Split (1)

train · 5.09k rows

pmcid stringlengths 6 6	title stringlengths 9 374	abstract stringlengths 2 4.62k ⌀	fulltext stringlengths 167 106k	file_path stringlengths 64 64
518964	Is remission of depressive symptoms in primary care a realistic goal? A meta-analysis	Background A primary goal of acute treatment for depression is clinical remission of symptoms. Most meta-analyses of remission rates involve randomized controlled trials (RCTs) using patients from psychiatric settings, but most depressed patients are treated in primary care. The goal of this study was to determine remission rates obtained in RCTs of treatment interventions for Major Depressive Disorder (MDD) conducted in primary care settings. Methods Potentially relevant studies were identified using computerized and manual search strategies up to May 2003. Criteria for inclusion included published RCTs with a clear definition of remission using established outcome measures. Results A total of 13 studies (N = 3202 patients) meeting inclusion criteria were identified. Overall remission rates for active interventions ranged between 50% and 67%, compared to 32% for pill placebo conditions and 35% for usual care conditions. Conclusions Remission rates in primary care studies of depression are at least as high as for those in psychiatric settings. It is a realistic goal for family physicians to target remission of symptoms as an optimal outcome for treatment of depression.	Background Major depressive disorder (MDD) is one of the most common and disabling of medical conditions [ 1 ]. The Canadian Community Health Survey recently reported a one-year prevalence rate of 4.5% for MDD, indicating that over 1.1 million Canadians suffer significant distress and impairment in function due to depression [ 2 ]. The economic costs of depression are estimated at over $5 billion annually [ 3 ]. Depression is currently the fourth-ranked medical condition contributing to global burden of disease, and is estimated to rise to second overall by the year 2010 [ 4 ]. There are many effective treatments for MDD, including psychotherapy and antidepressants. Traditionally, efficacy in randomized controlled trials (RCTs) for depression has been determined on the basis of score changes in rating scales such as the Hamilton Depression Rating Scale (HDRS) [ 5 ] or the Montgomery-Asberg Depression Rating Scale (MADRS) [ 6 ]. Clinical outcome has been usually assessed by clinical response rates, typically defined as a 50% or greater reduction from baseline scores on these rating scales [ 7 ]. Although obtaining clinical response represents an important therapeutic milestone, it does not necessarily indicate a complete recovery from MDD, since many patients with clinical response will still be left with substantial residual symptoms of depression. Studies have shown that the presence of residual symptoms after an episode of MDD is associated with higher risk of relapse, recurrence, chronicity, suicide, development of cardiovascular disease, and poor quality of life [ 8 - 10 ]. Such findings suggest that the goals of acute treatment (approximately the first 8–12 weeks or so of treatment) for MDD should be clinical remission, a clinical state distinguished by minimal residual symptoms, rather than just response [ 11 - 13 ]. Clinical remission is typically defined as a score within the normal range on a given outcome measure (e.g., 17-item HDRS score of 7 or less; MADRS score of 12 or less; Clinical Global Impression [CGI] [ 14 ] severity score of "Normal, not at all ill"), although there is still some uncertainty as to the validity of these cutoff scores for symptom remission [ 15 ]. The achievement of remission is of considerable clinical importance as it predicts decreased risk of relapse and greater psychosocial functioning than typically observed in patients who have achieved clinical response alone [ 16 - 18 ]. Clinical remission is now identified and promoted as a clinical target for successful management of MDD in many clinical practice guidelines [ 13 , 19 - 21 ]. Increasing numbers of treatment studies are now explicitly reporting both clinical response and remission rates in assessment of outcome. A meta-analysis of 8 antidepressant studies of venlafaxine versus selective serotonin reuptake inhibitors [SSRIs] and placebo reported mean remission rates of 45%, 35%, and 25%, respectively [ 22 ]. A subsequent meta-analysis of 32 RCTs comparing venlafaxine, SSRIs and other antidepressants reported a mean overall remission rate of 42% [ 23 ]. Finally, a meta-analysis of 6 RCTs comparing antidepressants and psychotherapy in patients with MDD reported mean remission rates of 46% for each treatment [ 24 ]. All the studies in these systematic reviews involved patients in psychiatric or mixed settings. However, most people suffering from MDD will be managed in the primary care setting [ 25 ]. Approximately 5% to 10% of all patients consulting a general practitioner have MDD, with prevalence estimates being two to three times higher when other depressive disorders (i.e., minor depression or dysthymia) are included [ 26 ]. It remains unclear whether the remission rates reported in psychiatric settings can be extrapolated to primary care environments, although it is of clinical importance for primary care physicians to know whether obtaining remission is a realistic goal for their patients. There has been a recent surge in studies assessing a variety of treatment interventions for depression in primary care settings, making this an opportune time to perform a meta-analysis to address this question. Hence, the primary objective of this study was to determine remission rates obtained in RCTs of treatment interventions for MDD conducted in primary care settings. Methods Potentially relevant studies were identified using computerized and manual search strategies. The computerized search conducted in June, 2003 included the databases: Medline, Psych Info, Embase, Biosis, Cochrane Database of Systematic Reviews, and Cochrane Controlled Trials Register and Current Controlled Trials (1981–May 2003). The search terms used were 'depressive disorder' or 'depression' combined with 'primary care' and 'remission' and/or variants. The bibliographies of relevant articles were also manually searched. Two reviewers (MYD and RWL) collected and independently assessed abstracts for inclusion criteria. Disagreements were resolved with consensus. Inclusion criteria Studies were included if they were RCTs with original data comparing one or more interventions (e.g., antidepressant vs. cognitive behavioral therapy) and published in English. Only studies of predominantly adult populations, as opposed to exclusively child or elderly patient populations, were included. Although the focus was principally upon patients with MDD (studies primarily dealing with minor depression and dysthymia were excluded), the criteria for a diagnosis of MDD was intentionally broad in order to capture the heterogeneity of the sample and allow the results to be as generalizable as possible. Included studies also had to use a standardized outcome measure (e.g., HDRS, MADRS, Beck Depression Inventory [BDI] [ 27 ]) and provide explicit criteria for remission. While the definition of remission varied among the studies (Table 1 ), for the purpose of this meta-analysis we accepted each study's definition of remission, which usually was a score within the normal range on the outcome measure. Table 1 Summary of included studies in meta-analysis of remission rates. Study Diagnostic Criteria Follow up Period Remission Criteria Total N Intervention Intervention Remission Rate Remission % Psychological Intervention Only Dowrick et al., 2000 [31] DSM-IV criteria for MDD or Adjustment Disorder 6 months No MDD detected by SCAN interview 425 • PST • Usual Care •Prevention course • 58/128 • 76/189 • 44/108 • 45 • 38 • 41 Antidepressant Intervention Only Benkert et al., 2000 [32] DSM-IV criteria for MDD and HDRS ≥ 18 6 weeks HDRS ≤ 7 275 • Mirtazapine • Paroxetine • 52/139 • 42/136 • 37 • 31 Patris et al.,1996 [33] DSM-IIIR criteria for MDD 8 weeks MADRS ≤ 12 357 • Citalopram • Fluoxetine • 114/173 • 110/184 • 66 • 60 Wade et al., 2002 [34] DSM-IV criteria for MDD 8 weeks MADRS ≤ 12 380 • Escitalopram • Placebo • 92/191 • 64/189 • 48 • 34 Psychological Intervention + Antidepressants Chilvers et al., 2001 [35] Diagnosed as MDD by GP 12 months RDC <4, BDI <10, or clear documentation in GP notes that patient is well 103 Randomised only: • Antidepressant • Counselling • 39/51 • 33/52 • 76 • 63 Mynors-Wallis et al., 1995 [36] Diagnosed as MDD by GP 12 weeks HDRS ≤ 7 or BDI ≤ 8 91 • PST • Amitriptyline • Placebo • 18/30 • 16/31 • 8/30 • 60 • 52 • 27 Mynors-Wallis et al., 2000 [37] RDC criteria for MDD 12 months HDRS ≤ 8 151 • PST-group •PST-RN • Antidepressant • PST+antidepressant • 24/39 • 23/41 • 20/36 • 23/35 • 62 • 56 • 56 • 66 Schulberg et al., 1998 [38] DSM-IIIR criteria for MDD 8 months HDRS ≤ 7 184 • IPT • Nortriptyline • 49/93 • 52/91 • 57 • 53 Scott et al., 1992 [39] DSM-IIIR criteria for MDD 4 months HDRS ≤ 7 121 • CBT • Counselling • Amitriptyline • Usual care • 12/30 • 22/30 • 18/31 • 14/30 • 40 • 73 • 58 • 47 Program Interventions Katon et al.,1999 [40] Diagnosed as MDD by GP 6 months Presence of 0 or 1 SCID-assessed symptoms 228 • Collaborative care • Usual Care • 50/114 • 35/114 • 44 • 31 Katzelnick et al., 2000 [41] Diagnosed as MDD by GP and HDRS ≥ 15 12 months HDRS ≤ 7 407 • Depression management • Usual care • 92/218 • 49/189 • 42 • 26 Kutcher et al., 2002 [42] Diagnosed as MDD by GP 29 weeks 8 weeks or longer with HDRS ≤ 10 269 • Sertraline • Sertraline + adherence program • 84/138 • 88/131 • 61 • 67 Rost et al., 2002 [43] Diagnosed as MDD by GP 24 months CES-D ≤ 16 211 • Enhanced depression care • Usual care • 85/115 • 39/96 • 74 • 41 (Abbreviations: BDI – Beck Depression Inventory, CBT – Cognitive Behavioural Therapy, CES-D – Centre for Epidemiological Studies – Depression Scale, HDRS – Hamilton Depression Rating Scale, HSCL-D-20 – 20-item Hopkins Symptom Check List, IPT-Interpersonal Psychotherapy, MADRS – Montgomery-Asberg Depression Rating Scale, MDD – Major Depressive Disorder, PST – Problem Solving Therapy, PST-PC – Problem Solving Therapy, administered by Primary Care Physician, PST-RN – Problem Solving Therapy, administered by Registered Nurse, RDC – Research Diagnostic Criteria, SCAN – Schedules of Clinical Assessment in Neuropsychiatry, SCID – Structured Clinical Interview for DSM-III-R.) Data extraction Two independent reviewers (MYD and EEM) extracted data from studies using a checklist developed for this study, with disagreements resolved by a third reviewer (RWL). A conservative measure of remission rate was calculated from each study using an intent-to-treat analysis [ 28 ], even if this method was not used in the study. For example, some studies calculated remission rates using only patients who returned for one follow-up visit post-randomization, or who had completed a course of treatment. The denominator used for remission rate was the total number of patients randomized to treatment, whether or not they were counted in the ensuing analysis. The numerator was the number of patients in remission reported in the study, regardless of the denominator used in the study analysis. The type of intervention was classified as placebo, "usual care" by clinician (standard treatment by a patient's own physician), psychotherapy treatment only, antidepressant treatment only, psychotherapy plus antidepressant treatment, or program intervention (e.g., collaborative care using other health professionals; educational programs targeted at quality improvement for prescribing practices). Statistics Each set of rates was pooled based on a Bayesian approach to meta-analysis using the Fastpro software program (version 1.7) by Eddy and Hasselblad. Readers interested in a more detailed discussion of this approach should refer to Eddy et al [ 29 ]. The pooled means and confidence intervals were calculated using Jeffrey's prior and a random effects model. Results The initial electronic and bibliographic search found 63 articles of which 47 warranted more detailed review based on the published abstract. Of these, 34 articles were excluded due to methodology (not RCTs, 4 studies), lack of remission criteria (18 studies), diagnostic criteria (not MDD, 11 studies) and age criteria (geriatric, 4 studies) (some studies were excluded for multiple reasons, see Additional File 1 ). A final count of 13 studies met the full inclusion criteria (Table 1 ). In total, 3202 primary care outpatients (75% female, 25% male) were included in the analysis. The mean age of the participants was 32.1 years (range 18–73 years). The average length of follow-up was 32 weeks (range 6–104 weeks). The study interventions and methodologies were too heterogeneous to allow for a meaningful statistical comparison of results between treatments. Figure 1 shows mean remission rates for specific interventions. Overall remission rates for active interventions, regardless of type, ranged between 50% and 67%, compared to 32% for pill placebo conditions and 35% for usual care conditions. There were a sufficient number of antidepressant arms in the studies to permit the summary of remission rates by duration of follow-up period. For antidepressant studies with follow-up of 6 months or less, mean remission rate was 51.4% (95% C.I., 43.1%–59.6%); for antidepressant studies with greater than 6 months of follow-up, mean remission rate was 62.3% (95% C.I., 48.9%–74.8%). Figure 1 Remission rates for specific treatment conditions from randomized controlled trials (RCTs) of interventions for depression in primary care settings. The white lines represent the mean remission rates and the boxes represent the 95% confidence interval. N is the number of treatment arms in the RCTs (Note: Psychotx = Psychotherapy, Antidepr = Antidepressants, pts = patients). Discussion This review of research assessing remission of depressive symptoms in primary care populations identified 13 studies meeting the inclusion criteria. Overall remission rates (regardless of type of intervention but excluding placebo or usual care arms) ranged between 50% and 67%. These rates are equivalent to, or indeed greater than, those reported in meta-analyses of studies examining pharmacological or psychological interventions for depression in psychiatric populations, in which the overall remission rates ranged between 35% and 46% [ 22 - 24 ]. On the one hand, we might have predicted this finding as studies conducted in primary care settings tend to include more patients with mild to moderate depression (although we excluded studies that focused exclusively upon minor depression or dysthymia), whereas patients referred to psychiatric settings are more likely to have moderate to severe depression. Primary care treatment trials also tend to be longer, favouring a higher remission rate; whereas the mean follow-up period of studies included in the current analysis was 9 months, it was only 7 weeks and 10 weeks in the two previous meta-analyses of pharmacological interventions for MDD [ 22 , 23 ], and 16 weeks in the meta-analysis of antidepressant versus psychotherapeutic interventions [ 24 ]. Conversely, we might have predicted that we would observe lower remission rates in the current meta-analysis as it included a number of studies with more lenient exclusion criteria than typically used in psychiatric clinical trials. In particular, the program intervention studies tend to include more heterogeneous patient populations as they do not routinely exclude patients with psychiatric or medical comorbidities, factors that may lessen the likelihood of obtaining remission of depressive symptoms [ 30 ]. While it was not within the scope of the current study to compare the effectiveness of different treatment interventions in improving remission rates, we can report on the trends we observed in the data. Antidepressant and psychotherapy interventions delivered in isolation showed similar remission rates (54% for both). Combination antidepressant plus psychotherapy interventions showed somewhat higher rates (67%), although this category included only 1 arm with only 35 patients. Program interventions had a mean remission rate of 50%, and all treatment interventions fared better than either placebo (32%) or usual care (35%). The studies identified in our review were quite heterogeneous in nature, ranging from those that looked solely at the effects of a particular pharmacological agent, through to complex program initiatives that incorporated a variety of interventions at different levels of care. This heterogeneity limits our ability to make broad comments about remission rates in primary care, but was not unexpected, as we wanted to capture the diversity of treatment interventions for depression currently being tested in this setting. Other potential limitations of the study include that fact that we only assessed published studies written in English and that we used a conservative measure of remission rate. Finally, we also used the definition of remission as specified by each individual study. While these definitions were similar to those widely used in RCTs conducted in psychiatric settings, and thus are useful for comparison, there is current controversy about depression scales and which cutoff scores indicate true remission of symptoms [ 15 ]. Conclusions This meta-analysis serves to answer an important clinical question about the feasibility of obtaining remission of symptoms of MDD in primary care patients. Our results indicate that this is a realistic goal in this population, although further research is still required to determine whether certain treatment modalities (or combinations of treatment interventions) are superior to others in achieving higher remission rates. Future research should also focus upon developing pragmatic strategies for general practitioners to implement evidence-based guidelines concerning the treatment of depression to clinical remission. Authors' contributions MYD and EEM conducted the data extraction, wrote the initial draft of the manuscript, interpreted results, and revised the manuscript. PW provided statistical consultation and analysis, and revised the manuscript. JEA interpreted the results and revised the manuscript. RWL conceived the initial idea, developed the method, interpreted results, revised the manuscript, and provided financial resources for the study. All authors read and approved the final manuscript. Competing interests RWL is on advisory/speaker boards or has received research funds from: AstraZeneca, Biovail, Canadian Network for Mood and Anxiety Treatments, Eli Lilly, GlaxoSmithKline, Janssen-Ortho, Litebook, Inc., Lundbeck, Merck, Organon, Roche, Shire, Servier, and Wyeth. Pre-publication history The pre-publication history for this paper can be accessed here: Supplementary Material Additional File 1 Studies excluded from the review Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC518964.xml
406409	The European Research Council—A European Renaissance	European scientists are pressing for the creation of an independent body to fund European research - driven by the pursuit of scientific excellence	Science looks set for a fundamentalist revival within the European Union. Its leading proponents are taking advantage of unprecedented political upheaval—as ten new Member States accede to the Union—to press their case for funding of basic research that is driven solely and independently by investigators themselves in the pursuit of excellence. Bob May, professor of mathematical biology at the University of Oxford, president of the Royal Society, and former UK Chief Scientist The broad thrust of their appeal calls for the setting up of a new agency, most commonly referred to as a ‘European Research Council’. The ERC could be an entirely new organisation or a new division within an established body, run by a small staff able to draw on the best expertise available. It would administer a new fund from EU coffers, tagged the European Fund for Research Excellence, that would be valued modestly, initially at least, at much less than half of the EU's existing budget for research. Most importantly, dispersal of that fund would reflect the wishes of eminent peer reviewers, assessing competitive bids in search of the best science, rather than the judgements of Eurocrats, looking for the most politically and economically expedient solutions and operating on a lead time of two years or more. Although the modus operandi of the proposed ERC has still to be worked out, European scientists have been looking to the United States and at the way that the National Science Foundation and the National Institutes of Health operate, as well as to private institutions such as the Howard Hughes Medical Institute in the United States and the Wellcome Trust in the United Kingdom. In particular, they seek the independence and excellence achieved outside of the EU framework. More to the point, they are weary of the bureaucratic formulations that determine how the EU's research budget, currently known as the Sixth Framework Programme (2002–2006) and worth around €4.4 billion/year (or just over 5% of all public spending on nonmilitary research in the region), is spent and distributed. The EU's guiding principle is often one of juste retour , or fair reward, in which Member States traditionally seek to recover grants at least equal to their contributions to the EU pot (see Box 1 ). Bernard Larrouturou, director general of France's National Centre of Scientific Research (CNRS) in Paris ‘Most of the Anglo-Saxon countries in Europe—the Scandinavian countries, the United Kingdom, the Netherlands—operate a peer review process and a research funding council process that's very similar to best practice in North America,’ says Michael Morgan, a consultant to the Wellcome Trust on European issues and former chief executive of the Trust's Genome Campus at Hinxton, near Cambridge, United Kingdom. ‘The French and Germans and others have elements of that but they also have what you might call more “state-funded science”, scientists as civil servants, and there is obviously much greater possibility of science being funded for less than the best scientific reasons,’ notes Morgan, referring to the opportunities for greater political influence on decision-making. ‘I'm not suggesting that that is the case, but it is the possibility,’ he adds. ‘What we need in Europe is something that should strictly adhere to the international standards of research funding and be evaluated by peer review,’ says Peter Gruss, professor of molecular cell biology at the University of Göttingen and president of the Max Planck Society in Munich, Germany. ‘The sole criterion has to be quality, not geographical distribution, not management capacity,’ he adds, alluding to the EU practice of juste retour . ‘We want to encourage excellence in Europe. We want to have as a benchmark a European standard that should be as high as the standard is in the US.’ Kai Simons, president of the ELSO and director of the Max Planck Institute for Cell Biology and Genetics Gruss acknowledges the tensions that the ERC proposal has generated among Member States: ‘I'm not saying that there aren't countries that have this standard—like the UK, parts of Germany, Sweden, and some other Nordic countries—but of course this is not the general European standard, and in order to get one and the same, the common standard, we need a common structure.’ A Fund for Excellence The European Commission now appears ready to accept the need for a common structure that would have, as the Commission puts it, ‘more open and less binding’ programmes of basic research, in contrast to the Framework Programme, whose emphasis is on applied research with commercial objectives. The Commission expects to publish its endorsement of the ERC proposal this month, so that approval by the Council of the EU should follow later this year. On this timetable, setting up of the ERC could begin in 2006 when the next five-year Framework Programme, FP7, gets underway. Over the ERC's first five years, its grant is expected to grow from around €500 million/year to €2 billion/year, and to derive from a reallocation of funds within the EU's budget rather than from any top-up contributions from Member States. Furthermore, Gruss released a legal opinion in March that advised how an ERC need not be an executive agency of the Commission, as many scientists had feared it would have to be under the terms of the EU Treaty, but could be established as an independent and autonomous body. The opinion is a real coup for the ERC lobbyists. Origins of the ERC Moves to establish an ERC are founded in a ‘new strategic goal’ for the EU that the leaders of its 15 Member States set during their European Council in Lisbon in March 2000. Over the first decade of the new millennium, they urged the EU ‘to become the most competitive and dynamic knowledge-based economy in the world’. They enthusiastically endorsed a notion, floated by the European Commission, of a European Research Area (ERA). ‘Research activities at national and [European] Union level must be better integrated and co-ordinated to make them as efficient and innovative as possible, and to ensure that Europe offers attractive prospects to its best brains,’ concluded the EU leaders, eager to reverse the flow of trained talent abroad, notably to North America. All appropriate means, they added, ‘must be fully exploited to achieve this objective in a flexible, decentralised and non-bureaucratic manner’. Two years later, at the European Council in Barcelona, the EU leaders went one step further by defining the target more precisely. ‘In order to close the gap between the EU and its major competitors,’ they said, ‘overall spending on R & D and innovation in the Union should be increased with the aim of approaching 3% of GDP by 2010. Two-thirds of this new investment should come from the private sector.’ The scale of the challenge is illustrated by the latest figures for R & D expenditure, published in February by the Statistical Office of the European Communities (Eurostat). The EU's estimated R & D spending in 2002 was 1.99% of GDP, still far short of the US (2.80%) and Japan (2.98% in 2000), and a long way from the target of 3%. Emphasising the UK's uneasiness about the EU's escalating enthusiasm for a regional science base, the Royal Society (the UK national academy of science) poured scorn on the ‘ambitious’ GDP target by noting how the UK alone would have needed an extra £11 billion in 2000, or more than 60% of total spending on R & D, to lift its ratio of 1.85% to the 3% target. The Royal Society also noted how public funding of R & D in the EU matches that in the US and Japan, with the disparity among GDP ratios reflecting the differentials in private investment in R & D, over which the EU has little control. Nevertheless, the challenge could not be ignored. According to Bob May, professor of mathematical biology at the University of Oxford, president of the Royal Society, and former UK Chief Scientist, such initiatives might be ‘driven more by political expediency than common sense, but the moment you see that train beginning to roll, there's a chance to do something useful with it’. Among the leading proponents of an ERC is Bernard Larrouturou, director general of the National Centre of Scientific Research (CNRS) in Paris, France. For Larrouturou, a biomathematician currently engaged in streamlining the organisation, the changes at the European level are a breath of fresh air. However, he is not convinced that funded investigators should expect to exclude Commission strategists entirely from their lives. The scientific community should lead an ERC, says Larrouturou, ‘but I do not like the idea that this should be completely under the guidance and wisdom of the scientific community with no strategy guidance. You cannot ask for 1 or 2 billion Euros every year and say there will not be any strategy and [that it will be done solely] on this basis of excellence.’ And Larrouturou distances himself from the idea that basic and applied research can be treated separately because this suggests, wrongly he says, a conflict between the two. On these issues, Larrouturou moves onto some common ground with John Taylor, former director general of Research Councils UK and now chairman of Roke Manor Research, a UK subsidiary of Siemens, the German electronics group. Research Councils UK oversees spending of Britain's national research councils (currently, just over £2 billion from its 2004–2005 Science Budget of nearly £2.7 billion). Interactions across disciplines and between scientists and technologists ‘are not helped by making artificial distinctions between this kind of research and that kind of research,’ says Taylor. ‘The distinctions I make are much more between top-down and bottom-up.’ While Taylor is a joint architect of one proposal to create an ERC, he remains unconvinced that the research funding system is broken, especially from the UK's perspective, and needs to be fixed. Nor is he convinced that EU funds for an ERC will not affect national R & D budgets. ‘I'm middle of the road,’ he says. ‘Much greater collaboration is good. It has to be a slow process, with all the different cultures involved. Collaboration on various areas of science is an excellent way to go, provided you don't try to organise it from the top and legislate for it all to happen in a particular way and to a particular timescale. Excellence is key.’ Taylor's cautions reflect his experience of the EU's Framework Programme and his reservations that any initiative from Brussels can be free of red tape. ‘If you want to do research, then you can't lay out beforehand all the answers you're going to get,’ he says. ‘And if you try to get people to stick rigorously to a plan, then you get a lot of silly things going on. If you try to form very complex bureaucratic organisations to do the research, you get a lot of delays and so on, so things are not very timely.’ But the Framework Programme's failures need not spell disaster for the fledgling funding council, insists Lennart Philipson, former director general of the European Molecular Biology Laboratory (EMBL) in Heidelberg, Germany, and now an emeritus professor at the Karolinska Institute in Stockholm, Sweden. Drawing on his 11 years as head of EMBL, until 1993, Philipson recalls how ‘pan-European peer review was the best method for distributing the funds of EMBL and EMBO [European Molecular Biology Organization]’. The continuing high status of the two organisations, he says, is testimony that the system works. In fact, EMBO is mentioned as a possible incubator for an ERC, in spite of its specialisation. Other proponents of the proposed research changes in the EU include 45 Nobel Laureates from Europe or of European origin, who headed a petition organised by EMBO. The European Life Scientists Organization (ELSO) organised another. Its president, Kai Simons, also the Director of the Max Planck Institute for Cell Biology and Genetics in Dresden, Germany, says research funding in Europe is just not working. ‘It's not geared for basic research—it has other aims,’ he notes. EU funds are ‘not grants, they are contracts with in-built milestones that have nothing to do [with basic research]. Basic research doesn't work like that.’ The evaluation and peer review system is falling apart, continues Simons. He says that the best people are not interested in peer reviewing a system that doesn't work: ‘You're not attracting the peer reviewers that you need to maintain quality.’ But at last, acknowledges Simons, someone in Brussels is listening. ‘In the past two years there has been enormous progress.’ Many Questions Remain Within a month of the Barcelona Council in 2002, the European Science Foundation (ESF), which brings together the funding agencies of 29 countries and acts as a bridge to Brussels, had formed a High Level Working Group to review the case for an ERC and how it might operate. The group, chaired by Sir Richard Sykes, Rector of Imperial College, London, United Kingdom, reported a year later, in April 2003. It endorsed the creation of an ERC as ‘the cornerstone for the ERA and the key approach to developing a locus for…long-term fundamental curiosity-driven research judged on the basis of excellence and merit’. The Sykes group also proposed, controversially, an enhanced ESF as the most effective medium for establishing an ERC swiftly. ‘Some people say that the ESF has no experience in funding large amounts… for research,’ acknowledges Enric Banda, director general of the Catalan Research Foundation in Barcelona, Spain, who finished a five-year term as the ESF's chief executive at the end of 2003 and is credited with ‘waking up’ the foundation. ‘But certainly if you create a new [organisation], that's the same thing. So the ESF is in a good position because its member organisations are the funding agencies.’ Bertil Andersson, who was a member of the Sykes Group before taking over from Banda at the ESF in January, also stakes the ESF's claim to nurture a fledgling ERC. But he accepts that any one of the respected national funding agencies, such as the German Research Foundation (DFG), or even a specialist body, such as EMBO, could do the job. ‘We don't need a new skyscraper in Brussels, but a lot of… peer review and running of the ERC could be done by existing bodies. ‘Compared to soccer, we have only the national leagues—we don't have the Champions League [the league of Europe's best teams],’ says Andersson. There is no competition for basic research grants across national boundaries in Europe, he insists. ‘The Swedish league is exciting, but the Champions League is more exciting.’ In the meantime, while the Sykes group was still deliberating, the Council of the EU appointed another group of experts to evaluate the case for an ERC. Chaired by Federico Mayor, former director general of the United Nations Educational, Scientific, and Cultural Organization (UNESCO), the ERC Expert Group also delivered its verdict—a resounding endorsement—within 12 months. ‘The first and main task for the ERC should be to support investigator-driven research of the highest quality selected through European competition,’ concluded the Mayor report, published in December 2003. ‘In doing so, the ERC should create and support nodes of excellence in European universities and research institutions, strengthening the knowledge-base that underpins economic, industrial, cultural and societal development, and thereby stimulating European competitiveness and innovative capacity at all levels.’ While few disagreed with the Mayor report's sentiments, the absence of a detailed analysis exposed underlying tensions over the rationale for an ERC. In the UK, in particular, some scientists seemed concerned that their mature and respected system for funding research risked dilution. ‘The British have always had doubts about what goes on in Europe,’ notes Kai Simons. ‘They always think that they can do it better. But the big problem for the British is that they are also too small to fund a new innovative area,’ he says. ‘Of course, we can do it without Britain, but they are an important part of Europe and it would be sad if they're not part of it.’ The agnostic John Taylor, who was a member of the Mayor group, recalls his early reservations when the group convened. ‘I'm way beyond the euphoria; I'm into practical pragmatics,’ he notes. ‘My major input into the whole thing has been to get them to “get real” instead of just philosophising. They've been using the sort of, dare I say it, Gallic approach… of thinking about the reasons why, and the philosophy, and not thinking about what you would actually do.’ Taylor dismisses the notion that wariness of the ERC is representative of a general antipathy in Britain towards European integration. ‘What we're saying is that science in the UK is not yet well-funded enough to say we would rather do this [the ERC] instead of the things that we're already trying to get done in the UK scene.’ Anticipating the Mayor report's publication, the Royal Society quickly pulled together a detailed background paper late last year that identified ‘a number of problems that need resolution, although not necessarily through the establishment of any major new institutions within Europe’. An addendum followed in March, in direct response to the Mayor report. That addendum highlighted what it saw as the paucity of solid evidence in the Mayor report and, in some cases, the confusing data in the report's case for an ERC. On balance it looked as though the Royal Society, and as such the British science establishment as a whole, had weighed the disadvantages of an ERC as greater than its advantages, but Bob May is quick to refute this charge. ‘My vision and the Royal Society's vision of the ERC is that it will fund the very best science,’ he insists. ‘The Mayor committee itself was really good people who'd produced basically a good report…. I'm basically in favour of this European Research Council… provided it can be set up properly, which is by no means certain.’ For May, and other scientists on the continent, the ERC offers a real chance to redress the balance of fortune in favour of young scientists. ‘The way to encourage science is to get the best people and set them free to express their creativity while they are young, which means bring them into the best laboratories—don't let them get entrained in hierarchies of deference to second-rate people,’ says May. ‘The most important single thing to create one Europe in science is a flexible postdoctoral programme that gets the best young people wherever they are and lets them go to the best places,’ enthuses May. An ERC will then foster those collaborations, he forecasts. ‘It won't ask whether they're juste retour , whether they're serving some industrial purpose, it will just try to fund the best science. But I hope increasingly the best projects will involve collaborations, as they do in Britain, collaborations among institutions within Europe.’ Box 1. Glossary of Europe Council of the European Union – Ruling organisation (along with European Parliament), and not to be confused with the European Council (see below). It comprises ministers from governments of the Member States, which have varying voting powers led by France, Germany, Italy, and the UK. Euro (€) – Common European currency launched on 1 January 2002 in 12 participating Member States (the UK, Sweden, and Denmark chose to postpone adoption of the currency indefinitely). European Commission – Executive organisation, mainly based in Brussels, run by 20 Commissioners and around 24,000 civil servants. European Council – Body that brings together leaders of Member States to define broad policy objectives for the EU's six main institutions (Parliament, Council, Commission, Court of Justice, Court of Auditors, and Ombudsman). Meets twice a year in the Member State holding the Council's presidency, which changes every six months. European Parliament – Elected organisation, based in Strasbourg, France, that rules the EU (jointly with the Council of the EU, see top) and will have 732 Members after the accession of the ten new Member States in May 2004. European Research Area – Commissioner Philippe Busquin's vision for the future of research in Europe, and the main focus of the 6th Framework Programme. It aims to achieve ‘scientific excellence, improved competitiveness and innovation through the promotion of increased co-operation, greater complementarity and improved co-ordination between relevant actors, at all levels’. European Union – Evolving political, social, and economic union of an increasing number of European countries, or Member States. First proposed in 1950 during rehabilitation after the Second World War and formally created by the Maastricht Treaty in 1992. Grew from six nations in 1951 (Belgium, France, Germany [then West Germany], Italy, Luxembourg, and the Netherlands) to nine in 1973 (addition of Denmark, Ireland, and the UK), to ten in 1981 (addition of Greece), to 12 in 1986 (addition of Spain and Portugal), to 15 in 1995 (addition of Austria, Finland, and Sweden), with a total population of 380 million people (cf. 290 million for US; 130 million for Japan). Ten more countries (Cyprus, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Malta, Poland, Slovakia, and Slovenia) join in May 2004, which will lift the EU's population to 450 million people. Bulgaria and Romania are due to join in 2007, which will add another 50 million people. Framework Programme – The EU's principal mechanism for funding research in Member States, proposed by the Commissioner for Research (Philippe Busquin) and adopted by the Council and Parliament. Framework Programmes have four-year budgets but cover five-year periods, so consecutive programmes overlap, and are prescribed two years before they begin. The 6th programme (FP6) is worth €17.5 billion (or about 4% of the EU's total budget and 5.4% of all public, nonmilitary research spending in Europe) and runs from the beginning of 2003 to the end of 2006. Juste retour (fair reward) – Claim made by Member States for rewards at least equal to their share of the cost of any programme or initiative; critics say it promotes bureaucracy and uncompetitiveness.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC406409.xml
