PTEN and its physiological activity [<xref rid="pcbi-0020082-b041" ref-type="bibr">41</xref>]. Because PTEN leaves the membrane of <italic>Dictyostelium</italic> after stimulation of the cells with chemoattractant and because the relatively high membrane abundance of PIP<sub>2</sub> is not likely to be substantially decreased upon chemotactic stimulation, we assume in our model that the chemotactic signal leads to phosphorylation of PTEN that interferes with its membrane binding capabilities and its phosphatase activity. In mammalian cells, PTEN phosphorylation has been shown to be mediated by Src [<xref rid="pcbi-0020082-b024" ref-type="bibr">24</xref>], so we assume in our signaling network a similar role for a Src-like kinase that in turn is stimulated via active Gβγ. Activation of Gα leads to activation of “SrcAct,” the component that activates Src. Active Src (“Src_act”) phosphorylates PTEN. Phosphorylated PTEN (“pPTEN,” or membrane-bound: “pPTEN_bnd”) has lost its phospholipid phosphatase activity and rapidly dissociates from the membrane. Csk deactivates Src_act. Csk is brought into close proximity with membrane-bound Src through its capability to bind to phosphorylated paxillin (“pPaxillin”). SHP2 dephosphorylates pPaxillin, thereby reducing the amount of membrane-proximal Csk. SHP2 is recruited to the membrane by binding to Src-phosphorylated Gab1 (“pGab1”) [<xref rid="pcbi-0020082-b042" ref-type="bibr">42</xref>], which in turn binds to PIP<sub>3</sub>. The PIP<sub>3</sub>-dependent recruitment of SHP2 and SrcAct thus represents a positive feedback for the processes leading to phosphorylation (and deactivation) of PTEN; the more PTEN becomes deactivated, the more PIP<sub>3</sub> that can be produced by PI3K. Increasing concentrations of PIP<sub>3</sub> lead to recruitment of more SHP2 and SrcAct and result in faster deactivation of PTEN. This explains (in our model) the efficiency of the stimulus-induced translocation of PTEN from the cell's front to its back.</p><p>(49 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sg003.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sg004"><label>Figure S4</label><caption><title>Predicted and Measured Dose Dependence of the Time T<sub>max</sub> until the PIP<sub>3</sub> Accumulation Reaches Its Maximum after Homogeneous Stimulation with cAMP</title><p>With increasing concentrations of homogeneously applied cAMP, the cells adapt more rapidly to the stimulus.</p><p>(10 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sg004.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sg005"><label>Figure S5</label><caption><title>Local Adaptation Processes, When Combined with Diffusive Communication between the Different Sides of the Cell, Can Explain the Intracellular Amplification of the External cAMP Gradient</title><p>The primary signal (ligation of the receptor, activation of the G proteins) is shown in light blue. Excitatory processes are shown in orange, inhibitory processes in gray. The processes that are part of the feedback mechanism establishing the intracellular polarization are shown in yellow.</p><p>(81 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sg005.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sg006"><label>Figure S6</label><caption><title>Global Dataset of Changes in Membrane-Bound PH-GFP in Cells Exposed to 2:1 Gradients of cAMP</title><p>Mean cAMP concentrations at the cell surface: (A) 1 μmol (15 datasets); (B) 100 nmol (18 datasets); (C) 10 nmol (25 datasets).</p><p>Due to the considerable variation in the responses of single cells, the average behavior of membrane-bound PH-GFP shows less pronounced “dips” in the cell front after the first peak than single-cell responses. For the comparison with simulation results, we chose single-cell responses that have their characteristic features (minima and maxima) at approximately the same time points as the multi-cell averages.</p><p>(25 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sg006.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sg007"><label>Figure S7</label><caption><title>Global Dataset of Changes in Membrane-Bound GFP-PTEN in Cells Exposed to 2:1 Gradients of cAMP</title><p>Mean cAMP concentrations at cell surface: (A) 1 μmol (four datasets); (B) 100 nmol (12 datasets); (C) 10 nmol (six datasets).</p><p>Due to the considerable variation in the responses of single cells, the average behavior of membrane-bound GFP-PTEN shows less pronounced minima in the front of the cell than single-cell responses. For the comparison with simulation results, we chose single-cell responses that have their characteristic features (minima and maxima) at approximately the same time points as the multicell averages.</p><p>(25 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sg007.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sg008"><label>Figure S8</label><caption><title>The Influence of Parameter Variation on the Signaling Dynamics</title><p>This screenshot of Simmune shows the results of simulated exposure to a 2:1 cAMP gradient (100 nmol mean concentration) with an automated variation of the parameters determining the number of Csk molecules per cell and the association rate for the activation by Gβγ of the phosphatase (PTENph) that dephosphorylates pPTEN. Csk varies (vertical axis) from 80,000 (upper row) to 160,000 (lower row) molecules per cell. The PTENph-Gβγ association rate varies (horizontal axis) from 50,000 1/(mols) (left column) to 500,000 1/(mols) (right column). While for all parameter sets the overall behaviors of membrane-bound GFP-PH and PTEN show the characteristic minima and maxima, their positions in terms of time points and molecule amounts change significantly with varying parameters. While the qualitative features are thus “robust” to the examined range of parameter values, the quantitative details are sensitive to these numerical changes. With parameter set (1), for example, the second peak of the accumulation of GFP-PH at the high-concentration side of the applied cAMP gradient (red) does not reach the experimentally observed value. On the other hand, the increase in PTEN at the low-concentration side of the cell (blue) for this set of parameters is too strong. Set (3), with a high PTENph-Gβγ association rate and a high concentration of Csk, reaches the second peak in the accumulation of GFP-PH at the side with high cAMP concentration too late. The dynamics obtained with set (2) (which is the parameter set chosen for all simulations reported here) correspond well to the experimentally observed behavior of membrane-bound GFP-PH and PTEN.</p><p>A comparison of the simulated behavior of membrane-bound PTEN after homogeneous stimulation with 100 nmol cAMP: the curves (1), (2), and (3) were obtained with the same parameter sets as the correspondingly labeled simulation results in (A). Set (2), which shows the correct behavior under homogeneous stimulation, also produces quantitatively and qualitatively correct results for stimulation with a cAMP gradient [see (A)]. See also <xref ref-type="supplementary-material" rid="pcbi-0020082-sv007">Video 7</xref> or <xref ref-type="supplementary-material" rid="pcbi-0020082-sd013">Text S13</xref>.</p><p>(138 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sg008.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sg009"><label>Figure S9</label><caption><title>Modifications of PI3K Membrane Attachment Result in Changed Behavior of PTEN</title><p>Stimulation with a homogeneous concentration of 100 nmol of cAMP leads to a transient loss of membrane-bound PTEN. For this dose, there is an 80% recovery of membrane-bound PTEN within 40 s. The dashed curve shows the simulated behavior of membrane-bound PTEN for a cell with membrane attachment of not only activated but also nonactivated PI3K with basal enzymatic activity. In agreement with previously reported data [<xref rid="pcbi-0020082-b007" ref-type="bibr">7</xref>], PTEN shows a slower, only partial return to the membrane.</p><p>(11 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sg009.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sg010"><label>Figure S10</label><caption><title>Behavior of PTEN with an Alternative Mechanism Regulating PTEN Membrane Attachment</title><p>See <xref ref-type="supplementary-material" rid="pcbi-0020082-sd002">Text S2</xref>, Section 3 for details.</p><p>(A) Loss and recovery of membrane-bound PTEN after stimulation with a homogeneous concentration of 400 nmol (solid line) and 40 nmol (dashed line). For both concentrations, PTEN shows a qualitatively correct behavior.</p><p>(B) Translocation of PTEN after stimulation with a linear gradient of 40 nmol in the “front” and 10 nmol in the “back” of the cell. The applied gradient leads to a pronounced accumulation of PTEN in the back and loss of PTEN in the front.</p><p>(C) Translocation of PTEN after stimulation with a linear gradient of 400 nmol in the “front” and 100 nmol in the “back” of the cell. The mean concentration of the applied gradient is closer to the saturation dose of the cAMP receptor. The receptor signal strengths in front and back are not strongly different. Due to the lack of an amplifying element, the applied gradient fails to induce a strong polarization of membrane-bound PTEN across the cell diameter.</p><p>(22 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sg010.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sg011"><label>Figure S11</label><caption><title>Mechanisms of Cellular Behavior in Simmune (Composite Screenshot)</title><p>Cells created in Simmune contain specified numbers of molecules and molecular complexes in their cytosol and on their membrane, based on the definition of their biochemistry by the modeler. The molecules will then diffuse and react according to their diffusion coefficients and reaction rates. Such intracellular reaction-diffusion networks determine the low-level behavior of the cells. Simmune offers a variety of possibilities for inspection of the intracellular biochemistry of simulated cells (see <xref ref-type="fig" rid="pcbi-0020082-g004">Figures 4</xref> and <xref ref-type="supplementary-material" rid="pcbi-0020082-sg001">S1</xref>). The higher-level behavior of cells in Simmune is defined by the <italic><bold>stimulus-response mechanisms</bold></italic> they possess (see <xref ref-type="supplementary-material" rid="pcbi-0020082-sd004">Text S4</xref>). In the example illustrated here, the stimulus consists of the ligation of the receptor for the green molecules. A second condition used in the simulation is the absence of a suprathreshold amount of ligated receptors for the red molecules. If both conditions are fulfilled, the cell secretes the blue molecules. In the simulation screenshot which shows a two-dimensional cut through a three-dimensional extracellular compartment, it can be seen that only cells within regions of high concentration of the green molecules and relatively low concentrations of the red molecules secrete the blue molecules. The cells in the region indicated by the white circle sense a sufficiently high concentration of green molecules but also a high concentration of red molecules. They do not respond.</p><p>(240 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sg011.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sg012"><label>Figure S12</label><caption><title>The “Complex Definition” Window of Simmune (Composite Screenshot)</title><p>Molecule complexes consist of molecules that are connected through interactions between their binding sites. During a simulation, the program creates representations for all possible molecule complexes (to an upper limit in the number of molecular components that can be set by the user), based on the binding interactions defined for the molecules in the model. The modeler does not have to define each complex by hand.</p><p>Wherever the local biochemistry allows the formation of a complex in some part of the simulated system, this complex will be taken into account for the simulation of the molecular reactions. Importantly, the set of (partial) differential equations that are processed during a simulation only includes equations for such complexes that are actually part of the local biochemistry. For example, the differential equations describing the reactions and the diffusional exchange of molecules for a cytosolic region of the cell never include components that describe the reaction dynamics of plasma membrane receptors. This minimizes computational cost.</p><p>Specific molecule complexes can be created for use in the definition of cellular mechanisms (see <xref ref-type="supplementary-material" rid="pcbi-0020082-sd004">Text S4</xref>) or for the definition of enzymatic transformations (see Simmune Tutorial 3 in <xref ref-type="supplementary-material" rid="pcbi-0020082-sv004">Video S4</xref>, or <xref ref-type="supplementary-material" rid="pcbi-0020082-sd011">Text S11</xref>). If the local concentrations of particular molecule complexes are to be tracked during a simulation, these complexes have to be defined first.</p><p>The structures in the left list of defined molecules within the “complexes” window can be used as building blocks to create molecular complexes. Molecule names from the list can be dragged and dropped into the white area of the “complex definition” part of the window. The light gray squares in the area indicate different possible positions of the molecules within a complex. Depending on the properties of the molecule (whether it is a receptor or cytosolic molecule) and the presence of other molecules that have already been put into the complex definition field, some positions within the field are possible/allowed while others are not. Simmune indicates the possible positions by turning the allowed squares green while a molecule is being dragged into the complex definition area. Once a molecule has been dropped into an allowed square, the program will indicate all possible binding interactions with the other molecules within the “complex definition” field. Double-clicking the small gray square attached to the potential connection between two molecular binding sites establishes the connection. The binding sites involved in the connection will snap together (see inset). Once all the connections for the complex the modeler wishes to build have been defined, a name can be specified and the complex saved. If the molecules are not properly connected or if the complex is isomorphic to an already existing complex, Simmune will issue an error message.</p><p>(55 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sg012.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sg013"><label>Figure S13</label><caption><title>Comparison of Simmune's Diffusion with Analytical Solution</title><p>See <xref ref-type="supplementary-material" rid="pcbi-0020082-sd007">Text S7</xref> for details.</p><p>(55 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sg013.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-st001"><label>Table S1</label><caption><title>Model Parameters</title><p>(48 KB XLS)</p></caption><media xlink:href="pcbi.0020082.st001.xls"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sd001"><label>Text S1</label><caption><title>Rationale for Model Building</title><p>(21 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sd001.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sd002"><label>Text S2</label><caption><title>Modular Analysis of the Chemosensing Signaling Network</title><p>(32 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sd002.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sd003"><label>Text S3</label><caption><title>Note on the Determination of Parameter Values for the Chemosensing Model</title><p>(16 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sd003.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sd004"><label>Text S4</label><caption><title>From Molecular Interactions to Cellular Behavior—How Simmune Works</title><p>(32 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sd004.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sd005"><label>Text S5</label><caption><title>Simmune's Internal Representation of Molecular Complexes</title><p>(13 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sd005.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sd006"><label>Text S6</label><caption><title>Automatic Creation of the Reaction Network</title><p>(23 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sd006.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sd007"><label>Text S7</label><caption><title>Discretization of Extracellular and Intracellular Space in Simmune</title><p>(35 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sd007.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sd008"><label>Text S8</label><caption><title>Simmune Tutorial Part 1</title><p>Defining molecules, binding and debinding events, and allosteric molecular modifications.</p><p>(310 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sd008.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sd009"><label>Text S9</label><caption><title>Simmune Tutorial Part 2</title><p>Defining transmembrane receptors and transmembrane signaling events.</p><p>(217 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sd009.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sd010"><label>Text S10</label><caption><title>Simmune Tutorial Part 3</title><p>Defining specific molecular complexes and enzymatic transformations.</p><p>(733 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sd010.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sd011"><label>Text S11</label><caption><title>Simmune Tutorial Part 4</title><p>Defining cells and extracellular space; running simulations.</p><p>(1.7 MB PDF)</p></caption><media xlink:href="pcbi.0020082.sd011.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sd012"><label>Text S12</label><caption><title>Simmune Tutorial Part 5</title><p>Simulating the Dictyostelium chemosensing model.</p><p>(513 KB PDF)</p></caption><media xlink:href="pcbi.0020082.sd012.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sd013"><label>Text S13</label><caption><title>Simmune Tutorial Part 6</title><p>Automated parameter variation for the chemosensing model.</p><p>(1.0 MB PDF)</p></caption><media xlink:href="pcbi.0020082.sd013.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sd014"><label>Text S14</label><caption><title>Simmune Tutorial Part 7</title><p>Building and simulating a simple model of cellular gradient sensing.</p><p>(2.3 MB PDF)</p></caption><media xlink:href="pcbi.0020082.sd014.pdf"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sv001"><label>Video S1</label><caption><title>Simmune's Display of Signaling Dynamics within the Reaction Network</title><p>(3.8 MB MOV)</p></caption><media xlink:href="pcbi.0020082.sv001.mov"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sv002"><label>Video S2</label><caption><title>Simmune Tutorial Part 1</title><p>Defining molecules, binding and debinding events, and allosteric molecular modifications.</p><p>(2.8 MB MOV)</p></caption><media xlink:href="pcbi.0020082.sv002.mov"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sv003"><label>Video S3</label><caption><title>Simmune Tutorial Part 2</title><p>Defining transmembrane receptors and transmembrane signaling events.</p><p>(2.3 MB MOV)</p></caption><media xlink:href="pcbi.0020082.sv003.mov"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sv004"><label>Video S4</label><caption><title>Simmune Tutorial Part 3</title><p>Defining specific molecular complexes and enzymatic transformations.</p><p>(5.8 MB MOV)</p></caption><media xlink:href="pcbi.0020082.sv004.mov"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sv005"><label>Video S5</label><caption><title>Simmune Tutorial Part 4</title><p>Defining cells and extracellular space; running simulations.</p><p>(6.8 MB MOV)</p></caption><media xlink:href="pcbi.0020082.sv005.mov"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sv006"><label>Video S6</label><caption><title>Simulating the Dictyostelium Chemosensing Model (Tutorial Part 5).</title><p>(6.9 MB MOV)</p></caption><media xlink:href="pcbi.0020082.sv006.mov"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material><supplementary-material content-type="local-data" id="pcbi-0020082-sv007"><label>Video S7</label><caption><title>Automated Parameter Variation for the Chemosensing Model (Tutorial Part 6).</title><p>(6.9 MB MOV)</p></caption><media xlink:href="pcbi.0020082.sv007.mov"><caption><p>Click here for additional data file.</p></caption></media></supplementary-material></sec>
Children’s IQs: Trasande et al. Respond	Could not extract abstract	<contrib contrib-type="author"><name><surname>Trasande</surname><given-names>Leonardo</given-names></name></contrib><contrib contrib-type="author"><name><surname>Landrigan</surname><given-names>Phillip J.</given-names></name></contrib><aff id="af1-ehp0114-a00400">Center for Children’s Health and the Environment, Department of Community Medicine, Mount Sinai School of Medicine, New York, New York, E-mail: <email>[email protected]</email></aff>	Environmental Health Perspectives	<p>Schwartz makes a number of claims regarding our methodology (<xref rid="b14-ehp0114-a00400" ref-type="bibr">Trasande et al. 2005</xref>) that are inaccurate and based on a selective reading of the literature.</p><p>In our article (<xref rid="b14-ehp0114-a00400" ref-type="bibr">Trasande et al. 2005</xref>), we estimated the health and economic consequences of prenatal methylmercury (MeHg) exposure in the 2000 U.S. birth cohort. Our major findings were that at least 316,588 children in that birth cohort suffered IQ (intelligence quotient) loss of 0.2–24.4 points as a result of MeHg toxicity sustained <italic>in utero</italic>. This loss of intelligence causes diminished economic productivity that will persist, and this lost productivity is the major monetary consequence of methylmercury toxicity. We used the most up-to-date publicly available data on mercury exposures and health outcomes, applied a risk assessment approach developed by the National Research Council (<xref rid="b12-ehp0114-a00400" ref-type="bibr">NRC 1994</xref>), and made conservative assumptions throughout.</p><p>To compute decrements in IQ that resulted from prenatal mercury exposures, we used data from <xref rid="b8-ehp0114-a00400" ref-type="bibr">Mahaffey et al. (2004)</xref> on percentages of women of childbearing age in 1999–2000 with mercury concentrations ≥ 3.5, 4.84, 5.8, 7.13, and 15.0 μg/L. These data most closely reflect exposure to women in the years 1999–2000, when toxicity to the developing brains of children in the 2000 birth cohort would have occurred. We then applied logarithmic and linear models to these data, and we calculated a range of IQ decrements for each subpopulation born with a cord blood mercury concentration > 5.8 μg/L. To assess a range of possible outcomes, we conducted a sensitivity analysis in which we applied a range of IQ decrements for each increase in mercury concentration. We described our methods in great detail (<xref rid="b14-ehp0114-a00400" ref-type="bibr">Trasande et al. 2005</xref>). Through this series of calculations, we generated upper and lower ranges of possible IQ decrements for each subpopulation among the most highly exposed children in the 2000 U.S. birth cohort.</p><p>In his letter, Schwartz asserts that it is impossible to impute effects on children’s intelligence of prenatal exposures to mercury near the U.S. Environmental Protection Agency’s (EPA) reference dose (RfD). In proffering this assertion, he appears to ignore a recent meta-analysis of the three studies that confirmed a dose–response relationship between low-level prenatal MeHg exposure and IQ (<xref rid="b2-ehp0114-a00400" ref-type="bibr">Cohen et al. 2005</xref>). A recent U.S. cohort study has also detected decrements in visual recognition memory among children exposed prenatally to MeHg (<xref rid="b11-ehp0114-a00400" ref-type="bibr">Oken et al. 2005</xref>).</p><p>Schwartz suggests that we should have used the U.S. EPA benchmark dose level (BMDL) of 58 μg/L as a cutoff. He apparently assumes that no injury occurs to fetal brains from exposure to MeHg below that level. That approach does not reflect biologic or epidemiologic reality. We based our selection of 5.8 μg/L as a no adverse effect level on the epidemiologic evidence, not on the U.S. EPA’s regulatory documents (<xref rid="b1-ehp0114-a00400" ref-type="bibr">Budtz-Jorgensen et al. 2004</xref>; <xref rid="b4-ehp0114-a00400" ref-type="bibr">Grandjean et al. 1999</xref>; <xref rid="b5-ehp0114-a00400" ref-type="bibr">Kjellstrom et al. 1986</xref> Kjellstrom et al. 1989). We relied especially upon the NRC’s report on prenatal exposure to MeHg (<xref rid="b13-ehp0114-a00400" ref-type="bibr">NRC 2000</xref>), which concluded that the likelihood of subnormal scores on neuro-developmental tests increased as cord blood mercury concentrations increased from levels as low as 5 μg/L. Methylmercury exposure has also been associated with persistent delays in peak I–III brainstem-evoked potentials at cord blood levels < 5 μg/L (<xref rid="b10-ehp0114-a00400" ref-type="bibr">Murata et al. 2004</xref>).</p><p>Schwartz misrepresents <xref rid="b3-ehp0114-a00400" ref-type="bibr">Crump et al.’s findings (1998)</xref>, stating that they “superseded previous reports and found no IQ reduction.” In fact, the <xref rid="b13-ehp0114-a00400" ref-type="bibr">NRC (2000)</xref> stated that Crump et al.</p><disp-quote><p>reported nonsignificant results from a regression analysis on all the children in the New Zealand cohort, but [that these results became significant] after omission of a single child whose mother’s hair Hg concentration was 86 ppm (4 times higher than that of the next highest exposure level in the study).</p></disp-quote><p>Schwartz misrepresents our characterization of the Seychelles Islands study (<xref rid="b7-ehp0114-a00400" ref-type="bibr">Landrigan and Goldman 2003</xref>; <xref rid="b9-ehp0114-a00400" ref-type="bibr">Myers et al. 2003</xref>), accusing us of stating that it had half the statistical power of the Faroe Islands study (<xref rid="b4-ehp0114-a00400" ref-type="bibr">Grandjean et al. 1999</xref>). In actuality, we stated that the Seychelles study “had only 50% statistical power to detect the effects observed in the Faroes” (<xref rid="b14-ehp0114-a00400" ref-type="bibr">Trasande et al. 2005</xref>). Schwartz asserts that the NRC’s choice not to apply the Seychelles data in setting an RfD represents equivocation about the health effects of MeHg. In actuality, the NRC came to the same conclusion as we did: “[t]he weight of the evidence of developmental neurotoxic effects from exposure to MeHg is strong” (<xref rid="b13-ehp0114-a00400" ref-type="bibr">NRC 2000</xref>).</p><p>Recent work (<xref rid="b15-ehp0114-a00400" ref-type="bibr">Trasande et al. 2006</xref>) suggests that our calculation of the economic costs (<xref rid="b14-ehp0114-a00400" ref-type="bibr">Trasande et al. 2005</xref>) may, in fact, be an underestimate. The new study indicates that downward shifts in IQ are also associated with thousands of excess cases of mental retardation (defined as IQ < 70) in the United States each year. Care of these children is associated with needs for health care, special education, and other services that impose a great burden on society.</p><p>All of these adverse consequences can be prevented by prevention of prenatal exposure to MeHg.</p>
Erratum	Could not extract abstract	Could not extract contributor	Environmental Health Perspectives	<p>A line of text was inadvertently omitted from the June 2006 Innovations article (“Plant vs. Pathogen: Enlisting Tobacco in the Fight Against Anthrax,” <related-article id="d35e47" related-article-type="corrected-article" xlink:href="16759974" ext-link-type="pubmed" vol="114" page="A364"><italic>EHP</italic> 114:A364–A367 [2006]</related-article>). The last sentence on page A365 should read: “The current anthrax vaccine works on this very principle by introducing nonvirulent PA into the body so antibodies are created.” <italic>EHP</italic> regrets the error.</p>
Causality and the Interpretation of Epidemiologic Evidence	<p>There is an ongoing debate regarding how and when an agent’s or determinant’s impact can be interpreted as causation with respect to some target disease. The so-called criteria of causation, originating from the seminal work of Sir Austin Bradford Hill and Mervyn Susser, are often schematically applied disregarding the fact that they were meant neither as criteria nor as a checklist for attributing to a hazard the potential of disease causation. Furthermore, there is a tendency to misinterpret the lack of evidence for causation as evidence for lack of a causal relation. There are no criteria in the strict sense for the assessment of evidence concerning an agent’s or determinant’s propensity to cause a disease, nor are there criteria to dismiss the notion of causation. Rather, there is a discursive process of conjecture and refutation. In this commentary, I propose a dialogue approach for the assessment of an agent or determinant. Starting from epidemiologic evidence, four issues need to be addressed: temporal relation, association, environmental equivalence, and population equivalence. If there are no valid counterarguments, a factor is attributed the potential of disease causation. More often than not, there will be insufficient evidence from epidemiologic studies. In these cases, other evidence can be used instead that increases or decreases confidence in a factor being causally related to a disease. Even though every verdict of causation is provisional, action must not be postponed until better evidence is available if our present knowledge appears to demand immediate measures for health protection.</p>	<contrib contrib-type="author"><name><surname>Kundi</surname><given-names>Michael</given-names></name></contrib><aff id="af1-ehp0114-000969">Institute of Environmental Health, Center for Public Health, Medical University of Vienna, Austria</aff>	Environmental Health Perspectives	<p>The principle of causality, so deeply embedded in humans’ minds that it has been thought of as immediately evident, is the very foundation not only of all three monotheistic world religions but also of the first staggering steps of science [<italic>de nihilo nihil</italic> (nothing can be born of nothing); <xref rid="b16-ehp0114-000969" ref-type="bibr">Lucretius 1951</xref>]. <xref rid="b11-ehp0114-000969" ref-type="bibr">Hume (1739)</xref> was the first to note that there is no logical foundation in the assumption that if in the past every event has had a cause, this will also be the case in the future and, furthermore, that what we perceive in daily life as well as in science is only a sequence of events but not cause and effect. Although Hume deeply believed in the truth of the principle of causality, he pointed to the role of the human mind in constructing reality and the futility of scientifically proving its validity. Kant (1791), as he became acquainted with Hume’s thoughts, was awakened from his metaphysical slumber, or so he kept saying, and set out to solve the problem of how Newton’s physics, which he thought of as eternally true, could be possible in the face of Hume’s demonstration that it cannot be inferred from experience. The Copernican turn in Kant’s reasoning was to imply the principle of causality from the assumption that it is among the conditions of every experience. Indeed, if A is a necessary condition of B, then B is a sufficient condition of A. Hence, if for every experience we make (B) it is a precondition that everything has a cause (A), then from the fact that we do have experiences (B), it follows that everything has a cause (A). However, to make this a logically coherent theory, Kant had to sacrifice “objective knowledge”—that is, the <italic>Ding an sich</italic> (the “thing in itself”) remains incomprehensible for the human mind. For more than 100 years, the philosophy of science circled around either the assumptions or the (untoward) consequences of Kant’s solution. When in 1905 Einstein published his special theory of relativity and his theory of the interaction of electrons and light (<xref rid="b6-ehp0114-000969" ref-type="bibr">Einstein 1905a</xref>, <xref rid="b7-ehp0114-000969" ref-type="bibr">1905b</xref>), the very foundation of Kant’s philosophy was called into question: the universal truth of Newton’s mechanics (<xref rid="b19-ehp0114-000969" ref-type="bibr">Newton 1726</xref>) and the validity of the deterministic concept. These considerations not only profoundly changed modern science but also resulted in an open-ended controversy within epistemology. And last but not least, epidemiology and the interpretation of epidemiologic evidence are deeply connected to these fundamental considerations about the nature of human knowledge.</p><sec><title>Defining Cause and Causality</title><p>The most advanced sciences, physics and chemistry, have altogether abandoned the concepts of cause and effect. These terms are no longer used in these sciences. Newton had already replaced cause and effect with functional relationships; however, to make himself understood to his contemporaries, in the third book of his <italic>Principia</italic> (1726) he spoke about causes (especially to defend his position of what can be called a minimal sufficient cause). Nevertheless, “cause and effect” remained terms used in physics, somewhat anachronistically, especially for scholarly purposes until the end of the 19th century. Mach (1883), alluding to Hume, stressed the psychological nature of these concepts and pointed out that “in nature there is no cause and no effect” and that these concepts are results of an economical processing of perceptions by the human mind.</p><p>The notion that diseases have natural causes and are not God’s punishments or trials or curses of malicious beings or results of supernatural forces has not even fully penetrated Western culture, let alone become the prevailing view worldwide. Despite its metaphysical character, the etiologic axiom that every disease has an endogenous and/or exogenous cause was extremely successful and is still the foundation of scientific medicine. However, what actually “causes” a disease has from the very beginning been a matter of controversy. Indeed, a single clinical phenomenon can have quite different “causes,” and one “cause” can have quite different clinical consequences (<xref ref-type="table" rid="t1-ehp0114-000969">Table 1</xref>). These facts are not consistent with the original concept of causation, which states that a cause is an object that is followed by another, and where all objects similar to the first are followed by objects similar to the second (<xref rid="b11-ehp0114-000969" ref-type="bibr">Hume 1739</xref>). Not even for infectious diseases does this (strong) concept of causation hold. (Hume gave several “definitions” of a cause, among these also what has been called the counter-factual approach, discussed below.)</p><p>How, then, should cause and causation be defined? In a review of definitions of “causation” in epidemiologic literature, <xref rid="b20-ehp0114-000969" ref-type="bibr">Parascandola and Weed (2001)</xref> delineated five categories. However, all of these definitions (summarized in <xref ref-type="table" rid="t1-ehp0114-000969">Table 1</xref>) have severe deficits. Not totally unexpected, the definitions found in the literature are insufficient to provide a basis for the notion of disease causation. As pointed out above for physical phenomena, it is also impossible for disease processes to draw an ontologic demarcation within the indefinite stream of events between causal and noncausal associations.</p><p>Consider a human being as a complex input–output system that is described by a path through a state space (of likely very high dimensionality) that may or may not explicitly depend on time. The task is to solve the equations that relate the input stream, the output stream, and the internal states to each other. The solution could give the probability that the human being will be in some internal state of disease at some point in time given a set of initial and/or side conditions. If we were in possession of such a tool, we would not need the crutch of a concept of causation. Meanwhile, in a pragmatic sense, it is reasonable to stay with this concept but hold in mind that it is just an economical way to organize the otherwise unfathomable stream of events and to take the necessary steps to counteract or prevent the disease process. The process of diagnosis itself is one of abstraction and generalization because no two diseased human beings given the same diagnosis have exactly the same features.</p><p>In this pragmatic sense, disease cause can be defined as follows: Given two or more populations of subjects that are sufficiently similar for the problem under study, a disease cause is a set of mutually exclusive conditions by which these populations differ that increase the probability of the disease. In some cases, the similarity must be high, such that only homozygous twins can be studied; in other cases, maybe only sex and age must be considered, or the state of immunity. To avoid encumbering the definition with unnecessary complexity, we use the term “conditions” and the active verb “increase.” What is meant is that a number of extrinsic and/or intrinsic factors (i.e., conditions) can be discerned that are present before diagnosis of the disease and that prevail at a time and for a duration that is compatible with what is known about the natural history of the disease. Hence, this temporal relation is a precondition for an agent to be considered a causal factor. The “conditions” must be mutually exclusive (e.g., groups of males characterized by one of the following conditions: smoking or having smoked cigarettes, cigars, pipes only, more than one of these, or none), because otherwise the increase in the probability of the disease cannot be uniquely related to any one of them.</p><p>This definition is in line with the main designs of epidemiologic studies: the cohort, the case–control, and the randomized controlled trial. It is also in line with the pragmatic definition that assessment of causality affords more than just the observation of an increased incidence or prevalence in some group or the other. This is the point from which Sir Austin Bradford Hill started his considerations that led to what are now commonly called the “<xref rid="b3-ehp0114-000969" ref-type="bibr">Bradford Hill criteria” (1965)</xref>.</p></sec><sec><title>Taking Refuge in Causality</title><p>It seems that the first time causality entered the discussion on epidemiologic results was during the tobacco controversy in the late 1950s and early 1960s. In particular, the criticism of <xref rid="b8-ehp0114-000969" ref-type="bibr">Fisher (1959)</xref> concerning the conclusions drawn from the British Doctors Study by <xref rid="b5-ehp0114-000969" ref-type="bibr">Doll and Bradford Hill (1954)</xref> initiated a detailed consideration of the concept of causality that led to the famous presidential address by Bradford Hill to the Section of Occupational Medicine of the Royal Society of Medicine in 1965. In this talk, Bradford Hill discussed nine issues that should be addressed when deciding whether an observed association is a causal relationship. These issues, now called the “Bradford Hill criteria”—although they were not intended as criteria and not all of them have stood the test of time—are still the starting point of many a treatise on the subject today.</p><p>The Bradford Hill criteria were established such that, in the case they are met for a specific factor, this would increase our confidence in this factor being causally related to the disease. However, they were not intended to dismiss a factor as potentially causing the disease: “None of my nine viewpoints can bring indisputable evidence for or against the cause-and-effect hypothesis and none can be required as a <italic>sine qua non</italic>” (<xref rid="b3-ehp0114-000969" ref-type="bibr">Bradford Hill 1965</xref>).</p><p>Some statements in the past few years about the relationship between environmental or occupational factors and human health have used the terms “causality” or “causal” in a negative sense—that is, claiming that there is no evidence for a causal relationship. First, one has to discriminate between evidence for no causal relationship, and no evidence of a causal relationship (<xref rid="b2-ehp0114-000969" ref-type="bibr">Altman and Bland 1995</xref>). The former expresses an important piece of evidence that may have substantial consequences on steps taken to prevent health hazards, whereas the latter simply expresses lack of knowledge. It is, however, often misunderstood as an exculpation of the agent in question and is readily misused by interested parties to claim that exposure is not associated with adverse health effects.</p><p>Some examples of such statements illustrate the point:</p><list list-type="bullet"><list-item><p>A “formal causation analysis based on an application of the Hill criteria confirms that there is no causal relationship between diesel exhaust and multiple myeloma” (<xref rid="b26-ehp0114-000969" ref-type="bibr">Wong 2003</xref>).</p></list-item><list-item><p>“Applying a weight-of-evidence evaluation to the PCB [polychlorinated biphenyl] epidemiologic studies can only lead to the conclusion that there is no causal relationship between PCB exposure and any form of cancer” (<xref rid="b9-ehp0114-000969" ref-type="bibr">Golden et al. 2003</xref>).</p></list-item><list-item><p>“Results of these studies to date give no consistent or convincing evidence of a causal relation between RF [radiofrequency] exposure and any adverse health effect” (<xref rid="b1-ehp0114-000969" ref-type="bibr">Ahlbom et al. 2004</xref>).</p></list-item></list><p>There are significant differences between these statements. The last one claims that there is no “consistent or convincing evidence” (whatever this may be) of a causal relation. Hence, it points mainly to the lack of knowledge accumulated so far. The second one goes a step further: It claims that risk assessment based on the weight-of-evidence approach [as applied by the U.S. Environmental Protection Agency (<xref rid="b25-ehp0114-000969" ref-type="bibr">U.S. EPA 1999</xref>) or the International Agency for Research on Cancer (<xref rid="b12-ehp0114-000969" ref-type="bibr">IARC 2004</xref>)] leads to the conclusion of no causal relationship. However, there is no category of this type in the weight-of-evidence approaches. Either the category “not likely carcinogenic to humans” (<xref rid="b25-ehp0114-000969" ref-type="bibr">U.S. EPA 1999</xref>) or “evidence suggesting lack of carcinogenicity” (<xref rid="b12-ehp0114-000969" ref-type="bibr">IARC 2004</xref>) may be used. Because of the by far higher demands on quality and size of studies set out to dismiss the assumption of carcinogenicity, there is an inherent imbalance of classification concerning carcinogenicity and lack of carcinogenicity. The first statement goes still further: It claims that an analysis based on the Bradford Hill criteria confirms that there is no causal relationship. Because the only Bradford Hill criterion that is essential is “temporal relation,” the only way to confirm—based on these so-called criteria—that there is no causal relation is to demonstrate that exposure commenced after disease onset. All other evidence may reduce the weight in favor of a causal relationship but cannot confirm that there is no causal relationship.</p></sec><sec><title>Are There Criteria for Causation?</title><p>During the past decades, Bradford Hill’s criteria have played almost the same role in occupational and environmental risk assessment as Koch’s postulates for microbiology (<xref rid="b15-ehp0114-000969" ref-type="bibr">Koch 1882</xref>). As was the case with Koch’s postulates, which cannot be fulfilled for many infectious agents, so Bradford Hill’s criteria are supportive (for the assumption of a causal relation) only if fulfilled, but cannot be used to dismiss the assumption of a causal relation. It is a complete misinterpretation of the nine issues considered by Bradford Hill that they can be a type of checklist to establish causation. But it may turn out that they owe their popularity, still persisting after 40 years, exactly to this misconception.</p><p>Because the definition of a disease cause given above affords the existence of mutually exclusive conditions, in a strict sense, causation can be indicated only by (experimental) production and control of all (relevant) conditions. This, however, leads to ethical problems if the factor is potentially debilitating or lethal. And it is practically impossible if the latency is long, as it is for chronic diseases. Resorting to animal experimentation can reduce some of these problems but introduces new ones, because inference from results in animals to effects in humans is far from trivial. Hence, we are often left with a number of problems that cannot be optimally solved, and therefore there is no set of criteria that, if fulfilled, would result in attributing a factor as either causally related or not. This does not mean that we cannot, to the best of our present knowledge, come to a decision concerning the relationship of an agent and a disease. Or, as <xref rid="b3-ehp0114-000969" ref-type="bibr">Bradford Hill (1965)</xref> said 40 years ago:</p><disp-quote><p>All scientific work is incomplete—whether it be observational or experimental. All scientific work is liable to be upset or modified by advancing knowledge. That does not confer upon us a freedom to ignore the knowledge we already have, or to postpone the action that it appears to demand at a given time.</p></disp-quote></sec><sec><title>A Pragmatic Approach</title><p>Concerning a particular chemical or physical factor, general medical knowledge may suffice to attribute it as harmful and as causing illness or death (but even in extreme cases such derivations may not be altogether valid—e.g., the statement that it is impossible to climb Mt. Everest without respiratory aid). But in a developed society, obviously, hazardous conditions are likely to have been detected already and are subject to an individual and/or public risk–benefit evaluation. So we are dealing with either less obvious hazards or those that occur only rarely or in a small proportion of the population. The evidence may stem from all kinds of sources, but often we start only from the pessimistic assumption that an agent either not present in the natural environment or present only at much lower levels may be harmful to health. Or it may be that during routine surveillance, a high prevalence of a (rare) disease is observed that coincides with a (rare) environmental condition. How should we come to a conclusion whether the suspected environmental condition is causing disease? It might be worthwhile to stress that there are cases where we do not need the verdict of causation before we take action (e.g., a not very important food additive may be banned on weak evidence of harmful effects). An important part, and a much ignored one, of Bradford Hill’s article deals with such situations, as <xref rid="b21-ehp0114-000969" ref-type="bibr">Phillips and Goodman (2004)</xref> pointed out.</p><p>Starting from the definition of a disease cause stated above, it is obvious that three main issues need to be addressed (to simplify the discussion, let us speak of the set of exclusive conditions as of an agent or determinant A):</p><list list-type="bullet"><list-item><p>Is the probability of the disease conditional on the presence of A higher than in the absence of A? (association)</p></list-item><list-item><p>Is the set of conditions to which the source populations are exposed sufficiently similar except for A? (environmental equivalence)</p></list-item><list-item><p>Are the features of the populations that differ with respect to exposure to A such that, for the problem under investigation, they can be considered equivalent? (population equivalence).</p></list-item></list><sec><title>Association</title><p>Although we can to some degree rely on statistical decision theory concerning an observed difference, some problems need to be addressed: First, there are cases where we observe an incidence only in those exposed to A and contrast it to the overall incidence in the population (as was the case with hepatic angiosarcoma in workers exposed to vinyl chloride monomer). If the disease is extremely rare in the population, it may not be feasible to do a conventional epidemiologic study. However, if a plausible mechanism of action can be delineated, the observation of an unexpectedly high incidence of the disease may suffice for a verdict of causation. Second, in the case–control approach, we estimate not the conditional probabilities of the disease but their ratio. Furthermore, it is questionable whether statistical decision theory based on random sampling can be applied without further consideration. Typically, all cases of the target disease occurring within a specified region (or even only those diagnosed in one or several hospitals) and during a specified period of time are intentionally included, and only controls are sampled (either from the population or from hospital cases presenting with other than the target disease). To apply statistical decision theory, we have to assume that the cases are a random sample from the distribution of all samples related to all time/space intervals. Furthermore, the population from which the cases and controls originate has, in general, not been stable during the relevant past. Cases of the target disease that occurred before study onset are not included, and also migration in and out of the target area may play an important role, as might deaths from other and maybe related causes. Because of these circumstances and the additional problem of reliably assessing the presence of A retrospectively, case–control studies are often denied the potential to form the basis of a causal interpretation. However, this is exaggerating the difficulties associated with this study type. Especially if several case–control studies from different areas and time periods are available, a generalization about the ratio of incidences can be made if the different sources of bias have been thoroughly addressed. Finally, even if the relative risk (whether estimated from rate ratios, odds ratios, or hazard ratios) is high, statistical significance may not be reached if the number of cases exposed to A is low.</p></sec><sec><title>Environmental equivalence</title><p>Ideally, those exposed to A should share the same conditions, besides A, with those not exposed to A. If not, all relevant conditions that are potentially related to both A and the outcome (i.e., confounding conditions) must be included in the data set to account for them in the analysis. Failing to do so—that is, controlling for some but not others—may increase confounding instead of removing it (e.g., <xref rid="b18-ehp0114-000969" ref-type="bibr">Maldonado and Greenland 2001</xref>); on the other hand, controlling for a variable that is downstream of A may remove the effect of A (<xref rid="b14-ehp0114-000969" ref-type="bibr">Kaufman and Poole 2000</xref>). Because the number of potentially confounding factors is indefinite and judgment about the degree of similarity between environmental conditions depends on limited experience, there is always the possibility that an observed association is due to confounding. On the other hand, the mere suspicion that an observed association is due to confounding does not conform to scientific reasoning because it cannot be refuted by a finite sequence of empirical tests. Analysis of uncontrolled confounding (<xref rid="b10-ehp0114-000969" ref-type="bibr">Greenland 2003</xref>; <xref rid="b22-ehp0114-000969" ref-type="bibr">Robins et al. 1999</xref>) can give an idea about the strength of the association between the confounding variable and both A and the outcome required to substantially alter inferences about the existence of an association between A and the outcome. These approaches may replace the earlier procedures, as already applied by Bradford Hill.</p></sec><sec><title>Population equivalence</title><p>The counter-factual approach to causality (last statement in <xref ref-type="table" rid="t1-ehp0114-000969">Table 1</xref>), although of questionable empirical content, has great heuristic strengths. A counterfactual cause is defined as something that leads to a difference in the disease propensity with respect to the same target (population). Although, of course, it is then impossible to ever empirically demonstrate such a cause, it points to the importance of considering all features of the populations that are substitutes for the target exposed to A or not exposed to A, respectively. Ideally, all features of these substitutes should be equal. However, this would afford restriction to homozygous twin studies with twins who shared the same experiences except for exposure to A. However, for practical purposes, it will suffice to demonstrate equivalence with respect to the features that determine susceptibility to A, disposition to develop the target disease, and the interaction between disposition and susceptibility (i.e., the joint distribution of these features).</p><p>Unfortunately, as a National Cancer Institute workshop has stressed (<xref rid="b4-ehp0114-000969" ref-type="bibr">Carbone et al. 2004</xref>), there is insufficient evidence to stratify populations based on susceptibility to develop cancer. For other chronic diseases, such as atherosclerosis, Alzheimer’s disease, and obstructive pulmonary disease, there might be even fewer evidence-based criteria for disposition and susceptibility. Therefore, a still more modest approach must be followed that is embedded in the universal scientific scheme of bold trial-and-error correction. As a minimum requirement, we must address the features that are known to be related to disease incidence (in most cases, age will be among these features); features that indicate early steps of the target disease (e.g., polyposis for colon cancer), thereby keeping in mind that agent A may be effective only during certain steps of the pathologic process; and features that may determine the potential to counteract or aggravate the disease (e.g., social class). Scientific discussion may reveal that potentially important features have been left out. In this case, considerations of the potential bias thereby introduced may reveal that the effect of A has been underestimated (e.g., if those exposed to A can be considered less prone to develop the target disease). If the investigation resulted in a positive association between A and the target disease, we might conclude that no further investigation is needed; if, on the other hand, no association was revealed, there is indeed a need for error correction. An analogue procedure follows from a suspected overestimation of the association.</p><p>Environmental equivalence and population equivalence are usually termed the <italic>ceteris paribus</italic> condition and are often jointly discussed. It is, however, important to discriminate between environmental and population characteristics. Only the former can be targets of change; the latter, although not stationary at all, must be taken as side conditions that can be controlled only by active selection. It is also important to consider self-selection processes in observational studies where features of the environment may determine to some degree features of the population and vice versa.</p><p>It goes without saying that all investigations that are assessed for a causal interpretation must be scrutinized for potential biases (especially exposure and outcome misclassification and response or observer bias). However, it is insufficient merely to point to a potential bias without considering the effect this bias may have had on the results. For example, in cohort studies, exposure misclassification can lead to a bias only in the opposite direction of the reported association.</p><p>Under the precondition that all investigations have been thoroughly assessed concerning association, environmental equivalence, and population equivalence, and potential biases, and still the following set of statements can be derived, then it is reasonable to allocate A among the potentially causal factors of the target disease:</p><list list-type="bullet"><list-item><p>The temporal relationship between exposure to A and disease onset (or diagnosis) conforms to what is known about the natural history of the disease.</p></list-item><list-item><p>There is an association between exposure to A and the target disease.</p></list-item><list-item><p>Environmental characteristics in which exposed and unexposed populations live can be considered equivalent during the etiologically relevant period except for A.</p></list-item><list-item><p>Characteristics of exposed and unexposed populations are sufficiently similar to consider them equivalent.</p></list-item></list><p>Only the first two statements are essential; the latter two can be substituted by evidence from experimental or other research demonstrating a mechanism of action that does not depend on individual characteristics or environmental factors. Furthermore, if it is impossible to demonstrate the equivalence condition, then other considerations and evidence can be substituted to support the assumption of a causal relation (see below).</p><p>Temporal relation, association, and environmental and population equivalence suffice for a verdict of potential causation. This assertion can only be refuted by the following:</p><list list-type="bullet"><list-item><p>Evidence that demonstrates that A is a downstream condition of some other factor B (e.g., <italic>Helicobacter pylori</italic> infection instead of gastritis as a potential causal factor for atherosclerosis)</p></list-item><list-item><p>Evidence that A is associated with B, the essential causal agent (e.g., technical tetrachloroethene contaminated with epoxybutane)</p></list-item><list-item><p>Evidence that essential side conditions have been overlooked that need to be present to make A effective or to make non-A preventive (e.g., a specific receptor phenotype).</p></list-item></list><p>It is not necessary to demonstrate a mechanism of action. <xref rid="b3-ehp0114-000969" ref-type="bibr">Bradford Hill (1965)</xref> and others pointed to the landmark 1854 study of John Snow, who demonstrated that the rate of cholera deaths in London was 14 times higher in households supplied with water from the Southwark and Vauxhall Company compared with households supplied with water from the Lambeth Company (<xref rid="b23-ehp0114-000969" ref-type="bibr">Snow 1855</xref>). Although Snow suspected a living organism contaminating drinking water by proximity to sewage, another 30 years elapsed before Robert Koch isolated <italic>Vibrio cholerae</italic>, and more than 100 years before the mechanism of action of the cholera toxin was established. The original observation of Snow sufficed to state that something in the water supplied by one company potentially caused cholera and to take appropriate action (closing the pump), and there was no need to wait until a mechanism of action had been demonstrated (thereby probably sacrificing the lives of thousands of people). However, if a mechanism of action can be established, the requirements for epidemiologic evidence outlined above can be somewhat relaxed.</p><p>Because of difficulties inherent in observational studies, it may be impossible to demonstrate environmental and/or population equivalence to a sufficient degree, and therefore additional evidence and considerations are necessary to support the notion of a causal relation between agent A and the target disease. There is no possible evidence beyond the three points stated above that will refute epidemiologic evidence in favor of a causal relation besides more and “better” epidemiologic evidence. Stakeholders tend to “flood” the scientific literature with inconclusive (powerless and/or biased) studies in the hope that the balance of evidence will turn in favor of a less strong association between agent A and the target disease. Assessment of evidence must take this into consideration and make proper use of such information (which in most cases will result in disregarding it altogether).</p><p>There is an extensive literature about “criteria” for causal inferences in the health sciences, most of which goes back to the seminal work of <xref rid="b3-ehp0114-000969" ref-type="bibr">Bradford Hill (1965)</xref> and Mervyn <xref rid="b24-ehp0114-000969" ref-type="bibr">Susser (1973)</xref>. Although neither author meant to establish a checklist, but only to formulate issues that aid in this task, application has been more or less schematically following these criteria. However, there is no rule that can guide the decision. How many of the criteria must be fulfilled? Is one counting more than the other? What to do if none is fulfilled? There is no straightforward answer to these questions, and every single case merits its own specific line of argumentation.</p><p><xref ref-type="table" rid="t2-ehp0114-000969">Tables 2</xref> and <xref ref-type="table" rid="t3-ehp0114-000969">3</xref> propose a dialogue approach to causal inference. It is assumed that epidemiologic evidence has been put forward that is evaluated along the criteria outlined above. A scientific dialogue of conjecture and refutation at first tries to dismiss the notion of a causal relation between agent/determinant A and disease D along the four issues “temporal relation,” “association,” “environmental equivalence,” and “population equivalence.” There are valid and invalid counterarguments. If the dialogue ends without valid counterarguments, no further evidence for the verdict of causation is necessary. More often than not, epidemiologic evidence will be insufficient (e.g., due to short duration of exposure). In this case, other evidence may support or weaken the assumption of a causal relation between A and D. The most important of these arguments favoring or against causation are shown in <xref ref-type="table" rid="t3-ehp0114-000969">Table 3</xref>. Arguments against causation are often not symmetrical to arguments in favor of causation. For example, a long-term experiment in animals that results in a higher incidence of the target disease in exposed animals supports causal inference, whereas a negative result does not support the assumption of no causal relation, because the tested species or strain may lack a decisive feature (e.g., an enzyme) that is present in humans and necessary for A to produce D. There are, however, cases where a positive result in animal experiments cannot be taken as evidence for causation because of processes not present in humans.</p><p>Most risk assessment procedures demand that for chronic diseases such as cancer there must be epidemiologic evidence before an extrinsic agent can be ascribed a hazardous potential for human health. Considering the long latencies involved in these diseases, there is a need to define procedures that give answers about a potential causal relationship in a more rapid fashion. Traditional epidemiologic evidence can be provided only <italic>ex post</italic>, when the health impairment has already occurred in a significant fraction of the exposed population. There is an urgent need to connect the disciplines of molecular biology and epidemiology (<xref rid="b4-ehp0114-000969" ref-type="bibr">Carbone et al. 2004</xref>). Such collaboration should result in <italic>a</italic>) a better characterization of the study participants with respect to susceptibility and <italic>b</italic>) early markers of responses to the agent in question that can be assessed long before occurrence of manifest disease. With regard to such new approaches, it is of paramount importance to investigate the mechanism of interaction of the extrinsic agent with the organism in order to define potential cofactors and sensitive end points. For chemical substances, <italic>in silico</italic> methods and structure–activity considerations may provide first answers to a potential path of action (e.g., binding to a receptor). For physical factors such as electromagnetic fields, knowledge is more limited, and new approaches must be designed.</p><p>Despite its metaphysical character, the principle of causation or, more specifically, the notion that every disease has a cause has been of great heuristic value and likely will govern our future endeavors for better understanding of the relationship between the environment and human health until we have accumulated more knowledge and may describe the process by a system of equations. However, the complexity of the problem may be too great ever to lend itself to complete description.</p></sec></sec>
Legislation: California Enacts Safe Cosmetics Act	Could not extract abstract	<contrib contrib-type="author"><name><surname>Washam</surname><given-names>Cynthia</given-names></name></contrib>	Environmental Health Perspectives	<p>Californians frustrated with what they consider the FDA’s loose control over cosmetic safety have taken matters into their own hands with the country’s first state cosmetics regulatory act, which takes effect in January 2007. The California Safe Cosmetics Act of 2005 will require manufacturers to report the use of potentially hazardous ingredients to the state Department of Health Services (DHS), which in turn will alert consumers. The DHS has the authority to investigate whether the product could be toxic under normal use and to require that manufacturers submit health effects data. Manufacturers that continue marketing products deemed unsafe in California could face legal action.</p><p>“The legislation’s sponsors believe that the basis of the law is the public’s right to know,” says Kevin Reilly, DHS deputy director of prevention services. The new law uses the list of toxicants drawn up under California’s Proposition 65, which mandates that the governor publish a list, updated at least yearly, of chemicals that are known to the state of California to cause cancer, birth defects, or other reproductive harm.</p><p>Although the new act applies only in California, its effects are likely to reverberate nationwide. Consumer advocates predict that manufacturers seeking to avoid negative publicity will remove, rather than report, suspect ingredients. Those formulas would then be marketed coast to coast.</p><p>Impetus for the law stems from consumers’ concerns over long-term exposure to certain cosmetic ingredients. Cosmetic use has not been linked to chronic illnesses, but some products do contain carcinogens (such as formaldehyde, used in nail treatments), teratogens (such as lead acetate, used in two hair dyes), and other reproductive toxicants (such as di-<italic>n</italic>-butyl phthalate, used in nail treatments and dandruff shampoos).</p><p>Studies in recent years have shown that humans absorb and inhale sometimes surprisingly high levels of toiletry ingredients. In the November 2005 issue of <italic>EHP</italic>, a team led by Susan M. Duty of the Harvard School of Public Health demonstrated that urine concentrations of phthalate metabolites increased by 33% with each personal care product—hair gel or spray, lotion, deodorant, cologne, aftershave—that subjects used.</p><p>Historically, cosmetics safety has been in the hands of manufacturers; the FDA requires no premarket testing. Each year, an expert panel convened by the industry-funded Cosmetic Ingredient Review (CIR) identifies priority ingredients for which it conducts literature reviews and analyses to determine safety. The panel—made up of independent academic researchers and representatives from industry, consumer interests, and the FDA—has declared 9 of the 1,286 ingredients reviewed since 1976 unsafe for normal cosmetic use. But manufacturers are not obligated to eliminate any ingredients—at least one ingredient identified as unsafe by CIR, hydroxyanisole, is still used.</p><p>Safety advocates see evidence of any harm in any use as reason enough for a ban. “Ingredients suspected of causing cancer shouldn’t be used in cosmetics,” says spokesman Kevin Donegan of the Breast Cancer Fund, a San Francisco–based nonprofit that promoted the California bill.</p><p>F. Alan Andersen, director and scientific coordinator of CIR, counters that the dose creates the danger. “We don’t subscribe to the notion that if there’s ever an adverse effect, [a chemical] must not be in a product people use,” he says. “It doesn’t make sense to us to apply the precautionary principle. Instead, we use a risk assessment approach, and the wide margins of safety that we have found for chemicals such as phthalates using this approach assure us that actual use of cosmetics is safe.”</p><p>The law drew fierce opposition from individual companies and the Cosmetic, Toiletry, and Fragrance Association (CTFA) as it worked its way through the California legislature. “CTFA supports strong federal regulation by the FDA,” says Kathleen Dezio, executive vice president of public affairs and communications for the association. “For this reason, CTFA has generally opposed state-specific legislation that would undermine this national approach and lead to an unworkable state-by-state patchwork of rules . . . or unjustified, extreme requirements that are well beyond those placed on any other category of food, beverages, drugs, or consumer products.” She adds that CTFA has met with the DHS and “pledged our cooperation in accomplishing the requirements” of the law.</p><p>Some manufacturers have already ceded to public pleas for safer products. In the past two years, almost 350 of them signed a pledge promoted by the Campaign for Safe Cosmetics, a coalition of health and environmental groups, to use no chemicals linked to cancer or birth defects. Industry leaders L’Oréal and Revlon broke new ground last year when they promised that products they sold in the United States would meet more stringent European Union standards. In 2004 Europe enacted a ban on suspected carcinogens, mutagens, and reproductive toxicants in personal care products.</p><p>“We’re definitely seeing a shift in the attitude of manufacturers,” Donegan says. “They’re starting to see the benefits of removing anything that could cause cancer.”</p>