516791	Human oligodendroglial cells express low levels of C1 inhibitor and membrane cofactor protein mRNAs	Background Oligodendrocytes, neurons, astrocytes, microglia, and endothelial cells are capable of synthesizing complement inhibitor proteins. Oligodendrocytes are vulnerable to complement attack, which is particularly observed in multiple sclerosis. This vulnerability may be related to a deficiency in their ability to express complement regulatory proteins. Methods This study compared the expression level of complement inhibitor mRNAs by human oligodendrocytes, astrocytes and microglia using semi-quantitative RT-PCR. Results Semi-quantitative RT-PCR analysis showed that C1 inhibitor (C1-inh) mRNA expression was dramatically lower in oligodendroglial cells compared with astrocytes and microglia. The mRNA expression level of membrane cofactor protein (MCP) by oligodendrocytes was also significantly lower than for other cell types. Conclusion The lower mRNA expression of C1-inh and MCP by oligodendrocytes could contribute to their vulnerability in several neurodegenerative and inflammatory diseases of the central nervous system.	Background Resident brain cells including oligodendrocytes [ 1 , 2 ], astrocytes, astrocytomas, microglia, glioblastomas [ 3 - 14 ], neurons [ 15 , 16 ], neuroblastomas [ 17 , 18 ] and endothelial cells [ 19 , 20 ] express mRNAs for complement proteins. Although the role of complement expression by these cells remains unclear, local complement activation in the central nervous system (CNS) might damage these cells and contribute to the pathology in several inflammatory and neurodegenerative diseases including multiple sclerosis, Alzheimer's disease and progressive supranuclear palsy. For self-protection, resident brain cells also express complement inhibitors, such as membrane cofactor protein (MCP), decay-accelerating factor (DAF), CD59, and C1-esterase inhibitor (C1-inh). The human HOG oligodendroglial cell line produces MCP, DAF, CD59, C1-inh and S-protein, but not complement receptor 1 (CR1) [ 1 ]. Human oligodendrocytes have been reported to express CD59 [ 21 ] and DAF, but not MCP, CR1, homologous restriction factor (HRF: C8 bp) or clusterin [ 22 ]. Astrocytes [ 23 ], neurons and Schwann cells have been reported to express CD59 [ 24 ] and neuroblastoma cell lines C1-inh [ 18 ]. Astrocytoma cell lines have been reported to express MCP, DAF, and CD59 [ 25 , 26 ]. In this study, the expression level of mRNAs for various complement inhibitors by human oligodendrocytes, astrocytes and microglia were compared by semi-quantitative PCR. We show that oligodendrocytes express extremely low levels of mRNA for C1-inh and significantly lower levels of mRNA for MCP compared with astrocytes and microglia. The expression level of mRNAs for CD59 and DAF showed no significant differences between the three cell types. Methods Cell culture: microglial- and astrocyte-enriched cultures Human microglial and astrocytic cells were isolated from surgically resected temporal lobe tissues. We thank Dr. J. Maguire, Department of Pathology and Laboratory Medicine, Vancouver General Hospital for providing the surgical specimens. Isolation protocols described by De Groot et al . [ 27 , 28 ] were used with minor modifications. Tissues were placed in a sterile Petri dish, rinsed with Hank's balanced salt solution, and visible blood vessels were removed. After washing tissues two more times with Hank's balanced salt solution, tissues were chopped into small (<2 mm 3 ) pieces with a sterile scalpel. The fragments were transferred into a 50 ml centrifuge tube containing 10 ml of 0.25% trypsin solution (Gibco-BRL, Life Technologies, Burlington, ON, Canada), and incubated at 37°C for 20 min. Subsequently DNase I (from bovine pancreas, Pharmacia Biotech, Baie d'Urfé, PQ, Canada) was added to reach a final concentration of 50 μg/ml. Tissues were incubated for an additional 10 min at 37°C. The cell suspension was diluted with 10 ml of Dulbecco's modified Eagle's medium (DMEM) and nutrient mixture F12 ham (DMEM-F12; Sigma-Aldrich, Oakville, ON, Canada) with 10% fetal bovine serum (FBS; Gibco-BRL, Life Technologies), and gently triturated by using a 10 ml pipette with a wide mouth. After centrifugation at 275 × g for 10 min, the cell pellet was resuspended in serum containing medium, triturated several times, and passed through a 100 μm nylon cell strainer (Becton Dickinson, Franklin Lakes, NJ). The cell suspension was then centrifuged once more (275 × g for 10 min), resuspended into 10 ml of DMEM-F12 with 10% FBS containing gentamicin (50 μg/ml, from Sigma), and plated onto uncoated 10 cm tissue culture plates (Becton Dickinson). Plates were placed in a humidified 5% CO 2 , 95% air atmosphere at 37°C for 2 hr in order to achieve adherence of microglial cells. Non-adherent cells with myelin debris were removed from these microglia-enriched cultures and transferred into poly-L-lysine coated 10 cm tissue culture plates in order to achieve adherence of astrocytes. Plates were incubated for 48 hr, after which the culture medium containing myelin debris and non-adherent cells was removed and used to prepare oligodendroglial cell cultures as described below. Both microglial- and astrocyte-enriched cultures were grown for 6 to 7 days before their mRNAs were extracted. Immunostaining with antibodies against CD68 (Dako, Mississauga, ON, Canada) which stains microglia as well as macrophages, and glial fibrillary acidic protein (GFAP, Dako), which is a marker of astrocytes, showed that the microglia-enriched cultures contained 93.5 ± 3.6 % (N = 4) microglial cells, while astrocyte-enriched cultures contained 85.7 ± 3.4 % (N = 4) astrocytes. Cell culture: oligodendroglial cells These were prepared as described before [ 2 ]. Briefly, cell culture media containing myelin debris and non-adherent cells that were removed from astrocyte-enriched cultures were used to extract oligodendroglial cells. The non-adherent cells were collected by centrifugation at 275 × g for 10 min and replated onto uncoated 10 cm tissue culture plates for another 24 hr. Subsequently, the cell culture medium containing floating cells was transferred to 50 ml tubes and Lymphoprep solution (Axis-Shield, Oslo, Norway) used to reduce the amount of contaminating myelin debris. For this purpose, 10 ml of Lymphoprep solution was carefully placed under the oligodendrocyte cell suspension and the density gradient was centrifuged at 325 × g for 10 min. The interphase was collected and transferred to a 50 ml centrifuge tube. Fresh culture medium was added and the suspension was centrifuged at 275 × g for 7 min. The cell pellet was resuspended and the oligodendrocyte cultures seeded onto 60 mm plastic culture dishes. Immunostaining with anti-O4 antibody (Chemicon International, Temecula, CA), which is a marker of oligodendrocytes, showed that the oligodendrocytes-enriched cultures contained 95.3 ± 4.4 % (N = 4) oligodendrocytes. RNA isolation and cDNA synthesis by reverse transcription Total RNA from oligodendroglial cells, microglia, and astrocytes were isolated by the acid guanidium thiocyanate-phenol-chloroform method. Two μg of the RNA was then used to prepare cDNA. RNA was treated with 10 U of DNase I (Gibco BRL, Life Technologies) for 60 min at 37°C in 25 μl of 1 × reverse transcriptase buffer (50 mM Tris-HCl, 75 mM KCl, 3 mM MgCl 2 ) containing 40 U of RNase inhibitor (Pharmacia Biotech) and 1 mM dithiothreitol (DTT), following by incubation at 85°C for 5 min to inactivate the enzyme. Reverse transcription was performed at 42°C for 90 min in 50 μl of the following mixture: 1 × reverse transcriptase buffer containing 2 μg of RNA, 5 mM DTT, 0.2 μg random hexamer primers (Pharmacia Biotech), 1 mM deoxynucleotides (Gibco BRL, Life Technologies), 40 units of RNase inhibitor, and 400 units of SuperScript II reverse transcriptase (Invitrogen Life Technologies, Burlington, ON, Canada). At the end of the incubation period, the enzyme was inactivated by heating at 65°C for 10 min [ 29 ]. Polymerase chain reaction PCR amplification was carried out in a 25 μl reaction mixture containing 1 × GeneAmp PCR buffer II (Perkin Elmer, Foster City, CA), 1.25 units AmpliTaq Gold DNA polymerase (Perkin Elmer), 2 mM MgCl 2 (Perkin Elmer), 200 μM dNTPs (Gibco BRL, Life Technologies) and 0.5 μM of each specific primer (Table 1 ). The mixture was prepared before the addition of 1.25 μl of cDNA. PCR amplification was carried out using an MJResearch (Boston, MA) programmable thermal controller. The amplification program consisted of an initial denaturation step at 94°C, which was extended to 9 min in order to activate AmpliTaq Gold enzyme. The remaining cycles were 1 min at 94°C, 1 min at 55°C and 1 min at 72°C. The number of cycles performed was 27 for glyceraldehyde-3-phosphate dehydrogenase (G3PDH), 30 for CD59, C1-inh and MCP, and 32 for DAF. After amplification, PCR products were separated on a 6% polyacrylamide gel and visualized by incubation for 10 min in a solution containing 10 ng/ml of ethidium bromide. Polaroid photographs of the gels were taken. Table 1 Oligonucleotide primers used for PCR, and the corresponding restriction endonucleases used for product confirmation. Gene Sequence (5' → 3') Fragment size (introns) Genbank accession No Restriction enzymes used and the expected sizes of digestion products (bp) C1 inh-F GTT GGG GGA TGC TTT GGT AGA TTT C 332 M13690 Sau 3AI (246, 86) C1 inh-R TTA GGA CTC TGG GGC TGC TGC TGT A (2 introns) CD59-F CTG CTG CTC GTC CTG GCT GTC TTC T 280 M34671 Pst I (233, 47) CD59-R TCC CAC CAT TTT CAA GCT GTT CGT T (2 introns) MCP-F CAA TTC AGT GTG GAG TCG TGC TGC 265 Y00651 Sau 3AI (193, 72) MCP-R TGA GGC ACT GGA CGC TGG AGA T (unknown) DAF-F GTA CTG TGA ATA ATG ATG AAG GAG 364 M30142 Hae III (330, 34) DAF-R TCT TAA CTC TTC TTT GGC TAA GTC (unknown) G3PDH-F CCA TGT TCG TCA TGG GTG TGA ACC A 251 X01677 Dde I (168, 83) G3PDH-R GCC AGT AGA GGC AGG GAT GAT GTT C (2 introns) PCR primer design and restriction analyses Primers were designed to span introns so that cDNA-derived PCR products would be of different sizes to those produced if genomic DNA was amplified (see Table 1 ). DAF and MCP were exceptions, since only cDNA sequences were available. Primers were synthesized either by Sigma-Aldrich or ID Labs (London, ON, Canada). The primer sequences and predicted PCR fragment sizes are listed in Table 1 , along with the names of the enzymes used for restriction digest analysis of each PCR fragment. The restriction digestion reactions were carried out at 37°C for 2 hr in the presence of 1 × the appropriate buffer provided by the suppliers (Invitrogen, Life Technologies and New England Biolabs, Mississauga, ON, Canada). The digested PCR products were analyzed on a 6% polyacrylamide gel (data not shown). In all cases the restriction fragments observed were of the predicted size (see Table 1 ). Statistical analysis The data are presented as means ± s.e.m. The significance of difference between values was estimated by means of one-way analysis of variance (ANOVA) with Fisher's LSD post-hoc test. P < 0.05 was considered to show statistically significant differences. Double fluorescence immunocytochemical analysis Oligodendrocytes, astrocytes, and microglia were harvested and air-dried on glass slides. Cells were then fixed with 4% paraformaldehyde for 10 min and permeabilized with 0.2% Triton X-100 in phosphate-buffered saline (PBS) for 5 min. For inactivation of endogenous peroxidase, cells were incubated with 0.3% H 2 O 2 for 30 min. Blocking was performed for 1 hr at room temperature in 5% skim milk. For double fluorescence immunostaining, cells were incubated at room temperature overnight with a primary antibody in 1% normal serum. The primary antibody and the dilution used in the first cycle were as follows: O4 (Chemicon International, 1: 100) for oligodendrocytes, GFAP (Dako, 1: 10,000) for astrocytes, CD68 (DAKO, 1: 50) for microglia. Cells were then treated for 2 hr at room temperature with a biotin conjugated anti-mouse IgM (Vector Laboratories, Burlingame, CA, 1: 200) secondary antibody for O4, a biotin conjugated anti-rabbit IgG (Vector Laboratories, 1: 200) secondary antibody for GFAP and a biotin conjugated anti-mouse IgG (Vector Laboratories, 1: 200) secondary antibody for CD68. Then cells were incubated with Texas Red Avidin DCS (Vector Laboratories) for 1 hr. The primary antibody and the dilution used in the second cycle were as follows: for C1-inh, goat anti-C1-inhbitor (Quidel, San Diego, CA, 1: 50); for CD59, mouse anti-CD59 (Serotec Ltd, Oxford, UK, 1: 10) or rat anti-CD59 (Serotec, 1: 25). Cells were incubated at 4°C for 3 days with a primary antibody in 1% serum corresponding to the secondary antibody type. Cells were then treated for 2 hr at room temperature with FITC-conjugated anti-mouse IgG (Vector Laboratories, 1: 200), anti-goat IgG (Santa Cruz Biotechnology, Santa Cruz, CA, 1: 200), or anti-rat IgG (Cappel, Durham, NC, 1: 200). The glass slides were then rinsed with distilled water, and a drop of Vectashield mounting medium (Vector Laboratories) placed on the slide. Results RT-PCR RT-PCR was carried out using primers for C1-inh, CD59, DAF and MCP. The housekeeping gene G3PDH was amplified in parallel with each RT-PCR run as an internal standard. Figure 1 illustrates the bands obtained for each of the RT-PCR products from oligodendrocytes (Fig. 1A ), astrocytes (Fig. 1B ) and microglia (Fig. 1C ). Specificity of each of the products was established by endonuclease digestion (Table 1 ). Figure 1 Demonstration of RT-PCR products. Polaroid photographs of typical ethidium bromide-stained gels of RT-PCR products from oligodendrocytic (Fig. 1A), astrocytic (Fig. 1B) and microglial (Fig. 1C) RNA extracts. Lanes for individual mRNA products are indicated in the legend at the top. Size markers are in the right lanes. MCP, membrane cofactor protein (265 bp); DAF, decay-accelerating factor (364 bp); CD59 (280 bp); C1-inh, C1-esterase inhibitor (332 bp); G3PDH, glyceraldehyde-3-phosphate dehydrogenase (251 bp). Semi-quantitative RT-PCR analysis To compare the ratio of each of the complement inhibitors to G3PDH, statistical analysis was performed by means of one-way ANOVA with Fisher's LSD post-hoc test (Fig. 2 ). The overall mean ± s.e.m. for C1-inh/G3PDH was 0.55 ± 0.12 (N = 5) in astrocytes, 0.58 ± 0.09 (N = 3) in microglia and 0.09 ± 0.06 (N = 12) in oligodendrocytes (Fig. 2A ). Oligodendrocytes showed a highly significant difference from astrocytes and microglia (Fig. 2A ; P < 0.001 by one-way ANOVA with Fisher's LSD post-hoc test). For MCP/G3PDH, the ratios were 0.80 ± 0.22 (N = 5) in astrocytes, 0.93 ± 0.10 (N = 3) in microglia and 0.44 ± 0.19 (N = 12) in oligodendrocytes. Oligodendrocytes showed a significant difference from astrocytes and microglia (Fig. 2B ; P = 0.002 vs. astrocytes and P = 0.001 vs. microglia by one-way ANOVA with Fisher's LSD post-hoc test). The corresponding means for CD59/G3PDH were 0.73 ± 0.10 (N = 5) in astrocytes, 0.83 ± 0.03 (N = 3) in microglia and 0.76 ± 0.09 (N = 14) in oligodendrocytes (Fig. 2C ). The corresponding means for DAF/G3PDH were 0.67 ± 0.07 (N = 5) in astrocytes, 0.67 ± 0.07 (N = 3) in microglia and 0.66 ± 0.15 (N = 14) in oligodendrocytes (Fig. 2D ). There were no significant differences between the three cell types for CD59 and DAF. Each N represents a different patient. Figure 2 A comparison of relative complement inhibitor expression level between oligodendrocytes, astrocytes and microglia. The data were estimated by one-way analysis of variance (ANOVA) with Fisher's LSD post-hoc test (A and B; P < 0.05 was considered to show statistically significant differences). Double fluorescence immunohistochemistry In order to establish identity between oligodendroglial cells, astrocytes or microglia and cells expressing the complement inhibitor proteins CD59 or C1-inh, double fluorescence immunostaining was carried out. Oligodendrocytes were detected by O4 staining with a Texas Red tagged secondary antibody (Fig. 3A and 3D ) in the first cycle and CD59 (Fig 3B ) or C1-inh staining (Fig. 3E ) detected with a green FITC tagged antibody in the second cycle. Astrocytes were detected by GFAP staining with a Texas Red tagged secondary antibody (Fig. 3G and 3J ) in the first cycle and CD59 staining (Fig 3H ) or C1-inh staining (Fig. 3K ) detected with a green FITC tagged antibody in the second cycle. Microglia were detected by CD68 staining with a Texas Red tagged secondary antibody (Fig. 3M and 3P ) in the first cycle, and CD59 staining (Fig 3N ) or C1-inh staining (Fig. 3Q ) detected with a green FITC tagged antibody in the second cycle. With double fluorescent excitation, all cells fluoresced yellow (Fig. 3C,3F,3I,3L,3O,3R ), indicating colocalization of O4 with CD59 or C1-inh, GFAP with CD59 or C1-inh, and CD68 with CD59 or C1-inh. Figure 3 Double fluorescence immunohistochemistry of oligodendrocytes, astrocytes and microglia. Double fluorescence immunostaining for O4 and CD59 or C1-inh is demonstrated in A-F. In A and D, cells of typical oligodendroglial morphology were stained in the initial cycle for the specific oligodendroglial marker O4. Detection is by a Texas Red-conjugated secondary antibody. Second cycle staining for CD59 (B) and C1-inh (E) are shown. The detections are by an FITC-linked green fluorescent secondary antibody. In C and F, double immunofluorescences are shown in which the cells appear yellow, demonstrating colocalization of O4 with CD59 or C1-inh. Double fluorescence immunostaining of astrocytes for GFAP and CD59 or C1-inh is demonstrated in G-L. In G and J, cells of typical astrocytic morphology are stained in the initial cycle for the specific astroglial marker GFAP. Detection is by a Texas Red-conjugated secondary antibody. Second cycle staining for CD59 (H) and C1-inh (K) is shown with an FITC-linked green fluorescent secondary antibody. In I and L, double immunofluorescences are shown in which the cells appear yellow, demonstrating colocalization of GFAP with CD59 or C1-inh. Double fluorescence immunostaining for microglia using the specific marker CD68 and CD59 or C1-inh is demonstrated in M-R. In M and P, cells of typical microglial morphology are stained by CD68 with detection by a Texas Red-conjugated secondary antibody. Second cycle staining for CD59 (N) and C1-inh (Q) are shown. The detections are by an FITC-linked green fluorescent secondary antibody. In O and R, double immunofluorescences are shown in which the cells appear yellow, demonstrating colocalization of CD68 with CD59 or C1-inh. (Magnification: × 200) Discussion This report shows that human oligodendrocytes express a much lower level of mRNA for C1-inh than astrocytes and microglia, and a significantly lower level of mRNA for MCP. The mRNA levels of CD59 and DAF were comparable in all the three cell types. Overall our data suggest that oligodendroglial cells, in common with other cell types, can produce complement inhibitors, but at a significantly lower level for C1-inh and MCP. It has already been reported that human neurons and Schwann cells [ 24 ], neuroblastoma cell lines [ 18 ], astrocytes [ 23 ], astrocytoma cell lines [ 25 , 26 ], the HOG human oligodendroglial cell line [ 1 ] and oligodendrocytes [ 21 , 22 ] produce some or all of the complement inhibitor proteins and their mRNAs. Activation of the complement cascade and deposition of activated complement fragments occur in non-infectious diseases such as multiple sclerosis, Pick's disease, Alzheimer's disease and other neurodegenerative conditions [ 15 , 16 , 30 - 34 ]. Complement inhibitors may play an important role in preventing such pathology. Full activation of the complement cascade requires overcoming a series of endogenous inhibitory factors. Oligodendrocytes are vulnerable to complement attack, which is particularly observed in multiple sclerosis [ 35 - 37 ] and this vulnerability may be related to a deficiency of their ability to express complement regulatory proteins [ 22 ], particularly C1-inh. Sporadic complement attack, in the form of complement activated oligodendroglia (CAO) is also observed in a number of neurodegenerative conditions [ 38 , 39 ], including Alzheimer's, Pick's, Huntington's and Parkinson's diseases, amyotrophic lateral sclerosis, progressive supranuclear palsy, Shy-Drager syndrome, argyrophilic grain dementia and pallido-nigral luysial atrophy [ 38 , 39 ]. The source of the complement proteins that become activated is unknown, but the data presented here suggest that oligodendrocytes are vulnerable to complement attack because of a low expression of C1-inh and MCP. Conclusions These results suggest that the lower expression of C1-inh and MCP by oligodendrocytes could contribute to their vulnerability in several neurodegenerative and inflammatory diseases of the central nervous system, particularly multiple sclerosis. List of abbreviations analysis of variance (ANOVA) central nervous system (CNS) complement activated oligodendroglia (CAO) complement receptor 1 (CR1) decay-accelerating factor (DAF) dithiothreitol (DTT) fluorescein isothiocyanate isomer (FITC) glyceraldehyde-3-phosphate dehydrogenase (G3PDH) glial fibrillary acidic protein (GFAP) homologous restriction factor (HRF) membrane cofactor protein (MCP) phosphate-buffered saline (PBS) Competing interests None declared. Authors' contributions MH was responsible for the majority of the experimental studies, and for writing the manuscript. AK contributed to the cell culture and the editing of the manuscript. PLM contributed to the conception, interpretation of results and the writing and editing of the manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC516791.xml
340940	Functional Bias and Spatial Organization of Genes in Mutational Hot and Cold Regions in the Human Genome	The neutral mutation rate is known to vary widely along human chromosomes, leading to mutational hot and cold regions. We provide evidence that categories of functionally related genes reside preferentially in mutationally hot or cold regions, the size of which we have measured. Genes in hot regions are biased toward extracellular communication (surface receptors, cell adhesion, immune response, etc.), while those in cold regions are biased toward essential cellular processes (gene regulation, RNA processing, protein modification, etc.). From a selective perspective, this organization of genes could minimize the mutational load on genes that need to be conserved and allow fast evolution for genes that must frequently adapt. We also analyze the effect of gene duplication and chromosomal recombination, which contribute significantly to these biases for certain categories of hot genes. Overall, our results show that genes are located nonrandomly with respect to hot and cold regions, offering the possibility that selection acts at the level of gene location in the human genome.	Introduction Because of the abundant availability of mouse and human genome data ( International Human Genome Sequencing Consortium 2001 ; Mouse Genome Sequencing Consortium 2002 ), it has come to light that mutation rates vary widely across different regions of the human genome ( Matassi et al. 1999 ; Mouse Genome Sequencing Consortium 2002 ; Hardison et al. 2003 ), in agreement with a number of smaller-scale studies ( Wolfe et al. 1989 ; Casane et al. 1997 ; Perry and Ashworth 1999 ). Regions of unusually high or low substitution rates have been observed from 4-fold sites and ancestral repeat sequences, two of the best candidates for measuring neutral rates of mutation in mammals ( Sharp et al. 1995 ; Mouse Genome Sequencing Consortium 2002 ; Hardison et al. 2003 ). The reasons for such regional variability are unclear, since structural characterizations of the mutation rate are nascent. Whatever the reason for these hot and cold regions, their existence suggests a question that has intriguing consequences for molecular evolution: does the organism take advantage of these hot and cold spots? One way to take advantage of a hot region would be to place genes there for which the hotness is useful—an intuitive example would be receptor proteins, which must respond to a constantly changing ligand set. Similarly, it could be beneficial to place delicate genes in a cold region, to reduce the possibility of deleterious mutations. These potential advantages offer the possibility that regional mutation rates affect the spatial organization of genes. The idea of such organization in mouse and human is bolstered by recent findings of gene organization in yeast. For example, Pal and Hurst (2003 ) showed that yeast genes are organized to take advantage of local recombination rates, which is particularly relevant since mutation rate and recombination rate are known to be correlated ( Lercher and Hurst 2002 ). If the local mutation rate—equivalent to the synonymous (amino acid preserving) substitution rate K S if synonymous substitutions are neutral—affects gene organization, this would constitute a type of selection complementary to traditional selection on point mutations ( Graur and Li 2000 ). We studied whether local mutation rates affect gene locations by measuring the mutation rates of genes and their organization in the human genome. First, we analyzed the substitution rates of the genes in each of the families defined by the Gene Ontology (GO) Consortium ( Ashburner et al. 2000 ). If the organism is taking advantage of varying K S , gene families should be biased toward regions of appropriate rate. In fact, we observe that several functional classes of genes preferentially occur in hot or cold regions. Some of the notable hot categories we observe are olfactory genes, cell adhesion genes, and immune response genes, while the cold categories are biased toward regulatory proteins such as those involved in transcription regulation, DNA/RNA binding, and protein modification. Also, to better characterize the hot and cold regions, we measured the length scale over which substitution rates vary. While rough limits on the size of hot and cold regions are known ( Matassi et al. 1999 ; Hardison et al. 2003 ), this paper presents the first known quantitative calculation of their length scale. Because mutation rates are regional, mutation rates in genes categories could be influenced by events altering the organization of genes in the genome, such as gene relocation or gene duplication. We therefore analyzed mechanisms by which functional categories of genes may have become concentrated in hot or cold regions. A clustering analysis reveals that the hotness of some categories is enhanced by local gene duplications in hot regions. However, there are strong functional similarities among the hot categories—both clustered and unclustered—as well as among the cold categories. These functional similarities imply that the instances of duplicated categories are not random; i.e., selection may have affected which genes have duplicated and persisted. Results Mutation Rates Have Regional Biases Recently, substitution rates between Mus musculus and Homo sapiens have been measured by several groups on a genome-wide scale ( Kumar and Subramanian 2002 ; Mouse Genome Sequencing Consortium 2002 ; Hardison et al. 2003 ). These substitution rates vary significantly across the genome ( Mouse Genome Sequencing Consortium 2002 ; Hardison et al. 2003 ), suggesting that neutral mutation rates may have regional biases as well. A popular proxy for neutral mutation rates is the substitution rate at 4-fold sites (a recent example is found in Kumar and Subramanian [2002] ), base positions in coding DNA that do not affect protein sequence and that should hence be under less selective pressure than other sites. The 4-fold sites also offer the advantage of being easily alignable. For these reasons, we estimated the neutral mutation rate from substitution rates at 4-fold sites (which we use interchangeably with the term K S in this paper). This identification is not without complexities, however, since there are processes that can in principle selectively affect the 4-fold sites. For example, some have argued that exogenous factors such as isochore structure influence the silent sites ( Bernardi 2000 ), and codon usage adaptation has been shown to affect silent sites in bacteria and yeast ( Sharp and Li 1987 ; Percudani and Ottonello 1999 ). So far, such selective effects have been difficult to detect in mammals ( Smith and Hurst 1999a ; Duret and Mouchiroud 2000 ; Iida and Akashi 2000 ; Kanaya et al. 2001 ). Recently, Hardison et al. (2003 ) showed that several functionally unrelated measures of mutation rate, including SNP density, substitutions in ancestral repeats, and substitutions in 4-fold sites, are correlated in genome-wide mouse–human comparisons—suggesting that these measures have common neutral aspects. We constructed our own dataset of the 4-fold substitution rates for 14,790 mouse/human orthologous genes, using data from the ENSEMBL consortium. In order to properly account for stochastic finite-size effects, we mapped the observed substitution rates to a normalized value, based on the assumption that all 4-fold sites mutate at the same rate (see Materials and Methods ). Under this assumption, it was expected that the normalized substitution rates would follow the normal distribution (a Gaussian distribution with σ = 1). Contrary to these expectations, the distribution of ortholog substitution rates was found to be highly biased toward high or low rates, indicating that 4-fold mutation rates vary substantially by location and on a scale larger than the typical size of a gene. Figure 1 shows the distribution of substitution rates for all mouse/human orthologs. The observed distribution has excesses of genes at both high and low substitution rates. These results are in agreement with the findings of Matassi et al. (1999 ), who reported significant mutation rate correlations between neighboring genes. This is not a compositional effect—the distribution remained the same even when corrections for the gene's human base composition were made (see Materials and Methods ). We further verified that substitution rates of neighboring genes were correlated using an analysis qualitatively similar to Matassi et al. (1999 )—though with approximately 20 times more orthologs—finding that gene substitution rates are correlated with their neighbors with a p -value of 10 −189 (see Materials and Methods ). These results imply that substitution rates have regional biases, acting both within a gene and over longer length scales. Figure 1 Distribution of Normalized Substitution Rates Histogram of substitution rates based on 14,790 orthologous mouse and human genes (black curve). The rate distribution has significantly more genes at high and low rates than the expected Normal distribution (red curve). This bias toward high and low rates remains even when rates are corrected for human base composition (green curve). Some Gene Categories Are Biased toward Hot or Cold Regions We next considered whether there is a relationship between gene locations and their functions, i.e., whether functional categories of genes have biases for being in regions of particular mutation rate. To test whether such biases exist, we performed an analysis of the GO assignments for each ortholog pair ( Ashburner et al. 2000 ), using data from the ENSEMBL human ENSMART database to assign genes to GO categories. For each GO category, we calculated a z -score to measure the overall substitution rate, based on the substitution rates of the genes in the category (see Materials and Methods ). The 21 GO categories having statistically significant positive values of z are shown in Table 1 . In terms of 4-fold substitution rates, the hot category rate averages were found to range from 0.346 (integral to membrane) to 0.468 (internalization receptor activity), while the genome-wide average was 0.337 (with a genewise standard deviation of 0.08). For a category with several genes, the effective standard deviation is much smaller, equal to 0.08/ √ N GO , where N GO where N GO is the number of genes in the category, so these rate biases are extremely significant. Hot gene categories were focused mainly in receptor-type functions, along with a few other categories such as “proteolysis” and “microtubule motor activity.” Some preferences were partially because categories have genes in common; e.g., eight genes are shared among the categories “dynein ATPase activity,” “dynein complex,” and “microtubule-based movement.” However, there were several categories of similar function that were independent; e.g., “membrane” and “olfactory receptor activity” shared no genes, and “cell adhesion” and “immune response” shared only 5% of their genes. Overall, there was a clear bias for the larger hot categories to contain receptor-type proteins: e.g., “receptor activity,” “olfactory receptor activity,” “G-protein coupled receptor protein signaling pathway,” “membrane,” and “immune response.” For the set of all 1,488 genes where the string “receptor” is part of the GO description, the average 4-fold substitution rate was found to be 0.347. The probability that a random set of 1,488 genes would have an average rate this high is 10 −6 . Table 1 Statistically Significant Hot GO Categories Listed are the categories with z > 0 having at least five genes and p z ≤ 10 −3 , sorted by statistical significance (−log 10 p z ). There is a bias toward proteins involved in extracellular communication. Several of the categories have an unusual number of clustered genes (−log 10 p cluster >3) The 36 statistically significant GO categories with negative z scores, are shown in Table 2 . The 4-fold rate averages for the cold categories ranged from 0.220 (“mRNA binding activity”) to 0.326 (“protein serine/threonine kinase activity”). The coldest gene categories included “nuclear proteins,” “transcription regulation,” “DNA and RNA binding,” “oncogenesis,” “phosphatases,” and “kinases,” all of which are important to regulatory processes. Many of these genes are also housekeeping genes ( Hsiao et al. 2001 ). For the set of all 1,704 genes where the string “regulat” is part of the GO description, the average 4-fold substitution rate was found to be 0.325. The probability that a random set of 1,704 genes would have an average rate this low is 10 −9 . Table 2 Statistically Significant Cold GO Categories Listed are the categories with z < 0 having at least five genes and p z ≤ 10 −3 , sorted by statistical significance. There is a bias toward proteins involved in DNA, RNA, or protein regulation. None of the cold categories have statistically significant clustering We repeated our z -score classifications using several other measures of mutation rate and in each case inferred similar hot and cold categories. For example, under the normalized rate model that accounts for human base composition, the same set of 23 hot categories were found. Of the 37 cold categories, 33 remained classified as cold. The four lost were “regulation of transcription from Pol II promoter,” “development,” “neurogenesis,” and “translation regulator activity.” There were six new categories, and these were also largely regulatory: “nucleic acid binding activity,” “translation initiation factor activity,” “ubiquitin C-terminal hydrolase activity,” “collagen,” “RNA processing,” and “negative regulation of transcription.” We also calculated several maximum likelihood (ML) measures of K S using mutation models in the Phylogenetic Analysis by Maximum Likelihood (PAML) package ( Yang 1997 ), including the Nei and Gojobori (1986 ) codon-based measure and the TN93 ( Tamura and Nei 1993 ) and REV ( Tavere 1986 ) models. We again found qualitatively similar sets of hot and cold categories—receptor genes at high substitution rates and regulatory genes at low substitution rates—though there were changes in the numbers of significant categories. For example, for the TN93 model, we observed ten hot categories—“induction of apoptosis by extracellular signals,” “G-protein coupled receptor protein signaling pathway,” “olfactory receptor activity,” “receptor activity,” “apoptosis,” “enzyme activity,” “chymotrypsin activity,” “trypsin activity,” “integral to membrane,” and “dynein ATPase activity”—and eight cold categories: “calcium-dependent protein serine/threonine phosphatase activity,” “ribonucleoprotein complex,” “protein serine/threonine kinase activity,” “RNA binding activity,” “protein amino acid dephosphorylation,” “intracellular protein transport,” “protein transporter activity,” and “nucleus.” The categories inferred from our original z -score analysis are probably more accurate than those from ML methods, because ML methods tend to produce strong outliers at high substitution rate, skewing calculations of the variance in the z -score analysis. Can Gene Duplications Explain the Hot and Cold Categories? Given the existence of hot and cold gene categories, the question then becomes: why do these biases exist? One potentially nonselective factor that could affect category rate biases is local gene duplications. New genes generally arise by duplication, in which a new copy of a gene is generated nearby to the preexisting gene by a recombinatorial event such as unequal crossing-over, followed by evolution to a novel, but often related function ( Graur and Li 2000 ). Such local duplications can cause many genes with similar function to be clustered together. Because there are regional biases in mutation rate (discussed in the section on Block Structure of the Substitution Rate), these functionally related genes will tend to have similar mutation rates. GO categories containing these genes will then be biased toward the mutation rate of the region surrounding the genes. We tested the effect of gene duplications on category rates through a clustering analysis (see Materials and Methods ). If gene duplications are not important to category rates, genes in a hot (cold) gene category would be expected to be distributed randomly throughout the many hot (cold) regions around the genome; i.e., clustering of genes would be weak. However, if gene duplications are relevant, we would expect hot (cold) genes of the same category to be tightly clustered since many of these genes would have arisen by local duplications. We therefore studied the location distribution of each of the gene categories and analyzed the significance of its clustering, using the short-range correlation length τ ∼ 10 6 basepairs (see the section on Block Structure of the Substitution Rate) as a defining length scale. This analysis was similar to that of Williams and Hurst (2002 ), who studied clustering of tissue-specific genes, though we analyzed a larger number of more narrowly defined gene families. We found that some of the hot gene categories were indeed clustered, but that none of the cold gene categories were. The results of the clustering for the hot and cold categories are displayed in Tables 1 and 2 , with the clustering p-values shown via their −log 10 values. Of the 21 statistically significant hot categories, ten categories had statistically significant clustering (−log 10 p cluster > 3). For example, the “olfactory receptor activity” category has 223 genes, with a randomly expected number of clustered genes equal to 30.6. The actual number of clustered genes was found to be 190, which has a p -value of less than 10 −16 . In the set of 37 cold gene GO categories, none had statistically significant clustering. The clustering significance is plotted versus the substitution score z for all the GO categories with at least five members in Figure 2 . There were many categories of hot genes with significant clustering (−log 10 p cluster > 3), but virtually no cold ones. Figure 2 Clustering versus Substitution Rate for GO Categories Containing at Least Five Members Virtually all clustered gene categories have higher than average substitution rates ( z > 0). As an example of clustering in the hot gene categories, we considered the olfactory receptors. It is well-established that olfactory receptors occur in clusters throughout the human genome ( Rouquier et al. 1998 ), and we likewise observed the olfactory receptors to be highly clustered in three regions near the head, middle, and tail of Chromosome 11 ( Figure 3 ). The central cluster is displayed in Figure 4 . This clustering provided evidence that local gene duplications have influenced the high category rate of the olfactory genes. Figure 3 Clustering of Olfactory Genes on Human Chromosome 11 The olfactory genes are clustered into three regions along the chromosome. The substitution rates of the olfactory genes are almost all hot, while the nonolfactory genes are distributed around r = 0. Figure 4 Olfactory Genes Lie in a Mutational Hot Spot Substitution rates of the olfactory genes in the central region of human Chromosome 11. The substitution rate of ancestral repeat sequences is higher in the region where the olfactory genes lie. We next attempted to determine whether the high olfactory rates are due to a regional bias. The substitution rates of all genes are plotted in Figure 4 , with the olfactory genes in red. As expected, the olfactory genes exhibited an obvious bias for higher substitution rates than other genes. We next calculated the mutation rate of the region as determined from an independent measure, the substitution rates between ancestral repeat sequences (green curve in Figure 4 ), using data published by Hardison et al. (2003 ) (see Materials and Methods ). The repeat sequence mutation rate was notably higher in the regions where the olfactory genes occur, showing that the hotness of the olfactory genes is a regional property and not specific to the genes. Similar clustering and regional hotness were observed for other hot gene categories. We plot the substitution rates of a cluster of homophilic cell adhesion genes on Chromosome 5 in Figure 5 , along with the rates of nearby genes and the ancestral repeat sequence substitution rates. The same features observed for the olfactory genes were also present for the cell adhesion genes: clustering, high substitution rates, and an elevated ancestral repeat substitution rate. The repeat substitution rate exhibited a plateau-like behavior over the region defined by the homophilic cell adhesion genes. These factors support the interpretation that significant numbers of hot genes have arisen by duplications in inherently hot regions of the genome. Figure 5 Homophilic Cell Adhesion Genes Also Lie in a Hot Spot Substitution rates of a cluster of homophilic cell adhesion genes on human Chromosome 5, along with substitution rates of other genes and ancestral repeat sequences. The repeat sequence substitution rate plateaus at a higher level in this region. Block Structure of the Substitution Rate Several explanations have been proposed that could account for the regional biases in mutation rate ( Mouse Genome Sequencing Consortium 2002 ), including recombination-associated mutagenesis ( Perry and Ashworth 1999 ; Lercher and Hurst 2002 ), strand asymmetry in mutation rates ( Francino and Ochman 1997 ), and inhomogeneous timing of DNA replication ( Wolfe et al. 1989 ; Gu and Li 1994 ). The structure of regional biases could be considered from the perspective of amino acid changing substitutions as well, since linked proteins have been known to have similar substitution rates ( Williams and Hurst 2002 ). However, the silent sites may be easier to comprehend, since protein sequences are more likely to be complicated by nonneutral pressures. To shed light on the structural properties of the hot and cold mutational regions, we measured the length scale over which substitution rates are correlated. Previously, correlations have been observed in blocks of particular physical (5 Mb) ( Hardison et al. 2003 ) or genetic (1, 2, 5, and 200 cM) ( Matassi et al. 1999 ; Lercher et al. 2001 ) size. While these studies have focused on whether correlations exist at certain length scales, it is informative to measure the decay of correlations with distance. We therefore measured the length scale of substitution rate correlation, using an analysis of the correlation function ( Huang 1987 ) where r ( t ) is the substitution rate of a gene t basepairs downstream of a gene with substitution rate r (0), and <…> indicates an average over the available data (see Materials and Methods ). We expect that at small t , the correlation function will be positive and then decrease with t as rates become decoupled. The length scale over which this decay occurs serves as a measure of the typical size of hot or cold regions. The rate correlation function is plotted in Figure 6 versus both the human and mouse values for t . Figure 6 Correlation Length Analysis of Substitution Rates Correlation of substitution rates in syntenous blocks as a function of distance between genes measured along the human chromosome (top) and measured along the mouse chromosome (bottom). There are two length scales of correlation decay: a short one of 1 Mb and a long one of 10 Mb. The curve fits are for < r (0) r ( t )> = A 0 exp (− t/τ ) + A ∞ for the region t ∈ [0, 10000000] . We observed two notable behaviors: (1) a strong correlation that decays over a region of approximately 1 Mb, and (2) a longer range correlation which plateaus over a region of approximately 10 Mb. At larger distances, correlations are weaker. For example, the human curve first dips below the < r (0) r ( t )> = 0 threshold at approximately 11 Mb, and the mouse curve first crosses it at approximately 9 Mb. This suggests that there are multiple phenomena that control the mutation rate of regions, both long (10 Mb) and short (1 Mb) length scale. We also measured the characteristic short-range correlation length using an exponential fit. The correlation length τ was determined by fitting the data to the functional form where A ∞ is the correlation at long distances and ( A 0 + A ∞ ) is the correlation at zero distance. Because of the observed plateauing behavior of the data, we performed our curve fit over the region t ∈ [0, 10000000] . For the human data, we obtained A 0 = 0.83, τ = 1.21 × 10 6 , A ∞ = 0.39. For the mouse data, we found values of a similar magnitude ( A 0 = 1.08, τ = 0.73 × 10 6 , A ∞ = 0.32), suggesting that short-range mutational processes may be alike in mouse and human. The long-range correlation A ∞ was at least an order of magnitude larger than would be expected by chance at all distances up to 10 Mb (see Materials and Methods ). It is unclear what factors are responsible for these two length scales of rate correlation, though some guesses are possible. For the short-range effect, one process that occurs on the appropriate length scale is DNA replication ( Alberts et al. 1994 ). Replication origins in a concerted unit activate under similar timing and similar cell conditions and could have a common regulatory mechanism, making it a reasonable to expect the DNA in such a unit to have similar mutation rates. Long-range correlations have previously been observed at chromosomal-size distances in particular regions of the genome; e.g., it is known that Chromosome 19 is generally hotter than other chromosomes ( Lercher et al. 2001 ; Castresana 2002 ). However, the 10 Mb correlation was not simply due to selection on chromosomes. We removed the respective chromosomal average from each substitution rate and repeated the correlation analysis, finding that A ∞ retained a significant value of approximately 0.2. One possible mechanistic explanation for the long-range correlation is suggested by the finding of Lercher and Hurst (2002 ) that recombination rate and substitution rate are correlated even in blocks extending to 30 Mb. Therefore, if large regions of similar recombination rate exist, they could be related to the long-range 4-fold correlation effects we observed. Discussion Evidence for Selection Recently, there has been evidence for selective factors influencing gene location in yeast ( Pal and Hurst 2003 ). This suggests the possibility that similar phenomena affect gene locations in mouse/human as well. We therefore considered whether regional mutation rates could have selectively influenced the types of genes occurring in different loci in the genome. Selection due to the local mutation rate would require different mechanisms than that observable through the traditional measure K A /K S , which quantifies selection on point mutations. For example, regional mutation rates could have influenced the fitness of the genome after events that cause gene relocation, such as gene transposition or chromosomal recombination. Or perhaps the duplication of certain genes provided a fitness benefit (a mechanism possibly relevant for the hot clustered categories). Differential duplication rates could force a category to have a mutation rate bias, due to the block structure of the mutation rate and the fact that duplications occur locally. The observed categories of hot and cold genes suggest gene locations have been selectively influenced by regional mutation rates. This is because if mutation rates were unrelated to gene function, then the lists of hot and cold categories would be expected to be random; i.e., the lists shown in Tables 1 and 2 would have been evenly sampled from all possible GO categories. However, this was not the case, as the hot and cold categories each had strong internal commonalities. The hot categories were found to be biased toward receptor activities or roles in extracellular communication. Intriguingly, arguments based on protein-level effects appear applicable to the silent-site hotness of these categories. Cellular receptors and those involved in extracellular communication are the proteins that most directly interact with the environment and are therefore the most likely to have experienced a dynamically changing set of selection pressures. This variability of selection pressures would have made it favorable for them to be in hot regions, in order that new mutations be possible to deal with new stimuli. Examples of hot categories with known protein-level diversification pressures include the olfactory receptors ( Lane et al. 2001 ), immune genes ( Papavasiliou and Schatz 2002 ), and cell adhesion genes ( Uemura 1998 ; Tasic et al. 2002 ). Arguments normally applied to protein-level selection were found to be appropriate for cold mutation rate categories as well. Cold categories were often related to transcription or other regulatory processes. Regulatory proteins should be tuned to interact with many different nucleic acid or protein targets, in contrast with receptor proteins, which typically interact with only a particular ligand. Mutations to regulatory proteins would therefore be expected to be more deleterious, and hence it would be beneficial for them to have low mutation rates. Strong conservation pressures in the cold categories could also be related to their roles as housekeeping genes ( Zhang and Li 2003 ) or as essential genes. For example, in the dataset of Winzeler et al. (1999 ), 81 out of 356 essential yeast genes were involved in transcription, whereas only four were involved in signal transduction, the function most similar to extracellular communication for which data were available. The applicability of protein-level arguments to synonymous mutation rates suggests that K S and K A are under similar pressures. This is consistent with what would be expected if gene locations have evolved to make use of the block structure of the mutation rate, since relocation to a hot (cold) spot would increase propensities for both high (low) K A and K S . More quantitatively, we observed that K S category biases were similar to category biases caused by selection on amino acid changing point substitutions—i.e., selection observable through the ratio K A /K S . We performed a GO z -score analysis on K A /K S (for consistency, the CODEML method in PAML was used to calculate both K A and K S ). There were eight hot categories common to both the 4-fold and K A /K S classifications (“immune response,” “proteolysis receptor activity,” “peptidolysis receptor activity,” “integral to membrane,” “chymotrypsin activity,” “cell adhesion,” “trypsin activity,” “olfactory receptor activity”) and 17 common cold categories (“nucleus,” “regulation of transcription,” “transcription factor activity,” “RNA binding activity,” “development,” “ribonucleoprotein complex,” “protein transporter activity,” “protein serine/threonine kinase activity,” “ubiquitin conjugating enzyme activity,” “GTP binding activity,” “ubiquitin-dependent protein catabolism,” “translation regulator activity,” “intracellular protein transport,” “neurogenesis,” “ubiquitin cycle,” “cytoplasm,” “regulation of transcription from Pol II promoter”). The strong commonalities between the two types of classification suggest that the selective forces that influenced amino acid changing point mutations also influenced gene locations. The hot and cold categories derived from K A /K S are available as Dataset S1 and Dataset S2 . Selection on gene locations would provide an evolutionary explanation for the puzzle of why K A and K S are correlated beyond levels expected by neutral evolutionary theory ( Mouchiroud et al. 1995 ; Ohta and Ina 1995 ). Assuming 4-fold sites are neutral, locational selection would have to be realized through the influence of the local mutation rate K S on the amino acid changing mutation rate K A . Thus, locational selection and point mutation-based amino acid selection would behave similarly with respect to positive or negative selection on protein sequence, increasing the correlation of K A and K S , even if mutations to any individual 4-fold site did not provide a fitness benefit. One caveat is that other, not necessarily exclusive, explanations for the strong correlation of K A and K S have been proposed as well—most notably simultaneous substitutions at adjacent sites, so-called tandem substitutions ( Smith and Hurst 1999b ). Tandem substitutions were not sufficient to explain our hot and cold categories, however. We rederived sets of hot and cold categories after correcting for tandem effects (see Materials and Methods ) and once again found similar results. For example, the six hottest categories (of 22 significant) were “dynein ATPase activity,” “receptor activity,” “homophilic cell adhesion,” “olfactory receptor activity,” “integral to membrane,” and “calcium ion binding activity.” The six coldest (of 36) were “nucleus,” “regulation of transcription, DNA dependent,” “RNA binding activity,” “transcription factor activity,” “development,” and “ribonucleoprotein complex.” Mechanisms For the hot clustered categories, it may be that high mutation rates and high rates of gene duplication are tied to a hidden variable that imposes both phenomena simultaneously. One possibility is the recombination rate along the genome, which Pal and Hurst (2003 ) found to have selective effects in yeast. For example, two mechanisms for diversification, gene duplication and mutation, can both be accelerated by recombination ( Graur and Li 2000 ; Lercher and Hurst 2002 ). High recombination rates are relevant for a number of the hot gene categories we have studied, as they have been suggested for the protocadherins ( Wu et al. 2001 ), immune response ( Papavasiliou and Schatz 2002 ), and olfactory families ( Sharon et al. 1999 ). Because both gene duplication and point mutation are useful for diversifying a family, it is difficult to separate the significance of mutation rate and recombination rate. Pal and Hurst (2003 ) offered preliminary evidence that in yeast, selection acts on the recombination rate, but not point mutation rates. However, we have observed unusual rate biases for nonclustered gene categories as well, for which recombination would not be expected to play a role. Cold gene categories are not clustered; therefore, the existence of cold categories (as well as nonclustered hot categories) cannot be attributed to duplication events. One alternate phenomenon that could cause cold category biases is gene relocation to cold regions. The concept of relocation brings up a number of questions. First, if cold genes have relocated, this leaves one wondering in what sort of environment cold genes originated. One speculative possibility is that these genes developed in regions of high recombination (the hot regions), which would have allowed for fast duplication and functional diversification, and later dispersed to cooler regions as their functions became fixed. Second, it is unclear whether gene relocations occur frequently enough to account for the observed rate biases. This issue is complicated by the fact that genes have arisen at different times. Many of the cold gene categories occur in diverse sets of tissues and have important regulatory effects, suggesting they should be relatively old. This old age may have allowed them enough time to redistribute through the genome. We verified the correlations of substitution rates along the genome and showed that these correlations lead to an excess of hot and cold genes, confirming studies by Matassi et al. (1999 ) and Hardison et al. (2003 ). Our results appear to disagree with those of Kumar and Subramanian (2002 ), who reported that mutation rates are uniform in the genome. While our rate measurements were qualitatively similar to those of Kumar and Subramanian (2002 ), one beneficial addition we made was the use of a normalized rate that accounts for the length dependence of rate variance, allowing genes of differing lengths to be treated equally in Figure 1 . Our correlation length analysis revealed two scales of rate correlation: a short decay length of 1 Mb and a long-range length extending along a syntenous block up to distances of 10 Mb. We have very speculatively proposed that DNA replication units and DNA recombination may be relevant to these length scales. More generally, it is hoped that these scale determinations will be helpful in placing constraints on possible processes that control mutation rate. Some data issues suggest topics for further exploration. First, the resolution of our analysis is dictated by the structure of the GO taxonomy, which currently has 16,000 categories but is evolving. Our category inferences should become more specific as GO gene assignments improve. Second, multispecies data will be invaluable in revealing the mutations that have occurred in each lineage. One promising early result from human–chimpanzee comparisons, based on a set of 96 orthologs derived from HOVERGEN release 44 ( Duret et al. 1994 ), is that olfactory receptors are a hot category. Unfortunately, this is the only statistically significant hot or cold category at present, owing to the lack of data. However, inferences should improve rapidly as more chimpanzee gene identifications become available. Materials and Methods Ortholog generation. We downloaded a list of the available 37,347 human and 27,504 mouse peptides from the ENSEMBL sequence database ( www.ensembl.org ), then used BLAST ( Altschul et al. 1990 ) to find orthologous peptide sequences between the genomes. The peptides studied were the set of all known or predicted peptides in the ENSEMBL 12.31.1 human and 12.3.1 mouse datasets. Sequences were designated as orthologous if the two peptides were each other's mutual best hit in the opposing databases, as determined by BLASTALL, and the E-value for the match was 10 −10 (using the higher score as a worst-case bound) or better. We chose this method of ortholog determination to get a one-to-one relationship between proteins. We found 14,790 ortholog pairs, a coverage rate of approximately 50% in mouse and 40% in human. The observed E-values between orthologs have a median value of 0.0 (<1e − 180). The aligned peptide orthologs were then used in conjunction with ENSEMBL cDNA data to determine aligned orthologous cDNA. For the chimpanzee–human comparison, human genes from ENSEMBL were compared to chimpanzee genes from HOVERGEN. A mutual best-hit criterion was used to determine the set of 96 orthologs. We manually inspected the mouse–human synteny of the olfactory gene cluster of Figure 4 to verify that orthologs were assigned correctly. This was to address the concern that orthologs are more difficult to assign in gene categories with many homologous members, since incorrect assignments could distort substitution rates. The synteny structure was found to be almost totally conserved for these genes, as it was for the cell adhesion genes in Figure 5 . Calculation of substitution rates. We calculated the distribution of substitution rates between the mouse and human genomes using the 4-fold sites of orthologous genes; 4-fold sites are the third bases of codons for which the amino acid is specified by the first two bases. For each of the orthologous gene pairs, we calculated p , the fraction of 4-fold sites in which the mouse base differs from the human base. The average value of p over all 4-fold sites in all orthologs was < p > = 0.337. The average 4-fold substitution rate on a genewise basis was 0.338 with a standard deviation of 0.080. These rates were in agreement with substitution rates measured in other studies of 4-fold sites or in ancestral repeats ( Mouse Genome Sequencing Consortium 2002 ; Hardison et al. 2003 ). Because genes are of finite length, stochastic effects can cause substitution rates to vary from gene to gene, even if all 4-fold sites mutate at the same rate. We defined a normalized substitution rate to correct for these finite-size effects. A gene with N 4-fold sites was modeled as having N independent events in which substitution can occur with probability < p >. This formulation can fit both the Jukes–Cantor one-parameter or the Kimura two-parameter model for mutation matrices ( Durbin et al. 1998 ). Although this model is not as sophisticated as other more modern treatments (e.g., see Tavere 1986 ; Tamura and Nei 1993 ; Li 1993 ; Goldman and Yang 1994 ), it gives an easily falsifiable prediction that the rate distribution, in the absence of regional correlation, can be approximated by a standard Normal distribution, due to the central limit theorem ( Rice 1995 ). Under this model, at each N the distribution of substitution rates can be described by a binomial distribution with a standard deviation of σ( N ) = √ < p >(1 − < p >)/ N . Therefore, gene substitution rates were normalized by their respective σ( N ) to get one universal rate distribution, which in the limit of many datapoints should approach the Normal distribution (2π) −½ exp ( x 2 /2). We defined the normalized substitution rate to be where p is the actual 4-fold substitution rate in the gene. The values of r for all ortholog pairs were used to calculate the distribution shown in Figure 1 . The actual rate distribution in genes was found to be skewed toward high or low mutation rates, as shown in Figure 1 . The observed distribution had a standard deviation of 2.04, significantly higher than the expected σ = 1. Similar excesses of hot and cold genes were found even when corrections were made for base composition. To verify this, we calculated a normalized mutation rate using a four-parameter model in which each site of type A, C, G, or T in the human sequence has its own substitution probability. For each human base (A, C, G, and T), we measured the substitution rate at the corresponding 4-fold locations, yielding 4 rates < p A >, < p C >, < p G >, < p T >. Based on these rates, we then calculated the expected frequency and variance of substitutions for a gene given the gene's base composition at the 4-fold sites. This yielded a distribution nearly identical to that in the one-parameter model (see Figure 1 ). We also tested whether neighboring genes have similar substitution rates. The orthologs were ordered by their location along the human genome, after which we calculated the Pearson correlation of a gene's substitution rate r with that of its following gene. We used only neighboring genes that were in syntenous blocks, as defined by all three conditions of monotonicity (the genes are ordered the same in both species), consistent strand orientation (a block is either in the same strand orientation in both species or completely reversed), and consistent chromosome (no chromosome changes in either species in a block), yielding a dataset of 11,087 neighbor pairs. Under this condition, the Pearson correlation was 0.26, corresponding to a highly significant p -value of 10 −189 . z -score calculation for GO categories. For each GO category, we calculated a normalized substitution rate ( z -score) based on the substitution rates of all members of that category. Of the genes in our ortholog set, 9,966 had GO classifications available. The z -score was defined to be where < r > GO is the average substitution rate r for the genes in the GO category, < r > all is the average r for all of the genes with GO classifications, σ all is the genewise standard deviation, and N GO is the number of genes in the category. The p -value for z was determined from the probability that a Gaussian-distributed variable takes on a value ≥ z . To reduce the problem of outliers, we limited our analysis to the GO categories containing at least five genes, of which there are 997, and accordingly set a p -value cutoff of 1/997 ∼ 10 −3 . We expressed the significance in terms of −log 10 p z , which should have a value larger than 3 to be statistically significant. z -scores corrected for tandem substitutions were calculated by first removing all possible tandem substitution sites from the dataset. That is, 4-fold sites were only accepted into the dataset if both the preceding and following bases matched in the two species. After culling the dataset, we calculated rates and category z -scores as before. Clustering analysis. To measure clustering, for each gene in a GO category we tested whether it had another category member downstream of it within the short-range correlation length of τ = 10 6 basepairs. In each GO category, we calculated the number of genes satisfying this condition, defining this to be the number of “clustered genes.” This “downstream” criterion (rather than a symmetric one) was used to avoid the problem of double counting of genes when several are close together. To test the statistical significance of the number of clustered genes in a category, we used bootstrapping. For each GO category, we performed 5,000 random trials of selecting N GO random genes from the entire set of orthologs, where N GO is the number of genes in the GO category. In each trial, we counted the number of clustered genes in this randomly selected group. The average number of clustered genes was used to approximate the random number of clustered genes by a Poisson distribution. These Poisson statistics were then used to calculate the significance of the number of clustered genes for the GO category. A Poisson distribution is appropriate as long as clustering of neighbors is a rare event, i.e., as long as N GO << N allgenes , which was generally the case. The random distributions were visually inspected and found to agree with the shape of the Poisson curve. To generate the data for Tables 1 and 2 , we also limited ourselves to the 997 categories with at least five genes, implying that −log 10 p cluster > 3 is the cutoff for significance. Calculation of repeat sequence mutation rates. Aligned repeat sequences between mouse and human were obtained from the dataset of Hardison et al. (2003 ). For each repeat, positions in which a base was defined for both the mouse and human sequence were used to calculate a normalized substitution rate, in analogy with the method used for the 4-fold sites. The genome-wide average value of p in these repeat sequences was 0.33, which was very close to the value for 4-fold sites, 0.34. The start position of each repeat sequence was used to define its location in the genome. In order to determine the locations of repeat sequences (based on the June 2002 UCSC genome map) along the physical map used for the gene sequences (based on the ENSEMBL May 2003 map), gene locations according to the two maps were compared. Repeat sequence locations were then corrected using the location differences of nearby genes. For clarity, the ancestral repeat values shown in Figure 4 and Figure 5 were smoothed using a moving-window average of 20 repeat sequences. Correlation length calculation. We considered all pairs of genes on continuous orthologous blocks, starting from the first neighbor up to the 35th gene downstream. This allowed us to get hundreds of measurements of r (0) r(t) for t values even as large as several megabases. We binned these data into 100 uniformly spaced groups covering t ∈ [0, 15000000] and then averaged over each of these bins to determine the correlation function < r (0) r(t) >. The data were plentiful enough for the averaged values shown in Figure 6 to be statistically significant. It was difficult to extend to larger values of t since the amount of data decreases with t , a fact manifested in the increasing fluctuations at larger t in Figure 6 . For example, the value of the average correlation < r (0) r(t) > at t = 15 Mb in the human data of Figure 6 was based on only 79 measurements, whereas at t = 75,000 it was based on 22,860 measurements. For genes with alternative splicings, only one of the genes was used, in order to avoid spurious effects caused by reuse of DNA. Orthologous block boundaries were defined by genes at which the chromosome changes in either species. Monotonicity and consistent strand orientation were ignored in order to obtain blocks with large values of t . Most of the r (0) r(t) data comes from blocks at least several megabases long. Approximately 5% is in blocks of size less than 10 6 basepairs, 55% is in blocks of size between 10 6 and 10 7 basepairs, and the remaining 40% is in larger blocks. The long-range correlation shown in Figure 6 was statistically significant. Theoretically, fluctuations in < r ( i ) r ( j )> should be of the order ∼ O (1/ √ N , where N is the number of data samples in a bin. At a distance of 10 Mb, there were approximately 400 samples, corresponding to an uncertainty of approximately 0.05. This uncertainty was an order of magnitude smaller than the observed value of A ∞ = 0.4. Supporting Information Dataset S1 Hot Gene Categories Based on K A /K S Gene categories with significant positive selection on amino acid changing point mutations. (23 KB XLS). Click here for additional data file. Dataset S2 Cold Gene Categories Based on K A /K S Gene categories with significant negative selection on amino acid changing point mutations. (21 KB XLS). Click here for additional data file. Accession Numbers The Gene Ontology ( http://www.geneontology.org/ ) ID numbers for the categories discussed in this paper are as follows: brain development (GO:0007420), calcium-dependent cell adhesion molecule activity (GO:0008014), calcium-dependent protein serine/threonine phosphatase activity (GO:0004723), calcium ion binding activity (GO:0005509), carbohydrate metabolism (GO:0005975), cell adhesion (GO:0007155), cell growth and/or maintenance (GO:0008151), chymotrypsin activity (GO:0004263), CTD phosphatase activity (GO:0008420), cytoplasm (GO:0005737), development (GO:0007275), DNA binding activity (GO:0003677), dynein ATPase activity (GO:0008567), dynein complex (GO:0030286), enzyme activity (GO:0003824), G-protein coupled receptor protein signaling pathway (GO:0007186), GTP binding activity (GO:0005525), heterogeneous nuclear ribonucleoprotein (GO:0008436), homophilic cell adhesion (GO:0007156), immune response (GO:0006955), integral to membrane (GO:0016021), internalization receptor activity (GO:0015029), intracellular protein transport (GO:0006886), magnesium-dependent protein serine/threonine phosphatase activity (GO:0004724), membrane (GO:0016020), metabolism (GO:0008152), microtubule-based movement (GO:0007018), microtubule motor activity (GO:0003777), mRNA binding activity (GO:0003729), myosin phosphatase activity (GO:0017018), neurogenesis (GO:0007399), nucleus (GO:0005634), olfactory receptor activity (GO:0004984), oncogenesis (GO:0007048), protein amino acid dephosphorylation (GO:0006470), protein phosphatase type 2A activity (GO:0000158), protein phosphatase type 2B activity (GO:0030357), protein phosphatase type 2C activity (GO:0015071), protein serine/threonine kinase activity (GO:0004674), protein transporter activity (GO:0008565), proteolysis and peptidolysis (GO:0006508), receptor activity (GO:0004872), regulation of transcription, DNA-dependent (GO:0006355), regulation of transcription from Pol II promoter (GO:0006357), regulation of translational initiation (GO:0006446), ribonucleoprotein complex (GO:0030529), RNA binding activity (GO:0003723), RNA polymerase II transcription factor activity (GO:0003702), RNA splicing (GO:0008380), transcription coactivator activity (GO:0003713), transcription factor activity (GO:0003700), transcriptional activator activity (GO:0016563), translation regulator activity (GO:0045182), trypsin activity (GO:0004295), ubiquitin conjugating enzyme activity (GO:0004840), ubiquitin cycle (GO:0006512), and ubiquitin-dependent protein catabolism (GO:0006511).	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC340940.xml
515298	HLA DR phenotypic frequencies and genetic risk of Type 1 diabetes in west region of Algeria, Tlemcen	Background The main genomic region controlling the predisposition to type 1 diabetes is the Human Leukocyte Antigens (HLA) class II of the major histocompatibility complex. Association with different HLA types depends also on the studied populations. In our investigation, we tried to measure the phenotypic HLA class II association frequencies of DR3 and/or DR4 antigens, using a serologic method called microlymphocytotoxicity analysis, in diabetic and nondiabetic (ND) subjects originating from the west-Algerian region of Tlemcen. The aim of the present study was to determine which HLA DR antigens represent a high susceptibility to develop the disease in this area. Using a case-control retrospective study design, we randomly recruited ninety-one related subjects, 39 type 1 diabetics and 52 ND as controls, at the Internal Medicine Board of Medical Centre University of Tlemcen. Results DR3 antigen frequencies were comparable between the type 1 diabetics and the ND subjects and showed no association with the disease ( p = 1.000, OR = 0.95), whereas DR4 and DR3DR4 antigens were associated with susceptibility to develop type 1 diabetes (DR4; OR = 2.10, DR3DR4; OR = 1.30). Also, no incidence for DR3 ( p = 0.2646) or DR3DR4 ( p = 0.0699) antigen frequencies was related to the sex ratio. However, significant differences in HLA DR4 frequencies between type 1 diabetics and ND were found to be related to sex ( p = 0.0085). Conclusion Taken together, our investigation showed that the strongest association with type 1 diabetes was noticed in the presence of HLA DR4 antigens followed by DR3DR4 antigens. This study highlighted a characteristic of Tlemcen population; a history of consanguineous marriages. Association studies between the disease and genetic polymorphisms should be undertaken in a population where consanguinity is more limited to reduce confounding in result interpretations.	Background The type 1 diabetes, previously called insulin-dependent diabetes mellitus (IDDM), is the consequence of progressive and selective destruction of pancreatic β cells by an immune-mediated process [ 1 ], resulting in an absolute lack of insulin [ 2 ]. It is well established that this self-destruction is primarily provoked by the activation of the autoreactive T lymphocytes by the production of T-helper 1 cytokines [ 3 , 4 ]. In addition, the difference of the epidemiological data from one region to another could largely explain why the release of the autoimmunity is stimulated under the influence of one or more environmental factors [ 5 , 6 ], in genetically predisposed subjects. It is currently obvious that the strongest genetic susceptibility of predisposition is allotted to the IDDM1 alleles located in the HLA locus of the chromosome 6p21 [ 7 - 9 ], and the non-HLA alleles, particularly the IDDM2 polymorph gene located in the region 5' of the insulin gene ( INS ) promoter situated on the chromosome 11p15 [ 10 , 11 ]. Other regions of the genome were identified as IDDM3, coding for the IGF1 receptor, IDDM4, located near the fibroblast growth factor 3 gene, IDDM5, near the estrogens receptor gene [ 7 ] etc. The majority of these regions have no possible statistical criteria allowing them to be clearly linked to the disease [ 11 ]. In all cases, it is mainly HLA alleles that present a high risk of contracting the disease compared to non-HLA alleles [ 6 , 12 ]. Due to this, they are qualified as high genetic predictive markers of developing type 1 diabetes in families having a type 1 diabetic member. Interestingly, the disease prediction offered with success the possibility of clinical trials to delay, or even prevent the appearance of type 1 diabetes. However, the different HLA types associated with diabetes depend also on the population. The purpose of our study is to measure HLA DR3 and/or DR4 antigen frequencies and their association in diabetic and nondiabetic subjects originating from the west-Algerian region of Tlemcen. Using a case-control retrospective study design, we attempt to determine which is the greatest HLA DR susceptibility contributing to developing type 1 diabetes. Ninety-one (91) eligible subjects (thirty-nine (39) type 1 diabetics and fifty-two (52) nondiabetics with relatives with type 1 diabetes as controls), were recruited at the Internal Medicine Board of the Medical Centre University of Tlemcen. Results Table 1 summarizes HLA DR3, DR4 and DR3DR4 antigen frequencies in type 1 diabetics and their nondiabetic relatives. Table 1 HLA DR phenotypic frequencies according to p -values and Odds ratio in type 1 diabetic and nondiabetic subjects. Type 1 diabetics n = 39: 15 M/24 F Nondiabetic controls n = 52: 21 M/31F Frequency (Proportion and %) HLA M F M F p Odds ratio (95 % CI) DR3 3 (7.69) 7 (17.9) 5 (9.62) 9 (17.31) 0.2646 a 0.95 (0.48–1.87) 10 (25.64) 14 (26.92) 1.000 DR4 4 (10.26) 9 (23.08) 6 (11.54) 4 (7.69) 0.0085 a, * 2.10 (1.04–4.24) 13 (33.33) 10 (19.23) 0.0361* DR3DR4 4 (10.26) 4 (10.26) 2 (3.85) 7 (13.46) 0.0699 a 1.30 (0.60–2.80) 8 (20.51) 9 (17.31) 0.5887 DR3 and/or DR4 11 (28.21) 20 (51.28) 13 (25.0) 20 (38.46) 0.5791 a 2.21 (1.13–4.36) 31 (79.49) 33 (63.46) 0.0194* X/X 4 (10.26) 4 (10.26) 8 (15.38) 11 (21.15) 0.0111 a, * - 8 (20.51) 19 (36.54) 0.0194* a p -Values related to the sex ratio comparison (male = M, female = F). * p < 0.05. X = non-DR3, non-DR4. DR4 and DR3DR4 antigens showed an association with susceptibility to type 1 diabetes (DR4; OR = 2.10, DR3DR4; OR = 1.30, that is respectively OR confidence interval 1.04–4.24 and 0.60–2.80, 95% CI), in contrast, DR3 antigens showed no association with the disease (OR = 0.95, OR confidence interval 0.48–1.87, 95% CI). It is important to notice that the strongest association is found in DR4 phenotype as indicated in Figure 1 . Figure 1 HLA DR antigen associations with type 1 diabetes. The block boxes represent the odds ratios (OR). The horizontal lines represent the lower and upper confidence limit of OR (confidence interval, 95 % CI). In addition, the phenotypic frequency of DR4 or DR3DR4 molecules is higher in diabetic group than in the control one, although the difference did not reach significance level in DR3DR4 frequencies ( p = 0.0361 and p = 0.5887 respectively). On the contrary, the DR3 molecules frequency is slightly decreased in type 1 diabetic patients compared to the controls and presents no statistically significant difference ( p = 1.000). Furthermore, no incidence was related to the sex criteria for the frequencies of DR3 and DR3DR4 molecules ( p > 0.05). However, significant differences in HLA DR4 frequencies are linked to the female sex and present a value definitely higher in type 1 diabetic patients compared to those of nondiabetics for the same sex ( p < 0.05) (Figure 2 ). Figure 2 HLA DR antigen frequencies compared to the sex ratio. * p < 0.05, Type 1 diabetics vs. control subjects. Discussion Type 1 diabetes is a polygenic disease which results from the interaction between environmental (viral, toxic, nutritional, socioeconomic [ 6 ]) and genetic factors. It is the form of diabetes which occurs mainly in children and young adults [ 13 ]. Fortunately, molecular epidemiology offers the hope of the possibility of preventing the disease in the future, by evaluating the potential factors of risk of developing this pathology [ 6 ]. Although almost 90 % of new cases of type 1 diabetes occur sporadically, studies of individuals with a diabetic relative in their family are essential [ 4 ]. It is well established that associations between type 1 diabetes and certain HLA antigens largely facilitate the identification of the subjects having a potential risk to develop the disease. Among the putative HLA molecules known to confer a high susceptibility are DQA1 (0301)-DQB1 (0302) (= DQ8), DQA1 (0501)-DQB1 (0201) (= DQ2), DRA-DRB1 (0401) (= DR4) and DRA-DRB1 (0301) (= DR3) [ 14 ]. However, epidemiological studies showed that the different HLA types associated with the diabetes depend also on the various populations. For instance, the risk of developing type 1 diabetes in Caucasians is greater