Propanil Exposure Induces Delayed but Sustained Abrogation of Cell-Mediated Immunity through Direct Interference with Cytotoxic T-Lymphocyte Effectors	<p>The postemergent herbicide propanil (PRN; also known as 3,4-dichloropropionanilide) is used on rice and wheat crops and has well-known immunotoxic effects on various compartments of the immune system, including T-helper lymphocytes, B lymphocytes, and macrophages. It is unclear, however, whether PRN also adversely affects cytotoxic T lymphocytes (CTLs), the primary (1°) effectors of cell-mediated immunity. In this study we examined both the direct and indirect effects of PRN exposure on CTL activation and effector cell function to gauge its likely impact on cell-mediated immunity. Initial experiments addressed whether PRN alters the class I major histocompatibility complex (MHC) pathway for antigen processing and presentation by antigen-presenting cells (APCs), thereby indirectly affecting effector function. These experiments demonstrated that PRN does not impair the activation of CTLs by PRN-treated APCs. Subsequent experiments addressed whether PRN treatment of CTLs directly inhibits their activation and revealed that 1° alloreactive CTLs exposed to PRN are unimpaired in their proliferative response and only marginally inhibited in their lytic activity. Surprisingly, secondary stimulation of these alloreactive CTL effectors, however, even in the absence of further PRN exposure, resulted in complete abrogation of CTL lytic function and a delayed but significant long-term effect on CTL responsiveness. These findings may have important implications for the diagnosis and clinical management of anomalies of cell-mediated immunity resulting from environmental exposure to various herbicides and other pesticides.</p>	<contrib contrib-type="author"><name><surname>Sheil</surname><given-names>James M.</given-names></name></contrib><contrib contrib-type="author"><name><surname>Frankenberry</surname><given-names>Marc A.</given-names></name></contrib><contrib contrib-type="author"><name><surname>Schell</surname><given-names>Todd D.</given-names></name></contrib><contrib contrib-type="author"><name><surname>Brundage</surname><given-names>Kathleen M.</given-names></name></contrib><contrib contrib-type="author"><name><surname>Barnett</surname><given-names>John B.</given-names></name></contrib><aff id="af1-ehp0114-001059">Department of Microbiology, Immunology, and Cell Biology, West Virginia University School of Medicine, Morgantown, West Virginia, USA</aff>	Environmental Health Perspectives	<p>Numerous studies concerning the health-related effects of environmental toxicants demonstrate that the immune system, in addition to other organ systems including the reproductive, nervous, pulmonary, and circulatory systems, is often compromised (<xref rid="b7-ehp0114-001059" ref-type="bibr">Carpy et al. 2000</xref>; <xref rid="b12-ehp0114-001059" ref-type="bibr">Costa 1997</xref>). Our understanding of such adverse immunologic effects, however, is largely limited to the immediate and early consequences after exposure to such agents. The principal contribution of this present article to the field of immunotoxicology research is its demonstration that the potential long-term impact of propanil (PRN) exposure on cell-mediated immunity is far more severe than its short-term consequences. This delayed appearance of irreversible PRN-induced immunotoxic effects may be important for diagnostic and therapeutic measures in assessing exposure to environmental toxicants in general.</p><p>PRN is a postemergent herbicide used extensively around the world in the cultivation of rice and wheat crops. Its particular effectiveness is due to the high level of acylamidase expression in a rice plant that allows it to detoxify PRN, whereas common grass-type weeds lack this enzyme and are killed by this herbicide (<xref rid="b20-ehp0114-001059" ref-type="bibr">Matsunaka 1968</xref>). PRN is routinely applied several times during a growing season without detrimental effects to the plant (<xref rid="b8-ehp0114-001059" ref-type="bibr">Casida and Lykken 1969</xref>; <xref rid="b33-ehp0114-001059" ref-type="bibr">Smith 1961</xref>), with 3–6 lb/acre applied annually in the United States (<xref rid="b12-ehp0114-001059" ref-type="bibr">Costa 1997</xref>; <xref rid="b20-ehp0114-001059" ref-type="bibr">Matsunaka 1968</xref>). Thus, a high environmental exposure of humans to PRN normally occurs as an occupational risk.</p><p>Earlier reports by Barnett and co-workers (<xref rid="b3-ehp0114-001059" ref-type="bibr">Barnett et al. 1992</xref>; <xref rid="b2-ehp0114-001059" ref-type="bibr">Barnett and Gandy 1989</xref>; <xref rid="b13-ehp0114-001059" ref-type="bibr">Frost et al. 2001</xref>; <xref rid="b37-ehp0114-001059" ref-type="bibr">Theus et al. 1993</xref>; <xref rid="b41-ehp0114-001059" ref-type="bibr">Xie et al. 1997</xref>; <xref rid="b43-ehp0114-001059" ref-type="bibr">Zhao et al. 1995</xref>, <xref rid="b42-ehp0114-001059" ref-type="bibr">1998</xref>) indicated that PRN exposure results in adverse effects on most compartments of the immune system, including macrophages, B lymphocytes, and T-helper lymphocytes. Curiously, however, there appeared to be little, if any, effect on cellular immunity mediated by cytotoxic T-lymphocyte (CTL) effectors (<xref rid="b3-ehp0114-001059" ref-type="bibr">Barnett et al. 1992</xref>; <xref rid="b2-ehp0114-001059" ref-type="bibr">Barnett and Gandy 1989</xref>).</p><p>Given that the responsiveness of the other immune compartments examined is inhibited by PRN exposure, we hypothesized that acute PRN exposure might yet impair CTL function, albeit in a manner that is initially difficult to detect under the <italic>in vitro</italic> conditions used. To test this hypothesis we considered that the adverse immunotoxic effects of PRN exposure on cell-mediated immunity might be observed in one or more of three parameters: <italic>a</italic>) presentation of peptide antigen to CTLs by antigen-presenting cells (APCs), <italic>b</italic>) proliferation and differentiation of CTLs, and/or <italic>c</italic>) functional lytic response of activated CTL effectors.</p><p>The immune activation and functional responsiveness of CTLs can be examined and assessed independently <italic>in vitro</italic>. CTL activation is based on the capacity of APCs to efficiently process and present peptide antigens to CTLs and thus indirectly affects CTL responsiveness. Conversely, the functional lytic response of CTLs emerges as a result of the differentiation of naive CD8<sup>+</sup> T cells into effector CTLs capable of responding through lysis of the target cell, thereby serving as a direct measure of CTL activation.</p><p>In this present article, we demonstrate three important consequences of PRN exposure on the <italic>in vitro</italic> parameters of CTL activation and their functional activity as effectors of cell-mediated immunity: <italic>a</italic>) antigen presentation to CTLs is not impaired, <italic>b</italic>) the functional lytic activity of primary (1°) CTLs is only marginally impaired, and <italic>c</italic>) upon restimulation of 1° CTLs in the absence of PRN, the secondary (2°) CTL response is completely abrogated. On the basis of these observations, we conclude that the immunotoxic effects of PRN exposure on CTLs are delayed in their appearance and directly impair the functional activity of these effectors of cell-mediated immunity. These results may have serious and important direct implications for both diagnosis and clinical management of the acute and chronic effects of PRN exposure. Furthermore, these findings warrant examining similar acute versus delayed exposure effects with respect to the immunotoxic potential of other environmental toxicants.</p><sec sec-type="materials\|methods"><title>Materials and Methods</title><sec><title>Animals</title><p>In this study we used C57BL/6 (B6; H-2<sup>b</sup>) and BALB/c (H-2<sup>d</sup>) female mice 10–12 weeks of age from Charles River Breeding Laboratories, Inc. (Wilmington, MA) or from our own breeding colony at the West Virginia University Health Sciences Vivarium. All animals used in this study were treated humanely and with regard for alleviation of suffering.</p></sec><sec><title>Cell lines</title><p>Two tumor cell lines, designated P815 (H-2<sup>d</sup>) and EL4 (H-2<sup>b</sup>), were used as targets for alloreactive cytotoxic T cells. The EL4 cell line expresses class I H-2<sup>b</sup> molecules and is derived from a B6 lymphoma originally induced in a C57BL/6N mouse by 9,10-dimethyl-1,2-benzanthracene (<xref rid="b17-ehp0114-001059" ref-type="bibr">Gorer 1950</xref>). P815 is a cell line derived from a mastocytoma in DBA/2 (H-2<sup>d</sup>) mice, and it expresses class I H-2<sup>d</sup> molecules (<xref rid="b23-ehp0114-001059" ref-type="bibr">Plaut et al. 1973</xref>; <xref rid="b27-ehp0114-001059" ref-type="bibr">Ralph and Nakoinz 1974</xref>). Both the EL4 and P815 cell lines have been used extensively by us and others as suitable targets for lysis in cytotoxic T-cell assays. N1 is derived from EL4 cells transfected with the vesicular stomatitis virus nucleoprotein (VSB-N) gene (<xref rid="b26-ehp0114-001059" ref-type="bibr">Puddington et al. 1986</xref>).</p></sec><sec sec-type="methods"><title>Monoclonal antibodies and fluorescence activated cell sorting analysis</title><p>We used the following H-2K<sup>b</sup>-specific monoclonal antibodies (mAb): 5F1 (<xref rid="b32-ehp0114-001059" ref-type="bibr">Sherman and Randolph 1981</xref>), Y-3 (<xref rid="b19-ehp0114-001059" ref-type="bibr">Jones and Janeway 1981</xref>), EH144 (<xref rid="b5-ehp0114-001059" ref-type="bibr">Bluestone et al. 1985</xref>; <xref rid="b15-ehp0114-001059" ref-type="bibr">Geier et al. 1986</xref>), Y-25 (<xref rid="b19-ehp0114-001059" ref-type="bibr">Jones and Janeway 1981</xref>), and 28-13-3 (<xref rid="b22-ehp0114-001059" ref-type="bibr">Ozato and Sachs 1981</xref>). Fluorescein isothiocyanate goat anti-mouse immunoglobulin (heavy- and light-chain–specific) was purchased from Southern Biotechnology Associates, Inc. (Birmingham, AL).</p></sec><sec><title>Citric acid treatment of APCs</title><p>The acid treatment protocol used in these studies to strip the APC cell surfaces of class I peptide/major histocompatibility complex (pMHC) complexes is essentially the same as that described by <xref rid="b35-ehp0114-001059" ref-type="bibr">Sugawara et al. (1987)</xref>, as modified by <xref rid="b34-ehp0114-001059" ref-type="bibr">Storkus et al. (1993)</xref>. Briefly, APCs are <italic>a</italic>) collected and pelleted by centrifugation; <italic>b</italic>) resuspended in 0.5 mL citrate-phosphate buffer, pH 3.0 (citrate-phosphate buffer consists of a 1:1 mixture of 0.263 M citric acid, pH 1.8, and 0.123 M Na<sub>2</sub>HPO<sub>4</sub>); <italic>c</italic>) incubated in citrate-phosphate buffer for 1 min at room temperature; <italic>d</italic>) resuspended in 10 mL RP-10 media, pelleted, and washed with Hank’s balanced salt solution; and <italic>e</italic>) resuspended to appropriate concentration in RP-10 media. RP-10 tissue culture media consists of RPMI-1640 media plus 10% fetal calf serum, with supplemental vitamins, nonessential amino acids, and HEPES buffer.</p></sec><sec><title>Effector cells</title><p>VSV-N peptide–specific CTLs were maintained <italic>in vitro</italic> by weekly stimulation with the target VSV-N peptide, p52–59. Briefly, 4 × 10<sup>5</sup> CTL clone 33 cells were incubated in a 24-well flat-bottom plate with 5 × 10<sup>6</sup> irradiated (2,000 rads) B6 spleen cells plus 2 μM VSV-N p52–59 peptide suspended in RP-10 media. CTL clone 33 cells were analyzed for antigen-specific lytic reactivity with <sup>51</sup>Cr-labeled N1 transfectant targets on day 5 and subsequently restimulated on day 7 of culture.</p><p>Alloreactive CTLs were induced by 1° stimulation of B6 spleen cells with irradiated (2,000 rads) spleen cells from BALB/c mice. Briefly, spleens were removed and processed into single-cell suspension preparations; BALB/c spleen cell suspensions were irradiated in a Gammacell 1000 cesium-137 irradiator (Atomic Energy of Canada Ltd., Kanata, Ontario, Canada) to deliver 2,000 rads. For 1° alloreactive stimulation, 25 × 10<sup>6</sup> B6 spleen cells per flask were added to upright T-25 flasks with 25 × 10<sup>6</sup> BALB/c irradiated spleen cells in 10 mL RP-10 media. Alloreactive cultures were placed in a 37°C humidified incubator at 7% CO<sub>2</sub> for 7 days.</p><p>Secondary alloreactive cultures were prepared similarly in RP-10 media except that 2.5 × 10<sup>6</sup> 1° effectors per flask were added together with 25 × 10<sup>6</sup> irradiated BALB/c spleen cells to upright T-25 flasks. Cultures were incubated for 7 days in the same manner as the 1° alloreactive cultures. Subsequent cultures beyond the 2° alloreactive effectors were maintained in 24-well dishes (Corning-Costar; Corning Life Sciences, Corning NY,) by the addition of 1 × 10<sup>5</sup> effectors plus 1 × 10<sup>6</sup> irradiated BALB/c spleen cells per well in 2 mL RP-10 media supplemented with 5% rat concanavalin A supernatant as a source of interleukin-2.</p></sec><sec><title>Mixed lymphocyte reaction assay</title><p>To measure the extent of alloreactive T-cell stimulation in mixed lymphocyte cultures (MLCs) and the effect of adding PRN on the induction of alloreactive CTL effectors, we used the mixed lymphocyte reaction (MLR) assay, as previously described (<xref rid="b31-ehp0114-001059" ref-type="bibr">Sheil et al. 1987</xref>). T-cell proliferation was determined in a one-way MLR assay on day 4 of culture by the incorporation of tritiated thymidine (<sup>3</sup>H-TdR) by proliferating T cells. Briefly, after 72 hr of culture, 5 × 10<sup>5</sup> viable 1° MLC cells in 100 μL plus 1 μCi <sup>3</sup>H-TdR in 100 μL RP-10 were added per well to four wells per sample in a 96-well plate (Corning-Costar; Corning Life Sciences). After incubation for 18–24 hr at 37°C in a 7% CO<sub>2</sub> humidified incubator, the cells were harvested, and the amount of proliferation was determined by measuring <sup>3</sup>H-TdR uptake, as reflected by the total radioactive counts per sample in liquid scintillation fluid.</p></sec><sec><title><sup>51</sup>Cr-release assay</title><p>We determined the lytic activity of peptide-specific and alloreactive effector CTLs using a standard 4-hr <italic>in vitro</italic> <sup>51</sup>Cr-release assay, as previously described (<xref rid="b31-ehp0114-001059" ref-type="bibr">Sheil et al. 1987</xref>). Briefly, tumor cells to be used as targets were labeled with radioactive sodium chromate (Na<sup>51</sup>Cr) and mixed with titrated doses of peptide-specific or alloreactive CTLs in 200 μL RP-10/well in 96-well round-bottom microtiter plates (Costar-Corning; Corning Life Sciences). The plates were incubated at 37°C in 7% CO<sub>2</sub> for 4 hr and centrifuged, and 100 μL supernatant was collected from each well. The amount of specific lysis was determined according to the following formula: % specific lysis = (experimental release – spontaneous release) ÷ (maximum release – spontaneous release) × 100.</p></sec><sec><title>PRN exposure</title><p>PRN (3,4-dichloroproprionaniline; > 97% purity) was purchased from Chem Service, Inc. (West Chester, PA) and dissolved in 70% ethanol (EtOH). Exposure of alloreactive effectors to PRN <italic>in vitro</italic> was accomplished by the addition of PRN concentrations of 16, 33, or 66 μM to the culture media at the initiation of culture (day 0) for 1° MLCs; for 2° MLCs, the PRN concentrations used were 66 and 165 μM. PRN remained in the media for the duration of the culture incubation period—usually 7 days. APCs and target cells were exposed <italic>in vitro</italic> to PRN (200 μM) for a period of either 18 hr or 2 hr at 37°C in 7% CO<sub>2</sub>, after which the PRN is washed out of the cultures.</p></sec><sec sec-type="methods"><title>Statistical analysis</title><p>All results shown are representative of at least three repeated experiments, and the sample points in each experiment were run in triplicate. Thus, we performed all statistical analyses using triplicate samples for data points within each experiment. Statistical evaluation was conducted using the Student’s <italic>t</italic>-test analysis, and significance in observed differences as described in the text was established as being at a level of <italic>p</italic> ≤ 0.01.</p></sec></sec><sec sec-type="results"><title>Results</title><p>To determine how PRN exposure might adversely affect cell-mediated immunity, we designed the initial experiments of this study to address its potential impact both indirectly (on <italic>in vitro</italic> antigen presentation to CTLs) and directly (on CTL lytic function). We first considered the requirement that antigen-specific CTLs must respond to a peptide antigen exposed on the surface of APCs bound to a self class I MHC molecule. If there is an adverse indirect effect on cell-mediated immunity due to PRN exposure, it could result from altered antigen processing and presentation characteristics of the APC. Alternatively, the exposure of potential effector CTLs to PRN might directly interfere with CTL proliferation and/or effector function. To address whether PRN exposure has a discernible effect on antigen presentation, we examined the well-characterized CTL response to the single antigenic peptide, VSV-N p52–59, in the context of the class I H-2K<sup>b</sup> molecule (<xref rid="b31-ehp0114-001059" ref-type="bibr">Sheil et al. 1987</xref>; <xref rid="b38-ehp0114-001059" ref-type="bibr">Van Bleek and Nathenson 1990</xref>). To examine the possible direct effects on CTL proliferation and/or differentiation, B6 anti-BALB/c MLC-derived CTLs were used as alloreactive effectors.</p><sec><title>PRN-exposed APCs are recognized efficiently by VSV-N peptide–specific CTLs</title><p>We examined the functional capacity of PRN-exposed APCs to determine whether PRN exposure of APCs <italic>in vitro</italic> adversely affects their ability to process and/or present antigen in the class I MHC pathway. The VSV-N transfectant model system (<xref rid="b26-ehp0114-001059" ref-type="bibr">Puddington et al. 1986</xref>) was used, as described previously (<xref rid="b31-ehp0114-001059" ref-type="bibr">Sheil et al. 1987</xref>), to determine whether exposure of VSV-infected cells to PRN interferes with their ability to effectively present viral peptide antigens to CTLs. In these experiments, CTL clone 33, specific for VSV-N p52–59 (<xref rid="b31-ehp0114-001059" ref-type="bibr">Sheil et al. 1987</xref>), was tested against the target VSV-N transfected EL4 (H-2<sup>b</sup>) tumor cell line, designated N1. Initially, N1 cells were exposed to PRN for 18 hr before their use as CTL targets; however, because of undesirable levels of toxicity to the N1 cells (i.e., up to 30%), the period of incubation with PRN was decreased to 2 hr. After incubation in the presence of PRN for either 2 or 18 hr, N1 cells were tested as targets for lysis by clone 33. The results depicted in <xref ref-type="fig" rid="f1-ehp0114-001059">Figure 1</xref> demonstrate that incubation of N1 cells with PRN does not adversely affect their capacity to serve as targets for lysis by VSV-N peptide–specific CTL effectors.</p><p>Another possible effect of PRN is its interference with the ability of N1 cells to serve as stimulators for the induction of VSV-N peptide–specific CTL effectors, even though they are undiminished in their capacity to serve as targets for CTL lysis. To address this possibility, we used N1 cells as APC stimulators for the VSV-N p52–59 peptide–specific, H-2K<sup>b</sup>-restricted CTL clone 33. After a 2-hr exposure to 200 μM PRN, N1 cells were added to culture flasks as stimulators for clone 33 CTLs. The 4-hr <sup>51</sup>Cr-release assay results depicted in <xref ref-type="fig" rid="f2-ehp0114-001059">Figure 2</xref> demonstrate that PRN-treated N1 cells are effective stimulators, in that they are undiminished, compared with control EtOH-treated N1 cells, in their capacity to stimulate lytic activity in clone 33 CTLs. This conclusion is reinforced by the observation that cell viability (as determined by trypan blue dye exclusion) and proliferative capacity (as determined by <sup>3</sup>H-TdR uptake in the MLR assay) of clone 33 CTLs are not significantly different after culture with either control EtOH-treated or PRN-exposed N1 cells (data not shown).</p></sec><sec><title>Citric acid–treated EL4 cells are recognized efficiently by VSV-N peptide–specific CTLs after PRN exposure</title><p>Previous studies have demonstrated that a CTL effector requires only very few pMHC complexes on the surface of an APC to become activated by and lyse the APC as its target (<xref rid="b6-ehp0114-001059" ref-type="bibr">Brower et al. 1994</xref>; <xref rid="b9-ehp0114-001059" ref-type="bibr">Christinck et al. 1991</xref>). Thus, the ability of PRN-treated N1 cells to efficiently present antigen to CTLs and to serve as targets for lysis may correspond to their expression of this minimal number of pMHC complexes needed to target CTL lysis, even though it may be much lower than is normally expressed on N1 cells not exposed to PRN. To address this possibility, we subjected EL4 cells (from which the N1 cell line is derived) to citric acid treatment, an approach that strips most pMHC complexes from the cell surface (<xref rid="b34-ehp0114-001059" ref-type="bibr">Storkus et al. 1993</xref>; <xref rid="b35-ehp0114-001059" ref-type="bibr">Sugawara et al. 1987</xref>). As shown in <xref ref-type="table" rid="t1-ehp0114-001059">Table 1</xref>, treatment of EL4 cells in this manner results in a similarly dramatic decrease in MHC class I expression, regardless of their subsequent exposure to PRN. This decreased MHC class I expression, however, does not adversely affect the presentation of peptide antigen to VSV-N peptide–specific CTLs, as shown by their undiminished lysis of acid-treated EL4 targets either with or without PRN exposure (<xref ref-type="fig" rid="f3-ehp0114-001059">Figure 3</xref>). Note that the exquisite sensitivity of this peptide-specific lysis is unaffected even with the addition of a concentration of target peptide as low as 78 pM. These results clearly indicate that the class I MHC antigen presentation pathway after PRN exposure, under these conditions, remains intact and fully functional.</p></sec><sec><title>Primary alloreactive CTLs show limited inhibition after PRN exposure</title><p>Given that we observed no overt adverse effects of PRN exposure on APC function, we directed our attention to whether PRN exposure interferes directly with CTL function itself. To address this point, an alloreactive C57BL/6 (B6) anti-BALB/c mouse model system, as described in “Materials and Methods,” was used to obtain CTL effectors. The effect of PRN exposure on alloreactive CTL activation was examined in two ways. First, we measured the proliferative capacity of BALB/c-stimulated B6 CTLs in a standard <sup>3</sup>H-TdR uptake assay. Second, to detect any functional changes in CTL effector activity, we measured allospecific CTL lysis of P815 targets in an <italic>in vitro</italic> 4-hr <sup>51</sup>Cr-release assay (<xref rid="b31-ehp0114-001059" ref-type="bibr">Sheil et al. 1987</xref>).</p><p>The <italic>in vitro</italic> PRN-exposed 1° B6 anti-BALB/c CTLs are unchanged in their proliferative capacity compared with EtOH-treated control CTLs (<xref ref-type="fig" rid="f4-ehp0114-001059">Figure 4A</xref>), thereby indicating that there is no effect on their ability to proliferate in response to antigen stimulation after exposure to a range (16, 33, or 66 μM) of PRN concentrations. Furthermore, their functional lytic reactivity is only marginally inhibited after PRN exposure, and only at the highest (66 μM) concentration used in these experiments, as shown by the CTL lytic response in a 4 hr <italic>in vitro</italic> <sup>51</sup>Cr-release assay (<xref ref-type="fig" rid="f4-ehp0114-001059">Figure 4B</xref>).</p></sec><sec><title>Proliferation and reactivity of secondary CTLs are markedly impaired after PRN exposure</title><p>We next addressed whether the subsequent <italic>in vitro</italic> exposure of these alloreactive CTLs to PRN, during their 2° stimulation, might reveal an increased adverse effect on CTL proliferation and/or CTL lytic activity. To examine this possibility, 1° alloreactive CTLs were harvested on day 7 of culture, washed, and restimulated with the addition of fresh PRN as described in “Materials and Methods.”</p><p>In the same manner as with alloreactive CTLs from 1° MLCs, we tested 2° alloreactive CTLs for their proliferative capacity (<xref ref-type="fig" rid="f5-ehp0114-001059">Figure 5A</xref>) and their lytic responsiveness (<xref ref-type="fig" rid="f5-ehp0114-001059">Figure 5B</xref>). Secondary B6 anti-BALB/c MLCs were exposed to 66 or 165 μM PRN during <italic>in vitro</italic> 2° stimulation. Because we saw no effects at the lower PRN concentrations of 16 and 33 μM in 1° MLCs, they were excluded from further analysis with 2° CTLs. Instead, 2° MLCs were set up using the effective 66 μM PRN concentration, as well as a higher concentration of 165 μM.</p><p>We tested secondary CTLs on day 4 for proliferation (<xref ref-type="fig" rid="f5-ehp0114-001059">Figure 5A</xref>) and on day 5 for lytic activity against syngeneic P815 targets (<xref ref-type="fig" rid="f5-ehp0114-001059">Figure 5B</xref>). We included an important control group in which PRN-exposed 1° MLC effectors were washed and restimulated in 2° MLCs without additional exposure to PRN. As shown in <xref ref-type="fig" rid="f5-ehp0114-001059">Figure 5A</xref>, the proliferative capacity of the 1° EtOH-treated/2° 66 μM PRN group (second bar) is fully intact, whereas the 1° EtOH-treated/165-μM PRN group (third bar) shows no proliferative capacity above background (i.e., media control). Interestingly, for the lytic response of the 1° EtOH-treated/2° 66-μM PRN group (<xref ref-type="fig" rid="f5-ehp0114-001059">Figure 5B</xref>, left), the level of inhibition also is approximately double that seen with CTLs exposed to 66 μM PRN in 1° MLC (<xref ref-type="fig" rid="f4-ehp0114-001059">Figure 4B</xref>). On the basis of this observation, we are presently examining whether activated CTLs in 2° MLC might be more susceptible to PRN-mediated inhibition than are naive CD8<sup>+</sup> T cells.</p><p>It is unlikely that the unresponsiveness observed in these groups is due to a generalized PRN-induced toxicity to the exposed CTLs because the viable cell yield of the 1° EtOH-treated/165-μM PRN group is approximately 70% of the EtOH-treated group, and that of the 1° EtOH-treated/2° 66-μM effectors is approximately 90% that of the EtOH-treated group. Furthermore, even the marginal decrease in cell viability observed in these groups has been taken into account in determining the total number of viable cells used in both the MLR and <sup>51</sup>Cr-release assays. Cell populations used in both assays were equalized based on these total viable cell determinations; thus, the number of cells added is the same for each group. It is possible, however, that these cell viability determinations do not take into account damaged cells whose cell membranes are still intact because these cells would exclude the trypan blue dye until such point as membrane damage has occurred.</p><p>Another important, although initially unanticipated, finding concerns the control alloreactive CTLs in the 1° 66-μM PRN/2°-EtOH–treated group. These CTLs were initially exposed to 66 μM PRN during 1° stimulation, followed by restimulation in 2° MLC in the absence of PRN. The lytic activity of this group is almost completely ablated (<xref ref-type="fig" rid="f5-ehp0114-001059">Figure 5B</xref>, middle), even though the 1° response is only marginally inhibited compared with EtOH-treated 1° alloreactive CTLs (<xref ref-type="fig" rid="f4-ehp0114-001059">Figure 4B</xref>). Thus, the CTL lytic response of this group is nearly 20-fold lower in than that of 66 μM PRN-treated 1° CTLs.</p><p>To determine whether this unanticipated decline in CTL reactivity is irreversible, 3° MLCs were established without the addition of PRN as was done for the 2° MLCs. So, in this case we have alloreactive CTLs that have been activated multiple times, but they were exposed to PRN only during their initial activation. Among the 3° MLC-derived CTL effectors, both proliferation and lytic activity remained severely diminished (data not shown), as seen with the 2° CTLs. Thus, the profound PRN-induced defect incurred during their 1° MLC stimulation appears to render these CTLs irreversibly impaired.</p></sec></sec><sec sec-type="discussion"><title>Discussion</title><p>Most herbicides and other pesticides exert a diverse array of immunotoxic effects on exposed individuals, including compromised humoral and cellular immunity (<xref rid="b1-ehp0114-001059" ref-type="bibr">Banerjee et al. 1996</xref>; <xref rid="b29-ehp0114-001059" ref-type="bibr">Rodgers 1995</xref>; <xref rid="b39-ehp0114-001059" ref-type="bibr">Vial et al. 1996</xref>; <xref rid="b40-ehp0114-001059" ref-type="bibr">Voccia et al. 1999</xref>). Earlier studies on the immunotoxic effects of PRN exposure by Barnett and co-workers (<xref rid="b3-ehp0114-001059" ref-type="bibr">Barnett et al. 1992</xref>; <xref rid="b2-ehp0114-001059" ref-type="bibr">Barnett and Gandy 1989</xref>) indicate that, although other important immune parameters are adversely diminished, the immunotoxic effects of PRN do not include impairment of cell-mediated immunity. This apparent anomaly in the immunotoxic impact of PRN exposure on different immune compartments prompted us to consider whether the effects of PRN on cell-mediated immunity might be more subtle or less easily detectable than effects on other immune compartments.</p><p>In the present study, we addressed the immunotoxic potential of the herbicide PRN on the effector cells of cell-mediated immunity, CTLs. A rigorous <italic>in vitro</italic> analysis of CTL activation and function was applied to determine whether and how PRN might induce immunotoxic effects in this regard. We approached this problem with the understanding that impaired cell-mediated immunity can result from the inhibition in antigen presentation to CD8<sup>+</sup> T cells and/or, more directly, from a diminished functional CTL response.</p><p>Antigen processing and presentation defects have been implicated as the basis for impaired cell-mediated immunity induced by viruses (<xref rid="b14-ehp0114-001059" ref-type="bibr">Fruh et al. 1997</xref>; <xref rid="b18-ehp0114-001059" ref-type="bibr">Hewitt and Dugan 2004</xref>; <xref rid="b21-ehp0114-001059" ref-type="bibr">Mylin et al. 1995</xref>) and by antioxidants (<xref rid="b16-ehp0114-001059" ref-type="bibr">Gong and Chen 2003</xref>; <xref rid="b25-ehp0114-001059" ref-type="bibr">Preynat-Seauve et al. 2003</xref>), as well as in tumor development (<xref rid="b4-ehp0114-001059" ref-type="bibr">Bennink et al. 1993</xref>; <xref rid="b10-ehp0114-001059" ref-type="bibr">Cohen et al. 2003</xref>; <xref rid="b28-ehp0114-001059" ref-type="bibr">Restifo et al. 1993</xref>; <xref rid="b30-ehp0114-001059" ref-type="bibr">Seliger et al. 1998</xref>) and aging (<xref rid="b24-ehp0114-001059" ref-type="bibr">Plowden et al. 2004</xref>). The indirect consequences of these agents on antigen presentation can adversely affect the proliferation, differentiation, and effector functions of T lymphocytes—including cell signaling mechanisms, cytokine secretion, developmental maturation, and target cell lysis by CD8<sup>+</sup> CTLs.</p><p>Thus it was important to examine the indirect immunotoxic effects of PRN exposure on the antigen processing and presentation component of cell-mediated immunity. In addition, the most common direct measures of CTL activation are proliferation and lytic activity. The experiments conducted in this study incorporate both indirect and direct approaches to determining the immunotoxic effects of PRN exposure on antigen presentation and CTL activation.</p><p>The most important findings of this study are that <italic>a</italic>) exposure to PRN during 1° CTL activation results in a dramatic delayed abrogation of CTL lysis that is irreversible, and <italic>b</italic>) the immunotoxic effects of PRN exposure under these conditions are limited to the functional activity of CTLs and do not affect antigen processing and presentation to CTLs. This study is unique in that it demonstrates such a striking difference between the short-term and delayed appearance of the immunotoxic effects of this herbicide. The issue of potentially delayed immunotoxic effects of pesticides has not been a focus of most studies, although some changes have been reported after <italic>in utero</italic> exposure that manifested after development (<xref rid="b11-ehp0114-001059" ref-type="bibr">Colosio et al. 1999</xref>; <xref rid="b39-ehp0114-001059" ref-type="bibr">Vial et al. 1996</xref>). This study, however, relates directly to the impaired activation of mature effectors of cell-mediated immunity. It is also important that these effects impair the proliferation and lytic activity of CTLs without interfering with the presentation of antigen by APCs.</p><p>In the initial approach to address whether PRN exposure inhibits cell-mediated immunity, we examined its possible impact on antigen processing and presentation in the class I MHC antigen presentation pathway, and indirectly on CTL induction and responsiveness. This approach used VSV-N gene-transfected N1 cells treated with PRN as targets for CTL-mediated lysis by CTL clone 33 (<xref rid="b31-ehp0114-001059" ref-type="bibr">Sheil et al. 1987</xref>), which is H-2K<sup>b</sup> restricted and specific for the VSV-N p52–59 peptide. Results depicted in <xref ref-type="fig" rid="f1-ehp0114-001059">Figures 1</xref> and <xref ref-type="fig" rid="f2-ehp0114-001059">2</xref> reveal that exposure of APCs to PRN does not interfere with their ability to target CTL-mediated lysis in an antigen-specific manner. Nevertheless, it is possible that PRN could adversely affect the ability of APCs to effectively stimulate CTLs in culture. The results depicted in <xref ref-type="fig" rid="f2-ehp0114-001059">Figure 2</xref> demonstrate that the responses between clone 33 CTLs stimulated with EtOH-treated (<xref ref-type="fig" rid="f2-ehp0114-001059">Figure 2A</xref>) or PRN-exposed (<xref ref-type="fig" rid="f2-ehp0114-001059">Figure 2B</xref>) N1 cells are similar in their responsiveness to N1 targets, indicating that PRN exposure also does not interfere with antigen presentation in a stimulatory capacity. Thus, the antigen presentation characteristics of N1 cells both as stimulators and as targets for clone 33 CTLs are unaltered by PRN exposure.</p><p>The absence thus far of any adverse effects of PRN exposure on antigen presentation, however, could be misleading because of the large number of potential pMHC complexes on N1 cells that can be engaged by the clone 33 T-cell receptors (TCRs). Previous studies have shown that minimally approximately 50–200 pMHC complexes need to be engaged for a CTL effector to lyse its target (<xref rid="b9-ehp0114-001059" ref-type="bibr">Christinck et al. 1991</xref>; <xref rid="b36-ehp0114-001059" ref-type="bibr">Sykulev et al. 1994</xref>); it is likely that many more pMHC complexes are formed and available for engagement by the clone 33 TCR on the N1 cells. Thus, if PRN exposure only partially interferes with antigen processing or presentation, its adverse effect may be masked under these <italic>in vitro</italic> assay conditions. To circumvent this problem, we replaced N1 cells as targets in the CTL lysis assay with untransfected EL4 cells plus titrated amounts of the target peptide VSV-N p52–59 (<xref ref-type="fig" rid="f3-ehp0114-001059">Figure 3</xref>). With the addition of lower peptide concentrations during this titration assay, fewer pMHC complexes will be formed. If there is a defect in antigen presentation, it should become apparent at the lower peptide concentrations.</p><p>As shown in <xref ref-type="fig" rid="f3-ehp0114-001059">Figure 3</xref>, we observed significant lysis even when a peptide concentration as low as 7.8 pM is added to EL4 targets, yet there is no significant difference in the level of activity in PRN-exposed groups compared with the EtOH-treated control group. With the higher concentrations of added peptide (up to 32 μM), the apparent difference between EtOH-treated and PRN-exposed EL4 targets is not significant. And even so, the critical point to be made is that at lower concentrations with fewer surface pMHC complexes formed, the sensitivity of the assay is much greater, and at peptide concentrations < 125 pM, the experimental and control groups are virtually indistinguishable (<xref ref-type="fig" rid="f3-ehp0114-001059">Figure 3</xref>). Thus, there is no observable effect of PRN exposure on <italic>in vitro</italic> antigen presentation, and we concluded that APCs exposed to PRN are unimpaired in their ability to serve both as stimulators and as targets for peptide-specific CTLs.</p><p>In the next phase of our study, we addressed whether there is a direct effect of PRN exposure on CTL reactivity alone. One potential complication with the peptide-specific CTL model system is that when peptide is added to <italic>in vitro</italic> cultures, it can bind to MHC molecules expressed on the surface of CTLs themselves as well as to those on the APCs, thereby complicating the interpretation of experimental results. To circumvent this problem we used an alloreactive B6 anti-BALB/c MLC model system to examine the effect of PRN on CTL proliferation and function.</p><p>In this model the alloreactive B6 CTLs respond directly to the allogeneic class I MHC molecules expressed on the BALB/c stimulator cells without the need for added peptide. As shown in <xref ref-type="fig" rid="f4-ehp0114-001059">Figure 4B</xref>, alloreactive CTLs exposed to PRN during 1° MLC are largely unaffected in their lytic reactivity, with only a limited decrease in reactivity observed at the highest (66 μM) concentration tested. Although there is a 2.5-fold difference between the response of the 66 μM PRN-exposed group and control CTLs, given the similarities in magnitude of their overall response, this apparent difference is probably minimal. We also noted the concomitant observation that the proliferative responses, as measured by <sup>3</sup>H-TdR uptake in the <italic>in vitro</italic> MLR assay, among all three PRN exposure groups are not significantly different from the EtOH control group (<xref ref-type="fig" rid="f4-ehp0114-001059">Figure 4A</xref>). These findings are similar to those reported initially by Barnett and co-workers (<xref rid="b3-ehp0114-001059" ref-type="bibr">Barnett et al. 1992</xref>; <xref rid="b2-ehp0114-001059" ref-type="bibr">Barnett and Gandy 1989</xref>) and seem to support their suggestion that PRN might have very little, if any, effect on cell-mediated immunity.</p><p>We next examined the impact of prolonged PRN exposure on these alloreactive CTL effectors by adding fresh PRN during their restimulation in 2° <italic>in vitro</italic> MLCs. The overall effect of this longer-term exposure to PRN is the appearance of a significantly increased adverse effect on both CTL proliferation and lytic activity. Although PRN exposure during 1° MLC activation has only a limited effect on these parameters, the subsequent exposure to PRN during 2° MLC activation has a much greater adverse effect on CTL effectors. Thus, the exposure of 2° MLC-derived CTLs to 66 μM PRN induced an almost 70-fold greater inhibition than the same PRN concentration used in 1° MLCs. This finding establishes the importance of long-term symptoms in the diagnosis and management of immunotoxic effects resulting from exposure to environmental contaminants, particularly with respect to repeated or chronic exposure over an extended time period.</p><p>Even more striking is the nearly complete abrogation of CTL function of 2° CTLs that were exposed to PRN in the 1° MLC but not to additional PRN upon 2° stimulation. The nearly 8-fold difference in proliferation plus greater than 20-fold difference in lytic activity upon restimulation of PRN-exposed 1° CTLs into 2° CTL effectors indicates that some early PRN-induced adverse event(s) must have occurred during 1° activation that has a greater long-term functional impact on cell-mediated immunity. Thus, the studies reported here highlight the dramatic differences between acute and chronic exposure effects as an important consideration when assessing the immunotoxic potential of environmental agents. This often-neglected yet clinically important parameter could have significant diagnostic and treatment ramifications for the detection and management of pathologic anomalies associated with exposure to environmental contaminants. It is also worth noting that the delayed appearance of immunotoxic effects after PRN exposure in the 1° MLC might provide insight into questions of environmentally relevant doses or concentrations used in studies of various toxicants. This finding suggests that the lack of observable toxic effects might be due to a delayed onset in the appearance of such effects rather than to the administration of an inadequate toxic dose.</p></sec>
TRI: Corroding Its Original Intent?	Could not extract abstract	<contrib contrib-type="author"><name><surname>Bazilchuk</surname><given-names>Nancy</given-names></name></contrib>	Environmental Health Perspectives	<p>If knowledge is power, as the proverb goes, then the EPA’s Toxics Release Inventory (TRI) is a powerful tool indeed. Firefighters and first responders used this nearly 20-year-old public database of toxic chemical emissions to identify potential contamination hot spots after the floods of Hurricane Katrina. Residents have used it to find out what kinds of pollutants are being emitted by nearby industries. Investment companies use it to evaluate whether or not to purchase a company’s stocks. Even the Internal Revenue Service uses it to collect a pollution tax from companies that release ozone-damaging chlorofluorocarbons.</p><p>Given the TRI’s extensive use, it should come as no surprise that an EPA proposal to streamline TRI regulations for the 23,000-plus facilities that report under the law has proved highly controversial. The EPA’s plan must be reported, a move that critics say would affect the value of the TRI database for the public at large. But proponents argue that the cost savings that businesses would realize from the relief in paperwork would justify any loss of data.</p><p>The arguments matter, because the power of the TRI lies in the information it provides. Authorized by the 1986 Emergency Planning and Community Right-to-Know Act, the TRI doesn’t limit emissions of the more than 650 chemicals it now covers, but merely requires that they be reported by the companies that manufacture, use, or process them. However, when residents find out what is discharged by industries in their neighborhoods, they can and have used the facts to force change. Companies have altered their practices when managers see their facilities top the list for particular chemical discharges. In fact, the myriad of uses of the TRI, and its success in influencing business practices, has surprised both supporters and opponents of the original law [see “Now That You Know,” <italic>EHP</italic> 105:38–43 (1997)].</p><p>Between 1998 and 2004, the latest year for which data are available, the industries and federal facilities that report TRI data have voluntarily cut total on- and offsite disposal and other releases of TRI chemicals to the air, water, and land by 45%, or some 3 billion pounds. Since 1988, industries have cut releases of the 299 chemicals covered by the original law by nearly 60%. Because the TRI is so different from traditional end-of-pipe regulatory programs, which put limits on how much pollution can be released, it has drawn widespread praise. “Any program that the States, the Sierra Club and Monsanto can all praise is no doubt a true environmental success story,” wrote 12 state attorneys general in their comments on the proposed TRI changes.</p><sec><title>117,000 Comments and Counting</title><p>The proposed changes would increase from 500 pounds to 5,000 pounds the threshold at which facilities would be allowed to use a brief certification form (Form A) instead of a detailed reporting form (Form R) to report on their toxic chemical waste. This threshold is based on the amount of chemical wastes handled by the facility, not the amount released to the environment. Form R requires a complete accounting of a chemical’s fate—the amount on the site; the amount released to the land, air, or water as emissions; the amount recycled or burned for energy recovery or destruction; and the amount shipped from the plant for treatment or disposal. In contrast, Form A simply certifies that a toxic chemical was used at the facility in at least the regulatory threshold amount, but provides no other details.</p><p>The EPA’s plan also contains changes regarding a special subset of 20 chemicals and chemical compounds including mercury, lead, and polycyclic aromatic compounds. Previously, none of these “persistent, bioaccumulative, and toxic” (PBT) chemicals could be reported on Form A. Under the new rule, however, a company may file Form A for PBTs if 500 pounds or less is recycled, used for energy recovery, or treated for destruction. If any amount is released or emitted, however, the company must still use the detailed form. Furthermore, dioxins must still always be reported on the detailed form.</p><p>In a separate filing, the EPA notified Congress that it is considering changing the frequency of TRI reporting from yearly to every other year. Even though there has been considerable response to this third proposal, there has been little substantive debate. Federal law requires the EPA to warn Congress a year before beginning rule making on TRI reporting frequency, so the agency is still developing the details for this proposal.</p><p>When the EPA’s proposed threshold and PBT changes were published in the 4 October 2005 issue of the <italic>Federal Register</italic>, they unleashed a flood of responses—some 70,000 responses by the 13 January 2006 deadline for public comments. Even after the deadline passed, the response didn’t stop; more than 117,000 comments have been filed with the federal agency to date.</p><p>Twelve state attorneys general have called on the EPA to abandon the proposal, and a half-dozen U.S. senators and more than 50 U.S. representatives have also written the agency to question the assumptions of the plan. Recalling the TRI’s genesis in the aftermath of the 1984 Bhopal industrial disaster, Representatives Stephen Lynch (D–MA), Henry Waxman (D–CA), and Dennis Kucinich (D–OH) wrote that the plan “is particularly troubling” in view of a recent petrochemical plant explosion in China that ultimately polluted the drinking water supply for millions of people. The congressmen noted that the EPA’s own analysis showed that allowing industries to use the higher threshold of 5,000 pounds for Form A would allow companies nationwide to release a total of 246,092 pounds of benzene—without reporting the release.</p><p>Industry and small business community representatives have countered, however, that the EPA’s proposals meet the intent of the law while saving companies time and money (the TRI already has a small business exemption that allows facilities with fewer than 10 employees—including farms, dry cleaners, and others—to completely skip reporting and data collection). The U.S. Small Business Administration’s Office of Advocacy has been among the most vocal proponents for the changes, arguing that the expanded use of Form A is exactly the kind of incentive that will encourage good waste management.</p><p>“The current program does not reward the best environmental performers,” says Kevin Bromberg, assistant chief counsel for environmental policy at the Office of Advocacy. “Under the current system, if you run a facility with perfect chemical management techniques and discharge no highly toxic chemicals, you must still fill out the long Form R. Small businesses that are top environmental performers should be rewarded through less paperwork—the short Form A.”</p></sec><sec><title>Saving Money, Same Data?</title><p>A change in reporting thresholds clearly changes the amount of detail available from the TRI; the question is how this change affects the utility of the inventory. For example, the EPA has stated that none of the detailed data now reported for 26 chemicals or chemical classes (such as chromium compounds) would be available under the proposed 5,000-pound limit for non-PBT chemicals. Most of the chemicals for which detailed reporting would be lost are pesticides.</p><p>But the EPA claims that Form A reports will remain meaningful because the public will still know that the chemical is present at a facility at levels under the proposed thresholds. “The Form A certifications for these chemicals will provide a range by which waste management quantities and practices may be estimated,” the agency wrote in its proposal.</p><p>All told, the EPA estimates that the two threshold changes for Form A would save companies a combined total of about 164,000 hours a year and about $7.4 million in filing costs. The EPA’s economic analysis estimates the annual savings at the facility level for each form avoided is approximately $430 for each non-PBT chemical and $790 for each PBT chemical—or between $2 and $4 per day. This savings would come at a loss of detailed information on more than 12,000 releases and disposals of chemicals around the country, which total 14 million pounds of non-PBT chemicals released to the environment—just 0.34% of the total amount released. Given the PBT chemical exception, however, the EPA proposal permits no loss of such information for releases of those chemicals into the environment.</p><p>These savings free up environmental managers to focus on solving problems instead of filling out forms, according to Jeff Gunnulfsen, manager of government relations for the Synthetic Organic Chemical Manufacturers Association, a trade group that supports the changes. “Most of our members may have one regulatory person handling many, many issues such as hazardous waste, TRI, air issues, safety, and FDA, so any burden reduction may help them focus on more pressing matters,” Gunnulfsen says.</p><p>Still, official comments filed by several companies suggest that not everyone in the business world thinks the changes will save time or money. Under the law, companies must still track the same information and make the same calculations, even if they end up filing the short form. The company must be able to demonstrate to the EPA, if ever called upon, that they know their forms to be correct.</p><p>Indeed, in comments submitted in response to the <italic>Federal Register</italic> notice, Mark Herwig of GE Corporate Environmental Programs wrote, “An analysis of TRI data from 2003 suggests that EPA’s estimated burden reduction resulting from the proposed rule could be overstated by over 50% for all facilities. . . . There are several areas of EPA’s burden analysis that need improvement to accurately characterize TRI reporting burden.” According to a fact sheet compiled by OMB Watch, a nonprofit government watchdog group, many other corporations have expressed similar feelings.</p><p>Sean Moulton, director of federal information policy for OMB Watch, says communities lose even if just a small percentage of the total data is lost. For example, because mining and electric utilities report extremely large emissions to the TRI, “they swamp everything,” Moulton says. “In comparison to national totals, releases in Delaware may look small. But if you live in Delaware and are looking for what might affect me and my family, then Delaware is huge.” He adds that many of the chemicals tracked under the TRI—such as arsenic and benzene—are dangerous even in small quantities. So focusing strictly on the relative low number of pounds lost may be a poor measure of the situation.</p><p>Mike Flynn, director of the EPA’s Office of Information Analysis and Access within the Office of Environmental Information, which oversees the TRI, says the effect of the changes on communities is an issue the agency takes very seriously.</p><p>“The goal is to provide information for communities—that is an important central tenet,” Flynn says. But 99% of the data would still be available, he adds, and data losses would be offset by the “clear benefits in providing incentives for these companies to cut their emissions more. This is one of the issues where we have to find the right balance.”</p></sec><sec><title>State Program Effects</title><p>Some states have reacted strongly to the EPA proposal, partly because their pollution prevention and monitoring programs rely on the data provided by facilities for the TRI.</p><p>For example, in Washington state, if the 5,000-pound threshold is implemented for non-PBT chemicals, 40% of all chemicals now filed on Form R could be reported on Form A, which would include a loss of detail about the fate of 46,000 pounds of carcinogens, says Idell Hansen, TRI coordinator for the Washington State Department of Ecology. “We will only have the name of the chemical and the location of the facility, and we’ll lose all ability to track that chemical,” she says. “Under the proposed rule, we’d lose all information on eight of the top forty facilities with the greatest relative risk based on 2002 [TRI] data,” including data on some of the highest-risk chemicals such as methyl isocyanate—the chemical behind the Bhopal incident.</p><p>An analysis by the nonprofit National Environmental Trust showed that roughly 900 zip codes nationwide—10% of those that are home to a TRI reporting facility—would lose all numerical toxic emissions data. The New York State Attorney General’s office explored the impacts of this loss on 45,000 residents in Tonawanda, New York, a Lake Erie community surrounded by several industrial facilities. According to that analysis, changed thresholds would mean that this one community would be subject to unreported releases of 8,100 pounds of neurotoxic chemicals and 3,100 pounds of chemicals that cause respiratory problems, among other releases.</p><p>Jessica Emond, an EPA spokeswoman, says it is important to realize that even if a chemical release is not reported to the TRI, the release is almost always regulated by other environmental laws that protect air and water quality (although Moulton points out these limits frequently apply to only a single medium, such as just water or just air, leaving a loophole for releases to other media). “The EPA sets a high bar for companies,” Emond says. “Even with proposed changes, this doesn’t affect the amount of chemicals that a company would be allowed to release under state and federal laws.”</p><p>The EPA’s timetable calls for finalizing its proposed rule changes by December 2006. However, congressional action before then might preempt the agency’s rule making. Three U.S. senators have asked the Government Accountability Office to examine the EPA’s proposal. Additionally, in mid-May the House of Representatives approved an amendment to the Interior Appropriations Bill that would prevent the EPA from spending money to finalize the proposal until October 2008. The fate of that amendment will be decided in conference committee later this year.</p></sec>
Guest Editorial: Autism and the Environment	Could not extract abstract	<contrib contrib-type="author"><name><surname>Daniels</surname><given-names>Julie L.</given-names></name></contrib><aff id="af1-ehp0114-a00396">University of North Carolina School of Public Health, Chapel Hill, North Carolina, E-mail: <email>[email protected]</email></aff>	Environmental Health Perspectives	<p>Speculation that the environment plays a role in the development of autism primarily comes from two observations: <italic>a</italic>) although concordance among monozygotic twins is high, it is not perfect, and a specific “autism gene” or set of genes has not yet been identified; and <italic>b</italic>) the prevalence of autism is higher than previously thought—if it is rising, the rise might be associated with a shift in the environment.</p><p>Autism is a complex neurodevelopmental disorder defined by impaired social interaction, communication deficits, restricted interests, and repetitive behavioral patterns. These traits can range from mild to very severe, and may be accompanied by cognitive impairment and other comorbidities. The autism spectrum disorder (ASD) classification includes three disorders: autistic disorder, Asperger disorder, and pervasive developmental disorder not-otherwise-specified; however, there is no evidence that these diagnostic labels represent etiologically homogeneous groups.</p><p>The high concordance rates among monozygotic twins and recurrence in families support a strong genetic contribution to ASDs (<xref rid="b2-ehp0114-a00396" ref-type="bibr">Bailey et al. 1995</xref>; <xref rid="b4-ehp0114-a00396" ref-type="bibr">Folstein et al. 1977</xref>; <xref rid="b6-ehp0114-a00396" ref-type="bibr">Ritvo et al. 1985</xref>; <xref rid="b8-ehp0114-a00396" ref-type="bibr">Steffenburg et al. 1989</xref>). There is also a growing acceptance that subtle autism-like traits, such as atypical communication and aloof personality style, more commonly cluster in the nonautistic family members of individuals with autism than in the general population (<xref rid="b5-ehp0114-a00396" ref-type="bibr">Murphy 2000</xref>). The segregation of the milder traits in family members may indicate the presence of some, but not all, of the factors (genetic or environmental) necessary to develop an ASD.</p><p>To date, no specific genes or combination of genes have been consistently associated with autism. Discrepancies in gene-discovery studies might be, in part, because ASDs result from a variety of gene–gene and gene–environment combinations. Despite the lack of a specific genetic mechanism, most researchers agree that the etiology of autism is heterogeneous and polygenetic, and for some susceptible individuals, might involve environmental triggers.</p><p>Much of the concern surrounding environmental factors and autism comes from the perception that the prevalence of autism is increasing. There has clearly been a rise in the number of individuals who are actually diagnosed with an ASD; however, there are few systematically collected data in the same population over time that can be used to evaluate true prevalence rate trends (Fombonne 2003; <xref rid="b7-ehp0114-a00396" ref-type="bibr">Rutter 2005</xref>). Many factors could contribute to increases in prevalence estimates over time, including changes in diagnostic criteria, increasing availability of specialized diagnostic tools, improved case ascertainment, and true changes in the prevalence.</p><p>Real shifts in prevalence could result from environmental changes. Systematically monitoring temporal ASD prevalence trends in the same population over time is a necessary step to identifying true changes in prevalence. However, ecologic associations between environmental changes and rising autism rates are not sufficient to infer causation for such a complex disorder.</p><p>It is unlikely that one or even a few specific environmental agents are responsible for the majority of ASDs. It is more likely that some individuals have enhanced susceptibility to insults from the environment that may, in combination with their genetic predisposition, lead to autism. It is rarely possible to distinguish these complex relationships by simply evaluating trends in the general population.</p><p>The much publicized concern over vaccines and autism has primarily been based on such ecologic trends. More rigorous studies evaluating vaccine-related hypotheses are needed to incorporate individual-level exposure data, account for alternate exposures to metals, and evaluate susceptible subgroups of the population. However, attention should also be given to other environmental hypotheses.</p><p>Other environmental exposures found to be associated with autism include thalidomide, valproic acid, and infections such as rubella (<xref rid="b1-ehp0114-a00396" ref-type="bibr">Arndt et al. 2005</xref>; <xref rid="b3-ehp0114-a00396" ref-type="bibr">Chess 1971</xref>). These relatively rare exposures have been evaluated in small studies that have reported subtle effects. Yet, such findings support the plausibility that exposure to an environmental agent during a critical window of development can be associated with development of an ASD. The characteristic traits of autism are rarely distinguished before 2–3 years of age, but the cascade of events that leads to autism probably occurs much earlier, most likely during early gestation. Research focused on environmental exposures during critical periods of neurodevelopment should be prioritized.</p><p>Little is currently known about the etiology of autism, except that it is complex and multifactorial. The interaction between genetic and nongenetic factors during critical periods of neurodevelopment warrants further investigation. Until specific susceptibility genes are discovered, the identification of environmental risk factors that primarily affect susceptible subgroups may require us to refine ASD subgroup classifications using specific phenotypic patterns or the clustering of ASDs in families.</p><p>Given the complexity of autism, we will not find a magic bullet (genetic or environmental) to blame for most cases. There are probably many combinations of genes and environmental factors that contribute to the constellation of autistic traits. Future investigations of hypotheses involving environmental exposures need to carefully characterize cases, improve exposure assessment, focus on critical windows of neurodevelopment, and ensure sufficient power to conduct subgroup analyses and assess interactions. These considerations have been accommodated in a few well-planned epidemiologic studies that are, or soon will be, in progress. As we await advances in genetic and behavioral research, these studies offer hope for advancing our understanding of the potential role environmental factors play in the development of autism.</p>
Women’s Health: Endometriosis and PCB Exposure	Could not extract abstract	<contrib contrib-type="author"><name><surname>Potera</surname><given-names>Carol</given-names></name></contrib>	Environmental Health Perspectives	<p>Endometriosis may be related to exposure to persistent pollutants such as polychlorinated biphenyls (PCBs), according to research published in the May 2006 issue of <italic>Chemosphere</italic>. This gynecological disorder linked to infertility afflicts 10% of U.S. women of reproductive age. Researchers measured blood PCB levels in women undergoing laparoscopy for suspected endometriosis or other gynecological conditions. Higher levels of PCBs were detected in women with histologically confirmed endometriosis compared with controls.</p><p>Toxicologist Elena De Felip of the Istituto Superiore di Sanità in Rome and her colleagues measured 11 PCB congeners that are most abundant in human tissue. In 80 women aged 20 to 40, the sum of all congeners was 1.6 times higher in the 40 women diagnosed with endometriosis than in controls. Three congeners, PCBs 138, 153, and 180, were particularly higher in women with endometriosis. These three congeners have been reported to have estrogenic activity and to interfere with hormone-regulated processes.</p><p>PCBs have been used since the 1930s, mainly in electrical equipment. Although no longer manufactured, these persistent chemicals accumulate in the food chain; today meat, fish, eggs, and milk are chief sources of PCBs. But diet seems unable to explain the difference in PCB levels detected in the two groups of women, since “the dietary habits of the women were basically the same,” says De Felip.</p><p>De Felip suspects that differences in how women detoxify and eliminate PCBs from the body may explain the disparity. These processes are mediated by polymorphic enzymes; therefore, she says, differences in toxicokinetic activity may represent the basis for the higher concentrations detected in women with endometriosis and may also be related to higher or lower susceptibility to that condition.</p><p>Studies of PCBs and endometriosis face several limitations. Researchers typically measure only a few widespread congeners that are selected because of their toxicological activities, including an association with cancer shown in animal models. “So we’re only getting part of the picture,” says Germaine Buck Louis, chief of the Epidemiology Branch at the National Institute of Child Health and Human Development.</p><p>In a study described in the January 2005 issue of <italic>Human Reproduction</italic>, Buck’s team measured 62 congeners in 84 women undergoing laparoscopy. Levels of 4 antiestrogenic congeners were 3.77 times higher in women diagnosed with endometriosis than in controls. “We don’t fully understand the role of estrogenic and antiestrogenic PCBs,” she says, but complex interactions of many PCBs as well as other chemicals may be involved in developing endometriosis.</p><p>Recent advances in PCB detection methods allow more congeners to be measured at lower concentrations. “Women with endometriosis may have low levels of a particular congener not found in [other women],” says Louis. Moreover, breastfeeding reduces PCB levels in women, so women without endometriosis may have lower blood levels because they become pregnant and breastfeed more often. “There’s no ideal comparison group for endometriosis studies,” says Louis, and “there are no easy answers.”</p>
Chemical Exposures: No Dental Dilemma for BPA	Could not extract abstract	<contrib contrib-type="author"><name><surname>Josephson</surname><given-names>Julian</given-names></name></contrib>	Environmental Health Perspectives	<p>Among the many uses of bisphenol A (BPA) is the manufacture of resin-based dental composites and sealants. Recently a team of researchers from the CDC sank their teeth into questions about whether BPA monomer leaching from sealants could be harmful to people. The results of their human study, presented in the March 2006 issue of the <italic>Journal of the American Dental Association</italic>, suggest that although leaching does occur, sealants are still a safe means of preventing dental cavities.</p><p>Low-level exposures to BPA monomer in pregnant rodents, at a level that humans could potentially receive from dental sealants, have been shown to disrupt reproductive development in their fetuses, and concerns have emerged about the possibility of human health effects from dental exposures. Scientific exploration of this question has yielded inconsistent results, says Renée Joskow, first author of the March paper. Much of this is due to limitations in laboratory detection and translation of animal studies to human health effects, as well as insufficiently addressing the parameters of exposure in a clinical dental setting.</p><p>The CDC team, led by Joskow (now of the U.S. Public Health Service) and Dana Barr, looked at 14 nonsmokers receiving their first resin-based sealants as part of their routine dental care. Each subject received one of two brands of dental sealant manufactured by two well-established dental equipment and material supply firms. Then their saliva and urine were tested for BPA.</p><p>All the patients had BPA in their saliva and urine, even before treatment. For patients receiving Helioseal F sealants, saliva BPA doubled immediately after treatment and returned to baseline within 1 hour. Urine BPA more than tripled 1 hour after treatment and returned to baseline within 24 hours. For patients receiving Delton LC sealants, saliva BPA increased nearly 126 times immediately after application and was still 23 times higher after 1 hour. Urine BPA jumped 10 times 1 hour after treatment and was still elevated 24 hours later. Both levels eventually returned to baseline.</p><p>Barr believes the patients’ baseline BPA came from background exposures from environmental sources such as water and food packaging. These, she suggests, could be “a more chronic low-level source of exposure” than dental sealants. Barr adds that in her view, although point-source exposure from dental sealants might approach levels that induce health effects in rodents, “[it] is not the most significant source of exposure in humans.” Moreover, she holds that exposure to BPA from dental sealants, already variable and short-lived in the body, could be easily reduced further by having the patient spit frequently in the hours after application.</p>