if they are carrying the HLA A8 and B15, while DQ6 alleles are protectors [ 11 , 14 , 15 ]. Among Japanese, it is the association with the HLA B54 that confers a higher susceptibility to develop the disease, while the strongest association was found with HLA DR and HLA DQ locus [ 11 ]. In Algeria, and especially in its western region (Tlemcen), which is known for its history of consanguineous marriage, there is a high rate of consanguinity. This fact could increase the risk of developing type 1 diabetes by favouring the transmission of HLA haplotypes and recessive genes of susceptibility except for HLA antigens that are common in both parents. For these reasons, we were interested in checking whether phenotypically DR3DR4 antigens would involve less risk to develop the disease, in comparison with DR3 or DR4 antigens, knowing the fact that the presence of a probable consanguinity could reduce the frequency of DR3DR4 polymorph phenotype compared to that of DR3 or DR4 phenotype. On a purely comparative basis, similarities seem to be found between our results and those of other investigators [ 5 , 13 , 15 ] with regards to HLA DR4 or DR3DR4 frequencies, which are higher in the type 1 diabetic than the nondiabetic population. On the contrary, the HLA DR3 antigens showed comparable frequencies in both groups of our sample. Consequently, these observations associating DR3 phenotype to a protector effect against type 1 diabetes in our studied population are thus do not conform with those reported in the literature [ 15 - 18 ]. However, DR4 and DR3DR4 antigens are obviously associated with susceptibility of developing the disease. It should also be noted that DR3DR4 antigens might represent a weaker predictive value of disease risk. We concluded from this that the non-excess of DR3DR4 antigens, or comparable frequencies of DR3 molecules between type 1 diabetic patients and ND relative controls, might be a strong indices of consanguinity in our sample. A recent study carried out in Sweden showed that HLA DR3 is associated to the development of type 1 diabetes and the incompatibility of blood group ABo [ 19 ]. One can thus note that the role of the HLA DR3 antigens in conferring risk for type 1 diabetes can be masked in the homogeneous populations. Due to the restricted size of our sample, which may influence our interpretation, we should not consider the odds ratio of HLA DR3 antigens or the eventual consanguinity of our studied population. Moreover, many evidences [ 20 - 23 ] incriminate DR3 and/or DR4 antigens in the susceptibility to type 1 diabetes and their association with the detected autoantibodies in this disease. Indeed, the marks of autoimmunity are much more observed in the diabetic patients carrying DR3 or DR4 antigens than in the diabetic patients where DR3 or DR4 antigens were absents [ 13 ]. Furthermore, recent research indicates that a subject with HLA DR4 or DR3 alleles has three or four times more chance of developing type 1 diabetes compared to the general population; DR3DR4 is associated with the highest risk (20 to 40 times more) [ 11 ]. In Algeria, very few investigations have been undertaken to study the impact of the genetic background on the risk to develop type 1 diabetes in its population, where an annual average incidence of 4.7 per 100000 has already been listed [ 24 ]. In a similar study carried out on Algerian unrelated type 1 diabetics (n = 50) and nondiabetics controls (n = 46), the presence of DR3-DQ2 (linkage disequilibrium) in 45 % of patients and in 13 % of controls was detected by molecular genotyping method using PCR (polymerase chain reaction) and SSO (sequence specific oligonucleotide). DR4-DQ8 was found in 37 % of diabetic cases and in 4 % of control groups [ 25 ]. Finally, association with type 1 diabetes attributed to DR-B1*0405 (alleles of DR4 antigens expression) susceptibility showed a match with our results; however, the DR4 antigens were found to be linked to the female sex. According to the results showed in Table 1 , there is no difference between men and women patients with type 1 diabetes carrying DR4 antigens, but a significant difference was noticed in female sex, either women were contracted or not with diabetic disease. It is certain that screening of HLA class II sub-types and determination of DNA coding sequences allows more precise characterization of ethnic groups, especially, because the same coded molecules can differ by the position of few amino acids. For example, it is well established that DQA1 and DQB1 alleles (sub-types of HLA DQ) code respectively for the alpha and beta chain of the DQ molecule [ 6 ]. Thus, the combination of DQ alpha with Arg in position 52 (Arg-52) and DQ beta in position 57 without Asp (non-Asp 57) is called a diabetogenic heterodimer which is the biggest risk factor of type 1 diabetes in Caucasian. Nevertheless, among the Japanese population, type 1 diabetes is particularly associated with HLA DQ alpha Arg-52, but not with HLA DQ beta non-Asp 57 [ 5 ]. Moreover, it is true that the search of an association with a candidate gene allows a better characterization for most of the frequent multifactorial diseases, because the candidate genes are directly implied in the pathological processes. Today, it is reported that the CTLA-4 (cytotoxic T-lymphocyte antigen-4) allele, localised on the chromosome 2p33, former IDDM12, is largely associated with the susceptibility of numerous common complex diseases, such as the common autoimmune disorders Graves' disease, the autoimmune hypothyroidism and type 1 diabetes [ 26 ]. Combined, these pertinent observations still open new perspectives for debating this crucial subject concerning the public health. Conclusion Type 1 diabetes, or youth diabetes, is a multifactorial disease occurring on a genetic ground of predisposition and starts under certain environmental conditions. The most effective preventive strategy must be designed at the pre-diabetes stage, since immune and/or genetic markers can easily indicate subjects with high risk before the clinical symptoms. Thus, the genetic analysis of HLA class II associations has allowed a screening of the contributing molecules to the type 1 diabetes development and the selection of subjects, which are likely to contract the disease. In this study, we were confronted with difficulties in interpretation of our results which are mainly due to the presence of some indices of consanguinity in our studied population, showing a non excess of DR3DR4 phenotype compared to the DR3 or DR4 phenotype, and a no link of DR3 phenotype to the disease. This result could be due to an ethnic characteristic of Tlemcen population, a history of consanguineous marriages. Nevertheless, these preliminary results made it possible to answer the asked questions. Thus, the strongest type 1 diabetes association is statistically revealed with phenotypic expression of DR4 followed by DR3DR4 phenotype. Association studies between the disease and genetic polymorphisms should be carried out on the population having more limited consanguinity to reduce confusions in result interpretations. Methods Subjects The study was conducted on ninety-one first-degree related subjects (brothers, sisters and siblings), randomly recruited at the Internal Medicine Board of the Medical Centre University of Tlemcen (west-Algeria). The sample included thirty-nine patients with type 1 diabetes (15 males, 24 females), and fifty-two healthy subjects (21 males, 31 females) selected from diabetes relatives as controls. Prior medical histories and personal characteristics were obtained from participants via a questionnaire. The patients' mean (± Standard Deviation) age at clinical onset was 12.28 ± 5.97 years with range of 5 to 22 years and median of 11 years. Subjects who were not first degree related and who were not originating from Tlemcen region were excluded. The use of first-degree relatives eliminates exposures of environmental factors such as food items and viruses since first-degree relatives usually share the same milieu. For execution of the protocol, the informed consent was obtained from all the participating subjects to the designated study. HLA phenotyping (standard complement-dependent assay) The applied serologic technique lies on the aptitude of the antibodies to recognize allotropic determinants of HLA molecules on cellular surface [ 27 ]. This method is sensitized by a reaction of microlymphocytotoxicity [ 28 , 29 ], which uses specific anti-HLA DR antisera and rabbit complement of commercial typing tray (Biotest, Germany). Initially, the peripheral blood lymphocytes (PBL) were separated from the other illustrated elements of venous blood (collected in EDTA-containing tubes) by density gradient centrifugation on Ficoll-Hypaque [ 30 ]. B lymphocytes were isolated by using nylon-wool-separated columns [ 31 ]. Statistical analysis The comparison of phenotypic frequencies was obtained by using chi-square analysis with Yates' correction or by Fisher's Exact test, whenever appropriate. The application of the observed χ 2 vs. expected χ 2 was employed to show significance of frequency differences with the sex ratio. A p value of less than 0.05 was considered statistically significant (two-by-two table: degrees of freedom (df) = 1, chi-square ≥ 3.84) [ 32 ]. The association between HLA antigens and type 1 diabetes was performed by determination of odds ratio (OR) [ 33 , 34 ] (confidence interval, 95 % CI). All statistical analyses were performed using the Epi Info 2000 Version 1.0 for Windows 95, 98, NT, and 2000 computers (Epi Info, Atlanta, Georgia, USA) and STATISTICA Version 5.0, '97 (STATISTICA, StatSoft, Paris, France). Author's contributions AM drafted the manuscript, performed statistical analyses and carried out the bibliography research. MS participated by coordinating and orienting the designated study. BA carried out HLA phenotyping and participated in the study design. KM recruited the eligible subjects. All authors read and approved the final version of the manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC515298.xml
515307	An unusual presentation of anetoderma: a case report	Background Anetoderma is a benign condition with focal loss of dermal elastic tissue resulting in localized areas of flaccid or herniated saclike skin. Currently, anetoderma is classified as either primary (idiopathic), or secondary anetoderma (which is associated with a variety of skin conditions, penicillamine use, or neonatal prematurity). Lesions appear on the upper arms, trunk, and thighs. Case presentation We report a 14-year-old boy, which was noticed to have had multiple, white, non-pruritic areas on the acral sites of upper and lower extremities for two years. In physical examination, the patient had normal mental development. Skin lesions consisted of scattered, white to skin-colored papules, less than 1 cm in diameter, and with central protrusion, with distribution on dorsal part of the index finger, forearms, distal portion of thighs and calves. Lesions were detected neither on the trunk nor the proximal areas of extremities. There are no sensory changes associated with the lesions. Otherwise, his general health was good. He did not have any medication consumption history. Family history was negative. Laboratory examinations were within normal limits. Skin biopsy from one of his lesions was done, that confirmed the diagnosis of anetoderma. Conclusions In summary, we report a case of anetoderma on unusual sites of the skin. We could not find similar reports of anetoderma developing on distal extremities without involvement of the upper trunk and proximal arms, in the medical literature.	Background The term anetoderma (anetos = slack) refers to a circumscribed area of slack skin associated with a loss of dermal substance on palpation and a loss of elastic tissue on histological examination [ 1 ]. In the past, cases of primary anetoderma were divided into the Jadassohn-Pellizari type, in which the lesions are preceded by erythema or urticaria, and the Schweninger-Buzzi type, in which there are no preceding inflammatory lesions. This is now of historical interest only, because in the same patient some lesions may be preceded by inflammation and the others may not, and the prognosis and histology are identical in the two types [ 2 , 3 ]. Anetoderma is a rare disorder that in the most usual form develop on the trunk, thighs and upper arms, less commonly on the neck and face and rarely elsewhere. The scalp, palms and soles are usually spared. We report a patient with anetoderma whose lesions present on distal extremities consisting of hands and calves. Case presentation A 14-year-old boy was noticed to have had multiple white, non pruritic area on his distal extremities for two years. The lesions consisted of whitish papules and depressed areas with central protrusion. On clinical examination, the otherwise healthy looking patient's general appearance and mental state had lumbar lordosis, laxity in large joints, tibia vara, high-arched palate, and dental misalignment. Skin lesions consisted of scattered, white to skin-colored papules, less than 1 centimeter in diameter, and with central protrusion, that were distributed on the dorsum of fingers (Fig 1 ), forearms (Fig 2 ), distal portion of the thighs and on the calves (Fig 3 ). No lesions on the trunk or proximal areas of extremities were detected. Palms, soles, dorsum of feet and mucosal membranes were spared. No sensory changes associated with the lesions. He did not have any history of medication consumption. Family history was negative. Figure 1 Solitary lesion over dorsum of left index finger. Figure 2 An anetodermic lesion on forearm. Figure 3 An anetodermic lesion on lateral aspect of left lower leg (before cryotherapy) Figure 4 Anetoderma lesions some minute after cryotherapy Laboratory examinations consisting of complete blood count, urinalysis, and blood chemistry including erythrocyte sedimentation rate and liver function tests were within normal limits. Antinuclear antibody was negative. The patient had no risk factor for AIDS or syphilis, so we did not request HIV or VDRL test. Hepatitis B surface antigens were not detected. Immunological assays consisting of IgM, IgA, IgG, IgE, C3, and C4 levels were normal. An induration of 0.5 centimeter in diameter was observed after tuberculin testing. Chest x-ray film was normal. The skin biopsy was done. Haematoxylin and eosin stained section showed faintly eosinophilic separated collagen fibers in the upper and mid-dermis (Fig 6 ). Verhoeff-vanGieson stained sections showed a marked decrease or in some areas total absence of elastic fibres, in both superficial and mid-dermis. Elastic fibres around the vessels in the affected areas are also fragmented and markedly decreased (Fig 7 , 8 ). We tried to treat the patient with liquid nitrogen cryotherapy by means of cotton-tip applicator for two freeze-thaw cycles (freeze time, 10–15 seconds), in 6–8 sessions weekly; which obtained moderate improvement in some early-onset lesions with no frank atrophy (Fig 4 and 5 ). Figure 5 The same view as Fig. 4 after several sessions of cryotherapy (the 2 lower lesions near completely been resolved). Figure 6 Faintly eosinophilic separated collagen bundles in upper dermis (H&E ×250). Figure 7 Severely decreased elastic fibres in superficial and mid dermis (Verhoeff-van Gieson stain ×250). Figure 8 Severely decreased elastic fibres in superficial dermis (Verhoeff-van Gieson stain ×400). Discussion Anetoderma, which was first described by Jadassohn in 1892, is characterized by localized areas of loss of substance and elastic tissue with flaccid skin and often leads to a herniation phenomenon [ 2 ]. We could not find similar reports (other than anetoderma-like changes on distal extremities secondary to hamartomatous congenital melanocytic naevi) [ 4 ] of anetoderma developing on distal extremities without involvement of the upper trunk and proximal arms, in the medical literature. This rare disorder occurs mainly in women aged 20–40 years, but is occasionally reported in younger and older patients of both sexes. It is perhaps more frequent in central Europe than elsewhere, which suggests a possible relationship to chronic atrophic acrodermatitis (due to Borrelia species) in some cases. In the most usual form, crops of round or oval, pink macules 0.5–1 centimeter in diameter develop on trunk, thighs and upper arms, less commonly on the neck and face and rarely elsewhere[ 3 ]. The scalp, palms and soles are usually spared. Each macule extends for a week or two to reach the size of 2–3 centimeter [ 3 ]. Sometimes there are larger plaques of erythema, and nodules have also been reported as primary lesion [ 5 ]. The number of lesions varies widely, from less than five to one hundred or more [ 3 ]. The lesions remain unchanged throughout life, and new lesions often continue to develop for many years. If the lesions coalesce, they form large atrophic areas, which are indistinguishable from acquired cutis laxa [ 3 ]. They may become confluent, to cover large areas, especially at the roots of the limbs and on the neck [ 3 ]. Although infrequently reported, anetoderma may occur in families, and patient must be examined for associated systemic abnormalities for thorough assessment of their skin disorders. In familial anetoderma, there were associated ocular, gastrointestinal or orthopedic anomalies in the affected patients or in any other family members, but causes without them have been reported [ 8 ]. Although isolated and perhaps coincidental, these abnormalities could be related to the same process that produces the lesions of anetoderma [ 3 ]. Primary anetoderma can be inherited, but it has also been described in association with prematurity, lupus erythematosus, antiphospholipid syndrome, and with decreased serum levels of alpha-1-antitrypsin [ 6 , 7 , 9 , 10 ] and [ 11 ]. Secondary anetoderma develops over other dermatoses are shown in Table 1 . Table 1 Dermatoses Associated with Anetoderma Syphilis [3] Sarcoidosis [17] Granuloma annulare [25] Tuberculosis [3] Acne vulgaris [17] Hepatitis B virus immunization [26] Xanthomas [3] Leprosy [17] Primary Sjogren's syndrome [27] Nodular amyloidosis [3] Lupus erythematosus [17, 21] Lichen planus [28] Melanocytic naevi [4] Pilomatricoma [18, 19] Insect bites [28] Low serum level of α-1-antitrypsin [6] Prurigo nodularis [20] Lupus profundus [30, 31] Antiphospholipid syndrome [6, 11, 1nd 24] Cutaneous plasmacytoma [21] Discoid lupus (with herediyary C2 defficiency) [32] Recurrent deep vein thrombosis [7] Benign cutaneous lymphoid hyperplasia [21] Pityriasis versicolor [33] History of Graves' disease [7] Urticaria pigmentosa [22, 23] Dermatofibroma [34] Familial type [8] Perifolliculitis [23] Penicillamine-induced [34, 40] Prematurity [10] Varicella [24] HIV-infection [42] The differential diagnosis of anetoderma includes other focal dermal atrophies and miscellaneous diseases that must be differentiated from the skin herniation phenomenon of anetoderma [ 12 ], are shown in Table 2 . Table 2 Differential Diagnosis of Anetoderma [12-16] Atrophic scars Discoid lupus erythematosus Lichen sclerosus et atrophicus Atrophoderma of Pasini and Pierini Corticosteroid-induced atrophy Perifollicular macular atrophy Morphea Perifollicular atrophoderma Atrophoderma vermiculare Striae distensae Focal dermal hypoplasia Naevus lipomatosus Connective tissue naevus Neurofibromas Cutis laxa postinflammatory elastolysis mid-dermal elastolysis Granulomatous slack skin acrodermatitis chronica atrophicans Atrophoderma of Pasini and Pierini is a major source of confusion both etymologically and clinically. Patients have larger lesions with a sharp peripheral border dropping into a depression with no outpouching. On biopsy, elastin is normal, while collagen may be thickened, but this finding is difficult to quantify [ 12 ]. Perifollicular atrophoderma is most prominent on the dorsa of the hands and often is associated with multiple basal cell carcinomas and hair abnormalities in the Bazex syndrome [ 13 ]. Perifollicular atrophy also has been described in extreme forms of keratosis pilaris, in which large keratin plugs may produce a dilated patulous follicle. This condition usually found on the cheeks of young children. Both of these lesions mimic perifollicular anetoderma but lack elastin changes [ 12 ]. In focal dermal hypoplasia thinning or absence of dermis, rather than changes in elastin fibres, accounts for the proximity of the subcutis to the epidermis [ 12 ]. Cutis laxa, postinflammatory elastolysis [ 14 ], and mid-dermal elastolysis [ 15 ] share with anetoderma the property of cryptogenic loss of elastic fibres. Elastase-producing strains of staphylococcus epidermidis have been held responsible for perifollicular macular atrophy. Anetoderma has also been reported in 5 patients with false-positive syphilis serology, 3 of who also fulfilled the criteria for the antiphospholipid syndrome [ 29 ]. Its pathogenesis is not yet clearly established, but immunological mechanisms could play an important role in dermal elastolysis [ 35 ]. The association of primary anetoderma with decreased levels of alpha-1-antitrypsin may be of significance: Alpha-1-antitrypsin inhibits neutrophil elastase and its reduction may cause increased elastic activity and elastin breakdown. Phagocytosis of elastic fibres by macrophages has been found in primary anetoderma [ 36 ]. No antibodies have been demonstrated against elastic fibres [ 37 ]. Venencie et al. [ 38 ], suggested that the degradation of elastic fibres in patients with anetoderma is caused by enhanced expression of progelatinases A and B and production of the activated form of gelatinase A, and that the lack of control of these enzymes by tissue inhibitors of metalloproteinases is probably a key factor in the development and duration of anetodermic lesions. Ghomrasseni et al. [ 39 ], demonstrated that for the five samples of anetodermic skin, matrix metalloproteinase-1 (MMP-1) levels were significantly higher compared with the uninvolver cultures and the healthy samples. A significant increase of tissue inhibitors of metalloproteinase (TIMP-1) expression was also observed in the affected cultures of explants. The study demonstrated a significant increase in the production of gelatinase A (MMP-2), and no significant production of TIMP-2 in lesional skin compared with the samples from the two healthy donors. Penicillamine-induced anetoderma has also been reported [ 3 , 34 ] and [ 40 ]. Penicillin and the antifibrinolytic drug ε-aminocaproic acid have been advocated [ 41 ], but Venencie et al. [ 3 ] studied 16 patients and found that no treatment was beneficial once the atrophy had developed. However, the wrinkled skin appearance in our patient had been present for 2 years, and his lesions did not show any signs of inflammation or pre-existing conditions like melanocytic naevi. Conclusions In summary, we report a case of anetoderma with lesions on unusual sites. We did not find similar reports of acral anetoderma in the medical literature. According to this paper liquid nitrogen cryotherapy has moderate efficacy in the treatment of some of the early anetoderma lesions, without frank atrophy. Competing interest None declared. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC515307.xml
512291	Hedgehog-interacting protein is highly expressed in endothelial cells but down-regulated during angiogenesis and in several human tumors	Background The Hedgehog (Hh) signaling pathway regulates a variety of developmental processes, including vasculogenesis, and can also induce the expression of pro-angiogenic factors in fibroblasts postnatally. Misregulation of the Hh pathway has been implicated in a variety of different types of cancer, including pancreatic and small-cell lung cancer. Recently a putative antagonist of the pathway, Hedgehog-interacting protein (HIP), was identified as a Hh binding protein that is also a target of Hh signaling. We sought to clarify possible roles for HIP in angiogenesis and cancer. Methods Inhibition of Hh signaling by HIP was assayed by measuring the induction of Ptc-1 mRNA in TM3 cells treated with conditioned medium containing Sonic hedgehog (Shh). Angiogenesis was assayed in vitro by EC tube formation on Matrigel. Expression of HIP mRNA was assayed in cells and tissues by Q-RT-PCR and Western blot. HIP expression in human tumors or mouse xenograft tumors compared to normal tissues was assayed by Q-RT-PCR or hybridization of RNA probes to a cancer profiling array. Results We show that Hedgehog-interacting protein (HIP) is abundantly expressed in vascular endothelial cells (EC) but at low or undetectable levels in other cell types. Expression of HIP in mouse epithelial cells attenuated their response to Shh, demonstrating that HIP can antagonize Hh signaling when expressed in the responding cell, and supporting the hypothesis that HIP blocks Hh signaling in EC. HIP expression was significantly reduced in tissues undergoing angiogenesis, including PC3 human prostate cancer and A549 human lung cancer xenograft tumors, as well as in EC undergoing tube formation on Matrigel. HIP expression was also decreased in several human tumors of the liver, lung, stomach, colon and rectum when compared to the corresponding normal tissue. Conclusion These results suggest that reduced expression of HIP, a naturally occurring Hh pathway antagonist, in tumor neo-vasculature may contribute to increased Hh signaling within the tumor and possibly promote angiogenesis.	Background The Hedgehog family of genes encodes secreted signaling molecules that regulate cell proliferation and cell fate determination. In mammals, there are three such genes: Sonic hedgehog ( Shh ), Indian hedgehog ( Ihh ), and Desert hedgehog ( Dhh ). All three Hedgehog (Hh) proteins function by binding to the transmembrane receptor, Patched-1 (Ptc-1), leading to the de-repression of the membrane-bound inhibitor, Smoothened [ 1 , 2 ]. This results in activation of the transcription factor Gli-1, which induces expression of target genes that include Ptc-1 and Gli-1 itself. The increase in expression of Ptc-1 may limit the range of action of Hh by sequestering it at the surface of Hh-responsive cells [ 3 ]. Hedgehog-interacting protein (HIP) was discovered by screening a mouse cDNA expression library for proteins that bound to Shh [ 4 ]. HIP binds all three Hh proteins with an affinity equal to that of Ptc-1, and in mouse embryos it is expressed in cells adjacent to those expressing Shh, positioning it appropriately for in vivo interactions. Ectopic expression of Shh leads to ectopic HIP expression, indicating that HIP is a transcriptional target of Hh signaling [ 4 ]. Transgenic mice that overexpress HIP in the endochondral skeleton displayed a phenotype similar to that of Ihh knockout mice, consistent with an inhibitory role for HIP in Hh signaling [ 4 ]. Although it has been shown that overexpression of HIP in cells making Shh reduced the amount of Shh secreted into the media [ 5 ], no data has been published specifically demonstrating that expression of HIP in responding cells inhibits the activation of the Shh signaling pathway. During gastrulation in the mouse, Ihh is secreted by the endoderm and is sufficient to activate hematopoiesis and vasculogenesis [ 6 ]. In addition to its role in developmental processes, Shh was shown to induce angiogenesis in a murine corneal angiogenesis model, probably through the induction of the angiogenic factors VEGF, Ang-1, and Ang-2 [ 7 ]. Moreover, inhibition of Shh in the ischemic hindlimb of mice through the use of a neutralizing antibody inhibits endogenous angiogenesis [ 8 ]. Hh is also required for normal angiogenesis in the murine yolk sac, as Ihh -/- mice can initiate vasculogenesis and hematopoiesis but are defective in vascular remodeling to form blood vessels [ 9 ]. In the murine cornea, fibroblasts were identified as the Shh-responsive cells, while endothelial cells in the corneal neovessels, as well as human umbilical vein endothelial cells and microvascular endothelial cells cultured in vitro , were unable to respond to Shh, even though they express the receptor Ptc-1 [ 7 ]. Misregulation of the Hh signaling pathway has been implicated in several different types of cancer, including basal cell carcinomas (BCCs), medullablastomas, and gliomas (reviewed in [ 10 ]. Mutations in Ptc-1, which lead to constitutive activation of the Hh pathway, are the underlying factor in basal cell nevus syndrome, a familial condition characterized by a predisposition to BCC development [ 11 ]. In addition, Gli-1 was originally identified as a gene overexpressed in a human glioma line [ 12 ]. Recently, it was reported that Hh pathway activity is upregulated in digestive tract, pancreatic, and small-cell lung tumors and is required for the growth of these tumors [ 13 - 15 ]. Here we show that overexpression of HIP in cultured mouse Leydig cells inhibits the ability of these cells to respond to exogenous Shh, confirming that HIP can block Hh signaling when expressed in the same cells as the receptor Ptc-1. We have also shown that HIP is expressed predominantly in vascular endothelial cells and is downregulated in HUVEC during in vitro angiogenesis. Strikingly, HIP mRNA levels were decreased or even absent in several types of human tumors, as well as in highly vascularized human tumors grown in nude mice. Taken together, these results indicate a correlation between angiogenesis and decreased expression of HIP, lending support to the role of HIP as a naturally occurring regulator of Hh signaling and neovessel formation in the adult organism. Methods Conditioned media production Adherent cultures of 293-EBNA cells were transfected with a Shh-pCDNA3 expression plasmid kindly provided by Dr. Pao-Tien Chuang (UCSF) using LipofectAMINE 2000 and OptiMem reduced-serum medium (Invitrogen). Three to five hours after transfection, cells were re-fed with serum-containing growth medium, and three or four days later conditioned medium was collected and filtered through 0.2-μm cellulose acetate filters. Endothelial cell tube formation assays For RNA isolation, human aortic endothelial cells (HAEC) were cultured in 12-well plates with or without Matrigel (BD Biosciences) for 16 hours, then cells were lysed, digested with proteinase K, and total RNA was extracted using the RNeasy Mini Kit (Qiagen). HIP expression was assayed by real-time quantitative reverse transcriptase PCR (Q-RT-PCR) and normalized to 18s rRNA (see below). Transfection and conditioned medium assays The mouse testicular epithelial cell line, TM3, was grown in complete medium (DMEM/F12 with 2.5% FBS, 5% horse serum, 15 mM Hepes, and 4 mM glutamine) at 37°C and 5% CO 2 . Cells were transfected in 12-well or 24-well plates at 80–90% confluence using LipofectAMINE 2000 transfection reagent and OptiMem reduced-serum medium (Invitrogen). For conditioned medium (CM) assays, cells were transfected with the plasmid MycHIP-pCDNA3 (gift of Dr. Pao-Tien Chuang, UCSF), which encodes full-length mouse HIP, or empty vector (mock). Mock or Shh CM was added to cells 36 hours after transfection, then RNA was harvested 36–48 hours later. Collection of mouse tissues PC3 human prostate cancer cells were implanted subcutaneously into male nude mice (strain nu/nu), and on day 39 the resulting tumors were excised, weighed, placed on dry ice, then stored at -80°C prior to RNA isolation. A549 lung cancer cells were implanted subcutaneously into female nude mice and harvested as described above on day 9. Normal livers or skin from 5- to 6-week-old male or 4- to 5-week-old female nude mice were excised immediately following cervical dislocation and frozen in liquid nitrogen, then stored at -80°C. RNA isolation RNA was extracted from cultured cells using the RNeasy kit (Qiagen), including on-column DNase I digestion, according to the manufacturer's instructions. To extract RNA from frozen normal mouse tissues and tumors, hard-frozen tissue samples of 100–300 μg were pulverized over dry ice, placed in denaturing buffer, and disrupted using a Mixer Mill (Retsch) or rotor-stator homogenizer (Omni International). RNA was isolated from the homogenate either by phenol-chloroform extraction followed by precipitation with isopropanol and washing with 70% ethanol (ToTALLY RNA kit, Ambion) or by using the RNeasy kit (Qiagen). All RNA samples were treated with DNase I (Qiagen) prior to quantitative RT-PCR. Concentrations of RNAs were determined by reading the absorbance at 260 nm. Quantitative RT-PCR Quantitative RT-PCR was performed using the ABI PRISM™ 7700 Sequence Detection System and TaqMan™ chemistry. Reactions contained 0.3 μM of each primer, 0.1 μM probe, 0.25 U/ml MultiScribe reverse transcriptase, 0.4 U/ml RNase inhibitor, and 1x PCR master mix (Applied Biosystems). Primers and probes used to quantitate the mRNA level of various genes were as follows: mouse Ptc-1 (forward primer 5'GCCAATGGCCTAAACCGACT, reverse primer 5'AAACCGGACGACACTTGGAG, probe 5'6FAM-CCCACTCCTTCGCCTGAGCCG-TAMRA), mouse HIP (forward primer 5'CCACTGACCTCCGATTGCTC, reverse primer 5'TGCAGCAGCACTTGCCAG, probe 5'6FAM-CGGCTCTGTCGAAACGGCTACTACACC-TAMRA), mouse vWF (forward primer 5'CCGGAAGCGACCCTCAGA, reverse primer 5'CGGTCAATTTTGCCAAAGATCT, probe 5'6FAM-TGGCCTCTACCAGTGAGGTTTTGAAGTACACAC-TAMRA), mouse CD146 (forward primer 5'GGGCCTCAGGCAACTTCA, reverse primer 5'TTGGTGCACACGGAAAATCA, probe 5'6FAM-CTCCTTGTGAATCAAAAACCAGTCCACTTGG-TAMRA), human HIP (forward primer 5'CCCACACTTCAACAGCACCA, reverse primer 5'GCACATCTGCCTGGATCGT, probe 5'6FAM-CCCCGAAGTGTTTGCTCATGGGCT-TAMRA), human vWF (forward primer 5'TGAAGTATGCGGGCAGCC, reverse primer 5'GCGGTCGATCTTGCTGAAG, probe 5'6FAM-CCTCCACCAGCGAGGTCTTGAAATACAC-TAMRA). Mouse and human GAPDH and ribosomal RNA probe and primer sets were purchased from Applied Biosystems. RNA quantities were determined from a standard curve of serially diluted total cellular or tissue RNA run in parallel with each set of reactions. Standard curves had a slope between -3.1 and -3.3 and correlation coefficients of 0.98 or greater. Western blot Total cell lysates were made by lysing cells in 2% SDS, 10% glycerol, and 0.063 M Tris-HCl (pH 6.8) containing a cocktail of protease inhibitors (Pierce). Lysates were heat denatured at 95°C for 10 minutes, passed through a 26G needle to shear DNA, and centrifuged at 10,000 g for 30 minutes to remove insoluble material. Protein concentrations were determined using the BCA assay (Pierce). Conditioned medium samples were diluted 1:1 in Laemmli sample buffer (Sigma). 50 μg of cell lysate or 20 μl of conditioned medium/sample buffer was resolved on 4–15% polyacrylamide gels (BMA) and transferred to nitrocellulose (Invitrogen). Even loading of protein samples was verified by staining with Ponceau S. After overnight blocking with 5% nonfat dry milk, 0.05% Tween-20 in phosphate-buffered saline, blots were incubated with antibodies against Shh (sc-1194, Santa Cruz), Myc tag (Invitrogen), or HIP (provided by Dr. Pao-Tien Chuang, UCSF). Antibody-antigen complexes were visualized using horseradish peroxidase-conjugated secondary antibodies (Jackson ImmunoLabs, Santa Cruz) and SuperSignal West Pico chemiluminescent substrate (Pierce). Hybridization of RNA probes to cancer array Linearized templates containing a 716-bp fragment of human HIP cDNA (bases 521–1236 of human HIP coding sequence) or a 777-bp fragment of human vWF cDNA (bases 7301–8077 of human vWF coding sequence) were radiolabelled with 32 P-dUTP by in vitro transcription using the MAXIscript kit (Ambion). The array was pre-hybridized in Clontech ExpressHyb solution with 100 μg/ml boiled, sheared salmon testis DNA (SS-DNA) for 2–12 hours at 68°C with rotation. Pre-hybridization solution was replaced with fresh hybridization solution with SS-DNA, plus a total of 1.5–1.7 × 10 7 cpm of 32 P-labelled HIP or vWF probe. Hybridization of the probe proceeded for 16–18 hours at 68°C, followed by four washes in 2X SSC/0.5% SDS at 68°C, one wash in 0.2X SSC/0.5% SDS at 68°C, and one wash in 2X SSC at room temperature. The array was exposed to a phosphor-screen for 24–72 hours and scanned in a Storm phosphor-imager at a resolution of 50 μm. After hybridization to the HIP probe, the same array was stripped by boiling in 0.5% SDS and re-scanned to verify the absence of residual probe before hybridization to the vWF probe. Relative levels of HIP and vWF expression were determined using ImageQuant software (Molecular Dynamics). Statistical analysis All results are expressed as mean ± standard deviation unless otherwise noted. Statistical significance of differences was determined using a two-tailed Student's t-test. Results HIP is highly expressed in endothelial cells Ptc-1 and Smo mRNA are expressed in human aortic endothelial cells (HAEC) at levels comparable to that of GAPDH, as determined by comparing the threshold cycle numbers obtained by quantitative RT-PCR (Q-RT-PCR) (data not shown). Thus HAEC express components of the Hh pathway that are known to mediate a response to Hh ligands. However, when HAEC or human umbilical vein endothelial cells (HUVEC) were treated with conditioned medium containing functional Shh, the Hh pathway was not activated as determined by measuring the levels of mRNA for Ptc-1 and Gli-1, two genes known to be responsive to Hh signaling (data not shown). Shh mRNA levels in HAEC and HUVEC were near the limits of detection by the Q-RT-PCR assay and could therefore be estimated at less than 10 -3 copies per cell based on the typical sensitivity of this assay. Hedgehog-interacting protein (HIP) binds Hh proteins with an affinity equal to that of Ptc-1 and can function as an antagonist of Ihh signaling in vivo [ 4 ]. Thus we hypothesized that HIP may be expressed in EC and function to block Hh signaling. Indeed, we found that HAEC and HDMEC expressed amounts of HIP mRNA similar to that of GAPDH, as measured by Q-RT-PCR. Endothelial cell-predominant expression of HIP was demonstrated by Q-RT-PCR analysis of a variety of human and mouse primary cell strains and immortalized cell lines and human cancer cell lines representing different tissues. When normalized to the level of GAPDH mRNA, the level of HIP mRNA was 100- to 10,000-fold higher in vascular endothelial cells than in the other cell types examined (Figure 1 ). Expression of HIP protein in HAEC, but not in ZR75-1 or HT-29, was confirmed by Western blot (Figure 1 , inset). The observation that HIP mRNA is expressed at a high level in human endothelial cells suggests that HIP may inhibit Hh signaling in these cells and may explain why these cells are unresponsive to Shh. HIP inhibits Ptc induction by Shh Conditioned media from 293 cells co-transfected with Shh and full-length (membrane-anchored) HIP contained a reduced amount of Shh and had a reduced ability to induce differentiation in C3H10T1/2 cells compared to cells transfected with Shh alone, demonstrating that HIP can sequester Shh [ 5 ]. However, there is no direct evidence that cells in which HIP is overexpressed are less responsive to Shh. In order to test this directly, 293-EBNA cells were transfected with a Shh expression plasmid. Western blot analysis demonstrated the presence of Shh protein in the conditioned medium from these cells (Figure 2 , inset left). Exposure of TM3 mouse Leydig cells, a cell line with no detectable endogenous HIP mRNA (Figure 1 ), to increasing amounts of conditioned media containing Shh resulted in induction of the Shh pathway, as measured by a dose-dependent increase in the amount of endogenous Ptc-1 mRNA (Figure 2 ). When TM3 cells were transfected with a plasmid expressing full-length, N-terminally Myc-tagged mouse HIP, the induction of Ptc-1 mRNA by Shh was reduced by 79 percent (p < 0.05; Fig. 2 ), demonstrating that HIP functions as an inhibitor of the Shh pathway when expressed in the responding cells. Expression of HIP in transfected TM3 cells was verified by Western blot using a Myc antibody (Fig. 2 , inset right). HIP is downregulated in endothelial cells during tube formation in vitro Shh has been shown to induce angiogenesis [ 7 ], and HIP is able to antagonize the Shh pathway (Figure 2 ). Therefore, we tested whether HIP expression is downregulated in EC during tube formation, an essential step in the angiogenic process. Expression of HIP mRNA in HAEC that had formed tubes on Matrigel as measured by Q-RT-PCR was 2.9-fold lower (p < 0.05) than in the same cells cultured in standard plastic dishes (Figure 3A ). To investigate whether there was a coordinated downregulation of Hh-regulated genes during tube formation on Matrigel, the mRNA level of Ptc-1, a known Hh-responsive gene, was measured. In contrast to HIP, levels of Ptc-1 mRNA increased 2.5-fold (p < 0.05) in cells on Matrigel compared to plastic (Figure 3B ). HIP is downregulated in human tumors compared to normal tissues Angiogenesis is a common feature of tumor growth (reviewed in [ 16 , 17 ]. It is conceivable that HIP is downregulated in this process, similar to the reduction we observed in endothelial cells during tube formation. Therefore, we measured HIP mRNA levels in a variety of tumor samples. Tumor and corresponding normal tissue RNA samples (normal and tumor tissues were from different individuals) were obtained from two different commercial sources (Ambion and BioChain), and the endothelial cell marker vWF was used to normalize HIP expression to the number of endothelial cells present in the tumors. In all tissue and tumor RNA samples assayed, vWF mRNA was detectable. In both pairs of liver tissue examined, HIP mRNA was expressed in normal tissue but was undetectable in the tumor (Table 1 ). One of two kidney tumor-normal pairs showed no difference in normalized HIP expression, while the other showed a 5.9-fold reduction of normalized HIP in tumor relative to normal tissue. Of the two breast tissue pairs, one had undetectable levels of HIP in both normal and tumor tissues (data not shown), while the other showed a 3.7-fold decrease in HIP expression in tumor compared to normal. In one lung tissue pair, vWF levels in normal tissue were over 900-fold lower than in the tumor, such that normalization of HIP to vWF resulted in an apparent 30-fold increase in HIP in tumor compared to normal tissue. In the other lung tissue pair, normalized HIP was 8.1-fold lower in tumor than in normal tissue. To further investigate the frequency of the change in HIP expression in human tumors, a cancer profiling array (Clontech) containing cDNA from 154 tumor and corresponding normal tissues from individual patients was used to compare the expression of HIP in a larger set of samples. A 32 P-labelled HIP probe corresponding to bases 521–1236 of human HIP cDNA was hybridized to the array, and relative levels of HIP expression were quantitated on a phosphoimager. HIP expression levels were normalized to the expression of the endothelial marker vWF, by re-probing the same array with a fragment of the human vWF cDNA, as different tumors and normal tissues may contain different degrees of vascularization. In liver, stomach, colon, and rectum, all or most of the paired samples showed a reduction of HIP in tumor compared to normal tissue (Table 2 ). In total, 28 samples had a decrease in the tumor, while only 3 samples showed a slight increase (1.3- to 1.7-fold), and 2 samples showed no change. The lung samples did not show a consistent patten of HIP mRNA expression. Five had a decrease of HIP mRNA in the tumor, while four had an increase and one showed no change. The remaining sets of paired normal and tumor tissues, including samples from breast, ovary, vulva, uterus, cervix, prostate, testis, thyroid, skin, bladder, small intestine, and pancreas, showed no marked difference in HIP expression between normal and tumor, or had undetectable HIP expression in both normal and tumor. HIP expression is very low or undetectable in xenograft tumors Mouse xenograft tumors resulting from transplantation of PC3 (human prostate cancer cell line) or A549 (human lung cancer cell line) cells into nude mice are known to have high levels of neovascularization as a result of angiogenesis, which is required for their continued growth [ 18 , 19 ], so we hypothesized that HIP expression would be lower in these tumors than in normal tissues that also contain normal vessels. RNA extracted from PC3 or A549 tumors weighing an average of 350 mg that were excised 39 (PC3) or 9 (A549) days after subcutaneous implantation, or from liver or skin from non-implanted nude mice, was assayed for murine HIP, vWF, CD146, and GAPDH mRNA by Q-RT-PCR. The Q-RT-PCR assay for murine HIP is species-specific, i.e. does not detect human HIP mRNA (data not shown). Therefore, any HIP mRNA detected using this assay must be from host mouse cells present in the tumor, not from the human tumor cells themselves. Murine rather than human HIP was assayed because tumors have been shown to incorporate vasculature from the surrounding vessels [ 20 - 22 ] and there is evidence that tumors can recruit endothelial and hematopoietic precursor cells [ 23 ] to form their own vasculature. RNA from tumors and normal liver and skin contained measurable amounts of the murine endothelial cell markers vWF and CD146, indicating the presence of endothelial cells (Figure 4 ). For the PC3 xenograft experiment, the levels of vWF and CD146 mRNA normalized to GAPDH were actually higher in the tumors than in the normal liver, consistent with the highly vascularized nature of these tumors. In contrast, mouse HIP mRNA was abundant in all liver and skin samples examined but was 16- to 30-fold lower in A549 tumors and undetectable in any of the PC3 tumor samples (Figure 4A ). These data demonstrate that although the tumors contained endothelial cells, they expressed markedly reduced or undetectable amounts of HIP mRNA. Discussion Our demonstration that HIP functions to inhibit Ptc-1 upregulation in mouse testicular epithelial cells (TM3) exposed to Shh-containing conditioned media is the first direct evidence to date that Shh signaling is attenuated in cells that express full-length, membrane-bound HIP. The decrease in Ptc-1 induction caused by transfection with HIP is consistent with previous reports indicating that HIP can bind and reduce the availability of Shh and presumably prevent it from signaling through Ptc-1 [ 4 , 5 ]. In human aortic endothelial cells (HAEC), which express high levels of HIP, the addition of Shh-containing conditioned medium did not cause induction of Ptc-1, even though these cells express the receptor components Ptc-1 and Smo. In addition, others have shown that although Shh induces angiogenesis in a mouse model of hindlimb ischemia, it has no effect on endothelial cell proliferation or migration in cell culture [ 7 ]. Our finding that vascular EC express abundant amounts of HIP mRNA may explain the inability of these cells to respond to Shh. An analysis of various human cell lines and primary cells indicated that HIP is absent or expressed at low levels in other cell types, suggesting that in adults HIP is expressed primarily in EC. These results are supported by gene chip data analysis of more than 30 normal human tissues showing that HIP is most highly expressed in blood vessels or in vascular-rich tissues such as liver, lung, brain, and pancreas (data not shown). These results suggest a role for HIP in the normal function of blood vessels. Several lines of evidence support the hypothesis that HIP expression is decreased in EC participating in angiogenesis. Firstly, we describe that HIP is downregulated in HAEC forming tubes on Matrigel. In contrast, Ptc-1 mRNA levels were increased in EC forming tubes on Matrigel, suggesting that the decrease in HIP mRNA under these conditions does not reflect a general downregulation of Hh-responsive genes. Given that transcription of HIP and Ptc-1 are both activated by Hh signaling, it is likely that the decrease in HIP expression in EC on Matrigel is mediated by a pathway independent of Hh signaling. Secondly, we have observed that HIP mRNA levels are decreased in human tumors and xenograft human tumors grown in mice, situations in which angiogenesis occurs. Previous studies have demonstrated a requirement for Shh in angiogenesis in the ischemic mouse limb and embryonic yolk sac [ 8 , 9 ]. The pro-angiogenic activity of exogenously supplied Shh protein has been attributed to the induction of factors such as VEGF and angiopoietin in fibroblasts [ 7 ]. Our data suggest the possibility that in situations where HIP expression is downregulated, Hh may also act directly on endothelial cells. The Shh pathway has been implicated in several different types of cancer, where upregulation of signaling, associated with mutations in Patched or amplification of Gli, is often the suspected cause of malignancy [ 11 , 12 , 24 - 26 ]. In at least one type of human cancer, basal cell carcinoma (BCC), HIP upregulation has been demonstrated [ 27 , 28 ], probably as a result of dysregulation of the Shh signaling pathway. More recently it has been reported that Hh pathway activity is activated in a wide variety of digestive tract tumors by overexpression of the Hh ligand, and is essential for tumor growth [ 13 ]. Aberrant expression of Shh and activation of Hh signaling was also demonstrated to occur in pancreatic cancer [ 14 ] and small cell lung cancer [ 15 ]. Hh ligands secreted by tumor cells have the potential to induce Hh pathway activation in nearby non-tumor cells, including fibroblasts and endothelial cells. Our observation that HIP mRNA expression decreases in several human tumor types relative to normal tissues suggests that modulation of HIP expression may also promote increased Hh signaling in certain tumors and contribute to cancer progression. However, since expression of HIP is known to be induced by activation of the Hh pathway, an alternative interpretation is that a reduced level of HIP mRNA in tumors is indicative of reduced Hh signaling, as suggested by Hu et al. [ 29 ]. Our finding that, in endothelial cells in Matrigel, HIP mRNA levels decrease while Ptc-1 levels increase demonstrates that a mechanism other than Hh signaling can modulate the expression of these Hh target genes. Thus, the down-regulation of HIP in tumors may be unrelated to Hh pathway activity. The precise mechanism of HIP regulation by the Hh pathway in the context of tumors remains to be elucidated by future studies. Among the tumor types represented on the Clontech Cancer Profiling Array, liver and several areas of the digestive tract (stomach, colon, rectum) exhibited consistent downregulation of HIP in the tumor relative to normal tissue. Tumors from other tissues, including breast, testis, and kidney, did not show a significant change in HIP expression from the normal tissue, suggesting that this phenomenon may be restricted to certain tumor types. Differences in the degree of vascularization of the various tumors may also explain why a decrease in HIP was not seen in all tumor types. Xenotransplantation of PC3 or A549 cells into immunocompromised mice is a commonly used animal model for prostate or lung cancer, respectively. Extensive neovascularization occurs in the resulting tumors, and it has been demonstrated that angiogenesis inhibitors slow their growth [ 30 , 31 ]. The discovery that vWF-normalized mouse HIP mRNA levels were markedly decreased in A549 tumors and undetectable in all of the PC3 tumors we examined gives further weight to the hypothesis that decreased HIP expression is associated with tumor angiogenesis. Angiogenesis has been well-characterized as a prerequisite to continued tumor growth and metastasis [ 16 , 17 , 32 ], and many approaches have been developed to attempt to control the progression of cancer by inhibiting angiogenesis [ 32 , 33 ]. A more complete analysis of the gene expression changes in endothelial cells during developmental and pathological processes should help to identify additional signaling pathways that are altered and that could be manipulated to control angiogenesis and disease progression. The identification of Shh as an angiogenic signaling molecule [ 7 ], along with the characterization of HIP as an inhibitor of Shh that is downregulated in certain tumors, strongly suggests that inhibition of the Shh pathway may be of therapeutic use in various cancers that rely on angiogenesis for their continued growth. Conclusions We have shown that HIP is expressed predominantly in EC but at low or undetectable levels in other cell types, and that the high expression of HIP in EC may be responsible for their inability to respond to Shh. Expression of HIP in TM3 cells attenuated their response to Shh, demonstrating that HIP antagonizes Hh signaling when expressed in the responding cell. The reduction of HIP in a variety of human tumor samples as well as PC3 and A549 xenograft tumors, and in EC undergoing tube formation on Matrigel, suggests that modulation of HIP may play a role in tumor angiogenesis. Competing interests None declared. Authors' contributions CLO participated in experimental design and carried out CM production and assays, tissue and RNA processing, Q-RT-PCR, Western blot, and cancer array hybridization. PPH participated in the design of the study and performed tube formation assays, tissue and RNA processing, Q-RT-PCR, and Western blot. JG participated in initial characterization of HIP and provided supporting data. GMR was involved in the planning and organization of the study and reviewed the manuscript. ARB initiated, planned, and supervised the entire study. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC512291.xml
546039	Aging and Death in an Organism That Reproduces by Morphologically Symmetric Division	In macroscopic organisms, aging is often obvious; in single-celled organisms, where there is the greatest potential to identify the molecular mechanisms involved, identifying and quantifying aging is harder. The primary results in this area have come from organisms that share the traits of a visibly asymmetric division and an identifiable juvenile phase. As reproductive aging must require a differential distribution of aged and young components between parent and offspring, it has been postulated that organisms without these traits do not age, thus exhibiting functional immortality. Through automated time-lapse microscopy, we followed repeated cycles of reproduction by individual cells of the model organism Escherichia coli, which reproduces without a juvenile phase and with an apparently symmetric division. We show that the cell that inherits the old pole exhibits a diminished growth rate, decreased offspring production, and an increased incidence of death. We conclude that the two supposedly identical cells produced during cell division are functionally asymmetric; the old pole cell should be considered an aging parent repeatedly producing rejuvenated offspring. These results suggest that no life strategy is immune to the effects of aging, and therefore immortality may be either too costly or mechanistically impossible in natural organisms.	Introduction That populations survive indefinitely while individuals grow old and die requires that young offspring be produced at the expense of old parents, and this has classically been explained in terms of an immortal germ line passing through a transient and disposable soma, or body [ 1 ]. However, with the discovery of aging in single-celled organisms with no clear separation of these two constituents, it has been proposed that reproduction by asymmetric division is a prerequisite for aging [ 2 ] and that organisms that reproduce without a distinction between parent and offspring do not age, thus exhibiting functional immortality [ 3 , 4 ]. One difficulty with this distinction is that an asymmetry sufficient for aging may not necessarily result in a visible difference during cell division. Therefore, what may be a more useful common feature of the organisms that have currently been found to age is the presence of a juvenile phase (developmental asymmetry) in their life cycle. Juvenile cells are either smaller or undifferentiated and must go through a period of growth or differentiation before becoming capable of reproducing. This is similar to the idea that the critical requirement for aging in unicellular organisms is the presence of a parent cell that provides for a smaller offspring cell [ 2 ]. Consistent with this, the primary single-celled organisms that have been shown to age (the yeast Saccharomyces cerevisiae and the bacterium Caulobacter crescentus ) share the traits of a visibly asymmetric division and an identifiable juvenile phase [ 3 , 5 , 6 ]. The report of aging in the binary fission yeast Schizosaccharomyces pombe has reinforced these apparent characteristics of aging organisms. The same study that found indications of aging in this organism also found that, contrary to expectations, its cell division process was visibly asymmetric, resulting in enlarged mother cells after only a few divisions, possibly indicating that S. pombe development is similar to that of budding yeast [ 7 ]. Here we look for evidence of aging in Escherichia coli , that is, in cells that do undergo a morphologically symmetrical division and do not exhibit a juvenile phase, in order to determine if such organisms age. Previous studies of senescence in E. coli have mostly focused on the loss of viability over time during nutrient depletion (conditional senescence) [ 4 ]. However, these studies do not address aging in terms of parent and offspring (reproductive life span), nor do they address growing cells that are not under conditions of starvation. To identify reproductive life-span effects, it is necessary to follow individual cells as they grow and reproduce over time. The bacterium E. coli grows in the form of a rod, which reproduces by dividing in the middle. One new end (or pole) per progeny cell is produced during this division event ( Figure 1 ). Therefore, one of the ends of each cell has just been created during division (termed the new pole), and one is pre-existing from a previous division (termed the old pole). Old poles can exist for many divisions, and if cells are followed over time, an age in divisions can be assigned to each pole, and hence to each cell. While experiments following the partitioning of cell constituents have found uniform distributions of DNA and lipids in daughter cells [ 8 ], it is known that components of the cell wall turn over slowly and are conserved in the poles where they are formed [ 9 ]. More generally, any cell constituent with limited diffusion and a long half-life may be expected to accumulate at the old pole, so there may exist a physiological (rather than morphological) asymmetry between the old and new poles. Limited by the manual measurements required, individual cells growing under the microscope have been followed in the past for a small number of divisions [ 10 , 11 , 12 ]. The only difference between poles was seen in a single experiment that indicated that old pole cells might divide later than their new pole siblings, hinting at a possible inequality between these cells [ 13 ]. To determine if E. coli experiences aging related to the inheritance of the old pole, we followed individual exponentially growing cells in an automated fluorescence microscopy system through up to nine generations of growth and reproduction, measuring the physical parameters of each cell over time. We present conclusive evidence for aging in the old pole cell, including cumulatively slowed growth, less offspring biomass production, and an increased probability of death. Figure 1 The Life Cycle of E. coli During cell division, two new poles are formed, one in each of the progeny cells (new poles, shown in blue). The other ends of those cells were formed during a previous division (old poles, shown in red). (A) The number of divisions since each pole was formed is indicated by the number inside the pole. Using the number of divisions since the older pole of each cell was formed, it is possible to assign an age in divisions to that cell, as indicated. Similarly, cells that consecutively divided as a new pole are assigned a new pole age, based on the current, consecutive divisions as a new pole cell. (B) Time-lapse images of growing cells corresponding to the stages in (A). False color has been added to identify the poles. Results Cells were grown from one cell into a monolayer microcolony that contained up to 500 cells, and time-lapse images (see Video S1 for an example film) were analyzed with custom automated software designed for this purpose. We followed 94 such colonies, resulting in the complete record from division to division, of 35,049 cells. As the history and physical parameters of each cell in the microcolony are known, and the identity of each pole is tracked, the complete lineage can be determined. The resulting lineages from each film were averaged by each unique cell position within the lineage. This can be represented as a single, bifurcating tree, where each branch point is an average cell for that position in the lineage, and the length of the lines connecting cells to their progeny are proportional to the growth rate of the cell ( Figure 2 ). At each division in the tree, the cell inheriting the old pole of the progenitor cell is represented on the right branch of the sibling pair (in red), while the cell inheriting the new pole is on the left branch (in blue). Figure 2 Average Lineage Showing Old Pole Effect on Growth Rate The first division in the microcolonies is not represented, as the identity of the poles is not known until after one division (hence each initial cell gives rise to two lineages that are tracked separately, and subsequently combined from all films to create the single average lineage shown here). The lengths of the lines connecting cells to their progeny are proportional to the average growth rate of that cell; a longer line represents a higher growth rate for that cell. At each division, the cell inheriting the old pole is placed on the right side of the division pair, and shown in red, while new poles are placed on the left side of each pair, and shown in blue (note that this choice of orientation is not the same as that of Figure 1 , to compare more easily old and new pole lineages). Because the position of the start of the growth line for each new generation is dependent on the generations that preceded it, the difference in growth rates is cumulative. Green lines indicate the point at which the first cell divides in the last four generations. Nine generations from 94 films encompassing 35,049 cells are included in this tree. The average growth rate of all the cells corresponds to a doubling time of 28.2+/−0.1 min. The data used to generate the average lineage are provided in Dataset S1 . The pattern of fast and slow growth rates in this average lineage gives striking evidence for reproductive asymmetry between the progeny cells, as the cells that show a cumulatively slowed rate of growth (shorter lines) are those cells that have more often inherited an old pole during their ancestry. To verify that this pattern in the average lineage is actually due to a difference in growth rate between new and old pole cells, we performed a pairwise comparison of every set of sister cells that was produced at the eighth generation in each of the films. As sister cells share temporal and spatial surroundings, this comparison controls for potential environmental variation within the microcolony. The comparison (two-tailed t -test) includes cells of all division ages and shows that the average growth rate of the old pole progeny cell is 2.2% (+/−0.1%) slower than that of the new pole cell. This analysis, performed on 7,953 cell pairs, conclusively demonstrates ( p < 0.00001, t = 14.40, df = 7952) that the cell that inherits the old pole grows slower than the new pole cell produced in the same division. Two factors from this same dataset demonstrate the lack of a juvenile phase in E. coli. First, comparison of the progeny cells shows that the new pole cell is slightly larger on average (0.9+/−0.1%; p < 0.00001, t = 5.62, df = 7952) than the old pole cell (the contrary would be expected in the presence of a juvenile phase). Second, the young pole cell is marginally more likely to divide sooner than the old pole cell (in about 15% of the cases cells divide within the same 2-min time point; of those where the two cells divide in different time points, 54% of the time the new pole cell divides first [significant; p < 0.00001, t = 5.02, df = 4812]), which is also not consistent with a phase where the young cell must grow or differentiate before reproduction. These differences are consistent for all generations during steady-state growth (data not shown). Therefore, while a juvenile phase is absent, there is a consistent functional asymmetry between the two progeny cells that is disadvantageous to the old pole. Each cell is defined not just by its preceding division, however, but also by all previously recorded divisions, back to the initial cell in the analysis. Therefore, each old pole cell can be categorized by the number of consecutive, final old pole divisions that occurred to produce the current cell (thus giving the age in divisions of its old pole). Equivalently, each new pole cell can be assigned a number of divisions that it sequentially divided as the new pole cell. Comparing these values with the growth rates of the cells, we find that the older the old pole of a cell is, the slower the growth rate of that cell, while cells with more consecutive new pole divisions exhibit increasing growth rates ( Figure 3 A). Furthermore, a pairwise comparison shows that the difference in the growth rate between the old pole sibling (the mother cell) and the new pole sibling (daughter cell) increases with the increasing age of the mother ( Figure 3 B). Therefore, the difference between pairs of progeny cells, as well as the pattern seen in the average lineage, is not only due to a decrease in growth rate of cells that have inherited the old pole, but also to an increase in the growth rate (for at least three divisions; subsequent divisions do not detectably improve) of cells that have repeatedly inherited the new pole. Figure 3 The Effects of Consecutive Divisions as an Old or New Pole on Growth Rate (A) The cellular growth rate, represented on the y-axis, is normalized to the growth rate of all cells from the same generation and geography in each film. On the x-axis consecutive divisions are seen as either a new pole (open circles), showing rejuvenation, or an old pole (closed circles), showing aging. Cells represented at each point: new pole divisions 1–7: 7 , 730; 3 , 911; 1,956; 984; 465; 211; 89; old pole divisions 1–7: 4,687; 3,833; 1,933; 956; 465; 213; 75. (B) Pair comparison of the growth rates of sibling cells. The division age of the old pole sibling (the mother cell) is shown on the x-axis. The percentage difference between the growth rate of the new pole sibling (the daughter cell) and this cell is shown on the y-axis. A positive difference corresponds to a faster growth rate for the new pole cell. Cell pairs represented at each point, ages 1–7: 9,722; 4,824; 2,409; 1,202; 601; 282; 127. In both graphs, cells are from all 94 films. The error bars represent the standard error of the mean. The old and new pole growth rates in (A) and the pair differences in (B) are fitted to a line to show the trend; however, the actual progressions may not be linear ( R 2 old poles = 0.97, new poles = 0.83, pair difference = 0.94). To determine the longer-term effect of inheriting the old pole, we performed a second pairwise analysis, comparing the total length of offspring produced by sister cells from the fifth generation until the end of tracking (this generation was chosen as each cell has the opportunity to progress through about three divisions). As the bacteria are rod shaped, the total length of cells produced is proportional to the biomass of the offspring. The results show that old pole cells produce less offspring biomass compared to their new pole sisters (3.1+/−0.3% less, p < 0.00001, t = 9.29, df = 1565). Therefore, it appears that the slower growth rate of the old pole cells also results in a longer-term decreased ability to produce offspring biomass. Another long-term effect of aging is the probability of survival of the organism over time. During the growth of the microcolonies, sixteen cells were observed to cease growing; these cells never resumed growth during the course of the experiment. We have defined these cells as potentially dead cells and have analyzed their locations in the lineages. While these apparent deaths may ultimately be due to stochastic events, they show a statistically significant bias ( p = 0.01; see Materials and Methods ) toward increased divisions spent as an old pole (over the total observation history). This observation is consistent with the hypothesis that aged cells are more susceptible to harmful events and/or less likely to survive them. It is unlikely that these cells represent a growth arrested “persister” state, as it has recently been demonstrated that persister cells that arise during exponential growth occur at a frequency of approximately 1.2 × 10 −6 [ 14 ]; the appearance of apparently dead cells in our study (about 4.6 × 10 −4 ) is almost 400 times more frequent. Discussion We find that the old pole is a significant marker for multiple phenotypes associated with aging, namely, decreased metabolic efficiency (reduced growth rate), reduced offspring biomass production, and an increased chance of death. Thus, E. coli, an organism with a morphologically symmetrical division, no juvenile phase, and no identified separation between germ line and soma, is susceptible to aging. Unlike the process of clonal senescence [ 15 ], where an entire population progressively declines in fitness, here the life potential of the lineage is continually renewed through young offspring cells (the process of rejuvenation) that are produced at the expense of aged parent cells. That the two cells are not equal, despite appearances, indicates a functional asymmetry that may be explained by a number of mechanisms, such as those that result in the polar localization of cell components [ 16 , 17 ]. In the simplest example, any component localized to the cell poles will have more time to accumulate in an old pole than in a young one. However, the pole effect may involve more active processes (such as differential turnover or accumulation) because, in addition to the aging effect of the parent, new pole cells show a concomitant increase in their growth and reproduction over several divisions. This result is not without precedent in aging organisms. In the budding yeast S. cerevisiae , the daughter cell of a young mother cell has a greater reproductive life span compared to the daughter cell of an old mother cell [ 18 ]. A likely explanation for this effect in yeast is the loss of segregation control of detrimental cellular components, such as extrachromosomal rDNA circles [ 19 ], and possibly damaged mitochondria as well [ 20 ]. Additionally, in fruit flies (Drosophila melanogaster) , several generations of daughters are required to recover from the effects of chromosome damage [ 21 ]. In our study, the new pole cells can apparently benefit from the preferential distribution of cellular components at the expense of the old pole cell for at least three divisions. After further divisions as a new pole cell, it is not clear that there is a continuing benefit to the cell. The observation that an optimum is not reached in a single division implies that the mechanism responsible for an age-related bias in cell component inheritance is not absolute; repeated rounds of sorting continue to improve the condition of new pole cells. On the other hand, the observed rate of decline in old pole cells (about 1% of the initial value per division) would result in a complete cessation of growth after about 100 divisions. While the behavior of cells with more than seven old pole divisions cannot be determined from these data, it is interesting to note that the observed rate of decline in offspring production in C. crescentus (a bacterium with an obviously asymmetric division and a developmentally significant juvenile phase) is of about the same magnitude as the decline in growth rate measured here [ 3 ]. In S. cerevisiae , the rate of offspring production also declines, resulting in cell cycle times that are as much as five times longer for mother cells that have divided 30 times than for young cells. This is an accelerating decline, however, and there is no detectable difference in cell cycle time for the first ten divisions [ 22 ]. Our results show that a juvenile phase is not required for the process of aging any more than the presence of a germ line or a visibly asymmetric division is. In contrast, we demonstrate the presence of a physiological asymmetry in E. coli, which is essential for the process of aging. The cost of the aging process in lost growth to the population under these conditions is about 2%. In competition, these cells would be rapidly displaced by competitors that did not age, but only if the cost of avoiding senescence were not equal or greater. It has been proposed that one such cost is the rigorous level of maintenance and repair that would be required to prevent the decline and eventual extinction of a perfectly symmetrical organism [ 23 ]. The physiological asymmetry during division may therefore represent the disposal by preferential partitioning of cellular damage that is expensive or impossible to repair. Concerning the evolutionary origin of the aging process seen here, it is possible that an asymmetry of division existed before aging appeared as a life history trait in these cells, and that such an inequality may have therefore allowed (or forced) aging to occur. However, as we have detected this asymmetry solely through phenotypes that can be linked to aging, it is equally possible that the necessity of aging is itself the evolutionary cause of the asymmetry. If this latter explanation is found to be correct, then the occurrence of aging in a single-celled organism that is apparently meticulously symmetric otherwise may indicate that it is either not cost effective in general to produce an immortal life form, or it is impossible to achieve perfect molecular maintenance through natural selection. The use of the model organism E. coli provides an excellent genetic platform for studying the fundamental mechanisms of cellular aging and may provide insight into the costs and evolutionary roots of repair, maintenance, and longevity. Simple model organisms such as yeast and nematodes have already proven their value in identifying evolutionarily conserved pathways that regulate life span in higher organisms [ 24 , 25 ]. As has been seen with many fundamental life functions, it may be that the primary processes involved in aging in E. coli will also be conserved in other forms of life. Materials and Methods Strain and growth conditions The sequenced wild-type strain of E. coli, MG1655 [ 26 ], was modified to express the gene encoding yellow fluorescent protein under the control of the lactose operon repressor and the Pl promoter of lambda phage (gene construct from M. Elowitz [ 27 ]). Cells were inoculated onto microscope cavity slides from exponentially growing liquid cultures, such that the colonies and cells grew exponentially in a single plane on the surface of a solid matrix of LB-agarose (NaCl concentration of 5 g/l, supplied by SdS, Peypin, France; other components were DIFCO from Becton Dickinson, Franklin Lakes, New Jersey, United States; agarose from QA-Agarose, Qbiogene, Irvine, California, United States; plus 1 mM IPTG from Qbiogene). The slide cavities were sealed with silicone grease and contained sufficient oxygen and nutrients to allow undiminished growth and fluorescence for the length of the experiment. The slides were incubated in a temperature-controlled (Cube and Box incubation system, Life Imaging Services, Reinach, Switzerland) automated microscope (Zeiss 200M; Zeiss, Jena, Germany) at 30 °C for up to 6 h. The entire microscope was contained within the incubator, eliminating temperature gradients. These conditions resulted in an excess of nutrients and a protected environment without external causes of cell mortality. Microscopy Up to seven fields containing one to four cells each were manually identified at the start of the experiment, and stored in the MetaMorph microscope control software (Universal Imaging, Downingtown, Pennsylvania, United States). Fluorescent images were recorded at each field with time points taken generally every 4 min for the first 160 min, then every 2 min for the remaining time. A subset of six colonies was recorded every 40 s for improved time resolution. The excitation light did not affect the cellular growth rate (data not shown). Images were taken with a CoolSNAP HQ (Princeton Instruments, Trenton, New Jersey, United States) at 100X magnification; the resulting images have a spatial dimension of 0.064 μ per pixel. Excitation light (480 to 520 nm) was limited to 5% of the output of a 100-W Hg vapor lamp, with an exposure of 2 s. Emission wavelengths were 505 to 565 nm (filters from Chroma, Rockingham, Vermont, United States). See Video S1 for a sample film. Image analysis The custom analysis software (BHV) was developed by integration of open-source software under a central shell scripted in Python [ 28 ]. For pixel-intensive operations, C routines were written and linked to the Python shell. To segment the cells in the images, local minima were identified to outline the cells, then eroded to remove remaining connections, and dilated back to size. The measurement process calculated the second moments of the cell's pixel intensity/location distribution and matched it to a rectangle with the same parameters. Long, curved cells that could not be approximated to a rectangle were measured manually, using tools built into the software. Frames were then compared at successive time points, and cells were identified from frame to frame by their overlap with the previous frame, taking into account predicted cell movement due to growth of the colony. This process tracked 80% of the cells into the ninth generation without manual intervention, measuring approximations to their length, width, fluorescence intensity, orientation, geographical location within the colony, and the complete lineage history of each cell. The rate of change of each of these parameters was calculated, with the growth rate being represented by an exponential fit to length over time. Length is proportional to cell mass, as cells did not increase in width during growth. The limits to the cell-tracking process were expansion of the colony beyond the image frame and the formation of multiple layers of cells. Manual measurements were used to correct tracking errors, which allowed us to produce datasets for colonies that are up to 100% tracked. Therefore, the complete history of each cell, including how many times it divided and its relationship with the other cells in the lineage, was unambiguously determined. Statistical analysis We verified that our results are not the effect of bias unrelated to pole age in two ways. First, we compared pairs of daughters of the same cell using a pairwise t -test. Since the sibling cells occupy similar points in space and time, this eliminates influences coming from overall changes in the colony growth rate or potential variations in environment across the colony. Second, we created control datasets (data not shown) from the colonies in which all the properties of the cells and lineages are preserved, with the exception of pole identity, which is randomly assigned at each division. In this case, we tested the null hypothesis that the pole age of a cell is not a factor in cell physiology by comparing the observed result with the expected normal binomial distribution derived from the null hypothesis. In both cases, we determined from these tests the probability that the old pole effect is due to random fluctuations, expressed as a p -value. We determined if the apparently dead cells were biased toward divisions as old pole cells by examining their complete recorded division histories, which describe how many times each cell divided as an old pole or new pole cell. We compared the average number of old pole divisions of the population of 16 dead cells (mean = 3.44) with the distribution of averages (mean = 2.75, SD = 0.29) from 1,000,000 randomly generated sets of 16 cells with the same number of total divisions (total divisions = 6, 6, 4, 4, 7, 4, 5, 7, 8, 3, 6, 6, 7, 7, 6, 2). This yielded the one-tailed probability (the hypothesis was that these cells would be enriched in old pole divisions, based on our observations of growth rates) of such an old pole bias arising by chance. All error ranges in the text and figures represent the standard error of the mean. The values for Figure 3 A were normalized by finding the mean growth rate of all cells that shared similar conditions and dividing the individual growth rates of each cell by this mean. Similar conditions were defined as being in the same colony, with the same number of generations since the start of the film, and either in the interior of the colony or on the border (border cells exhibited a slight artifact in size measurement due to light spread in the optics; both interior and exterior populations showed a similar old pole effect). Supporting Information Dataset S1 Table of Data for the Average Lineage Shown in Figure 2 The lineage is represented as a series of letters, where “O” indicates a division as an old pole cell, and “N” a division as a new pole cell. The first cell in the tree is labeled “–.” The number of cells indicates how many individual cell growth rates comprise the average rate. At early points in the lineage, there are many more than 94 cells (the number of films), as each initial cell gives rise to two lineages (as it is not possible to assign old/new pole status to the initial cell), and some films start with more than one cell. The average growth rates approximately correspond to microns per minute. (52 KB XLS). Click here for additional data file. Video S1 Film of Growing Microcolony This film shows 305 min (114 frames) of the growth of a microcolony condensed to 7 s. For the first 40 frames (approximately 3 s), images were taken every 4 min; for the remaining frames, images were taken every 2 min. The complete lineage history of the entire microcolony from the single initial cell in frame 1 to all 505 cells in frame 114 has been tracked and recorded, allowing pole ages to be assigned to every cell. (286 KB WMV). Click here for additional data file.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC546039.xml