Disease-First: A New Paradigm for Environmental Health Science Research	Could not extract abstract	<contrib contrib-type="author"><name><surname>Wilson</surname><given-names>Samuel H.</given-names></name><degrees>MD</degrees></contrib><aff id="af1-ehp0114-a00398">Deputy Director, NIEHS and NTP E-mail: <email>[email protected]</email></aff>	Environmental Health Perspectives	<p>In recent years, many observers have advocated the adoption of a new paradigm in the environmental health sciences—a shift toward a sharper focus on understanding human disease and improving human health by integrating knowledge from environmental health research with that from the broad spectrum of medical research. We term this the “disease-first” approach. The first step in this approach is to prioritize specific common diseases according to the public health burden they pose. Next researchers will gather information on molecular changes that accompany the pathogenesis of each condition, including cellular and tissue changes that occur over time. We will then work to link these biological responses to environmental exposures including toxicants, metals, toxins, and lifestyle and dietary factors that eventually lead to disease. The fundamental goal of the NIEHS is to learn how this knowledge can be used to reduce morbidity and extend longevity. We believe the disease-first approach will allow our field to greatly reduce the burden of human disease.</p><p>The common diseases that account for the bulk of the public health burden of disease are chronic, disabling, and widely prevalent. Diseases such as asthma, obstructive lung diseases, and diabetes mellitus are leading causes of death in the United States and are increasing in incidence. These diseases are multifactorial and it is recognized that environmental exposures are a key risk factor in all of them, but large gaps exist in our ability to characterize the complex associations among environmental factors, genetic factors, and health outcomes. Gene–gene, gene–environment, and gene–vector–environment interactions all play important roles in any given condition. Thanks to advancing technologies in areas such as genomics, proteomics, and metabolomics, as well as in computational biology and bioinformatics, we now have the tools to redefine exposure–response relationships, develop quantitative models to support risk assessment, and reach more comprehensive understanding of the diverse and complex responses leading to pathogenesis. One of the key elements of the disease-first approach is to more fully leverage knowledge from these burgeoning technologies to vastly improve clinical outcomes.</p><p>The first priority of the disease-first approach is to more systematically track population health status in the United States, both temporally and spatially. Improved surveillance of health status can point to differences in exposures, which can be used as starting points for causality research. Although a comprehensive health status monitoring system does not yet exist, recent developments at the federal level suggest that it could soon be a reality [see Kyle et al., p. 980 this issue].</p><p>Such monitoring will allow for human body burden measurements of hazardous substances and xenobiotic metabolites. The disease-first approach will integrate this information with ongoing measurements from traditional environmental exposure assessments. These might include measurement of atmospheric pollutants and research on the fate and transport of hazardous agents in the ecosystem. The NIEHS will develop a number of exposure-based bioassays that will classify exposures (e.g., metals, chemicals, air pollution, diet) and link them to risk factors for developing clinical disease or to early steps in a disease process. Use of improved animal models of human disease will enable research on temporal aspects of dose–response relationships for harmful agents or lifestyle factors, facilitating the discovery of molecular signatures of thresholds separating the normal stress response from pathology. Such knowledge of thresholds will allow us to predict responses to stressors and to understand how individual susceptibility affects those responses.</p><disp-quote><p>We believe the disease-first approach will allow environmental health science to greatly reduce the burden of human disease.</p></disp-quote><p>Both population-based research and clinical research on individuals will be needed to ascertain correlations between “real world” exposures over ranges of dose, time, and disease, while accounting for genetic variations among subpopulations. By linking traditional exposure information with exposure responses, we can gain knowledge of diseases and early disease processes that will accelerate our understanding of both individual susceptibility and disease pathogenesis.</p><p>Many new tools will be required to accomplish these goals, some of which are now becoming available and others of which will soon be possible. We have seen the rapid maturation of novel, high-throughput analytical methodologies in the various “-omic” sciences, as well as advances in nanotechnology, imaging, and bioinformatics. Development in these areas will continue with support from both the public and private sectors. Improved understanding of disease at the molecular level and of pathophysiologic processes, along with research innovations such as RNA interference and subcellular imaging, will allow more precise analysis of mammalian pathophysiology. Geographic information systems and sensor-based technologies will enable greater precision in environmental and personal exposure monitoring, integrating previously unavailable data into research.</p><p>The NIEHS research portfolio will emphasize understanding of molecular mechanisms of pathogenesis and directly link environmental exposures with common diseases. We have established a new Office of Translational Biomedicine at the NIEHS to facilitate interdisciplinary and cross-disciplinary research endeavors, and we will seek to collaborate more broadly across the medical research community to maximize the potential gains from this approach.</p><p>This new paradigm for research in the environmental health sciences is our challenge and our vision for a healthier future.</p>
New Books	Could not extract abstract	Could not extract contributor	Environmental Health Perspectives	<p>Advances in Genetics</p><p>Jeffrey Hall, ed.</p><p>Burlington, MA:Elsevier, 2006. 224 pp. ISBN: 0-12-017656-4, $149.95</p><p>Artists-in-Labs: Processes of Inquiry</p><p>Jill Scott, ed.</p><p>New York:Springer, 2006. 136 pp. ISBN: 3-211-27957-1, $25.95</p><p>Bioinformatics: Genomics and Post-Genomics</p><p>Frédéric Dardel, François Képès</p><p>Hoboken, NJ:John Wiley & Sons, Inc., 2006. 272 pp. ISBN: 0-470-02001-6, $79.95</p><p>Data Mining for Biomedical Applications</p><p>Jinyan Li, Qiang Yang, Ah-Hwee Tan, eds.</p><p>New York:Springer, 2006. 155 pp. ISBN: 3-540-33104-2, $58</p><p>Enhancing Philanthropy’s Support of Biomedical Scientists: Proceedings of a Workshop on Evaluation</p><p>George R. Reinhart, ed.</p><p>Washington, DC:National Academies Press, 2006. 146 pp. ISBN: 0-309-10097-6, $29.25</p><p>Environmental Chemistry at a Glance</p><p>Ian Pulford, Hugh Flowers</p><p>Malden, MA:Blackwell Publishing, 2006. 144 pp. ISBN: 1-405-13532-8, $29.95</p><p>Environmental Impacts of Treated Wood</p><p>Timothy G. Townsend, Helena Solo-Gabriele</p><p>Boca Raton, FL:CRC Press, 2006. 520 pp. ISBN: 0-8493-6495-7, $139.95</p><p>Fundamental Molecular Biology</p><p>Lizabeth Allison</p><p>Malden, MA:Blackwell Publishing, 2006. 600 pp. ISBN: 1-405-10379-5, $109.95</p><p>Geoenvironmental Sustainability</p><p>Raymond N. Yong, Catherine N. Mulligan, Masaharu Fukue</p><p>Boca Raton, FL:CRC Press, 2006. 400 pp. ISBN: 0-8493-2841-1, $129.95</p><p>Human Developmental Toxicants: Aspects of Toxicology and Chemistry</p><p>James L. Schardein, Orest T. Macina</p><p>Boca Raton, FL:CRC Press, 2006. 472 pp. ISBN: 0-8493-7229-1, $159.95</p><p>Human Genetics and Genomics, 3rd ed.</p><p>Bruce R. Korf</p><p>Malden, MA:Blackwell Publishing, 2006. 488 pp. ISBN: 0-632-04656-2, $54.95</p><p>Mercury Hazards to Living Organisms</p><p>Ronald Eisler</p><p>Boca Raton, FL:CRC Press, 2006. 336 pp. ISBN: 0-8493-9212-8, $169.95</p><p>Natural Disasters as Interactive Components of Global-Ecodynamics</p><p>Kirill Ya Kondratyev, Vladimir F. Krapivin, Costas A. Varostos</p><p>New York:Springer, 2006. 579 pp. ISBN: 3-540-31344-3, $209</p><p>Pesticides: Health, Safety and the Environment</p><p>G. A. Matthews</p><p>Malden, MA:Blackwell Publishing, 2006. 248 pp. ISBN: 1-405-13091-1, $159.99</p><p>Principles of Gene Manipulation and Genomics, 7th ed.</p><p>Sandy Primrose, Richard Twyman</p><p>Malden, MA:Blackwell Publishing, 2006. 668 pp. ISBN: 1-405-13544-1, $94.95</p><p>Scaling and Uncertainty Analysis in Ecology</p><p>J. Wu, K. B. Jones, H. Li, O. L. Loucks, eds.</p><p>New York:Springer, 2006. 313 pp. ISBN: 1-4020-4662-6, $119</p><p>The Molecular Biology of Cancer</p><p>Stella Pelengaris, ed.</p><p>Malden, MA:Blackwell Publishing, 2006. 500 pp. ISBN: 1-405-11814-8, $74.95</p><p>The Regulatory Genome: Gene Regulatory Networks In Development And Evolution</p><p>Eric Davidson</p><p>Burlington, MA:Elsevier, 2006. 304 pp. ISBN: 0-12-088563-8, $69.95</p><p>To Recruit and Advance: Women Students and Faculty in U.S. Science and Engineering</p><p>Committee on the Guide to Recruiting and Advancing Women Scientists and Engineers in Academia, Committee on Women in Science and Engineering, National Research Council</p><p>Washington, DC:National Academies Press, 2006. 150 pp. ISBN: 0-309-10204-9, $33.26</p><p>Toxicity Testing for Assessment of Environmental Agents: Interim Report</p><p>Committee on Toxicity Testing and Assessment of Environmental Agents, National Research Council</p><p>Washington, DC:National Academies Press, 2006. 270 pp. ISBN: 0-309-10092-5, $36</p>
EHPnet: Cure Autism Now	Could not extract abstract	<contrib contrib-type="author"><name><surname>Dooley</surname><given-names>Erin E.</given-names></name></contrib>	Environmental Health Perspectives	<p>In 1995, parents of children with autism joined together to form the nonprofit organization Cure Autism Now (CAN). Since then, its membership has grown to include clinicians and scientists committed to accelerating the pace of biomedical research in autism. CAN raises and distributes funds for research on the causes, prevention, and treatment of autism, as well as education and outreach. As a resource for everyone interested its work, CAN has a website located at <bold><ext-link ext-link-type="uri" xlink:href="http://www.cureautismnow.org/">http://www.cureautismnow.org/</ext-link></bold>.</p><p>So far, CAN has committed over $25 million to research funding and has established and continues to support the Autism Genetic Resource Exchange (AGRE). Clicking on the Research link at the top of the CAN homepage takes visitors to an overview of the CAN science program, which includes six initiatives that the group believes will yield the most effective treatment for individuals with autism.</p><p>The Genomics Initiative focuses on gene mapping and microarray work. CAN’s goal is to identify several genes involved in autism within the next three years. Closely related to the Genomics Initiative is the AGRE, an open gene bank with a large collection of immortalized cell lines and DNA samples gathered from families with more than one autistic child. Available on the AGRE page is a link to research updates published since 2001.</p><p>The goal of the Innovative Technology for Autism Initiative is to stimulate development of products that provide realistic solutions to the issues encountered by those with autism, their families, educators, health care specialists, and researchers. The initiative offers multiyear grants, fast-track “bridge” grants, and educational programs. It also sponsors a workgroup within which investigators can meet, share, and collaborate, and which also serves to actively bring new investigators into the field.</p><p>One major hurdle that autism researchers are working to overcome is the lack of any biomarker for diagnosis. The CAN Biomarkers Initiative has yielded two preliminary findings of possible autism bio-markers—one a novel protein in the urine of children with autism and some of their unaffected relatives, and the other a distinct lipid profile that was seen in 20 AGRE samples. CAN has launched a study in an effort to replicate and confirm these results.</p><p>In the past few years, new findings on neuroplasticity, the ability of the brain to grow and change throughout life, have led to significant breakthroughs in the treatment of stroke and dyslexia through a process called neural retraining. To apply these same ideas to the treatment of autism, CAN has established the Neural Retraining Initiative. The initiative’s first project, led by Michael Merzenich of the University of California, San Francisco, is working to design, produce, and test nonpharmaceutical tools and techniques, including one to prevent the emergence of full-blown autism in at-risk infants.</p><p>CAN has also awarded several grants through its Environmental Factors in Autism initiative to study the neurotoxicity of mercury and how it may factor in the development of autism. Thimerosal, which contains ethylmercury, has been widely used as a preservative in vaccines and other health and medical products, and has been raised as a potential contributor to autism.</p>
A Meeting of the Minds on Mice	Could not extract abstract	<contrib contrib-type="author"><name><surname>Hood</surname><given-names>Ernie</given-names></name></contrib>	Environmental Health Perspectives	<p>If genetics research is ever to fulfill its promise of revolutionizing medicine, genotypes must be linked to phenotypes—that is, individual genomic characteristics must be identified and associated with outcomes in the forms of disease susceptibility and/or development; individual responses to drugs, infectious agents, or environmental exposures; or other individual characteristics such as behavioral tendencies. Why does the person who never smoked develop lung cancer, while the three-pack-a-day smoker remains healthy? Why do some people become addicted to drugs, while other users are never hooked? Why does a particular medication work wonders in some people, but not work at all in others? These and countless similar questions represent the enormous challenge still facing researchers as they strive to make personalized medicine a clinical reality.</p><p>The answers to many of these questions may yet be discovered in the genomes of mice, our diminutive mammalian relatives. That’s certainly the hope and belief of the members of the mouse genetics community, 180 of whom gathered 6–9 May 2006 in Chapel Hill, North Carolina, for the fifth annual meeting of the Complex Trait Consortium (CTC), a loosely woven international organization tightly knit in its dedication to elucidating human characteristics by identifying their genetic counterparts in mice.</p><p>The “complexity” of complex traits derives from the fact that they are polygenic—multiple genes interact to cause these conditions, and the genes involved may not interact additively. Ninety-three reports presented at the CTC meeting updated progress in the hunt for the multiple genes and quantitative trait loci (or chromosomal “hot spots”) associated with a wide variety of complex traits such as heart failure, tumor resistance, obesity, drug and alcohol addiction, and schizophrenia. Sponsored by the NIEHS, the UNC–Chapel Hill, and Agilent Technologies, the conference brought together a diverse group of mouse geneticists, molecular biologists, statisticians, and bioinformaticists from 10 countries.</p><p>The CTC is all about collaboration and interaction. “It’s unquestionably the best meeting that I go to every year,” says Abraham Palmer, an assistant professor of human genetics at the University of Chicago. “The opportunities to communicate with other geneticists working in other fields and with the people who develop our methodology are critically important, and accelerate by months or even years the rate at which the field can move forward,” he says.</p><p>Karlyne Reilly, a principal investigator in the Mouse Cancer Genetics Program at the National Cancer Institute, agrees. “It brings together a wide variety of science around techniques and how you solve the problems that are common to these different areas,” she says. “I always come away with new tools to play with, that I can apply to my own research.”</p><sec><title>Building a Better Mouse Line</title><p>The CTC is presently at the midpoint of building a resource that should prove enormously valuable in the effort to associate genotypes with phenotypes. The Collaborative Cross (CC) is a carefully planned and controlled mouse recombinant inbreeding program that began in 2005 with eight genetically heterogeneous strains. Upon its expected completion in about four years, 1,000 lines closely modeling the breadth of human genetic diversity will have been generated. According to conference keynote speaker Jean-Louis Guénet, a professor emeritus of mouse genetics at the Institut Pasteur in Paris, it will be “one of the most important pages in the book of genetics of the future.”</p><p>Armed with several powerful new bioinformatic and biostatistical tools being developed specifically to take full advantage of the resource, the CC will enable researchers to hunt far more precisely and efficiently for the multiple genes and quantitative trait loci that constitute complex traits, and will allow the community to share and integrate their raw data sets far more effectively.</p><p>“The idea is to accumulate as much diverse data as possible for relatively fixed strains, what we call the ‘genetic reference population,’” says Robert Williams, a professor of anatomy and neurobiology at the University of Tennessee Health Science Center and one of the founders of the CTC. “The hope is that everybody will use their own tools—their own methods and their own phenotypes—but the Collaborative Cross will provide a way to bind those results together by using the same animal resource.”</p><p>According to conference co-organizer David Threadgill, an associate professor of genetics at UNC–Chapel Hill, the goal is for the CC to evolve to become “the central resource for experimental mammalian biology.” With a fixed genetic reference population and common tools, he says, “it will be the resource that everybody turns to,” because every piece of data collected through the CC will be immediately comparable to any other piece of data in the database.</p></sec><sec><title>CC Riding</title><p>The CC will enable a so-called systems genetics approach, as opposed to the traditional, laborious effort to identify one gene at a time. As Guénet points out, diseases that are the consequences of the alteration of a single gene—one example is cystic fibrosis—tend to be marginal in terms of frequency. However, polygenic diseases tend to be much more widespread, he says: “Next door to you, you probably have someone with asthma, dermatitis, or autoimmune disease. . . . So we have to work hard to understand the genetic determinants of these complex diseases, and presumably what we are going to learn from the mouse can be transposed to the human being, because we share ninety-eight percent of our genes with the mouse.”</p><p>Williams shares Guénet’s optimism about the tremendous potential of the CC to shed useful light on common human diseases. “You have to understand the function of the gene and its products in a complex milieu, in a mouse or human—not only a mouse or human, but many different mice and many different humans,” he says. “We think [the CC] will provide the resource to do that.” He adds that the ability to conduct experimental population-based research with the CC should allow much more comprehensive exploration of the genetics associated with gene–environment interactions.</p><p>That exploration will also be enhanced by the completion of a mouse genetics initiative undertaken by the NIEHS and Perlegen Sciences to identify the genetic variants in 15 diverse strains of laboratory mice, including SNPs (single-nucleotide polymorphisms), indels (insertions/deletions), and haplotypes (blocks of related SNPs). The database, a project of the recently established NIEHS Center for Rodent Genetics, is scheduled to be unveiled in September 2006, and is anticipated to be a rich and robust source of information for the mouse genetics community.</p><p>Signs of early but significant progress in the CC initiative were among the highlights of the meeting. Conference co-organizer John E. French, an NIEHS research physiologist, is encouraged by results emerging from pilot studies. “There’s at least been a proof of principle established that it’s going to be a very effective tool,” he says. “We are only seeing the beginning evidence of that—there’s a long way to go—but some of the promise has been identified and, I think, validated.” According to Williams, the pilot project is now of sufficient size (two recombinant inbred sets, LXS and BXD, with 80 member strains) that “it provides the community with a good flavor of what this will look like when we have an order of magnitude more strains than we do now.”</p><p>Threadgill is excited by the flavor that’s already emerging. “The major things that are starting to come out are the results of integrating data sets, integrating genetic variations, and integrating gene expression patterns,” he says. The new knowledge that’s coming out of that—the identity of new genes that are potential master modulators of genetic networks, and how those may actually also be very important for mediating disease processes—speak to the remarkable potential that will be realized when the CC is completed.</p></sec><sec><title>A Case in Point</title><p>Research results presented by Palmer on his group’s work at the University of Chicago illustrate the broad outlines of the types of studies being undertaken by mouse geneticists. Palmer and colleagues are investigating the genetic underpinnings of susceptibility or resistance to drug addiction; given today’s working definition of “the environment,” recreational drug use is fast becoming a xenobiotic exposure of great interest. An understanding of the genotypic differences between addiction susceptibility and resistance could lead to new targets for therapeutic drugs or preventive interventions.</p><p>The team selectively bred mice to have very high or very low sensitivity to locomotor stimulation, a particular behavioral effect of methamphetamine that is a characteristic animal response to drugs of abuse. They then measured the expression of more than 14,000 genes in a region of the animals’ brain known to be involved in response to the drug. Ultimately, they arrived at a candidate gene that was found to be very differentially expressed in the high- and low-sensitivity mice—casein kinase 1 epsilon (<italic>Csnk1e</italic>). It was a gene already known to be involved in locomotor stimulant response of animals to various drugs. But the question then became, was it important in humans?</p><p>Fortuitously, thanks to colleague Harriet de Wit of the University of Chicago Department of Psychiatry, Palmer had access to DNA from a cohort of 100 healthy human volunteers. In a double-blind study, the subjects received 0-, 10-, and 20-mg doses of amphetamine in a randomized order. Responses were measured by standardized questionnaires, and were then compared to results of genotyping tests, to see whether there was a correlation between response to the drug and polymorphisms in <italic>Csnk1e</italic>.</p><p>“We found a statistically significant association between this gene, <italic>Csnk1e</italic>, and people’s sensitivity to the euphoric effects of the drug,” says Palmer. “So the people with one genotype ‘got a buzz,’ while people with another genotype didn’t. We hypothesize that that may have implications for the likelihood of a person with one genotype who samples the drug to continue to use the drug, and that of course would put them at grave risk for developing an abusive relationship with the drug.”</p><p>Palmer suspects that polymorphisms in <italic>Csnk1e</italic> may also be important in a variety of other systems whose mechanisms might be similar to that of addiction. These include the manic phase of bipolar disorder and the use of stimulants to treat attention deficit/hyperactivity disorder.</p><p>“I think we’re now at a point where it’s just about to become easy to go from a phenotype to identifying some of the genes that are involved in that phenotype,” says Palmer. “To get all of them is going to take longer, and it’s going to require further refinements in our methodology, but I actually think that the story I told is going to become a common story. . . . In the same way that molecular biology took a long time to mature, and now is unbelievably central to the way we think about the progress of medicine and health sciences, I think this field of genetics is right at that turning point.”</p><p>Knowledge gained from the genomes of our mammalian cousins by groups like the CTC may provide the vital information to eventually usher in the much-anticipated era of personalized medicine. Says Threadgill, “What it really comes down to is being able to predict which individuals are going to be susceptible to certain environmental exposures or disease processes, which individuals are going to respond adversely to combinations of alleles, so that interventions and preventive medicine can be applied where they need to be applied, rather than in global fashion.”</p></sec>
A Clear Solution for Dirty Water	Could not extract abstract	<contrib contrib-type="author"><name><surname>Lougheed</surname><given-names>Tim</given-names></name></contrib>	Environmental Health Perspectives	<p>Turning water into wine may be among the most venerable of miracles, but for Greg Allgood, the real miracle has been turning dirty water into drinkable water. He once wowed an audience in a Malawi village, where hundreds of inhabitants along with the country’s Minister of Health watched him transform a sample of the only local source of drinking water. “There were gasps of excitement when the water turned from this horrible, muddy dark color to crystal clear and safe,” he recalls.</p><p>Allgood was demonstrating PUR<sup>™</sup>, a modest-looking packet of powder that quickly turns turbid, health-threatening water into the kind of liquid most of us would pay to drink out of a bottle. PUR was developed in the late 1990s by household products giant Procter & Gamble (P&G) and shares its name—but not its technology—with home tap water filters sold by that company in developed nations. Now PUR occupies a place at the forefront of P&G’s Children’s Safe Drinking Water Program, a philanthropic initiative that Allgood directs.</p><p>Allgood spends about a third of his time in places like Malawi where people have limited or no access to treated, potable water sources. Worldwide, as many as 2 billion people drink water extracted from shallow wells or polluted lakes and rivers, with nothing like the municipal treatment systems that are taken for granted in most of North America and Europe. In the few developing locales where such infrastructure might exist—and indeed, even in the richest nations on the planet—this resource can be ruined suddenly by a natural disaster like a hurricane, earthquake, or tsunami, creating an immediate, desperate, and widespread need for safe drinking water.</p><sec><title>The Stuff of Life</title><p>Water can be the key to keeping death and disease at bay. Hydration is fundamental to bodily functions, including the ability to retain nutrients. Infants, the elderly, and immunocompromised persons are especially vulnerable to dehydration caused by diarrhea, which is in turn spawned by bacteria or viruses acquired from tainted drinking water. In African countries ravaged by HIV/AIDS, large portions of the adult population could likewise succumb to even limited numbers of parasites found in relatively clean water. “While [a healthy person] might take a couple of weeks to get over <italic>Giardia</italic>, it could be fatal to a person that has a reduced immune system,” says Allgood. As opposed to dealing with these ailments once they appear, purifying water can keep them from appearing at all.</p><p>The CDC became interested in point-of-use treatment when cholera exploded in Peru in 1991 and spread rapidly throughout Latin America. A dependence on questionable drinking water lay at the heart of this epidemic, and the Pan American Health Organization estimated that it would take some $200 billion and more than a decade to install the necessary municipal infrastructure to alleviate the problem throughout the region. The CDC sought alternatives to help affected populations in the meantime.</p><p>Chlorine bleach was among the most widely available disinfectants, although people had difficulty gauging how much was needed to treat a given amount of water without creating an unpleasant taste or harmful concentrations. The agency therefore supported development of special bottles of dilute bleach—the bottle caps were designed to hold just the right amount of solution to safely treat one jerry can of water.</p><p>These efforts caught the attention of P&G, the leading manufacturer of bleach in many of the affected countries. But while this approach continues to be used in many parts of the world, it does not remove suspended material from the water, leaving users with water that is microbe-free but can still look dirty. So in the mid-1990s, P&G struck a formal Cooperative Research and Development Agreement with the CDC, focusing on how drinking water could be even better treated at the point of use.</p></sec><sec><title>Floccing Toward Solutions</title><p>P&G researchers tackled the challenge with flocculants, agents that promote molecular aggregation and can cause colloids or loose particles in a liquid to amass in clumps that sink to the bottom. Combined with large-particle calcium hypochlorite—essentially, powdered bleach—the result was PUR, a proprietary formulation that Allgood describes as reverse-engineering the municipal water treatment process.</p><p>Using PUR is like making a batch of powdered soft drink mix. Each packet of powder is designed to treat 10 liters of water. One simply tears open the packet, pours the powder directly into the water, and stirs. Within a matter of seconds, any floating material will start to flocculate into clumps that sink to the bottom. In no more than five minutes, all of the water is clear, and after standing for about 20 minutes, it will be completely disinfected. If desired, the solid remnants can be removed with the most basic of filters, such as a simple piece of cloth.</p><p>“The large particle size makes [the powder] slowly dissolve, so in essence it acts like a time-released formula of chlorine disinfectant,” Allgood says. “That’s important, because this product is meant to treat a huge range of waters, from clear to extremely contaminated.”</p><p>Even seasoned observers, including the scientists who initially refined and tested PUR, agree that its action is nothing less than dramatic.</p><p>“It was extremely impressive, and the most impressive thing about it was its simplicity,” notes John Perry, a microbiologist at Freeman Hospital in Newcastle upon Tyne, United Kingdom. He and his colleagues spent two years working closely with P&G, putting PUR through its paces in the laboratory.</p><p>“We would take a bucket of clean water and contaminate it with all sorts of things—lots of different types of bacteria, but also viruses, protozoan cysts, and they’d also put a lot of soil in it to mimic the kind of conditions that you get in the field,” Perry says. “We did a very detailed analysis of what came out at the end of the process, and all of these bacteria, viruses, and cysts had magically disappeared.”</p><p>These results were recounted in a paper coauthored by Perry that appeared in the June 2003 issue of the <italic>Journal of Water and Health</italic>. Other investigators have also published findings from applications of PUR in various settings, ranging from ongoing rural development activities in Kenya and Guatemala to crises like that in Haiti following Tropical Storm Jeanne in September 2004. Just a few months after Jeanne struck, various aid agencies purchased 13 million packets of PUR and transported them to parts of Sri Lanka, Indonesia, and the Maldives when they were struck by the great tsunami of December 2004.</p></sec><sec><title>One Option of Many</title><p>In addition to its humanitarian value in disaster relief, the product is also being marketed as a household commodity in many other parts of the world where large portions of the population lack reliable water treatment. The pricing of such a good varies widely from one market to another, based on what the local market will be thought to bear. Sally Cowal, a senior vice president with the Washington, DC–based nonprofit firm Population Services International (PSI), oversees the complex dynamics of advertising and selling PUR in different countries.</p><p>“Because we’re in social marketing, we have a great belief that if you pay for something, you’re much more likely to use it than if it’s handed to you,” she says. Of PSI’s alliance with P&G, she says, “We’re learning a lot from one another. They don’t know particularly well how to reach the bottom of the pyramid in the countries we work in; that’s what we know really well. But they know things about brands and brand management and sophisticated marketing and sales techniques that we [can] learn from them.”</p><p>Neither of these organizations present PUR as a single, definitive answer to water treatment under any and all circumstances. Eric Mintz, chief of the Diarrheal Diseases Epidemiology Section of the CDC’s Foodborne and Diarrheal Diseases Branch, points out that dilute bleach, membrane filters, and solar (ultraviolet) disinfection each have their appropriate niche.</p><p>“We think those all have a place, and they all have advantages and disadvantages,” Mintz says. “Allowing people to choose from different options is also good.” He notes that using PUR can be somewhat more expensive and cumbersome than other methods. For example, although the 13¢ needed to buy a packet of PUR in the Dominican Republic sounds cheap, this may be much more on a per-liter basis than a family would pay for the CDC’s dilute bleach treatment. Plus, the PUR system requires more components—two containers, a stirrer, a filter—than most other systems. The optimal option, Mintz adds, is undoubtedly the kind of built infrastructure found in the developed world.</p><p>But Steve Luby, who heads up the CDC’s work in Bangladesh, observes that much of the developing world has waited four or five decades for permanent water treatment systems to arrive. He argues that too many lives are at risk for measures such as PUR to be ignored.</p><p>“The numbers [of people at risk] are just huge, and if we wait to build infrastructure we’ll lose a generation,” he says. “We can do something good here, and it also gets people understanding the importance of water and the importance of <italic>clean</italic> water, and the need to actually invest in making water clean. We view this as a step toward community empowerment, toward central infrastructure solutions.”</p></sec>
Remember <italic>Pfiesteria</italic>?: Occupational Exposure Unlikely to Cause Cognitive Effects	Could not extract abstract	<contrib contrib-type="author"><name><surname>Tillett</surname><given-names>Tanya</given-names></name></contrib>	Environmental Health Perspectives	<p>Case reports have suggested that exposure to the dinoflagellate <italic>Pfiesteria</italic> may contribute to deficits in human learning and memory. Until now, however, there has been no clear, objective documentation of health effects associated with regular occupational exposure to this organism. The results of the first systematic, multiyear study of <italic>Pfiesteria</italic>’s human health effects now demonstrate that commercial fishermen (“watermen”) likely do not face significant health risks from routine occupational exposure to the organism <bold>[<italic>EHP</italic> 114:1038–1043; Morris et al.]</bold>.</p><p><italic>Pfiesteria</italic> is a common inhabitant of estuarine waters in the U.S. mid-Atlantic region in the summer and fall. In 1997, watermen working along the Pocomoke River, an estuary off Chesapeake Bay, experienced a pattern of neuropsychological deficits in association with fish kills linked to <italic>Pfiesteria</italic> outbreaks. Researchers studying <italic>Pfiesteria</italic> in a lab environment had reported similar memory and learning deficits.</p><p>Using a cohort of 88 healthy watermen with regular occupational exposure to Chesapeake Bay waters and 19 controls with minimal contact to the waters (matched to the watermen by zip code, age, and educational level), a team of Maryland researchers collected data over four summers, from 1999 through 2002. They questioned the subjects biweekly about symptoms like those reported in the 1997 episode and about their exposure to the waters and to known chemical toxicants. Subjects were tested at the beginning and end of each summer season on sensory and motor functions, attention and concentration, memory, visual functions, and verbal functions. In addition, the research team analyzed more than 3,500 water samples taken from Chesapeake Bay to monitor the presence of <italic>Pfiesteria</italic> and other harmful species.</p><p><italic>P. piscicida</italic> was found in water samples drawn from a number of locations in all four years of the study, and <italic>P. shumwayae</italic> (recently renamed <italic>Pseudopfiesteria shumwayae</italic>) was found in the last two years. However, the investigators found no decline in neurological function among the watermen in any year of the study.</p><p>The scientists note that unique, isolated instances of <italic>Pfiesteria</italic> outbreaks or unusually toxic strains of the dinoflagellate may have been associated with the marked, reversible health effects documented in the past. They point out that the present study is congruent with similar studies in North Carolina and Virginia in providing reassurance that in the absence of these conditions, watermen do not appear to face significant health risks from routine occupational exposure to estuarine waters that contain <italic>Pfiesteria</italic>.</p>
Beyond the Bench: Continuing Education for Nurses on Environmental Genetics and Complex Diseases	Could not extract abstract	<contrib contrib-type="author"><name><surname>Tillett</surname><given-names>Tanya</given-names></name></contrib>	Environmental Health Perspectives	<p>Many people find it hard to fit professional development and continuing education into their busy work lives. Now help is just a mouse-click away for nurses seeking flexible, self-paced training in the growing field of environmental genetics. The Community Outreach and Education Core (COEC) of the Center for Environmental Genetics at the University of Cincinnati, in collaboration with the Genetics Education Program for Nurses (GEPN) of Cincinnati Children’s Hospital Medical Center, has created an online Environmental Genetics and Complex Diseases educational module that introduces nurses to the principles of environmental genetics, and also teaches them how to apply those principles in nursing practice.</p><p>Online since December 2005, the module is useful for all nurses in clinical practice, but especially targets those who work extensively with minority or medically underserved patients. The module focuses on alcoholism, lead, and asthma, three challenging public and environmental health problems in underserved communities.</p><p>“The module is designed to prepare nurses in underserved communities to identify people who are at risk for environmental genetic conditions and help those people gain access to community services that emphasize prevention and early treatment strategies,” says COEC director M. Kathryn Brown.</p><p>Cynthia Prows, a clinical nurse specialist in genetics and the principal investigator of the web program, says the module organizes information into useful and manageable resources. “There is a tremendous amount of information on the Internet about genetics and about environmental health. But how do nurses who have limited knowledge in the topic areas locate the various sites, sift through all the information, decide what information is current and accurate, and then use that information for learning purposes? The answer is, most nurses don’t because they don’t have the time or the necessary foundational knowledge in genetics to mine the overwhelming mass of information that is accessible through the Internet.”</p><p>The module developers have done that work for the nurses, and have organized the content in a way that helps nurses develop foundational knowledge in environmental genetics using high-quality resources that are applicable to their practice. Once learners create a unique username and password, they can access the module free of charge, and can re-enter it at any time at the place they last exited. Those who wish to earn 4.8 nursing continuing education contact hours after completing the module and associated evaluations pay a minimal processing fee.</p><p>The module offers nurses background information on gene–environment interactions, and teaches them environmental and sociodemographic risk factors for common diseases. It also provides screening tools and community resources for nurses treating patients with recognizable genetic and environmental risk factors. Each of the three learning tracks also offer prenatal, pediatric, and adult case studies and self-assessments with each content area.</p><p>After completing the module, nurses are able to approach their communities armed with valuable knowledge of gene–environment interactions and insight into how those interactions can affect human health. They are also equipped with a wealth of online resources that can be accessed long after they complete the training module.</p><p>“Making sense of the fast-growing literature about how the health impacts of environmental exposures through the life span are mediated by our genetics is a challenge for health care professionals,” says Brown. “We hope that the vast array of resources identified in these self-paced, online modules will be helpful to primary care practitioners trying to make sense of new developments in genetic screening tests, environmental prevention strategies, and treatment options.”</p><p>The module is available at <ext-link ext-link-type="uri" xlink:href="http://gepn.cchmc.org/">http://gepn.cchmc.org/</ext-link>. Three additional genetics education modules currently available include Promoting Informed Decision-Making about Genetic Testing, Ethical and Social Issues Related to Genetic Testing, and Interpreting Family History. Two new modules are also in the pilot testing phase: Genetics Is Relevant Now––Nurse Views and Patient Stories, and Nurses’ Role in Pharmacogenetics/Pharmacogenomics.</p>
Anogenital Distance: Defining “Normal”	Could not extract abstract	<contrib contrib-type="author"><name><surname>Weiss</surname><given-names>Bernard</given-names></name></contrib><aff id="af1-ehp0114-a0399a">University of Rochester, Rochester, New York, E-mail: <email>[email protected]</email></aff>	Environmental Health Perspectives	<p>In their letter to <italic>EHP</italic>, <xref rid="b3-ehp0114-a0399a" ref-type="bibr">McEwen and Renner (2006)</xref> dismissed the findings of <xref rid="b6-ehp0114-a0399a" ref-type="bibr">Swan et al. (2005)</xref>, who reported a significant relationship between a measure of anogenital distance (AGD) in boys and levels of phthalate metabolites in their mothers’ urine during pregnancy. AGD is a sexually dimorphic index that, on average, is twice as great in males as in females, so it serves as a marker of proper male development. McEwen and Renner based their argument on an idiosyncratic form of logic. They asserted that</p><disp-quote><p>All male infants evaluated in the study appeared normal … there is no evidence for potential adverse effect in the test population. … no conclusion can be drawn whether the reported values are normal or abnormal. The range of AGD values … likely represents typical biologic variation that would be expected to occur among normal study subjects.</p></disp-quote><p>McEwen and Renner seem to be wholly unfamiliar with the meaning of a modest or even a slight shift in the mean of an index that reflects the distribution of susceptibility in a population. I have pointed out (<xref rid="b7-ehp0114-a0399a" ref-type="bibr">Weiss 1988</xref>) that even a 5-point (5%) reduction in mean intelligence quotient in a population of 100 million increases the number of individuals classified as retarded from 6 million to 9.4 million. It is this kind of relationship that eventually prompted the Centers for Disease Control and Prevention (CDC) to lower its definition of elevated lead risk levels in blood, set at 40 μg/dL in 1970, to 10 μg/dL in 1991 (<xref rid="b2-ehp0114-a0399a" ref-type="bibr">CDC 1991</xref>). <xref rid="b1-ehp0114-a0399a" ref-type="bibr">Bellinger (2006)</xref> put it this way:</p><disp-quote><p>A small change in the mean signals predictable accompanying changes in the proportions of individuals in the source population who fall into the tails of the distribution, where individuals who meet diagnostic criteria are found. Thus, the importance of a shift in group mean lies not in what it indicates about the average change among members of the study sample, but what it implies about the changes in the tails of the distribution in the population from which the study sample was drawn.</p></disp-quote><p>He noted, based on <xref rid="b4-ehp0114-a0399a" ref-type="bibr">Rose (1981)</xref>, that in a population with a prevalence of clinically defined hypertension of 15%, a 5-mm reduction in mean systolic blood pressure would result in a 33% decrease in prevalence (<xref rid="b1-ehp0114-a0399a" ref-type="bibr">Bellinger 2006</xref>). Epidemiologists recognize that a slight decrease in mean blood pressure in a population is translated into a major decrease in the incidence of serious cardiovascular events such as heart attacks.</p><p>We already know that shortened AGD at birth is one element, the leading edge, as it were, of the “phthalate syndrome” in rats, which is marked by testicular pathology, reduced spermatogenesis, hypospadias, and cryptorchidism, a compilation of signs indicating disordered male development that <xref rid="b5-ehp0114-a0399a" ref-type="bibr">Sharpe (2001)</xref> and others have noted to be on the increase in industrialized nations. An almost imperceptible shift to a lower mean AGD in the human male would foreshadow a heightened prevalence of reproductive system dysfunction. Is that the connection now emerging in the clinic?</p><p>If <xref rid="b3-ehp0114-a0399a" ref-type="bibr">McEwen and Renner’s (2006)</xref> criteria for “normal” were to govern the way in which we define the health risks of lead exposure, we would be basing our criteria on the number of children brought into hospital emergency rooms with lead poisoning rather than on the threats it poses to their neurobehavioral development. No parent, and no community, would tolerate such a definition these days.</p>
Environmental Polymorphism Registry: Banking DNA to Discover the Source of Susceptibility	Could not extract abstract	<contrib contrib-type="author"><name><surname>Spivey</surname><given-names>Angela</given-names></name></contrib>	Environmental Health Perspectives	<p>Walking outside on a day when ozone levels are at “code orange” doesn’t bother one person, but for someone else, it can result in chest pain, coughing, wheezing, or lung and nasal congestion. Why? Polymorphisms, tiny interindividual variations in genes, may be part of the reason. Providing a pool of information to help researchers determine how these variations interact with the environment to cause disease is the ultimate goal of the Environmental Polymorphism Registry (EPR), which is sponsored by the NIEHS and conducted in collaboration with the University of North Carolina at Chapel Hill’s General Clinical Research Center.</p><p>Any North Carolinian over 18 years of age can donate a sample—about a tablespoon of blood—to the EPR. Rather than recruiting donors with a particular disease, the EPR aims to gather, over five years, samples from 20,000 people who represent the general state population. The regional nature of the effort facilitates recruitment and follow-up.</p><p>“Recruitment is monitored to ensure that the EPR population is representative of the North Carolina population,” says Patricia Chulada, one of the four principal investigators of the EPR and a health scientist administrator at the NIEHS. “If we see deficiencies in certain groups, then we can increase efforts targeted to those particular groups.”</p><p>This approach will help researchers find out which polymorphisms are most common. “We want to look at people’s genetic material and find variations, and then go back and figure out what those variations mean,” says another EPR principal investigator, Paul B. Watkins, a professor of medicine at UNC–Chapel Hill and director of the General Clinical Research Center.</p><p>Chulada and Perry Blackshear, the NIEHS director of clinical research, initiated the registry by approaching Watkins and Susan Pusek, director of training and career development at the General Clinical Research Center. Watkins says the institute—and Blackshear himself—realized that “this is an essential direction of research to understand why some people are healthy and some are sick.”</p><p>There are multiple DNA registry efforts in the United States. Two major DNA banking efforts include Northwestern University’s NUgene Project and the Marshfield Clinic’s Personalized Medicine Research Project, both launched in 2002. International DNA banks are even more common, Chulada says. For instance, Iceland’s deCODE Project has recruited more than 80,000 subjects and has published findings on genes associated with arthritis and many other common conditions.</p><p>The EPR is unique, however, because it is designed to focus on environmentally responsive genes—those that increase the risk of disease when combined with an environmental exposure. The registry was created with the express intent of facilitating clinical studies of polymorphisms in these genes. Being affiliated with the NIEHS, where scientists are already studying such interactions, makes the EPR a natural resource for these investigations.</p><sec><title>Protecting Participants</title><p>Unlike with anonymous DNA databases, EPR donors provide their names and contact information so they can be asked to participate in follow-up studies if their DNA contains a polymorphism of interest. Participation in follow-up studies is optional, and donors can drop out of the database at any time.</p><p>Donors learn about the steps taken to ensure confidentiality in a 6-page consent form. Study interviewers at recruitment tables also discuss this information with potential donors, Chulada says.</p><p>Donors’ names and other information are stored separately from samples. When a sample is collected, it’s assigned a personal identification number. The code key that links the sample to identifying information is kept separate from the sample and from all other data in a computer system that’s password-protected. Access to this system is limited to only a few people directly involved in the EPR. Researchers can obtain contact information for potential participants only after approval by the EPR Oversight Committee.</p><p>To receive samples, researchers must sign a material transfer agreement, in which the researcher’s institution agrees to several conditions. “They can only use the samples for what they outlined in the agreement,” Chulada says. “They can’t give the samples to others. And they have to destroy the samples within a certain amount of time [which varies on a case-by-case basis].”</p><p>In addition, the NIH has granted the EPR a Certificate of Confidentiality, which protects researchers from being required, even by subpoena, to disclose research data or other information about an individual to an outside party such as an insurance company, an employer, or a civil or criminal court. “This is another layer of protection built into this system,” Chulada says.</p></sec><sec><title>Stepping Up Recruitment</title><p>The EPR has already accumulated about 4,000 samples—not far behind the 5,000 collected by NUgene since its launch. The EPR’s goal of 20,000 samples is the minimum needed to conduct certain types of studies with adequate statistical power, Chulada says. For example, if a researcher was interested in a rare genetic variant that occurs in only 1% of the population, the variant should be present in 200 samples from a registry of 20,000. “That would give us adequate statistical power to test for a phenotypic association of low to moderate effect, depending on other factors,” Chulada says.</p><p>When the EPR began, it recruited exclusively at two clinics at UNC–Chapel Hill. It has since expanded recruitment to Rex Hospital in Raleigh and is applying for approval to recruit at Duke University Medical Center in Durham. However, Chulada says, “Although recruiting at medical clinics gave us a diverse population in terms of health and other characteristics, we learned that we could increase both recruitment rates and diversity by recruiting outside of the clinic setting.”</p><p>A recruitment fair held for five days at the NIEHS campus in Research Triangle Park yielded about 420 donors. “We were ecstatic with the response of the NIEHS community,” Chulada says. The general public also can donate through study drives at corporations and health fairs in Research Triangle Park. Potential participants can visit the EPR web-site (<ext-link ext-link-type="uri" xlink:href="http://dir.niehs.nih.gov/direpr/">http://dir.niehs.nih.gov/direpr/</ext-link>) to find out about upcoming drives.</p></sec><sec><title>A DNA Goldmine</title><p>John Hollingsworth, a scientist working in the Environmental Lung Disease Group in the NIEHS Laboratory of Respiratory Biology, is one of the first investigators to apply for use of EPR samples. Hollingsworth and colleagues want to identify people who have a polymorphism in a certain gene, Toll-like receptor 4 (<italic>TLR4</italic>), known to be important in innate immune responses.</p><p><italic>TLR4</italic> was first identified as a candidate gene for response to ozone by NIEHS scientist Steven Kleeberger, leader of the Environmental Genetics Group in the Laboratory of Respiratory Biology. Subsequently, Hollingsworth and colleagues have demonstrated that mice deficient in <italic>TLR4</italic> are protected against airway hyper-responsiveness after exposure to ozone. “We want to determine if this gene is important in people in the biologic response to inhaled ozone,” Hollingsworth says. “We’re trying to validate what we’ve seen in mice in a human cohort.”</p><p>Hollingsworth calls the EPR a “gold-mine.” He says, “It’s a perfect situation. We have a cohort willing to be genotyped, rather than doing a mass screening of people for a single project, which is what we’ve had to do in the past.”</p></sec>
Radiation: Tanning Trippers Get UV High	Could not extract abstract	<contrib contrib-type="author"><name><surname>Burton</surname><given-names>Adrian</given-names></name></contrib>	Environmental Health Perspectives	<p>It has long been suspected that cutaneous endorphins are produced during exposure to UV light. Now research published in the April 2006 issue of the <italic>Journal of the American Academy of Dermatology</italic> suggests that frequent users of tanning beds may become addicted to these endorphins. Moreover, blocking the effects of the endorphins could lead to withdrawal symptoms.</p><p>“This might explain why some people appear to be hooked on sunbathing and why frequent users of tanning beds say they experience a positive mood change or are more relaxed after a session,” says coauthor Steven Feldman, a professor of dermatology at Wake Forest University School of Medicine.</p><p>Feldman’s team thought that blocking this endorphin rush might cause such people to lose some of their tanning enthusiasm; what they didn’t expect was for some to develop withdrawal symptoms.</p><p>The subjects included eight frequent tanners (who used tanning beds 8 to 15 times per month) and eight infrequent tanners (who used them up to 12 times per year). The researchers administered either a placebo or 5, 15, or 25 mg of naltrexone, a central and peripheral opioid receptor blocker; this blockage causes withdrawal symptoms in opioid drug–addicted people but not in nonaddicted people. The subjects were then asked to lie for 10 minutes on each of two tanning beds, one a true UV bed, the other rigged not to deliver UV light. Afterwards, the subjects, who were blind to the test conditions, were asked to describe which session made them feel best.</p><p>With the placebo and the 5-mg naltrexone dose, the frequent tanners showed a clear preference for the UV bed—and more strongly so than the infrequent tanners. But this preference fell away with the 15- and 25-mg doses of naltrexone, “suggesting that light-induced endorphins are reinforcing [frequent tanners’] behavior,” says report coauthor Mandeep Kaur, also a dermatology professor at Wake Forest University School of Medicine.</p><p>Further evidence of this was seen when half of the frequent tanners developed nausea and jitteriness with the 15-mg dose. “These are common [opioid drug] withdrawal symptoms,” explains Feldman, “and they were bad enough for two subjects to drop out.” Although there were no further problems at the 25-mg dose, Feldman says these results suggest that frequent tanners suffer some degree of dependency on endorphins.</p><p>“Clearly tanning is not as addictive as smoking,” remarks Robert Dellavalle, an associate professor of dermatology at the University of Colorado Health Sciences Center. “Just look at the prevalence of smoking in middle age—twenty percent in the UK and the United States. In contrast, there is a steep drop-off in the prevalence of tanning as people age.”</p><p>Still, says, Feldman, although it’s not time for the Drug Enforcement Administration to raid beauty parlors, “these results do raise questions about the safe use of tanning beds.”</p>
Anogenital Distance: Bailey and Renner Respond	Could not extract abstract	<contrib contrib-type="author"><name><surname>Bailey</surname><given-names>John E.</given-names></name></contrib><aff id="af1-ehp0114-a0399b">Cosmetic, Toiletry, and Fragrance Association, Washington, DC, E-mail: <email>[email protected]</email></aff>	Environmental Health Perspectives	<p>In his letter, Weiss misrepresents the arguments presented in our letter (<xref rid="b1-ehp0114-a0399b" ref-type="bibr">McEwen and Renner 2006</xref>) regarding the study of <xref rid="b2-ehp0114-a0399b" ref-type="bibr">Swan et al. (2005)</xref>. We pointed out that a value for “normal” anogenital distance (AGD) is not known and that without this information, “abnormal” AGD values cannot be determined. <xref rid="b2-ehp0114-a0399b" ref-type="bibr">Swan et al. (2005)</xref> measured AGD in a limited number of subjects (134 boys) who varied widely in age, height, and weight. This small sample size is inadequate to determine a normal AGD value, and there are no historical control data for AGD in male human infants using a definition of AGD comparable to the one used by Swan et al.</p><p>Although the significance of AGD values in humans, if any, is unknown, it is clear that a meaningful study with AGD as the end point of interest requires knowledge of normal values as a prerequisite. Further, the lack of knowledge of normal AGD values is only one of the significant limitations of the study by <xref rid="b2-ehp0114-a0399b" ref-type="bibr">Swan et al. (2005)</xref>; others were identified in our previous letter (<xref rid="b1-ehp0114-a0399b" ref-type="bibr">McEwen and Renner 2006</xref>).</p>
Headliners: Autism: Misfolded Protein Presents Potential Molecular Explanation for Autism Spectrum Disorders	Could not extract abstract	<contrib contrib-type="author"><name><surname>Phelps</surname><given-names>Jerry</given-names></name></contrib>	Environmental Health Perspectives	<p>De Jaco A, Comoletti D, Kovarik Z, Gaietta G, Radiæ Z, Lockridge O, et al. 2006. A mutation linked with autism reveals a common mechanism of endoplasmic reticulum retention for the α,β-hydrolase fold protein family. J Biol Chem 281:9667–9676.</p><p>Currently, there is only very limited information available on the etiology and biological basis of the autism spectrum disorders, although a mutation in the <italic>neuroligin 3</italic> gene has caught researchers’ attention in recent studies. Now NIEHS grantees Mark H. Ellisman and Palmer Taylor at the University of California, San Diego, and their colleagues have determined that homologous mutations in the genes coding the proteins butyrylcholinesterase (BChE) and acetylcholinesterase (AChE) cause defects in protein expression similar to those seen with <italic>neuroligin 3</italic>, shedding further light on a potential molecular mechanism underlying autism.</p><p>The neuroligins, BChE, and AChE are members of the α,β-hydrolase fold family of proteins. The <italic>neuroligin 3</italic> mutation, an arginine-to-cysteine substitution, was identified in a set of twins and has been shown to result in most of the expressed protein being retained within the endoplasmic reticulum. The small amount of protein that does reach the surface of the cell shows little binding affinity for its partner, β-neurexin, suggesting possible misfolding of the protein. Misfolded proteins are known to cause endoplasmic reticulum stress. This, in turn, can trigger cell death and contribute to human diseases including neurodegeneration, heart disease, and diabetes mellitus.</p><p>In the current study, the researchers used confocal fluorescence microscopy and analysis of oligosaccharide processing to observe whether an arginine-to-cysteine mutation affected AChE and BChE similarly despite the proteins having differing oligomerizing capacities. By inserting homologous mutations in the AChE and BChE cDNAs, they found that the mutation also resulted in endoplasmic reticulum retention of the two cholinesterases. The proteins were then likely degraded in the proteasome. The authors speculate that altering intracellular oxidation/reduction parameters may assist in the proper folding and export of these proteins.</p>
Tracing the Origins of Autism: A Spectrum of New Studies	Could not extract abstract	<contrib contrib-type="author"><name><surname>Szpir</surname><given-names>Michael</given-names></name></contrib>	Environmental Health Perspectives	<p>The etiology of a medical condition might seem an unlikely subject to arouse intense feelings. Yet few medical disorders have stirred up as much passion and divisiveness among scientists and the general public as autism has in recent years. The heat of the controversy has even attracted attention from periodicals such as <italic>The Wall Street Journal</italic>, the <italic>Columbia Journalism Review</italic>, and <italic>Wired</italic> magazine—seemingly improbable forums for a medical debate. Why all the furor?</p><p>At the eye of the storm is the startling climb in the numbers of children who have been diagnosed with one of the autism spectrum disorders (ASDs). The most severe ASD is autistic disorder (which often is called simply “autism”); other forms include Asperger syndrome and the much rarer childhood disintegrative disorder. In the United States, the diagnosis of ASDs increased roughly 10-fold over the course of a decade, from 4–5 children per 10,000 in the 1980s to 30–60 children per 10,000 in the 1990s, according to a report in the August 2003 <italic>Journal of Autism and Developmental Disorders</italic>. The 5 May 2006 issue of <italic>Morbidity and Mortality Weekly Report</italic> describes the results of two parent surveys from 2003 and 2004, which suggested that 55–57 children per 10,000 had autism (however, an editorial note points out that, due to the nature of the surveys, parents of children with other ASDs may have reported their children as having autistic disorder).</p><p>Some scientists believe that much of the upsurge is the result of increased awareness of ASDs or changes in diagnostic criteria, which would suggest that the true prevalence of the disorders has been stable over time. Others disagree. “It is premature to state that there is no increase in prevalence,” says W. Ian Lipkin, a professor of neurology, anatomy, and neurobiology at Columbia University. “None of the studies to date has been designed to definitively address the issue.”</p><p>The prevalence of ASDs plays into the fundamental question of what causes these disorders. If the number of cases is truly on the rise, then it would seem likely that some change in the environment is driving up the total. That’s partly what has divided scientists into opposing camps—they cannot agree on the relative importance of genetic and environmental factors in the disorders’ etiology.</p><p>Alas, answering the prevalence question might not end that debate. “Even if the prevalence of autism were stable,” says Lipkin, “you would not be able to rule out the possibility of an environmental trigger.” That’s because very little is known about the mechanisms that cause autism, be they environmental or genetic.</p><p>“The study of autism was, until recently, largely dominated by the field of psychology, where characterizing the behaviors and developing reliable instruments for diagnosis have been major areas of research over the past few decades,” says Irva Hertz-Picciotto, an epidemiologist at the University of California, Davis.</p><p>Indeed, the core symptoms of ASDs—social disinterest, repetitive and overly focused behavior, and problems in communication, usually appearing before 3 years of age—have been well described. Much less research has focused on the causes of these symptoms.</p><p>Several investigations dating back to the 1970s indicate that identical twins have a much higher concordance rate of ASDs than fraternal twins, according to a report in the Spring 1998 issue of <italic>Mental Retardation and Developmental Disabilities Research Reviews</italic>. Those studies provide some of the best evidence that these disorders have a strong genetic component. But the identity of the genes involved, much less how they produce ASDs, has not been established. Moreover, the concordance rate for identical twins is not 100%, which suggests that at least some cases must be associated with environmental or epigenetic factors.</p><p>A few cases of ASDs have been clearly linked to environmental insults. These include prenatal exposure to chemical agents such as thalidomide and valproic acid, as well as to infectious agents such as the rubella and influenza viruses. Here again, the concordance rate is not 100%, which suggests that a genetic predisposition is necessary for chemical and microbial factors to act as triggers.</p><p>Tantalizing clues like these are prompting scientists to reconsider the research agenda for ASDs. Martha Herbert, a pediatric neurologist at Harvard Medical School, and her colleagues have been applying the methods of genomics to identify environmentally responsive genes that might be important in these disorders.</p><p>“When you realize that the widespread changes we’re seeing in autistic brains may occur in parallel with or even downstream from widespread changes in the body—such as in the immune system—and that these changes may be environmentally triggered, you start looking for ways to think more broadly about genetic vulnerability. It can’t be just about ‘brain genes,’” Herbert says.</p><p>Some new epidemiological studies also are looking for gene–environment interactions. According to Diana Schendel, an epidemiologist and project officer for autism research at the CDC, which funds one of the projects, these initiatives will be able to examine many possible causal pathways to ASDs, including both genetic and environmental causes that may lead to the development of the disorders in different subgroups of children.</p><p>Some of these projects are already under way, whereas others will begin soon. All of the scientists involved, however, believe their research will finally provide some of the answers that everyone has been looking for.</p><sec><title>CHARGE</title><p>The Childhood Autism Risks from Genetics and the Environment (CHARGE) project is unique among the large ASD epidemiological studies. It focuses solely on autistic disorder, and it emphasizes a search for environmental factors—including a broad array of chemicals in food, consumer products, and ambient air, as well as infectious and medical exposures—that might be linked to the disorder. The study is funded by the NIH.</p><p>CHARGE is a case–control study in which a group of autistic children aged 2 to 5 years is compared to a group of age-matched controls in a population-based study. “Because of the California Department of Developmental Services’ system of Regional Centers [nonprofit corporations that coordinate health care services and support for citizens with developmental disabilities], we have a handle on enumerating a high proportion of the children newly diagnosed with autism in our defined area over a specific time period,” says Hertz-Picciotto, the principal investigator of the CHARGE study. “Similarly, we can enumerate the children in the same area and time period who are not cases. We then sample from both.”</p><p>The project was initiated in 2002 with the goal of recruiting 1,000 to 2,000 children. Half of the children will be autistic. The other half will make up two control groups: one group of children with developmental delays (but not an ASD) and a second group of children selected from the general population without regard to developmental characteristics.</p><p>The advantage of the case–control design is that scientists can acquire large numbers of children with the disorder. By comparison, in a cohort design researchers would need a very large sample size, given the prevalence of autism, to acquire the same number of cases.</p><p>Hertz-Picciotto expects to have enrolled nearly 700 children by August 2006, the end of the first funding period. “I’ve applied for another five-year grant,” she says, “and I hope to be funded to enroll nine hundred in that round, which would bring us to sixteen hundred children.”</p><p>The CHARGE team is looking at possible exposures during the prenatal period and early childhood. Some of the data will be gathered through comprehensive interviews with parents, but Hertz-Picciotto admits that this is not the best way to look for exposures. “You ask people questions, and their answers may be colored by the fact that they know they have a child with a condition,” she says. “They may spend a lot of time thinking about what they might have done or what might have gone wrong, and they may have preconceived ideas about what caused [the disorder]. They might not be as objective.” Such problems with postdiagnosis interview information are recognized as a weakness of retrospective studies.</p><p>The scientists are getting around this issue by examining each child’s medical records and those of the mother during pregnancy and delivery—nonsubjective data gathered in the course of routine obstetric care. They are also collecting blood, urine, and hair specimens that will be analyzed in the laboratory.</p><p>The study has already provided some intriguing leads. “We’re finding that the immune system seems to function at a lower level in autism,” says Hertz-Picciotto. “That’s an important clue. It could mean that whatever causes autism also disrupts the immune system, or it could be that the immune system disrupts neural development so that something goes awry in laying down brain circuitry prenatally or in the early postnatal period.” [For more information on the CHARGE study, see p. 1119, this issue.]</p></sec><sec><title>ABC</title><p>The Autism Birth Cohort (ABC) Study, now under way in Norway, is a large prospective design that is expected to gather information on 100,000 babies. The work is being led by scientists at the Mailman School of Public Health at Columbia University, who are collaborating with colleagues at the Norwegian Institute of Public Health, with funding from the U.S. National Institute of Neurological Disorders and Stroke.</p><p>“When you want to know why some people are more at risk than others in a population, then that’s best answered using a cohort design,” says Ezra Susser, an epidemiologist at Columbia University and a co-investigator on the ABC project. “When we think about environmental causes of [ASDs], we’re probably interested in phenomena that occur prior to birth or perhaps shortly after birth. So you want to collect prospective data from people as early as possible in pregnancy.” Because ASDs are not common, the study will need large numbers of children to have enough statistical power, according to Susser.</p><p>So far the ABC team has recruited 75,000 pregnant Norwegian mothers, but Susser is hoping for more. “We’ve got enough to look for an environmental risk factor, but you need larger numbers for studying gene–environment interactions, which could turn out to be important,” he says. It’s possible the team could acquire greater numbers by collaborating with other studies. One candidate for collaboration is the Avon Longitudinal Study of Parents and Children in the United Kingdom, which is looking at the complex ways in which environmental features may relate to optimal development and health in children. But there’s been no agreement yet, Susser says.</p><p>Even so, the ABC scientists are optimistic about their study. “Little is known about the natural history of [ASDs],” says Lipkin, who is the principal investigator of the project. “By starting prenatally, we’re collecting detailed, critical information about environmental exposures in an unbiased fashion.”</p><p>The scientists are also collecting plasma, serum, RNA, and DNA. “We have extraordinary biological materials,” says Lipkin. “We can pursue biomarkers as well as exposure to toxicants and infection. We also have maternal DNA, paternal DNA, and the child’s DNA [so-called trio data]; thus we can look for the appearance of novel mutations,” he adds.</p><p>The ABC researchers will follow the children through time, with parents answering questionnaires about the health and social interactions of their children as they reach 6, 18, and 36 months of age. “It may be that the developmental trajectory tells us much more than a single time point can ever tell us about the pathogenesis of [ASDs],” says Mady Hornig, a physician-scientist at Columbia University who participates in the project.</p><p>Despite their enthusiasm for the project’s potential, the ABC scientists feel they could accomplish much more if they only had the funding. “The pity of it is we have no money to do the biological work,” says Lipkin. “We can collect the samples and do the questionnaires, but we’ve been unable to get funding to look for any of the environmental factors. We’re collecting blood, but we won’t know whether there’s a biomarker until we do a biomarker analysis. We have funds to collect RNA, but in order to do the transcript profiling we need approximately four hundred dollars per sample,” he says.</p><p>Lipkin adds that there’s only so much that one can do with questionnaire data. “We do ask about infection and diet, but that’s not the same as having a lab value that can validate what was reported, and then look at a direct correlation with the outcome,” he says.</p><p>Lipkin believes that part of the problem is that searching for environmental factors goes against the current research paradigm in ASDs. “The focus is on genetic factors,” he says. “Infectious diseases, toxicology, and immunology receive short shrift. The ABC is clearly the right opportunity to pursue these other leads because we have the ideal samples to survey prenatally and postnatally,” he says.</p><p>The scientists are just now receiving the responses to the 36-month questionnaire. “It’ll probably be another two years before we have our first report,” Hornig says. Funds are now in place to study the children at 36 months; however, the team hopes to follow them for a lifetime, according to Hornig.</p></sec><sec><title>CADDRE</title><p>In response to the Children’s Health Act of 2000, the CDC established and funds six Centers for Autism and Developmental Disabilities Research and Epidemiology (CADDRE) to investigate potential risk factors for ASDs. The multisite approach offers a study group that is geographically and demographically more representative of the general U.S. population than a smaller regional study could provide, according to Craig Newschaffer, an epidemiologist and principal investigator at the Johns Hopkins Bloomberg School of Public Health CADDRE site.</p><p>According to Newschaffer, the CADDRE sites will use a case cohort design in which the exposure patterns of the ASD cases are compared to a random sample of children living in the same geographic area. A third study group, consisting of neurodevelopmentally impaired children who do not have an ASD, will round out the sample populations. The investigators hope to enroll a total of 650 to 900 children, aged 3 to 5 years, in each study group across all the sites, making CADDRE the largest study of its kind in the United States, says Newschaffer. A uniform protocol across the sites will allow the scientists to pool their data.</p><p>CADDRE will collect and archive blood, cheek cell, and hair samples from the children in order to investigate a broad range of potential risk factors. “We’re not focused on the environment as much as CHARGE is,” says Newschaffer, “but we are collecting data on questionnaires and reviewing medical records on exposure, in addition to the biosampling for exposures.”</p><p>The scientists should have sufficient numbers to look at gene–environment interactions. “We are collecting DNA from the parents and the kids from each of the groups. We’ll have trio data in each of the three groups, a potentially powerful design,” says Newschaffer.</p><p>CADDRE scientists will also characterize the behavior of the children, as well as describe any comorbid medical conditions and atypical physical features. The goal is to sort out different etiologic subgroups within the autism spectrum. As Newschaffer explains, “There are a lot of possible reasons why we’ve had a hard time coming up with genetic and nongenetic risk factors. One of them is that autism is likely a heterogeneous condition, with different etiologies producing kids with what appear to be similar phenotypic profiles. If you don’t separate out the different etiologic groups, it’s going to be very hard to find an association with a gene or an exposure. If we limit our analyses to kids that have a certain profile, we’re going to be able to make some informed guesses about what profiles might allow risk factors to emerge,” he says. The CADDRE sites will begin recruiting children into the study in the fall of 2006.</p><sec><title>More Studies, More Acronyms</title><p>There are several other smaller epidemiological studies in the works. In California, scientists are tapping into specimen banks that have stored blood samples taken from mothers during pregnancy and from their children at birth. The Early Markers for Autism (EMA) study employs a case–control design, with about 100 children with an ASD (primarily autism), 100 who are developmentally delayed, and 200 from the general population. “We can correlate what’s happening in the mom and the baby, which is really exciting,” says Lisa Croen, a perinatal epidemiologist at the Kaiser Permanente Division of Research in California and the project’s principal investigator.</p><p>EMA is a multidisciplinary collaboration with epidemiologists, geneticists, immunologists, neurovirologists, and endocrinologists, according to Croen. “Because autism is so complex, it’s important for all these researchers to communicate with each other. I think EMA is a model for how to do research in autism,” she says. EMA is unique, according to Croen, because the study will be looking for biological markers of ASDs very early in development, during gestation, and at birth. “This allows us to focus on mechanisms that may be leading to autism rather than mechanisms that are consequences of having autism,” she says.</p><p>The EMA scientists are investigating genetic and nongenetic factors, with a focus on the immune dysregulation hypothesis of ASDs. “We’re measuring different kinds of immune markers, including immunoglobulin levels and antibodies to specific infectious agents, cytokines, and autoantibodies,” says Croen. “We’re looking for things that distinguish kids who are subsequently diagnosed with autism from those who aren’t. This will help us understand the pathobiology of autism—the mechanisms that are leading to the dysregulation in development.”</p><p>The three-year EMA is currently in its last year. “We still have lots of analyses to do,” says Croen, “but we’re beginning to write some papers. We’re finding differences between the children in levels of certain proteins measured in the circulating blood collected from mothers during pregnancy. I think the study has much to contribute to our understanding of the biology of what might be going wrong.”</p><p>Croen is also an investigator on the California Autism Twin Study (CATS), which expects to recruit 300 identical and fraternal twin pairs born between 1987 and 1999 in which at least one of the twins has an ASD. Comparing the twin pairs will allow the scientists to estimate the heritability of ASDs—the relative genetic and environmental contributions to the disorder. “Knowing the behavioral and developmental differences between the twins might help us understand the effects of gene expression, the <italic>in utero</italic> environment, and environmental triggers,” Croen says.</p><p>Hertz-Picciotto is also excited about a five-year study that she and her colleagues hope to begin soon. Unlike CHARGE, the new effort, called MARBLES (Markers for Autism Risk in Babies—Learning Early Signs), will be a prospective study in which data will be gathered before the children are diagnosed. Pregnant women who already have at least one child with autism will be enrolled right at the beginning of pregnancy. The mothers will keep diaries about their symptoms and health-related events, and the researchers will collect cord blood samples and placentas.</p><p>Based on previous research, Hertz-Picciotto expects that about 1 in 10 siblings of the autistic children will also have the disorder, and perhaps 1 in 4 or 5 will be “on spectrum” with a related but less severe condition such as Asperger syndrome, or with some symptoms of the broad behavioral phenotype, such as language delays and atypical social skills. “This work is complementary to the case–control approach, and should provide us with a lot of information that will build on what we find in CHARGE. It should be a phenomenal resource,” she says.</p></sec><sec><title>You Say You Want a Revolution</title><p>In April 2004, the U.S. DHHS issued a publication, <italic>Congressional Appropriations Committee Report on the State of Autism Research</italic>, describing recommendations made by a panel of expert scientists convened by the Interagency Autism Coordinating Committee (IACC). The IACC panel suggested an ambitious agenda, which included the goal of identifying environmental risk factors and their associated developmental windows within a four- to six-year period, as well as identifying genetic and nongenetic causes of ASDs and their interactions within seven to ten years.</p><p>Hertz-Picciotto, a member of the IACC panel, thinks these goals should be taken with a grain of salt. “I’m optimistic that we will have identified some environmental risk factors, and may have excluded a few others, between 2008 and 2010—but by no means will we have the final word. The genetics and the gene–environment interactions may be even tougher. Unfortunately, I don’t see enough groups working on the environmental contribution to autism, so it may be slower than projected,” she says.</p><p>Mark Blaxill, vice president of SafeMinds, a parent-led advocacy group, also believes that environmental risk factors don’t receive enough consideration. “The CDC has not addressed the crisis in autism responsibly,” he says. “They should be raising the alarm, and they have failed to do so. They should be asking why so many children are sick. Instead, they’ve tried to suggest a degree of doubt about the increases, and that diverts attention and funding from environmental causes.”</p><p>Schendel responds, “It is clear that more children than ever before are being classified as having an ASD. It is important that we treat common developmental disorders, and especially the ASDs, as conditions of urgent public health concern. The CDC’s efforts in addressing this public health concern include funding for ASD monitoring programs to understand ASD trends, funding for research into the genetic and environmental causes of ASDs, and education and outreach programs to promote early identification and timely intervention for all children with developmental problems.”</p><p>Despite the promise of the new epidemiological studies, some researchers are still dismayed, as one scientist put it, that “geneticists are running the show, and ignoring the environmental aspects.” What would it take for things to change? Blaxill invokes the ideas of philosopher Thomas Kuhn, who suggested that scientific revolutions occur when an old paradigm is replaced by a new one. “I believe we’re in the middle of a paradigm shift,” Blaxill says. “The dramatic explosion of autism rates does not fit the genetic model. It’s an anomaly that will kill the old paradigm.”</p></sec></sec>
The Beat	Could not extract abstract	Could not extract contributor	Environmental Health Perspectives	<sec><title>Now Broadcasting: green.tv</title><p>green.tv, the world’s first Internet-based broadband channel dedicated to environmental issues, started broadcasting in March 2006 from its website at <ext-link ext-link-type="uri" xlink:href="http://www.green.tv/">http://www.green.tv/</ext-link>. The channel, developed with support from UNEP, is also available as a podcast on iTunes. green.tv will carry films from around the world, produced by NGOs, community film makers, public sector agencies, and environmentally minded corporations. The site features seven subchannels focused on air, land, water, climate change, people, species, and technologies. Each subchannel will run a feature film, a news item, and a story for children. The channel’s first offerings include films from Water Aid, the Sierra Club, the Eden Project, the Women’s Environment Network, Farm Africa, and others.</p><fig id="f1-ehp0114-a0403b" position="anchor"><graphic xlink:href="ehp0114-a0403bf1"/></fig></sec><sec><title>UNEP Promotes Sustainable Building and Construction</title><p>In February 2006 UNEP announced the launch of the Sustainable Building and Construction Initiative to promote environmentally friendly practices in the construction industry. Three of the world’s largest construction companies—Lafarge, Skanska, and Arcelor—have signed on to the effort. The construction sector employs over 100 million people worldwide and contributes 10% of the global gross domestic product. Yet the industry also plays a serious role in problems such as climate change, waste generation, and depletion of natural resources. The new initiative will address these issues, and also lobby for laws and building standards to support sustainable practices.</p></sec><sec><title>Pediatric Environmental Health in Argentina</title><p>WHO statistics show that approximately 33% of diseases affecting children under age 5 are linked to environmental risk factors. To address this threat, the Argentine and Buenos Aires governments have set up new “pediatric environmental health units” in Buenos Aires and several provinces of Argentina. The units are made up of pediatricians, nurses, social workers, teachers, and others who work as a team to uncover and remediate risk factors in children’s environments, often at the request of a referring physician. The units have the authority to work with schools, public works, and neighborhood residents if they believe a specific hazard exists. The units will also train other professionals within the hospitals where they are based and conduct research on child environmental health issues.</p><fig id="f2-ehp0114-a0403b" position="anchor"><graphic xlink:href="ehp0114-a0403bf2"/></fig></sec><sec><title>Unlimited Mileage from the Drive-Thru?</title><p>A company offering rental cars powered entirely by biodiesel opened its doors in Los Angeles in February 2006. The cars get 400 to 800 miles per tank on 100% biodiesel made from recycled cooking oil. Bio-Beetle Eco Rental Cars first started in Hawaii in 2003 with a single car, and now offers 16 at that location, while the LA location is starting with 4 vehicles. Company founder Shaun Stenshol hopes to open two more U.S. locations by the end of the year. Other biodiesel rentals may not be far behind: Enterprise is pilot-testing an offering of biodiesel Jeeps in Portland, Oregon.</p></sec><sec><title>Goldman Environmental Prize 2006</title><p>For 17 years, the $125,000 Goldman Environmental Prize has been awarded to activists dedicated to effecting environmental change in their home countries. The six winners for 2006 are:</p><list list-type="bullet"><list-item><p>Yu Xiaogang, of China, who created groundbreaking watershed management programs while documenting the socioeconomic impact of dams on Chinese communities. China’s central government now considers social impact assessments for major dam developments.</p></list-item><list-item><p>Anne Kajir, of Papua New Guinea, who uncovered government corruption that allowed rampant illegal logging of the region’s largest remaining intact parcel of tropical rain forest. As a novice lawyer, she successfully defended a Supreme Court appeal forcing the logging industry to pay damages to indigenous land owners.</p></list-item><list-item><p>Tarcísio Feitosa da Silva, of Brazil, who led efforts to create the world’s largest area of protected tropical forest regions in a remote area of northern Brazil that was threatened by illegal logging.</p></list-item><list-item><p>Craig E. Williams, of Kentucky, who convinced the Pentagon to halt plans for burning old chemical weapons that had been stockpiled around the United States.</p></list-item><list-item><p>Olya Melen, of Ukraine, who used the legal system to temporarily halt the construction of a massive canal through the rich wetlands of the Danube Delta.</p></list-item><list-item><p>Silas Kpanan’Ayoung Siakor, of Liberia, who revealed evidence that former Liberian president Charles Taylor used profits from unchecked logging to pay for a 14-year civil war. The revelation led the UN Security Council to ban the export of Liberian timber, part of wider ongoing trade sanctions.</p></list-item></list><fig id="f3-ehp0114-a0403b" position="anchor"><graphic xlink:href="ehp0114-a0403bf3"/></fig></sec>
Mercury from Fish Does Not Reduce Children’s IQs	Could not extract abstract	<contrib contrib-type="author"><name><surname>Schwartz</surname><given-names>Joel</given-names></name></contrib><aff id="af1-ehp0114-a0399c">American Enterprise Institute, Sacramento, California, E-mail: <email>[email protected]</email></aff>	Environmental Health Perspectives	<p><xref rid="b13-ehp0114-a0399c" ref-type="bibr">Trasande et al. (2005)</xref> concluded that pre-natal methylmercury (MeHg) exposure is reducing children’s IQs (intelligence quotients), costing $8.7 billion/year. They achieved this high estimate <italic>a</italic>) by assuming that IQ reductions occur at MeHg exposures near or even below the 5.8 μg/L reference dose (RfD), although there is no evidence for IQ reductions even at much higher exposures; and <italic>b</italic>) by overstating by nearly a factor of three the fraction of newborns with MeHg exceeding the RfD. I believe that their analysis is flawed, invalid, and not appropriate as an input to policy decisions.</p><p><xref rid="b13-ehp0114-a0399c" ref-type="bibr">Trasande et al. (2005)</xref> assumed that 10% of newborns are exposed prenatally to MeHg exceeding the RfD. However, the appropriate value is 3.6%. Trasande et al. made two errors. First, they used a lower RfD than 5.8 μg/L, based on the observed enrichment of MeHg in umbilical cord blood relative to maternal blood. However, the current RfD already accounts explicitly for this enrichment through an uncertainty factor of 3.15 applied to the benchmark dose lower limit [<xref rid="b14-ehp0114-a0399c" ref-type="bibr">U.S. Environmental Protection Agency (EPA) 2001</xref>]. Second, they assumed that women 16–49 years of age measured during 1999–2000 accurately represented MeHg levels in pregnant women (<xref rid="b8-ehp0114-a0399c" ref-type="bibr">Mahaffey et al. 2004</xref>). National Health and Nutrition Examination Survey (NHANES) data collected during 1999–2002 (<xref rid="b5-ehp0114-a0399c" ref-type="bibr">Jones et al. 2004</xref>), available before <xref rid="b13-ehp0114-a0399c" ref-type="bibr">Trasande et al. (2005)</xref> submitted their manuscript, show the 95th percentile MeHg level for pregnant women to be 32% below Trasande et al.’s value.</p><p>If any MeHg exposure above the RfD reduced IQ, there would still be cause for concern. However, there is no evidence for IQ reductions even at exposures several times the RfD.</p><p>Previous studies in the Seychelles Islands (<xref rid="b9-ehp0114-a0399c" ref-type="bibr">Myers et al. 2003</xref>) and New Zealand (<xref rid="b3-ehp0114-a0399c" ref-type="bibr">Crump et al. 1998</xref>) did not find IQ reductions at any MeHg exposure. A study in the Faroe Islands (<xref rid="b4-ehp0114-a0399c" ref-type="bibr">Grandjean et al. 1999</xref>) did not measure IQ. Many children in these studies had prenatal MeHg exposures exceeding 10 times the RfD. The claim of IQ reductions in Americans is even weaker because Americans’ MeHg exposures are far lower. Of 629 pregnant women measured by NHANES, the highest exposure was 3.7 times the RfD (<xref rid="b1-ehp0114-a0399c" ref-type="bibr">Centers for Disease Control and Prevention 2005</xref>). Among those exceeding the RfD, 75% were below twice the RfD.</p><p><xref rid="b13-ehp0114-a0399c" ref-type="bibr">Trasande et al. (2005)</xref> cited results from the Faroe Islands (<xref rid="b4-ehp0114-a0399c" ref-type="bibr">Grandjean et al. 1999</xref>) to claim IQ reductions, but this study is less compelling than the Seychelles study (<xref rid="b10-ehp0114-a0399c" ref-type="bibr">Myers et al. 2003b</xref>) for assessing Americans’ risks: <italic>a</italic>) the Seychellois are exposed to MeHg through ocean fish, similar to Americans, whereas the Faroese are exposed through whale meat (<xref rid="b10-ehp0114-a0399c" ref-type="bibr">Myers et al. 2003b</xref>); <italic>b</italic>) the Seychellois are ethnically diverse, but the Faroese are homogeneously Scandinavian (<xref rid="b12-ehp0114-a0399c" ref-type="bibr">Rice et al. 2003</xref>); and <italic>c</italic>) the Seychelles study used hair MeHg to measure exposure, and the Faroes study used cord blood. Hair MeHg has been calibrated with fetal brain levels, but cord blood has not (<xref rid="b2-ehp0114-a0399c" ref-type="bibr">Cernichiari et al. 1995</xref>; <xref rid="b9-ehp0114-a0399c" ref-type="bibr">Myers et al. 2003a</xref>).</p><p>Despite the advantages of the Seychelles study, <xref rid="b13-ehp0114-a0399c" ref-type="bibr">Trasande et al. (2005)</xref> dismissed it, claiming that the National Research Council (<xref rid="b11-ehp0114-a0399c" ref-type="bibr">NRC 2000</xref>) “opined that the most credible of the three prospective epidemiologic studies was the Faroe Islands investigation.” In reality, referring to all three studies, the <xref rid="b11-ehp0114-a0399c" ref-type="bibr">NRC (2000)</xref> concluded that “each of these studies was well designed and carefully conducted.” Nevertheless, the NRC “concluded that a well-designed study with positive effects provides the most appropriate public-health basis for the RfD.” The NRC thus excluded the Seychelles study not because of the quality of the study but because the study found that MeHg did not cause any harm.</p><p><xref rid="b13-ehp0114-a0399c" ref-type="bibr">Trasande et al. (2005)</xref> also made other errors:</p><list list-type="bullet"><list-item><p>They claimed that the New Zealand study reported IQ reductions, citing <xref rid="b6-ehp0114-a0399c" ref-type="bibr">Kjellstrom et al. (1986</xref>, <xref rid="b7-ehp0114-a0399c" ref-type="bibr">1989)</xref>. However, they omitted <xref rid="b3-ehp0114-a0399c" ref-type="bibr">Crump et al.’s (1998)</xref> reanalysis, coauthored with Kjellstrom, which superseded previous reports and found no IQ reduction.</p></list-item><list-item><p>They claimed that the Seychelles study had only half the statistical power of the Faroes study. The studies actually have similar power (<xref rid="b10-ehp0114-a0399c" ref-type="bibr">Myers et al. 2003</xref>; <xref rid="b11-ehp0114-a0399c" ref-type="bibr">NRC 2000</xref>).</p></list-item><list-item><p>They claimd the NRC concluded that MeHg reduces IQs even at exposures lower than the RfD. However, the NAS cautioned that the cohort studies were incapable of assessing effects of exposures near the RfD, because hardly any children had such low MeHg exposures (<xref rid="b11-ehp0114-a0399c" ref-type="bibr">NRC 2000</xref>).</p></list-item></list><p>The weight of the evidence indicates that MeHg, even at exposures substantially greater than the highest U.S. levels, does not reduce children’s IQ. The evidence against IQ reductions is particularly strong for MeHg exposures from fish.</p><p><xref rid="b13-ehp0114-a0399c" ref-type="bibr">Trasande et al. (2005)</xref> relied on mistaken assumptions regarding exposures to and effects of MeHg, and misinterpreted or omitted contrary evidence. Therefore, I consider their analysis to be fundamentally flawed and invalid.</p>
The NIH ENDGAME Consortium	Could not extract abstract	Could not extract contributor	Environmental Health Perspectives	<p>The NIH has recently established a highly interactive consortium of 11 research groups on Enhancing Development of Genome-wide Association M<bold>e</bold>thods (ENDGAME) to advance the utility of genome-wide association studies. The consortium (funded by the NHLBI, NIEHS, NCI, NHGRI, and NIGMS) brings together expertise in genetics, epidemiology, biostatistics, and bioinformatics to develop and test innovative, informative, and cost-effective study designs and analytical strategies for performing genome-wide association studies on complex diseases. Available resources such as the International Haplotype Mapping (HapMap) data and single nucleotide polymorphism (SNP) discoveries along with improvements in genome technologies have increased the feasibility of genome-wide association studies for complex diseases. However, it has become increasingly apparent that a major barrier to successfully completing these studies is a lack of both appropriate analytical tools and understanding of which study designs and computational methods are most appropriate for particular study scenarios.</p><p>The NIEHS is most interested in the development of analytical tools and approaches that would allow identification of environmental components or covariates of complex diseases in genome-wide association studies. Although most common chronic diseases are the result of complex interactions between genes (G) and environmental (E) factors, most analytical approaches adopted for whole genome scans do not incorporate interactive effects with environmental factors. Studies have indicated that failure to account for G × E interactions in complex disease association analyses can decrease the power to find genetic disease loci and underestimate both the genetic and environmental effects of the disease. The NIEHS is therefore co-funding with the NCI two applications in this consortium, led by Dr. Duncan Thomas of the University of Southern California and Dr. Charles Kooperberg of the Fred Hutchinson Cancer Center, that specifically focus on identifying study designs and analytical methods that will enhance the possibility of identifying gene–gene and gene–environment interactions. All strategies and tools developed through this consortium will be made available to the entire scientific community. The long-term goals of ENDGAME are to accelerate the identification of genetic susceptibility factors in human disease and the ultimate development of novel and individual disease prevention and treatment strategies through the advancement of genome-wide association study methodologies.</p><sec><title>Contact</title><p><bold>Kimberly A. McAllister, Ph.D.</bold> \| <email>[email protected]</email></p></sec>
Near and Not-So-Dear TRI Facilities: Prenatal Proximity and Later Brain Cancer	Could not extract abstract	<contrib contrib-type="author"><name><surname>Brown</surname><given-names>Valerie J.</given-names></name></contrib>	Environmental Health Perspectives	<p>The most clearly established environmental risk factor for childhood brain cancer is therapeutic radiation exposure (not including diagnostic X-rays). New research now suggests that children of mothers who lived near an EPA Toxics Release Inventory (TRI) facility while pregnant may be more likely to later develop brain cancer, especially if the site released carcinogens <bold>[<italic>EHP</italic> 114:1113–1118; Choi et al.]</bold>.</p><p>Prenatal exposure to chemicals can have profound long-term effects, as some toxic chemicals that are stopped by the blood–brain barrier in adults may reach the fetus via the placenta. This work is the first to specifically examine brain cancer risk in children and potential exposure to TRI releases, although some previous research has suggested slight increases in risk for certain birth defects associated with such emissions.</p><p>Of the more than 650 toxic chemicals listed in the TRI, 193 are known or suspected carcinogens, according to the EPA. Fifty-five known, probable, or possible carcinogens were actually released within 2 miles of the study participants. However, it is very difficult to accurately assess exposure to TRI releases. The TRI itself shows only the type and mass of chemicals released in a given year, not where the chemicals went or precisely when they were released. Because of the uncertainty built into using these data, studies such as this must be interpreted with caution.</p><p>The study included 382 children diagnosed with brain cancer before age 10 and an equal number of cancer-free controls analyzed as pairs. Mothers of children whose brain cancer was diagnosed before age 10 years were nearly 50% more likely to have lived within 1 mile of such a site during pregnancy; the likelihood was nearly 75% higher for children diagnosed before age 5. However, when looking at risk for two major childhood brain cancer types in particular, astrocytoma and primitive neuroectodermal tumors, there was no difference.</p><p>The team used EPA Region III’s chronic toxicity index, which combines total mass released with toxicity factors including carcinogenic weight of evidence and cancer potency factors. For this study, inhalation and oral cancer potency factors were included. Other potential factors, such as mothers’ exposures in the work-place during pregnancy, children’s postnatal exposure, and exposure through contaminated drinking water, were not taken into account. The authors therefore caution that their results are not conclusive, but should be replicated and expanded using improved exposure measures.</p>
Fellowships, Grants, & Awards	Could not extract abstract	Could not extract contributor	Environmental Health Perspectives	<sec><title>Institutional Patient-Oriented Career Development Programs in the Environmental Health Sciences (K12)</title><p>The National Institute of Environmental Health Sciences (NIEHS) is committed to the support and early career promotion of new researchers who will serve in leadership roles to promote the understanding of the impact of environmental exposures on human health. The need for additional researchers is particularly acute in the integration of the environmental health sciences with research in human disease. This RFA is intended to support the early career development of doctoral-level scientists who can make substantial contributions to the understanding of human health and disease by using environmental sciences to study the etiology, pathogenesis, progression, and epidemiology of human disease.</p><p>Concerns about the declining role of physician scientists in the research workforce began to emerge over 20 years ago, and since 1994, Institute of Medicine and National Research Council reports have recommended that the National Institutes of Health increase its commitment to the development of scientists who are prepared to engage in clinical research and who are conceptually prepared to engage in research designed to shorten the time to clinical application of important basic research. Emphasis has been placed on increasing the ranks of clinical researchers and, in particular, individuals who are engaged in patient-oriented research.</p><p>In 1995, NIH convened a committee to review the status of recruitment and training of future clinical researchers. The report of this panel, issued in 1997, recommended the support of clinical research programs aimed at medical students, such as combined degree programs (MD/PhD) programs for clinical researchers; ensuring that postdoctoral training grants include formal course work or degree programs in clinical research; development of new support mechanisms for young and mid-career clinical investigators; and taking steps to reduce clinical researchers’ educational debts.</p><p>In response to these reports, in 1999, the NIH introduced three new types of career development awards: The Mentored Patient-Oriented Career Development Award (K23); the Mid-Career Patient-Oriented Research Career Development Award (K24); and the Clinical Research Curriculum Award (K30) to provide institutions with the funds to develop or expand formal course work in areas related to clinical research.</p><p>The Clinical Research Curriculum Award (K30) is designed to support the development or improvement of core courses which provide in-depth instruction in the fundamental skills, methodology, theories, and conceptualizations necessary for the well-trained, clinical researcher. This includes formal course work in the design of clinical research projects, hypothesis development, biostatistics, epidemiology, and the legal, ethical, and regulatory issues related to clinical research. The program is complementary to the Mentored Clinical Investigator (K08, K23) programs since the funding is provided for curriculum development and not for the support of the participants in the program.</p><p>More recently, the importance of the cultivation and training of a sufficiently large clinical research workforce to facilitate bench-to-bedside research and to work in the interdisciplinary, team-oriented environments that characterize current research programs has been further emphasized and promoted by a number of activities within the NIH Roadmap to accelerate medical discoveries to improve health. Under the Re-engineering the Clinical Research Enterprise, the Multidisciplinary Clinical Research Career Development Programs Request for Applications was first announced in 2003 to support the early career development of clinical researchers from a variety of disciplines engaged in all types of clinical research. The basis of training is a core curriculum, such as those developed in the Clinical Research Curriculum Award.</p><p>At the same time the NIH was recognizing the need to encourage the support of greater numbers of clinical researchers, the NIEHS was seeking to define and enlarge the role of clinical research in its programs. A series of meetings was held in 1999 to try to define the issues and develop strategies for future directions. In addition to the issues addressed by the NIH as a whole, particular impediments to the development of the field of clinical environmental health sciences research were noted. First is the lack of a discrete population base from which to attract talented physician and other patient-oriented research trained scientists. Meeting attendees recognized that the environmental health sciences need to become closely intertwined with the stream of clinical training in a variety of departments and to be made attractive to physicians and other clinical researchers with a wide set of backgrounds. Second, it must be recognized that the research collaborations of a patient-oriented research scientist may not naturally coincide with the clinical home department. Third, the importance of strong mentorship implies that the development of junior scientists engaged in patient-oriented research cannot proceed in the absence of strong environmental medicine and environmental health science role models and research funding to support clinical projects for research training. Lastly, an impediment to attracting physicians and other patient-oriented research scientists to the environmental health sciences has been that in many cases research has focused on mechanism based effects of environmental exposures so that the contribution of an environmental exposure to what is seen clinically is not well delineated. Therefore, more research into the etiology, pathophysiological progression, epidemiology, or human biological link between an environmental exposure and a clinical decrement in function is needed as a basis for the development of environmental medicine.</p><p>The objective of this Institutional Patient-Oriented Career Development Program in the Environmental Health Sciences is to establish strong, multidisciplinary institution-based programs that promote the career development and career transition of doctoral level scientists in multidisciplinary team settings to design and direct projects in patient-oriented environmental health sciences research.</p><p>This specific K12 initiative supports an institutional patient-oriented career development program in the environmental health sciences and is intended to build upon training and career development programs initiated by the NIH Roadmap activities in Re-engineering the Clinical Research Enterprise, and to complement the refocusing of the research efforts of the training and research programs of the NIEHS to include a greater emphasis on integration of basic and clinical research with a focus on environmental exposures and unique scientific opportunities relevant to the mission of NIEHS.</p><p>Programs supported by this initiative will be expected to be characterized by: 1) a core of formal didactic core experiences in research methodology for conducting hypothesis based integrative research and specific courses in environmental health sciences; 2) a base of high-quality research in environmental health sciences focusing on a particular aspect of disease etiology, pathogenesis, progression, or epidemiology; 3) a pool of talented early career patient-oriented research scholars from a variety of disciplines who plan to pursue careers in the clinical environmental health sciences; and 4) a unifying seminar series or topics course, along with grand rounds presentations and journal club, which will instill in the clinical scholars an appreciation for integrative research and research approaches to problems of environmental exposures and human biology, human pathophysiology, and human disease.</p><p>The Institutional Patient-Oriented Research Career Development Program in the Environmental Health Sciences should be fully integrated with the clinical research infrastructure of the institutional setting. A medical school and an accredited graduate school must be co-primary participants in the application. Other clinically related institutions, such as schools of public health, pharmacy, nursing, etc., are encouraged to participate. Programs are expected to incorporate a multidisciplinary approach to the development of the program and cover a spectrum of scientific disciplines critical to innovative patient-oriented research in the environmental health sciences. This includes physicians trained in numerous clinical specialties and subspecialties, and individuals with doctoral degrees in disciplines such as human genetics, pharmacology, nursing, statistics, epidemiology, psychology, veterinary medicine, and engineering. Programs should be sufficiently flexible to allow either physicians to be trained in basic environmental sciences or those with a PhD in basic science to become proficient and redirect their careers to patient-oriented research problems in the environmental health sciences.</p><p>Applications responsive to this RFA will also be expected to integrate and build on other NIEHS supported resources at the institutional setting, such at training grants (T32) (<ext-link ext-link-type="uri" xlink:href="http://www.niehs.nih.gov/dert/training/t32.htm">http://www.niehs.nih.gov/dert/training/t32.htm</ext-link>), research grants, NIEHS Core Centers (P30) (<ext-link ext-link-type="uri" xlink:href="http://grants.nih.gov/grants/guide/rfa-files/RFA-ES-05-008.html">http://grants.nih.gov/grants/guide/rfa-files/RFA-ES-05-008.html</ext-link>), NIEHS DISCOVER Centers (<ext-link ext-link-type="uri" xlink:href="http://grants.nih.gov/grants/guide/rfa-files/RFA-ES-06-001.html">http://grants.nih.gov/grants/guide/rfa-files/RFA-ES-06-001.html</ext-link>), Children’s Centers (<ext-link ext-link-type="uri" xlink:href="http://grants.nih.gov/grants/guide/rfa-files/RFA-ES-05-004.html">http://grants.nih.gov/grants/guide/rfa-files/RFA-ES-05-004.html</ext-link>), etc.</p><p>Programs must include didactic and practical training in the design, conduct and analysis of patient-oriented research along with an emphasis on methods and problems in clinical environmental health sciences research. Programs must propose to use a didactic core curriculum which is presented to all Clinical Research Scholars, such as those developed by the K30 program or the NIH Roadmap Multidisciplinary Clinical Research Career Development Program as well as courses in environmental sciences (toxicology, environmental medicine, environmental physiology, etc.)</p><p>Examples of the multidisciplinary core curriculum include: 1) clinical research methodology (including hypothesis generation, protocol design, etc.); 2) epidemiology; 3) biostatistics; 4) informatics; 5) ethical issues (with specific information on application to environmental health issues); 6) ensuring the safety of subjects (application to environmental health issues); 7) compliance with regulatory requirements for clinical research; 8) team building, leadership and management skills; 9) strategic, tactical, and negotiation skills; 10) grant writing and career development; 11) interactions with industry; 12) project/laboratory management.</p><p>This standard clinical research curriculum should be augmented with courses and other didactic experiences that are specifically designed to give the scholar a firm grounding in the environmental health sciences and a research focus on environmental exposures and the particular aspect of the disease etiology, pathogenesis, progression, or epidemiology which is central to the theme of the program.</p><p>The first and second year of the Clinical Scholar’s program should be equally divided between the core curriculum and a specific research project that uses environmental sciences to understand human disease. The third through the fifth year of the program should primarily focus the scholar’s time on their research project.</p><p>NIH defines human clinical research (<ext-link ext-link-type="uri" xlink:href="http://grants1.nih.gov/grants/policy/hs/glossary.htm">http://grants1.nih.gov/grants/policy/hs/glossary.htm</ext-link>) as: 1) patient-oriented research; 2) epidemiologic and behavioral studies; and 3) outcomes research and health services research. Studies falling under Exemption 4 for human subjects research are not considered clinical by the NIH definition. Patient-oriented research is further defined as research conducted with human subjects (or on material of human origin such as tissues, specimens, and cognitive phenomena) for which an investigator (or colleague) directly interacts with human subjects. Patient-oriented research includes: 1) mechanisms of human disease; 2) therapeutic interventions; 3) clinical trials; or 4) development of new technologies. Excluded from this definition are <italic>in vitro</italic> studies that utilize human tissues that cannot be linked to a living individual.</p><p>In order to be responsive to this announcement, all of the clinical scholars supported by the program must be engaged in patient-oriented research. Programs which propose research training in only epidemiologic research, behavioral studies, and outcomes, or intervention research will not be considered responsive to this announcement and will not be considered for review or funding.</p><p>In order to provide a cohesive focus for the program and the scholars supported, the application must describe a plan to impart an understanding of current research in the environmental health sciences, particularly the clinical and integrative aspects and emphasizing the relative roles of environmental exposures, genetics, and other co-factors in the etiology of human diseases. This may include currently available coursework, plus a seminar series, a topics course, journal clubs, discussions of case studies, grand round presentations, etc. which are specifically developed for the participants in the program.</p><p>In addition, programs should propose a unifying core of career development opportunities specifically for environmental health patient-oriented research scholars which facilitate the interactions and exchange among scholars, mentors, and the program directors, and collaboration among scholars.</p><p>Research Projects of the Scholars supported by this program are expected to have a defined impact on the environmental health sciences and be responsive to the mission of the NIEHS, which is distinguished from that of other institutes by its focus on research programs seeking to use environmental sciences to understand the cause, mechanisms, moderation, or prevention of a human disease or disorder, or relevant pathophysiologic process. Scholars who are supported by this K12 Institutional Patient-Oriented Career Development Program in the Environmental Health Sciences will also be expected to pursue research projects involving the study of an NIEHS mission relevant environmental exposure, and to pursue projects which can evolve in Clinical Career Development Award applications (K08, K23, K01, K99/R00, or research grant applications (R01, R03, R21) within the defined mission area of the NIEHS. Examples of environmental agents relevant to the NIEHS mission include industrial chemicals or manufacturing byproducts, metals, pesticides, herbicides, air pollutants and other inhaled toxicants, particulates or fibers, fungal, bacterial or biologically derived toxins. Agents considered to be outside the mission of the NIEHS which would not be appropriate research areas for scholars supported by this program include, but are not limited to: alcohol, chemotherapeutic agents, ionizing radiation, smoking (except for second hand smoke in children), drugs of abuse, pharmaceuticals, and infectious or parasitic agents, except when these are disease cofactors to an environmental toxicant exposure to produce the biological effect. Studies using model compounds are only responsive when proposals to extend the research to a relevant compound are included in the protocols.</p><p>As part of the research career development experience, scholars should understand the relevance of the exposure paradigm to human exposure, and the biological and clinical rationale for the link between the exposure and the relevant human disease. Research projects of scholars should emphasize the translational and integrative aspects of the environmental health sciences.</p><p>Each Institutional Patient-Oriented Career Development Program in the Environmental Health Sciences application should include a Governance Committee composed of scientists from the sponsoring Institution who have clinical and environmental health science research expertise, and including the Program Director and Co-Director(s). The Committee may use institutional or outside consultants if needed. The Governance Committee is responsible for making recommendations regarding the appointment of scholars to the program, monitoring scholar progress and making recommendations to the Program Director regarding their continuation, evaluating ongoing research activities for merit and relevance to the program’s theme, and making recommendations for the addition or deletion of mentors from the program. The Governance Committee is a formal part of the structure of the program. It should meet regularly and keep written minutes, which will be reviewed as part of a competing or noncompeting application. In addition, an annual evaluation is recommended.</p><p>Each Environmental Health Clinical Scholar appointed to the K12 award must be supervised by a mentoring team of at least two mentors from two different disciplines. One should have research expertise in the environmental exposure proposed in the research project and one should provide expertise in the clinical and patient-oriented research. Either through interactions with the mentoring team, or other career development activities, the mentoring team should insure the clinical scholar has appropriate scientific training in the basic and mechanistic scientific foundation for the clinical research problem under investigation.</p><p>The purpose of the K12 is to provide systematic support for the transition of clinical scientists from trainee to new mentored faculty. Therefore, how this transition is to be accomplished and the progress of the scholars monitored should be addressed in the application. Benchmarks for progress of the scholars should be outlined in the application.</p><p>Within the application, applicants must present a recruitment plan. The application should describe the potential pool of scholars, including the types of prior clinical and research training. The criteria to be used for candidate evaluation and selection should be described. In addition a plan for recruiting scholars with economically, socially, or culturally disadvantaged backgrounds, individuals with disabilities or from racial or ethnic groups that are currently underrepresented in biomedical, behavioral, or clinical sciences should be included in the recruitment plan described in the application.</p><p>Applicants should also describe a comprehensive evaluation and tracking component that will review the effectiveness of all aspects of the program (including scholars, courses, mentors, co-directors, mentoring effectiveness and institutional characteristics, and a system for tracking graduates throughout their career to determine the success rate of applying for and obtaining research support, and publications.</p><p>This funding opportunity will use the NIH Mentored Clinical Scientist Development Program (K12) institutional award mechanism.</p><p>As an applicant, you will be solely responsible for planning, directing, and executing the proposed program.</p><p>This funding opportunity uses the just-in-time budget concepts. It also uses the nonmodular budget format described in the PHS 398 application instructions (see <ext-link ext-link-type="uri" xlink:href="http://grants.nih.gov/grants/funding/phs398/phs398.html">http://grants.nih.gov/grants/funding/phs398/phs398.html</ext-link>). A detailed categorical budget for the Initial Budget Period and the Entire Proposed Period of Support is to be submitted with the application PHS 398 application instructions.</p><p>The PHS 398 application instructions are available at <ext-link ext-link-type="uri" xlink:href="http://grants.nih.gov/grants/funding/phs398/phs398.html">http://grants.nih.gov/grants/funding/phs398/phs398.html</ext-link> in an interactive format. Applicants must use the currently approved version of the PHS 398. For further assistance contact GrantsInfo, 301-435-0714 (telecommunications for the hearing impaired: TTY 301-451-0088) or by e-mail: <email>[email protected]</email>.</p><p>Applications must be prepared using the current PHS 398 research grant application instructions and forms. Applications must have a D&B Data Universal Numbering System (DUNS) number as the universal identifier when applying for Federal grants or cooperative agreements. The D&B number can be obtained by calling 866-705-5711 or through the web site at <ext-link ext-link-type="uri" xlink:href="http://www.dnb.com/us/">http://www.dnb.com/us/</ext-link>. The D&B number should be entered on line 11 of the face page of the PHS 398 form.</p><p>The letter of intent receipt date for this RFA is September 23, 2006, with the application of receipt date October 23, 2006. The complete version of this RFA is available at <ext-link ext-link-type="uri" xlink:href="http://grants.nih.gov/grants/guide/rfa-files/RFA-ES-06-005.html">http://grants.nih.gov/grants/guide/rfa-files/RFA-ES-06-005.html</ext-link>.</p><p>Contact: Carol Shreffler, Division of Extramural Research, National Institute of Environmental Health Sciences, EC-23, Building 4401, Room 3411, P.O. Box 12233, Research Triangle Park, NC 27709 USA, 919-541-1445, fax: 919-541-5064, e-mail: <email>[email protected]</email>.</p><p>Reference: RFA-ES-06-005.</p></sec>
Potential Residential Exposure to Toxics Release Inventory Chemicals during Pregnancy and Childhood Brain Cancer	<sec><title>Background</title><p>Although the susceptibility of the developing fetus to various chemical exposures is well documented, the role of environmental chemicals in childhood brain cancer etiology is not well understood.</p></sec><sec><title>Objectives</title><p>We aimed to evaluate whether mothers of childhood brain cancer cases had greater potential residential exposure to Toxics Release Inventory (TRI) chemicals than control mothers during pregnancy.</p></sec><sec sec-type="methods"><title>Methods</title><p>We included 382 brain cancer cases diagnosed at < 10 years of age from 1993 through 1997 who were identified from four statewide cancer registries. One-to-one matched controls were selected by random-digit dialing. Computer-assisted telephone interviews were conducted. Using residential history of mothers during pregnancy, we measured proximity to TRI facilities and exposure index, including mass and chemicals released. We calculated odds ratios (ORs) and 95% confidence intervals (CIs) using conditional logistic regression to estimate brain cancer risk associated with TRI chemicals.</p></sec><sec><title>Results</title><p>Increased risk was observed for mothers living within 1 mi of a TRI facility (OR = 1.66; 95% CI, 1.11–2.48) and living within 1 mi of a facility releasing carcinogens (OR = 1.72; 95% CI, 1.05–2.82) for having children diagnosed with brain cancer before 5 years of age, compared to living > 1 mi from a facility. Taking into account the mass and toxicity of chemical releases, we found a nonsignificant increase in risk (OR = 1.25; 95% CI, 0.67–2.34) comparing those with the lowest versus highest exposure index.</p></sec><sec><title>Conclusions</title><p>Risk of childhood brain cancers may be associated with living near a TRI facility; however, this is an exploratory study and further studies are needed.</p></sec>	<contrib contrib-type="author"><name><surname>Choi</surname><given-names>Hannah S.</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001113">1</xref><xref ref-type="aff" rid="af2-ehp0114-001113">2</xref></contrib><contrib contrib-type="author"><name><surname>Shim</surname><given-names>Youn K.</given-names></name><xref ref-type="aff" rid="af2-ehp0114-001113">2</xref></contrib><contrib contrib-type="author"><name><surname>Kaye</surname><given-names>Wendy E.</given-names></name><xref ref-type="aff" rid="af2-ehp0114-001113">2</xref></contrib><contrib contrib-type="author"><name><surname>Ryan</surname><given-names>P. Barry</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001113">1</xref></contrib>	Environmental Health Perspectives	<p>There is significant concern about exposure of the fetus to environmental pollutants, food additives, and drugs, which may reach the fetus through the mother and affect the brain at critical stages of development. The developing central nervous system (CNS) is much more susceptible to chemical exposures than the adult CNS, and the brain is also the major target of toxicity for congenital effects. Some toxic agents impenetrable to the adult brain freely enter the developing brain because the blood–brain barrier of the fetus is not fully developed and is not completed until approximately 6 months after birth (<xref rid="b1-ehp0114-001113" ref-type="bibr">Adinolfi 1985</xref>; <xref rid="b8-ehp0114-001113" ref-type="bibr">Johanson 1980</xref>; <xref rid="b19-ehp0114-001113" ref-type="bibr">Rodier 1994</xref>, <xref rid="b20-ehp0114-001113" ref-type="bibr">1995</xref>). Exposures to chemicals early in life are likely to have a greater impact on health outcomes such as cancer, neurodevelopmental impairment, and immune dysfunction (<xref rid="b23-ehp0114-001113" ref-type="bibr">Thomas 1995</xref>).</p><p>Although the susceptibility of the developing fetus to various chemical exposures is well documented, the role of environmental chemicals in childhood brain cancer etiology is not well understood. The best established environmental risk factor for childhood brain cancer is radiation exposure (Harvey et al. 1985; <xref rid="b10-ehp0114-001113" ref-type="bibr">Kuijten and Bunin 1993</xref>; <xref rid="b11-ehp0114-001113" ref-type="bibr">MacMahon 1962</xref>; <xref rid="b14-ehp0114-001113" ref-type="bibr">Mole 1974</xref>; <xref rid="b21-ehp0114-001113" ref-type="bibr">Ron et al. 1988</xref>). Therapeutic cranial irradiation (X rays) has repeatedly been linked to childhood brain cancer, whereas diagnostic X rays with their usual low dose and short exposure periods were not enough to result in disease outcome (<xref rid="b10-ehp0114-001113" ref-type="bibr">Kuijten and Bunin 1993</xref>).</p><p>Some studies investigated the risk for childhood cancer and birth defects among people living near hazardous waste sites and as a result of chemical exposures from the environment. Residential location has been a concern because toxic chemicals in landfill may disperse into the air or soil, eventually leading to human exposure (<xref rid="b5-ehp0114-001113" ref-type="bibr">Elliot et al. 2001</xref>). In Clinton County, Pennsylvania, increased risk for cancer death was observed near the Drake Superfund site (<xref rid="b3-ehp0114-001113" ref-type="bibr">Budnick et al. 1984</xref>). Other studies have attempted to find a link between childhood cancers and residential proximity to hazardous waste sites, with mixed results (<xref rid="b9-ehp0114-001113" ref-type="bibr">Knox 2000</xref>; <xref rid="b31-ehp0114-001113" ref-type="bibr">White and Aldrich 1999</xref>). In the Dover Township case–control study, a significantly increased odds ratio was observed for potential exposure to ambient air pollutants among female children with leukemia. However, no significant risks were found for brain cancer (<xref rid="b17-ehp0114-001113" ref-type="bibr">New Jersey Department of Health and Senior Services and Agency for Toxic Substances and Disease Registry 2003</xref>).</p><p>One source of information on environmental contaminants is the Toxics Release Inventory (TRI), which is managed by the U.S. Environmental Protection Agency (<xref rid="b27-ehp0114-001113" ref-type="bibr">U.S. EPA 2006</xref>). The TRI was established by a mandate of the <xref rid="b6-ehp0114-001113" ref-type="bibr">Emergency Planning and Community Right-To-Know Act (EPCRA) of 1986</xref>. The TRI database contains an annual report of chemical releases to the environment and transfers of chemicals to off-site locations. The TRI captures the mass of specific compounds released into the environment (routinely or by accident) and those otherwise managed as waste. The mass of compounds released is considered to be relatively constant over the reporting period because what is released routinely and as waste usually exceeds what is accidentally released. As of 2002, > 650 toxic chemicals and chemical compounds are required to be reported.</p><p>The TRI database has been used in several studies. A link to slight increases in risk for certain birth defects associated with toxic releases (<xref rid="b13-ehp0114-001113" ref-type="bibr">Marshall et al. 1997</xref>) has been suggested, but potential links to childhood cancers have not yet been investigated. The most common use of the TRI database has been to add up the total mass of TRI chemicals released to identify the most problematic polluters. However, this method emphasizes volume without regard to toxicity or environmental fate. Further, the total mass of chemicals released does not equal actual concentrations in the environment nor actual exposures to populations (<xref rid="b15-ehp0114-001113" ref-type="bibr">Neumann 1998</xref>).</p><p>Previous studies used several different methods of categorizing exposure. <xref rid="b13-ehp0114-001113" ref-type="bibr">Marshall et al. (1997)</xref> evaluated the risk of CNS and musculoskeletal birth defects from exposure to solvents, metals, and pesticides from hazardous contaminant sites including TRI sites in New York State. This case–control study rated the probability of exposure as “high,” “medium,” “low,” or “unknown” for each contaminant group, using a standard, 1-mi radius template divided into 25 sectors. The templates were centered on the geographic coordinates of each contaminant site, overlaying with residential address at birth. This study found that residing within 1 mi of a TRI facility that released solvents had a significantly elevated risk for CNS defects with an odds ratio (OR) of 1.3.</p><p><xref rid="b16-ehp0114-001113" ref-type="bibr">Neumann et al. (1998)</xref> attempted to create a method for incorporating the toxicity factors so that the TRI data are more useful in estimating concentrations in the environment and potential effects from exposure. The chronic toxicity index was developed by the U.S. EPA’s Region III Air Radiation and Toxics Division using the TRI databases and chronic oral toxicity factors and total mass for both carcinogens and noncarcinogens to estimate the relative hazards of TRI chemicals. The investigators used oral reference doses and cancer potency factors for the chronic toxicity index and ranked TRI chemicals on the basis of total mass versus total chronic toxicity index. The results varied greatly (<xref rid="b16-ehp0114-001113" ref-type="bibr">Neumann et al. 1998</xref>). Even though the chronic toxicity index has its own limitations, it is likely to be a better indicator of potential risk than the use of mass alone.</p><p>The primary objective of this study was to investigate whether mothers of childhood brain cancer cases had greater potential residential exposure to TRI chemicals than control mothers during pregnancy. We assessed potential exposure by considering residential proximity to TRI facility during pregnancy, whether carcinogens were emitted, and a comparative ranking system for TRI chemical releases by combining toxicity information and total mass of release.</p><sec sec-type="materials\|methods"><title>Materials and Methods</title><sec sec-type="methods"><title>Study population</title><p>Subjects who participated in the U.S. Atlantic Coast childhood brain cancer study, a population-based case–control study of environmental risk factors (<xref rid="b2-ehp0114-001113" ref-type="bibr">Agency for Toxic Substances and Disease Registry 2004</xref>), were eligible for the TRI study. Briefly, cases eligible for the original Atlantic coast childhood brain cancer study included all incident cases of first primary brain cancer [<italic>International Classification of Diseases for Oncology</italic> (ICD-O-2) (<xref rid="b32-ehp0114-001113" ref-type="bibr">World Health Organization 1990</xref>) codes C71.0–C71.9 including all morphologic codes with a behavior code of 3, excluding lymphomas] (<xref rid="b18-ehp0114-001113" ref-type="bibr">Percy et al. 1990</xref>) diagnosed at < 10 years of age between 1993 and 1997, born in the United States, and a resident of one of the four states (Florida, New Jersey, New York excluding New York City, and Pennsylvania) at the time of diagnosis. In addition, an eligible case had to have the biological mother available for an interview in English and a telephone in the household. During the computer-assisted telephone interview, a standardized screening questionnaire was used to verify eligibility and obtain mothers’ consent to participate in the study. The study protocol was approved by the Centers for Disease Control and Prevention and four state institutional review boards. The four statewide cancer registries initially identified 937 case children. Eligibility screening interviews were not completed for 228: three (0.3%) physician refusals, two (0.2%) out-of-state children, 176 (18.8%) unable to be traced, 39 (4.2%) mother refusals, and eight (0.9%) with language barriers. Of the 709 case children for whom screening interviews were completed, 662 met the eligibility criteria, and 535 mothers of the 662 agreed to participate. Of the 535, nine were excluded because of difficulties in finding matched controls, and 526 were included in the original study (56.1% of the originally identified 937 cases or 79.5% of the 662 eligible case children).</p><p>Potential controls in the original study were identified from the study base population through random-digit dialing (RDD) (<xref rid="b28-ehp0114-001113" ref-type="bibr">Wacholder et al. 1992</xref>; <xref rid="b29-ehp0114-001113" ref-type="bibr">Waksberg 1978</xref>). Eligible controls had to be born in the United States, be free of cancer, have the biological mother available for an interview in English, and have a telephone in the household. An equal number of controls were selected by matching individually to cases on sex, race (white, black, or other), birth year (± 1 year), and state of residence at the time of cases’ diagnosis. The age at diagnosis of each case was used as a reference age for the corresponding control. Among the 20,802 RDD numbers prescreened for nonworking and nonvoice numbers, each of 3,553 (17.1%) households had a child meeting the eligibility criteria for the control selection. Of the 3,553 children, 820 (23.1%) met the matching criteria. Of the 820 meeting the matching criteria, 122 did not have a matching case available, 102 mothers refused to participate, and 526 agreed to participate (2.5% of the 20,802 working residential numbers or 83.8% of the 628 eligible control children for whom a matching case child was available).