517510	Estimating age conditional probability of developing disease from surveillance data	Fay, Pfeiffer, Cronin, Le, and Feuer ( Statistics in Medicine 2003; 22; 1837–1848) developed a formula to calculate the age-conditional probability of developing a disease for the first time (ACPDvD) for a hypothetical cohort. The novelty of the formula of Fay et al (2003) is that one need not know the rates of first incidence of disease per person-years alive and disease-free , but may input the rates of first incidence per person-years alive only. Similarly the formula uses rates of death from disease and death from other causes per person-years alive. The rates per person-years alive are much easier to estimate than per person-years alive and disease-free. Fay et al (2003) used simple piecewise constant models for all three rate functions which have constant rates within each age group. In this paper, we detail a method for estimating rate functions which does not have jumps at the beginning of age groupings, and need not be constant within age groupings. We call this method the mid-age group joinpoint (MAJ) model for the rates. The drawback of the MAJ model is that numerical integration must be used to estimate the resulting ACPDvD. To increase computational speed, we offer a piecewise approximation to the MAJ model, which we call the piecewise mid-age group joinpoint (PMAJ) model. The PMAJ model for the rates input into the formula for ACPDvD described in Fay et al (2003) is the current method used in the freely available DevCan software made available by the National Cancer Institute.	Background Fay, Pfeiffer, Cronin, Le, and Feuer [ 1 ] showed how to calculate the age-conditional probabilities of developing a disease (ACPDvD) from registry data. Throughout this paper we use "cancer" as our disease of interest, but the method applies to specific types of cancer as well as other diseases where information is collected by population based surveillance methods. Fay et al [ 1 ] provided a formula (see equation 1 below) to calculate ACPDvD after inputing the rate function by age of (1) first incidence of cancer per person-years alive, (2) death from cancer per person-years alive, and (3) death from other causes per person-years alive. Fay et al [ 1 ] used a simple piecewise constant model for the three rate functions, which have constant rates within each age group. Here we detail two more complicated models for the rates. The first model is a segmented regression model or joinpoint model for the rates, where the rate function is a series of linear functions that join at the mid-points of the age groups, and the rate function is constant before the first mid-point and after the last "mid-point" (because the last interval goes to infinity, the last "mid-point" is not really a mid-point at all, see below). We will call this model the MAJ (mid-age group joinpoint) model for the rates. In Figure 1 we show how both the piecewise constant model and the mid-age group joinpoint model apply to all invasive cancer incidence from the Surveillance Epidemiology and End Results (SEER) program of the U.S. National Cancer Institute in 1998–2000. Figure 1 uses the SEER 12 registries which cover about 14 percent of the U.S. population, covering 5 states (Connecticut, Hawaii, Iowa, New Mexico, Utah), 6 metropolitan areas (Atlanta, Detroit, Los Angeles, San Francisco-Oakland, San Jose-Monterey, Seattle-Puget Sound) and the Alaska Native Registry (see [ 2 ]). Similar graphs showing the MAJ model can be made for the other rates required in the calculations, death from cancer and death from other causes per person-years alive. Figure 1 SEER 12 all invasive cancer incidence rates, 1998–2000, all races, both sexes: Piecewise constant and mid-age joinpoint methods. Notice that the MAJ model gives a more smoothly changing and probably a better modeled rate. The only place where the MAJ model may not perform better than the piecewise constant model is at peaks or valleys, where there may be some bias. In Figure 1 we see that the smoothness of the MAJ appears to produce more plausible estimates for ages 0 through 85 and from ages 90 and above, and the only age group with a noteworthy bias problem is 85 to 90. Thus, for almost all of the age range the MAJ model is more plausible. A problem with the mid-age group joinpoint model is that it requires numeric integration for its calculation. The second model uses a series of piecewise constant values to approximate the mid-age group joinpoint model. We call this second model the PMAJ (piecewise mid-age group joinpoint) model. The PMAJ does not require numeric integration, so it is much faster than the MAJ model. The PMAJ model is a piecewise constant model that only differs from the piecewise constant model of Fay et al [ 1 ] in that the pieces are smaller and the corresponding values of the rates are motivated by the MAJ model. Starting with version 5.0, the freely available DevCan software [ 3 ] uses the PMAJ method. (There was a small calculation error in versions 5.0 and 5.1 that has been corrected in version 5.2). DevCan calculates ACPDvD or age conditional probability of dying from a disease for U.S. cancer data or for user supplied data. The outline of this paper is as follows. The review and overview section reviews the issues in estimating the age conditional probability of developing disease from surveillance data. This section includes a motivation for using this type of statistic to describe population data. The review and overview section additionally gives graphical descriptions of the MAJ and PMAJ methods. The paper is structured so that readers not interested in the details may skip the next two sections and the appendix, which give precise and notationally involved definitions of the MAJ estimators. The examples and discussion section gives examples of the estimator of ACPDvD using three different methods for estimating the rates, the simple piecewise constant method proposed in Fay et al [ 1 ], the MAJ method, and the PMAJ method. In supplimental material [see Additional file 1 ] we compare the PMAJ method with the method of Wun, et al [ 4 ], since the latter method was the method used in versions of the DevCan software before version 5.0. Review and overview Consider a surveillance program like the SEER program of the U.S. National Cancer Institute. This program attempts to count every incidence of cancer within the catchment area of the program. Because cancer is a disease in which the rates of the disease are highly dependent on age, in order to give interpretability to the counts within the SEER registries, we must somehow account for the age distribution in the popoulation. One simple and popular statistic is the age adjusted rate or directly standardized rate (DSR). In the SEER Cancer Statistics Review [ 2 ] DSRs are used to compare different cancer sites, trends on specific cancer sites over time, and rates by sex and race. The DSR is calculated by a simple weighted sum of the age specific rates for each 5 year age group, where the weights are proportional to the U.S. 2000 population. Thus, the DSR may be interpreted as the rates adjusted as if all the populations being compared had age distributions similar to the U.S. 2000 population. The DSRs are useful for gaining an overall picture of how the incidence and mortality of each cancer effects different populations (e.g., different races, SEER population at different times), while controling for the effect of differing age distributions between populations being compared. A disadvantage of the DSR is that it is hard to relate to an individual's risk. For example, Table I-4 of the SEER Cancer Statistics Review, 1975–2000 [ 2 ] states that the DSR for breast cancer for females for the years 1996–2000 is 135 per 100,000 person-years. The average American woman may wonder, how does that relate to my risk? Will I be likely to get breast cancer in my lifetime? If I am 40 years old now, what is my risk of getting breast cancer in the next 10 years given that I have survived to this old without getting it? These questions are the motivation for using the age conditional probability of developing disease (ACPDvD), and in order to estimate the ACPDvD for female breast cancer, we require information not only about the rate of female breast cancer but also about the rates of dying from female breast cancer and dying from other causes. The ACPDvD uses cross-sectional incidence and mortality rates to estimate the age-conditional probabilities of developing disease in a hypothetical cohort in which we assume the age specific rates do not change over time. This gives a personal interpretation to the cross-sectional data, allowing statements like the following: if the incidence and mortality rates remain at their present values (as observed in SEER 12, 1998–2000), then a female born today would have a 13.5% chance of developing breast cancer over her lifetime (see Table 2 ). We can also calculate ACPDvD over intervals. For example, a female who has reached 40 years old without developing breast cancer has a 1.5% chance of developing breast cancer by the time she is 50. Table 2 Age Conditional Probability of Developing Different Types of Invasive Cancers (in Percent) from SEER 12, 1998–2000 Start Age End Age Model All Invasive (Both Sexes) Prostat(Male) Breast (Female) Acute Lymphocytic Leukemia (Both Sexes) 0 20 Piecewise const 0.3158 0.0009 0.0015 0.0669 PMAJ, interval = .5 0.3260 0.0011 0.0021 0.0633 MAJ 0.3260 0.0011 0.0021 0.0633 0 50 Piecewise const 4.0690 0.2002 1.9188 0.0837 PMAJ, interval = .5 4.1657 0.2550 1.9492 0.0808 MAJ 4.1657 0.2550 1.9492 0.0808 40 50 Piecewise const 2.5260 0.2032 1.5131 0.0053 PMAJ, interval = .5 2.5976 0.2579 1.5169 0.0055 MAJ 2.5975 0.2579 1.5169 0.0055 0 Inf Piecewise const 42.0876 17.4952 13.6471 0.1154 PMAJ, interval = .5 41.7547 17.3375 13.5477 0.1121 MAJ 41.7574 17.3389 13.5485 0.1121 60 61 Piecewise const 1.2340 0.5989 0.3822 0.0009 PMAJ, interval = .5 1.0852 0.4946 0.3627 0.0009 MAJ 1.0852 0.4946 0.3627 0.0009 64 65 Piecewise const 1.2758 0.6131 0.3872 0.0009 PMAJ, interval = .5 1.4453 0.7440 0.4045 0.0010 MAJ 1.4453 0.7440 0.4045 0.0010 60 65 Piecewise const 6.0331 2.9128 1.8777 0.0042 PMAJ, interval = .5 6.0622 2.9492 1.8758 0.0044 MAJ 6.0622 2.9492 1.8759 0.0044 Calculation of the ACPDvD is somewhat complicated, and we describe the complications in relation to the simple DSRs. Consider first the age specific incidence rates which are used to calculate the DSRs. These rates simply count the number of incident cases of a particular disease (e.g., female breast cancer) within each age group and divide by the total number of person-years estimated by the population. For counts of a single year, the person-years are estimated by the mid-year population of the catchment area (for sex-specific cancers like prostate cancer or female breast cancer, we only use the population of the appropriate sex). Note that the incident cases may include individuals who have previously been diagnosed with the cancer and have developed a new primary cancer. For the ACPDvD for any specific disease we would like the rate of first incidence per person-years alive and disease-free. Thus, there are two difficulties, (1) the usual age specific incidence rates include persons with multiple primary cancers, and (2) the denominators include persons who have previously been diagnosed. Merrill and Feuer [ 5 ] discuss both difficulties and adjust for them creating risk-adjusted cancer incidence rates. Merrill and Feuer [ 5 ] study the effect of these adjustments for several cancer sites. To handle the first difficulty, (similar to [ 5 ]) we can remove cases where we have a record of a previous diagnosis of that particular type of cancer. Because the registries in SEER were not all begun at the same time, to avoid bias the DevCan program only searches the records for previous cancers back until the year when the last registry was added. This year is denoted the follow-back year. (If the disease of interest is any malignant cancer, then the difficulty is handled differently. Although at each cancer record we do not record what specific types of cancers were previously diagnosed for the person, we do know whether any tumors were previously diagnosed. Thus, if the disease of interest is any malignant cancer and if the record states there was a previously diagnosed tumor, then we assume that the previously diagnosed tumor was malignant, and do not count that case as a first incidence.) To handle the second difficulty, the additional person-years in the denominator, Merrill and Feuer [ 5 ] adjust the denominator by multiplying the age-specific population by 1 minus an estimate of the prevalence of the disease in the population. Merrill and Feuer [ 5 ] also estimate the prevalence of medical procedures which remove individuals from the at-risk population, such as hysterectomy which removes the risk of uterine cancers. In calculating the ACPDvD we use only first incident of the disease of interest as in [ 5 ], but we correct for the denominators in a different way using an assumption and some mathematics from the theory of competing risks. This second correction is detailed with precise mathematical notation in Fay et al [ 1 ]; here we give more heuristic arguments. In the following let the disease of interest be "cancer". The ACPDvD between ages x and y , given alive and cancer-free at age x , may be written as the fraction, To calculate the numerator, we integrate over the probability that the first cancer occurred at exactly age a. In math notation this probability is where f c ( a ) is a probability function representing the probability that the first cancer occurred at exactly age a. One key result described in Fay et al [ 1 ] is that f c ( a ) can be written as the product of two functions, λ c ( a ) = the probability that the first cancer occurred at exactly age a, given the individual is alive just before age a , and S a ( a -) = the probability that the individual is alive just before age a . The function λ c ( a ) is known as a cause-specific hazard function, and it is estimated by some function of the age-specific rates, such as the piecewise constant model of Fay et al [ 1 ] or the MAJ model introduced in this paper (see Figure 1 ). Using standard results for continuous survival data, we can write S a ( a -) as where λ a ( u ) ( = the probability that the individual died at age u , given the individual is alive just before age u ) is the usual hazard function. We estimate λ a ( u ) using some function of the age-specific rates. Thus, the numerator can be written as If we use the MAJ for both hazard functions, then there is no closed form solution. To see why this is so, note that within the exponential, the integral of a piecewise linear function is the sum of a series of quadratic functions, and the overall integral has no closed form solution. This problem motivates the piecewise mid-age joinpoint (PMAJ) model, where we use a series of piecewise constant functions to approximate the MAJ model. Figure 2 gives the PMAJ model together with the piecewise constant model used by Fay et al [ 1 ] for 70 to 90 year olds from the SEER 12, 1998–2000 rates for all invasive (first) cancer incidence rates per person-years alive. Remember, although both Figure 1 and Figure 2 plot incidence rates, we additionally need similar rate functions for mortality rates to calculate the ACPDvD. Figure 2 SEER 12 all invasive cancer incidence rates, 1998–2000, all races, both sexes: Piecewise constant and PMAJ methods. Now consider the denominator of the ACPDvD, the probability of being alive and cancer-free at age x , denoted . For reference, in Table 1 we give the notation. The only change from the notation in Fay et al [ 1 ] is that we use the subscript a to represent all causes of events instead of a blank subscript. For example, we let S* ( u ) = . Other notation in this paper is defined as it is introduced. Fay et al [ 1 ] assumed that the risk of death from other causes does not change if you have previously been diagnosed with cancer, then used the key result mentioned above together with some algebra and calculus to derive the denominator. Then the ACPDvD between the ages of x and y given alive and cancer-free just before age x is Table 1 Notation Random Variables and Parameters T = age at death T * = age at first cancer or death before cancer J = type of death J * = type of event ( J = d ) = death from cancer ( J * = c ) = first cancer ( J = o ) = death from other causes ( J * = o ) = death before first cancer λ c ( t ) = rate at t for first cancer given alive = rate at t for first cancer given alive and cancer-free λ o ( t ) = rate at t for death before cancer given alive = rate at t for death before cancer given alive and cancer-free λ d ( t ) = rate at t for death from cancer given alive λ a ( t ) = rate at t for death given alive = rate at t for first cancer or death before first cancer given alive and cancer-free Observations Within the age interval, [ a i , a i +1), and within the calendar interval of interest we observe... c i = number of first cancer incident cases = estimate of person-years alive associated with j = c , d , o (DevCan uses the sum of mid-year populations during the calendar interval of interest) d i = number of cancer deaths o i = number of other deaths The details of the MAJ and the PMAJ models are given in the next two sections. Readers only interested in the practical ramifications of the choice in models may skip to the examples and discussion section. Mid-age group joinpoint estimator In Fay et al [ 1 ], the rates were estimated by a piecewise constant model. Here we use a mid-age group joinpoint (MAJ) model, where we draw lines connecting the midpoints of the intervals except the first and last interval. The first interval is constant until the midpoint, and the last interval is constant after a nominal "midpoint". This nominal "midpoint" is half the length of the previous age interval from the beginning of the last interval, and would be the midpoint if the last age interval was the same length as the previous interval. We introduce new notation for breaking up the ages. Fay et al [ 1 ] used 0 = a 0 < a 1 < ··· < a k < a k +1 = ∞. Here we use a joinpoint model with joins at the midpoints (and nominal midpoint), Let (The indices start at -1 so that the index values for the rate estimators, , match up with the count notation of [ 1 ].) The MAJ estimator for the rate of event j (for j = c, d, or o ) at t i (for i = 0,1,..., k) is where j i is either c i , d i , or o i as defined in Table 1 . (Note that , where is the piecewise constant function used in [ 1 ]). We define and . For j = a , MAJ estimator for the rate at t i is Then for t ∈ [ t i , t i +1 ) for i = 1,..., k , we define as the point on the line defined by connecting the points ( t i , ) and ( t i +1 , ). In other words, Where and Thus, α j ,-1 = and β j ,-1 = 0, and similarly by taking limits as t k +1 → ∞ then α j , k = and β j , k = 0. Now for u ∈ [ t i , t i +1 ) is Note that (for ℓ = 0,1,..., k ) so that for i = 0,1,...,k, Also notice that (when u < ∞) Therefore when u ∈ [ t i , t i +1 ), Let ( x, y ) be the estimator of A ( x, y ) using the MAJ model. The two integrals we need to estimate for ( x, y ) are of the type, where in the numerator of ( x, y ) we need (i.e., j = c and h = a in equation 7), and in the denominator of ( x, y ) we need . Suppose, without loss of generality, that t ∈ [ t i ,t i +1 ), then where R j , h ( t ℓ , v ) (for ℓ = - 1,0,1,2,..., i and v ≤ t ℓ+1 ) is defined implicitly (see the Appendix). Then, Piecewise mid-age group joinpoint estimator In the MAJ model we divided up the age line into k + 2 intervals. Here we define those intervals in both the t i notation and the a i notation. In the MAJ model the rates for the first and the last intervals are represented by lines with zero slope, and the rates for the i th interval ( i = 1,..., k ) for the j th rate type ( j = a , c , d , o ) is a line defined by connecting the points ( t i -1 , ) and ( t i , ) (see equations 2 and 3 for definition of ). In the PMAJ model we divide the i th interval into m i equal sized intervals, and use a piecewise constant estimate on each of those m i intervals. One way to define m i is to chose m i so that each equal sized interval is 1/2 year long. In other words, m i = 2( t i - t i -1 ). This is the definition of m i that we use for the DevCan software (starting with version 5.0, see [ 3 ]), but all the following holds for arbitrary m i . In Figure 2 we show the PMAJ model with half-year intervals and the piecewise constant model for the US all invasive cancer mortality rates for ages 70 through 90 years. Here are the details. Consider the h th (for h = 1,..., m i ) of the m i intervals within interval i (for i = 1,..., k ) for rate type j (for j = a , c , d , o ). This interval is For convenience we introduce new notation for the ends of this interval, let so that t i -1,0 = t i -1 and = t i . At the beginning of this interval the value of the rate is (see equations 4 and 6 for definitions of α j , i -1 and β j , i -1 ). Similarly at the end of this interval the rate is For the PMAJ model we simply assume a constant rate equal to the average of the beginning and the end values of the rate over this interval. In other words, under the PMAJ model for any t ∈ [ t i -1, h -1 ,t i -1, h ) we estimate the rate with Since the PMAJ model is a piecewise model, we can use Appendix A of [ 1 ] to express the estimator of age conditional probability of developing cancer. The only hard part is correctly defining the starting and ending of each piecewise interval. The ends of these intervals are For convenience write these interval ends with only a single index as where and m 0 = 1. In other words, t -1 = τ 0 and for i = 0,1,..., k , then t i = τ g ( i ) and t i , h = τ g ( i )+ h , where . Now we can follow very similar notation to Appendix A of [ 1 ]. We now repeat that Appendix with the modifications to notation required for the PMAJ model. Let the estimator of A ( x,y ) under the PMAJ model be denoted ( x,y ). Let τ i ≤ x < τ i +1 and τ j < y ≤ τ j +1 for x < y,i ≤ j , and j ≤ M + 2. For convenience we regroup the ages after inserting group delimiters at x and y . Let the new delimiters be 0 = b 0 ≤ b 1 ≤ b 2 ≤ ··· ≤ b M +3 = ∞ where b 0 = τ 0 ,..., b i = τ i , b i +1 = x, b i +2 = τ i +1 ,..., b j +1 = τ j , b j +2 = y, b j +3 = τ j +1 ,..., b M +3 = τ M +1 = ∞. We let and similarly and . In this notation, the probability of developing cancer by age y given survival until age x is A ( x, y ) = A ( b i +1 , b j +2 ), and under the PMAJ model we estimate it with Because or may equal zero and b ℓ+1 may equal infinity, we let . These integrals are where the case λ = 0 and b ℓ+1 = ∞ is one of the "impossible" hypothetical cohorts (see Section 3.1 of [ 1 ]). Thus, we obtain, Examples and discussion In this section we explore several different methods for estimating the rate functions, all using the formula of Fay et al [ 1 ] (e.g., all using equation 1). This comparison explores the differences between the piecewise constant method proposed in Fay et al [ 1 ], the PMAJ method, and the MAJ method. A different comparison emphasizing differences between versions of the DevCan software is described in the supplemental material [see Additional file 1 ]. For all of the examples we use data from 1998–2000 [ 6 ]. The incidence data come from the Surveillance, Epidemiology, and End Results (SEER) program of the (U.S.) National Cancer Institute, and mortality data from the (U.S.) National Center for Health Statistics. We use the SEER 12 registries which cover about 14 percent of the U.S. population. We only use the mortality data covering the same area as the SEER 12 registries cover. Because the SEER 12 registries have complete coverage only back through 1992, we only look back in the database until 1992 to delete any incident case that had previously been diagnosed with the cancer of interest. These incident cases are deleted so that they are not counted when estimating the counts of first cancer incidence (the c i values). The mid-year population estimates (the n i values) come from the sum U.S. Census estimates of mid-year populations from 1998, 1999, and 2000 for the SEER 12 catchment areas for the appropriate sex group (e.g., males for prostate cancer). In Table 2 we show the results for all invasive cancers and acute lymphocytic leukemia for both sexes, prostate cancer for males, and breast cancer for females. We see the PMAJ values approximate the MAJ values very well. In conclusion, we have described several methods for estimating rates for input into a formula to calculate ACPDvD, and we have shown that the PMAJ method provides fast and reasonable estimators for the rates. Appendix: Calculation of R function Recall that R j,h ( t ℓ , v ) represents an integral with 4 parameters. We can write it as To simplify notation substitute let t ℓ = u and α j ℓ = α j ,β j ℓ = b j , α h ℓ = a h , and β h ℓ = b h . Thus, Case 1: b j = 0 and b h = 0 For our application, whenever v → ∞ then b j = 0 and b h = 0, so this is an important special case. When b j = 0 and b h = 0 and a h = 0 and we obtain which goes to ∞ when v → ∞. When b j = 0 and b h = 0 and a h ≠ 0 and we obtain which goes to a j / a h when v → ∞. Case 2: General Case with v < ∞ To calculate the integral, R ( u, v, a j , b j , a h , b h ) for finite v , we can use an adaptive use of Romberg's algorithm for numeric integration (we follow closely Lange [ 7 ], pp. 210–211). Let Divide the interval [ u, v ] into n equal subintervals of length ( v - u )/ n , and let Then lim n →∞ T n = R ( u, v, a j , b j , a h , b h ). A more accurate approximation uses Romberg's algorithm, Let be our estimate of R . The algorithm we use to calculate is as follows: 1. Choose n . 2. Calculate T n . 3. Calculate T 2 n . 4. For i = 1 to I max do: • If then let and stop. • Otherwise calculate , and continue. For example, one could use n = 100 and δ = 10 -5 and I max = 100. Supplementary Material Additional File 1 Comparing the method of Wun, Merrill, and Feuer (1998) to the PMAJ method. We calculate lifetime risks of developing certain cancers for different race and sex combinations. For each lifetime risk we give the old method of Wun, Merrill, and Feuer [4], the PMAJ method, and the percent difference. In general, the two methods agree to within about 2 percent. Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517510.xml
516168	Subcellular distribution of the V-ATPase complex in plant cells, and in vivo localisation of the 100 kDa subunit VHA-a within the complex	Background Vacuolar H + -ATPases are large protein complexes of more than 700 kDa that acidify endomembrane compartments and are part of the secretory system of eukaryotic cells. They are built from 14 different (VHA)-subunits. The paper addresses the question of sub-cellular localisation and subunit composition of plant V-ATPase in vivo and in vitro mainly by using colocalization and fluorescence resonance energy transfer techniques (FRET). Focus is placed on the examination and function of the 95 kDa membrane spanning subunit VHA-a. Showing similarities to the already described Vph1 and Stv1 vacuolar ATPase subunits from yeast, VHA-a revealed a bipartite structure with (i) a less conserved cytoplasmically orientated N-terminus and (ii) a membrane-spanning C-terminus with a higher extent of conservation including all amino acids shown to be essential for proton translocation in the yeast. On the basis of sequence data VHA-a appears to be an essential structural and functional element of V-ATPase, although previously a sole function in assembly has been proposed. Results To elucidate the presence and function of VHA-a in the plant complex, three approaches were undertaken: (i) co-immunoprecipitation with antibodies directed to epitopes in the N- and C-terminal part of VHA-a, respectively, (ii) immunocytochemistry approach including co-localisation studies with known plant endomembrane markers, and (iii) in vivo -FRET between subunits fused to variants of green fluorescence protein (CFP, YFP) in transfected cells. Conclusions All three sets of results show that V-ATPase contains VHA-a protein that interacts in a specific manner with other subunits. The genomes of plants encode three genes of the 95 kDa subunit (VHA-a) of the vacuolar type H + -ATPase. Immuno-localisation of VHA-a shows that the recognized subunit is exclusively located on the endoplasmic reticulum. This result is in agreement with the hypothesis that the different isoforms of VHA-a may localize on distinct endomembrane compartments, as it was shown for its yeast counterpart Vph1.	Background Vacuolar H + -ATPases are large multi-heteromeric protein complexes located at endomembranes of all eukaryotic cells. Plant V-ATPase has been identified at the vacuolar and various other endomembranes, and also at the plasma membrane [ 1 - 3 ]. The total molecular mass of V-ATPase is estimated to surpass 700 kDa. V-ATPase pumps protons into membrane-surrounded intracellular compartments at the expense of hydrolysis energy of ATP [ 4 ]. V-ATPases share a common structure composed of a ball-like head, a membrane-intrinsic part and connecting stalks similar to ATP-producing F-ATP synthases. Biochemical examinations of the subunit composition revealed that V-ATPases are built from up to 14 subunits. The protein subunits are denominated VHA-A through H for soluble components protruding into the cytoplasm (V 1 -part), and VHA-a, c, c', c", d and e for membrane-associated subunits (V 0 -part) [ 5 ]. In plants, the molecular characterisation of VHA-subunits is only beginning to include also the genes that were identified recently [ 6 , 7 ]. Following the cloning of cDNA sequences coding for VHA-A, -B, -c, and -E until 1995, VHA-D, -F, -C and -G were cloned from A. thaliana , oat and barley [ 5 , 7 ]. However, the first complete set of subunit sequences only became available with the sequencing of the genome of A. thaliana [ 9 ]. The second set was cloned from the halotolerant, facultative CAM plant Mesembryanthemum crystallinum [ 10 ], and a third set is now available from rice. A comparative analysis of the genes/ESTs revealed that Arabidopsis and Mesembryanthemum express a similar set of subunits [ 7 ]. A detailed analysis of the sequences from both species showed that all examined subunits share the same properties in binding of ATP and conducting protons through the proteolipid ring [ 10 ]. An important question concerns the composition of the plant V 0 -sector since its subunit composition is not resolved yet. VHA-a is the subunit with the highest molecular mass within the V-ATPase and reveals a bipartite structure. As first shown in yeast, VHA-a consists of a 50 kDa hydrophilic N-terminal domain and a hydrophobic, membrane-spanning C-terminus [ 11 - 13 ]. Interestingly, Li and Sze [ 14 ] could not observe the VHA-a subunit in functional V-ATPase of oat. Therefore, they suggested a role of VHA-a in assembly of plant V-ATPase. The conclusion is in contrast to results in yeast where site directed mutagenesis had allowed to identify amino acids involved in the mechanism of proton translocation across the membrane. Mutant yeast cells devoid of VHA-a or complemented with modified VHA-a variants exhibited the conditional phenotype of V-ATPase deficiency [ 15 ]. Another novel subunit in plants that was only addressed recently is VHA-H. VMA13p being the yeast orthologue of VHA-H was cloned and represents the only crystallized subunit of V-ATPase at present [ 16 ]. VHA-H is considered to activate and regulate V-ATPase by functionally coupling ATP hydrolysis to proton flow through the V o -domain [ 17 ]. The aim of the study was to improve the understanding of V-ATPase distribution in plant cells with emphasis on the localization of VHA-a and VHA-H in the V-ATPase structure. Three specific questions were answered using different methodology: (i) Is VHA-a part of the V-ATPase structure? (ii) Where are different subunits localized within plant cells? (iii) Is FRET a suitable method to investigate subunit interaction within the complex, for example between VHA-a and VHA-c, and VHA-H and VHA-B. Results Primary structure of VHA-a From the database dbEST (NCBI, USA) a cDNA fragment from M. crystallinum was identified (AI822404) with similarity to known genes coding for VHA-a in yeast. Using RACE-PCR a full length cDNA was obtained with a size of 2783 bp. Its open reading frame encoded a hypothetical protein of 93.1 kDa. Database search with the program FASTA revealed the highest similarities to the entries At2g21410 (77%), At4g39080 (76 %) and At2g28520 (60 %) from the A. thaliana genomic database and also significant similarity on the amino acid level to the genes Stv1 (34 %) and Vph1 (38 %) from S. cerevisiae . These two latter genes are coding for isoforms of the yeast 100 kDa subunit whereas all other subunits of the yeast vacuolar ATPase are encoded by one gene each. The sequence alignment between the newly cloned amino acid sequence from M. crystallinum , the three isogenes from A. thaliana and the two isogenes from yeast (Fig. 1 ) indicates a structure with two distinct segments distinguished by their degree of similarity: the N-terminus with a low degree of sequence conservation (amino acid 1 to 400) and the C-terminus (amino acid 401 to 816) with a higher degree of conservation between the sequences. This corresponds to the above mentioned domain structure of S. cerevisiae VHA-a with a cytoplasmic and a membrane-integrated domain. The heterogeneity in the N-terminus is characterised through a high number of deletions and insertions most remarkably between the amino acid positions Gly84 to Ile85, and Gly141 to Gln142. In contrast to all other VHA-sequences, in yeast-Stv1 this region contains additional sequence insertions of approximately 20 amino acids. Sequence variation between VHA-a from distinct species and paralogues within same species continues until Glu193. Other domains of the N-terminus vary to a lesser extent. The heterogeneity decreases in direction to the membrane-spanning C-terminus. Figure 1 Amino acid sequence alignment of plant and yeast VHA-a. Comparison of amino acid sequences deduced from the three coding regions detected in the A. thaliana genome, the sequence of M. crystallinum and the two yeast gene products Vph1 and Stv1 using ClustalW-algorithm. Identical amino acids are marked in dark grey, those similar with M. crystallinum in light grey. The programs JPRED (EMBL, Hinxton) and THMM (EBS, Denmark) were used to predict transmembrane domains (labelled with +). Amino acids identified as essential for H + -pumping in yeast are marked with *. The N-terminal amino acids of VHA-a that are conserved in all species are characterised by an above average portion of amino acids with an acid or aromatic character. A comparison of the first 400 amino acids of VHA-a from 10 different species showed the presence of 34 amino acids conserved throughout all species, 10 of which have acidic and 8 aromatic residues. In all sequences compared the representation of aromatic amino acids is higher than the average. In yeast it was shown that sequence motifs with aromatic amino acid residues are involved in the targeting of proteins [ 18 ]. An assignment of VHA-a sequences from A. thaliana and M. crystallinum to the distinct yeast isoforms Vph1 and Stv1, in order to define orthologous genes, was not possible on basis of the amino acid sequence alignment. The comparison of the C-termini revealed a higher degree of sequence conservation interrupted through various short insertions or deletions. The sequence was analysed for secondary structures (Predict Protein, EMBL, Heidelberg) and membrane topology (THHM, Denmark). The prediction correlated regions of high sequence conservation with the localisation of putative transmembrane domains (Fig. 1 , marked with +). Furthermore, these hypothetical transmembrane domains are in accordance with the membrane-topology suggested by Leng et al. [ 19 ] for the yeast Vph1 protein. Mutations in single amino acids of Vph1 have previously allowed the identification of 5 charged amino acids in the membrane spanning helices of the C-terminus whose mutation strongly affected (Lys734, His743, Glu789 and Arg799) or fully inhibited (Arg735) transmembrane H + -transport, although they did not affect the assembly of the vacuolar ATPase [ 19 - 23 ]. These functional amino acids are also conserved in Mc-VHA-a. Detection of VHA-a in membrane preparations in vitro For further examination of McVHA-a, antisera were raised against two polypeptide fragments, i.e. a N-terminal 42 kDa polypeptide and a 13 kDa domain located between predicted transmembrane helices 3 and 4. Both were expressed heterologously in E. coli using the primer combination a-nterm-f and a-nterm-r, and a-memb-f and a-memb-r, respectively. The derived PCR-products were cloned into the expression vector pCRT7-NT (Invitrogen, Holland), introduced and expressed in E. coli . The 42 kDa and 13 kDa polypeptides were purified to homogeneity by chromatography on Ni-NTA and by preparative SDS-PAGE. The 42 kDa polypeptide was recognized by a purified antibody against yeast Vph1 [ 13 ] (not shown). The antiserum produced in guinea pig reacted with the heterologously expressed protein a N-term (not shown). The presence of VHA-a was investigated in plant endomembranes and soluble fractions rapidly prepared from 5 week old M. crystallinum plants. In the soluble and the membrane-fraction, the serum against McVHA-a N-term identified two polypeptides with apparent molecular masses between 65 and 70 kDa (Fig. 2 ). After increasing the NaCl concentration to 500 mM, the apparently full size 95 kDa band was detected in the membrane fraction corresponding to the expected size of untruncated McVHA-a (Fig. 2 ). To test whether the addition of NaCl to the membrane preparation affected the purification efficiency of other V-ATPase subunits from the V1-sector, membranes and soluble fractions purified with buffer containing either 100 or 500 mM NaCl were also reacted with antibodies against VHA-E [ 24 ] and anti VHA-D i recognizing VHA-B and D i [ 25 ]. The latter authors demonstrated that this antibody marked both VHA-A and -B. The increased NaCl concentration had no effect on the labelling strength of VHA-E and VHA-A/B (not shown). Interestingly, also McVHA-A, -B and -E were detected both in the membrane and soluble fractions. Figure 2 Immunodetection of VHA-a in membrane and soluble fractions of M. crystallinum leaves. Membranes were isolated in the presence of 100 (1, 2), 500 mM NaCl (3, 4), and without (5) or with (6) protease inhibitor complete ® , respectively. Membranes and soluble fraction were separated rapidly for (1)–(4) whereas the standard membrane isolation procedure with ultracentrifugation was undertaken for (5) and (6). The 95 kDa band is seen as the dominant protein in the membrane fraction isolated at high salt and to some part in the isolate obtained in the presence of protease inhibitor. To examine the sensitivity of VHA-a against proteases, membranes were isolated in second approach, in the absence and presence of a protease inhibitor cocktail (complete ® , Roche, Mannheim, Germany). Without protection from proteolysis, a band at about 50 kDa and a doublet band above 60 kDa were predominant. In the sample with protease inhibitor, the 95 kDa band appeared although as one of five bands of similar intensity if decreasing transfer efficiency of high molecular mass polypeptides from the gel to the membrane is assumed. Apparently, VHA-a is prone to degradation. However, the results also indicate that VHA-a is part of V-ATPase. To further prove that tentative conclusion, immunoprecipitation was performed using anti VHA-a N-term and anti VHA-a memb followed by Western blot analysis with anti VHA-E and anti-VHA-A (Fig. 3 ). Each control, i.e. immunoprecipitation without serum and with preimmune