</p><p>This TRI study included 764 subjects (382 case–control pairs) of the 1,052 (526 case–control pairs) participants in the original study: 222 subjects born before 1988 were excluded because the reporting of TRI information began in 1987; 34 subjects who had incomplete pregnancy residential information or dates of residence were excluded; 32 subjects missing their matched case or control counterparts were excluded.</p></sec><sec><title>Computer-assisted telephone interview</title><p>The biological mothers of cases and controls were interviewed in English using a computer-assisted telephone interview system. Bilingual (Spanish) interviewers were available. Mothers were asked to provide information on residential history of the parents and child from 24 months before the child’s birth until the age of diagnosis or reference age (i.e., age at diagnosis for counterpart case) for controls. Interviewers were instructed to obtain residential addresses and to take nearest intersecting street names when the street numbers were unavailable. The questionnaire also included information on demographic characteristics and on mothers’ smoking habits during pregnancy.</p></sec><sec><title>Exposure assessment</title><p>Addresses of mothers during pregnancy for the 10 months before birth were geocoded with latitude and longitude coordinates using GeoCoder (version 3.4b; GeoAccess Inc., Lenexa, KS). This software package was used with the TRI facilities’ geographic coordinates to determine exact distances from each residence to all facilities within a 2-mi radius. We geocoded 624 of 928 (67.2%) pregnancy addresses in the first round. We located 288 of 928 (31%) unable to be geocoded in the first round because of invalid addresses or zip codes using database records, public records, court records, and calls to post offices; these were gecoded in the second or third round. A total of 912 of 928 (98.3%) pregnancy addresses were successfully geocoded.</p><p>We extracted the TRI data from the U.S. EPA TRI CD-ROM containing information for the years 1987–1997 (<xref rid="b24-ehp0114-001113" ref-type="bibr">U.S. EPA 2000</xref>). To assess the quality of geocoded data obtained from the TRI database, we randomly selected approximately 5% of the TRI facilities’ addresses used in this study and matched those to addresses on the Streetmap 2000 street layer residing on the spatial data engine using ArcGIS (version 8.1; Environmental Systems Research Institute, Inc., Redlands, WA). Distance measurements for both were calculated. The range of difference was between 0.017 and 0.534 mi, with a mean value of 0.3455 mi (0.060, 0.119, 0.343 mi for 25th, 50th, and 75th percentiles respectively). The original plan to use the 0.5-mi radius or cutoff point as a potential exposure category was abandoned because these distances were deemed unstable. We retained the 1.0-mi and 2.0-mi radii as proximity measures.</p><p>We calculated the distance from the mother’s residence during any point in pregnancy to the nearest TRI facility and categorized the exposure levels as residing ≤ 1 mi versus > 1 mi, and ≤ 2 mi versus residing > 2 mi of any facility. Next, we investigated whether any carcinogen was released to the air from facilities within 1 mi versus > 1 mi and within 2 mi versus > 2 mi of any facility. The TRI air emissions of any class of carcinogens were categorized as dichotomous variables without regard to the amount released to the air. Air emissions included stack and fugitive air releases. Carcinogens as defined by the U.S. EPA included all known, probable, and possible human carcinogens (<xref rid="b25-ehp0114-001113" ref-type="bibr">U.S. EPA 2002a</xref>); EPCRA section 313 lists toxic chemicals that meet the Occupational Safety and Health Administration carcinogen standard and are associated with the 0.1% <italic>de minimis</italic> concentration limit when in a mixture (<xref rid="b25-ehp0114-001113" ref-type="bibr">U.S. EPA 2002a</xref>).</p><p>Finally, we chose a hazard-screening tool for exposure assessment. To comparatively rank TRI chemical releases, we adapted the chronic toxicity index developed by the U.S. EPA’s Region III (<xref rid="b16-ehp0114-001113" ref-type="bibr">Neumann et al. 1998</xref>). The screening tool uses the TRI databases combining toxicity factors and total mass to estimate the relative hazards of TRI chemical releases with a separate algorithm for carcinogens and noncarcinogens. For carcinogens, the carcinogenic weight of evidence (WOE) and cancer potency factors (CPF) and the pounds of chemicals released are included in the index calculation. The WOE data were obtained from the Integrated Risk Information System (<xref rid="b26-ehp0114-001113" ref-type="bibr">U.S. EPA 2002b</xref>), a database of human health effects that may result from exposure to environmental substances. We used the U.S. EPA Region II Risk-Based-Concentration Table (<xref rid="b25-ehp0114-001113" ref-type="bibr">U.S. EPA 2002a</xref>) to obtain the CPF for the inhalation or ingestion routes of exposure. Although the likely exposure route would be through the inhalation route, chemicals with only oral CPF were included in the index using the oral CPF value.</p><p>We modified the chronic toxicity index to include the duration of residence and the distance to the TRI facility. With some subjects’ pregnancy period spanning 2 calendar years, duration of residence at each address during pregnancy for each calendar year was calculated separately to match it with the appropriate year-specific TRI data. Because the TRI data report the total amount of emissions during a calendar year, the number of months a woman lived at a particular address for the particular year while pregnant was divided by 12 months and then multiplied into the chronic toxicity index. Only known, probable, and possible carcinogens, as defined by the <xref rid="b25-ehp0114-001113" ref-type="bibr">U.S. EPA (2002a)</xref>, that were released within 2 mi of pregnancy residence and having the appropriate carcinogenic WOE and CPF information available were included. We incorporated the duration of exposure, and residential distance to the facilities to the chronic toxicity index:</p><disp-formula><graphic xlink:href="ehp0114-001113e1.jpg" position="float" mimetype="image"/></disp-formula></sec><sec sec-type="methods"><title>Statistical analysis</title><p>We used conditional logistic regression analyses to achieve maximum likelihood estimates of ORs and 95% confidence intervals (CI) for the exposure variables. Exposure variables for residential proximity and residing near a facility releasing carcinogens were categorized as ≤ 1 mi versus > 1 mi, and ≤ 2 mi versus > 2 mi. We categorized the exposure index into three levels using the following cut-point values: zero; greater than zero but less than median index value among controls; and greater than median index value among controls. The potential confounders examined included mother’s education, household income level, and mother’s pregnancy age. Because there were no substantial confounding effects from these variables, judged by the change-inestimate methods (i.e., 10% change in OR), unadjusted ORs are presented. Because it is possible that the effect of potential gestational exposure may be more relevant to cancer development in earlier childhood or to particular histological subtype of childhood brain cancer, we repeated the analysis by reference age (< 5 and ≥ 5 years) and by two major histological subtypes, primitive neuroectodermal tumors (PNET) and astrocytomas [ICD-O-2 codes 9400–9441 and 9470–9473, respectively (<xref rid="b18-ehp0114-001113" ref-type="bibr">Percy et al. 1990</xref>)]. All statistical analyses were conducted using SAS software (version 8.02; SAS Institute Inc., Cary, NC).</p></sec></sec><sec sec-type="results"><title>Results</title><sec sec-type="intro\|methods"><title>Demographics and histopathologic characteristics of the study population</title><p><xref ref-type="table" rid="t1-ehp0114-001113">Table 1</xref> shows the distribution of the histopathologic types of the brain tumor among cases and controls. Most of the case and control children were white (88%); 11% were black and only 1.6% were classified as other. There were 233 pairs (61%) with a reference age (age at diagnosis for cases) of < 5 years. Most of the children (72%) were born before 1993. The distribution of mothers’ age at pregnancy was similar in cases and controls (<xref ref-type="table" rid="t1-ehp0114-001113">Table 1</xref>). Case mothers’ education levels were slightly higher than control mothers’ education levels, but the household income levels were slightly higher for the controls. Astrocytomas were the most common type; about half the cases had astrocytomas whereas 29% had PNETs (<xref ref-type="table" rid="t2-ehp0114-001113">Table 2</xref>).</p><p>Overall, 635 case and control mothers lived at one address for the entire pregnancy. The remaining 129 (17%) mothers had lived at more than one address during pregnancy: 121 mothers with two and eight with three addresses. The resulting total was 901 addresses. Mothers’ residences during pregnancy were located in 23 different states for the case mothers and 18 different states for control mothers. However, 94% of both case and control mothers lived during the entire pregnancy in one of the four states—Florida, New Jersey, New York (excluding New York City), or Pennsylvania.</p></sec><sec><title>Residential proximity to TRI facilities during pregnancy</title><p>We identified a total of 1,624 different TRI facilities within 2 mi of any of the case and control residences. The case mothers had a higher frequency of living within 1 and 2 mi of any TRI facility than control mothers at any point during pregnancy. <xref ref-type="table" rid="t3-ehp0114-001113">Table 3</xref> shows the results of analyses comparing cases and controls living within 1 mi versus > 1 mi from TRI facilities and living within 2 mi versus > 2 mi. Living within 1 mi of any TRI facilities during pregnancy showed slightly increased OR for all reference ages (OR 1.32; 95% CI, 0.96–1.80) and a statistically significant OR for those < 5 years of age at diagnosis (OR 1.66; 95% CI, 1.11–2.48) compared to living > 1 mi. For living within 1 mi versus > 1 mi from a TRI facility releasing carcinogens, the OR was 1.48 (95% CI, 1.01–2.17) for all ages, and 1.72 (95% CI, 1.05–2.82) for those < 5 years of age at diagnosis. Analysis by tumor types, astrocytoma and PNET, was associated with increased risk estimates, but the results were not statistically significant (<xref ref-type="table" rid="t4-ehp0114-001113">Table 4</xref>). For astroycytoma, living within 1 mi of any TRI facility had an OR of 1.18 (95% CI, 0.77–1.82) compared to living > 1 mi from any facility, and living within 1 mi of a facility releasing carcinogens had an OR of 1.32 (95% CI, 0.79–2.22) compared to living > 1 mi from a facility releasing carcinogens.</p></sec><sec><title>Exposure index</title><p>Of 193 TRI compounds classified as known, probable, or possible carcinogens, 55 compounds were actually released within 2 mi of residences of the study population during pregnancy. From those 55 compounds, we obtained information on 26 compounds and calculated the exposure indices for them. The most common compounds released within 2 mi of residence for individuals in the study population were dichloromethane, nickel and nickel compounds, styrene, lead, trichloroethylene (TCE), formaldehyde, and di(2-ethylhexyl) phthalate. Compounds with the highest exposure index values for residential addresses were 1,3-butadiene, ethylene oxide, dichloromethane, chloroform, and vinyl chloride.</p><p>There was an increasing risk trend as the exposure index level increased for those with a reference age of < 5 years: Compared to subjects with an exposure index of zero, the ORs were 1.24 (95% CI, 0.67–2.28) for subjects with an exposure index of greater than zero and less than the median and 1.25 (95% CI, 0.67–2.34) for subjects with an exposure index of greater than the median (<xref ref-type="table" rid="t5-ehp0114-001113">Table 5</xref>). However, the increasing trend was not statistically significant (<italic>p</italic> = 0.38). No increasing trend in risks for two major subtypes of brain cancer, astrocytoma and PNETs, was observed by the increasing exposure index level (<xref ref-type="table" rid="t6-ehp0114-001113">Table 6</xref>).</p><p>Because some of the carcinogens did not have the appropriate toxicity information, a separate analysis was conducted by calculating the exposure index only with the mass of compounds released, duration at each residence, and distance to the facility. However, the results did not differ and elevated risk was not observed.</p></sec></sec><sec sec-type="discussion"><title>Discussion</title><p>Environmental epidemiology studies constantly struggle with ways to assess past exposure. Although a number of databases include information on the release of chemicals, these were collected mostly for regulatory purposes and therefore lack the individual specificity desired for these studies. Nonetheless, it is important to try to use these data in creative ways if we are to have any information at all on past exposures. Because of the uncertainty built into using these data, studies such as this must be interpreted with caution. In this study we used data from the TRI to assess exposure in three different ways: living within a specified distance of a TRI facility (1 or 2 mi), living within a specified distance of a TRI facility emitting a carcinogen (1 or 2 mi), and a toxicity index that took into consideration the toxicity of the chemical released and the duration of the exposure in addition to distance from a TRI facility. Actual individual exposure measures for specific chemicals were not available for this study.</p><p>We observed an elevated risk for mothers living within 1 mi of a TRI facility and living within 1 mi of a facility releasing carcinogens for having children with brain cancer diagnosed before 10 years of age. The odds ratios were higher for brain cancer cases diagnosed before age 5 years. For the exposure assessment using the exposure index, we observed an increasing risk trend as the exposure index level increased, although the trend was not statistically significant. Nevertheless, since the number of subjects that actually had a positive exposure index value was small, <italic>p</italic>-values would have been affected by the small sample size.</p><p>It is not feasible to compare the results of this current analysis with previous studies because similar studies linking childhood brain cancers with TRI releases are not available. However, similar methods of exposure assessment were used in previous studies on central nervous system birth defects from possible exposure to TRI sites (<xref rid="b4-ehp0114-001113" ref-type="bibr">Croen et al. 1997</xref>; <xref rid="b13-ehp0114-001113" ref-type="bibr">Marshall et al. 1997</xref>). <xref rid="b13-ehp0114-001113" ref-type="bibr">Marshall et al. (1997)</xref> observed an increased risk for CNS defects associated with living within 1 mi of a facility emitting either solvents or metals into the air; however, they did not observe a dose–response trend as distance to TRI facilities was reduced. It is interesting that the 1-mi cutoff for exposure categorization resulted in significant risk for CNS defects (<xref rid="b13-ehp0114-001113" ref-type="bibr">Marshall et al. 1997</xref>), but there was a lack of association when distance was further subcategorized within 1 mi.</p><p>Although prenatal residential proximity to TRI facilities resulted in a statistically significant increased risk for childhood brain cancer, it is imprudent to associate that with actual exposure to any compounds released, so results should be interpreted accordingly. Several issues concerning exposure assessment must be taken into account. Some of the limitations of this analysis include concern over accuracy of residential history data, limitations of the TRI data themselves, and methods of exposure assessment.</p><p>Residential history information used in this analysis comprised self-reported responses from mothers of cases and controls. There is potential for recall and reporting bias that is further compounded by the fact that some subjects had to provide information dating back 10 years. Inaccurate address information for cases and controls that made it impossible to assign geocoding information meant that distances to TRI facilities could not be determined, so that some cases and controls had to be excluded from the study. The concern here is selection bias, because subjects who were living in rural areas, less educated, or frequent movers may have been more likely to be excluded (<xref rid="b30-ehp0114-001113" ref-type="bibr">Ward et al. 2000</xref>). However, only 11 of the 830 children born after 1988 were missing information on distance to TRI facilities, and 23 of the 830 children were missing mothers’ pregnancy residential information, for a total of only 34 of the 830 (4%), which is likely too small of a number to introduce such a bias.</p><p>Another limitation of this study lies with the TRI data themsleves: They are self-reports from companies and it is difficult to assess the accuracy of the data. Facilities with < 10 full-time employees or those not meeting TRI quantity thresholds are not required to report releases. Thus exposure experienced by both cases and controls may be higher than estimated through the TRI, because such facilities also may contribute to the overall pollutant burden in the community. The variability in exposure arising from these unreported emissions relative to those arising from TRI facilities is unknown. Also, chemical releases and waste generation are estimated and do not provide measurement of actual concentrations in the environment (<xref rid="b16-ehp0114-001113" ref-type="bibr">Neumann et al. 1998</xref>).</p><p>For the first two levels of analysis using proximity and the release of carcinogens, residing near multiple facilities or multiple compounds released was not accounted for, although an attempt was made to include them in the exposure index. The exposure index has its own limitations because not all the TRI compounds have a toxicity value necessary for obtaining the chronic index. Several compounds lacked the inhalation data requiring oral toxicity factors to be used to estimate the index. However, preliminary findings suggest that substituting oral factors for inhalation did not change the final rank of TRI emission using the chronic index approach (<xref rid="b16-ehp0114-001113" ref-type="bibr">Neumann et al. 1998</xref>).</p><p>This study did not account for other potential confounders such as mother’s exposure to chemicals at the workplace during pregnancy. The TRI is just one source of information on environmental releases. Other sources of air pollution such as toxic emissions from cars or other hazardous waste sites were not included. Only TRI air emissions data were extracted for the analysis, so we did not explore possible exposure through contaminated drinking water. The pathway of exposure through contaminated drinking water is more difficult to assess for each individual; the location of TRI sites may or may not have resulted in water contamination because municipal water wells are not directly related to location of residences (<xref rid="b12-ehp0114-001113" ref-type="bibr">Marshall et al. 1993</xref>). We would need to know whether private wells or municipal water wells were the principal source of water and determine if they were possibly contaminated by TRI chemical releases.</p><p>Although we used only the period of 10 months before birth, many mothers and their children lived in the same residential address long after birth but these exposure data were not included in the analysis. Therefore, it is difficult to rule out effects of potential exposure after birth. Further studies may be conducted to determine whether children who had lived at the same address from pregnancy to early childhood may have been exposed to further environmental chemical releases and possibly had a higher risk than those exposed only prenatally. Furthermore, because the TRI facilities report the annual releases and transfer without indicating the specific time and date of the release, it is possible that the actual releases occurred outside of the 10-month pregnancy period we examined.</p><p>There are several strengths in this study. This is the only study to date to examine the role of TRI releases and childhood brain cancer. In addition, this study included a large number of cases and controls drawn from the general population. We attempted to improve and build on previous exposure assessment methods. Most previous studies such as those dealing with environmental equity have compared populations using census tracts and circular zones of different distances around hazardous waste sites and compared population characteristics within and outside of those boundaries (<xref rid="b22-ehp0114-001113" ref-type="bibr">Sheppard et al. 1999</xref>). Some have used ZIP-code boundaries (<xref rid="b31-ehp0114-001113" ref-type="bibr">White and Aldrich 1999</xref>); however, ZIP codes have irregular boundaries, which do not indicate any specific relation to the hazardous waste site. We used direct distance to the TRI facilities and attempted to incorporate the amount as well as the toxicity of compounds released through the use of the chronic toxicity index.</p><p>Most published studies relied on the address on the birth certificate, which may not give a true picture of residence throughout the entire pregnancy. In this study we used residential addresses during pregnancy that were obtained from a survey question on residential history, rather than using the address at time of birth, and included multiple addresses when applicable.</p><p>Our results suggest a possible relationship between living within 1 mi of any TRI facility or a TRI facility emitting carcinogens during pregnancy and a child’s later developing childhood brain cancer. However, there are many uncertainties as to why such a relationship exists and why the same relationship was not found for living within half a mi of a facility. Most of the limitations discussed would be expected to bias the risk estimates toward the null and obscure any true association; however, it is unclear how other limitations might affect the risk estimate.</p><p>Despite the inherent limitations in using these data for epidemiology studies, research in this area needs to continue to refine their use. Further studies need to be conducted to explore whether these results can be replicated and also address and improve on some of the limitations described. Although this was a large study of childhood brain cancer including > 300 cases and 300 controls, this study was not designed to focus on specific chemicals because the number of cases and controls with potential to exposure specific carcinogens would be too small to warrant meaningful analysis. Therefore, it is not possible to pinpoint the specific agents that may have increased the risk for brain cancer. There is the potential for further improving on exposure assessment methods by using an exposure index using a larger sample size or by obtaining more complete toxicity and exposure information for the compounds.</p></sec><sec><title>C<sc>orrection</sc></title><p>In <xref ref-type="table" rid="t2-ehp0114-001113">Table 2</xref>, the value for “All other” has been corrected from 17 (4.5), as published online, to 75 (19.6); in <xref ref-type="table" rid="t5-ehp0114-001113">Table 5</xref>, the value for “All reference ages” exposure index level II has been changed from 1.91 to 0.91.</p></sec>
A Killer Smell: Mold Toxin Destroys Olfactory Cells in Mice	Could not extract abstract	<contrib contrib-type="author"><name><surname>Wakefield</surname><given-names>Julie</given-names></name></contrib>	Environmental Health Perspectives	<p>Mold seems ubiquitous: it permeates spaces made damp by leaking water lines, faulty roofs, or storm flooding. Although no one contests that its slimy presence is a general nuisance, its related adverse health effects have been the subject of some controversy. Now researchers at Michigan State University’s Center for Integrative Toxicology have found that a toxin produced by the black mold <italic>Stachybotrys chartarum</italic> can damage nerve cells key to the sense of smell, at least in the noses of mice <bold>[<italic>EHP</italic> 114:1099–1107; Islam et al.]</bold>. The study is the first to probe how inhaling black mold toxins affects nasal passages.</p><p>Other researchers have previously reported links between <italic>S. chartarum</italic> exposure and human health effects including upper and lower respiratory illnesses. There is also evidence of an association between exposure to fungi in a damp indoor environment and effects such as asthma symptoms in sensitive individuals. However, in a recent Institute of Medicine report, a panel of experts concluded that there is limited or insufficient evidence to determine whether an association exists for other suggested health outcomes such as chronic obstructive pulmonary disease, neuropsychiatric symptoms, skin symptoms, and immune diseases.</p><p>The Michigan team found that a single low dose of satratoxin G administered directly into the noses of mice selectively killed sensory neurons involved in detecting odors and sending signals to the olfactory bulbs in the brain. Satratoxins are a type of mycotoxin found in the spores and other parts of <italic>S. chartarum</italic>. The toxins killed the olfactory neurons by apoptosis while apparently leaving bystander cells unharmed. The mice that inhaled the fungal toxins also developed inflammation of the nasal passages and rhinitis (“runny nose” symptoms), as well as milder inflammation of the olfactory bulbs.</p><p>It is still unclear how these findings apply to humans exposed to molds. Moreover, before broader health impacts may be assessed, both the amounts of mycotoxins in the air and the nature of human exposure need to be better understood, as do the effects of mold toxins on humans’ sense of smell and nasal inflammation. On first examination, however, these mouse studies suggest that exposure to airborne mold toxins may adversely affect people’s ability to smell. At a minimum, the study raises new questions about the hazards of exposure to black mold in water-damaged buildings.</p>
Does Living Near a Superfund Site Contribute to Higher Polychlorinated Biphenyl (PCB) Exposure?	<p>We assessed determinants of cord serum polychlorinated biphenyl (PCB) levels among 720 infants born between 1993 and 1998 to mothers living near a PCB-contaminated Superfund site in Massachusetts, measuring the sum of 51 PCB congeners (∑PCB) and ascertaining maternal address, diet, sociodemographics, and exposure risk factors. Addresses were geocoded to obtain distance to the Superfund site and neighborhood characteristics. We modeled log<sub>10</sub>(∑PCB) as a function of potential individual and neighborhood risk factors, mapping model residuals to assess spatial correlates of PCB exposure. Similar analyses were performed for light (mono–tetra) and heavy (penta–deca) PCBs to assess potential differences in exposure pathways as a function of relative volatility. PCB-118 (relatively prevalent in site sediments and cord serum) was assessed separately. The geometric mean of ∑PCB levels was 0.40 (range, 0.068–18.14) ng/g serum. Maternal age and birthplace were the strongest predictors of ∑PCB levels. Maternal consumption of organ meat and local dairy products was associated with higher and smoking and previous lactation with lower ∑PCB levels. Infants born later in the study had lower ∑PCB levels, likely due to temporal declines in exposure and site remediation in 1994–1995. No association was found between ∑PCB levels and residential distance from the Superfund site. Similar results were found with light and heavy PCBs and PCB-118. Previously reported demographic (age) and other (lactation, smoking, diet) correlates of PCB exposure, as well as local factors (consumption of local dairy products and Superfund site dredging) but not residential proximity to the site, were important determinants of cord serum PCB levels in the study community.</p>	<contrib contrib-type="author"><name><surname>Choi</surname><given-names>Anna L.</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001092">1</xref></contrib><contrib contrib-type="author"><name><surname>Levy</surname><given-names>Jonathan I.</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001092">1</xref></contrib><contrib contrib-type="author"><name><surname>Dockery</surname><given-names>Douglas W.</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001092">1</xref></contrib><contrib contrib-type="author"><name><surname>Ryan</surname><given-names>Louise M.</given-names></name><xref ref-type="aff" rid="af2-ehp0114-001092">2</xref></contrib><contrib contrib-type="author"><name><surname>Tolbert</surname><given-names>Paige E.</given-names></name><xref ref-type="aff" rid="af3-ehp0114-001092">3</xref></contrib><contrib contrib-type="author"><name><surname>Altshul</surname><given-names>Larisa M.</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001092">1</xref></contrib><contrib contrib-type="author"><name><surname>Korrick</surname><given-names>Susan A.</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001092">1</xref><xref ref-type="aff" rid="af4-ehp0114-001092">4</xref></contrib>	Environmental Health Perspectives	<p>Polychlorinated biphenyls (PCBs) are persistent synthetic organic chemical pollutants found in air, water, sediments, and soil. Because of concern over their toxicity and persistence in the environment, the manufacture of PCBs was banned in the United States in 1977, resulting in declines in environmental PCB levels (<xref rid="b26-ehp0114-001092" ref-type="bibr">Longnecker et al. 1997</xref>). However, exposure to PCBs continues because of their presence in products manufactured before 1977, the disposal of PCB-contaminated products in landfills and hazardous waste sites, and their environmental persistence and bioaccumulative characteristics.</p><p>The developing fetus is particularly vulnerable to exposure to environmental toxins (<xref rid="b10-ehp0114-001092" ref-type="bibr">Fein et al. 1983</xref>). PCBs readily cross the placenta, and prenatal PCB exposure has been associated with decreased birth weight (<xref rid="b9-ehp0114-001092" ref-type="bibr">Fein et al. 1984</xref>; <xref rid="b30-ehp0114-001092" ref-type="bibr">Patandin et al. 1998</xref>) and decrements in cognitive function in childhood (<xref rid="b16-ehp0114-001092" ref-type="bibr">Jacobson et al. 1996</xref>; <xref rid="b34-ehp0114-001092" ref-type="bibr">Stewart et al. 2003</xref>; <xref rid="b40-ehp0114-001092" ref-type="bibr">Vreugdenhil et al. 2002</xref>). However, some studies have not demonstrated adverse associations of early-life PCB exposures with prenatal growth or childhood cognition (<xref rid="b12-ehp0114-001092" ref-type="bibr">Gladen and Rogan 1991</xref>; <xref rid="b13-ehp0114-001092" ref-type="bibr">Gray et al. 2005</xref>). Given their potential health hazards, it is important to understand risk factors (including potentially remediable ones) for PCB exposure among infants and children.</p><p>Among general population samples, diet, particularly consumption of contaminated fish and other animal products, is a major source of PCB exposure. Other potential pathways for PCB exposure include inhalation and dermal contact, both occupationally and in the ambient environment (<xref rid="b6-ehp0114-001092" ref-type="bibr">DeCaprio et al. 2005</xref>; <xref rid="b25-ehp0114-001092" ref-type="bibr">Löffler and Bavel 2000</xref>). Among reproductive-age women, reported correlates of serum PCB levels include older age, alcohol consumption, parity, and lactation (<xref rid="b17-ehp0114-001092" ref-type="bibr">Jacobson et al. 1984</xref>; <xref rid="b31-ehp0114-001092" ref-type="bibr">Rogan et al. 1986</xref>). However, risk factors for nonoccupational PCB exposure vary among populations and regions. Residential proximity to a contaminated site may be an important risk factor for PCB exposure, as it may capture both direct exposure pathways (inhalation or dermal contact) and socioeconomic- or lifestyle-related exposure risks.</p><p>The New Bedford Harbor in southeastern Massachusetts is contaminated with PCBs as a result of waste disposal from local industry from the 1940s until 1977. The most PCB-contaminated sediment or “hot spot” was in the harbor estuary adjacent to a capacitor manufacturer (<xref ref-type="fig" rid="f1-ehp0114-001092">Figure 1</xref>) (<xref rid="b41-ehp0114-001092" ref-type="bibr">Weaver 1984</xref>). In 1982 the harbor was designated a Superfund site. As part of the remediation plan, the most contaminated sediments were dredged between April 1994 and September 1995 [U.S. Environmental Protection Agency (U.S. EPA) 1999].</p><p>The present study was undertaken to assess whether residential proximity to this PCB-contaminated site or related local factors (e.g., consumption of locally produced foods) were associated with higher cord serum PCB levels among infants of mothers living near the harbor. We characterized PCB exposure incorporating known exposure pathways and individual risk factors with geographic information system methods to assess spatial correlates of cord serum PCB levels. We evaluated exposure pathways as well as temporal variability in exposure for PCBs and, because of their unique toxicologic properties, dioxin toxic equivalent (TEQ) levels for dioxin-like PCBs (<xref rid="b37-ehp0114-001092" ref-type="bibr">Van den Berg et al. 1998</xref>).</p><sec sec-type="materials\|methods"><title>Materials and Methods</title><sec sec-type="methods"><title>Study population</title><p>Study participants were part of an ongoing cohort study of PCBs and child development. Mother–infant pairs were recruited just after birth at St. Luke’s Hospital in New Bedford, Massachusetts, between March 1993 and December 1998. Dredging of PCB-contaminated New Bedford Harbor sediments occurred in the middle of the study recruitment period (April 1994 to September 1995). Participation was limited to consenting mothers (18 or more years of age) who had resided in one of the four towns (New Bedford, Acushnet, Fairhaven, Dartmouth) bordering New Bedford Harbor for the duration of their pregnancy. Infants born by cesarean section were excluded. Infants who required high-grade neonatal care or were otherwise not available for a newborn examination were not included in the study.</p><p>Of the 788 participants in the birth cohort study, 37 did not have a cord serum sample (not collected at birth or lost during laboratory sample preparation). We excluded one infant whose mother’s address was missing, two younger twins, and 28 younger siblings, leaving a total of 720 infants available for analysis.</p><p>The human subjects committees of Harvard School of Public Health, Brigham and Women’s Hospital of Boston, and St. Luke’s Hospital of New Bedford, Massachusetts, approved the protocol of this study. Data were collected after written informed consents were obtained from study mothers.</p></sec><sec><title>Cord blood PCB levels</title><p>Cord blood samples were obtained at birth in Vacutainer tubes and centrifuged, and the serum fraction was removed. The serum was stored in solvent-rinsed glass vials with Teflon-lined caps at –20°C until extraction. Analyses were performed by the Harvard School of Public Health Organic Chemistry Laboratory. Cord blood analytic methods and quality control procedures are described elsewhere (<xref rid="b19-ehp0114-001092" ref-type="bibr">Korrick et al. 2000</xref>). Briefly, 51 individual PCB congeners were measured using liquid–liquid extraction and extract analysis by capillary column gas chromatography with electron capture detection. Confirmatory analyses were done with microelectron capture detection and a capillary column of different polarity. Serum lipids were not measured because of insufficient sample volume. PCB concentrations were reported as the sum of 51 congeners (∑PCB) in units of nanograms of analyte per gram of serum. We also grouped PCB levels into light PCBs (sum of 14 mono- to tetrachlorinated biphenyls) and heavy PCBs (sum of 37 pentato decachlorinated biphenyls) according to their elution order and relative volatility (<xref rid="b5-ehp0114-001092" ref-type="bibr">Cullen et al. 1996</xref>). These two groups were chosen <italic>a priori</italic> based on the hypothesis that PCB exposure pathways may vary by their relative volatility.</p><p>The 51 congeners were chosen based on their toxicity, persistence in the environment or human samples, and presence in New Bedford environmental samples; these included a subset of mono-<italic>ortho</italic> dioxin-like PCBs (congeners 105, 118, 156, 167, and 189). The dioxin TEQ concentration for the dioxin-like PCBs was calculated (<xref rid="b37-ehp0114-001092" ref-type="bibr">Van den Berg et al. 1998</xref>) and expressed in parts per trillion (ppt) lipid, assuming 0.17% lipid for cord serum based on our laboratory’s data and published values (Altshul LM, personal communication; <xref rid="b7-ehp0114-001092" ref-type="bibr">Denkins et al. 2000</xref>). PCB-118 was chosen a priori for individual assessment. It was prevalent in harbor sediments consistent with the predominant Aroclors used by the area’s industries (<xref rid="b3-ehp0114-001092" ref-type="bibr">Brown and Wagner 1990</xref>; <xref rid="b41-ehp0114-001092" ref-type="bibr">Weaver 1984</xref>). In addition, it was disproportionately prevalent in our serum samples; cord serum levels of PCB-118 were comparable with levels observed in other population-based surveys (<xref rid="b19-ehp0114-001092" ref-type="bibr">Korrick et al. 2000</xref>) despite overall PCB levels being substantially lower than most other populations (<xref rid="b27-ehp0114-001092" ref-type="bibr">Longnecker et al. 2003</xref>).</p></sec><sec><title>Dietary assessment</title><p>Mothers completed a semiquantitative food-frequency questionnaire during a home evaluation of the child at age 2 weeks. The mothers reported diet histories before and during pregnancy. Twenty-four items from the food-frequency questionnaire were collapsed into six groups: meat (including organ meat), poultry, dairy, eggs, grains, and fish. We further considered fish in four subcategories: tuna; dark-meat fish (mackerel, blue fish, salmon, sardines, and swordfish); other fish (including catfish), and shellfish. In addition, mother’s self-reported consumption of locally grown produce, dairy products (including eggs), meat (including chicken), fish, game, and wine were determined as binary (yes/no) variables.</p></sec><sec><title>Occupation, gardening, and other potentially PCB-exposure–related activities</title><p>Mothers’ potential occupational PCB exposure (including working with paints, sealants, caulking compounds, and lubricants), gardening, and other potentially PCB-exposure–related activities (including use of pesticides and fertilizers) were determined by interviewer-administered questionnaire at the 2-week evaluation. Total person-years of exposure were calculated separately for occupation, gardening, and other. Potential exposures were reported as the sum of years from self-reports of engagement in these activities for at least 1 day per week. For each exposure pathway, we divided person-years of potential exposure into three categories: zero and below and above the 75th percentile of nonzero values.</p></sec><sec><title>Other risk factors</title><p>We determined maternal age, birthplace, race, education, marital status, reproductive history, pregnancy smoking and alcohol consumption, residential history, household income, and infant’s race and sex from the 2-week questionnaire and maternal and infant medical records.</p></sec><sec><title>Geographic information systems</title><p>Home address for the duration of the mother’s pregnancy was geocoded by Mapping Analytics (Rochester, NY), a commercial geocoding firm previously shown to have good (96%) accuracy (<xref rid="b21-ehp0114-001092" ref-type="bibr">Krieger et al. 2001</xref>). We used the geocoded residence location for mapping, calculating distance from the Superfund site, and retrieving Census block group data.</p><p>A map of New Bedford Harbor PCB levels (<xref rid="b36-ehp0114-001092" ref-type="bibr">U.S. EPA 2001</xref>) was aligned to the Massachusetts town boundaries (<xref rid="b28-ehp0114-001092" ref-type="bibr">MassGIS 2002</xref>; scale 1:25,000 meter units) using ArcGIS (ESRI Inc., Redlands, CA) to estimate the latitude and longitude of the harbor hot spot. Residential distance (in miles) from the hot spot was used as an index of potential site-related PCB exposure.</p><p>Indoor PCB sources include pre-1977 sealants, electrical appliances, and light fixtures (<xref rid="b2-ehp0114-001092" ref-type="bibr">Balfanz et al. 1993</xref>; <xref rid="b39-ehp0114-001092" ref-type="bibr">Vorhees et al. 1997</xref>). We did not have information about individual home characteristics, but as a proxy, we calculated the fraction of houses built between 1940 and 1979 compared with the total number of houses built through 1990, using 1990 Census block group data.</p><p>We constructed neighborhood socioeconomic indices based on 1990 Census block group data (<xref rid="b20-ehp0114-001092" ref-type="bibr">Krieger et al. 2003</xref>): <italic>a</italic>) crowding—percentage of households with more than one person per room; <italic>b</italic>) poverty—percentage of persons below the federally defined poverty line ($12,647 for a family of four in 1989); <italic>c</italic>) low income—percentage of households with income less than 60% of the U.S. median household income ($18,000); <italic>d</italic>) median household income; <italic>e</italic>) high education—percentage of persons, 25 or more years of age, with at least 4 years of college; and <italic>f</italic> ) low education—percentage of persons, 25 or more years of age, with less than a 12th grade education.</p></sec><sec sec-type="methods"><title>Statistical analysis</title><p>The cord serum PCB levels were highly positively skewed and were log<sub>10</sub> transformed for linear regression analyses. Univariate and bivariate associations were explored. Associations between log PCB levels and continuous covariates were assessed using scatter plot smoothing (<xref rid="b38-ehp0114-001092" ref-type="bibr">Venables and Ripley 1997</xref>) to examine any nonlinear relationships.</p><p>Potential exposure risk factors were divided into those associated with exposure pathways—dietary, inhalation, and dermal exposure sources—and those related to individual characteristics. A set of core individual characteristics was included in each exposure pathway analysis: maternal age and birthplace, smoking during pregnancy, previous lactation, child’s date of birth and sex, dredging period, and household income. Individual socioeconomic indicators (maternal education and race) were also included in models assessing neighborhood socioeconomic indicators and PCB levels. Multivariate models for log PCB included the core individual characteristics and exposure pathway covariates significant (<italic>p</italic> < 0.10) in at least one of the individual pathway models for at least one of the PCB measures (∑PCB, heavy PCBs, light PCBs and PCB-118). Regression results are reported as the relative (percent) increase in PCB level associated with each predictor, calculated as the antilog of the regression coefficient and 95% confidence intervals.</p><p>PCB levels were mapped and a smoothed surface was fitted by kriging (<xref rid="b4-ehp0114-001092" ref-type="bibr">Cressie 1991</xref>) using ArcGIS Geostatistical Analyst (ESRI Inc.). We estimated the surface by an inverse-distance weighted average of 25 neighboring points chosen on the basis of a small-prediction mean square error and a reasonable area to detect local spatial variability. We restricted this mapping to residences within a 5-mile radius of the hot spot. Similar mapping was performed for multivariate model residuals to provide information on any unmeasured spatial correlates of PCB exposure. To protect the confidentiality of participants, each residence location was offset by a random amount generated from a normal distribution with mean zero and standard deviation (SD) equal to 1% of the SD of residence latitudes and longitudes.</p><p>Generalized additive models (Hastie and Tibshirani 1990) were fit in S-Plus (version 3.4; Insightful Corp., Seattle, WA) to assess temporal variability in PCB levels. The span parameter with the lowest Akaike information criterion (AIC) for each PCB measure was chosen. Linear regression models were fit in SAS (version 8.2; SAS Institute Inc., Cary, NC).</p></sec></sec><sec sec-type="results"><title>Results</title><p>Cord serum PCB levels had geometric means (SDs) as follows: ∑PCB, 0.40 (2.02) ng/g with a range of 0.068–18.14 ng/g; heavy PCBs, 0.33 (2.09) ng/g with a range of 0.035–11.91 ng/g; light PCBs, 0.063 (2.12) ng/g with a range of 0.0074–6.23 ng/g; PCB-118, 0.035 (2.37) ng/g with a range of 0–2.05 ng/g; and dioxin-like PCB TEQs, 4.40 (2.39) ppt lipid, with a range of 0–151.5 ppt lipid.</p><p>Maternal, infant, and household characteristics are shown in <xref ref-type="table" rid="t1-ehp0114-001092">Table 1</xref>. Twenty percent of mothers were born outside of the United States (14% from Portugal, the Azores, or Cape Verde). Most (58%) had an educational level of high school or less, and 70% had an annual household income of < $40,000; 75% of the study population resided within 3.9 miles of the hot spot. Half of the infants were born after the harbor was dredged (October 1995 and later).</p><p>Maternal age at the infant’s birth was strongly associated with cord serum PCB levels, which declined over time, with additional declines after the harbor dredging was completed (<xref ref-type="table" rid="t1-ehp0114-001092">Table 1</xref>, <xref ref-type="fig" rid="f2-ehp0114-001092">Figure 2</xref>). Mothers who were born in Portugal, the Azores, or Cape Verde and female infants had significantly higher cord serum PCB levels (<xref ref-type="table" rid="t1-ehp0114-001092">Table 1</xref>). After adjustment for maternal age, prior lactation and higher household income were associated with lower cord serum PCB levels. Maternal smoking during pregnancy was also associated with lower PCB levels (<xref ref-type="table" rid="t1-ehp0114-001092">Table 1</xref>). These parameters were defined as core covariates and included in subsequent analyses. Maternal marital status, alcohol consumption during pregnancy, and infant race were not associated with serum PCB levels.</p><p>PCB associations with maternal diet before and during pregnancy were essentially the same. We report the results of analyses assessing diet during pregnancy. Maternal intake of organ meats (liver, tripe, kidney, bone marrow) was significantly associated with higher PCB levels (<italic>p</italic> < 0.05) for ∑PCB, light PCBs, and PCB-118 after adjustment for the base model covariates (<italic>p</italic> = 0.05 for heavy PCBs) (<xref ref-type="table" rid="t2-ehp0114-001092">Table 2</xref>). Consumption of local dairy products (including eggs) was associated with significantly higher levels of ∑PCB and heavy PCBs. Consumption of dark fish was positively associated with PCB levels, but this association was only marginally significant for light PCBs and PCB-118 (<xref ref-type="table" rid="t2-ehp0114-001092">Table 2</xref>).</p><p>Mothers who were long-term gardeners had infants with lower heavy PCB and PCB-118 levels than infants of mothers who did not garden; however, this association was based on a very small sample size (<italic>n</italic> = 9) and therefore was not included in our final multivariate model. Otherwise, we found no consistent association of cord serum PCB levels with PCB-related occupations or activities, distance of residence from the hot spot, or age of homes in the child’s neighborhood (<xref ref-type="table" rid="t3-ehp0114-001092">Table 3</xref>). Although there was a tendency for infants born to mothers living in poor or low-income neighborhoods to have higher light PCB levels than those born to mothers living in other neighborhoods, these associations were not significant (<xref ref-type="table" rid="t4-ehp0114-001092">Table 4</xref>).</p><p>We constructed multivariate models including core covariates and significant covariates from the pathway analyses (<xref ref-type="table" rid="t5-ehp0114-001092">Table 5</xref>). Maternal age and birthplace (in Portugal, the Azores, or Cape Verde) remained the strongest predictors of cord serum PCB levels (<italic>p</italic> < 0.001). In addition, infants born late in the study had significantly lower PCB levels than infants born early in the study (<xref ref-type="table" rid="t5-ehp0114-001092">Table 5</xref>). Even with adjustment for infant birth date, infants born after dredging had significantly lower light PCB and PCB-118 levels, with near significance for ∑PCB levels (<xref ref-type="table" rid="t5-ehp0114-001092">Table 5</xref>). Covariate-adjusted smoothed plots of ∑PCB, heavy PCB, light PCB, and PCB-118 levels by infant date of birth corroborate the apparent independent dredging effect (<xref ref-type="fig" rid="f2-ehp0114-001092">Figure 2</xref>). Mother’s prior lactation and smoking during pregnancy were significantly associated with lower PCB levels, and maternal consumption of organ meat and locally produced dairy were associated significantly with higher PCB levels (<xref ref-type="table" rid="t5-ehp0114-001092">Table 5</xref>).</p><p>Maps of unadjusted log<sub>10</sub>(∑PCB) levels (<xref ref-type="fig" rid="f1-ehp0114-001092">Figure 1A</xref>) and log<sub>10</sub>(∑PCB) residuals from the multivariate adjusted model (<xref ref-type="fig" rid="f1-ehp0114-001092">Figure 1B</xref>) showed spatial variability in PCB levels but no relationship to proximity of residence to the PCB hot spot. Similar results were found with heavy and light PCB levels.</p><p>Results of pathway analyses for dioxin-like PCB TEQs were similar to those of the four other PCB measures. Significant predictors of higher PCB TEQ concentrations included older maternal age; maternal birth in Portugal, the Azores, or Cape Verde; and consumption of red meat during pregnancy. Mother’s previous lactation, smoking during pregnancy, and infant birth at end of the study were associated with lower PCB TEQs.</p></sec><sec sec-type="discussion"><title>Discussion</title><p>We found no evidence that living closer to the New Bedford Harbor Superfund site was associated with increased cord serum PCB levels either in the crude unadjusted means or after adjusting for other risk factors for PCB exposure in the study population.</p><p>However, children born before or during dredging had consistently higher cord serum PCB levels than children born after dredging, even after we accounted for birth date (<xref ref-type="table" rid="t5-ehp0114-001092">Table 5</xref>, <xref ref-type="fig" rid="f2-ehp0114-001092">Figure 2</xref>), suggesting a possible effect of the PCB-contaminated site and its dredging on cord blood PCB levels. Serum levels of light PCBs were more strongly associated with dredging than were heavy PCBs (<xref ref-type="table" rid="t5-ehp0114-001092">Table 5</xref>). This finding suggests that differences in PCB volatility affect exposure risks potentially associated with the site. Furthermore, for PCB-118, a dioxin-like pentachlorinated biphenyl disproportionately prevalent in study samples, the dredging effect was more significant than the temporal decline, with near-constant concentrations before dredging, an increase during dredging, and a significant decline after dredging (<xref ref-type="fig" rid="f2-ehp0114-001092">Figure 2</xref>). Overall, these results support modest, transient increases in cord serum PCB levels during dredging, with significant declines in serum PCB levels observed after dredging, particularly for the more volatile PCBs and PCB-118 (<xref ref-type="table" rid="t5-ehp0114-001092">Table 5</xref>, <xref ref-type="fig" rid="f2-ehp0114-001092">Figure 2</xref>). The apparent differential effects of remediation on cord serum levels of various congeners are notable given possible congener-specific differences in toxicity.</p><p>In addition to the previously described dredging associations, maternal consumption of locally produced dairy products—an exposure risk factor potentially related to the contaminated site—was associated with higher cord serum PCB levels (<xref ref-type="table" rid="t2-ehp0114-001092">Tables 2</xref>, <xref ref-type="table" rid="t5-ehp0114-001092">5</xref>).</p><p>The most important predictor of elevated cord serum PCB levels was older maternal age at the birth of the study infant. Older age is a well-established risk factor for increased serum organochlorine concentrations, presumably as a consequence of cumulative exposure and temporal trends in exposure (<xref rid="b22-ehp0114-001092" ref-type="bibr">Kutz et al. 1991</xref>). In multivariate models, we found that older maternal age and earlier birth year were both associated with elevated cord blood PCB levels, indicating both cumulative exposure and temporal trend effects.</p><p>Mothers born in Portugal, the Azores, or Cape Verde had infants with substantially higher cord serum PCB levels than mothers born in the United States, Canada, or other countries, even after adjustment for diet or other lifestyle covariates that may vary by country of origin (<xref ref-type="table" rid="t5-ehp0114-001092">Table 5</xref>). Although this observed association may be a chance finding or consequent to residual confounding by diet or lifestyle, it is also consistent with potentially higher early-life exposure to PCBs resulting in higher serum levels during pregnancy. For example, PCB contamination is present in fish species from southern Europe and the Atlantic Ocean along the Azore Islands (<xref rid="b33-ehp0114-001092" ref-type="bibr">Stefanelli et al. 2004</xref>). Higher early-life exposure to PCBs has been associated with higher serum levels in adulthood among other populations (<xref rid="b32-ehp0114-001092" ref-type="bibr">Rylander et al. 1997</xref>).</p><p>We confirmed the previously reported association of prior lactation with lower serum PCB levels, which is likely due to PCB excretion in milk (<xref rid="b11-ehp0114-001092" ref-type="bibr">Fitzgerald et al. 1998</xref>; <xref rid="b18-ehp0114-001092" ref-type="bibr">Jensen 1991</xref>). Smoking during pregnancy was also associated with lower cord serum PCBs. Previous studies have been inconsistent regarding the association of PCBs with smoking; maternal smoking during pregnancy was associated with higher newborn PCB levels in one study (<xref rid="b23-ehp0114-001092" ref-type="bibr">Lackman et al. 2000</xref>) but not in others (<xref rid="b9-ehp0114-001092" ref-type="bibr">Fein et al. 1984</xref>; <xref rid="b31-ehp0114-001092" ref-type="bibr">Rogan et al. 1986</xref>). Smoking may decrease organochlorine concentrations by enhancing their metabolism via smoking-related induction of cytochrome P450 enzymes (<xref rid="b8-ehp0114-001092" ref-type="bibr">Deutch and Hansen 1999</xref>; <xref rid="b43-ehp0114-001092" ref-type="bibr">Zevin and Benowitz 1999</xref>). Long-term gardening was associated with lower cord serum PCBs (<xref ref-type="table" rid="t3-ehp0114-001092">Table 3</xref>), opposite to the hypothesized effect. However, the small number of long-term gardeners (<italic>n</italic> = 9) suggests that chance and/or confounding may explain this finding. Of note, the final multivariate model (<xref ref-type="table" rid="t5-ehp0114-001092">Table 5</xref>) was unchanged by the addition of gardening (data not shown).</p><p>In addition to these correlates of exposure, maternal consumption of organ meat and local dairy products (<xref ref-type="table" rid="t2-ehp0114-001092">Tables 2</xref>, <xref ref-type="table" rid="t5-ehp0114-001092">5</xref>) was associated with significantly higher cord blood PCB levels, but other potential dietary risks (including fish intake) were not. Although fish and animal products have been identified as important sources of general population exposure to PCBs and dioxins in some studies (<xref rid="b24-ehp0114-001092" ref-type="bibr">Laden et al. 1999</xref>; <xref rid="b29-ehp0114-001092" ref-type="bibr">Patandin et al. 1999</xref>), levels of PCBs and dioxins in fish and other foods have been declining (<xref rid="b15-ehp0114-001092" ref-type="bibr">Hays and Aylward 2003</xref>). Contaminated areas of the harbor were closed to fishing in 1979 (<xref rid="b1-ehp0114-001092" ref-type="bibr">Agency for Toxic Substances and Disease Registry 1995</xref>), 14 years before we started this study. A lack of association of local fish consumption with serum PCB levels is consistent with the lag between last likely intake of the most contaminated fish and our exposure assessment.</p><p>Other evaluated risk factors did not explain heterogeneity in cord serum PCB levels. Although serum concentrations of PCB-exposed workers are higher than those of the general population (<xref rid="b42-ehp0114-001092" ref-type="bibr">Wolff et al. 1982</xref>), the small number of mothers with potential occupational exposure limited the statistical power to detect such associations. Furthermore, the age of study mothers was such that most of their occupational (and other) activities occurred after the ban on PCBs.</p><p>Neighborhood socioeconomic status and age of housing were not associated with increased cord PCB levels. Although manufacturers incorporated PCBs in building materials and light fixtures during a well-defined time period (<xref rid="b2-ehp0114-001092" ref-type="bibr">Balfanz et al. 1993</xref>), house age was not a good predictor of indoor air PCB concentrations in previous studies in New Bedford (<xref rid="b39-ehp0114-001092" ref-type="bibr">Vorhees et al. 1997</xref>). This measure does not capture renovations or other potential indoor PCB sources such as electrical appliances or fluorescent lights. Moreover, the neighborhood distribution of home ages is an imperfect proxy for the age of the specific home of interest.</p><p>Correlates of cord serum PCBs did not vary much by the different congener groupings assessed. For example, the exposure pathways we observed for the heavy PCBs and ∑PCBs were quite similar (<xref ref-type="table" rid="t5-ehp0114-001092">Table 5</xref>). This is likely because the correlation of ∑PCB levels with heavy PCBs was much higher (<italic>r</italic> = 0.99) than with light PCBs (<italic>r</italic> = 0.76), consistent with the predominance of heavy PCBs in the sum. Except for diet, correlates of PCB TEQ exposure were also similar to other PCB concentration measures. Specifically, maternal consumption of red meat, but not organ meat, was associated with significantly higher PCB TEQs. <xref rid="b29-ehp0114-001092" ref-type="bibr">Patandin et al. (1999)</xref> also found meat to be a major contributor to dietary intake of PCB TEQs. Because the congeners are weighted by dioxin-like activity, these findings provide insights into the correlates of potential toxicity, about which very little is known.</p><p>There are several limitations in the interpretation of our findings. First, the median serum PCB level in our cohort was about one-quarter of the overall median in a recent review of 10 studies of PCBs and neurodevelopment (<xref rid="b27-ehp0114-001092" ref-type="bibr">Longnecker et al. 2003</xref>). Despite this limitation, our findings corroborate previously established correlates of serum PCB levels, including age and secular trends. In addition, the use of simplified proxies for some exposure pathways limited our ability to determine the relative contribution of various routes of exposure to cord serum PCB levels. In particular, it could be argued that residential distance from the site does not capture outdoor concentrations because it ignores prevailing winds. However, the maps of cord serum PCB levels and model residuals do not indicate any likely wind-related spatial patterns with this region’s prevailing wind direction from the south-southwest (<xref rid="b5-ehp0114-001092" ref-type="bibr">Cullen et al. 1996</xref>). In addition, it is possible that including household income and other demographic variables reduced our ability to characterize exposure pathways by overadjusting for these indirect correlates of exposure. However, sensitivity analyses demonstrated that this was not the case. For example, results of our pathway analyses were not substantially changed by removing income from the model. Lastly, the cross-sectional nature of this analysis limits the certainty of inferences regarding the observed temporal-and dredging-associated differences in serum PCB levels.</p><p>In conclusion, our findings among New Bedford area infants suggest that maternal residence near a Superfund site per se does not lead to higher cord serum PCB levels independent of other exposure risk factors, such as maternal age, birthplace, diet, previous lactation, pregnancy smoking, and infant date of birth. However, there was evidence of an important local impact on exposure risk as shown by increased cord serum PCB levels in association with maternal local dairy consumption and lower cord serum PCB levels after site dredging.</p></sec>
PCB-Related Alteration of Thyroid Hormones and Thyroid Hormone Receptor Gene Expression in Free-Ranging Harbor Seals (<italic>Phoca vitulina</italic>)	<p>Persistent organic pollutants are environmental contaminants that, because of their lipophilic properties and long half-lives, bioaccumulate within aquatic food webs and often reach high concentrations in marine mammals, such as harbor seals (<italic>Phoca vitulina</italic>). Exposure to these contaminants has been associated with developmental abnormalities, immunotoxicity, and reproductive impairment in marine mammals and other high-trophic-level wildlife, mediated via a disruption of endocrine processes. The highly conserved thyroid hormones (THs) represent one vulnerable endocrine end point that is critical for metabolism, growth, and development in vertebrates. We characterized the relationship between contaminants and specific TH receptor (<italic>TR</italic> ) gene expression in skin/blubber biopsy samples, as well as serum THs, from free-ranging harbor seal pups (<italic>n</italic> = 39) in British Columbia, Canada, and Washington State, USA. We observed a contaminant-related increase in blubber <italic>TR-</italic>α gene expression [total polychlorinated biphenyls (∑PCBs); <italic>r</italic> = 0.679; <italic>p</italic> < 0.001] and a concomitant decrease in circulating total thyroxine concentrations (∑PCBs; <italic>r</italic> = −0.711; <italic>p</italic> < 0.001). Consistent with results observed in carefully controlled laboratory and captive feeding studies, our findings suggest that the TH system in harbor seals is highly sensitive to disruption by environmental contaminants. Such a disruption not only may lead to adverse effects on growth and development but also could have important ramifications for lipid metabolism and energetics in marine mammals.</p>	<contrib contrib-type="author"><name><surname>Tabuchi</surname><given-names>Maki</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001024">1</xref><xref ref-type="aff" rid="af2-ehp0114-001024">2</xref></contrib><contrib contrib-type="author"><name><surname>Veldhoen</surname><given-names>Nik</given-names></name><xref ref-type="aff" rid="af2-ehp0114-001024">2</xref></contrib><contrib contrib-type="author"><name><surname>Dangerfield</surname><given-names>Neil</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001024">1</xref></contrib><contrib contrib-type="author"><name><surname>Jeffries</surname><given-names>Steven</given-names></name><xref ref-type="aff" rid="af3-ehp0114-001024">3</xref></contrib><contrib contrib-type="author"><name><surname>Helbing</surname><given-names>Caren C.</given-names></name><xref ref-type="aff" rid="af2-ehp0114-001024">2</xref></contrib><contrib contrib-type="author"><name><surname>Ross</surname><given-names>Peter S.</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001024">1</xref></contrib>	Environmental Health Perspectives	<p>A wide range of chemicals produced either directly or indirectly as a result of human activities have contributed to the contamination of aquatic food chains around the world. Marine mammals occupying high trophic levels in aquatic food webs are often contaminated with relatively high concentrations of persistent organic pollutants (POPs), including polychlorinated biphenyls (PCBs), polychlorinated dibenzo-<italic>p</italic>-dioxins (PCDDs), polychlorinated dibenzofurans (PCDFs), and polybrominated diphenylethers (PBDEs). This is because of food-web–related biomagnification, the extent of which reflects the persistence of the chemical coupled with the long lifespan and limited detoxification capacity of marine mammals (<xref rid="b39-ehp0114-001024" ref-type="bibr">Ross et al. 2000</xref>; <xref rid="b50-ehp0114-001024" ref-type="bibr">Tanabe et al. 1988</xref>). Many studies have shown that exposure to these complex mixtures of POPs can lead to developmental abnormalities, reproductive impairment, endocrine disruption, and immunosuppression in harbor seals (<xref rid="b4-ehp0114-001024" ref-type="bibr">Brouwer et al. 1989</xref>; <xref rid="b12-ehp0114-001024" ref-type="bibr">De Swart et al. 1994</xref>; <xref rid="b36-ehp0114-001024" ref-type="bibr">Reijnders 1986</xref>) and other marine mammals (<xref rid="b45-ehp0114-001024" ref-type="bibr">Skaare et al. 2001</xref>; <xref rid="b47-ehp0114-001024" ref-type="bibr">Sonne et al. 2004</xref>). Marine mammals may therefore serve as indicators of marine environmental contamination, something that is relevant to both human and ecologic risk assessments (<xref rid="b38-ehp0114-001024" ref-type="bibr">Ross 2000</xref>).</p><p>Harbor seals (<italic>Phoca vitulina</italic>) are non-migratory (adult home range of ~ 50 km<sup>2</sup>) and abundant around the coast of British Columbia, Canada, and Washington State, USA (<xref rid="b9-ehp0114-001024" ref-type="bibr">Cottrell et al. 2002</xref>). The biology and physiology of this pinniped have been well documented, reflecting its wide distribution in temperate waters around the world and its relative ease of study. In British Columbia and Washington State, the harbor seal has been used to identify regional POP hotspots (<xref rid="b40-ehp0114-001024" ref-type="bibr">Ross et al. 2004</xref>) and as a model marine mammal species for evaluating the relationship between contaminants and health effects (<xref rid="b29-ehp0114-001024" ref-type="bibr">Levin et al. 2005</xref>; <xref rid="b43-ehp0114-001024" ref-type="bibr">Simms et al. 2000</xref>).</p><p>Although the concept of endocrine disruption in wildlife has garnered much international scientific attention (<xref rid="b7-ehp0114-001024" ref-type="bibr">Colborn et al. 1997</xref>), contaminant-related alteration of thyroid hormones (THs) and related processes may adversely affect vertebrates. The THs thyroxine (T<sub>4</sub>) and triiodothyronine (T<sub>3</sub>) play a crucial role in developmental processes and in the regulation of metabolism in adults. THs are produced in the thyroid gland, mainly in the form of T<sub>4</sub>, and are subsequently converted to the more bioactive T<sub>3</sub> form in peripheral (target) tissues through the action of deiodinases. Laboratory-based studies have indicated that some POPs and their metabolites interfere with TH physiology at multiple levels, including hormone synthesis, circulatory transport of TH, and TH metabolism in the liver and brain (<xref rid="b3-ehp0114-001024" ref-type="bibr">Brouwer et al. 1998</xref>; <xref rid="b28-ehp0114-001024" ref-type="bibr">Legler and Brouwer 2003</xref>). Reductions in circulating TH levels have been observed with increasing exposure to PCBs and related compounds in laboratory animals, aquatic birds, and both captive and free-ranging marine mammals, highlighting the sensitivity of this endocrine end point to disruption by environmental contaminants (<xref rid="b37-ehp0114-001024" ref-type="bibr">Rolland 2000</xref>).</p><p>In addition to effects on circulating THs, recent <italic>in vitro</italic> and laboratory animal evidence suggests that PCB and PCDD exposure can affect TH receptor (TR) activity and TH-responsive gene expression (<xref rid="b56-ehp0114-001024" ref-type="bibr">Zoeller 2005</xref>). THs (primarily through the biologically active form of T<sub>3</sub>) function as signaling molecules that interact with two nuclear receptors, TR-α and TR-β [whose genes are designated as <italic>THRA</italic> and <italic>THRB</italic>, respectively, in <xref rid="b17-ehp0114-001024" ref-type="bibr">GenBank (2006)</xref>], and alter their transcription activation and repression activities (<xref rid="b55-ehp0114-001024" ref-type="bibr">Yen 2001</xref>). THs are therefore critical to the regulation of the gene expression machinery required during different life stages of an animal.</p><p>Although circulating TH levels are often used as biomarkers of contaminant exposure in wildlife (<xref rid="b6-ehp0114-001024" ref-type="bibr">Chiba et al. 2001</xref>; <xref rid="b19-ehp0114-001024" ref-type="bibr">Hall et al. 1998</xref>; <xref rid="b24-ehp0114-001024" ref-type="bibr">Jenssen et al. 1995</xref>), gene expression end points that exploit the cellular functions of TH could provide a more sensitive and mechanistically based means to characterize the thyroid-toxic potential of complex contaminant mixtures in the real world. Such gene expression analysis might also form the basis of an early detection approach for POP exposure before the manifestation of higher-level health effects, such as developmental abnormalities and neurotoxicity, especially when results are consistent with laboratory-based observations.</p><p>Obtaining liver or blood from free-ranging marine mammals is generally fraught with logistical and ethical challenges. Skin/blubber biopsies have been used to generate useful information on contaminant concentrations and, more recently, on toxicologically relevant endocrine end points (<xref rid="b15-ehp0114-001024" ref-type="bibr">Fossi et al. 2003</xref>; <xref rid="b31-ehp0114-001024" ref-type="bibr">Miller et al. 2005</xref>; <xref rid="b32-ehp0114-001024" ref-type="bibr">Mos and Ross 2002</xref>). Gene expression analysis using small biopsies has the potential to become a useful, sensitive, and minimally invasive biomarker of contaminant exposure in seals and other wildlife.</p><p>The objectives of this study were <italic>a</italic>) to develop <italic>TR</italic> gene expression biomarkers using skin/blubber biopsies, <italic>b</italic>) to confirm the utility of using circulating THs as biomarkers of POP exposure, and <italic>c</italic>) to assess the feasibility of using TH-related gene expression biomarkers in free-ranging harbor seals.</p><sec sec-type="materials\|methods"><title>Materials and Methods</title><sec><title>Sample collection</title><p>A total of 39 healthy, young harbor seals (<italic>Phoca vitulina</italic>) of comparable body weight and condition were live-captured from five locations in southern British Columbia and northern Washington State during the summer of 2003 (<xref ref-type="fig" rid="f1-ehp0114-001024">Figure 1</xref>). These locations included three Canadian sites in Queen Charlotte Strait (QCS) (northeastern Vancouver Island, <italic>n</italic> = 10) and the Strait of Georgia (City of Vancouver, <italic>n</italic> = 8; Hornby Island, <italic>n</italic> = 7), and two U.S. sites in Juan de Fuca Strait (Smith Island, <italic>n</italic> = 7) and Puget Sound (Gertrude Island, <italic>n</italic> = 7). Both the accumulation of POPs and biologic end points in marine mammals are age dependent (<xref rid="b9-ehp0114-001024" ref-type="bibr">Cottrell et al. 2002</xref>), so we restricted our sampling to pups ranging in age from 3.5 to 5 weeks. Seals hauled out on sandy beaches were captured using a rapidly deployed seine net (<xref rid="b23-ehp0114-001024" ref-type="bibr">Jeffries et al. 1993</xref>), whereas those hauled out on rocky inlets were captured one at a time using a salmon-landing net (<xref rid="b44-ehp0114-001024" ref-type="bibr">Simms and Ross 2000</xref>). Seals were kept in hoop nets until sampling and then removed from the net and manually restrained for tissue and blood collection. Seals were typically captured at low tides (peak haul-out times), with time of capture during the day ranging from 0825 hr to 1540 hr across all sites.</p><p>Blood samples were taken from the extradural vein using a Vacutainer blood collection system with an 18-gauge needle and serum collection tube (Becton-Dickenson, Franklin Lakes, NJ, USA). All collected blood samples were stored at 4°C in the field. Blood samples were centrifuged within 5 hr after collection at 400 × <italic>g</italic> for 20 min. Serum was aspirated and stored in 1 mL aliquots in cryovials either on dry ice (−80°C) or in liquid nitrogen (−196°C) during transport, and in a freezer (−80°C) until analysis of TH concentrations was performed.</p><p>Skin/blubber biopsy samples were taken from an area 20 cm lateral to the spinal column and anterior to the pelvis. The area was shaved first with an electric shaver (Sculptor with type-50 blades; Oster, Niles, IL, USA) and cleaned using Betadine (Purdue Frederick, Pickering, Ontario, Canada) followed by 95% isopropyl alcohol. Two biopsy samples were collected: one 3.5 mm in diameter and the other 8 mm, with each approximately 2–3 cm in depth (Acuderm, Ft. Lauderdale, FL, USA). After sample collection, the biopsy area on the animal was disinfected using Betadine and Aquaphor (Beiersdorf, Wilton, CT, USA) iodine ointment. The 8-mm-diameter biopsy samples were wrapped in hexane-rinsed aluminum foil, placed in 2 mL cryovials, and stored immediately in liquid nitrogen in the field. The 3.5-mm-diameter biopsy samples were placed into 1.0 mL of the RNA stabilization solution RNAlater (Ambion, Houston, TX, USA) in RNase-free 1.5 mL cryovials and stored on wet ice in the field. Blubber samples frozen in liquid nitrogen were subsequently transferred to −80°C storage in the laboratory, and biopsy samples in RNAlater were stored at −20°C.</p><p>Animals were subsequently weighed, sexed, measured for length and axillary girth, assessed for general body condition, and then released. Captive time was approximately 15–20 min for captures using the landing net and less than 60 min for captures using the seine net method. All procedures were carried out under the auspices of the respective animal care committees and scientific research permits for researchers in British Columbia [Fisheries and Oceans Canada Animal Care Committee using guidelines from the Canadian Council on Animal Care (<xref rid="b5-ehp0114-001024" ref-type="bibr">CCAC 1997</xref>); Scientific Research Permit] and in Washington State under U.S. Marine Mammal Protection Act Scientific Research Permit No. 835 [<xref rid="b34-ehp0114-001024" ref-type="bibr">National Oceanic and Atmospheric Administration (NOAA) 1997</xref>] issued to the Washington Department of Fish and Wildlife by National Marine Fisheries Service.</p></sec><sec><title>Tissue homogenization</title><p>Because a possible stratification within blubber biopsies could influence our results, we evaluated the steady-state levels of the normalizer gene ribosomal protein L8 (<italic>L8</italic>) in skin and upper and lower blubber sections collected from all animals (<xref ref-type="fig" rid="f2-ehp0114-001024">Figure 2A</xref>). For this, each 3.5-mm-diameter tissue biopsy was separated into skin (~ 2 mm) and blubber sections using a razor blade before homogenization. Blubber samples were further divided into lower (close to the muscle) and upper (close to the skin) sections of 4 mm in depth.</p><p>All blubber samples were homogenized in TRIzol reagent (Invitrogen Canada Inc., Toronto, Ontario, Canada) using a Retsch MM301 mixer mill as described by <xref rid="b52-ehp0114-001024" ref-type="bibr">Veldhoen and Helbing (2001)</xref> and with the following modifications. Each blubber tissue sample was homogenized in a 1.5 mL microcentrifuge tube with the addition of 400 μL TRIzol and a 3-mm-diameter tungsten-carbide bead. For any given sample, an additional 3–6 min of mixing was performed if unhomogenized tissue remained after the initial 6 min homogenization period. Because of the presence of a substantial amount of connective tissue, the mixer mill procedure was incapable of efficient homogenization of the skin samples. These samples were homogenized using a PowerGen 125 tissue homogenizer (Fisher Scientific, Pittsburgh, PA, USA). Skin samples were minced with a razor blade and placed into a 2.0 mL microcentrifuge tube (Mic Rew Simport Plastics Ltd., CA, USA) containing 400 μL TRIzol. The shearing head was placed directly into each sample tube and gradually ramped from 8,000 rpm to approximately 30,000 rpm for a total of 3 min with 10-sec cooling period intervals every 15 sec of homogenization. To minimize heat production in the skin samples, tubes were kept on wet ice during the entire homogenization procedure.</p></sec><sec><title>Isolation of total RNA and preparation of cDNA</title><p>Total RNA was isolated from the tissue homogenates in TRIzol reagent as described by the manufacturer. After phase separation, 1 μL glycogen (Roche Diagnostics, Laval, QC, Canada) was added to each retained aqueous phase, and RNA was precipitated with the addition of isopropanol and a 1 hr incubation at −20°C. Total RNA was resuspended in diethyl pyrocarbonate–treated distilled, deionized H<sub>2</sub>O (20 μL for blubber and 40 μL for skin samples) and stored at −70°C.</p><p>Total cDNA was produced using Superscript II RNase H<sup>−</sup> reverse transcriptase as described by the manufacturer (Invitrogen Canada Inc.). One microgram of total RNA from each sample was annealed with 500 ng random hexamer oligonucleotide (Amersham Biosciences Inc., Baie D’urfe, Quebec, Canada) at 65°C for 10 min and placed on wet ice. The assembled 20 μL cDNA synthesis reactions were incubated at 42°C for 2 hr and diluted 20-fold before real-time quantitative polymerase chain reaction (QPCR) analysis.</p></sec><sec><title>Cloning of TR cDNA sequences</title><p>Target cDNA sequences representative of the gene transcripts <italic>TR-</italic>α and <italic>TR-</italic>β as well as our control gene, <italic>L8</italic>, were amplified using primers designed using Primer Premier software (version 4.1; Premier Biosoft International, Palo Alto, CA, USA) and synthesized by AlphaDNA (Montreal, Quebec, Canada) (<xref ref-type="table" rid="t1-ehp0114-001024">Table 1</xref>). Each 25 μL DNA amplification reaction included 2 μL of 20-fold total cDNA template, 20 pmol of each primer, 200 μM equimolar dNTPs (dATP, dCTP, dGTP, and dTTP), and 2.5 units of Taq DNA polymerase (Invitrogen Canada Inc.). DNA amplification was performed on a Gene Amp PCR System 9700 (PerkinElmer Biosystems, Foster City, CA, USA) using the following thermocycle conditions: denaturation at 95°C (5 min); 35 cycles of 94°C (1 min), 53°C (1 min), and 72°C (2 min); and an elongation step at 72°C (7 min). DNA products were separated on a 1.5% agarose gel and visualized with ethidium bromide staining on a ChemiImager 4000 (Alpha Innotech Corp., San Leandro, CA, USA). DNA bands representing <italic>TR-</italic>α [631 base pairs (bp)], <italic>TR-</italic>β (801 bp), and <italic>L8</italic> (602 bp and 126 bp) were excised and isolated by three repeated 5-min freeze/thaw cycles followed by a 10 min centrifugation at 12,000 × <italic>g</italic> (<xref rid="b46-ehp0114-001024" ref-type="bibr">Smith 1980</xref>). Isolated cDNA products (4 μL) were cloned into pCR2.1-TOPO vector using the TOPO TA Cloning Kit (Invitrogen Canada Inc.). Plasmid DNA was purified from selected transformants using the QIAprep Spin Miniprep Kit (Qiagen, Mississauga, Ontario, Canada), and the presence of insert sequence was confirmed by restriction analysis using Eco<italic>RI</italic> (Amersham Biosciences). The identity of each cloned cDNA was determined by DNA sequencing followed by BLASTn analysis (<xref rid="b33-ehp0114-001024" ref-type="bibr">National Center for Biotechnology Information 2006</xref>). Primer sequences are shown in <xref ref-type="table" rid="t1-ehp0114-001024">Table 1</xref>, and cloned sequences have been submitted to <xref rid="b17-ehp0114-001024" ref-type="bibr">GenBank (2006)</xref>.</p></sec><sec><title>QPCR assay</title><p>Primers specific for seal <italic>TR-</italic>α, <italic>TR-</italic>β, and <italic>L8</italic> were designed for the reverse-transcription QPCR assay (<xref ref-type="table" rid="t1-ehp0114-001024">Table 1</xref>). Quantitative DNA amplification reactions (15 μL) were performed on a MX4000 system (Stratagene, La Jolla, CA, USA) as described previously (<xref rid="b10-ehp0114-001024" ref-type="bibr">Crump et al. 2002</xref>). Each sample was prepared in quadruplicate, and the derived copy number values were averaged. The copy number for each gene transcript was determined from standard curves generated from the cloned plasmids in the previous section. <italic>TR-</italic>α and <italic>TR-</italic>β expression values were normalized to those of the expression of the <italic>L8</italic> internal control. The expression of this gene has been found to be invariant in many tissue types during developmentally associated changes in endogenous TH concentrations in reptiles (<xref rid="b26-ehp0114-001024" ref-type="bibr">Katsu et al. 2004</xref>) and amphibians (<xref rid="b42-ehp0114-001024" ref-type="bibr">Shi and Liang 1994</xref>).</p></sec><sec><title>TH assay</title><p>The concentrations of total T<sub>4</sub> (TT<sub>4</sub>), free T<sub>4</sub> (FT<sub>4</sub>), total T<sub>3</sub> (TT<sub>3</sub>), and free T<sub>3</sub> (FT<sub>3</sub>) were measured in animals from all five locations (<italic>n</italic> = 39) using related EIAgen enzyme-linked immunosorbent assay (ELISA) kits and by following the manufacturer’s recommended protocol (Adaltus, Montréal, Quebec, Canada). Frozen (undiluted) serum samples were thawed on wet ice, and all four TH measurements were obtained within 6 hr. For each ELISA assay, reactions were prepared in triplicate, and the signal intensity of seal serum samples and TH standards was measured at 450 nm on an MRX microplate reader (Dynatech Laboratories Inc., Chantilly, VA, USA). The sample data were subsequently averaged and compared with the standard curve in order to obtain representative TH concentration values.</p><p>Interassay variation was evaluated in two ways. First, we regularly included a pooled seal serum sample as a reference, and results were accepted for any given assay only when reference results were ± 20% of expected values. Second, total hormone measurements (TT<sub>3</sub> and TT<sub>4</sub>) were validated using the manufacturer’s reference standard (Thyroid Calver reagent; Casco Neal, Portland, ME, USA), and results were accepted for an assay only when concentrations were within ± 5% of expected values.</p><p>No purified harbor seal THs are commercially available. With this in mind, we validated the TH assays for harbor seals by conducting analyses of serial dilutions within a fixed sample volume and using incremental spikes of seal serum with Thyroid Calver reagent. Responses of serial dilutions of seal serum and standard additions of seal serum with the reference standard both produced linear results (data not shown).</p></sec><sec><title>Measurement of POP concentrations in blubber tissue</title><p>Each frozen (−80°C) 8 mm tissue biopsy was cut vertically, and the upper skin layer (~ 2 mm) removed. A portion of each blubber sample (100–300 mg wet weight) was used in the analysis for all PCB congeners and for specific PCDD and PCDF congeners using high-resolution gas chromatography and high-resolution mass spectrometry analysis at the Fisheries and Oceans Canada Regional Contaminant Laboratory (Institute of Ocean Sciences, Sidney, British Columbia, Canada). Details of the chromatography and mass spectrometry conditions, the criteria used for chemical identification and quantification, and the quality assurance and quality control practices have been previously described (<xref rid="b21-ehp0114-001024" ref-type="bibr">Ikonomou et al. 2001</xref>).</p><p>Although 154 PCB peaks were quantified (out of 209 theoretical congeners), many congeners were not detectable in all of the samples. ∑PCB concentration was therefore calculated using the following rules. If a congener was detected in > 70% of the sample population, the minimum detection limit substitutions were made. Where congeners were detected in < 70% of samples, the minimum detection limit was set at zero. Sample lipid values were also measured, and the concentration of POPs was expressed on a lipid weight (lw) basis. Total toxic equivalents (∑TEQs) to 2,3,7,8-tetrachlorodibenzo-<italic>p</italic>-dioxin were calculated for all dioxin-like PCBs (12 congeners), PCDD (7 congeners), and PCDF (10 congeners) using the most recently reported international mammalian toxic equivalency factors (<xref rid="b51-ehp0114-001024" ref-type="bibr">Van den Berg et al. 1998</xref>).</p></sec><sec><title>Statistical analyses</title><p>All statistical analyses were performed using SPSS software (version 12; SPSS Inc., Chicago, IL, USA). Results were evaluated by site and among all individuals. For the former, harbor seal pups from five sites were compared for circulating TH concentrations and steady-state mRNA expression levels in skin and blubber biopsy samples. Seals from the remote QCS, previously (and in this study) shown to be relatively uncontaminated (<xref rid="b40-ehp0114-001024" ref-type="bibr">Ross et al. 2004</xref>), were used as a reference group. For each group, values were examined for normality with the Shapiro-Wilk test and for homogeneity of variances using Levine’s test. Groups that were normal with equal variance were evaluated using one-way analysis of variance (ANOVA) to assess intergroup differences followed by Tukey’s honest significant difference (HSD) test. If the data were not normally distributed, a Kruskal-Wallis test was used followed by a Mann-Whitney <italic>U</italic> test for pairwise comparison of groups. Significance was defined as <italic>p</italic> < 0.05. Extreme outliers, defined as values more than three times the interquartile range, were removed.</p><p>For the among-individual assessment of the entire study group, correlation analysis was carried out using the Pearson method for normally distributed data or the nonparametric Kendall’s tau-b method when data were not normally distributed. Given our concern that body weight (~ age) of the harbor seal pups might influence either contaminant concentrations or thyroid end points, we conducted regressions between body weight and contaminant concentrations, TH concentrations, and TR levels. If body weight exhibited a significant relationship with contaminant or thyroid measurements, we conducted multiple regression analysis to identify the relative contribution of each variable.</p></sec></sec><sec sec-type="results"><title>Results</title><sec><title>Sample collection</title><p>Of the 39 harbor seals sampled, availability of adequate tissue quality and cost considerations for contaminant analyses resulted in the analysis of 39 serum samples for TH measurement, 35 biopsies (3.5 mm) for gene expression analysis, and 24 blubber biopsies (8 mm) for contaminant analysis.</p><p>The mean ± SE body weight was 20.6 ± 0.52 kg (range, 14.1–27.0 kg). Our ANOVA results revealed a significant difference among sites. A subsequent Tukey’s HSD test indicated that only Smith Island seals differed, being slightly heavier than QCS seals (<italic>p</italic> = 0.038).</p></sec><sec><title>TR <italic>gene sequences in the harbor seal</italic></title><p>Partial cDNA sequences were isolated from biopsied harbor seal blubber that represented expressed mRNA of <italic>TR-</italic>α and <italic>TR-</italic>β genes, as well as our control gene, <italic>L8</italic>. Both <italic>TR</italic> sequences obtained are predicted to encompass approximately half of the estimated open reading frame region within the mRNA transcripts and include sequence located between the encoded DNA-binding and ligand-binding domains. A comparison of harbor seal <italic>TR</italic> sequences with six other species using the ClustalW alignment program (<xref rid="b13-ehp0114-001024" ref-type="bibr">European Bioinformatics Institute 2006</xref>) indicated that the mRNA sequence for <italic>TR-</italic>α and <italic>TR-</italic>β are highly conserved among these vertebrates (<xref ref-type="table" rid="t2-ehp0114-001024">Table 2</xref>). This is particularly evident within mammals, where the putative protein sequences of harbor seal <italic>TR-</italic>α and <italic>TR-</italic>β display > 99% amino acid identity.</p></sec><sec><title><italic>Measurement of</italic> TR <italic>expression in skin/blubber biopsies</italic></title><p>Based on the cDNA sequence obtained, oligonucleotide primers were developed for QPCR analysis of specific gene expression biomarkers. Significant variation in L8 mRNA expression (<italic>p</italic> < 0.05, Tukey’s HSD or Mann-Whitney <italic>U</italic>-test) within the vertical plane of the biopsy tissue sample was observed for all sample sections, with the exception of QCS and Gertrude Island (<xref ref-type="fig" rid="f2-ehp0114-001024">Figure 2B</xref>). We then compared each section (skin, upper blubber, and lower blubber) separately across the animals from different locations. Both the skin and the upper blubber (adjacent to skin) sections showed a significant difference in L8 mRNA expression among sites (skin: <italic>p</italic> < 0.001, Kruskal-Wallis; upper blubber: <italic>p</italic> = 0.028, Kruskal-Wallis). However, <italic>L8</italic> steady-state transcript levels in the lower blubber region did not differ among sites (<italic>p</italic> = 0.05, ANOVA; <italic>p</italic> > 0.05, Tukey). Therefore, we chose the amount of <italic>L8</italic> transcript in the lower blubber as a suitable normalizer gene for the comparison of gene expression levels between the different seal populations. All subsequent QPCR analyses of <italic>TR</italic> transcript copy numbers are presented for lower blubber sections only.</p><p>TR-α mRNA abundance was found to be significantly higher than that of TR-β (<italic>p</italic> = 0.004, Tukey) in all the individuals examined (<xref ref-type="table" rid="t3-ehp0114-001024">Table 3</xref>). In addition, the relationship between TR-α and TR-β mRNA expression patterns was positively correlated (<italic>R</italic> = 0.651). In comparisons of animals from different geographic locations, Gertrude Island samples displayed a significant elevation in both <italic>TR-</italic>α (<italic>p</italic> < 0.001, Tukey) and <italic>TR-</italic>β (<italic>p</italic> = 0.011, Tukey) transcript levels compared with animals from the QCS reference site.</p></sec><sec><title>Serum TH levels</title><p>The concentrations of different TH forms were measured in serum collected from the seal pups by site (<xref ref-type="table" rid="t4-ehp0114-001024">Table 4</xref>). Among the seals sampled from different locations, Gertrude Island animals had significantly lower TT<sub>4</sub> and FT<sub>4</sub> compared with reference site QCS animals (<italic>p</italic> < 0.001, Tukey; <italic>p</italic> < 0.001, Mann-Whitney). A strong positive correlation was found between measured TT<sub>4</sub> and FT<sub>4</sub> levels (<italic>R</italic> = 0.844, <italic>p</italic> < 0.001) among all individuals, whereas no correlation existed between TT<sub>3</sub> and FT<sub>3</sub> serum concentrations (<italic>R</italic> = 0.260, <italic>p</italic> = 0.121). Negative correlations between circulating TT<sub>4</sub> and TR-α mRNA levels (<italic>R</italic> = −0.456, <italic>p</italic> < 0.05) and circulating FT<sub>4</sub> and <italic>TR-</italic>α expression (<italic>R</italic> = −0.481, <italic>p</italic> < 0.05) were detected. No correlation was observed between any serum TH measurement and <italic>TR-</italic>β transcript levels (data not shown).</p></sec><sec><title>POP concentrations in blubber</title><p>Of a total of 154 PCB congener peaks quantified, 135 peaks were detected in QCS seals, 142 peaks in Smith Island seals, and 153 peaks in Gertrude Island seals. Four of 24 seals were identified as extreme outliers in the TEQ calculations and were removed from further analysis. Seal pups located on Gertrude Island showed an approximate 10-fold higher ∑PCB concentration (6.2 ± 1.0 mg/kg, lw) compared with animals from the reference QCS (0.7 ± 0.1 mg/kg, lw; <italic>p</italic> < 0.001, Tukey) and 5-fold higher than animals from Smith Island (1.3 ± 0.2 mg/kg, lw; <italic>p</italic> < 0.001, Tukey). Calculated ∑TEQ values for PCBs, PCDDs, and PCDFs were also significantly higher in seal pups on Gertrude Island (70.4 ± 16.5 ng/kg) compared with animals from QCS (12.6 ± 2.2 ng/kg; <italic>p</italic> < 0.001, Mann-Whitney) but did not differ significantly from those from Smith Island (24.9 ± 11.6 ng/kg; <italic>p</italic> = 0.052, Mann-Whitney). PCBs were the major constituent measured among contaminant classes measured in study animals, which included PCBs, PCDDs, and PCDFs, and were also the dominant contributor to ∑TEQ (PCBs represented an average of 43.4% in QCS seals, 59.2% for Smith Island seals, and 90.1% for Gertrude Island seals). More details on contaminant levels and patterns in harbor seals from this region are available elsewhere (<xref rid="b40-ehp0114-001024" ref-type="bibr">Ross et al. 2004</xref>).</p></sec><sec><title><italic>Correlation of TH and</italic> TR <italic>end points with POP exposure</italic></title><p>∑PCB concentrations were negatively correlated with circulating TT<sub>4</sub> (<italic>R</italic> = −0.711, <italic>p</italic> < 0.001) (<xref ref-type="table" rid="t5-ehp0114-001024">Table 5</xref>, <xref ref-type="fig" rid="f3-ehp0114-001024">Figure 3</xref>) and FT<sub>4</sub> (<italic>R</italic> = −0.724, <italic>p</italic> < 0.001, <xref ref-type="table" rid="t5-ehp0114-001024">Table 5</xref>). In contrast, a positive correlation was observed between ∑PCB concentrations and the level of TR-α mRNA (<italic>R</italic> = 0.679, <italic>p</italic> < 0.01) (<xref ref-type="table" rid="t5-ehp0114-001024">Table 5</xref>, <xref ref-type="fig" rid="f4-ehp0114-001024">Figure 4</xref>). Similarly, negative correlations were also found between ∑TEQ and circulating TT<sub>4</sub> (<italic>R</italic> = −0.495, <italic>p</italic> < 0.01), and positive correlations with the level of <italic>TR-</italic>α expression in the blubber (<italic>R</italic> = 0.464, <italic>p</italic> < 0.01).</p><p>Although we limited our studies to seals of a similar body weight (~ age), the potential confounding influence of age on both PCB concentration and thyroid end points remained a concern. Subsequent regression analysis revealed that body weight did not correlate with TT<sub>3</sub>, <italic>TR-</italic>α, <italic>TR-</italic>β, ∑PCBs, or ∑TEQ (data not shown). However, there were negative correlations between body weight and TT<sub>4</sub>, FT<sub>3</sub>, and FT<sub>4</sub>. We found no correlation between FT<sub>3</sub> and PCBs, so we did not further evaluate this relationship. Multiple regression analysis showed that, although both PCB concentrations and body weight correlated with TT<sub>4</sub> and FT<sub>4</sub>, PCBs were the primary exploratory variable in the observed thyroid changes, whereas body weight was not significant for TT<sub>4</sub> (PCBs: partial <italic>R</italic> = 0.71, <italic>p</italic> < 0.001; body weight: partial <italic>R</italic> = 0.27, <italic>p</italic> = 0.08) or FT<sub>4</sub> (PCBs: partial <italic>R</italic> = 0.72, <italic>p</italic> < 0.001; body weight: partial <italic>R</italic> = 0.33, <italic>p</italic> = 0.14). There were no significant correlations between any of the endocrine end points and time of day for each capture (data not shown), suggesting that circadian rhythm did not unduly influence our results. Likewise, there was no correlation between time held (restraint) before release and any of the endocrine end points, suggesting that stress was not a factor (data not shown).</p></sec></sec><sec sec-type="discussion"><title>Discussion</title><p>A “weight of evidence” from laboratory-based studies, captive feeding studies of seals, and studies of free-ranging marine mammals highlights the endocrine-disrupting nature of complex mixtures of POPs and many of their constituents (<xref rid="b38-ehp0114-001024" ref-type="bibr">Ross 2000</xref>). Although mechanisms of action are often ill-defined in field studies, a common pattern of adverse health effects observed in contaminant-exposed individuals and populations includes developmental, immunologic, and reproductive effects. Despite having been banned in North America for three decades, the highly persistent PCBs continue to present a toxic risk to wildlife and dominated the contaminant profiles in our study of British Columbia and Washington State harbor seals. Gertrude Island (Puget Sound) harbor seals were particularly contaminated, having PCB concentrations that were several times higher than those sampled in the adjacent coastal waters of northern Washington State and southern and central British Columbia.</p><p>Elevated POP exposure has been associated with altered circulating vitamin A and immune function in free-ranging harbor seals sampled from the same study areas (<xref rid="b29-ehp0114-001024" ref-type="bibr">Levin et al. 2005</xref>; <xref rid="b43-ehp0114-001024" ref-type="bibr">Simms et al. 2000</xref>). Our observed negative relationship between circulating TT<sub>4</sub> and PCB concentrations in harbor seals contributes to the notion that PCBs represent a significant health concern at the top of the food chain. This finding is consistent with previous observations of contaminant-related reductions in TH concentrations in captive seals fed contaminated fish (<xref rid="b4-ehp0114-001024" ref-type="bibr">Brouwer et al. 1989</xref>) and in free-ranging pinnipeds (<xref rid="b6-ehp0114-001024" ref-type="bibr">Chiba et al. 2001</xref>; <xref rid="b11-ehp0114-001024" ref-type="bibr">Debier et al. 2005</xref>; <xref rid="b19-ehp0114-001024" ref-type="bibr">Hall et al. 1998</xref>; <xref rid="b24-ehp0114-001024" ref-type="bibr">Jenssen et al. 1995</xref>). Histopathologic lesions, including fibrosis and colloid depletion, in thyroid glands of seals inhabiting PCB-contaminated areas (<xref rid="b41-ehp0114-001024" ref-type="bibr">Schumacher et al. 1993</xref>) may explain reduced TH levels in our contaminated seals. However, laboratory animal studies provide more information on possible mechanisms of action, where altered hormone synthesis in the thyroid gland, disrupted circulatory transport, and altered metabolic enzyme activity have been observed (<xref rid="b3-ehp0114-001024" ref-type="bibr">Brouwer et al. 1998</xref>). Our findings suggest that the more contaminated seals from Gertrude Island may be considered hypothyroid (<xref rid="b20-ehp0114-001024" ref-type="bibr">Haulena et al. 1998</xref>), highlighting concerns about the health of high trophic level wildlife in this region.</p><p>THs play a critical role in regulating a wide range of physiologic processes such as growth, development, and metabolism, largely through binding to the nuclear receptors <italic>TR-</italic>α and <italic>TR-</italic>β, and modulate their activity on TH-responsive gene promoters (<xref rid="b54-ehp0114-001024" ref-type="bibr">Wu and Koenig 2000</xref>). PCBs can also directly affect TR activity and TH-responsive gene expression. The observed differential relationship between <italic>TR-</italic>α and <italic>TR-</italic>β transcripts in seal blubber samples relative to PCB levels may indicate a particular vulnerability of the <italic>TR-</italic>α gene. This may be related to the apparent hypothyroidism observed in the more contaminated animals. Interestingly, increased <italic>TR-</italic>α expression has been observed in the brains of hypothyroid compared with euthyroid rats (<xref rid="b8-ehp0114-001024" ref-type="bibr">Constantinou et al. 2005</xref>). Positively TH-regulated genes were up-regulated in postnatal and fetal rats brains after <italic>in utero</italic> exposure to the PCB mixture Aroclor 1254 despite a reduction in the dam’s circulating TH levels (<xref rid="b16-ehp0114-001024" ref-type="bibr">Gauger et al. 2004</xref>; <xref rid="b57-ehp0114-001024" ref-type="bibr">Zoeller et al. 2000</xref>). These results suggest that PCBs may interfere directly or indirectly with TH signaling, leading to changes in TH-responsive gene expression.</p><p>Altered circulating TH levels in PCB-exposed marine mammals provide evidence of an effect on this endocrine end point. However, obtaining blood samples is not always feasible, and skin/blubber biopsies essentially represent the only obtainable samples for many marine mammals, including cetaceans. In addition, circulating TH levels can be influenced by a number of natural factors and may not present a rigorous assessment of thyroid status. We therefore developed a gene expression biomarker approach using <italic>TR</italic> expression levels in blubber/skin biopsies in order to evaluate the utility of such an approach for harbor seals and other marine mammals. Using this biopsy-based sampling technique, we were able to quantify the expression of <italic>TR</italic> genes in blubber. Although we focused on <italic>TR</italic> expression in these studies, other emerging gene expression biomarkers could be examined in the same way. Expression levels of the aryl hydrocarbon receptor or cytochrome P450 as gene expression biomarkers in liver have already been suggested for marine mammals (<xref rid="b27-ehp0114-001024" ref-type="bibr">Kim and Hahn 2002</xref>).</p><p>The positive correlation between blubber <italic>TR-</italic>α and PCB concentrations in harbor seals suggests that contaminants either directly or indirectly affect this TH end point and may alter TH-regulated gene expression. The high degree of sequence conservation between harbor seals and other vertebrates accentuates the likely functional similarity of TRs between animal groups. Although our results would suggest that PCBs affect systemic thyroid homeostasis in harbor seals, our detection of contaminant-related alteration of <italic>TR</italic> gene expression in blubber raises a toxicologic concern of particular note for marine mammals. TH is known to play an important role in the maintenance and function of adipose tissue (<xref rid="b1-ehp0114-001024" ref-type="bibr">Ailhaud et al. 1992</xref>). T<sub>3</sub> treatment can induce adipocyte cell proliferation, fat cell cluster formation, lipid accumulation, and increased malic enzyme and glycerophosphate dehydrogenase activities in young rats as well as in preadipocyte cell lines (<xref rid="b14-ehp0114-001024" ref-type="bibr">Flores-Delgado et al. 1987</xref>; <xref rid="b18-ehp0114-001024" ref-type="bibr">Grimaldi et al. 1982</xref>). TRs within murine adipocytes predominantly are composed of the <italic>TR-</italic>α isoform, with little detectable <italic>TR-</italic>β isoform (<xref rid="b25-ehp0114-001024" ref-type="bibr">Jiang et al. 2004</xref>), consistent with our findings in blubber. Recently, several TH-regulated genes were identified in human and mouse adipose tissue that encode for protein products involved in lipid metabolism (<xref rid="b53-ehp0114-001024" ref-type="bibr">Viguerie et al. 2002</xref>).</p><p>Blubber is a specialized adipose tissue layer under the skin of marine mammals and is vital for energy storage, heat insulation, thermogenesis, and buoyancy control. Blubber is typically viewed as a storage depot associated with lipid reserves, within which lipids, lipid classes, and fatty acid profiles have been characterized in physiologic and energetic studies of marine mammals (<xref rid="b22-ehp0114-001024" ref-type="bibr">Iverson et al. 1997</xref>; <xref rid="b30-ehp0114-001024" ref-type="bibr">Mellish et al. 1999</xref>). However, blubber also represents an important storage site for micronutrients, holding as much as 66% of the body burden of vitamin A in harbor seals (<xref rid="b32-ehp0114-001024" ref-type="bibr">Mos and Ross 2002</xref>). Our finding of <italic>TR-</italic>α in blubber highlights the metabolically dynamic nature of this tissue because this receptor mediates the actions of TH-dependent metabolism and homeostasis. We speculate that contaminants might therefore present a risk to the structural and functional integrity of blubber because metabolism within adipocytes may be altered. The influence of TH-related processes on body weight in laboratory animals and in humans (<xref rid="b2-ehp0114-001024" ref-type="bibr">Baxter et al. 2004</xref>; <xref rid="b35-ehp0114-001024" ref-type="bibr">Pelletier et al. 2003</xref>) underscores the potential effects of a disruption of TR on such critical processes as energy storage and thermoregulation in marine mammals.</p><p>Our biomarker-based thyroid assessment may be applied to studies of other species for which blood samples are not available because of logistical constraints (e.g., cetaceans). The contaminant-associated decrease in circulating TH levels and concomitant up-regulation of <italic>TR-</italic>α expression in blubber of harbor seals may indicate an increased risk of TH-dependent health effects, such as developmental abnormalities and neurotoxicity. In addition, altered <italic>TR-</italic>α gene expression in blubber may have profound consequences for metabolic turnover and energetics in contaminated marine mammal populations.</p></sec>
Paraoxonase Polymorphisms, Haplotypes, and Enzyme Activity in Latino Mothers and Newborns	<p>Recent studies have demonstrated widespread pesticide exposures in pregnant women and in children. Plasma paraoxonase 1 (PON1) plays an important role in detoxification of various organophosphates. The goals of this study were to examine in the Center for Health Assessment of Mothers and Children of Salinas (CHAMACOS) birth cohort of Latina mothers and their newborns living in the Salinas Valley, California, the frequencies of five <italic>PON1</italic> polymorphisms in the coding region (<italic>192</italic><italic><sub>QR</sub></italic> and <italic>55</italic><italic><sub>LM</sub></italic>) and the promoter region (−<italic>162</italic><italic><sub>AG</sub></italic>, −<italic>909</italic><italic><sub>CG</sub></italic>, and −<italic>108</italic><italic><sub>CT</sub></italic>) and to determine their associations with PON1 plasma levels [phenylacetate arylesterase (AREase)] and enzyme activities of paraoxonase (POase) and chlorpyrifos oxonase (CPOase). Additionally, we report results of <italic>PON1</italic> linkage analysis and estimate the predictive value of haplotypes for PON1 plasma levels. We found that <italic>PON1</italic><sub>−</sub><italic><sub>909</sub></italic>, <italic>PON1</italic><sub>−</sub><italic><sub>108</sub></italic>, and <italic>PON1</italic><italic><sub>192</sub></italic> had an equal frequency (0.5) of both alleles, whereas <italic>PON1</italic><sub>−</sub><italic><sub>162</sub></italic> and <italic>PON1</italic><italic><sub>55</sub></italic> had lower variant allele frequencies (0.2). Nearly complete linkage disequilibrium was observed among coding and promoter polymorphisms (<italic>p</italic> < 0.001), except <italic>PON1</italic><italic><sub>192</sub></italic> and <italic>PON1</italic><sub>−</sub><italic><sub>162</sub></italic> (<italic>p</italic> > 0.4). Children’s PON1 plasma levels (AREase ranged from 4.3 to 110.7 U/mL) were 4-fold lower than their mothers’ (19.8 to 281.4 U/mL). POase and CPOase activities were approximately 3-fold lower in newborns than in mothers. The genetic contribution to PON1 enzyme variability was higher in newborns (<italic>R</italic><sup>2</sup> = 25.1% by genotype and 26.3% by haplotype) than in mothers (<italic>R</italic><sup>2</sup> = 8.1 and 8.8%, respectively). However, haplotypes and genotypes were comparable in predicting PON1 plasma levels in mothers and newborns. Most of the newborn children and some pregnant women in this Latino cohort may have elevated susceptibility to organophosphate toxicity because of their <italic>PON1</italic><italic><sub>192</sub></italic> genotype and low PON1 plasma levels.</p>	<contrib contrib-type="author"><name><surname>Holland</surname><given-names>Nina</given-names></name><xref ref-type="aff" rid="af1-ehp0114-000985">1</xref></contrib><contrib contrib-type="author"><name><surname>Furlong</surname><given-names>Clement</given-names></name><xref ref-type="aff" rid="af2-ehp0114-000985">2</xref></contrib><contrib contrib-type="author"><name><surname>Bastaki</surname><given-names>Maria</given-names></name><xref ref-type="aff" rid="af1-ehp0114-000985">1</xref></contrib><contrib contrib-type="author"><name><surname>Richter</surname><given-names>Rebecca</given-names></name><xref ref-type="aff" rid="af2-ehp0114-000985">2</xref></contrib><contrib contrib-type="author"><name><surname>Bradman</surname><given-names>Asa</given-names></name><xref ref-type="aff" rid="af1-ehp0114-000985">1</xref></contrib><contrib contrib-type="author"><name><surname>Huen</surname><given-names>Karen</given-names></name><xref ref-type="aff" rid="af1-ehp0114-000985">1</xref></contrib><contrib contrib-type="author"><name><surname>Beckman</surname><given-names>Kenneth</given-names></name><xref ref-type="aff" rid="af3-ehp0114-000985">3</xref></contrib><contrib contrib-type="author"><name><surname>Eskenazi</surname><given-names>Brenda</given-names></name><xref ref-type="aff" rid="af1-ehp0114-000985">1</xref></contrib>	Environmental Health Perspectives	<p>Organophosphate (OP) pesticide exposure remains widespread in the United States (<xref rid="b3-ehp0114-000985" ref-type="bibr">Barr et al. 2004</xref>; <xref rid="b9-ehp0114-000985" ref-type="bibr">Bradman et al. 2005</xref>; <xref rid="b34-ehp0114-000985" ref-type="bibr">Hill et al. 1995</xref>; <xref rid="b42-ehp0114-000985" ref-type="bibr">Loewenherz et al. 1997</xref>; <xref rid="b52-ehp0114-000985" ref-type="bibr">Simcox et al. 1999</xref>). Pregnant women, fetuses, and children in both urban (<xref rid="b5-ehp0114-000985" ref-type="bibr">Berkowitz et al. 2003</xref>; <xref rid="b60-ehp0114-000985" ref-type="bibr">Whyatt et al. 2003</xref>) and rural agricultural populations (<xref rid="b28-ehp0114-000985" ref-type="bibr">Eskenazi et al. 2004</xref>; <xref rid="b29-ehp0114-000985" ref-type="bibr">Fenske et al. 2002</xref>) are directly exposed to pesticides, and in some cases these exposures may exceed health-based reference doses (<xref rid="b9-ehp0114-000985" ref-type="bibr">Bradman et al. 2005</xref>; <xref rid="b14-ehp0114-000985" ref-type="bibr">Castorina et al. 2003</xref>). OP pesticide metabolites have also been detected in meconium (<xref rid="b59-ehp0114-000985" ref-type="bibr">Whyatt and Barr 2001</xref>) and amniotic fluid (<xref rid="b8-ehp0114-000985" ref-type="bibr">Bradman et al. 2003</xref>). OP exposure at high doses has profound effects, primarily on the central nervous system (<xref rid="b26-ehp0114-000985" ref-type="bibr">Eskenazi et al. 1999</xref>), and there is growing information in animals and humans suggesting that low-level chronic exposure may affect neurodevelopment (<xref rid="b26-ehp0114-000985" ref-type="bibr">Eskenazi et al. 1999</xref>; <xref rid="b61-ehp0114-000985" ref-type="bibr">Young et al. 2005</xref>).</p><p>The unique physiologic and behavioral characteristics of children may increase their exposures to environmental contaminants compared with adults (<xref rid="b44-ehp0114-000985" ref-type="bibr">National Research Council 1993</xref>). Young children eat, drink, and breathe more per unit of body weight than do adults, and they also explore their environment orally, engaging in extensive hand-to-mouth behavior (<xref rid="b44-ehp0114-000985" ref-type="bibr">National Research Council 1993</xref>). In addition, young children may be more susceptible to the adverse effects of OP exposure than are adults, because of their lower ability to metabolize and detoxify OP pesticides (<xref rid="b45-ehp0114-000985" ref-type="bibr">Padilla et al. 2000</xref>; <xref rid="b51-ehp0114-000985" ref-type="bibr">Sheets 2000</xref>).</p><p>The human paraoxonase 1 (PON1) enzyme (43 kDa, composed of 354 amino acids) is a polymorphic, high-density lipoprotein-associated esterase that metabolizes many different substrates, including OP compounds (<xref rid="b22-ehp0114-000985" ref-type="bibr">Davies et al. 1996</xref>; <xref rid="b33-ehp0114-000985" ref-type="bibr">Geldmachervon Mallinckrodt and Diepgen 1988</xref>), drugs, and oxidized lipids (<xref rid="b24-ehp0114-000985" ref-type="bibr">Draganov et al. 2005</xref>; <xref rid="b57-ehp0114-000985" ref-type="bibr">Watson et al. 1995</xref>). Studies of the PON1 enzyme, which detoxifies activated oxon forms of several OP pesticides, including diazinon, chlorpyrifos, and parathion, indicate that PON1 levels in newborns are on average 3- to 4-fold lower than those of adults (<xref rid="b2-ehp0114-000985" ref-type="bibr">Augustinsson and Barr 1963</xref>; <xref rid="b18-ehp0114-000985" ref-type="bibr">Chen et al. 2003</xref>; <xref rid="b19-ehp0114-000985" ref-type="bibr">Cole et al. 2003</xref>; <xref rid="b25-ehp0114-000985" ref-type="bibr">Ecobichon and Stephens 1973</xref>; <xref rid="b43-ehp0114-000985" ref-type="bibr">Mueller et al. 1983</xref>). Newborns reach a plateau near adult PON1 levels between 6 and 24 months of age, suggesting that newborn children and infants will be more susceptible to OP compounds (<xref rid="b19-ehp0114-000985" ref-type="bibr">Cole et al. 2003</xref>).</p><p>The <italic>PON1</italic> gene has been mapped to chromosome 7q21.3–22.1 (<xref rid="b35-ehp0114-000985" ref-type="bibr">Humbert et al. 1993</xref>; <xref rid="b46-ehp0114-000985" ref-type="bibr">Primo-Parmo et al. 1996</xref>) and contains nine exons. Recent studies suggest that some individuals may have specific <italic>PON1</italic> genotypes that are associated with low levels of plasma <italic>PON1</italic> (<xref rid="b11-ehp0114-000985" ref-type="bibr">Brophy et al. 2001b</xref>; <xref rid="b23-ehp0114-000985" ref-type="bibr">Deakin et al. 2003</xref>; <xref rid="b56-ehp0114-000985" ref-type="bibr">Suehiro et al. 2000</xref>). The hydrolytic catalytic efficiency of some <italic>PON1</italic> substrates is dependent on the single nucleotide polymorphism (SNP) <italic>Q192R</italic> (<xref rid="b41-ehp0114-000985" ref-type="bibr">Li et al. 2000</xref>). However, adults with the same <italic>PON1</italic><italic><sub>192</sub></italic> genotype can have at least a 13-fold difference in <italic>PON1</italic> activities (<xref rid="b22-ehp0114-000985" ref-type="bibr">Davies et al. 1996</xref>; <xref rid="b31-ehp0114-000985" ref-type="bibr">Furlong et al. 2002</xref>). The <italic>C-108T</italic> polymorphism, in a Sp1 binding site of the promoter region, has a major effect on the expression of the <italic>PON1</italic> gene. The <italic>C-108</italic> allele expresses on average twice as much <italic>PON1</italic> as does the <italic>T-108</italic> allele (<xref rid="b11-ehp0114-000985" ref-type="bibr">Brophy et al. 2001b</xref>; <xref rid="b37-ehp0114-000985" ref-type="bibr">James et al. 2000</xref>). Other polymorphisms in the promoter region (<italic>A-162G</italic>, and <italic>C-909G</italic>) may have less significant effects on PON1 expression and are in strong disequilibrium with <italic>C-108T</italic> (<xref rid="b21-ehp0114-000985" ref-type="bibr">Costa et al. 2002</xref>; <xref rid="b37-ehp0114-000985" ref-type="bibr">James et al. 2000</xref>). The <italic>PON1</italic><italic><sub>M55</sub></italic> allele has been associated with low <italic>PON1</italic> enzyme levels; however, most of this effect is related to its strong disequilibrium with the <italic>T-108</italic> allele. Recently, additional promoter polymorphisms have been identified (<xref rid="b50-ehp0114-000985" ref-type="bibr">SeattleSNPs 2005</xref>); however, their influence on PON1 levels has yet to be determined (Jarvik GP, personal communication). Limited information on <italic>PON1</italic> haplotypes (<xref rid="b17-ehp0114-000985" ref-type="bibr">Chen et al. 2005</xref>; <xref rid="b39-ehp0114-000985" ref-type="bibr">Koda et al. 2004</xref>; <xref rid="b58-ehp0114-000985" ref-type="bibr">Wetmur et al. 2005</xref>) suggests that haplotypes provide no significant improvement in predicting PON1 levels over a combination of <italic>PON1</italic> polymorphisms (<xref rid="b17-ehp0114-000985" ref-type="bibr">Chen et al. 2005</xref>).</p><p>The gene frequencies for specific alleles of <italic>PON1</italic> genes vary by ethnicity, implying differential susceptibility to pesticides among different ethnic groups (<xref rid="b1-ehp0114-000985" ref-type="bibr">Allebrandt et al. 2002</xref>; <xref rid="b12-ehp0114-000985" ref-type="bibr">Brophy et al. 2002</xref>). In a study of mothers and newborns from New York, a noticeable difference in haplotype frequency was observed among three ethnic groups (<xref rid="b17-ehp0114-000985" ref-type="bibr">Chen et al. 2005</xref>).</p><p>In the present study, we examined the frequencies and haplotypes of five <italic>PON1</italic> polymorphisms in coding regions (<italic>192</italic><italic><sub>QR</sub></italic> and <italic>55</italic><italic><sub>LM</sub></italic>) and promoter regions (−<italic>162</italic><italic><sub>AG</sub></italic>, −<italic>909</italic><italic><sub>CG</sub></italic>, and −<italic>108</italic><italic><sub>CT</sub></italic>) and their associations with PON1 plasma levels and enzyme activities in pregnant Latina women and their newborns living in the Salinas Valley, California, an agricultural community (<xref rid="b27-ehp0114-000985" ref-type="bibr">Eskenazi et al. 2003</xref>) where approximately 500,000 pounds of OP pesticides are used annually (<xref rid="b13-ehp0114-000985" ref-type="bibr">California Environmental Protection Agency 2002</xref>). Additionally, we report results of <italic>PON1</italic> linkage analysis for five <italic>PON1</italic> polymorphisms and estimate the predictive value of haplotypes, compared with <italic>PON1</italic> genotypes, for PON1 plasma levels. The present study follows our recent publications demonstrating that the Salinas Valley population has a relatively high level of exposure to OP compounds (<xref rid="b9-ehp0114-000985" ref-type="bibr">Bradman et al. 2005</xref>) and that OP exposure as assessed by maternal dialkyl phosphate metabolite levels was associated with shorter gestational age (<xref rid="b28-ehp0114-000985" ref-type="bibr">Eskenazi et al. 2004</xref>) and increased frequency of abnormal reflexes in neonates (<xref rid="b61-ehp0114-000985" ref-type="bibr">Young et al. 2005</xref>).</p><sec sec-type="materials\|methods"><title>Materials and Methods</title><sec sec-type="subjects"><title>Subjects and recruitment</title><p>Pregnant women (<italic>n</italic> = 130) and their newborns (<italic>n</italic> = 130) were randomly selected from the CHAMACOS (Center for the Health Assessment of Mothers and Children of Salinas) cohort, a longitudinal birth cohort study of the effects of pesticides and other environmental exposures on the health of pregnant women and their children living in the Salinas Valley, California. Women were eligible for enrollment in the CHAMA-COS study if they were ≥ 18 years of age, < 20 weeks’ gestation at enrollment, English- or Spanish-speaking, Medi-Cal eligible, and planning to deliver at the Natividad Medical Center (<xref rid="b9-ehp0114-000985" ref-type="bibr">Bradman et al. 2005</xref>; <xref rid="b27-ehp0114-000985" ref-type="bibr">Eskenazi et al. 2003</xref>, <xref rid="b28-ehp0114-000985" ref-type="bibr">2004</xref>; <xref rid="b61-ehp0114-000985" ref-type="bibr">Young et al. 2005</xref>). All women in the subcohort described here were representative of the CHAMACOS cohort; they were Latina by ethnicity, including 85% born in Mexico and the remainder in the United States. Most of the participants never smoked (> 92%), had relatively high pesticide exposures based on diethyl phosphate urinary metabolites (median, 20 nmol/L; range, 7–560 nmol/L), and worked in agriculture during pregnancy (39%). Fathers were more likely to smoke (11%) and work in agriculture (72%) than were mothers. Study protocols were approved by the University of California, Berkeley, and the University of Washington human-subject review committees in compliance with all applicable requirements. Written informed consent was provided by all subjects.</p></sec><sec><title>Biologic samples collection and processing</title><p>We collected blood from mothers at the time of their glucose tolerance test (26.1 ± 2.3 weeks) and in the hospital shortly before or after delivery. Blood samples were also collected from the umbilical cords by delivery room staff once the baby was safely delivered. Heparinized whole blood was centrifuged, divided into plasma, buffy coats, and red blood cells, and then stored at −80°C. BD Vacutainers (Becton Dickinson, Franklin Lakes, NJ) without anticoagulant were used to collect serum and clot. Processed plasma samples were stored at −80°C before being shipped on dry ice to the University of Washington, Seattle, for analysis of enzyme activity.</p><p>DNA was isolated from blood clots. Blood clots thawed in a 37°C water bath were first mechanically disrupted using ClotSpin tubes (Gentra Systems Inc., Minneapolis, MN). The Qiagen protocol (Qiagen Inc., Santa Clarita, CA) was slightly modified by prolonging the initial lysis and protease digestion step to overnight incubation. DNA concentration was measured using PicoGreen (Molecular Probes Inc., Eugene, OR), adjusted to 10 ng/μL, plated in 96-well plates, and stored at −80°C. Samples were transferred to 384-well plates for analysis of multiple SNPs, using robotic equipment to avoid manual pipetting errors and for time efficiency.</p></sec><sec><title>PON1 <italic>genotyping</italic></title><p>Genotyping was conducted by the University of California, Berkeley, and Children’s Hospital Research Institute Genotyping Core. Taqman real-time polymerase chain reaction method was used for genotyping of the −<italic>162</italic><italic><sub>AG</sub></italic>, <italic>55</italic><italic><sub>LM</sub></italic>, and <italic>192</italic><italic><sub>QR</sub></italic> polymorphisms. Briefly, primers for these SNPs were custom designed by Applied Biosystems Inc. (Foster City, CA). Amplifluor allele-specific primers were used for genotyping of −<italic>909</italic><italic><sub>CG</sub></italic> and −<italic>108</italic><italic><sub>CT</sub></italic>. Genotype calling was performed either manually using a spreadsheet (Chemicon AssayAuditor, for real-time data) or by automatic allele calling in SDS 2.1 (Applied Biosystems, for end-point data).</p><p>Quality assurance procedures included assessment of randomly distributed blank samples in each plate, duplicates of randomly selected samples with independently isolated DNA from the same subjects, and internal controls. Repeated analysis in several runs showed a high degree (96.5%) of concordance, and the most robust call was selected in the case of discordance (3.5%). Furthermore, the assays were repeated for all low-confidence samples until a reliable call was obtained, using a combination of the TaqMan and Amplifluor methods for a subset of samples. Additional analysis was performed independently at the University of Washington for 10% of the DNA samples for the <italic>192</italic><italic><sub>QR</sub></italic> polymorphism by standard polymerase chain reaction method (details given by <xref rid="b47-ehp0114-000985" ref-type="bibr">Richter et al. 2004</xref>) with approximately 95% concordance; all discrepancies were resolved by repeated runs. Quality control software was used to check data for Mendelian errors, and if those were noted, the whole run was repeated.</p></sec><sec><title>Enzyme assays</title><p>Plasma was frozen at −80°C until analysis. We measured three PON1 enzyme activities in plasma from mothers and children, using paraoxonase (POase), chlorpyrifos oxonase (CPOase), and phenylacetate arylesterase (AREase) according to published protocols (<xref rid="b38-ehp0114-000985" ref-type="bibr">Jarvik et al. 2003</xref>; <xref rid="b48-ehp0114-000985" ref-type="bibr">Richter and Furlong 1999</xref>; <xref rid="b47-ehp0114-000985" ref-type="bibr">Richter et al. 2004</xref>). We used PON1 plasma levels (AREase assay) to analyze the genetic effect because, unlike POase and CPOase levels, they are not affected by differential catalytic efficiency primarily controlled by the <italic>PON1</italic><italic><sub>192</sub></italic> SNP and have been shown to correspond with PON1 levels determined by immunologic methods (<xref rid="b7-ehp0114-000985" ref-type="bibr">Blatter-Garin et al. 1994</xref>; <xref rid="b32-ehp0114-000985" ref-type="bibr">Furlong et al. 1993</xref>). Together, these three assays provide comprehensive information about PON1 enzyme activities regarding different substrates. Assessment of PON1 activities in mothers was first conducted for 25 pregnant women at two different time points, at 26 weeks and at delivery. PON1 activities were not statistically different between the two time points for all three PON1 enzyme assays (<italic>r</italic> = 0.77–1.0, <italic>p</italic> < 0.0001). Therefore, we performed analyses of AREase, POase, and CPOase in the remainder of 105 Latina mothers at one time point only—at 26 weeks’ gestation. Children in the study were of both sexes, and girls represented 54%. No sex differences in PON1 enzyme levels or genotypes were observed, as is consistent with the available PON1 literature (<xref rid="b21-ehp0114-000985" ref-type="bibr">Costa et al. 2002</xref>).</p></sec><sec sec-type="methods"><title>Statistical analysis</title><p>Standard analyses for all genotype data included analysis for Hardy-Weinberg equilibrium, pairwise linkage disequilibrium (LD), and haplotype assignment using algorithms implemented in the publicly available Haploview software (Battett et al. 2005), including PYPOP (<xref rid="b40-ehp0114-000985" ref-type="bibr">Lancaster et al. 2003</xref>), tagSNPs (<xref rid="b55-ehp0114-000985" ref-type="bibr">Stram et al. 2003</xref>), and PHASE (<xref rid="b54-ehp0114-000985" ref-type="bibr">Stephens et al. 2001</xref>, <xref rid="b53-ehp0114-000985" ref-type="bibr">2003</xref>). The LD statistic <italic>D</italic>′ was calculated for each pair of five <italic>PON1</italic> SNPs, and <italic>R</italic><sup>2</sup> values were used to describe the haplotype structure of the <italic>PON1</italic> gene in our Latino cohort. PYPOP, tagSNPs, and PHASE software methods showed similar results. We used PHASE to generate the data reported in this article, because it has been shown to reduce error rates in haplotype reconstruction compared with the expectation maximization algorithm (<xref rid="b53-ehp0114-000985" ref-type="bibr">Stephens and Donelly 2003</xref>).</p><p>Subjects were grouped according to their imputed diplotypes (<xref rid="b17-ehp0114-000985" ref-type="bibr">Chen et al. 2005</xref>). When more than one diplotype was possible for an individual, only the most likely imputed haplotypes were used in this analysis. The distributions and descriptive statistics were established separately for each of the three PON1 enzyme assays in mothers and in their newborns for each of the five SNPs. The distributions of enzyme activities were approximately normal. Linear regression and backward regression models were used to determine whether the additional information for all five polymorphisms altered the effect of genotype on enzyme activity. Coefficients of determination (total <italic>R</italic><sup>2</sup>) were calculated for the proportion of variability in PON1 plasma levels explained by the five SNP genotypes (used as ordinal variables) and by imputed haplotypes. Each haplotype with > 5% frequency was coded as a variable in the linear regression model, where the values 0, 1, or 2 denoted the presence of zero, one, or two copies of the haplotype for a subject. Haplotypes with < 5% frequency were pooled into one group for this analysis. All analyses were conducted in STATA software (version 8.0; StataCorp., College Station, TX) and SAS software (version 9.1; SAS Institute Inc., Cary, NC).</p></sec></sec><sec sec-type="results"><title>Results</title><sec><title>PON1 <italic>polymorphisms</italic></title><p><italic>PON1</italic> gene frequencies were established for two coding polymorphisms (<italic>PON1</italic><italic><sub>192</sub></italic> and <italic>PON1</italic><italic><sub>55</sub></italic>) and three promoter region polymorphisms (<italic>PON1</italic><sub>−</sub><italic><sub>909</sub></italic>, <italic>PON1</italic><sub>−</sub><italic><sub>162</sub></italic>, <italic>PON1</italic><sub>−</sub><italic><sub>108</sub></italic>) (<xref ref-type="table" rid="t1-ehp0114-000985">Table 1</xref>). As expected, the five polymorphisms had similar allelic frequencies in 130 pregnant Latina women of Mexican descent and their newborns. All genotypes were consistent with Hardy-Weinberg equilibrium (data not shown). The SNPs at position <italic>PON1</italic><sub>−</sub><italic><sub>162</sub></italic> of the promoter region and <italic>PON1</italic><italic><sub>55</sub></italic> in the coding region had lower variant allele frequencies (−<italic>162</italic>A, <italic>55</italic>M) than did the major allele, whereas the other three polymorphisms (<italic>PON1</italic><sub>−</sub><italic><sub>909</sub></italic>, <italic>PON1</italic><sub>−</sub><italic><sub>108</sub></italic><sub>,</sub> and <italic>PON1</italic><italic><sub>192</sub></italic>) had approximately equal presence of both alleles in this population. Specifically, the frequencies of <italic>PON1</italic><italic><sub>192</sub></italic> alleles were <italic>Q</italic> = 0.46, <italic>R</italic> = 0.54 in mothers, and <italic>Q</italic> = 0.51, <italic>R</italic> = 0.49 in children, with overall population prevalence Q ~ R ~ 0.5. Frequencies for the major alleles of promoter polymorphisms PON1<sub>G−909</sub>, PON1<sub>G−162</sub>, and PON1<sub>C−108</sub> were, respectively, 0.52, 0.78, and 0.51 in mothers and 0.56, 0.81, and 0.55 in children, and the frequency of a major allele of the coding <italic>PON1</italic><italic><sub>L55</sub></italic> polymorphism equaled 0.82 in both age groups.</p><p>Results of linkage analysis between five <italic>PON1</italic> polymorphisms were also similar for Latina mothers and their newborns (<xref ref-type="table" rid="t2-ehp0114-000985">Table 2</xref>). We observed nearly complete LD among the three promoter polymorphisms (<italic>D</italic>′ = 0.8–1; <italic>p</italic> < 0.001). Strong LD (<italic>D</italic>′ = 0.87 and 0.94 in mothers and children, respectively; <italic>p</italic> < 0.001) was found between the two coding polymorphisms (<italic>PON1</italic><italic><sub>192</sub></italic> and <italic>PON1</italic><italic><sub>55</sub></italic>). There was a more complex relationship between coding and promoter region polymorphisms: although <italic>PON1</italic><italic><sub>55</sub></italic> had high LD with all three promoter polymorphisms (<italic>D</italic>′ = 0.74–1), only two (<italic>PON1</italic><sub>−</sub><italic><sub>108</sub></italic> and <italic>PON1</italic><sub>−</sub><italic><sub>909</sub></italic>) of the three promoter SNPs were linked to <italic>PON1</italic><italic><sub>192</sub></italic> (<italic>D</italic>′ = 0.22 and 0.19 in mothers, and <italic>D</italic>′ = 0.27 and 0.34 in children, respectively). These LDs were all modest but statistically significant. However, no linkage was demonstrated between SNPs at positions <italic>PON1</italic><italic><sub>192</sub></italic> and <italic>PON1</italic><sub>−</sub><italic><sub>162</sub></italic> (<italic>D</italic>′ = 0.0 in mothers and 0.18 and children; <italic>p</italic> > 0.4).</p></sec><sec sec-type="methods"><title>Haplotype analysis</title><p>Haplotype analysis revealed a total of 32 different combinations of alleles (<xref ref-type="table" rid="t3-ehp0114-000985">Table 3</xref>). However, their frequencies were noticeably different and fall into three distinct groups: <italic>a</italic>) a main group contributing approximately 93% of all haplotypes for this cohort, which is composed of seven haplotypes with individual frequencies ranging from 7 to 24%; <italic>b</italic>) a second group with individual frequencies ranging from 0.1 to 1.9%, which contributes 5.5–7.8% of all haplotypes in mothers and children; and <italic>c</italic>) a group of 17 rare haplotypes contributing a total of approximately 1% of haplotype variability.</p></sec><sec><title>PON1 <italic>enzyme activities</italic></title><p>AREase levels allow for a comparison of PON1 levels across genotypes because the catalytic efficiency of hydrolysis of phenylacetate is not affected by the <italic>PON1</italic><italic><sub>192</sub></italic> polymorphism (<xref ref-type="table" rid="t4-ehp0114-000985">Tables 4</xref>, <xref ref-type="table" rid="t5-ehp0114-000985">5</xref>). The AREase activity in mothers ranged from 19.8 to 281.4 U/mL and in newborns, from 4.3 to 110.7 U/mL. The mean AREase values for mothers were similar across three <italic>PON1</italic><italic><sub>192</sub></italic> genotypes (Q/Q = 151.9 U/mL; Q/R = 144.3 U/L; R/R = 152.2 U/L; <italic>p</italic> = 0.64). In cord samples, the <italic>Q192R</italic> polymorphism slightly influenced AREase levels with <italic>PON1</italic><italic><sub>R192</sub></italic> individuals having the highest average levels (Q/Q = 30.8 U/mL; Q/R = 35.8 U/mL; R/R = 42.9 U/mL), although the difference between genotypes was not statistically significant (<italic>p</italic> = 0.13).</p><p>AREase levels varied noticeably across <italic>C-108T</italic> genotypes in mothers (C/C = 163.6 U/mL; C/T = 147.1 U/mL; T/T = 134.8 U/mL; <italic>p</italic> = 0.04) with a larger gradient in newborns (C/C = 48.7 U/mL; C/T = 34.0 U/mL; T/T = 27.3 U/mL; <italic>p</italic> = 0.0003). AREase levels also differed by the <italic>L55M</italic> polymorphism in cord blood (<italic>p</italic> < 0.0001) but not in maternal blood (<italic>p</italic> = 0.44), with M/M homozygotes having the lowest levels in children (17.6 U/mL) and in mothers (135.6 U/mL). AREase levels were significantly higher in subjects with the <italic>G-909</italic> and <italic>A-162</italic> alleles in both mothers and children (<italic>p</italic> = 0.0003–0.03).</p><p>Mean POase and CPOase activities in mothers were significantly higher (1024.2 and 9358.3 U/L, respectively) than in newborns (315.1 and 2663.4 U/L, respectively) (<xref ref-type="table" rid="t4-ehp0114-000985">Tables 4</xref>, <xref ref-type="table" rid="t5-ehp0114-000985">5</xref>). Both POase and CPOase activity levels demonstrated a strong association with the <italic>PON1</italic><italic><sub>192</sub></italic> genotype. The lowest mean average POase activity was seen in <italic>192</italic><italic><sub>QQ</sub></italic> children (81.2 U/L) and the highest in <italic>192</italic><italic><sub>RR</sub></italic> mothers (1927.9 U/L), resulting in a 24-fold difference among these two genotype/age groups. Furthermore, the lowest overall POase level (10.3 U/L) was observed in the <italic>192</italic><italic><sub>QQ</sub></italic> newborn child, and the highest in <italic>192</italic><italic><sub>RR</sub></italic> adult (3014.2 U/L), bringing the overall difference among individuals from the same population to 300-fold. For CPOase, respective differences in enzyme activity between the lowest and highest levels in this cohort reached 70-fold. Both POase and CPOase activities also varied significantly by the other four <italic>PON1</italic> polymorphisms, with the lowest values for −909<sub>GG</sub>, −162<sub>GG</sub>, −108<sub>TT</sub>, and 55<sub>MM</sub> homozygote groups (<italic>p</italic> < 0.001 in children; <italic>p</italic> < 0.03 in mothers).</p><p>Overall, PON1 activity levels in maternal blood were significantly higher than in cord blood for all enzyme assays and all genotypes (<italic>p</italic> < 0.001). Specifically, maternal POase, CPOase, and AREase levels were 3.3-fold, 3.6-fold, and 4.0-fold higher, respectively, than those in newborn. As expected, AREase, CPOase, and POase levels correlated well within <italic>PON1</italic><italic><sub>192</sub></italic> genotypes. In mothers, the correlations between AREase and CPOase levels ranged between 0.62 and 0.74 in <italic>QQ</italic>, <italic>QR</italic>, and <italic>RR</italic> groups, and in newborns, these correlations were even stronger, 0.90–0.92 (all <italic>p</italic>-values < 0.0001 for both mothers and newborns). The correlations between AREase and POase, and between POase and CPOase were the highest in <italic>RR</italic> newborns (both 0.93) and mothers (0.7 and 0.95, respectively), and somewhat lower for other maternal and new-born <italic>PON1</italic><italic><sub>192</sub></italic> genotype groups (all <italic>p</italic>-values less than 0.001). AREase activity was not compared with either CPOase or POase across genotypes because of the differential effects of the <italic>PON1</italic><italic><sub>192</sub></italic> polymorphism on CPOase or POase activities. For example, the <italic>PON1</italic><italic><sub>Q192</sub></italic> alloform hydrolyzes paraoxon with a catalytic efficiency nine times lower than <italic>PON1</italic><italic><sub>R192</sub></italic> (<xref rid="b41-ehp0114-000985" ref-type="bibr">Li et al. 2000</xref>).