serum, gave the expected results of no response (Lanes 1 and 2 in Fig. 3A,3B,3C ). With serum, the 55 kDa band of VHA-A and the dimer of VHA-E was seen. Apparently, immunoprecipitation with antibody directed against the N-terminus as well as the membrane part of VHA-a pulled down a complex also containing subunits of the V 1 section, and thus probably the holocomplex. Figure 3 Co-immunoprecipitation of VHA-E and VHA-A with VHA-a. Tonoplast enriched membranes of M. crystallinum were solubilized in buffer supplemented with 2% (v/v) Triton X-100. Antibody directed against VHA-a, either the N-terminal or membrane part, was added and immunoprecipitation was performed. The pellet samples and part of the supernatant (10–15% of total) were loaded on a SDS-gel, and Western blot was performed with anti VHA-E or A. The band intensities related to loaded sample size indicate that only a fraction of total V-ATPase was immuno-precipitated with the anti-VHA-a antibodies. As controls, immunoprecipitation was performed without added serum and with preimmune serum. With anti-E, monomeric and dimeric band of 30 and 60 kDa was detected by Western blotting in the precipitate obtained with anti VHA-a N-term , with anti VHA-A, a 65 kDa band was labelled in separations of both, the precipitates obtained with anti VHA-a N-term and anti-VHA-a Memb , respectively. Immunochemical analysis of VHA-distribution A more detailed analysis of the distribution of the VHA-subunits in the plant cell was performed by immuno-labelling of maize root tip cells. To highlight the distribution of the different VHA-subunits in the cell, squashed maize root cells were incubated with anti VHA-A, anti VHA-E or anti VHA-a N-term . Fig 4 shows the staining patterns of the antibodies in young cells devoid of large vacuoles (Fig. 4A,4D,4G ), cells with beginning vacuolization (Fig. 4B,4E,4H ) and older cells with many vacuoles of 2–5 μm diameter (Fig. 4C,4F,4I ). In cells devoid of large vacuoles, anti VHA-E as well as anti-VHA-A marked punctuated compartments whereas in cells with developed vacuoles a nearly complete staining of the tonoplast and an unsteady staining of other small cellular compartments was observed. In a converse manner, anti-VHA-a N-term did not stain the tonoplast of cells of various vacuolisation state. The staining with anti-VHA-a N-term revealed a distinct reticulate pattern and a staining of the nuclear membrane reminding of ER-labelling. Figure 4 Localization of VHA-subunits in plant cells, using anti-VHA-A, anti-VHA-E and anti-VHA-a Nterm -antibody Confocal images represent single images of isolated maize root cells. Immuno-staining was performed on maize root cells. Secondary fluorescently labelled antibodies used were anti-rabbit-FITC (A-F) or anti-guineapig (Cy5). Images were colour coded with Adobe Photoshop. Scalebars are 10 μm. Labelling of tonoplast membranes with anti VHA-A (A to C) and anti VHA-E (D-F) in root cells of increasing age, i.e. non-vacuolized to vacuolized (left to right). Anti-VHA-a N-term labelling of root cells of increasing age (G-I). Note that the staining pattern of VHA-a N-term is distinct from the tonoplast labelling with VHA-A and VHA-E in all cases. A double immuno-labelling-technique [ 26 ] was then employed to identify the antibody-marked intracellular compartments in detail. The selected antibodies were directed against the aquaporin γ-Tip located in the tonoplast of the lytic vacuole (anti γ-Tip; [ 27 ]) and against marker components of the endoplasmic reticulum (ER) (anti-calreticulin). The results of this immuno-staining are shown in Fig. 5 . In cells with developed vacuoles, anti-γ-Tip-labelling of the tonoplast (Fig. 5A ) co-localised completely with anti-VHA-A (Fig. 5B,5C ). Anti VHA-E labelled similar structures as anti-VHA-A and co-localised also with anti-γ-Tip on the tonoplast (not shown). A co-staining of root cells with anti-VHA-A and the ER-marker anti-calreticulin was then performed. Anti-calreticulin marks a specific ER-network including the nuclear membrane (Fig. 5D,5G,5J ). The double labelling with antibodies against VHA-A (Fig. 5E ) or VHA-E (Fig. 5H ) showed no significant co-localisation of the typical tonoplast staining with the ER-marker. When performing a co-labelling of maize root cells with anti-calreticulin (Fig. 5J ) and anti-VHA-a N-term (Fig. 5K ) revealed a complete co-labelling of the two markers (Fig. 5L ). Figure 5 Co-localization of VHA-subunits with the tonoplast marker γ-Tip and calreticulin, a marker for the ER. Confocal images represent single images of isolated maize root cells which were immuno-probed with antibodies directed against marker polypeptides of the vacuolar membrane (γ-Tip) and endoplasmatic reticulum (calreticulin) and simultaneously treated with anti-VHA-antibodies. Secondary fluorescently labelled antibodies used were anti-rabbit-FITC (A), anti rabbit-Cy3 (B, E, H), anti-mouse-FITC (D, G, J) or anti-guineapig-Cy5 (K). Images were colour coded with Adobe Photoshop. Scalebars are 10 μm. In each row, the immuno-decoration with the marker, with the VHA-subunit specific antibody and the superposition of both is shown. (A-C) Root cell labelled with γ-Tip (A) and VHA-A (B). Note the complete co-localisation of the both markers on the tonoplast of small vacuoles (C). (D-F) Double-staining with calreticulin (D) and VHA-A. (G-I) Double-labelling of a root cell with calreticulin (G) and VHA-E (H) reveals a similar result as with VHA-A. Tonoplast labelling and ER-staining are distinct. (J-L) Co-labelling with calreticulin (J) and VHA-a N-term reveals a complete co-localisation of the two signals on the ER. For high resolution, immunogold labelling with anti VHA-a N-term and anti VHA-A was performed on ultra-thin cross sections of maize root cells (Fig. 6 ). With anti VHA-a pronounced label with gold particles was detected in ER membranes (Fig. 6D ), and occasionally a weak labelling of the Golgi apparatus (Fig. 6C ). A labelling of tonoplast membranes was not found, indicating that VHA-a N-term is predominantly located on the ER. For comparison, immuno-gold analysis of ultrathin cross sections with anti-VHA-A. revealed labelling of the tonoplast (not shown) and a labelling of the Golgi apparatus significantly stronger than with anti-VHA-a (Fig. 6A ). Figure 6 Immunogold-based localisation of VHA-a. Ultra-thin cross sections of maize root cells in 1 mm distance to the tip were decorated anti VHA- (A), premmune serum, (B), anti VHA-a N-term -antibody (C,D), respectively. Sections were washed, treated with secondary antibody linked to 15 nm gold particles and visualized in an electron microscope. Fluorescence resonance energy transfer between VHA-subunits in vivo FRET allows to investigate protein-protein interaction in vitro and in vivo . Both partners have to carry fluorescent labels with overlapping emission (donor fluorophore) and excitation spectra (acceptor) and need to be situated in close proximity. Half maximum energy transfer takes place at distance of the Förster radius R 0 . Cyan and yellow fluorescent proteins constitute such a FRET pair and were fused to the C-termini of various subunits of V-ATPase. Under the assumption of freely rotating fluorophores, R 0 is close to 5 nm. The size of the V-ATPase complex is about 15 nm (diameter) × 25 nm (length from lumen side to tip of head). Arabidopsis protoplasts were co-transformed with vector constructs of VHA-a fused to YFP and VHA-c fused to CFP under the control of the 35S promotor. Upon excitation of doubly labelled protoplasts at 458 nm both, CFP and YFP showed strong fluorescence (Fig. 7A ). Fluorescence emission spectra were recorded using a double dichroic mirror which exhibits high reflectivity at ~514 nm. Therefore, the two emission maxima were separated by a minimum (Fig. 7C ). This fact renders the quantitative analysis of the FRET efficiency due to the decrease in donor and increase in acceptor fluorescence more difficult. Alternatively, the method of acceptor bleaching can be used to verify energy transfer [ 28 ] and can be seen in Fig. 7B , the fluorescence intensity of YFP decreased rapidly within 10 scan cycles upon excitation at 514 nm due to photobleaching to a residual intensity attributable to auto-fluorescence. Simultaneously, the donor fluorescence increased substantially, thus providing direct evidence for energy transfer from CFP to YFP. From the increase in CFP fluorescence upon acceptor bleaching an overall FRET efficiency of ~0.45 is estimated. Assuming freely rotating fluorophores (K 2 = 2/3) this corresponds to a distance between VHA-a, and VHA-c of ~5.4 to 7.2 nm (mean 6.3 ± 0.8 nm). Here it has to be pointed out that each ring of the rotor V 0 contains 5 VHA-c-subunits and one VHA-c"-subunit. Hence, dependent on the position of the VHA-c/CFP-subunits in the ring different distances between CFP-labelled VHA-c subunits and the YFP-labelled VHA-a subunit will result. Figure 7 FRET between VHA-subunits co-expressed in protoplasts of A. thaliana and onion epidermis cells . (A) Mesophyll protoplasts of A. thaliana were simultaneously transformed with p35S::VHA-a/YFP and p35S::VHA-c/CFP using the polyethylene glycol method. After 20 h, fluorescence emission from protoplasts was measured following excitation at 458 nm and 514 nm, and image analysis in the range of 470 – 500 nm for CFP and 560 – 585 nm for YFP fluorescence, respectively. (B) Acceptor bleaching in dependence on scan numbers. For this experiment, protoplasts expressing VHA-a-YFP were excited at 514 nm and emission was recorded between 550 and 600 nm. (C) Emission spectra of VHA-c-CFP and VHA-a-YFP before (solid line) and after (broken line) acceptor bleaching. (D) FRET between VHA-A-YFP and VHA-B-CFP after co-expression in onion epidermis cells. Excitation was achieved at 458 nm, and 2D emission images were recorded in the range of 470 to 510 nm (CFP), and 550 to 600 nm (YFP). Similar experiments were performed with onion epidermis cells (Fig. 7D ) co-transformed with VHA-A/YFP and VHA-B/CFP, and VHA-B/CFP and VHA-H/YFP, respectively. This system was employed for two reasons, (i) to confirm and extent the results from protoplasts and (ii) to work in turgescent cells, not previously subjected to a protoplast isolation stress. Emission spectra were recorded before and after photobleaching of acceptor. Decreases in acceptor and increases in donor fluorescence intensities, respectively, were smaller for these pairs of VHA-fusions than for VHA-a-YFP and VHA-c-CFP (shown in Fig. 7C ). Nevertheless there was significant FRET in the case of VHA-A/YFP and VHA-B/CFP and some indication of FRET in the case of VHA-B/CFP and VHA-H/YFP, whereas the co-transformed pair of VHA-A/CFP and VHA-H/YFP gave no FRET. It should be noted that in addition to the highly expressed fusion proteins, untagged endogenous subunits still are present in the cell. Under such condition, formation of partial subcomplexes with possibly varied FRET properties may not be ruled out. Discussion The set of subunits that assemble plant V-ATPase has recently been completed by similarity searches in plant genomic and EST sequences using information on yeast VMA and other orthologues [ 5 , 6 ]. The presence of a subunit of about 100 kDa in the functionally active plant vacuolar ATPase has been under discussion for a long time [ 14 ]. Here, for the first time, a cDNA coding for a plant VHA-a was cloned and characterised. McVHA-a as well as the homologous Arabidopsis gene products contain all charged amino acids that have been shown to be essential for proton-translocation at conserved positions in the membrane spanning region of the C-terminus (Fig. 8 ). Additional evidence for an essential function of VHA-a in V-ATPase was first obtained for V-ATPase of Bos bovis . The specific V-ATPase-inhibitor bafilomycin [ 29 ] was shown to bind to the 100 kDa VHA-a subunit and not, as previously suggested to the proteolipid VHA-c. The results were confirmed for all tested species [ 30 ]. Bafilomycin also is a potent inhibitor of plant V-ATPase and is routinely used to distinguish V-ATPase-dependent ATP hydrolysis from background activity [ 31 ]. In 1999, Li and Sze [ 14 ] detected two unknown polypeptides with apparent molecular masses of 63 and 54 kDa in purified catalytically active V-ATPase but no polypeptide with a molecular mass of about 100 kDa. The authors hypothesized that plant V-ATPase is active in the absence of the 100 kDa subunit. Another explanation for this observation could be the sensitivity of VHA-a to degradation through proteases [ 32 ], and the detected unknown polypeptides could result from limited proteolysis of VHA-a, producing subunit-fragments still capable of transporting protons. The immunoblots with soluble fractions and membrane-preparations of M. crystallinum (Fig. 2 ) incubated with anti VHA-a N-term support this hypothesis. The 65 kDa fragment in the cytoplasmic fraction is likely to derive from a proteolytic processing, releasing the soluble portion of McVHA-a. Accordingly, high NaCl concentration in the purification medium could inhibit involved proteases. The hypthesis is supported by the experiment with protease inhibitor cocktail, where the 95 kDa subunit was detected as band with intermediate intensity, although only among other bands that were immuno-responsive to anti-VHA-a antibody. Thus even the protease inhibitors could not fully suppress proteolysis. Immunoprecipitation with the antibodies anti VHA-a N-term and anti VHA-a memb further proved that both the N-terminal hyrophilic portion and the membrane sector of VHA-a are associated with a complete V-ATPase complex. Figure 8 Membrane topology of plant VHA-a and location of the C-termini of individual subunits based on the FRET data. Based on amino acid sequence analysis and similarity with Vph1 (Leng et al. 1999), the topology of Mc-VHA-a is depicted in (A). The relative location of amino acid residues essential for proton pumping or structure are indicated with boxes. The numbers indicate the amino acid positions. In (B), the results from the FRET experiments are summarized: Asterisks tentatively mark the positions of the C-termini where the GFP variants have been fused to. In plants, immuno-cytochemical examinations of the subcellular localisation of single subunits or holocomplexes (mostly using antibodies directed against VHA-A) have previously indicated a distribution of V-ATPase among nearly all endomembranes of the secretory pathway including the plasmalemma [ 1 - 3 ], [ 33 - 35 ]. These findings were supported by measurement of bafilomycin-sensitive ATPase activity in selectively purified endomembranes of plants [ 36 - 38 ]. In our examination the antisera against VHA-A and -E showed a staining of the tonoplast and to a lesser extent also of the ER and Golgi-Apparatus (data not shown). VHA-A and E are part of the cytoplasmically exposed V 1 -structure of V-ATPase. The presence of both subunits on all these membranes shows the presence of fully assembled V-ATPase. The presence of the active V-ATPase in plants is not only necessary at the tonoplast, since Matsuoka et al. [ 38 ] could show that the activity of the vacuolar ATPase on the ER and the GA is necessary for correct targeting of soluble storage proteins to the vacuole. The presence of active V-ATPase on the prevacuolar compartment was concluded from the acid pH-optimum of enzymes involved in vacuolar transport, for example the activity of the vacuolar sorting receptor BP-80 whose action is strictly pH dependent [ 21 ]. For this reason a similar staining pattern of all used anti-VHA antisera would have been anticipated, but our results showed distinct staining patterns of anti-VHA-A and -E on the one hand and anti-VHA-a N-term on the other hand. Immuno-labelling indicates that VHA-a N-term antiserum exclusively labels the ER with some rare association on Golgi stacks. These results are surprising since the sequence features of McVHA-a suggest an essential involvement of VHA-a in proton transport. A possible explanation might be the presence of three different isoforms in A. thaliana and O. sativa which all share the localisation of the charged amino acids responsible for proton-translocation (Fig. 1 ). Based on our results, we suggest that the isoform of VHA-a recognised by anti VHA a- Nterm (an antibody which was generated against the isoform-specific N-terminus) is exclusively associated with V-ATPase localised on the ER. The hypothesis of a compartment-specific localisation of VHA-a subunits is supported through several findings in plants and yeast. In yeast all known subunits and chaperones of the V-ATPase are encoded by one gene. Only the 100 kDa VHA-a is encoded by two different isoforms (Vph1 and Stv1) [ 13 ]. In general the subunit-isogenes of the V-ATPase have a very high degree of similarity within each species [ 16 , 39 ]. In a converse manner, the sequences of the VHA-a isogenes are very heterogenic in S. cerevisiae and A. thaliana (Fig. 1 ) especially in the cytoplasmic region of the N-terminus [ 40 ]. A differential localisation was shown for the two yeast isoforms (Vph1 and Stv1) [ 12 ]. A detailed examination through Kawasaki-Nishi et al. [ 40 ] showed a localisation of Vph1 on the tonoplast, while Stv1 was detected on the late Golgi-apparatus and the prevacuolar compartment of yeast. By expressing chimeric proteins composed of the N-terminus of Stv1 and the C-terminus of Vph1, and vice versa, the authors were able to show that the N-terminus defines subcellular sorting. Following selective enrichment of the differentially localised complexes, both types of V-ATPases were shown to differ in their stability of the V 0 /V 1 -complex and in their coupling efficiency [ 41 ]. The N-terminus was responsible for the differential coupling activity, whereas the C-terminus mediated the differential dissociation. In plants, Matsuoka et al. [ 38 ] were able to distinguish between two different V-ATPase activities through their differential response to the V-ATPase inhibitor bafilomycin. These V-ATPase activities were localised in distinct membrane fractions of the secretory pathway including the vacuolar compartments. An antibody directed against the V-ATPase holoenzyme revealed significant differences in immuno-staining of endomembrane and vacuolar fractions. These and our results on sequence properties of the different plant VHA-a isoforms and the intracellular localisation of VHA-a support the hypothesis that at least two different V-ATPase activities exist in plants, differing in intracellular localisation and sensitivity to bafilomycin. The target of bafilomycin is VHA-a [ 29 ]. Thus, both types of V-ATPases might be distinguished through the presence of different isoforms of VHA-a. Our results from FRET analysis also allows to make some structural assignments: From crystal structure of F-ATP synthase, the C-termini of subunit β, homologous to VHA-A, and subunit α, homologous to VHA-B, are located in close vicinity and oriented to the membrane [ 42 ]. Occurrence of FRET between VHA-A/YFP and VHA-B/CFP supports the same structural arrangement in V-ATPase. Following crystalization of isolated yeast VHA-H [ 16 ], the structure was fitted in 3D reconstructions of plant V-ATPase based on electron microscopic analysis [ 43 ] and suggests localization of the C-terminus of VHA-H to the head structure in proximity to VHA-B. In vivo -FRET in protoplasts expressing VHA-B/CFP and VHA-H-YFP confirms the orientation of the C-termini of VHA-B and H in close vicinity. It should be noted that co-expression of other pairs of chimeric donors and acceptors such as VHA-E/CFP and VHA-c/YFP did not elicit FRET after excitation of CFP (not shown). The assumed location of the C-terminus of VHA-c in the lumen of the endomembrane compartments and the C-terminus of VHA-E most likely in the vicinity of the head is in agreement with the negative result, i.e. the absence of FRET between VHA-E/CFP and VHA-c/YFP. The studies exemplify the suitability of FRET to analyse structural features of V-ATPase in vivo . The efficient but variable FRET in cells expressing VHA-c/CFP and VHA-a/YFP allows two conclusions: First, the C-termini of VHA-c and VHA-a are likely to be located on the same, i.e. luminal, side of the endomembrane compartments supporting the topological model with nine transmembrane-domains of VHA-a in plants as previously suggested for yeast [ 22 , 23 ] (Fig. 8 ). Second, a significant portion of total VHA-a is located in the neighbourhood of VHA-c. The calculated distance of 5.4 to 7.2 nm between donor and acceptor fluorophore corresponds to the diameter of the proteolipid-ring of the rotor, consisting of 5 subunits of VHA-c and 1 VHA-c", respectively. The results confirm that VHA-a is part of the functional complex and not only involved in V-ATPase assembly. More than half (cf. Fig. 2 ), and possibly all V-ATPase complexes contain the holopolypeptide of 95 kDa. Conclusions The analysis of the primary structure of plant VHA-a revealed the presence of amino acid residues that are essential for proton pumping in yeast. Employing immuno-co-precipitation and FRET it could be demonstrated that subunits VHA-a and VHA-H are part of the V-ATPase complex of plants. Furtheron it is shown that one distinct VHA-a subunit isoform is localized on the ER. The study also shows the usefulness of FRET to study multisubunit protein structures in vivo and in vitro. Methods Plant growth Mesembryanthemum crystallinum and Arabidopsis thaliana were grown in hydroponics and soil culture, respectively, as described in [ 10 , 44 ]. Growth conditions were 120 μmol quanta m -2 s -1 , 60% relative humidity, 20°C, and a daily photoperiod of 12 h duration. Rosette leaves from 3- to 5-week-old Arabidopsis plants were taken for protoplast transfection. Zea mays and Hordeum vulgare were germinated on filter paper in the dark at 25°C for 48 h. Cells were isolated from the first 2 mm of the growing root tip. Onion epidermis was stripped from onion bulbs obtained from a local market. Membrane isolation Leaves (50 g) of M. crystallinum were homogenized in a buffer containing 250 mM sucrose, 50 mM Tris-Cl, pH 8.0, 4 mM ethylenediamine tetraacetic acid (EDTA), 4 mM dithiothreitol and a few crystals of phenylmethylsulfonylfluoride [ 45 ]. As indicated in a set of experiments, either NaCl was added at 100 or 500 mM concentration or complete protease inhibitor ® cocktail (Roche, Mannheim, Germany) was added throughout the procedure. Following differential sedimentation and gradient centrifugation, tonoplast enriched membranes were recovered from a 30%/35% sucrose interphase, sedimented, frozen in liquid nitrogen and stored at -80°C. Gel electrophoresis and Western blot detection Membrane proteins were separated on 12.5% sodium dodecylsulfate polyacrylamide gels, transferred to nitrocellulose and probed with anti-VHA-E [ 46 ], anti-VHA-A (kind gift of Dr. R. Ratajczak and Prof. U. Lüttge, TU Darmstadt, Germany) raised in rabbit or anti-VHA-a raised in guinea pig. Following incubation with primary and secondary antibody conjugated with peroxidase, detection was achieved with the lumilight ® system according to the supplier (Roche, Mannheim, Germany). Immunoprecipitation For immunoprecipitation, membranes were solubilised in 50 mM Tris-Cl, pH 7.5, 150 mM NaCl, 1 mM EDTA and 2% (v/v) Triton X-100, 5 μl anti VHA-a antiserum was added, and the samples were shaken at room temperature for 45 min. Then 150 μl protein A-sepharose equilibrated in the same buffer was added. After 15 min, the suspension was placed on a cushion of 1 ml of 40% sucrose and spun at 10,000 × g for 1 min. The sediment was washed thrice with 50 mM Tris-Cl, pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% (v/v) Triton X-100 and 0.1 % (w/v) SDS, and finally once in 125 mM Tris-Cl, pH 6.8. The sediment was boiled in loading buffer and analysed by Western blot using rabbit antisera raised against VHA-E or A. Anti-VHA- a antibody preparation and other antibodies used in this study Two antibodies against specific domains of VHA-a were raised in rabbits, and denominated anti-VHA-a N-term and anti-VHA-a Memb . For both the corresponding cDNA fragments of Mc-VHA-a were amplified by PCR using primer combinations a-nterm-f (ATG CGA TCG GAG CCG ATG CAA) and a-nterm-r (TTC ACC CAA CTC ATC GGT GG) encoding the 42 N-terminally located fragment, and a-memb-f (CTT CCA AAG CCC TTT ATT ATG) and a-memb-r (TCA CTC ATG TCC ACC ATG TCA ATC) encoding the polypeptide loop of about 13 kDa located between transmembrane domain 3 and 4 according to the topological model of Vph1p of S. cerevisiae [ 47 ]. The gene fragments were cloned into the vector pCR-T7-NT-Topo (Invitrogen, The Netherlands) and transformed into E. coli JM109. The 6x-his-tagged proteins were expressed, purified by chromatography on Ni-nitrilotriacetate columns, separated by preparative SDS-PAGE, excised as protein bands, eluted and used for immunization (Pineda, Berlin). In addition, antisera against subunits VHA-A (kind gift of Dr. R. Ratajczak and Prof. U. Lüttge, TU Darmstadt, Germany), VHA-E, calreticulin ([ 48 ], kindly provided by Andrew Smith, Oxford, UK), Jim 84 ([ 49 ] kindly provided by Chris Hawes, Oxford, UK), γ-TIP ([ 27 ]; kindly provided by John C Rogers, Washington State University, USA) were used in the co-localization studies. Construction of fusions between VHA subunits and variants of green fluorescence protein (GFP) Mc-VHA-a and -c were cloned into the vectors pECFP/pEYFP (Clontech, Palo Alto, USA) in a site-directed manner after amplification from cDNA [ 10 ] using the primers a-ges-BamHI-f (AAA AGG ATC CAT GCG ATC GGA GCC GAT GCA A) and a-ges-NcoI-r (AAA AAC ATG GCC TCT TCT TCT TCA CCA ATC GT), McVHA-c with c-ges-BamHI-f (AAA AGG ATC CAT GTC AAC CGT CTT CAA TGG) and c-ges-NcoI-r (AAA ACC ATG GCT GCC CTT GAC TGT CCA GCT CG). Mc-VHA-A and Mc-VHA-H were cloned as described in [ 10 ]. The constructs were introduced into the vector p35SGFP [ 50 ], so that the chimeric genes were placed under control of the 35S promoter and the original GFP gene was lost. The same strategy was used to produce Mc-VHA-A, -B and -H gene fusions with variants of GFP. Protoplast isolation and transformation methods Protoplasts were gently sedimented by centrifugation, resuspended in W5 medium, sedimented again, resuspended in MMG medium (0.4 M mannitol, 15 mM MgCI 2 , 4 mM morpholinoethane sulfonic acid, KOH , phl 5.7) and checked for sufficient intactness in the microscope. In short, 1 mm leaf slices of 3- to 5-week-old Arabidopsis plants were vacuum-infiltrated and cell walls were digested in media containing 1.5 % (w/v) cellulase R10 and 0.4 % (w/v) macerozyme R10. Protoplasts were gently sedimented by centrifugation, resuspended in W5 medium, sedimented again, resuspended in MMG medium (0.4 M mannitol, 15 mM MgCl 2 , 4 mM morpholinoethane sulfonic acid, KOH, pH 5.7) and checked for sufficient intactness in the microscope. 110 μl PEG-medium (4 % (w/v) polyethylene glycol 4000, 0.2 M mannitol, 0.1 M CaCl 2 ) and 20 μl plasmid DNA (3 μg/μl) were added to 100 μl protoplast suspension. The samples were incubated at room temperature for 15 min and then consecutively diluted with 0.5, 1, 2 and 4 ml W5-medium with 15 min incubation steps in between (154 mM NaCl, 125 mM CaCl 2 , 5 mM KCl, 2 mM morpholinoethane sulfonic acid, KOH, pH 5.7). Following 24 h incubation at 25°C, sedimented protoplasts were used for analysis. Cells of onion epidermis were placed on filter paper soaked with one-strength MS basal medium in petri dishes and were transiently transformed with a biolistic approach. Gold particles (1.6 μm, 60 mg/ml) were suspended in 50 % glycerol. 8.33 μl of the suspension were mixed with 8.33 μl plasmid DNA (1 μg/ μl), 8.33 μl 2.5 M CaCl 2 , 3.33 μl 0.1 M spermidine. Sedimented gold particles were consecutively washed with 70 % and 100 % ethanol and resuspended in 8 μl 100 % ethanol, loaded on a macro carrier for transformation with the Particle Delivery System using a rupture disc of 1100 psi (PDS-1000/He, Biorad, Hercules, USA). The distance between macrocarrier and tissue was 12 cm. The epidermis tissue was incubated for about 20 h at room temperature in the dark prior to analysis. Immuno-fluorescence labelling and image acquisition by confocal laser scanning microscopy (CLSM) Immuno-labelling was performed according to [ 26 ]. In brief, cells were fixed in 3.7 % para-formaldehyde (10 mM MgSO 4 , 10 mM EGTA, 1 × phosphate buffered saline, pH 6.8), washed, permeabilised in 0.5% Triton X-100 and washed again. Following blocking of non-specific binding sites with 1 % bovine serum albumin, primary antibody was added for over night at 4°C. Washed samples were incubated with secondary antibody labelled with Cy3, Cy5 or FITC for 1 h. Double labelling was performed by combined application of primary antibodies from rabbit and guinea pig. Slides were mounted with Citifluor Mounting Medium. Fluorescence analysis was performed with a confocal laser scanning microscope Leica TCS-SP2 (Leica, Heidelberg, Germany) equipped with three lasers and excitation wavelengths of 458, 476, 488, 514, 568 and 633 nm. The double dichroic mirror DD488/543 was used for fluorescein isothiocyanate (FITC), and for Cy5 the triple dichroic mirror TD488/543/633 was used. Background was controlled and photomultiplier voltage (800 V) selected for maximum sensitivity in the linear range. Immunogold-labelling and electron microscopy Cells were fixed in 2.5% glutaraldehyde in EM buffer (50 mM KH 2 PO 4 , 50 mM NaH 2 PO 4 , pH 7.0) for 45 min, washed with EM buffer and dehydrated with a series of increasing concentration of acetone. Samples were embedded in epoxyresin (Transmit EM, TAAB laboratories equipment, Berkshire, Great Britain), cut into ultra-thin cross-sections of 60–70 nm and immobilized on 200 mesh gold nets. Immuno-decoration was performed with antibody diluted in Tris-buffered saline (TBS, 10 mM bovine serum albumin and 0.05 % (w/v) NaN 3 ) for an hour. Samples were washed five times and incubated with secondary antibody conjugated to 15 nm gold particles. The samples were stained with 0.1 % (w/v) uranyl acetate for 5 s and afterwards with 2 % lead citrate. The samples were analysed with an electron microscope (H500, Hitachi, Japan) at 75 kV. Confocal microscopy of GFP-fusion proteins and FRET-measurement Transformed protoplasts and onion epidermis cells were examined for the localisation of the CFP/YFP-fused proteins using the same CLSM set-up as mentioned above. Autofluorescence of 10 protoplasts, as well as reference spectra of YFP and CFP-derived fluorescence were recorded in the spectral range of 480 to 700 nm, averaged and used for corrections. Excitation was recorded at 458 nm (CFP and FRET) and 514 nm (YFP), respectively. Scan speed was 800 Hz. Acceptor dye was bleached with 100 % laser intensity. Emission spectra were recorded and averaged from 20 transformed protoplasts. For a first estimate of transfer efficiency, a Foerster radius for green fluorescence protein variants of R o = 5 nm [ 52 ] was used to calculate the donor/acceptor distance via the equations E = (I CFP/bleached - I CFP/unbleached )/I CFP/unbleached and R = ((R o 6 /E)-R o 6 ) 1/6 , where E is the transfer efficiency, and I CFP the fluorescence emission intensity in the CFP peak. Abbreviations CFP: cyan fluorescence protein; CLSM: confocal laser scanning microscope; FRET: (Förster) fluorescence resonance energy transfer; PAGE: polyacrylamide gel electrophoresis; Stv1: VHA-a subunit isogene in yeast; VHA: vacuolar H + -ATPase; Vph1: VHA-a subunit isogene in yeast; YFP: yellow fluorescence protein Authors' contributions CK: VHA-a sequence analysis, immuno-cytochemistry, transient expression systems, preparation of anti VHA-a N-term ; TS: co-transfection of protoplasts and epidermis cells, CLSM analysis; SB: immuno-co-localisation and discussion; SS: co-immuno-precipitation; MH: preparation of anti VHA-a Memb ; BS-J: immuno-co-localisation and discussion; JR and MS: FRET analysis; DG: transformation and construct design, KJD: project design and supervision.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC516168.xml
528858	Add-on therapy options in asthma not adequately controlled by inhaled corticosteroids: a comprehensive review	Many patients with persistent asthma can be controlled with inhaled corticosteroids (ICS). However, a considerable proportion of patients remain symptomatic, despite the use of ICS. We present systematically evidence that supports the different treatment options. A literature search was made of Medline/PubMed to identify randomised and blinded trials. To demonstrate the benefit that can be obtained by increasing the dose of ICS, dose-response studies with at least three different ICS doses were identified. To demonstrate whether more benefit can be obtained by adding long-acting β 2 -agonist (LABA), leukotriene antagonist (LTRA) or theophylline than by increasing the dose of ICS, studies comparing these options were identified. Thirdly, studies comparing the different "add-on" options were identified. The addition of a LABA is more effective than increasing the dose of ICS in improving asthma control. By increasing the dose of ICS, clinical improvement is likely to be of small magnitude. Addition of a LTRA or theophylline to the treatment regimen appears to be equivalent to doubling the dose of ICS. Addition of a LABA seems to be superior to an LTRA in improving lung function. However, addition of LABA and LTRA may be equal with respect to asthma exacerbations. However, more and longer studies are needed to better clarify the role of LTRAs and theophylline as add-on therapies.	Introduction Inhaled corticosteroids (ICS) are the mainstay of current asthma management and should be used in all patients with persistent asthma. Many patients with persistent asthma can be controlled with regular ICS. However, a considerable proportion of patients treated with ICS remain symptomatic, despite the use of low to moderate doses (doses defined according to the ATS classification for adults [ 1 , 2 ]: beclomethasone dipropionate (BDP) 200 – 1000 μg/d, budesonide 200 – 800 μg/d or fluticasone propionate (FP) 100 – 500 μg/d) of ICS. Based on the differences in potency and pharmacokinetics the doses could also be defined differently [ 3 , 4 ]. Recent treatment guidelines [ 1 , 2 , 5 , 6 ] classify these patients as having moderate to severe persistent asthma (steps 3 and 4). According to the recent guideline [ 2 ] the typical clinical features of step 3 asthma include symptoms daily, nocturnal symptoms at least once a week, exacerbations that may affect activity or sleep, forced expiratory volume in one second (FEV 1 ) 60 – 80% of predicted or peak expiratory flow (PEF) between 60 and 80% of the personal best reading. Daily rescue therapy is usually needed. Typical findings include low values of PEF or FEV 1 , a marked variation in daily PEF recordings and/or a significant response to bronchodilators. Thus, asthma is not adequately controlled, and the treatment needs to be optimized. According to current guidelines the therapeutic options in the treatment of asthma not adequately controlled by low to moderate doses of ICS are as follows: 1. Increase in the dose of the ICS, 2. Addition of long-acting β 2 -agonist (LABA; formoterol or salmeterol), 3. Addition of a leukotriene receptor antagonist (LTRA; montelukast, pranlukast or zafirlukast) and 4. Addition of theophylline. Currently, the National Heart, Lung and Blood Institute guideline [ 2 ] recommends addition of LABA as the first choice and gives the other choices as secondary options, but leave the clinician alone to make the decision without offering comprehensive data to support the different options. Recently, this "step-3" dilemma on the different treatment options has gained attention [ 7 , 8 ]. Several of these options have been separately assessed in several reviews, systematic reviews and metaanalyses [ 7 , 9 - 16 ]. However, no comprehensive reviews exist on the subject. The aim of our article is to review the evidence that supports the increase in the dose of ICS and use of the different "add-on" options. Firstly to demonstrate the benefit that can be obtained by increasing the dose of ICS, dose-response studies with at least three different ICS doses were identified. Secondly, to demonstrate whether more benefit can be obtained by adding LABA, LTRA or theophylline to the treatment than by increasing the dose of ICS, we aimed to identify studies where the addition of a LABA, LTRA or theophylline to the treatment regimen was compared with the addition of a corresponding plabeco to an increased dose (usually doubled dose) of ICS. Thirdly, we aimed to identify studies comparing the different "add-on" options. In this review, we hope to help the clinician facing the "step-3 dilemma" by presenting in a systematic way the evidence obtained from randomised clinical trials that supports the use of these different treatment options. Methods The paper reviews studies where participants were adults or adolescents (≥12 years) with clinical evidence of asthma not adequately controlled with ICS. The general inclusion criteria in this review were: randomized, blinded and controlled trials with either parallel group or cross-over design published as a full-length paper. Steroid-tapering studies were not included as they are difficult to interpret. Studies published in abstract form only were not included. Similarly, studies lasting less than 4 weeks, containing less than 10 patients per group or studies containing a significant proportion (>10%) of patients using systemic steroids were excluded. Similarly "add-on" studies where a significant proportion (>10%) of patients were not using inhaled steroids were excluded. We made a search of Medline from January 1 1966 to October 2001. All searches were limited to studies published in the English language. To identify the latest studies published, another search was made by using the drug names (budesonide, beclomethasone, fluticasone, flunisolide, mometasone, triamcinolone, formoterol, salmeterol, montelukast, pranlukast, zafirlukast, theophylline) from Medline on October 2003. The searches were manually (HK) evaluated to identify studies fulfilling the inclusion criteria and full papers were retrieved. In the case of uncertainty based on the abstract full papers were retrieved. All studies fulfilling the inclusion criteria for the ICS dose-response studies or "add-on" studies (see below) were scored for quality using the method described by Jadad et al. [ 17 ]. Furthermore, relevant systematic reviews were identified from the Cochrane Library (Issue 2, 2003). In addition, some in vitro results or results from open, non-randomized or uncontrolled trials or meta-analysis of particular relevance to the present topic may be cited. Inclusion criteria for dose-response studies with ICS To find the dose-response studies with ICS the term "anti-inflammatory agents, steroidal" was combined with the term: "dose-response relationship, drug" (MeSH), which combination produced 249 papers. To demonstrate the dose-response effect of ICS only controlled studies with at least three different ICS doses and a parallel-group design were included. Studies using consecutive doses of steroids were not included because it makes it impossible to differentiate the dose-response relation from the time course relation of efficacy. Inclusion criteria for "add-on" studies with long-acting β 2 -agonists, leukotriene antagonists and theophylline When the basic search done with the term "anti-inflammatory agents, steroidal" was combined with another made with terms: "salmeterol OR formoterol" it produced 97 papers, when combined with a search made with a term "leukotriene antagonists" (MeSH), it produced 26 papers and when combined with a search with a term "theophylline" (MeSH) it produced 342 papers. Only studies where the addition of LABA, LTRA or theophylline to the treatment with inhaled steroid was compared with the addition of a corresponding placebo to an increased dose (usually double-dose) of inhaled steroid were included. In addition, studies comparing the different "add-on" options were identified. Increasing the dose of inhaled corticosteroid On the design of dose-response studies with ICS We identified 14 studies [ 18 - 31 ] assessing the dose-response relationship of ICS in the treatment of chronic asthma. All included studies were of fair to excellent quality (Jadad score 3–5). The main characteristics of these studies are presented in Table 1 (see Additional file 1 ). The inclusion criteria in most of these studies were moderate to severe chronic asthma but previous use of small to moderate doses of ICS was not required in all studies. The studies included patients with a relatively wide range of FEV 1 % predicted and based on that these patients belong to steps 2–4 according to the recent guideline [ 2 ]. In all except three studies a ≥12% reversibility in FEV 1 or PEF in response to a bronchodilator