</p></sec><sec><title><italic>Phenotypic effects of</italic> PON1 <italic>genotype and haplotype</italic></title><p>We constructed linear regression models to determine the proportion of the variance of AREase explained by the five <italic>PON1</italic> polymorphisms and the imputed haplotypes. The five <italic>PON1</italic> genotype polymorphisms explained 8.1 and 23.1% of the variance of AREase in mothers and newborns, respectively. The coefficient of variation (<italic>R</italic><sup>2</sup>) was similar for both promoter polymorphisms <italic>PON1</italic><sub>−</sub><italic><sub>909</sub></italic> and <italic>PON1</italic><sub>−</sub><italic><sub>162</sub></italic> in mothers (~5%) and children (~14%) after adjusting for <italic>PON1</italic><sub>−</sub><italic><sub>108</sub></italic>. <italic>PON1</italic> haplotypes did not significantly improve the amount of variance explained (total <italic>R</italic><sup>2</sup> = 8.8% for mothers, 26.3% for newborns). The genetic contribution to AREase levels was significantly higher in newborns than in their mothers (<italic>p</italic> < 0.01). Because POase and CPOase characterize PON1 catalytic efficiency and are primarily controlled by <italic>PON1</italic><italic><sub>192</sub></italic> polymorphism, a comparison of haplotype and genotype effects by these two assays was not relevant.</p></sec></sec><sec sec-type="discussion"><title>Discussion</title><p>To our knowledge, this is the first study to report three PON1 enzyme activity levels in a large cohort of newborns and mothers from an agricultural cohort with relatively high levels of OP exposure. Two main PON1 factors are likely to contribute to the risk of adverse health effects of OP exposure: the level of enzyme (as measured by AREase assay) and the ability of this enzyme to detoxify OP metabolites (as measured in this study by CPOase assay, and primarily affected by <italic>PON1</italic><italic><sub>192</sub></italic>). Thus, newborn children in this cohort, based on their lower PON1 plasma levels and detoxifying activities, are likely to be significantly more susceptible to OP exposure than are their mothers. Similar to results of a study from Mexico (<xref rid="b49-ehp0114-000985" ref-type="bibr">Rojas-Garcia et al. 2005</xref>), we found large interindividual variability in PON1 plasma levels in both mothers and children, with a 14-fold difference in AREase among mothers, a 25-fold difference in newborns, and an overall range of 65-fold in this cohort. Additionally, we observed a range of 70-fold for CPOase and 300-fold for POase. However, it is important to emphasize that POase variability does not reflect differential sensitivity to paraoxon exposure based on recent animal data (<xref rid="b41-ehp0114-000985" ref-type="bibr">Li et al. 2000</xref>). On the other hand, PON1 levels and <italic>PON1</italic><italic><sub>192</sub></italic> polymorphism are very important in determining sensitivity to chlorpyrifos and chlorpyrifos oxon exposure (<xref rid="b20-ehp0114-000985" ref-type="bibr">Cole et al. 2005</xref>; <xref rid="b30-ehp0114-000985" ref-type="bibr">Furlong et al. 2006</xref>). Further, AREase variability primarily defines sensitivity to diazoxon exposure (<xref rid="b30-ehp0114-000985" ref-type="bibr">Furlong et al. 2006</xref>; <xref rid="b41-ehp0114-000985" ref-type="bibr">Li et al. 2000</xref>). Moreover, given the wide range in enzyme levels, some of the mothers are predicted to have an elevated susceptibility because they have levels as low as most of the newborns.</p><p>In nine children followed longitudinally, PON1 reached plateaus comparable with mean adult levels between 6 and 24 months of age (<xref rid="b19-ehp0114-000985" ref-type="bibr">Cole et al. 2003</xref>). Our finding that CHAMACOS mothers had approximately 4-fold higher AREase levels than did their newborns confirms previous observations of lower PON1 activities in small groups of neonates compared with adults (<xref rid="b2-ehp0114-000985" ref-type="bibr">Augustinsson and Barr 1963</xref>; <xref rid="b25-ehp0114-000985" ref-type="bibr">Ecobichon and Stephens 1973</xref>; <xref rid="b43-ehp0114-000985" ref-type="bibr">Mueller et al. 1983</xref>). Our finding is also consistent with a recent report where neonates had 2.6- to 4.6-fold lower PON1 levels compared with mothers in three ethnic groups residing in New York (<xref rid="b18-ehp0114-000985" ref-type="bibr">Chen et al. 2003</xref>). However, no previous study has reported either POase or CPOase variability in such a large cohort of newborns.</p><p>In Latino newborns of the CHAMACOS cohort, all three <italic>PON1</italic> promoter polymorphisms as well as <italic>PON1</italic><italic><sub>55</sub></italic> were significantly associated with AREase levels in children, and a greater proportion of the variance in AREase enzyme levels was explained by genetic polymorphisms in newborns than in mothers. The association of these polymorphisms and AREase levels is in agreement with another study of PON1 levels in newborns (<xref rid="b17-ehp0114-000985" ref-type="bibr">Chen et al. 2005</xref>). There was a nearly complete LD among the three promoter region polymorphisms, as also observed in other studies (<xref rid="b10-ehp0114-000985" ref-type="bibr">Brophy et al. 2001a</xref>; <xref rid="b17-ehp0114-000985" ref-type="bibr">Chen et al. 2005</xref>; <xref rid="b37-ehp0114-000985" ref-type="bibr">James et al. 2000</xref>; <xref rid="b49-ehp0114-000985" ref-type="bibr">Rojas-Garcia et al. 2005</xref>). However, <italic>PON1</italic><italic><sub>192</sub></italic> was not in LD with promoter SNP <italic>PON1</italic><sub>−</sub><italic><sub>162</sub></italic> and was in weak LD with the <italic>PON1</italic><sub>−</sub><italic><sub>108</sub></italic> and <italic>PON1</italic><sub>−</sub><italic><sub>909</sub></italic>. The lack of strong LD between <italic>PON1</italic><italic><sub>192</sub></italic> and promoter polymorphisms is also in agreement with data from the Hispanic population in New York City (<xref rid="b17-ehp0114-000985" ref-type="bibr">Chen et al. 2005</xref>). We found stronger LD between the two coding-region SNPs, <italic>PON1</italic><italic><sub>192</sub></italic> and <italic>PON1</italic><italic><sub>55</sub></italic>, in both mothers and children (<italic>D</italic>′ = 0.88 and 0.94, respectively) of the CHAMACOS cohort compared with Hispanics in New York City (<xref rid="b17-ehp0114-000985" ref-type="bibr">Chen et al. 2005</xref>). The differences in linkage pattern may be attributed to variation among ethnic groups (<xref rid="b39-ehp0114-000985" ref-type="bibr">Koda et al. 2004</xref>).</p><p>It has been reported that the association of the <italic>M55</italic> allele with low PON1 levels is primarily attributable to LD with the inefficient <italic>T-108</italic> allele (<xref rid="b11-ehp0114-000985" ref-type="bibr">Brophy et al. 2001b</xref>). <italic>PON1</italic><italic><sub>M55</sub></italic> has also been reported to be somewhat less stable than <italic>PON1</italic><italic><sub>L55</sub></italic>, additionally affecting protein levels in plasma (<xref rid="b37-ehp0114-000985" ref-type="bibr">James et al. 2000</xref>). This may explain why in CHAMACOS mothers, who have about a 4-fold higher AREase levels than the newborns, the effect of <italic>PON1</italic><italic><sub>55</sub></italic> was not statistically significant.</p><p>Our analysis of five <italic>PON1</italic> SNPs in a Latino population of Mexican descent living in California suggests that these SNPs may be located on separate haplotype blocks because we found nearly complete LD among the coding SNPs but not between <italic>PON1</italic><italic><sub>192</sub></italic> and <italic>PON1</italic><sub>−</sub><italic><sub>162</sub></italic> The presence of several haplotypes blocks in the <italic>PON1</italic> gene has been previously reported for other ethnic groups (<xref rid="b36-ehp0114-000985" ref-type="bibr">International HapMap Consortium 2003</xref>; <xref rid="b39-ehp0114-000985" ref-type="bibr">Koda et al. 2004</xref>). This underscores the importance of further analysis of <italic>PON1</italic> genetic variability. The gene frequencies for specific alleles of <italic>PON1</italic> genes vary by ethnicity, implying different population susceptibility to pesticides (<xref rid="b21-ehp0114-000985" ref-type="bibr">Costa et al. 2002</xref>). The frequency of <italic>PON1</italic><italic><sub>192</sub></italic> alleles in our Latina cohort of Mexican descent (<italic>Q</italic> = 0.5) was similar to those observed in Caribbean Hispanic mothers and neonates in New York City (both <italic>Q</italic> = 0.5) (<xref rid="b18-ehp0114-000985" ref-type="bibr">Chen et al. 2003</xref>). <italic>PON1</italic><sub>−</sub><italic><sub>162</sub></italic> frequencies were also comparable in these two populations (~ 0.8). However, the frequencies for the <italic>PON1</italic><sub>−</sub><italic><sub>108</sub></italic>, <italic>PON1</italic><sub>−</sub><italic><sub>909</sub></italic>, and <italic>PON1</italic><italic><sub>55</sub></italic> were noticeably different between Latinos from California and New York. The <italic>PON1</italic><italic><sub>192</sub></italic> frequency in Hispanics from Washington State (<italic>Q</italic> = 0.6) was slightly higher than in New York and California (<xref rid="b12-ehp0114-000985" ref-type="bibr">Brophy et al. 2002</xref>). In previous studies, the allele frequencies for <italic>PON1</italic><italic><sub>192</sub></italic> polymorphism in Caucasians was <italic>Q</italic> = 0.7, whereas for African Americans and other groups of African descent, the <italic>PON1</italic><italic><sub>192</sub></italic> frequencies are reversed, <italic>Q</italic> = 0.3 (<xref rid="b12-ehp0114-000985" ref-type="bibr">Brophy et al. 2002</xref>).</p><p><xref rid="b1-ehp0114-000985" ref-type="bibr">Allebrandt et al. (2002)</xref> have compared a combination of the <italic>PON1</italic><italic><sub>192</sub></italic> and <italic>PON</italic><italic><sub>155</sub></italic> allele frequencies across various ethnic groups. Using this approach, Mexican Latinos of the CHAMACOS cohort appear to be equally differentiated from Caucasians, Asians, and African Americans, which is consistent with their Native American background, whereas Caribbean Hispanics from New York (<xref rid="b18-ehp0114-000985" ref-type="bibr">Chen et al. 2003</xref>) are closer to Africans and Caucasians (data not shown). This difference across ethnic groups corroborates genetic and historical information about these populations (Cavalli-Sforza et al. 1993).</p><p>An effect of both <italic>PON1</italic> genotype and haplotypes on PON1 phenotype as measured by AREase was stronger in CHAMACOS newborns. This is in agreement with another study (<xref rid="b17-ehp0114-000985" ref-type="bibr">Chen et al. 2005</xref>; <xref rid="b58-ehp0114-000985" ref-type="bibr">Wetmur et al. 2005</xref>) that evaluated the relationship of five <italic>PON1</italic> SNPs with enzyme activity in mothers and their newborns. It is also clear that polymorphisms characterized to date in the <italic>PON1</italic> gene account for only a portion of the variability in PON1 levels observed among individuals. Additional research needs to be carried out to identify other factors (e.g., trans-acting factors, other <italic>PON1</italic> polymorphisms including intronic and exonic splice enhancing sequences) that influence PON1 expression.</p><p>Individuals with low PON1 activity are hypothesized to be at higher risk for any adverse health effects of OP exposure. In the only study to date to directly examine this hypothesis, <xref rid="b6-ehp0114-000985" ref-type="bibr">Berkowitz et al. (2004)</xref> reported that in residents of east Harlem (the same cohort described by <xref rid="b18-ehp0114-000985" ref-type="bibr">Chen et al. 2003</xref>), low PON1 plasma levels were associated with smaller neonatal head circumference. Further, although prenatal levels of the urinary metabolite of chlorpyrifos—3,5,6-trichloro-2-pyridinol (TCP)—were not associated with any measure of fetal growth or length of gestation by itself, higher levels of TCP were associated with smaller head circumference in children whose mothers had low expression of PON1.</p><p>We previously reported in the CHAMA-COS cohort that OP exposure as measured by urinary dialkyl phosphate metabolite levels of the mother during pregnancy was associated with shorter gestational duration (<xref rid="b28-ehp0114-000985" ref-type="bibr">Eskenazi et al. 2004</xref>) and poorer neonatal reflexes (<xref rid="b61-ehp0114-000985" ref-type="bibr">Young et al. 2005</xref>). A recent publication links <italic>PON1</italic><italic><sub>RR</sub></italic> and <italic>PON2</italic><italic><sub>CC</sub></italic> genotypes in infants with increased risk of preterm delivery in China (<xref rid="b16-ehp0114-000985" ref-type="bibr">Chen et al. 2004</xref>). In future analyses, we aim to expand the analyses of PON1 to the entire CHAMACOS cohort and to determine whether PON1 levels modify the previously observed relationship between OP exposure and gestational duration and neonatal development.</p></sec>
Cancer Mortality in Workers Exposed to Organochlorine Compounds in the Pulp and Paper Industry: An International Collaborative Study	<p>The objective of this study was to evaluate cancer mortality in pulp and paper industry workers exposed to chlorinated organic compounds. We assembled a multinational cohort of workers employed between 1920 and 1996 in 11 countries. Exposure to both volatile and nonvolatile organochlorine compounds was estimated at the department level using an exposure matrix. We conducted a standardized mortality ratio (SMR) analysis based on age and calendar-period–specific national mortality rates and a Poisson regression analysis. The study population consisted of 60,468 workers. Workers exposed to volatile organochlorines experienced a deficit of all-cause [SMR = 0.91; 95% confidence interval (CI), 0.89–0.93] and all-cancer (SMR = 0.93; 95% CI, 0.89–0.97) mortality, with no evidence of increased risks for any cancer of <italic>a priori</italic> interest. There was a weak, but statistically significant, trend of increasing risk of all-cancer mortality with increasing weighted cumulative exposure. A similar deficit in all-cause (SMR = 0.94; 95% CI, 0.91–0.96) and all-cancer (SMR = 0.94; 95% CI, 0.89–1.00) mortality was observed in those exposed to non-volatile organochlorines. No excess risk was observed in cancers of <italic>a priori</italic> interest, although mortality from Hodgkin disease was elevated (SMR = 1.76; 95% CI, 1.02–2.82). In this study we found little evidence that exposure to organochlorines at the levels experienced in the pulp and paper industry is associated with an increased risk of cancer, apart from a weak but significant association between all-cancer mortality and weighted cumulative volatile organochlorine exposure.</p>	<contrib contrib-type="author"><name><surname>McLean</surname><given-names>David</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001007">1</xref><xref ref-type="aff" rid="af2-ehp0114-001007">2</xref></contrib><contrib contrib-type="author"><name><surname>Pearce</surname><given-names>Neil</given-names></name><xref ref-type="aff" rid="af2-ehp0114-001007">2</xref></contrib><contrib contrib-type="author"><name><surname>Langseth</surname><given-names>Hilde</given-names></name><xref ref-type="aff" rid="af3-ehp0114-001007">3</xref></contrib><contrib contrib-type="author"><name><surname>Jäppinen</surname><given-names>Paavo</given-names></name><xref ref-type="aff" rid="af4-ehp0114-001007">4</xref></contrib><contrib contrib-type="author"><name><surname>Szadkowska-Stanczyk</surname><given-names>Irena</given-names></name><xref ref-type="aff" rid="af5-ehp0114-001007">5</xref></contrib><contrib contrib-type="author"><name><surname>Persson</surname><given-names>Bodil</given-names></name><xref ref-type="aff" rid="af6-ehp0114-001007">6</xref></contrib><contrib contrib-type="author"><name><surname>Wild</surname><given-names>Pascal</given-names></name><xref ref-type="aff" rid="af7-ehp0114-001007">7</xref></contrib><contrib contrib-type="author"><name><surname>Kishi</surname><given-names>Reiko</given-names></name><xref ref-type="aff" rid="af8-ehp0114-001007">8</xref></contrib><contrib contrib-type="author"><name><surname>Lynge</surname><given-names>Elsebeth</given-names></name><xref ref-type="aff" rid="af9-ehp0114-001007">9</xref></contrib><contrib contrib-type="author"><name><surname>Henneberger</surname><given-names>Paul</given-names></name><xref ref-type="aff" rid="af10-ehp0114-001007">10</xref></contrib><contrib contrib-type="author"><name><surname>Sala</surname><given-names>Maria</given-names></name><xref ref-type="aff" rid="af11-ehp0114-001007">11</xref></contrib><contrib contrib-type="author"><name><surname>Teschke</surname><given-names>Kay</given-names></name><xref ref-type="aff" rid="af12-ehp0114-001007">12</xref></contrib><contrib contrib-type="author"><name><surname>Kauppinen</surname><given-names>Timo</given-names></name><xref ref-type="aff" rid="af13-ehp0114-001007">13</xref></contrib><contrib contrib-type="author"><name><surname>Colin</surname><given-names>Didier</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001007">1</xref></contrib><contrib contrib-type="author"><name><surname>Kogevinas</surname><given-names>Manolis</given-names></name><xref ref-type="aff" rid="af11-ehp0114-001007">11</xref></contrib><contrib contrib-type="author"><name><surname>Boffetta</surname><given-names>Paolo</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001007">1</xref></contrib>	Environmental Health Perspectives	<p>Pulp and paper production workers have been exposed to a complex mixture of hazardous substances, including known or suspected carcinogens such as wood dust, various wood extracts and associated bioaerosols, reduced sulfur compounds, talc, formaldehyde, combustion products, epichlorohydrin, acid mists, auramine and other benzidine-based dyes, and a range of chlorinated organic compounds (<xref rid="b17-ehp0114-001007" ref-type="bibr">Kauppinen et al. 1997</xref>, <xref rid="b16-ehp0114-001007" ref-type="bibr">2002</xref>). The patterns of exposure in the industry are complicated because of the range of different processes that have been used over time in the various stages of pulp and paper manufacture, which together with the relatively small numbers of workers within specific departments has limited the power of epidemiologic studies of mill-based cohorts. A number of studies, nevertheless, have suggested increased risks of gastrointestinal cancers (<xref rid="b11-ehp0114-001007" ref-type="bibr">Henneberger et al. 1989</xref>; <xref rid="b26-ehp0114-001007" ref-type="bibr">Milham and Demers 1984</xref>), respiratory system cancers (<xref rid="b26-ehp0114-001007" ref-type="bibr">Milham and Demers 1984</xref>; <xref rid="b30-ehp0114-001007" ref-type="bibr">Siemiatycki et al. 1986</xref>; <xref rid="b36-ehp0114-001007" ref-type="bibr">Toren et al. 1991</xref>), and certain lymphatic and hematopoietic neoplasms (<xref rid="b7-ehp0114-001007" ref-type="bibr">Coggon et al. 1997</xref>; <xref rid="b23-ehp0114-001007" ref-type="bibr">Matanoski et al. 1998</xref>) in pulp and paper industry workers. Despite the large number of studies conducted, there is still uncertainty about the exact nature and extent of cancer risks associated with work in this industry (<xref rid="b35-ehp0114-001007" ref-type="bibr">Toren 1996</xref>).</p><p>The International Agency for Research on Cancer (IARC) therefore coordinated an international collaborative cohort study to investigate mortality and cancer incidence in the pulp, paper, paperboard, recycled paper, and paper product industries. This study has combined cohorts from 13 countries, consisting of 98,665 workers (2,110,913 person-years), and included the development of a comprehensive database of exposure measurements for the retrospective assessment of study participants’ exposure (<xref rid="b17-ehp0114-001007" ref-type="bibr">Kauppinen et al. 1997</xref>, <xref rid="b16-ehp0114-001007" ref-type="bibr">2002</xref>). The results for mortality and incidence in selected national cohorts (<xref rid="b10-ehp0114-001007" ref-type="bibr">Fassa et al. 1998</xref>; <xref rid="b12-ehp0114-001007" ref-type="bibr">Henneberger and Lax 1998</xref>; <xref rid="b11-ehp0114-001007" ref-type="bibr">Henneberger et al. 1989</xref>; <xref rid="b14-ehp0114-001007" ref-type="bibr">Jäppinen and Pukkala 1991</xref>; <xref rid="b15-ehp0114-001007" ref-type="bibr">Jäppinen and Tola 1986</xref>; <xref rid="b20-ehp0114-001007" ref-type="bibr">Langseth and Andersen 1999</xref>, <xref rid="b21-ehp0114-001007" ref-type="bibr">2000</xref>; <xref rid="b25-ehp0114-001007" ref-type="bibr">McLean et al. 2002</xref>; <xref rid="b28-ehp0114-001007" ref-type="bibr">Rix et al. 1997</xref>, <xref rid="b27-ehp0114-001007" ref-type="bibr">1998</xref>; <xref rid="b29-ehp0114-001007" ref-type="bibr">Sala-Serra et al. 1996</xref>; <xref rid="b33-ehp0114-001007" ref-type="bibr">Szadkowska-Stanczyk et al. 1997</xref>; <xref rid="b34-ehp0114-001007" ref-type="bibr">Szadkowska-Stanczyk and Szymczak 2001</xref>; <xref rid="b39-ehp0114-001007" ref-type="bibr">Wild et al. 1998</xref>), and for separate analyses of exposure to sulfur dioxide (<xref rid="b22-ehp0114-001007" ref-type="bibr">Lee et al. 2002</xref>) and asbestos (<xref rid="b5-ehp0114-001007" ref-type="bibr">Carel et al. 2002</xref>) in the overall cohort, have been reported.</p><p>Workers in this industry experience exposure to chlorinated organic compounds, both volatile chlorinated hydrocarbons such as trichloroethylene, perchloroethylene, dichloromethane, and trichloromethane, and non-volatile organochlorine compounds such as chlorophenols and their salts [pentachlorophenol (PCP)], polychlorinated biphenyls (PCBs), and polychlorinated dibenzodioxins (PCDDs) or polychlorinated dibenzofurans (PCDFs). IARC has classified the volatile organochlorines trichloroethylene and perchloroethylene as probably carcinogenic to humans (group 2A) on the basis of limited evidence in humans of excess risks of cancer of the liver and biliary tract, non-Hodgkin lymphoma (NHL), and esophageal and cervical cancer, and dichloromethane and trichloromethane as possibly carcinogenic to humans (group 2B) based on sufficient evidence for carcinogenicity in animals (<xref rid="b31-ehp0114-001007" ref-type="bibr">Siemiatycki et al. 2004</xref>). Of the nonvolatile organochlorine compounds, IARC has classified the 2,3,7,8-substituted tetrachlorodibenzo-<italic>para</italic>-dioxins (TCDDs) as carcinogenic to humans (group 1) on the basis of both mechanistic evidence and limited evidence in humans of excess risk of all cancers combined rather than for any specific site cancer; PCBs as probably carcinogenic to humans (group 2A) because of limited evidence of excess cancers of the liver and biliary tract and of lymphatic and hematopoietic tissues; and PCPs and their sodium salts as possibly carcinogenic to humans (group 2B) based on findings of excess risk of soft tissue sarcoma and NHL (<xref rid="b31-ehp0114-001007" ref-type="bibr">Siemiatycki et al. 2004</xref>). The aim of the present study was to investigate the association between cancer mortality and exposure to chlorinated organic compounds in the IARC pulp and paper workers cohort.</p><sec sec-type="materials\|methods"><title>Materials and Methods</title><p>Workers employed for at least 1 year in the pulp and paper industry in 13 countries during the period 1920–1996 were included in the overall IARC study, with cohorts from Denmark, Finland, France, Japan, New Zealand, Norway, Poland, Spain, Sweden, Scotland, and the United States included in this analysis of organochlorine exposure. Brazil and South Africa were excluded from this analysis because of inadequacies in the quality of mortality data. Cohort members were identified from company personnel records, with work histories available for the full period of employment in that company.</p><p>The exposure assessment procedure used for this study has been described in detail elsewhere (<xref rid="b17-ehp0114-001007" ref-type="bibr">Kauppinen et al. 1997</xref>, <xref rid="b16-ehp0114-001007" ref-type="bibr">2002</xref>). Briefly, an international panel of industrial hygiene experts used their professional judgment to estimate exposure over different time periods to 27 main agents at the level of department (but not specific job titles within departments) in each of the mills studied, based on detailed company questionnaires on current and historical raw materials and production processes, and < 31,000 existing (mainly unpublished) occupational exposure measurements. Where sufficient measurement data were available, the expected prevalence and level of exposure to specific agents were quantified, and then depending on the range of estimated exposure levels for each agent, limits for low-, medium-, and high-exposure categories were assigned. For these agents, each worker’s weighted cumulative exposure was then estimated by combining the prevalence, level, and duration of exposure. Where only limited measurement data were available, a qualitative assessment of likely, unlikely, or unknown potential for exposure was made.</p><p>Exposure to volatile organochlorine compounds was defined as inhalatory exposure to chlorinated solvents or other specified compounds (indicator agents being trichloroethylene, perchloroethylene, dichloromethane, and trichloromethane) at a level exceeding the nonoccupational background level. These substances have been used as cleaning and degreasing agents in most departments, with the highest exposures occurring in maintenance, repair, and cleaning operations, and medium exposures in sulfate pulp and pulp bleaching departments. The high-exposure category was defined as workers in those departments in which > 50% were exposed and in which the mean level of exposure over the work year was estimated to exceed 1 ppm for trichloromethane, 2.5 ppm for perchloroethylene, and 5 ppm for dichloromethane and trichloroethylene [all 1/10th of the threshold limit value of the American Conference of Governmental Industrial Hygienists (<xref rid="b1-ehp0114-001007" ref-type="bibr">ACGIH 2005</xref>)].</p><p>Nonvolatile organochlorine exposure was assessed in qualitative terms as ever/never exposed, with exposure defined as potential dermal or inhalatory exposure to PCP or its salts, PCBs, PCDDs, PCDFs, or other non-volatile organochlorine compounds exceeding the nonoccupational long-term background level. PCP has been used to prevent sapstain in softwoods used for pulp, with consequent worker exposure during pulping operations. Exposure to PCBs is most likely to have occurred during maintenance and repair of electrical or hydraulic equipment. Both PCP and PCBs contain PCDDs and PCDFs as contaminants of their manufacture, and PCDDs and PCDFs may also be formed as by-products of the bleaching of pulp with chlorine compounds, resulting in occupational exposure both during bleaching and downstream for those involved in paper production (<xref rid="b19-ehp0114-001007" ref-type="bibr">Krishnan 1990</xref>).</p><p>Workers were followed up for mortality according to procedures specific to each country. The period of follow-up varied among countries, ranging from 12 to 50 years. Details on periods of employment and follow-up are reported in <xref ref-type="table" rid="t1-ehp0114-001007">Table 1</xref>. Causes of death were either abstracted from death certificates or obtained from mortality registries, and coded according to the <italic>International Classification of Diseases</italic>, <italic>9th Revision</italic> (ICD-9) [<xref rid="b37-ehp0114-001007" ref-type="bibr">World Health Organization (WHO) 1975</xref>]. Tabulation of person-years started at the beginning of the observation period or on day 1 of the second year of employment if this occurred after the start of the observation period. Standardized mortality ratios (SMRs) were calculated as the ratio of observed to expected deaths, with expected deaths being computed by multiplying the person-years in each sex-specific, age-specific, and 5-year calendar-period–specific stratum by the national reference rates using the Person Years program (<xref rid="b8-ehp0114-001007" ref-type="bibr">Coleman et al. 1986</xref>). National rates were derived from the WHO Mortality Database (<xref rid="b38-ehp0114-001007" ref-type="bibr">WHO 2001</xref>). Ninety-five percent confidence intervals (CIs) of the SMR were calculated under the assumption that the observed numbers of deaths follow a Poisson distribution. Internal analyses were conducted according to years since first employment (< 18, 18–27, 28–37, > 38 years) and duration of exposure (< 4, 4–10, 11–21, > 22 years), and for the volatile organochlorines also by cumulative exposure (∑ level × duration; < 3, 3–9, 10–29, > 30 ppm-years) and weighted cumulative exposure (∑ prevalence × level × duration: < 1, 1–17, > 18 ppm-years), with cut points for continuous measurements of exposure set at quartiles or tertiles depending on numbers available. Tests for linear trend in SMRs were performed using a method described by <xref rid="b4-ehp0114-001007" ref-type="bibr">Breslow and Day (1987)</xref>.</p><p>Poisson regression analysis was used to examine internal dose–response relations associated with exposure to volatile organochlorines and to explore the effect of potential confounding factors. Rate ratios (RRs) and 95% CIs derived from the analysis were adjusted for country, sex, age, calendar period, and employment status (i.e., whether person-years accumulated while workers were employed in the companies included in the study). The reference group for each RR was the first level of each variable.</p></sec><sec sec-type="results"><title>Results</title><p>The distribution of study participants by exposure status and country is shown in <xref ref-type="table" rid="t1-ehp0114-001007">Table 1</xref>. At the end of follow-up of the overall cohort, 79% of the workers were alive, 18% had died, 2% were lost to follow-up, and 1% had emigrated. Altogether, 60,468 workers (1,347,782 person-years) were classified according to volatile organochlorine exposure status, with 82% of the person-years classified as ever exposed and 17% (i.e., 9,628 workers or 229,434 person-years) as having high exposure. A total of 58,162 workers (1,259,780 person-years) were classified according to nonvolatile organochlorine exposure status, with 45% (i.e., 24,940 workers or 570,135 person-years) classified as ever exposed. There was significant overlap between the two exposure groups, with maintenance workers in particular often experiencing exposure to both volatile and nonvolatile organochlorines.</p><p>Cause-specific mortality for the workers classified according to exposure to volatile organochlorines is shown in <xref ref-type="table" rid="t2-ehp0114-001007">Table 2</xref>. Among exposed workers there was a deficit of all causes of death (9,350 deaths; SMR = 0.91; 95% CI, 0.89–0.93) and of all malignant neoplasms (2,285 deaths; SMR = 0.93; 95% CI, 0.89–0.97). Of the cancers of <italic>a priori</italic> interest, there were reduced SMRs for cancer of the esophagus (45 deaths; SMR = 0.74; 95% CI, 0.54–0.99), liver (33 deaths; SMR = 0.76; 95% CI, 0.53–1.07), and cervix (17 deaths; SMR = 0.99; 95% CI, 0.58–1.59); for neoplasms of lymphatic or hematopoietic tissues (189 deaths; SMR = 0.94; 95% CI, 0.81–1.08); and for NHL (52 deaths; SMR = 0.86; 95% CI, 0.64–1.13). Statistically significant excess mortality from pleural neoplasms was observed in the exposed group (20 deaths; SMR = 2.00; 95% CI, 1.22–3.09), mostly due to the even larger excess observed in the highly exposed subjects (8 deaths; SMR = 3.67; 95% CI, 1.58–7.23) who were maintenance workers also exposed to asbestos. The elevation observed in those never exposed was virtually identical (4 deaths; SMR = 1.91; 95% CI, 0.52–4.90) to the exposed group, possibly also due to asbestos exposure. Other statistically significant findings included excess mortality from cancer of the penis and other male genital organs (7 deaths; SMR = 2.51; 95% CI, 1.01–5.17) in the exposed group, and cancer of other respiratory organs (4 deaths; SMR = 3.84; 95% CI, 1.05–9.84) in the highly exposed group. The results reported here are for the overall cohort including both men and women. In general, there was little difference in the findings between sexes, although the results among women were based on a relatively small number of deaths.</p><p>Cause-specific mortality for the workers classified according to exposure to nonvolatile organochlorines is shown in <xref ref-type="table" rid="t3-ehp0114-001007">Table 3</xref>. All-cause mortality in the exposed workers was below expected (4,622 deaths; SMR = 0.94; 95% CI, 0.91–0.96), as was mortality from all cancer (1,145 deaths; SMR = 0.94; 95% CI, 0.89–1.00). Mortality from the other neoplasms of <italic>a priori</italic> interest was also below expected, including liver cancer (16 deaths; SMR = 0.69; 95% CI, 0.40–1.13), soft tissue sarcoma (4 deaths; SMR = 0.80; 95% CI, 0.22–2.04), lymphatic and hematopoietic tissue neoplasms in general (97 deaths; SMR = 0.99; 95% CI, 0.81–1.21), and NHL in particular (25 deaths; SMR = 0.86; 95% CI, 0.55–1.26). The only neoplasms showing elevated risks were penis and other cancer of male genital organs (5 deaths; SMR = 3.60; 95% CI, 1.17–8.40) and Hodgkin disease (17 deaths; SMR = 1.76; 95% CI, 1.02–2.82), and in both sites this was more than three times the rate observed in the workers never exposed. It is of interest to note, however, that the 3-fold excesses in risk of cancer of the penis and other male genital organs and of Hodgkin disease in workers not exposed to nonvolatile organochlorines are matched by deficits (of a similar magnitude) in risk of pleural and other respiratory cancers in those not exposed.</p><p>No consistent pattern of increasing risk with increasing exposure to either volatile or nonvolatile organochlorines was apparent for any cause of death after stratification by duration of employment or by years since first exposure to both volatile and nonvolatile compounds, or by weighted cumulative exposure to volatile organochlorines (data not shown). The Poisson regression analyses stratified according to weighted cumulative volatile organochlorine exposure (shown in <xref ref-type="table" rid="t4-ehp0114-001007">Table 4</xref>), and adjusted for sex, age, employment status, calendar year, and country, showed a weak (<italic>p</italic> = 0.002) trend of increasing risk of mortality from all cancer combined with increasing weighted cumulative exposure to volatile organochlorines (< 1 ppm-years: RR = 1; 1–17 ppm-years: RR = 1.12; 95% CI, 1.01–1.24; > 18 ppm-years: RR = 1.19; 95% CI, 1.016–1.34). No other site of <italic>a priori</italic> interest showed a statistically significant trend of increasing risk, although nonsignificant increases were suggested for liver cancer and for cancer of the pleura. The risk estimates, and the positive trend, for cancer of the pleura were essentially unchanged after adjustment for either exposure to or high exposure to asbestos.</p></sec><sec sec-type="discussion"><title>Discussion</title><p>In this large multicenter historical cohort study, which examined the risks associated with exposure to both volatile and nonvolatile organochlorines in the pulp and paper industry work environment, we found lower than expected overall mortality and all-cancer mortality rates. Internal comparisons based on duration of exposure and time since first exposure showed no consistent exposure–response trends of risk increasing with either volatile or nonvolatile organochlorines. A weak, but statistically significant, trend of increasing risk of all-cancer mortality with increasing weighted cumulative exposure to volatile organochlorines was observed. This finding is similar to the effect seen in other large cohorts with potential exposure to nonvolatile organochlorines contaminated with TCDD (<xref rid="b18-ehp0114-001007" ref-type="bibr">Kogevinas et al. 1997</xref>) but has not previously been reported for volatile organochlorine exposure.</p><p>As in most historical cohort studies of industrial workers, we found a deficit in overall mortality and cancer mortality in this study compared with rates expected in the national populations. This is common in occupational cohort mortality studies and has been observed in previous studies of pulp and paper workers (<xref rid="b3-ehp0114-001007" ref-type="bibr">Band et al. 2001</xref>; <xref rid="b7-ehp0114-001007" ref-type="bibr">Coggon et al. 1997</xref>; <xref rid="b23-ehp0114-001007" ref-type="bibr">Matanoski et al. 1998</xref>), due to the healthy worker effect that arises because healthy people are more likely to gain employment and to remain in employment (<xref rid="b6-ehp0114-001007" ref-type="bibr">Checkoway et al. 2004</xref>). The healthy worker effect is generally weaker for cancer than for other causes of mortality, as observed in this study.</p><p>The assessment of exposure was based on a pulp and paper industry exposure matrix developed by an expert team of industrial hygienists familiar with the pulp and paper industry, although relatively few quantitative data were available on organochlorine exposure (<xref rid="b17-ehp0114-001007" ref-type="bibr">Kauppinen 1997</xref>). Estimates of organochlorine exposure were therefore based largely on information available from company questionnaires about processes and raw materials used (e.g., time periods when chlorine bleaching was done or when PCBs were used in mill electrical equipment). Work histories were available only at the department level for most of the mills under study, so individual exposure estimates were based on the level and prevalence of exposure in the department worked in rather than on a more specific job title. The inability to take into account heterogeneity of exposure among workers in a department is likely to have resulted in significant nondifferential mis-classification of exposure, resulting in a tendency to underestimate any true elevation of risk associated with exposure. In addition, because the exposure assessment for the non-volatile organochlorines was qualitative, it was possible only to make internal comparisons based on the likelihood of ever being exposed rather than evaluating trends according to cumulative dose.</p><p>As with most historical cohort studies, there is also a lack of information on potential lifestyle confounders such as smoking. However, even for lung cancer, the relatively small differences in smoking status between groups of manual workers are unlikely to account for a relative risk of > 1.5 in studies involving a comparison with national mortality rates (<xref rid="b2-ehp0114-001007" ref-type="bibr">Axelson 1978</xref>), and the confounding effect is even weaker for internal dose–response analyses (<xref rid="b32-ehp0114-001007" ref-type="bibr">Siemiatycki et al. 1988</xref>). It is therefore unlikely that there is serious confounding by lifestyle factors in the present study, even regarding the findings for pleural cancer and other respiratory cancers. It is possible that the weak but statistically significant exposure–response relationship for all cancers associated with weighted cumulative exposure to volatile compounds could be due to confounding by lifestyle factors or to the exposure of many maintenance workers to other carcinogens, including asbestos.</p><p>Overall, we found little evidence of any increased risk of cancer mortality in pulp and paper workers exposed to organochlorines, apart from the weak but statistically significant exposure–response trend for cumulative exposure to volatile organochlorines. Although there is evidence of such an association for all cancers with exposure to phenoxy herbicides contaminated with TCDD (<xref rid="b18-ehp0114-001007" ref-type="bibr">Kogevinas et al. 1997</xref>), we are not aware of any evidence suggesting such an association for volatile organochlorine exposure of the type that occurs in pulp and paper mills. Although statistically significant, the association was relatively weak, and this finding should therefore be regarded as preliminary and requiring further investigation.</p><p>We found little evidence of increased risks for specific cancer sites that have been previously associated with organochlorine exposure, including cancer of the esophagus, liver, cervix, and NHL for volatile organochlorines, and cancer of the liver, soft tissue sarcoma, lymphatic, and hematopoietic tissue and NHL for nonvolatile organochlorines. Instead, the only sites that showed significant excess risks in those exposed to volatile organochlorines were cancer of the pleura, “other respiratory,” and “penis and other male genital organs,” and for all three sites the risk was higher in those with high exposure. Interestingly, although the excess risk for cancer of penis and other male genital organs was also elevated in those with exposure to non-volatile organochlorines, the risks for cancer of the pleura and other respiratory cancers were elevated only in those never exposed to nonvolatile organochlorines and below expected in those with exposure. For pleural cancer, this may be because many of the group classified as exposed to nonvolatile organochlorines were bleach plant operators, whereas the group with volatile organochlorine exposure were predominantly maintenance workers. So it is possible that these findings could be due to concomitant exposure to asbestos, because the highest exposures to these compounds occurred in maintenance, repair, and cleaning operations, and an excess risk of pleural cancer has already been reported in workers exposed to asbestos in this industry (<xref rid="b5-ehp0114-001007" ref-type="bibr">Carel et al. 2002</xref>). Unfortunately, joint analyses of asbestos and organochlorine exposure were equivocal because most of the pleural and other respiratory cancers occurred in workers exposed to both factors, and there were insufficient numbers to determine whether there were increased risks in workers exposed to nonvolatile organochlorines alone. When the organochlorine findings were adjusted for asbestos exposure (data not shown), the RRs were essentially unchanged. Although the elevated risk observed for both cancers of penis and other male genital organs and other respiratory cancers was statistically significant, in both cases the CIs were wide and the small number of cases precluded further analysis. It is possible, therefore, that these are chance findings. Increased risks of Hodgkin disease, however, have been consistently reported among pulp mill workers (<xref rid="b26-ehp0114-001007" ref-type="bibr">Milham and Demers 1984</xref>; <xref rid="b35-ehp0114-001007" ref-type="bibr">Toren et al. 1996</xref>), woodworkers (<xref rid="b24-ehp0114-001007" ref-type="bibr">McCunney 1999</xref>), and those with exposure to the herbicides 2,4-D (2,4-dichlorophenoxyacetic acid), 2,4,5-T (2,4,5-trichlorophenoxyacetic acid) and its contaminant TCDD, cacodylic acid, and picloram (<xref rid="b9-ehp0114-001007" ref-type="bibr">Dich et al. 1997</xref>; <xref rid="b13-ehp0114-001007" ref-type="bibr">Institute of Medicine 2000</xref>). The statistically significant increase in risk observed among workers with exposure to nonvolatile organochlorines in this cohort appears to be consistent with these earlier findings. Because the survival for NHL is relatively good, an analysis of incidence would have been more informative, but unfortunately cancer incidence data were available only from Denmark, Finland, New Zealand, Norway, and Sweden. The lack of incidence data and the small number of cases precluded any further analysis of this association.</p><p>In summary, there is little evidence that exposure to organochlorines at the levels experienced in the pulp and paper industry causes an increased risk of cancer, apart from a weakly statistically significant association with weighted cumulative volatile organochlorine exposure. There was little evidence of an increased risk for any specific cancer sites, apart from a statistically significant association between nonvolatile organochlorine exposure and Hodgkin disease and cancer of the pleura. The finding for pleural cancer is consistent with evidence that maintenance workers were exposed to asbestos.</p></sec>
Persistent Pesticides in Human Breast Milk and Cryptorchidism	<sec><title>Introduction</title><p>Prenatal exposure to some pesticides can adversely affect male reproductive health in animals. We investigated a possible human association between maternal exposure to 27 organochlorine compounds used as pesticides and cryptorchidism among male children.</p></sec><sec><title>Design</title><p>Within a prospective birth cohort, we performed a case–control study; 62 milk samples from mothers of cryptorchid boys and 68 from mothers of healthy boys were selected. Milk was collected as individual pools between 1 and 3 months postpartum and analyzed for 27 organochlorine pesticides.</p></sec><sec><title>Results</title><p>Eight organochlorine pesticides were measurable in all samples (medians; nanograms per gram lipid) for cases/controls: 1,1-dichloro-2,2-bis(4-chlorophenyl)ethylene (<italic>p,p</italic>′-DDE): 97.3/83.8; β-hexachlorocyclohexane (β-HCH): 13.6/12.3; hexachlorobenzene (HCB): 10.6/8.8; α -endosulfan: 7.0/6.7; oxychlordane: 4.5/4.1; 1,1,1-trichloro-2,2-bis(4-chlorophenyl)ethane (<italic>p,p</italic>′-DDT): 4.6/4.0; dieldrin: 4.1/3.1; <italic>cis</italic>-heptachloroepoxide (<italic>cis</italic>-HE): 2.5/2.2. Five compounds [octachlorostyrene (OCS); pentachlorobenzene, 1,1-dichloro-2,2-bis(4-chlorophenyl)ethane (<italic>p,p</italic>′-DDD); <italic>o,p</italic>′-DDT; mirex] were measurable in most samples (detection rates 90.8–99.2%) but in lower concentrations. For methoxychlor, <italic>cis</italic>-chlordane, pentachloroanisole (PCA), γ -HCH, 1,1-dichloro-2-(2-chlorophenyl)-2,2(4-chlorophenyl)ethane, <italic>trans</italic>-chlordane, α -HCH, and <italic>o,p</italic>′-DDE, both concentrations and detection rates were low (26.5–71.5%). Heptachlor, HCH (δ, ɛ ), aldrin, β-endosulfan and <italic>trans</italic>-heptachloroepoxide were detected at negligible concentrations and low detection rates and were not analyzed further. Seventeen of 21 organochlorine pesticides [<italic>p,p</italic>′-DDT, <italic>p,p</italic>′-DDE, <italic>p,p</italic>′-DDD, <italic>o,p</italic>′-DDT, HCH (α , β, γ ), HCB, PCA, α -endosulfan, <italic>cis</italic>-HE, chlordane (<italic>cis</italic>-, <italic>trans</italic>-) oxychlordane, methoxychlor, OCS, and dieldrin] were measured in higher median concentrations in case milk than in control milk. Apart from <italic>trans</italic>-chlordane (<italic>p</italic> = 0.012), there were no significant differences between cryptorchid and healthy boys for individual chemicals. However, combined statistical analysis of the eight most abundant persistent pesticides showed that pesticide levels in breast milk were significantly higher in boys with cryptorchidism (<italic>p</italic> = 0.032).</p></sec><sec><title>Conclusion</title><p>The association between congenital cryptorchidism and some persistent pesticides in breast milk as a proxy for maternal exposure suggests that testicular descent in the fetus may be adversely affected.</p></sec>	<contrib contrib-type="author"><name><surname>Damgaard</surname><given-names>Ida N.</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001133">1</xref></contrib><contrib contrib-type="author"><name><surname>Skakkebæk</surname><given-names>Niels E.</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001133">1</xref></contrib><contrib contrib-type="author"><name><surname>Toppari</surname><given-names>Jorma</given-names></name><xref ref-type="aff" rid="af2-ehp0114-001133">2</xref></contrib><contrib contrib-type="author"><name><surname>Virtanen</surname><given-names>Helena E.</given-names></name><xref ref-type="aff" rid="af2-ehp0114-001133">2</xref></contrib><contrib contrib-type="author"><name><surname>Shen</surname><given-names>Heqing</given-names></name><xref ref-type="aff" rid="af3-ehp0114-001133">3</xref></contrib><contrib contrib-type="author"><name><surname>Schramm</surname><given-names>Karl-Werner</given-names></name><xref ref-type="aff" rid="af3-ehp0114-001133">3</xref></contrib><contrib contrib-type="author"><name><surname>Petersen</surname><given-names>Jørgen H.</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001133">1</xref><xref ref-type="aff" rid="af4-ehp0114-001133">4</xref></contrib><contrib contrib-type="author"><name><surname>Jensen</surname><given-names>Tina K.</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001133">1</xref></contrib><contrib contrib-type="author"><name><surname>Main</surname><given-names>Katharina M.</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001133">1</xref></contrib>	Environmental Health Perspectives	<p>Studies published during the last decades have indicated that the birth prevalence of cryptorchidism may have increased in some regions (<xref rid="b6-ehp0114-001133" ref-type="bibr">Berkowitz et al. 1995</xref>; <xref rid="b8-ehp0114-001133" ref-type="bibr">Boisen et al. 2004</xref>; <xref rid="b21-ehp0114-001133" ref-type="bibr">John Radcliffe Hospital Cryptorchidism Study Group 1992</xref>; <xref rid="b36-ehp0114-001133" ref-type="bibr">Pierik et al. 2004</xref>; <xref rid="b53-ehp0114-001133" ref-type="bibr">Thong et al. 1998</xref>). Genetic factors may contribute to these findings. However, the short time interval of this increase suggests that environmental factors may be important (<xref rid="b41-ehp0114-001133" ref-type="bibr">Sharpe and Skakkebæk 2003</xref>).</p><p>Several organochlorine pesticides can cause adverse effects in the male reproductive system in animals (<xref rid="b11-ehp0114-001133" ref-type="bibr">Edwards et al. 2006</xref>; <xref rid="b57-ehp0114-001133" ref-type="bibr">Vos et al. 2000</xref>). An increased incidence of cryptorchidism in male panthers has been attributed to endocrine-disrupting chemicals in the environment, such as 1,1-dichloro-2,2-bis(4-chlorophenyl)ethane (<italic>p,p</italic>′-DDE) (<xref rid="b13-ehp0114-001133" ref-type="bibr">Facemire et al. 1995</xref>). <italic>In utero</italic> exposure of male rats and rabbits to dichlorodiphenyldichloroethylene (DDT) resulted dose dependently in reduced anogenital distance, hypospadias, cryptorchidism, and epididymal agenesis (<xref rid="b18-ehp0114-001133" ref-type="bibr">Gray et al. 2001</xref>, <xref rid="b17-ehp0114-001133" ref-type="bibr">2004</xref>; <xref rid="b34-ehp0114-001133" ref-type="bibr">Palmer et al. 2000</xref>; <xref rid="b65-ehp0114-001133" ref-type="bibr">Wolf et al. 1999</xref>; <xref rid="b66-ehp0114-001133" ref-type="bibr">You et al. 1998</xref>).</p><p>Increased rates of orchidopexy in areas with extensive pesticide use in agriculture have been reported (<xref rid="b15-ehp0114-001133" ref-type="bibr">Garcia-Rodriguez et al. 1996</xref>). One study also found higher pesticide levels in fat tissue samples from boys operated on for cryptorchidism than in children who were operated on for other reasons (<xref rid="b19-ehp0114-001133" ref-type="bibr">Hosie et al. 2000</xref>).</p><p>The aim of this study was to investigate, in a case–control study nested within a prospective cohort, whether individual <italic>in utero</italic> exposure to organochlorine pesticides estimated by measurements of maternal breast milk concentrations was associated with cryptorchidism among male children.</p><sec sec-type="materials\|methods"><title>Materials and Methods</title><p>We conducted a joint prospective, longitudinal birth cohort study in Finland (Turku University Hospital) and Denmark (National University Hospital at Copenhagen) from 1997 to 2001. In this study we aimed to describe regional prevalence rates and risk factors (lifestyle and exposure) for cryptorchidism by means of questionnaires and biologic samples (blood samples of mother and child, placentas, and one breast milk sample per child). The study was prospectively planned by both research groups as a joint venture in 1996. Recruitment and examinations were completely standardized. The cohort (antenatal recruitment and inclusion, inclusion criteria, and clinical examinations) has been described previously in detail (<xref rid="b8-ehp0114-001133" ref-type="bibr">Boisen et al. 2004</xref>). All boys in the cohort were examined by a group of specially trained doctors for signs of cryptorchidism. The examination technique and definition of cryptorchidism developed by Scorer were applied (<xref rid="b40-ehp0114-001133" ref-type="bibr">Scorer 1964</xref>). Standardization of the clinical procedures was achieved by repetitive workshops, and borderline cases were examined by two researchers from the national study groups. All boys were examined shortly after birth and again at 3 months of age. Boys born prematurely were examined at the expected date of delivery.</p><p>From the total biobank of breast milk samples, we included 65 samples from each country for organochlorine pesticide measurements. The number of samples was determined by funding. The samples represent 29 Danish and 33 Finnish cases defined as boys with cryptorchidism (unilateral or bilateral) at birth. Four Danish and 25 Finnish boys were still cryptorchid at 3 months of age, whereas the others (25/8) had spontaneous descent. Danish/Finnish controls (36/32), defined as boys without cryptorchidism at birth or 3 months, were included. In Denmark, the controls were selected randomly from the entire birth cohort of healthy boys. In Finland, the boys were selected prospectively by a case–control design in which the boys with cryptorchidism were matched to controls at birth for maternal parity, smoking (yes/no), diabetes (yes/no), gestational age (± 7 days), and date of birth (± 14 days). This design was chosen in Finland because of lack of sufficient funding to collect and store biologic samples from all. To ensure that all prospectively planned chemical analyses could be performed, only breast milk samples with a volume > 125 mL were included.</p><p>The study was conducted according to the Helsinki II Declaration (<xref rid="b64-ehp0114-001133" ref-type="bibr">World Medical Association 2004</xref>) and was approved by the local ethical committees in both countries (Finland: 7/1996, Denmark: KF01-030/97) and the Danish Data Protection Agency (registration no. 1997-1200-074). The families were included after oral and written informed consent had been obtained from the parents.</p><p>Human breast milk samples were collected from 1 to 3 months postpartum in Denmark and from 1 to 2 months postpartum in Finland as successive aliquots. All mothers were given oral and written instructions to feed the baby first and then sample milk aliquots (hind milk) by manual expression into a glass or porcelain container, avoiding the use of mechanical breast pumps. The aliquots were frozen consecutively in a glass bottle (250-mL Pyrex glass bottle with a Teflon cap (1515/06D, Bibby Sterilin, Staffordshire, England) and stored in household freezers. The samples were delivered frozen to the hospital at the 3 months’ examination and stored at –20°C.</p><p>Exposure measurements in biologic samples from boys with congenital cryptorchidism at birth and controls were prospectively planned to include persistent and nonpersistent chemicals. Twenty-seven organochlorine compounds were selected by the following criteria: previous worldwide use, suspicion of endocrine-disrupting activity from animal and/or <italic>in vitro</italic> studies, and highly sensitive analytic methods available. As part of other substudies, the same breast milk samples were planned to be analyzed for other compounds with suspected endocrine-disrupting activity.</p><p>The selected breast milk samples were thawed at room temperature for 12 hr, heated, and shaken for 30 min at 37°C to homogenize the samples, and then divided into smaller aliquots and refrozen at –20°C until further chemical analysis. Extraction, cleanup, and analysis of organochlorine pesticides in the milk samples were based on a method that has been described in detail elsewhere (<xref rid="b43-ehp0114-001133" ref-type="bibr">Shen et al. 2005</xref>). Milk samples (10 mL) were extracted with 250 mL of a mixture consisting of acetone and <italic>n</italic>-hexane (2:1 v/v) (<xref rid="b5-ehp0114-001133" ref-type="bibr">Beek 2000</xref>). The milk extracts were collected in flasks weighed in advance and evaporated using a rotary vacuum evaporator (water bath at 45°C). After evaporation, the flasks were placed into an exicator until stable weight was achieved. The lipid content was calculated on wet weight basis. The residual was dissolved in toluene, and gel permeation-chromatography followed by sandwich cartridge cleanup was used to remove lipids and other interferences from the extract. Finally, the organochlorine pesticides were measured by high-resolution gas chromatography/high-resolution mass spectrometry quantified by an isotope dilution method.</p></sec><sec><title>Statistical Analysis</title><p>Descriptive data of the mothers and the boys (anthropometric measurements) are reported as mean ± SD or number (percentage) (<xref ref-type="table" rid="t1-ehp0114-001133">Table 1</xref>). We tested differences between boys with and without cryptorchidism by unpaired <italic>t</italic>-tests or chi-square tests. Descriptive statistics of pesticide concentrations are given as medians and ranges (minimum and maximum) because of skewed distributions. The sum of DDT metabolites was calculated as the sum of all six metabolites. We calculated the DDE(dichloro-diphenyldichloroethylene)/DDT ratio and the enantiomeric median ratio (ER) by simple division: <italic>p,p</italic>′-DDE/1,1,1-trichloro-2,2-bis(4-chlorophenyl)ethane (<italic>p,p</italic>′-DDT) and (+)-isomer concentration/(–)-isomer concentration for the individual compounds, respectively. We tested the differences between cryptorchid and healthy boys using the Mann-Whitney <italic>U</italic>-test. We tested the differences in pesticide levels between cryptorchid and healthy boys using a logistic regression model in which the pesticide level (log-transformed) and, in some analyses, country and other potential confounders were entered as covariates. This approach had the advantage of allowing the inclusion of cases and controls based on their pesticide levels in the analysis. This selection does not introduce bias in the estimation. <italic>p</italic>-Values were not corrected for multiple testing.</p><p>To determine whether a given measurement of a sample indicated the presence of a pesticide, both the limit of detection (LOD) and the limit of quantification (LOQ) were determined for every sample. We defined the LOD as three times the background noise of the analytic instrument. Samples with values below the LOD were nondetectable. Samples above LOD, in which the pesticide concentration could not be reliably quantified, were assigned a value below the LOQ; the LOQ represents the level at which a compound concentration was determined with sufficient precision. We defined the LOQ as three times the value of a blank sample, which was the blank matrix used in sample preparation. Detection rate was defined as percentage of samples with a detectable and quantifiable value. Because statistical handling of measurements below the LOD or below the LOQ may influence results, we tried different selection schemes to test whether conclusions were sensitive to the actual values of the unquantifiable measurements. We performed three analyses to compare the exposures of cryptorchid and normal boys. The first included all data using the LOD for samples with nondetectable values and the detected value for unquantified samples. The second included all data except for samples with values below the LOQ. The third analysis excluded all nondetectable samples. Exposure levels reported in <xref ref-type="table" rid="t2-ehp0114-001133">Table 2</xref> are based on the third analysis excluding samples with values below the LOD or below the LOQ. Exposure patterns between cases and controls using the other two statistical approaches did not substantially differ from those in <xref ref-type="table" rid="t2-ehp0114-001133">Table 2</xref> (data not shown).</p><p>To investigate a combined effect of persistent pesticides, we used eight pesticides for which all individuals had measurable and quantifiable levels [<italic>p,p</italic>′-DDE, <italic>p,p</italic>′-DDT, β-hexachlorocyclohexane (HCH), hexa-chlorobenzene (HCB), α -endosulfan, <italic>cis</italic>-heptachloroepoxide (<italic>cis</italic>-HE), oxychlordane, and dieldrin]. A statistical test of the null hypothesis that there were no differences in the median exposure levels of cases and controls was carried out as a Monte Carlo permutation test. In the permutations, cases and controls were randomly assigned exposure profiles from the observed profiles, and median levels between cases and controls were compared within countries. We then compared the observed test statistic (the number of median exposures, which were higher among cases than controls) with the permutation distribution. The permutation scheme of the test takes into account the within-individual association structure between different persistent organochlorine pesticides—that some individuals tend to have high exposure to many pesticides, whereas others have low exposures.</p></sec><sec sec-type="results"><title>Results</title><p>Study population characteristics are given in <xref ref-type="table" rid="t1-ehp0114-001133">Table 1</xref>. We found no significant differences between the mothers giving birth to a cryptorchid boy versus a healthy boy. Gestational age of cryptorchid boys was slightly lower than in normal boys (Denmark: <italic>p</italic> = 0.029; Finland: <italic>p</italic> = 0.688). No systematic difference between cases and controls with regard to year of birth was observed (<italic>p</italic> = 0.543). In both countries, the selected controls did not differ from the healthy mothers and boys in the entire cohorts with respect to maternal age, parity, smoking, gestational age, and birth weight (data not shown).</p><p><xref ref-type="table" rid="t2-ehp0114-001133">Table 2</xref> shows the results of pesticide measurements in breast milk samples of mothers with cryptorchid and normal boys. The lipid content (percent weight per weight) did not differ significantly (<italic>p</italic> = 0.707) between cases [3.7 (range, 1.1–7.9)] and controls [3.8 (range, 0.4–10.1)]. Concentrations of persistent pesticides (nanograms per gram lipid) in breast milk showed large interindividual variations and skewed distributions as well as differences in absolute levels.</p><p>Eight compounds—<italic>p,p</italic>′-DDE, β-HCH, HCB, α -endosulfan, oxychlordane, <italic>p,p</italic>′-DDT, dieldrin, and <italic>cis</italic>-HE (listed with decreasing concentrations)—were quantifiable in all samples. The concentrations of these eight compounds were higher than those of any of the remaining 19 compounds, which were all measured in very low concentrations. The sum of all DDT metabolites was slightly higher for cases than controls, but the difference did not reach statistical significance. The ratio between <italic>p,p</italic>′-DDE and <italic>p,p</italic>′-DDT was higher among controls than cases, but not significantly. Seventeen of 21 organochlorine pesticides [<italic>p,p</italic>′-DDT; <italic>p,p</italic>′-DDE; 1,1-dichloro-2,2-bis(4-chlorophenyl)ethane (<italic>p,p</italic>′-DDD); 1,1,1-trichloro-2-(2-chlorophenyl)-2-(4-chloropheny l)ethane (<italic>o,p</italic>′-DDT); HCH (α , β, γ ); HCB; pentachloroanisole (PCA); α -endosulfan; <italic>cis</italic>-HE; chlordane (<italic>cis</italic>-, <italic>trans</italic>-); oxychlordane; methoxychlor; octachlorostyrene (OCS); and dieldrin] were measured in slightly higher median concentrations in milk from mothers giving birth to cryptorchid boys than in mothers giving birth to healthy boys with only <italic>trans</italic>-chlordane reaching statistical significance (<italic>p</italic> = 0.012) (<xref ref-type="table" rid="t2-ehp0114-001133">Table 2</xref>).</p><p><xref ref-type="table" rid="t3-ehp0114-001133">Table 3</xref> presents the overall exposure pattern of cryptorchid and healthy boys depending on inclusion of samples with values below the LOD and/or LOQ. The overall exposure pattern in the three analyses was comparable, showing that most compounds were measured in higher concentrations in milk from case mothers than in milk from control mothers.</p><p>A combined analysis (a Monte Carlo permutation test) of the eight most prevalent organochlorine pesticides revealed that the difference between milk from case mothers and control mothers was not likely due to chance (<italic>p</italic> = 0.032).</p><p>OCS, pentachlorobenzene (PeCB), <italic>p,p</italic>′-DDD, <italic>o,p</italic>′-DDT, and mirex were detected in almost all samples (detection rates 90.8–99.2%), but at low concentrations. Aldrin, β-endosulfan, <italic>trans</italic>-heptachloroepoxide, and ɛ -HCH were not detected in any sample. Heptachlor and δ-HCH were measured at very low concentrations in three and five samples, respectively. None of these six pesticides are included in <xref ref-type="table" rid="t2-ehp0114-001133">Table 2</xref>. For the remaining eight compounds [methoxychlor, <italic>cis</italic>-chlordane, PCA, γ -HCH, 1,1-dichloro-2-(2-chlorophenyl)-2,2(4-chlorophenyl)ethane (<italic>o,p</italic>′-DDD), <italic>trans-</italic>chlordane, α -HCH, 1,1-dichloro-2-(2-chlorophenyl)-2-(4-chlorophenyl)ethylene (<italic>o,p</italic>′-DDE)], the total detection rate varied between 26.9 and 71.5%. Detection rates did not differ significantly between cases and controls for single pesticides (data not shown).</p><p>For oxychlordane and <italic>cis</italic>-HE, the enantiomeric concentrations were available. The absolute concentrations of the enantiomeric isomers were higher, but not significantly, for cryptorchid boys than for controls (data not shown). The ER for oxychlordane (cases/controls) was 1.36/1.28. For <italic>cis</italic>-HE, the corresponding figures were 2.48/2.19 (<italic>p</italic> = 0.103 and <italic>p</italic> = 0.467, respectively).</p></sec><sec sec-type="discussion"><title>Discussion</title><p>Most persistent organochlorine pesticides were found in higher concentrations in boys with cryptorchidism than in controls, although no individual compound was significantly correlated with cryptorchidism. For eight chemicals (<italic>p,p</italic>′-DDE, <italic>p,p</italic>′-DDT, β-HCH, HCB, α -endosulfan, <italic>cis</italic>-HE, oxychlordane, and dieldrin), which were measurable in all samples, the differences between cases and controls were unlikely to be due to chance. Although we cannot exclude the possibility that individual chemicals alone may cause cryptorchidism, our study suggests that exposure to more than one chemical at low concentrations represents a risk factor for congenital cryptorchidism. There may also be simultaneous coexposure to other environmental chemicals that contribute to the effect on testicular descent; the pesticide measurements may represent a proxy marker (sentinel) for other exposures, such as brominated flame retardants.</p><p>We were able to quantify most organochlorine pesticides at low levels in breast-milk samples. This indicates that these chemicals are still relevant despite the fact that most have been banned or restricted in the study areas for many years. High DDE/DDT levels support the assumption the current exposure level primarily originates from previous contamination, environmental persistence, and long-range atmospheric dissipation, and less from imported food products and increased traveling activity to areas with ongoing use of pesticides such as DDT for malaria control.</p><p>Levels of the investigated pesticides have generally declined in the study area, whereas the birth prevalence of cryptorchidism appears to have increased (<xref rid="b8-ehp0114-001133" ref-type="bibr">Boisen et al. 2004</xref>). This may explain why no single compound was strongly correlated with cryptorchidism. However, low concentrations of a mixture of chemicals over time may still be harmful to the fetus. In addition, the overall exposure to other chemicals with endocrine-disrupting activity may have increased in the same period. Thus, the persistent pesticides investigated here may reflect current overall exposure: Women with the highest levels of persistent pesticides may also be the ones with the highest concentrations of other endocrine-disrupting chemicals. This hypothesis is supported by a previous study that found the concentrations of HCB, <italic>p,p</italic>′-DDE, <italic>p,p</italic>′-DDT, and β-HCH in individual samples increased with increasing poly-chlorinated biphenyl concentrations (<xref rid="b2-ehp0114-001133" ref-type="bibr">Andersen and Orbæk 1984</xref>).</p><p>To our knowledge, no previous studies have compared levels of persistent organochlo-rine pesticides in breast milk with the birth prevalence of cryptorchidism. Breast milk was chosen as a surrogate biomarker of previous maternal exposure to persistent pesticides because these compounds accumulate in lipid-rich tissue and thereby in breast milk (<xref rid="b5-ehp0114-001133" ref-type="bibr">Beek 2000</xref>; <xref rid="b20-ehp0114-001133" ref-type="bibr">Jensen and Slorach 1991</xref>). There is a dynamic equilibrium between levels of persistent compounds in maternal adipose tissue and breast milk (<xref rid="b10-ehp0114-001133" ref-type="bibr">Cerrillo et al. 2005</xref>; <xref rid="b22-ehp0114-001133" ref-type="bibr">Kanja et al. 1992</xref>; <xref rid="b38-ehp0114-001133" ref-type="bibr">Rogan et al. 1986</xref>; <xref rid="b47-ehp0114-001133" ref-type="bibr">Skaare et al. 1988</xref>; <xref rid="b61-ehp0114-001133" ref-type="bibr">Waliszewski et al. 2001</xref>); therefore, daily intake of pesticides during lactation has little influence on the levels measured in milk. Both human and animal studies have demonstrated that pesticides during pregnancy can be transferred to the fetus by crossing the placenta (<xref rid="b14-ehp0114-001133" ref-type="bibr">Foster et al. 2000</xref>; <xref rid="b26-ehp0114-001133" ref-type="bibr">Lange et al. 2002</xref>; <xref rid="b60-ehp0114-001133" ref-type="bibr">Waliszewski et al. 2000</xref>). Levels measured in breast milk were positively correlated to levels measured in umbilical cord samples (<xref rid="b10-ehp0114-001133" ref-type="bibr">Cerrillo et al. 2005</xref>; <xref rid="b22-ehp0114-001133" ref-type="bibr">Kanja et al. 1992</xref>; <xref rid="b47-ehp0114-001133" ref-type="bibr">Skaare et al. 1988</xref>; <xref rid="b61-ehp0114-001133" ref-type="bibr">Waliszewski et al. 2001</xref>). Concentrations of compounds in breast milk are a suitable proxy for fetal exposure during pregnancy. As persistent pesticides are accumulated in the lipid fraction of the breast milk, any variations in the lipid content may affect the levels measured. In our study, the women were carefully instructed to collect only hind milk, and they collected many small aliquots that were pooled over time. Thus, the breast milk samples in this study represent an average content over a long period. Because we found no differences in mean lipid content between cases and controls, the potential bias induced by lipid variation due to collection is negligible. In both countries, the selected controls did not differ from the healthy mothers and boys in the entire cohorts with respect to maternal age, parity, smoking, gestational age, and birth weight, and therefore we believe that the samples are representative.</p><p>Because of matching for the most common confounders, such as parity, in the Finnish population, there were no significant differences between cases and controls. In the Danish group, only gestational age differed slightly (<italic>p</italic> = 0.029). Primiparae, slim women, and smokers tend to have higher pesticide levels (<xref rid="b20-ehp0114-001133" ref-type="bibr">Jensen and Slorach 1991</xref>). As more mothers of healthy boys were primiparae and nonsmokers and had lower mean body mass index (BMI) than mothers of cryptorchid boys, our study may underestimate the effect of organochlorine pesticide exposure on cryptorchidism. No definitive relation between maternal age and the level of organochlorine pesticides in breast milk has been reported; some studies have found that older mothers have higher levels, whereas others have not (<xref rid="b20-ehp0114-001133" ref-type="bibr">Jensen and Slorach 1991</xref>). In our study, case mothers were slightly older than control mothers, but fewer of them were primiparae. Older mothers had higher concentrations [significant difference in three compounds (<italic>p,p</italic>′-DDE, oxychlordane, and α -endosulfan)]. However, older primiparae mothers [<italic>n</italic> = 7:3 cases (4.8%); 4 controls (5.9%)] especially contributed to this difference. Because they were equally distributed in the two groups, this cannot explain the difference between cases and controls. Milk from mothers giving birth to premature babies may be different in composition. One Danish study described higher levels of HCB in milk from mothers giving birth to premature infants (<xref rid="b20-ehp0114-001133" ref-type="bibr">Jensen and Slorach 1991</xref>). In our study, levels were higher in milk from mothers giving birth to premature infants. The differences were not significant, and the number of premature infants was low (five cases and two controls) and cannot therefore explain the difference we found.</p><p>The sensitivity of the analytic method allowed the detection of traces of organochlorine pesticides below the LOQ. This phenomenon was, as expected, most frequent for pesticides detected in generally low concentrations. Because the statistical handling of these measurements can profoundly change the results, we evaluated our data carefully with different approaches and found that the overall findings remained unchanged.</p><p>Few other studies have investigated the possible relationship between persistent pesticide levels in biologic samples and the prevalence of cryptorchidism. <xref rid="b19-ehp0114-001133" ref-type="bibr">Hosie et al. (2000)</xref> compared levels of pesticides [DDT and metabolites, toxaphene, HCH (α , β, γ ), HCB, PCA, PeCB, and several chlorinated cyclodienes such as heptachlor] in fat biopsies from 18 cryptorchid boys and 30 controls. Their findings were comparable with ours: Pesticide concentrations were higher among cases than controls, but reaching significance for only a few (HCB and heptachloroepoxide). However, the relatively limited total number of samples and especially the broad age range of boys (0.1–15 years) limits the interpretation of the data. For some participants, biomonitoring was conducted a long time after the relevant prenatal exposure window for testicular maldescent.</p><p>Two studies have been published comparing levels of DDE and DDT in maternal blood and cryptorchidism and hypospadias in the offspring based on two large birth cohorts conducted in the United States in the 1960s (<xref rid="b7-ehp0114-001133" ref-type="bibr">Bhatia et al. 2005</xref>; <xref rid="b28-ehp0114-001133" ref-type="bibr">Longnecker et al. 2002</xref>). Although both studies were based on biologic samples collected in a period during which DDT was still being used, neither study found firm associations. <xref rid="b7-ehp0114-001133" ref-type="bibr">Bhatia et al. (2005)</xref> did not find any association. <xref rid="b28-ehp0114-001133" ref-type="bibr">Longnecker et al. (2002)</xref> found adjusted odd ratios to be elevated, although not significantly, and concluded that the results were consistent with a modest to moderate association between DDE/DDT and cryptorchidism. Both studies were performed as nested case–control studies within large prospective birth cohort studies; the women were recruited during pregnancy and the diagnosis of cryptorchidism was well ascertained. Although these studies applied a different biologic matrix (blood) than ours (breast milk), the concentrations in ours were lower. This would also be expected, as the samples in both studies were collected in the period from 1959 to 1966, when DDT was still permitted.</p><p>Other studies without access to biologic samples have also reported associations between cryptorchidism and parental pesticide exposure (<xref rid="b15-ehp0114-001133" ref-type="bibr">Garcia-Rodriguez et al. 1996</xref>; <xref rid="b16-ehp0114-001133" ref-type="bibr">Garry et al. 1996</xref>; <xref rid="b25-ehp0114-001133" ref-type="bibr">Kristensen et al. 1997</xref>; <xref rid="b36-ehp0114-001133" ref-type="bibr">Pierik et al. 2004</xref>; <xref rid="b37-ehp0114-001133" ref-type="bibr">Restrepo et al. 1990</xref>; <xref rid="b62-ehp0114-001133" ref-type="bibr">Weidner et al. 1998</xref>). Most of these studies were register-based and retrospective. One was performed as a nested case–control study in a large birth cohort study comparing parental exposure (registered by interview of the parents after birth) among boys with cryptorchidism and hypospadias with the exposure of parents of normal boys (<xref rid="b36-ehp0114-001133" ref-type="bibr">Pierik et al. 2004</xref>). <xref rid="b15-ehp0114-001133" ref-type="bibr">Garcia-Rodriguez et al. (1996)</xref> conducted an ecologic study in which orchidopexy rates in different municipalities in Granada, Spain, were compared. Pesticide use was categorized into four groups; areas with high pesticide exposure had higher orchidopexy rates. Generally, the exposure assessment was indirect and based on job titles, pesticide purchase, or self-reported exposure with the possibility of recall bias in case–control studies. Distinction between maternal and paternal exposure is made only in some studies: <xref rid="b36-ehp0114-001133" ref-type="bibr">Pierik et al. (2004)</xref> found that cryptorchidism was significantly associated with paternal pesticide exposure but not maternal. <xref rid="b62-ehp0114-001133" ref-type="bibr">Weidner et al. (1998)</xref> described a significantly increased risk of cryptorchidism in sons of female gardeners. Finally, <xref rid="b37-ehp0114-001133" ref-type="bibr">Restrepo et al. (1990)</xref> found increased risk with maternal pesticide exposure during pregnancy (relative risk = 4.6); however, this did not reach significance. Because most of these studies have been conducted recently, the pesticide exposure may reflect newer, nonpersistent pesticides. In addition, register data of cryptorchidism are less reliable because of variation in ascertainment and reporting (<xref rid="b54-ehp0114-001133" ref-type="bibr">Toppari et al. 2001</xref>).</p><p>We found a large interindividual difference in concentrations of pesticides in breast milk. This was most likely related to individual differences in exposure and metabolism. Although no significant difference in factors such as lipid content in breast milk, maternal age, maternal BMI, parity, and smoking habits were found between the mothers of cryptorchid and healthy boys, these factors may have contributed to some of the interindividual differences (<xref rid="b20-ehp0114-001133" ref-type="bibr">Jensen and Slorach 1991</xref>; <xref rid="b46-ehp0114-001133" ref-type="bibr">Sim and McNeil 1992</xref>).</p><p>Compared with previous reports on Danish and Finnish breast milk, the average levels of persistent pesticides found in our study were low (<xref rid="b9-ehp0114-001133" ref-type="bibr">Bro-Rasmussen et al. 1968</xref>; <xref rid="b20-ehp0114-001133" ref-type="bibr">Jensen and Slorach 1991</xref>; <xref rid="b31-ehp0114-001133" ref-type="bibr">Mussalo-Rauhamaa et al. 1988</xref>; <xref rid="b52-ehp0114-001133" ref-type="bibr">Sundhedsstyrelsen 1999</xref>; <xref rid="b58-ehp0114-001133" ref-type="bibr">Vuori et al. 1977</xref>; <xref rid="b63-ehp0114-001133" ref-type="bibr">Wickstrom et al. 1983</xref>). This finding agrees with other reports on declining concentration of some persistent chemicals (<xref rid="b20-ehp0114-001133" ref-type="bibr">Jensen and Slorach 1991</xref>; <xref rid="b33-ehp0114-001133" ref-type="bibr">Noren and Meironyte 2000</xref>; <xref rid="b39-ehp0114-001133" ref-type="bibr">Schade and Heinzow 1998</xref>; <xref rid="b48-ehp0114-001133" ref-type="bibr">Solomon and Weiss 2002</xref>). The ERs indicated to what extent the chiral persistent pollutants were metabolized in the mother’s body after the exposure (<xref rid="b44-ehp0114-001133" ref-type="bibr">Shen et al. 2006</xref>), for example, because (–)-<italic>cis</italic>-HE is more readily degraded than (+)-<italic>cis</italic>-HE in the human body; and if the two samples have the same levels of <italic>cis</italic>-HE with different ERs, the higher ER sample could have been exposed more heavily in the past. It is an interesting result that the case samples have higher levels of <italic>cis</italic>-HE with higher ERs (medians). This means, considering the already metabolized part, that case mothers should have been exposed to <italic>cis</italic>-HE more heavily than control mothers in their exposure histories.</p><p>The most recent data (1993–1994) from the Danish National Board of Health (DNBH) of breast milk samples from 86 primiparae 25–29 years of age showed median concentrations of 178, 38, 33, and 8 ng/g lipid for <italic>p,p</italic>′-DDE, β-HCH, HCB, and dieldrin, respectively (<xref rid="b52-ehp0114-001133" ref-type="bibr">Sundhedsstyrelsen 1999</xref>). In our study, the corresponding figures were 135, 16, 12, and 5 ng/g lipid for primiparae of the same age. DNBH estimated that the average daily intake of <italic>p,p</italic>′-DDE by the child was 1.1 μg/kg/day; in our study, the corresponding value was 0.7 μg/kg/day. This value is still above the acceptable daily intake/tolerable daily intake (ADI/TDI) suggested in the same report (0.5 μg/kg/day) (<xref rid="b52-ehp0114-001133" ref-type="bibr">Sundhedsstyrelsen 1999</xref>). For μ-HCH, HCB, and dieldrin, estimated average daily intake in our study (0.09, 0.07, 0.03 μg/kg/day) was below the recommended ADI/TDI (0.6, 0.17, and 0.05 μg/kg/day).</p><p>In contrast to a Finnish study from 1988 in which mirex was not detected in any of the milk samples (<xref rid="b31-ehp0114-001133" ref-type="bibr">Mussalo-Rauhamaa et al. 1988</xref>), we were able to measure mirex in all samples, although in low concentrations. This difference could be due to our more sensitive analytic methods. To our knowledge, data on endosulfan, PCA, and PeCB in Danish and Finnish milk samples have not been published before. The levels were low compared with studies from other industrialized countries (<xref rid="b10-ehp0114-001133" ref-type="bibr">Cerrillo et al. 2005</xref>; <xref rid="b30-ehp0114-001133" ref-type="bibr">Mes et al. 1993</xref>; <xref rid="b32-ehp0114-001133" ref-type="bibr">Newsome and Ryan 1999</xref>; <xref rid="b33-ehp0114-001133" ref-type="bibr">Noren and Meironyte 2000</xref>).</p><p>In this study we focused on pesticides with suspected endocrine-disrupting activity or pesticides that have been used worldwide (<xref rid="b12-ehp0114-001133" ref-type="bibr">European Commission Directorate General Environment 2000</xref>; <xref rid="b55-ehp0114-001133" ref-type="bibr">Toppari et al. 1996</xref>). The relative potency of each pesticide is often much lower than that of natural hormones. However, several of the included compounds act as both estrogens and antiandrogens (<xref rid="b1-ehp0114-001133" ref-type="bibr">Andersen et al. 2002</xref>; <xref rid="b23-ehp0114-001133" ref-type="bibr">Kelce et al. 1995</xref>), which might increase their possible adverse effect on testicular descent. Furthermore, different compounds may interact, thereby enhancing their effects (<xref rid="b1-ehp0114-001133" ref-type="bibr">Andersen et al. 2002</xref>; <xref rid="b24-ehp0114-001133" ref-type="bibr">Kortenkamp and Altenburger 1998</xref>; <xref rid="b27-ehp0114-001133" ref-type="bibr">LeBlanc et al. 1997</xref>; <xref rid="b45-ehp0114-001133" ref-type="bibr">Silva et al. 2002</xref>). Existing data on possible mixture effects of the specific organochlorine pesticides <italic>in vitro</italic> are limited. A combination of 10 compounds including endosulfan<italic>,</italic> dieldrin, methoxychlor, and some DDT metabolites demonstrated a cumulative effect (<xref rid="b49-ehp0114-001133" ref-type="bibr">Soto et al. 1994</xref>, <xref rid="b50-ehp0114-001133" ref-type="bibr">1995</xref>). Similarly, a number of studies have found indications of additivity for some of the pesticides included in our study (<xref rid="b3-ehp0114-001133" ref-type="bibr">Arnold et al. 1997</xref>; <xref rid="b29-ehp0114-001133" ref-type="bibr">Merritt et al. 1999</xref>; <xref rid="b35-ehp0114-001133" ref-type="bibr">Payne et al. 2001</xref>; <xref rid="b42-ehp0114-001133" ref-type="bibr">Shekhar et al. 1997</xref>; <xref rid="b51-ehp0114-001133" ref-type="bibr">Sumpter and Jobling 1995</xref>; <xref rid="b56-ehp0114-001133" ref-type="bibr">Vonier et al. 1996</xref>). Others have not been able to demonstrate additivity between dieldrin and endosulfan (<xref rid="b4-ehp0114-001133" ref-type="bibr">Ashby et al. 1997</xref>; <xref rid="b59-ehp0114-001133" ref-type="bibr">Wade et al. 1997</xref>) or methoxychlor (<xref rid="b4-ehp0114-001133" ref-type="bibr">Ashby et al. 1997</xref>). Investigating the possible effects of mixtures is complicated because mechanisms of actions for individual compounds are often poorly known, and some chemicals may act through different routes depending on dosage.</p><p>In conclusion, our study suggests an association between congenital cryptorchidism and some persistent organochlorine pesticides present in mothers’ breast milk. Although our study cannot provide proof for a causal relationship, our data are in line with results from animal studies. Thus, prenatal exposure to persistent organochlorine pesticides may adversely affect testicular descent in boys.</p></sec>
Ozone and PM<sub>2.5</sub> Exposure and Acute Pulmonary Health Effects: A Study of Hikers in the Great Smoky Mountains National Park	<p>To address the lack of research on the pulmonary health effects of ozone and fine particulate matter (≤ 2.5 μm in aerodynamic diameter; PM<sub>2.5</sub>) on individuals who recreate in the Great Smoky Mountains National Park (USA) and to replicate a study performed at Mt. Washington, New Hampshire (USA), we conducted an observational study of adult (18–82 years of age) day hikers of the Charlies Bunion trail during 71 days of fall 2002 and summer 2003. Volunteer hikers performed pre- and posthike pulmonary function tests (spirometry), and we continuously monitored ambient O<sub>3</sub>, PM<sub>2.5</sub>, temperature, and relative humidity at the trailhead. Of the 817 hikers who participated, 354 (43%) met inclusion criteria (nonsmokers and no use of bronchodilators within 48 hr) and gave acceptable and reproducible spirometry. For these 354 hikers, we calculated the posthike percentage change in forced vital capacity (FVC), forced expiratory volume in 1 sec (FEV<sub>1</sub>), FVC/FEV<sub>1</sub>, peak expiratory flow, and mean flow rate between 25 and 75% of the FVC and regressed each separately against pollutant (O<sub>3</sub> or PM<sub>2.5</sub>) concentration, adjusting for age, sex, hours hiked, smoking status (former vs. never), history of asthma or wheeze symptoms, hike load, reaching the summit, and mean daily temperature. O<sub>3</sub> and PM<sub>2.5</sub> concentrations measured during the study were below the current federal standards, and we found no significant associations of acute changes in pulmonary function with either pollutant. These findings are contrasted with those in the Mt. Washington study to examine the hypothesis that pulmonary health effects are associated with exposure to O<sub>3</sub> and PM<sub>2.5</sub> in healthy adults engaged in moderate exercise.</p>	<contrib contrib-type="author"><name><surname>Girardot</surname><given-names>Steven P.</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001044">1</xref><xref ref-type="aff" rid="af2-ehp0114-001044">2</xref></contrib><contrib contrib-type="author"><name><surname>Ryan</surname><given-names>P. Barry</given-names></name><xref ref-type="aff" rid="af1-ehp0114-001044">1</xref><xref ref-type="aff" rid="af2-ehp0114-001044">2</xref></contrib><contrib contrib-type="author"><name><surname>Smith</surname><given-names>Susan M.</given-names></name><xref ref-type="aff" rid="af3-ehp0114-001044">3</xref></contrib><contrib contrib-type="author"><name><surname>Davis</surname><given-names>Wayne T.</given-names></name><xref ref-type="aff" rid="af4-ehp0114-001044">4</xref></contrib><contrib contrib-type="author"><name><surname>Hamilton</surname><given-names>Charles B.</given-names></name><xref ref-type="aff" rid="af3-ehp0114-001044">3</xref></contrib><contrib contrib-type="author"><name><surname>Obenour</surname><given-names>Richard A.</given-names></name><xref ref-type="aff" rid="af5-ehp0114-001044">5</xref></contrib><contrib contrib-type="author"><name><surname>Renfro</surname><given-names>James R.</given-names></name><xref ref-type="aff" rid="af6-ehp0114-001044">6</xref></contrib><contrib contrib-type="author"><name><surname>Tromatore</surname><given-names>Kimberly A.</given-names></name><xref ref-type="aff" rid="af3-ehp0114-001044">3</xref></contrib><contrib contrib-type="author"><name><surname>Reed</surname><given-names>Gregory D.</given-names></name><xref ref-type="aff" rid="af4-ehp0114-001044">4</xref></contrib>	Environmental Health Perspectives	<p>Both observational studies and controlled-chamber studies have been used to assess acute effects of air pollution on lung function in adults engaged in exercise or work (<xref rid="b6-ehp0114-001044" ref-type="bibr">Aris et al. 1991</xref>; <xref rid="b7-ehp0114-001044" ref-type="bibr">Avol et al. 1984</xref>; <xref rid="b9-ehp0114-001044" ref-type="bibr">Brunekreef et al. 1994</xref>; <xref rid="b15-ehp0114-001044" ref-type="bibr">Folinsbee et al. 1984</xref>, <xref rid="b16-ehp0114-001044" ref-type="bibr">1988</xref>; <xref rid="b19-ehp0114-001044" ref-type="bibr">Gong et al. 1986</xref>; <xref rid="b20-ehp0114-001044" ref-type="bibr">Hazucha 1987</xref>; <xref rid="b21-ehp0114-001044" ref-type="bibr">Horstman et al. 1990</xref>; <xref rid="b22-ehp0114-001044" ref-type="bibr">Kinney et al. 1996</xref>; <xref rid="b23-ehp0114-001044" ref-type="bibr">Korrick et al. 1998</xref>; <xref rid="b26-ehp0114-001044" ref-type="bibr">McBride et al. 1994</xref>; <xref rid="b27-ehp0114-001044" ref-type="bibr">McDonnell et al. 1993</xref>, <xref rid="b29-ehp0114-001044" ref-type="bibr">1995</xref>, <xref rid="b28-ehp0114-001044" ref-type="bibr">1997</xref>; <xref rid="b30-ehp0114-001044" ref-type="bibr">Naeher et al. 1999</xref>; <xref rid="b36-ehp0114-001044" ref-type="bibr">Pekkanen et al. 2002</xref>; <xref rid="b38-ehp0114-001044" ref-type="bibr">Selwyn et al. 1985</xref>; <xref rid="b41-ehp0114-001044" ref-type="bibr">Spektor et al. 1988</xref>; <xref rid="b43-ehp0114-001044" ref-type="bibr">Torres et al. 1997</xref>). Although fewer in number, observational studies offer the advantage of studying the effects of pollution on humans engaged in “real-world” activities in natural settings (<xref rid="b42-ehp0114-001044" ref-type="bibr">Thurston and Ito 2001</xref>). However, they also have significant methodologic challenges. These include <italic>a</italic>) identifying an accessible population at risk whose exposures can be defined and adequately characterized, <italic>b</italic>) specifying measurable health outcomes, <italic>c</italic>) collecting an adequate amount of suitable quality-assured data on exposure and health outcomes, <italic>d</italic>) collecting sufficient data on other factors that may influence the exposure–outcome relationship, and <italic>e</italic>) the logistical issues of employing properly trained and motivated field technicians, finding cooperative subjects, and having a large enough sample size to adequately power the statistical analyses (<xref rid="b24-ehp0114-001044" ref-type="bibr">Lippmann 1989</xref>).</p><p>In 1992 and 1993, Harvard University researchers performed a large observational study of day hikers at Mt. Washington in the White Mountain National forest of New Hampshire (<xref rid="b23-ehp0114-001044" ref-type="bibr">Korrick et al. 1998</xref>). The Mt. Washington area is a popular site for outdoor recreation but is plagued with episodically high levels of ozone and fine particulate matter (≤ 2.5 μm in aerodynamic diameter; PM<sub>2.5</sub>) due to transported air pollutants and their precursors from surrounding industrial and urban areas (<xref rid="b23-ehp0114-001044" ref-type="bibr">Korrick et al. 1998</xref>). Among the significant findings in the study were a 2.2% decline (<italic>p</italic> = 0.003) in forced vital capacity (FVC) and a 2.6% decline (<italic>p</italic> = 0.02) in forced expiratory volume in 1 sec (FEV<sub>1</sub>) for each 50 ppbv (parts per billion by volume) increment in mean O<sub>3</sub> and consistent associations of decrements in both FVC (0.4% decline, <italic>p</italic> = 0.001) and peak expiratory flow (PEF; 0.8% decline, <italic>p</italic> = 0.05) across the interquartile range for PM<sub>2.5</sub> concentration of 9 μg/m<sup>3</sup> after adjusting for age, sex, smoking status, history of asthma or wheeze, hours hiked, ambient temperature, and other covariates.</p><p>The Great Smoky Mountains National Park is also a popular outdoor recreation area where ongoing monitoring has revealed high levels of air pollutants. Located in the southern Appalachian Mountains, the park encompasses 2,100 km<sup>2</sup> (520,000 acres) on the border of western North Carolina and eastern Tennessee. Approximately 95% of this acreage is forested, and elevations range from 267 to 2,021 m. With an average of > 8 million annual visitors since 1990, the park is one of the nation’s most popular. Unfortunately, it also experiences levels of O<sub>3</sub> and PM<sub>2.5</sub> that exceed those in any other national park in the eastern United States and often exceed those in nearby cities (<xref rid="b32-ehp0114-001044" ref-type="bibr">National Park Service Air Resources Division 2002</xref>). As of 2004, the entire park was classified by the U.S. Environmental Protection Agency (EPA) as a nonattainment area for the 8-hr National Ambient Air Quality Standard (NAAQS) of 80 ppbv, and a portion of the park was classified as nonattainment for the 24-hour PM<sub>2.5</sub> NAAQS of 65 μg/m<sup>3</sup> (<xref rid="b33-ehp0114-001044" ref-type="bibr">National Park Service Air Resources Division 2005</xref>). Furthermore, between 1990 and 2003, the Great Smoky Mountains was one of six national parks or federal lands to experience statistically significant increases in O<sub>3</sub> (<xref rid="b45-ehp0114-001044" ref-type="bibr">U.S. EPA 2004b</xref>). As with the Mt. Washington area, the cause of these air quality problems is primarily the regional transport of air pollutants and their precursors from nearby metropolitan areas. For the Smoky Mountains, these areas include North Carolina, Georgia, Ohio, and Tennessee (<xref rid="b32-ehp0114-001044" ref-type="bibr">National Park Service Air Resources Division 2002</xref>; <xref rid="b37-ehp0114-001044" ref-type="bibr">Renfro 2002</xref>). Transported pollutants may then be sustained at elevated levels at higher elevation (> 1,000 m) sites, due primarily to geography and the lack of sources of nitric oxide to promote O<sub>3</sub> titration (<xref rid="b5-ehp0114-001044" ref-type="bibr">Aneja and Li 1992</xref>; <xref rid="b25-ehp0114-001044" ref-type="bibr">Malone 2003</xref>).</p><p>As a class I area protected under the federal <xref rid="b10-ehp0114-001044" ref-type="bibr">Clean Air Act (1990)</xref>, the park has air quality issues that have received much attention from the popular media (<xref rid="b8-ehp0114-001044" ref-type="bibr">Barringer 2004</xref>), advocacy groups (<xref rid="b34-ehp0114-001044" ref-type="bibr">National Parks Conservation Association 2004</xref>), the U.S. Congress (<xref rid="b46-ehp0114-001044" ref-type="bibr">U.S. General Accounting Office 2001</xref>), and multi-organizational research efforts (<xref rid="b39-ehp0114-001044" ref-type="bibr">Southern Appalachian Mountains Initiative 2002</xref>; <xref rid="b40-ehp0114-001044" ref-type="bibr">Southern Oxidants Study 2002</xref>). Despite this attention, to our knowledge, no formal studies have been conducted in the park to document the possible health impacts of air pollution on people who recreate there.</p><p>To address this lack of research and to add to the epidemiologic literature on acute health effects of air pollution, we assessed the effects of O<sub>3</sub> and PM<sub>2.5</sub> on the pulmonary function of hikers at a popular recreation site in the park. Specifically, our primary goals were to determine whether the high levels of O<sub>3</sub> and PM<sub>2.5</sub> frequently observed in the Great Smoky Mountains National Park were associated with decrements in lung function of adult day hikers and to compare these findings with those reported in the Mt. Washington study.</p><sec sec-type="materials\|methods"><title>Materials and Methods</title><p>We conducted an epidemiologic study of day hikers of the Charlies Bunion trail on 71 days over two periods: 10 August 2002 through 16 October 2002 (29 sampling days) and 17 June 2003 through 27 August 2003 (42 sampling days). The Charlies Bunion trail is an approximately 6.7 km portion (one-way) of the Appalachian Trail originating at Newfound Gap, a popular high-elevation (1.54 km) destination in the Great Smoky Mountains National Park.</p><p>Between 0900 and 1200 hr, we solicited adult (≥18 years of age) volunteers embarking on day hikes along the Charlies Bunion trail to participate in the study. In accordance with all federal guidelines governing use of human participants, we obtained written informed consent from those volunteers choosing to participate. This informed consent procedure was overseen by institutional review boards at both the University of Tennessee and Emory University. A participating hiker was then assigned a random four-digit code, and we obtained height and weight (with and without any hiking load) data. All researchers involved in data collection and analysis completed the National Institutes of Health <italic>Human Participant Protections Education for Research Teams</italic> online course (<xref rid="b31-ehp0114-001044" ref-type="bibr">National Institutes of Health 2006</xref>) and any additional human subject protection education programs required by their respective universities’ institutional review boards. Data collection days were rotated between two teams: one led by the University of Tennessee and one led by Emory University and Western Carolina University.</p><sec><title>Pulmonary function testing (spirometry)</title><p>To assess change in pulmonary function, we asked participants to perform spirometry both before their hike and when they returned from their hike. Spirometry technicians received 1–2 days of training by a licensed respiratory therapist in all aspects of performing spirometry. As part of this training, technicians were required to demonstrate proper techniques with mock volunteers and were trained by the respiratory therapist before being allowed to work on the study. Puritan-Bennett Renaissance II Spirometry Systems (Tyco Healthcare, Pleasanton, CA) were used to perform all spirometry.</p><p>Prehike pulmonary function tests were typically performed in the mornings (0900–1200 hr), and posthike tests were performed in the afternoons (1400–1900 hr) within 20 min of a hiker’s return to the Newfound Gap trailhead. All tests were performed at 1.54 km above mean sea level inside a retrofitted research van that was equipped with two spirometry stations. Participants were tested in the seated position wearing nose clips and performed a minimum of three and a maximum of eight FVC maneuvers as recommended by the American Thoracic Society (ATS) standards (<xref rid="b2-ehp0114-001044" ref-type="bibr">ATS 1995</xref>). Participants were required to have pre-and posthike testing performed by the same technician on the same machine.</p><p>On each sampling day, the spirometers were calibrated in the morning before prehike testing and in the afternoon before posthike testing using a fixed-volume, 3-L syringe. Tolerance limits for acceptable calibration were ± 3% (2.91–3.09 L) in accordance with American Association for Respiratory Care Clinical Practice Guidelines (<xref rid="b1-ehp0114-001044" ref-type="bibr">American Association for Respiratory Care 1996</xref>).</p><p>To determine whether a hiker’s pre- and posthike pulmonary function tests met the ATS acceptability criteria for inclusion in epidemiologic studies, each maneuver within both the pre- and posthike test sessions was evaluated by a pulmonary physician (R.A.O.). The physician, experienced with spirometry and blinded to the study hypothesis, inspected both the flow-volume and volume-time curves to ensure ATS standards were satisfied. Briefly, current (1994) ATS standards for acceptable spirometry include good start of test (an extrapolated volume of ≤ 5% of the FVC or 150 mL, whichever is greater), no hesitation or false start, a rapid start to rise time, no cough, especially during the first second of the maneuver, and no early termination of exhalation (unless there is no volume change for at least 1 sec or the subject cannot or should not continue to exhale further) (<xref rid="b2-ehp0114-001044" ref-type="bibr">ATS 1995</xref>). For each hiker who gave at least two acceptable prehike and at least two acceptable posthike maneuvers, we assessed FVC and FEV<sub>1</sub> reproducibility criteria set forth by the ATS. These criteria require that the largest two FVC values from among acceptable maneuvers be within 0.2 L of each other and the largest two FEV<sub>1</sub> values from among acceptable maneuvers be within 0.2 L of each other (<xref rid="b2-ehp0114-001044" ref-type="bibr">ATS 1995</xref>).</p><p>For each hiker who gave acceptable and reproducible pre- and posthike spirometry, we calculated the percentage change in five spirometric values: FVC, FEV<sub>1</sub>, FEV<sub>1</sub>/FVC, PEF, and mean flow rate between 25% and 75% of the FVC (FEF<sub>25–75%</sub>). Percentage change was defined as 100 times the difference of the posthike value minus the prehike value divided by the prehike value. For FVC and FEV<sub>1</sub>, we used the maximum prehike and posthike values from among those maneuvers that were acceptable and reproducible. Prehike and posthike values of FEV<sub>1</sub>/FVC, PEF, and FEF<sub>25–75%</sub> were taken from the single acceptable and reproducible maneuver with the maximum sum of FEV<sub>1</sub> and FVC (<xref rid="b2-ehp0114-001044" ref-type="bibr">ATS 1995</xref>).</p></sec><sec><title>Trip log diary</title><p>Each participant was given a trip log diary to complete during the hike. Along the Charlies Bunion trail there are four National Park Service signs marking various points. These are the Newfound Gap trail-head, Sweat Heifer Creek Trail (2.7 km from Newfound Gap trailhead), Boulevard Trail turnoff (1.6 km from Sweat Heifer Creek Trail), Ice Water Spring Shelter (0.3 km from Boulevard Trail turnoff), and Charlies Bunion (2.1 km from Ice Water Spring Shelter). We provided digital watches, demonstrated proper technique for taking a pulse (radial or carotid), and instructed hikers to record their time of arrival and 15-sec pulse at designated location on ascent (trailhead to highest destination reached) and then on descent (highest destination reached to trailhead) and to note any special circumstances or deviations from the trail. Hikers were not asked to record respiratory symptoms along the hike.</p></sec><sec sec-type="background"><title>Respiratory health symptoms and history questionnaire</title><p>After completing posthike spirometry, hikers responded to a modified version of the ATS Division of Lung Disease questionnaire (<xref rid="b14-ehp0114-001044" ref-type="bibr">Ferris 1978</xref>). The standardized questionnaire obtained information on respiratory illness symptoms (cough, wheeze, phlegm, shortness of breath), history of respiratory illness (chest injury, heart trouble, bronchitis, pneumonia, pleurisy, pulmonary tuberculosis, hay fever, bronchial asthma), use of a bronchodilator within 48 hr, frequency and intensity of weekly aerobic activity, demographics (race, sex, age, marital status, education level, occupation), smoking status (never, current, former), and smoking history (if applicable).</p></sec><sec><title>O<sub>3</sub> and PM exposure assessment</title><p>Real-time ambient O<sub>3</sub> and PM<sub>2.5</sub> concentrations, along with temperature and relative humidity, were monitored on-site at the Newfound Gap trail-head on each study day. One-minute average O<sub>3</sub> concentrations were measured using a ultraviolet-absorption–based O<sub>3</sub> monitor (model 202; 2B Technologies, Boulder, Colorado). Dynamic calibration of the monitor was performed at the Knox County, Tennessee, Department of Air Quality Management’s Air Quality Laboratory. We performed co-location studies at the Spring Hill Elementary monitoring site in Knoxville, Tennessee. Finally, because most of the Charlies Bunion trail is under forested canopy, we conducted a series of studies to assess a possible canopy effect—the potential reduction of O<sub>3</sub> concentration due to vegetative uptake and deposition. The details of these studies are presented elsewhere (<xref rid="b25-ehp0114-001044" ref-type="bibr">Malone 2003</xref>). Briefly, the portable O<sub>3</sub> monitor was used to measure concentrations on the trail (under the canopy) and at the trailhead (outside of the canopy). From these studies, a canopy correction factor was developed for the exposure calculations to ensure that the measured O<sub>3</sub> concentrations accurately reflected a hiker’s true O<sub>3</sub> exposure.</p><p>A β-attenuation filter-based mass monitor (E-BAM; Met One Instruments, Grants Pass, OR) measured 1-hr average PM<sub>2.5</sub> concentrations. Co-location studies were performed with a continuous PM<sub>2.5</sub> monitor (tapered element oscillating microbalance) at the Look Rock monitoring station, and flow, temperature, and system calibrations were performed throughout the study.</p><p>The O<sub>3</sub> monitor was small enough to be attached to the E-BAM, and two 12-V DC batteries connected in parallel provided sufficient power for the monitors to run for at least 12 hr. All data were downloaded from the monitors directly onto a laptop computer.</p><p>On days where either portable monitor was not operating, we substituted values from two permanent monitoring stations maintained by the National Park Service: Clingmans Dome for O<sub>3</sub> (a high-elevation site 6.4 km from Newfound Gap, 2.0 km above mean sea level) and Look Rock for PM<sub>2.5</sub> (located on the eastern border of the park, 0.80 km above mean sea level). In both cases, we corrected the park’s monitors to equivalent values for Newfound Gap based on correlations obtained from co-location studies. The correlation coefficients ranged from 0.6 to 0.9, indicating that correlations between the portable monitors and permanent monitors were adequate. Monitor failure occurred on approximately 15 sampling days for O<sub>3</sub> and 7 sampling days for PM<sub>2.5</sub>.</p><p>Concentrations for O<sub>3</sub> and PM<sub>2.5</sub> were reported as 15-min average concentrations for use in exposure calculations. A time-weighted average pollutant (O<sub>3</sub> or PM<sub>2.5</sub>) concentration for each hiker was calculated by multiplying the average pollutant concentration in each discreet interval along the hike by the fraction of time spent in that interval. Times spent in each of the interval were taken from the trip log diary data. O<sub>3</sub> canopy corrections were made for portions of the hike under the forested canopy. In general, a 13% decrease in O<sub>3</sub> concentration was observed within the canopy (<xref rid="b25-ehp0114-001044" ref-type="bibr">Malone 2003</xref>).</p><p>Fifteen-minute averages of temperature and relative humidity were measured at the trailhead on each study day, and an overall daily average was computed for use in all statistical models.</p></sec><sec sec-type="methods"><title>Statistical methods</title><p>To obtain an estimate of the relationship between O<sub>3</sub> and PM<sub>2.5</sub> exposure and change in pulmonary function, we used multiple linear regression, modeled by ordinary least squares estimation, as our primary method of analysis (PROC GLM; SAS Institute Inc., Cary, NC). The dependent variables in these analyses were the percentage change (posthike from prehike) in each of the five spirometric values: FVC, FEV<sub>1</sub>, FEV<sub>1</sub>/FVC, PEF, and FEF<sub>25–75%</sub>. The two pollutant exposure variables, O<sub>3</sub> and PM<sub>2.5</sub>, were considered the independent variables in the analysis.</p><p>To compare results between our study and the Mt. Washington study, we employed a similar modeling strategy. We fit separate regression models for each of the spirometric values as a function of each pollutant exposure. Both univariate and adjusted models were calculated. For the adjusted models, we selected <italic>a priori</italic> covariates based on those adjusted for in the Mt. Washington study. These included both continuous variables (age, hours hiked, and mean temperature) and categorical variables [sex, smoking status (former vs. never), history of asthma or wheeze symptoms, carrying a backpack, and reaching the summit]. In addition to these models, an adjusted piecewise linear regression model was fit for O<sub>3</sub> using an inflection point of 40 ppbv to determine whether or not different relationships were observed at higher concentrations.</p></sec></sec><sec sec-type="results"><title>Results</title><sec sec-type="methods"><title>Study population</title><p>Over the 71 sampling days, 905 hikers initiated participation in the study. Of these hikers, 79 did not return for the posthike testing and an additional nine withdrew (either during pre- or posthike testing). A total of 817 (90.3%) returned for posthike spirometry testing.</p><p>Initial eligibility criteria included adult age (≥18 years), nonsmoker (had never smoked or had not smoked for 1 year before testing), no use of bronchodilator or asthma medication within 48 hr of testing, and day hikers who hiked at least to the Sweat Heifer trail marker. Among the 817 hikers who completed the study, 96 (12%) violated at least one of the initial inclusion criteria, and 721 (88%) were retained for further consideration. The most significant reasons for exclusion were smoking (<italic>n</italic> = 43 current smokers) and use of a bronchodilator within 48 hr of the test (<italic>n</italic> = 34).</p><p>Pulmonary function tests of these 721 hikers were then evaluated for inclusion in the analysis population as described previously. Of these hikers, 367 (50.9%) were excluded for failure to provide at least two acceptable and reproducible pre- and posthike pulmonary function tests. The most common reason for spirometric test failure was failure to blow out hard enough or long enough (~ 30%). This resulted in a final sample size for the analysis population of 354 hikers.</p><p>Selected demographic data for hikers included in the analysis population as well as those excluded are shown in <xref ref-type="table" rid="t1-ehp0114-001044">Table 1</xref>. Most hikers were white (96%), never smoked (75%), and had no history of asthma or wheeze (82%). Sex was evenly divided, with a slight majority of females (56%). Age ranged from 18 to 82 years, with mean age of 43 years.</p><p>We tested for differences between those excluded due to spirometric test failure and those included in the analysis population using chi-square comparisons for categorical variables and two-sided <italic>t</italic>-tests for continuous variables. These results are shown in <xref ref-type="table" rid="t1-ehp0114-001044">Table 1</xref>. Statistically significant differences (at the 5% level) were seen in sex (more males excluded) and, as a result, in baseline FEV<sub>1</sub> and FVC. Otherwise, the excluded hikers did not differ substantially from the analysis population.</p></sec><sec><title>Exposure assessment</title><p>O<sub>3</sub> and PM<sub>2.5</sub> concentrations were lower than anticipated at the onset of the study, and despite a record of frequent violations in past years, there were no exceedances of the current 8-hr NAAQS (80 ppbv) or the 24-hr standard for PM<sub>2.5</sub> (65 μg/m<sup>3</sup>) during the study period (<xref rid="b44-ehp0114-001044" ref-type="bibr">U.S. EPA 2004a</xref>). The average daily O<sub>3</sub> concentration measured at the Newfound Gap trail-head on the 71 study days was 52.0 ± 13.4 ppbv with a range of 27.6–79.3 ppbv. The average daily PM<sub>2.5</sub> concentration was 13.9 ± 8.2 μg/m<sup>3</sup> with a range of 1.6–38.4 μg/m<sup>3</sup>.</p><p>Average daily temperature for the study days ranged from 2.6 to 24.1°C with a mean of 19.2 ± 4.4°C, and average daily relative humidity ranged from 48.2 to 93.9% with a mean of 73.6 ± 10.8%.</p><p>We computed O<sub>3</sub> and PM<sub>2.5</sub> concentrations for hikers included in the analysis data set (<italic>n</italic> = 354) using each hiker's time–weight average concentration including a correction for time spent under the canopy. (<xref ref-type="table" rid="t1-ehp0114-001044">Table 1</xref>). O<sub>3</sub> concentrations ranged from 25.0 to 74.2 ppbv with a group mean of 48.1 ± 12.0 ppbv during exercise. PM<sub>2.5</sub> concentrations ranged from 0.21 to 41.9 μg/m<sup>3</sup> with a group mean of 15.0 ± 7.4 μg/m<sup>3</sup> during exercise. For comparison, concentrations were also computed for excluded hikers and are shown in <xref ref-type="table" rid="t1-ehp0114-001044">Table 1</xref>.</p><p><xref ref-type="fig" rid="f1-ehp0114-001044">Figures 1</xref> and <xref ref-type="fig" rid="f2-ehp0114-001044">2</xref> show the hourly variation of PM<sub>2.5</sub> and O<sub>3</sub>, respectively, on study days. In contrast to strong diurnal patterns in urban O<sub>3</sub>, high-elevation sites typically display only small variation in O<sub>3</sub> concentrations throughout the day (<xref rid="b4-ehp0114-001044" ref-type="bibr">Aneja et al. 2000</xref>). These data reflect this high-elevation O<sub>3</sub> pattern. PM<sub>2.5</sub> concentrations were also fairly constant throughout the day, with increases in the late afternoon (1500 hr and later). For both pollutants, 2003 levels were slightly higher than those observed in 2002. This was expected because of the seasonal difference between the 2002 and 2003 sampling periods (2002 sampling period was mostly during the fall and 2003 mostly during the summer).</p><p>For the 354 included hikers, the mean O<sub>3</sub> concentrations were significantly (<italic>p</italic> < 0.0001) correlated with mean PM<sub>2.5</sub> concentrations (Spearman <italic>r</italic> = 0.67). However, both pollutants were weakly but significantly associated with average daily temperature and relative humidity (O<sub>3</sub>: Spearman <italic>r</italic> = 0.16, <italic>p</italic> = 0.0039, and Spearman <italic>r</italic> = –0.59, <italic>p</italic> < 0.0001, respectively; PM<sub>2.5</sub>: Spearman <italic>r</italic> = 0.38, <italic>p</italic> < 0.0001, and Spearman <italic>r</italic> = –0.31, <italic>p</italic> < 0.0001, respectively).</p></sec><sec><title>Exercise profile</title><p>From the trip log diaries, we determined each hiker’s highest destination reached, the total hiking distance (using the roundtrip distances from the National Park Service), and the total roundtrip hiking time (defined as time between prehike and posthike spirometry).</p><p>Selected exercise characteristics are also summarized in <xref ref-type="table" rid="t1-ehp0114-001044">Table 1</xref>. Most included hikers (79%) carried a backpack or other load during their hike, with the average load weighing 4.1 ± 2.6 kg. Most (71%) also reached the peak (Charlies Bunion), with the average hiking distance of 12.2 ± 2.4 km and average hiking time of 5.0 ± 1.2 hr. There were no significant differences in the exercise profile compared with excluded hikers.</p><p>From the pulse data, we determined each hiker’s maximum self-reported pulse (as number of beats per minute) and the percentage of age-predicted maximum pulse rates achieved, defined as 100 times the maximum self-reported pulse divided by 220 minus the hiker’s age. For hikers included in the study, the mean percent maximum pulse achieved was 68 ± 13% with a range of 35–100%.</p><p>We also determined each hiker’s baseline level of physical fitness by asking hikers about their typical exercise intensity and weekly frequency on the ATS-DLD questionnaire. Most (73%) indicated that they exercised at least 2 days per week, and most (72%) indicated that their exercise level was moderate or intense.</p></sec><sec><title>Pulmonary function response to exposure</title><p>The crude mean posthike percentage changes in each spirometric variable (FVC, FEV<sub>1</sub>, FEV<sub>1</sub>/FVC, FEF<sub>25–75%</sub>, PEF) were small and, in most cases, positive (<xref ref-type="table" rid="t2-ehp0114-001044">Table 2</xref>). Only two spirometric variables—PEF and FEV<sub>1</sub>/FVC— had negative overall mean posthike percentage changes: 1.08% and –0.003%, respectively. Crude mean changes for FVC, FEV<sub>1</sub>, and PEF were 0.24%, 0.15%, and 1.27%, respectively.</p><p>To explore a possible dose–response relationship between pollutant exposure and pulmonary function, we calculated the quintiles of the observed mean O<sub>3</sub> and PM<sub>2.5</sub> distributions and determined the mean posthike percentage change in selected spirometric variables—FVC, FEV<sub>1</sub>, and PEF—within each quintile. These results are summarized in <xref ref-type="table" rid="t2-ehp0114-001044">Tables 2</xref> and <xref ref-type="table" rid="t3-ehp0114-001044">3</xref> and displayed graphically in <xref ref-type="fig" rid="f3-ehp0114-001044">Figures 3</xref> and <xref ref-type="fig" rid="f4-ehp0114-001044">4</xref> for PM<sub>2.5</sub> and O<sub>3</sub>, respectively.</p><p>Across the quintiles of O<sub>3</sub> and PM<sub>2.5</sub> concentration, the prehike means of each of the pulmonary functions were similar. However, trends in mean posthike percentage changes across quintiles of either pollutant were not statistically significant for any spirometric variable. For FVC and FEV<sub>1</sub> with O<sub>3</sub>, mean posthike percentage changes were positive with the exception of the first two quintiles (corresponding to O<sub>3</sub> concentrations of 35.3 and 43.5 ppbv); for FVC and FEV<sub>1</sub> with PM<sub>2.5</sub>, only quintile 2 (corresponding to a PM<sub>2.5</sub> concentration of 11.1 μg/m<sup>3</sup>). As <xref ref-type="fig" rid="f3-ehp0114-001044">Figures 3</xref> and <xref ref-type="fig" rid="f4-ehp0114-001044">4</xref> show, the curves for FVC and FEV<sub>1</sub> are relatively constant, indicating little variation in response as a function of pollutant level.</p><p>The PEF response curves show a steady increase from –4.43% to 2.50% across quintiles of PM<sub>2.5</sub> concentration (<xref ref-type="fig" rid="f3-ehp0114-001044">Figure 3</xref>) and a steady increase from –1.51% to 1.99% across quintiles of O<sub>3</sub> concentration (<xref ref-type="fig" rid="f4-ehp0114-001044">Figure 4</xref>).</p></sec><sec><title>Multiple linear regression models</title><p>Results from multiple linear regression analyses of the percentage change between the pre- and posthike pulmonary function variables (FVC, FEV<sub>1</sub>, FEV<sub>1</sub>/FVC, PEF, and FEF<sub>25–75%</sub>) and the time-weighted average concentration of O<sub>3</sub> and PM<sub>2.5</sub> during the hike period are presented in <xref ref-type="table" rid="t3-ehp0114-001044">Table 3</xref>. Parameter estimates for the exposures, along with their respective <italic>p</italic>-values, are shown for both univariate and adjusted models. In the final adjusted models, we controlled for age, hours hiked, sex, smoking status (never or former), history of asthma or wheeze symptoms, carrying a backpack or other load, reaching the summit, and mean daily temperature. The adjusted models are based on a sample size of <italic>n</italic> = 339 because of missing temperature data for 15 hikers.</p><p>In most cases, regression slopes (in units of percent change/concentration) were small and not statistically significant. For example, the coefficient for the percent change in FEV<sub>1</sub> as a function of PM<sub>2.5</sub>, adjusted for covariates, was 0.003%/μg/m<sup>3</sup> with a <italic>p</italic>-value of 0.937, indicating that there was no association between PM<sub>2.5</sub> concentration and change in FEV<sub>1</sub> over the hike period. Similar interpretations of the coefficients of the other outcome variables and pollutant exposures may be made. Finally, <italic>F</italic>-tests for significant overall regression (data not shown) indicated that the adjusted models did not explain a significant amount of the variation in posthike pulmonary function change. The results from the piecewise model for O<sub>3</sub> with an inflection point of 40 ppbv did not produce different results. In all cases, except for PEF in the adjusted PM<sub>2.5</sub> models, the regression slopes were not statistically different from zero.</p><p>These conclusions were consistent across several subgroups. There was no change in statistical significance of the regression coefficients for those hikers with a self-reported history of asthma or wheeze (<italic>n</italic> = 62). To improve power, we defined two dichotomous categorical variables based on the ATS-DLD questionnaire responses: a respiratory symptom index based on a hikers’ reporting of any positive symptom of respiratory illness (e.g., cough, cough with phlegm, shortness of breath; <italic>n</italic> = 176) and a respiratory health history index based on whether a hiker reported any positive history of respiratory or cardiovascular illness (e.g., heart trouble, bronchitis, pneumonia, asthma; <italic>n</italic> = 173) (<xref rid="b17-ehp0114-001044" ref-type="bibr">Galizia and Kinney 1999</xref>). In both subgroups, mean lung function changes did not differ over the exposure levels, and both univariate and adjusted models resulted in no statistically significant associations. Finally, we restricted analyses to those > 50 years of age (<italic>n</italic> = 103), and our results were the same. We did not perform subanalyses on those with extreme lung function decrements (posthike percentage decrements of ≥ 5% in FVC or FEV<sub>1</sub>) because of lack of sufficient sample (<italic>n</italic> = 40).</p><p>To evaluate whether meteorologic variables may have confounded the relationship between exposure and outcome, we computed regression models both with and without average daily temperature and relative humidity. In both cases, results did not change. We included temperature in our final models, however, to compare findings with the Mt. Washington study. We also computed multi-pollutant models, adjusting simultaneously for O<sub>3</sub> and PM<sub>2.5</sub>. As expected, because of the high correlations between the two pollutants, it was not possible to separate the effects in these models.</p></sec><sec sec-type="methods"><title>Comparison with the Mt. Washington study</title><p><xref ref-type="table" rid="t4-ehp0114-001044">Table 4</xref> compares selected experimental variables between the Mt. Washington and Charlies Bunion (present) studies. The Mt. Washington study was performed on 74 days over 2 years. A total of 766 hikers initiated, with 530 (69%) meeting eligibility criteria. The Charlies Bunion study was performed on 71 days over 2 years. More hikers (<italic>n</italic> = 905) initiated the present study, but the inclusion rate was much smaller (39% compared with 69%). The primary reason for this difference in inclusion was spirometric test failure: fewer subjects in the Charlies Bunion study met ATS requirements for acceptability and reproducibility.</p><p>The demographics for both studies were similar. In both, most (96–97%) participants were white, never smokers (71–76%), and had no history of asthma or wheeze (82–92%). The average age was higher in the Charlies Bunion study: 46 compared with 35 in the Mt. Washington study. Finally, males composed a smaller percentage of included subjects (44% in the present study vs. 71% in the Mt. Washington study).</p><p>The exercise profile of included hikers in both studies was a significant point of difference. Although there were some similarities, including average maximum pulse rate (122 in the Mt. Washington study vs. 121 in the present study), percentage of age-predicted pulse (66% vs. 68% in the present study), and most reaching the summit and carrying a load, there was a significant difference in exercise (hiking) time. Mt. Washington hikers spent an average of 8 hr hiking, whereas Charlies Bunion hikers spent an average of 5 hr hiking. These differences are reflected in differing exposure levels. Despite similar air pollutant levels in both locations (Mt. Washington vs. Charlies Bunion, respectively: mean O<sub>3</sub>, 40 vs. 47 ppbv; mean PM<sub>2.5</sub>, 15 vs. 15 μg/m<sup>3</sup>), the fact that the Mt. Washington study participants spent more time exercising translated into a higher exposure to pollutants.</p><p>Pulmonary function testing between the two studies was similar. In both cases, spirometry was performed in the seated position with nose clips. Posthike testing time was slightly later for the Mt. Washington study because of the longer hike time. One important difference, however, was the coaching. In the Mt. Washington study, only one spirometry technician certified by the National Institute for Occupational Safety and Health (NIOSH) conducted all tests. In the present study, however, 13 technicians were employed. These technicians were predominantly graduate students who had received 1–2 days of training from a certified respiratory therapist. Because spirometry is a highly effort-dependent test, the additional number of technicians may have introduced more variability in the measurements. Finally, baseline values of FEV<sub>1</sub> and FVC were slightly higher in the Mt. Washington study as a direct result of the larger percentage of males in their analysis population.</p><p><xref ref-type="table" rid="t5-ehp0114-001044">Table 5</xref> directly compares selected findings for percentage change in pulmonary function as a function of ambient O<sub>3</sub> and PM<sub>2.5</sub> from the two studies. In the Mt. Washington study, adjusted linear models demonstrated statistically significant declines in FEV<sub>1</sub> (–0.051%/ppbv) and FVC (–0.043%/ppbv) with O<sub>3</sub> and statistically significant declines in FEV<sub>1</sub> (–0.041%/μg/m<sup>3</sup>), FVC (–0.043%/μg/m<sup>3</sup>), and PEF (–0.087%/μg/m<sup>3</sup>) with PM<sub>2.5</sub>. In the Charlies Bunion study, linear models adjusting for the same variables did not demonstrate significant associations between posthike change in FEV<sub>1</sub> and FVC and either pollutant. However, in both studies, there were no significant associations with PEF, FEV<sub>1</sub>/FVC, or FEF<sub>25–75%</sub> and O<sub>3</sub>.</p></sec></sec><sec sec-type="discussion"><title>Discussion</title><p>This study evaluated the hypothesis that exposure to ambient O<sub>3</sub> and PM<sub>2.5</sub> leads to acute respiratory effects, as measured by transient changes in pulmonary function, in healthy adults engaged in moderate exercise. Furthermore, we have added to the epidemiologic literature on acute health effects of air pollution by replicating another observational study of healthy adult hikers. To our knowledge, this was one of the first replications of a large-scale observational study of exercising adults. Although there were differences in findings between the two studies, consistent conclusions were reached.</p><p>We demonstrated that no statistically significant responses in pulmonary function occur when an average of 5.0 hr of outdoor exercise occurs at the levels of O<sub>3</sub> and PM<sub>2.5</sub> that we observed, some of which were substantially below the current NAAQS—80 ppbv for O<sub>3</sub> (8-hr) and 65 mg/m<sup>3</sup> for PM<sub>2.5</sub> (24-hr). Specifically, posthike percentage changes in FVC, FEV<sub>1</sub>, FEV<sub>1</sub>/FVC, FEF<sub>25–75%</sub>, and PEF were not associated with either O<sub>3</sub> or PM<sub>2.5</sub> exposure.</p><p>In studies where repeated pulmonary function tests are performed within the same day, it is important to assess confounding effects due to diurnal variation in lung function. It has been documented that expiratory flow and volume variables have minimum values early in the morning (0400–0600 hr) and peak around noon (<xref rid="b11-ehp0114-001044" ref-type="bibr">Dockery and Brunekreef 1996</xref>). In our study, however, spirometric measurements were made at the same times (prehike, 0900–1200 hr; posthike, 1400–1900 hr) on all study days, regardless of pollution levels. This ensured that this confounding did not occur, but we assessed it quantitatively by computing regression models that were restricted to hikers whose prehike spirometric measurements were taken before 1100 hr and posthike measurement taken after 1500 hr (<italic>n</italic> = 135). Our results did not change.</p><p>A potential source of bias in our study was with the spirometry. It has been demonstrated that exclusion of subjects with unacceptable and nonreproducible measurements in studies of pulmonary function and health outcomes may lead to removing subjects with a more accelerated loss of lung function (<xref rid="b12-ehp0114-001044" ref-type="bibr">Eisen et al. 1984</xref>). In this study, more than half of the participants were excluded because of spirometric test failure on either the pre- or posthike testing (or both). To assess this potential bias, we performed additional analyses of spirometric test failure using the full study population (<italic>n</italic> = 721). Full descriptions and results of these studies are presented elsewhere (<xref rid="b18-ehp0114-001044" ref-type="bibr">Girardot 2005</xref>), but the relevant findings are briefly discussed here. Of the full study population, 700 (97%) hikers provided three complete maneuvers during both the prehike and posthike sessions and were included in these analyses. Spirometric test failure, as defined by the 1994 ATS standards and including both acceptability and reproducibility criteria for the top three maneuvers, was exhibited by 439 (62.7%) participants during prehike sessions and by 424 (60.6%) participants during posthike sessions. For both sessions, reproducibility criteria (both FVC and FEV<sub>1</sub>) for the top two maneuvers were achieved by > 80% of participants (prehike, 84.9%; posthike, 82.3%). Fewer than half of the hikers could perform three acceptable maneuvers during a test session (prehike, 40.3%; posthike, 45.0%), and slightly more could perform at least two acceptable maneuvers during a test session (prehike, 59.7%; posthike, 55.0%). We also sought to examine the association between spirometric test failure and a number of hiker characteristics, including age, sex, body mass index, respiratory health status, and respiratory health history using both stratified analyses and logistic regression modeling, where spirometric test failure was treated as the outcome (coded dichotomously as yes or no). We found no statistically significant associations at the 5% level. Finally, we examined models that included a technician variable as a predictor of test failure. There was no association between technician and spirometric test failure.</p><p>These findings imply that the most likely cause of test failure was poor coaching techniques. It has been well argued that achieving quality spirometry depends largely on the “skill and perseverance of the technician” (<xref rid="b13-ehp0114-001044" ref-type="bibr">Enright et al. 2004</xref>). In our study, we were faced with the challenge of collecting data from unpaid volunteers in a nonclinical setting (on top of a mountain in a research van) who were generally unfamiliar with the technique and in a hurry to start their hike. Furthermore, we employed graduate students, senior undergraduates, and research assistants. Although they were all trained and approved by a certified respiratory therapist from the University of Tennessee, we realize that coaching volunteer participants—who were frequently uncooperative and/or hesitant—to achieve three acceptable and reproducible maneuvers was extremely difficult. As a result, our recommendations for any field study using spirometry is to employ only NIOSH-certified technicians and to minimize the number of technicians to help reduce the variability that could have been introduced by using different technicians on different days (<xref rid="b35-ehp0114-001044" ref-type="bibr">NIOSH 2004</xref>).</p><p>Despite the loss of sample size because of poor spirometry, we must point out that the excluded population did not differ substantially from the included population (<xref ref-type="table" rid="t1-ehp0114-001044">Table 1</xref>). For example, we did not have more hikers with asthma or wheeze excluded because of poor spirometry. In addition, our resulting sample size of <italic>n</italic> = 354 is higher than other studies examining similar hypotheses and is comparable with the Mt. Washington study population of <italic>n</italic> = 530. Finally, before being included in the analyses, each individual maneuver was carefully reviewed by an experienced pulmonary physician (R.A.O.) who was blinded to the study hypothesis. As a result, we feel that the conclusions reached would not differ had more participants been included in the analyses.</p><p>There were several additional limitations to our study. First, we could not assess minute ventilation of the hikers to determine a true pollution dose for each hiker. Maximum pulse was used as a proxy for exercise intensity (and hence dose), but this is not an adequate surrogate, because more fit subjects have lower minute ventilation and therefore receive a lower dose of pollutant. In addition, the study did not include children, and there was almost no participation from minority groups such as African Americans or Hispanics. Finally, by choosing to replicate the Mt. Washington study, we were constrained to follow similar protocols and procedures to allow the comparative analysis to be more meaningful. For example, one type of information not considered during this study or in the Mt. Washington study was an assessment of clinical symptoms of respiratory disease during the hike. The ATS, in defining what constitutes an adverse health effect, has stated that reduction in FEV<sub>1</sub> or FVC must be associated with clinical symptoms (e.g., cough or wheeze) (<xref rid="b3-ehp0114-001044" ref-type="bibr">ATS 2000</xref>). Another variable both studies failed to measure was prehiking levels of pollutants. It could be argued that elevated levels of pollutants before the start of a hike might affect the health outcome, especially if these levels were higher than those experienced during the hike. However, we feel that because all of our subjects began their hikes in the morning, when pollution levels are typically at their lowest (even in urban areas), prehike pollution exposure was likely to be minimal. Further, in our study, most hikers arrived in automobiles, which offered some slight protection from air pollution. As a result, we do not feel that this was an issue in either study.</p><p>Air quality conditions during the study differed from what was initially predicted based on historical data. During the two study periods, the park had some of the best air quality in many years, due primarily to heavy rainfall. Rainfall “washes out” air pollutants, resulting in good air quality. As a result, the focus of the study shifted from modeling health effects at levels higher than the federal standards to modeling health effects at levels below the current federal standards. The findings from this study directly address the question of whether current federal standards are protective for human health in a healthy, exercising population.</p><p>Both this study and the Mt. Washington study examined the respiratory effects of relatively low concentrations of O<sub>3</sub> and PM<sub>2.5</sub>. One key difference between the two studies was the exposure duration. Mt. Washington hikers averaged 8.0 hr of exercise, whereas hikers in this study averaged 5.0 hr. However, these exercise periods were longer than in many previous field studies, which average exercise times of less than 2 hr. Another key difference was the mean age of the study populations. In the present study, the average age of the hikers was 46 years, compared with 35 in the Mt. Washington study. This is an important point of comparison, because older individuals may be less responsive to O<sub>3</sub> and PM than younger individuals. Although the Mt. Washington study found significant decrements in FVC and FEV<sub>1</sub> with both pollutants, the magnitude of the mean changes was small, and as the authors point out, “unlikely to result in clinical symptoms in most individuals” (<xref rid="b23-ehp0114-001044" ref-type="bibr">Korrick et al. 1998</xref>). Furthermore, both studies failed to show significant associations in other spirometric variables—PEF, FEV<sub>1</sub>/FVC, or FEF<sub>25–75%</sub>—and O<sub>3</sub> and between FEV<sub>1</sub>/FVC or FEF<sub>25–75%</sub> and PM<sub>2.5</sub>. These findings are consistent with previous studies of lung function effects in nonasthmatic subjects. Relatively few observational studies have been conducted on healthy adults engaged in moderate exercise under typical outdoor conditions. For example, results of PM<sub>2.5</sub> peak flow analyses in several studies reported no consistent evidence for adverse health effects (<xref rid="b47-ehp0114-001044" ref-type="bibr">Vedal 1998</xref>).</p><p>This study is one of the first designed and conducted, in part, to compare findings from two observational studies of acute respiratory illness and low levels of air pollution in adults engaged in outdoor exercise. Because large-scale observational studies, which are typically expensive and time-consuming to run, are relatively rare, the results obtained from this type of comparative study are important in the epidemiologic literature because they provide evidence (or lack of evidence) of associations between environmental exposure and health effects for individuals in natural settings. Our findings suggest that low levels of pollutant exposure over several hours may not result in significant declines in lung function in healthy adults engaged in exercise or work. However, there is considerable variation in individual response to pollutant exposure, and findings from epidemiologic studies—which rely on testing group means and other indicators—may not be entirely indicative of a lack of individual risk for adverse health effects due to air pollution. Finally, it may be difficult to separate the effects of the exercise or activity itself from the air pollution effects.</p></sec>