was required. There was 1 study that assessed the dose-response of budesonide, 7 of FP, 1 of BDP, 3 of mometasone furoate, and 2 of triamcinolone acetonide. The studies utilized two main approaches to identify a dose-response relationship. Some studies considered dose-response relationship to be present if the results obtained with the lowest and highest dose of ICS were significantly different, whereas in others the presence or absence of dose-response relationship was characterized with more advanced statistical analysis (e.g. analysis for linear trend or Jonckheere's nonparametric trend test). In this review, both ways of analysis are accepted as evidence for the presence of dose-response. In the following discussion the important difference between the formal dose-response studies presented in this review and the results reported in some meta-analysis is that the data of the meta-analyses may result from studies assessing one or more doses of ICS and comparing their effects with placebo or baseline. Thus, the data derived from some the published meta-analyses [ 9 , 11 , 14 , 32 ], although showing a dose-response effect, is obtained by combining different doses from several studies, and is not resulting from a strict dose-response relationship study. In addition, the data obtained using meta-analysis may be derived only from one or two studies. Overview on lung function and symptoms in the 14 included studies Studies with ICS show a statistically significant dose-response effect for morning PEF and FEV 1 in the treatment of chronic asthma in 9 (69%) and 5 (31%) studies of the 14 studies included, respectively ( Table 2a, see Additional file 1 ). However, statistical analysis of dose-dependency fails to show any significant dose-related effect for FVC in 5 (71%) studies of 7 where it was analysed. Similarly, no statistical dose-dependency was found for evening PEF in 6 (50%) studies out of 12 where it was analysed ( Table 2a, see Additional file 1 ). The total or daytime symptom scores show a statistically significant dose-response effect in 5 (38%) out of 13 studies, whereas nighttime symptom score showed a dose-dependency in only three (25%) studies out of 13 where it was analysed. A dose-response for the rescue β 2 -agonist use was found in 4 (33%) out of 12 studies where it was analyzed ( Table 2b, see Additional file 1 ). The difference between the highest and the lowest dose of ICS was most often statistically significant for morning PEF (7/12 studies; 58%) and to a lesser extent for evening PEF (3/10 studies; 30%), FEV 1 and total or daytime symptom scores (both 2/12 studies; 16.7%), night-time symptom score and rescue β 2 -agonist use (both 1/11 studies; 9%) and FVC (0/6 studies; 0%). Similarly, the difference between the two consecutive doses of ICS was very seldom statistically significant ( Table 2ab, see Additional file 1 ). Thus, taken together, the results suggest that morning and evening PEF and FEV 1 are more sensitive to show a statistically significant dose-response effect for ICS, whereas symptom scores and rescue β 2 -agonist use are in general less sensitive to the increase in steroid dose. However, this conclusion may also be influenced by the duration of treatment. Inclusion of relatively short studies in this review, may either under- or over-estimate the dose-response differences depending on the outcome measure being used. Beclomethasone dipropionate – studies included in this systematic review The dose-response relationship of the effects of BDP (100 – 800 μg/d in two different formulations) was evaluated in asthmatic subjects who had deterioration in asthma control after discontinuation of ICS [ 18 ]. There was a statistically dose-dependent effect on morning PEF, FEV 1 , FVC, days free from wheeze or chest tightness and β 2 -agonist use, but not on evening PEF or nights free from asthma related sleep disturbance ( Table 2ab, see Additional file 1 ). The dose-response effects detected in this study may reflect the fact that the patient population was carefully identified to show a well-defined responsiveness to ICS. Thereafter ICS were withdrawn to induce a clinically meaningful deterioration of asthma control. Thus, the design may not directly reflect clinical practice, where a patient is symptomatic, despite the use of low to moderate doses of ICS. Beclomethasone dipropionate – other literature A recent meta-analysis [ 10 ] analysed the dose-response effect of BDP in the treatment of chronic asthma. Eleven studies with variable methodological quality involved 1614 subjects were included in the analysis. Most of the endpoints were based on only 1–2 studies. In asthmatic patients not treated with oral steroids a small advantage of BDP 800 μg/d over 400 μg/d was apparent for improvement in FEV 1 and morning PEF and reduction in night-time symptom score compared to baseline. Studies that assessed BDP 1000 v 500 μg/d and BDP 1600 v 400 μg/d demonstrated a significant advantage of the higher dose compared to the lower dose for percentage improvement in airway responsiveness to histamine and FEV 1 compared to baseline. No differences between higher and lower daily doses of BDP were apparent for daytime symptoms, withdrawals due to asthma exacerbations or oropharyngeal side effects. Budesonide – studies included in this systematic review A 6 weeks dose-response study in Japanese asthmatics previously not on ICS showed that increasing the dose of budesonide (200–800 μg/d) [ 19 ] results in a dose-related improvement in morning and evening PEF and daytime and nighttime symptom scores, but not for FEV 1 . In this study, there was no statistically significant difference between the doubling doses of budesonide ( Table 2ab, see Additional file 1 ). Instead, even the lowest dose of budesonide (200 μg/d) was superior to placebo in the case of morning and evening PEF and daytime and night-time symptom scores, but not for FEV 1 . Budesonide – other literature In a randomised, double-blind, placebo-controlled study of parallel-group design lasting 12 weeks four different doses of budesonide (200, 400, 800 and 1600 μg/d were compared in patients suffering from moderate to severe asthma. This study was not included in the systematic analysis due to a high proportion of patients on oral glucocorticoids (15.6%). Increasing the dose of budesonide [ 33 ] results in a dose-related improvement in morning PEF and FEV 1 , but not in evening PEF, FVC, symptom scores or rescue β 2 -agonist use. Instead, even the lowest dose of budesonide (200 μg/d) was superior to placebo for all parameters studied. The improvement induced by these low doses is much greater than the difference between the lowest and highest doses of budesonide studied, despite the 8-fold difference in the dose (Figure 1 ) [ 33 ]. There was a statistically significant difference only between the lowest (200 μg/d) and the highest (1600 μg/d) doses of budesonide when morning PEF or FEV 1 were analysed. Instead, the lowest (200 μg/d) or the highest dose (1600 μg/d) did not differ from the two medium doses (400–800 μg/d). When evening PEF, FVC, daytime or nighttime asthma symptom scores or the use of rescue medication were analysed, there was no significant differences between any of the studied budesonide doses [ 33 ]. Figure 1 Mean change from baseline in morning peak expiratory flow (PEF) in patients treated with placebo or various doses of budesonide. A significant dose-response effect is seen. However, it should be noted that the difference between placebo and low-dose budesonide is greater than the difference between low-dose budesonide and high-dose budesonide and that there is no statistically significant difference between the various doses of budesonide. Reproduced from reference 33 with permission. The dose-relationship of budesonide in the treatment of chronic asthma is a subject of a recent Cochrane review [ 12 ]. In this meta-analysis including both children and adults (n = 3907) in non-oral steroid-treated mild to moderately severe asthmatics no clinically worthwhile differences in FEV 1 , morning PEF, symptom scores or rescue β 2 -agonist use were apparent across a dose range of 200–1600 μg/d. However, in moderate to severe asthma there was a significant reduction in the likelihood of trial withdrawal due to asthma exacerbation with budesonide 800 μg/d compared with budesonide 200 μg/d. The reviewers also conclude that budesonide exhibits a significant improvements favouring high dose (1600 μg/d) over low dose (200 μg/d) for improvement in FEV 1 in severe asthma [ 12 ]. Another recent meta-analysis combining 3 placebo-controlled studies with at least two different budesonide doses demonstrated a statistically significant dose-response for morning PEF and FEV 1 but not for evening PEF [ 14 ]. Fluticasone propionate – studies included in this systematic review The dose-dependency of FP has been studied in seven studies in patients with mild to moderate asthma. In two of the studies, patients were previously not on ICS ( Table 1, see Additional file 1 ). The difference between the highest and lowest dose was 4- to 20-fold. In all studies almost all parameters improved significantly better with all doses of FP as compared with placebo. Only three studies [ 20 , 21 , 26 ] show a dose-response effect on morning PEF, only two studies [ 20 , 26 ] show a dose-response relationship for evening PEF and rescue medication use and only one study [ 20 ] shows a dose-response relationship for FEV 1 , FVC and daytime symptom score ( Table 2ab, see Additional file 1 ). When different doses of FP (50–200–1000 μg/d) were studied in a randomized, double-blind dose-response setting, there was no difference in FEV 1 , FVC, evening PEF, symptom scores, use of rescue medication or the number of night awakenings between the lowest and highest FP dose, despite a 20-fold difference in the dose [ 21 ]. Only for morning PEF was the high (1000 μg/d) dose of FP better than the two lower doses, whereas even the lowest dose of FP (50 μg/d) was significantly better than placebo in improving all these parameters. In a dose-response study [ 20 ] with patients with symptomatic chronic asthma (n = 672) patients were randomized to four different doses of FP (100, 200, 400, 800 μg/d). FP improved lung function and symptoms in a dose-related manner. The linear trend for doubling the dose of FP was calculated to be as follows: morning PEF increased 4.3 L/min (95% CI 1.8–6.8) and FEV 1 increased 0.03 L (95% CI 0–0.05 in two weeks). How does this translate into clinical practice? When assessing a response to a bronchodilator or when assessing a response to inhaled or oral steroid an improvement of 10–20% above the previous values is often considered significant. Thus, in the above study, this would mean >36 L/min increase in morning PEF values. Recently, the average minimal patient perceivable improvements have been estimated as 18.8 L/min for PEF and 0.23 L for FEV 1 [ 34 ]. Based on that the increase in lung function obtained by doubling the dose of fluticasone in the above study seems to be only of very limited clinical benefit. Fluticasone propionate – other literature In a recent meta-analysis [ 9 ] the dose-response relation of inhaled FP in adolescents or adults with asthma in eight studies [n = 2324] employing 2–3 different doses of inhaled FP were analysed. The dose-response curve for the raw data began to reach a plateau at around 100–200 μg/d and peaked by 500 μg/d. A negative exponential model for the data indicated that 80% of the benefit at 1000 μg/d was achieved at doses of 70–170 μg/d and 90% by 100–250 μg/d. A quadratic meta-regression showed that the maximum achievable efficacy was obtained by doses of around 500 μg/d. Another recent meta-analysis [ 11 ] of 28 studies with 5788 patients (children and adults) with chronic asthma evaluated the dose-response effect of FP, compared to placebo. Evidence for a dose-response effect was apparent for likelihood of trial withdrawal due to lack of efficacy, change in FEV 1 , morning PEF, evening PEF, nighttime awakening score and physician-rated efficacy. It is important to appreciate that this was only evident when improvements over placebo were compared for the highest dose of FP (1000 μg/d) and lowest dose of FP (100 μg/d). There were no significant differences when any other doses were compared (e.g. FP 200 v 100 μg/d, FP 500 v 200 μg/d, FP 1000 v 500 μg/d). Sixty percent (0.31 L; 95% CI 0.27–0.36 L) of the effect on FEV 1 with FP 1000 μg/d (0.53 L; 95% CI 0.43–0.63 L) was achieved with tenth of the dose. No dose-response effect was apparent for change in symptom score or for rescue β 2 -agonist use [ 11 ]. Another recent meta-analysis from the same authors [ 32 ] found a statistically significant advantage of FP 200 μg/d over 100 μg/d for morning PEF (6 L/min; 95% CI 1–10 L/min), evening PEF (6 L/min, 95% CI 2–11 L/min) and night-time awakening score (0.17, 95% CI 0.04 – 0.30), but not for FEV 1 , daily symptom score, night-time awakenings and daily use of rescue β 2 -agonist use. No significant advantage was obtained with the use of FP at doses of 400–500 μg/d over 200 μg/d for morning or evening PEF, FEV 1 , daily symptom score or rescue β 2 -agonist use. Patients treated with higher dose (800 – 1000 μg/d) of FP achieved significantly greater improvements in morning PEF (22 L/min, 95% CI 15–29 L/min) and evening PEF (13 L/min, 95% CI 6–19 L/min) compared to the lower dose (50–100 μg/d). Another recent meta-analysis [ 14 ] including eight trials with at least 2 different doses of FP demonstrated a statistically significant dose-response in morning PEF, evening PEF and asthma symptom score but not in FEV 1 or β 2 -agonist use. Mometasone furoate and triamcinolone acetonide – studies included in this systematic review Mometasone furoate is a corticosteroid closely related to FP and is being investigated in a dry powder inhalation formulation for the treatment of asthma [ 35 ]. Studies with mometasone furoate [ 27 - 29 ] show a dose-related efficacy in the treatment of mild to moderate asthma when morning PEF is analysed ( Table 2a, see Additional file 1 ). Interestingly, even doubling doses of mometasone furoate produced statistically significant improvements in morning and evening PEF ( Table 2a, see Additional file 1 ) [ 27 - 29 ]. Occasionally, a statistically significant dose-dependency or difference between the highest and lowest dose was found for evening PEF, FEV 1 or daytime or total symptom score. In contrast, no significant dose-dependency was found for FVC, nighttime symptom score or rescue β 2 -agonist use ( Table 2ab, see Additional file 1 ). Linear trend analyses showed a dose-response for triamcinolone acetonide (TAA) in the treatment of moderate to severe asthma across the dose-range of 150 to 600 μg/d or 200 to 1600 μg/d for most variables in the two studies included in this review ( Table 2ab, see Additional file 1 ) [ 30 , 31 ]. Occasionally, a statistically significant difference was reported even between two consecutive doses of TAA. As compared with placebo, therapeutic activity was generally evident at doses of 150–200 μg daily for all variables with significant clinical efficacy demonstrated for all doses. Mometasone furoate and triamcinolone acetonide – other literature A four-week randomised, double-blind, double-dummy and parallel group study [ 36 ] comparing the efficacy and safety of mometasone furoate administered by metered dose inhaler (112, 400 and 1000 μg/d) with BDP (336 μg/d) and placebo recruited adult patients with moderate asthma (n = 395). The patients were required to have a stable ICS dose, FEV 1 or 50–90% and a bronchodilator response of ≥15% in absolute FEV 1 at baseline. This study reported significantly better improvement in FEV 1 , FVC and morning PEF with doses of 400 and 1000 μg/d than with 112 μg/d. Also, physician's evaluation of asthma symptoms, but not salbutamol use was significantly better with dose 1000 μg/d than with 112 μg/d. This study, although fulfilling the criteria for dose-response study as defined in materials and methods, was excluded from the systematic evaluation, as the published statistical analysis did not include any formal dose-response analysis, and the reported difference between different mometasone doses always required a statistically significant difference to the active comparator BDP. In contrast to the results presented in this review ( Table 2ab, see Additional file 1 ), a meta-analysis [ 14 ] including 2 studies with mometasone furoate (200 μg/d versus 400 μg/d) failed to show any significant dose-response in FEV 1 . In the meta-analysis, there was not enough data to analyse other parameters than FEV 1 . The 3 studies [ 27 - 29 ] included in this review were not included in the meta-analysis [ 14 ]. The data suggests that 200 μg/d of mometasone furoate may be a relatively small dose. As both the inhaler device and mometasone have not been available for the treatment of asthma, it is difficult to define their exact position in the treatment of asthma, although there are data to suggest that a total daily dose of 400 μg of mometasone furoate administered with dry powder inhaler may be equal to total daily dose of 500 μg of FP via a Diskhaler or a daily dose of 800 μg budesonide via a Turbuhaler [ 28 , 29 ]. A placebo-controlled, double-blind parallel-group study assessed the effects of three different doses of TAA (450, 900 and 1800 μg/d for 12 weeks; delivered using a non-chlorofluorocarbon propellant) in patients with chronic symptomatic asthma and using ICS [ 37 ]. The data for all variables (FEV 1 , FEF 25–75 , morning and evening PEF, symptom scores and rescue salbutamol use) shows that even the lowest dose significantly differs from placebo, and there appears to be no clear dose-response. However, no formal statistical analysis was reported for the presence of a dose-response and thus this study is not included in Tables 1–2. A recent meta-analysis [ 14 ] including 3 studies with TAA, demonstrated a statistically significant dose-response in morning PEF, evening PEF and asthma symptom score, but not in FEV 1 . Conclusions on the effects of ICS on lung function and asthma symptoms Taken together these results indicate that the change in the ICS dose from low dose to moderate dose is at the flat part of the ICS dose-response curve for most lung function and symptom parameters studied (Figure 2 ). Furthermore, it appears that the low and moderate doses of currently used ICS are in the flat part of the steroid dose-response curve. Thus, it is predicted that doubling the dose of ICS is not sufficient to significantly improve lung function or reduce symptoms. Rather, the data suggest that the increase in the dose of ICS should be at least 4-fold to produce a clinically significant improvement in variables such as symptoms, use of rescue β 2 -agonists, PEF or lung function. However, the steepness of the dose-response curve for different outcomes may vary. For example, an open dose-response evaluation of different sequential doses of budesonide in patients with mild-to-moderate asthma (38) shows that the dose-response curves for FEV 1 /PEF and FEF 25–75 are not identical. Similarly, the dose-response curves of budesonide on adenosine monophosphate (AMP) and methacholine bronchial challenges were significantly different [ 38 ]. It should also be noted that patients often receive higher doses of ICS in their daily routine treatment than required [ 3 ]. Figure 2 The dose-response curve of inhaled glucocorticoids. The studies discussed above present mean data for groups of patients, but do not address the issue of differences in responsiveness to the anti-inflammatory effects of corticosteroids between individual patients. It may be possible that increasing the dose of ICS may be beneficial for some patients. Is there a dose-response in the anti-inflammatory effects of ICS? Studies included in this systematic review We were not able to identify any studies that would have studied the dose-dependency of the anti-inflammatory effects of ICS in asthma and would have satisfied the inclusion criteria for the present review. Other literature In a study [ 39 ] with patients with chronic asthma (n = 66) treated with moderate doses of ICS the dose-dependency of consecutive doses of budesonide (800, 1600 and 3200 μg/d) and FP (500, 1000 and 2000 μg/d) were studied. Budesonide increased methacholine PD 20 from 259 to 467 μg and FP from 271 to 645 μg, both showing a dose-dependency. However, no statistical comparison was made between individual doses. The PD 20 was increased 1.67-fold and 1.96-fold when the patients were switched from the lowest dose to the highest dose of budesonide and FP, respectively. An apparently dose-dependent decrease in the blood eosinophil count was obtained with budesonide but not with FP treatment [ 39 ]. In contrast, no significant differences were observed for either treatment, when morning or evening PEF, symptom scores, and consumption of β 2 -agonist were analysed. Allergen PC 15 and methacholine PC 20 values were determined before and after treatment with budesonide at 200, 400 and 800 μg/d for 7 days in a double-blind, randomized and cross-over study (6 day washout period) in eleven atopic subjects with inhalation allergy [ 40 ]. The allergen PC 15 and methacholine PC 20 were significantly larger for all doses of budesonide as compared with placebo, but there was no significant difference between the 3 doses of budesonide. In an open trial with patients with moderate to severe asthma the effects of progressively increasing doses of budesonide (400, 800, 1600 and 2400 μg/d) were studied [ 41 ]. Budesonide decreased the blood eosinophil count in a dose-dependent manner. In a double-blind, randomized placebo-controlled study combining two separate studies, the dose-dependency of the anti-inflammatory effects of budesonide (100, 400 and 1600 μg/d) was assessed in patients with mild asthma (n = 31). Based on trend analysis, there were dose-dependent changes in exhaled NO, sputum eosinophils and PC 20 to inhaled budesonide but a plateau response of exhaled NO was found at a dose of 400 μg/d [ 42 ]. In a study with a novel ICS ciclesonide, its effects were studied in a parallel-group, double-blind, placebo-controlled, randomized cross-over study (washout period 3–8 weeks) in patients (n = 29) with mild to moderate asthma [ 43 ]. Compared with placebo, ciclesonide for 14 days (100, 400 and 1600 μg/d) reduced airway responsiveness to AMP by 1.6, 2.0 and 3.4 doubling doses, respectively, and this effect was dose-dependent. A significant reduction in the percentage of eosinophils in induced sputum was observed after 400 and 1600 μg daily ciclesonide, but this was not dose-dependent. Sputum eosinophil cationic protein (ECP) was significantly reduced after 400 μg daily ciclesonide only, and no dose-dependent effect was seen. In a recent single-cohort, prospective placebo-controlled study with four 1 week periods with nonsteroid-treated asthmatic patients (n = 15) the effects of different doses of BDP (100, 400 and 800 μg/d) were measured on FEV 1 , exhaled nitric oxide (FENO) and methacholine PC 20 [ 44 ]. All doses of BDP resulted in a significant change in FEV 1 and methacholine PC 20 from baseline or placebo treatment, but with no significant separation of active BDP doses. All doses of BDP resulted in a significant change in FENO from placebo treatment, but with significant separation of only the 100 μg and 800 μg doses by FENO. Another study assessed the dose-response relationship of the anti-inflammatory effects of BDP (50, 100, 200 and 500 μg/d) in the treatment of mild to moderate asthma for 8 weeks in a randomised, placebo-controlled, double-blind trial of parallel-group design [ 45 ]. Maintenance ICS therapy was discontinued and patients were randomised to different treatment groups and inflammatory markers such as exhaled NO, sputum eosinophil counts and PD 15 to saline were followed. There was a significant linear relationship between BDP dose and exhaled NO concentration, FEV 1 and changes in sputum eosinophils at the end of treatment. In contrast no relationship was found between BDP dose and PD 15 to saline. However, the results of this study may be confounded because the patients were treated with oral prednisolone for two days in the beginning of the study. In a recent randomized and double-blinded study, 12 atopic mild stable asthmatic subjects were treated with placebo or mometasone furoate (100, 200 and 800 μg/d) for six days [ 46 ] in a cross-over fashion. All three doses of MF demonstrated similar attenuation of early responses and allergen-induced airway hyperresponsiveness relative to placebo with no dose-response relationship. In contrast, the late maximal % fall in FEV 1 after placebo treatment was 24% and was significantly reduced in a dose-dependent manner to 12%, 11% and 6% for the 100, 200 and 800 μg daily treatments. The allergen-induced sputum eosinophilia (×10 4 cells/ml) 24 h after challenge during placebo treatment was 60.2 and was significantly reduced to 24.0, 15.3 and 6.2 for the 100, 200 and 800 μg daily treatments, respectively. Although a statistically significant dose-response relationship was present, the difference between the lowest and highest dose (8-fold difference) for late maximal fall in FEV 1 or allergen-induced sputum eosinophilia was less than the difference between placebo and the lowest dose of MF. Taken together, the results suggest that there is tendency towards slightly higher anti-inflammatory efficacy with higher doses of ICS. At the moment there are only a few studies that assess the dose-dependency of the anti-inflammatory effects of ICS. Most of these studies included only small numbers of patients. However, despite the 4–8–16-fold differences in the doses of ICS studied, it has not been easy to demonstrate the dose-dependency of the anti-inflammatory effects of inhaled glucocorticoids. Thus, based on the scarce published evidence we would predict that doubling of the commonly used low to moderate doses of ICS is likely to produce only a small increase in the anti-inflammatory effect, suggesting that inflammation may be suppressed in most patients by relatively low doses of ICS. Is there a dose-response with the adverse effects of ICS? Glucocorticoids suppress corticotrophin levels, which may eventually lead to atrophy of the adrenal cortex and diminished levels of endogenous cortisol. The diminished levels of endogenous cortisol or reduced cortisol excretion have been used as markers of systemic activity of ICS. These systemic effects may include osteoporosis, behavioural effects, growth suppression, posterior subcapsular cataracts, risk for ocular hypertension and glaucoma as well as skin thinning and bruising [ 47 ]. In the following sections the literature on the dose-related effects of different steroids on HPA axis as well as on local adverse effects is discussed. Studies included in the systematic review Of the 14 studies included in this review, in 8 the effects on HPA-axis suppression were analysed. No data on the effects of BDP, budesonide or TAA on HPA-axis were reported. Six of the 7 randomised, double-blind dose-response studies with FP also analysed its effect on HPA axis, measuring either basal morning cortisol levels, post-cosyntropin stimulation test levels or urinary excretion of cortisol metabolites ( Table 2b, see Additional file 1 ). Only one study reported a statistically significant dose-response effect (3% decrease per doubling dose of FP) in morning plasma cortisol levels [ 20 ] and one study [ 21 ] reported slight transient reductions in urinary free cortisol and urinary 17-hydroxy steroids in the group receiving the highest dose of FP (1000 μg/d). However, in 5 studies made with FP, no dose-related effects on HPA-axis suppression were described ( Table 2b, see Additional file 1 ). There was no indication for the dose-dependent HPA-axis suppression in 2 studies with mometasone furoate. One needs to note that these studies were not planned and powered to detect differences in systemic or adverse effects. Beclomethasone dipropionate – other literature The dose-related effects of HFA-BDP (200–800 μg/d) were studied in 43 steroid-naïve asthmatic patients in a randomized double-blind fashion for 14 days [ 48 ]. When the HFA-BDP dose increased a greater decrease in the percent change from baseline in steady state 24 h urinary free cortisol was found suggesting a dose-response. Despite the observed statistically significant differences between placebo and the two highest dose-groups in mean percent change in 24 h urinary free cortisol, only one patient among all the treatment groups fell below the reference range for this parameter. In another small, randomized study 26 steroid-naïve asthmatic patients were treated with increasing doses of BDP (400 – 1600 μg/d) [ 49 ]. Only the highest dose of BDP produced a significant suppression of 24 h urinary free cortisol. In a recent Cochrane review [ 10 ], the dose-response relationship of BDP on HPA axis function was analysed. Only two small studies with adult patients not treated with oral steroids were identified, and showed no effect on morning plasma cortisol by two to five-fold increase in the BDP dose. Budesonide – other studies A randomized double-blind study with consecutive dose design [ 39 ] comparing FP (500–2000 μg/d) and budesonide (800–3200 μg/d) reported that budesonide, but not FP (or at least to a lesser extent) reduced 24 h urine cortisol excretion, plasma-cortisol and serum osteocalcin in a dose-related manner. Similar results have been reported from an open, randomized, parallel group trial with budesonide at doses of 400, 800, 1600 and 2400 μg/d for 2 weeks at each dose level, in adult patients with moderate to severe asthma [ 41 ]. Budesonide decreased the 24 h urinary cortisol excretion, serum cortisol and osteocalcin in a dose-dependent manner. In a randomized, double-blind parallel-group study [ 33 ], budesonide (1600 μg/d for 12 weeks) induced a mean change from baseline in synthetic corticotrophin (cosyntrophin)-stimulated plasma cortisol levels that was significantly different from placebo and the lowest dose of budesonide. However, the difference from placebo was only 10%, and all other doses of budesonide were not statistically different from placebo. In contrast, the mean basal morning plasma cortisol levels among different budesonide treatment groups and placebo did not differ. In a randomized cross-over study [ 50 ], budesonide (1600 μg/d) reduced serum osteocalcin and blood eosinophil count as compared with placebo, but these effects were not dose-dependent. In contrast, budesonide (400–1600 μg/d) had no significant effects on adrenal function as assessed by 8 am serum cortisol or overnight urinary cortisol excretion. In a recent open study, budesonide (400–1600 μg/d) was given to patients with mild to moderate asthma (n = 26) sequentially for 3 weeks each dose, a total of 9 weeks [ 38 ]. There was a significant dose-related suppression of morning cortisol levels and overnight urinary cortisol values, but not of serum osteocalcin. For example, the percentages of patients with a stimulated plasma cortisol response less than 500 nM were 7% at baseline, 13% at 400 μg/d, 40% at 800 μg/d and 66% at 1600 μg/d. The authors reported that the proportions of patients with a beneficial airway response together with a minimal systemic response – that is, an optimal therapeutic index – were approximately 50% at all three doses of budesonide. However, the proportion of patients with a good airway response together with a marked systemic response – that is, a suboptimal therapeutic index – increased from 4% at low dose to 38% at high dose [ 38 ]. In a recent Cochrane meta-analysis, statistically significant, dose-dependent suppression by budesonide of 24 hour urinary free cortisol excretion and serum cortisol post synthetic ACTH infusion over the dose range 800 – 3200 μg/d were apparent, but the authors concluded that the clinical significance of these findings is unclear [ 12 ]. Fluticasone propionate – other literature FP has also been shown to suppress 8 am serum cortisol and urinary cortisol/creatinine ratio in a dose-dependent manner in a single-blind placebo-controlled cross-over study for 9 days in patients (n = 12) with mild to moderate asthma [ 51 ]. Similar dose-dependent suppression of adrenocortical activity was reported in four other studies with patients with mild to moderate asthma from the same research group [ 52 - 55 ]. Interestingly, the suppressive effects of FP on adrenocortical activity were greater than those observed on osteocalcin or eosinophils. A Cochrane review [ 11 ] collected data on the effects of FP on HPA-axis function. Significant differences were not apparent between any daily dose of FP in the range of 100–1000 μg/d and placebo on basal plasma cortisol values or urinary cortisol excretion. However, the authors were not able to make a meta-analysis of the cortisol values. In another Cochrane review [ 32 ] the same authors found no evidence for dose-dependent suppression of HPA function. However, no decent meta-analysis could be made due to limited availability of data. In contrast to these findings another meta-analysis [ 47 ] found that FP exhibits a significantly steeper dose-related systemic bioavailability than BDP, budesonide, or triamcinolone when 21 studies of urinary cortisol levels and 13 studies of suppression of 8 am plasma cortisol levels were analysed. Thus, there clearly exists a discrepancy in the published literature concerning the systemic effects of FP. Based on the recent Cochrane review and meta-analysis [ 32 ] it seems obvious that there is a dose-response relationship in the appearance of local side-effect hoarseness and/or dysphonia so that FP at doses of 400–500 μg/d and 800–1000 μg/d has a significantly higher risk than at lower doses (50–100 μg/d). Similarly FP at doses of 50–100 μg/d induces significantly less oral candidiasis than at doses of 800–1000 μg/d. However, there seemed to be no significant difference in the incidence of sore throat/pharyngitis between any of the FP doses. Another systematic review [ 16 ] collected data from fluticasone studies and calculated NNT (number needed to treat) to prevent worsening of asthma and NNH (number needed to harm) to induce oral candidiasis. Three patients needed to be treated with fluticasone 100 μg/d to prevent worsening of asthma (NNT 3), and for fluticasone 1000 μg/d the NNT was 2.1 patients. In contrast, the dose-response curve for side effects was steep. For a dose of fluticasone 100 μg/d, oral candidiasis developed in one of every 90 subjects treated (NNH 90), whereas the NNH for fluticasone 1000 μg and 2000 μg daily were 23 and 6, respectively. Triamcinolone acetonide – other literature In two randomized studies, TAA in the dose range of 400–1600 μg/d [ 50 , 51 ] did not significantly affect 8 am serum cortisol or the 24 h or overnight urinary excretion of corticosteroid metabolites. In an open non-controlled 6 months study with 400–800–1600 μg/d TAA the plasma cortisol levels before and after cosyntrophin injection were analysed in patients with asthma [ 56 ]. Although all treatment regimens caused some reduction in the 24 h excretion of corticosteroid products, none of the mean values was below the normal ranges and no significant suppression in the cosyntrophin test was seen. The mean data indicated that TAA had overall no significant effect on adrenal function at any dose or at any time. However, three patients exhibited some reduction in adrenal function. In another small, randomized study 26 steroid-naïve asthmatic patients were treated with increasing doses of TAA (800 – 3200 μg/d) [ 49 ]. Only the highest dose of TAA produced a significant suppression of 24 h urinary free cortisol. Conclusions on the effects of ICS on HPA axis and local side effects Taken together, the data on the systemic adverse effects of ICS is conflicting and seems also to reflect the study design. Several studies have measured only the basal morning cortisol levels or levels after stimulation with high cosyntrophin doses. However, these may be insensitive markers for HPA-axis suppression [ 47 ]. Different, a possibly more sensitive endpoint could be plasma cortisol profile during 20–24 h period, which has been shown to be affected by a short course of fluticasone and/or budesonide or even after single inhaled doses [ 57 - 59 ]. There is disagreement between the relative potency of budesonide and FP on HPA-axis function. In addition to the different ways to measure HPA-axis function, this may be due to the use of different inhalers, duration of the treatment period, the selection of the patient group or different design and sponsoring of the studies by pharmaceutical companies. In addition there are differences in the delivery of ICS between normal subjects and patients with asthma and in patients with severe versus mild asthma [ 60 - 62 ]. Although generally safe, it appears that there is at least some degree of dose-dependency in the HPA-axis effects of inhaled steroids. Some smaller studies [ 39 , 41 , 54 ] suggest that there is a significant decrease in the therapeutic index with higher doses of ICS. Recently, a statistical meta-analysis using regression was performed for parameters of adrenal suppression in 27 studies [ 47 ]. Marked adrenal suppression, and thus a marked risk for systemic adverse effects, occurs at doses of ICS above 1500 μg/d (budesonide and BDP) or 750 μg/d (FP), although there is a considerable degree of inter-individual susceptibility. Meta-analysis showed significantly greater potency for dose-related adrenal suppression with FP compared with BDP, budesonide, or TAA. The author concludes that ICS in doses above 1500 μg/d (750 μg/d for FP) may be associated with a significant reduction in bone density [ 47 ]. Long-term, high-dose ICS exposure increases the risk for posterior subcapsular cataracts, and to a much lesser degree, the risk for ocular hypertension and glaucoma. Skin bruising, which correlates with the degree of adrenal suppression, is most likely to occur with high-dose exposure [ 47 ]. Adding a long acting-β 2 -agonist (LABA) The rationale LABA provide long-lasting relaxation of airway smooth muscle, while the ICS provide potent topical anti-inflammatory action. In addition to these complementary actions, β 2 -agonists may have several other actions that may contribute to their efficacy in relieving asthma symptoms. β 2 -Agonists inhibit plasma exudation in the airways by acting on β 2 -receptors on postcapillary venule cells. They inhibit the secretion of bronchoconstrictor mediators from airway mast cells and may inhibit release of mediators from eosinophils, macrophages, T-lymphocytes and neutrophils. In addition, β 2 -agonists may have an inhibitory effect on the release of neuropeptides from sensory nerves [ 63 ]. Corticosteroids may also increase the expression of β 2 -receptors in inflammatory cells to overcome the desensitisation in response to chronic β 2 -agonist exposure [ 64 ]. In addition, LABA may prime the glucocorticoid receptor facilitating activation by corticosteroids [ 65 , 66 ]. Design of 12 LABA add-on studies included in the review The literature search identified 3 studies with formoterol [ 67 - 69 ] and 9 studies with salmeterol [ 70 - 78 ]. All these studies included adult or adolescent patients with symptomatic asthma. Generally, patients used low to moderate doses of inhaled glucocorticoids. In two studies [ 68 , 73 ] previous use of ICS was not required. In all studies PEF or FEV 1 reversibility of at least 10–15% was required ( Table 3, see Additional file 1 ). Diurnal or period PEF variation >15% was required in four studies. FEV 1 of >(40)–50% of predicted and a clearly positive symptom score was required in most studies ( Table 3, see Additional file 1 ). In general, the mean FEV 1 (% predicted) varied between 61 and 87% in different studies, being 61–70% in 4 studies, 70–80% in 3 studies, 81–87% in two studies and was not reported in three studies. The mean absolute PEF values varied from 299 to 404 L/min and FEV 1 from 2.12 to 2.54 L ( Table 5, see Additional file 1 ). Thus, the patient population in these studies represents mainly those with moderate to severe persistent asthma. This as well as the fact that patients with recent exacerbations are excluded may produce a selection bias, compared with the real life. In one study [ 78 ] patients were required to have at least two exacerbations during the previous year to be eligible for the inclusion in the study. One study [ 68 ] was performed in patients mainly affected with mild persistent asthma. In salmeterol and formoterol studies, the comparison dose of ICS was increased 2–2.5 (-4)-fold, whereas in the formoterol study [ 67 ] the comparison dose of budesonide was 4-fold higher ( Table 4, see Additional file 1 ). Another significant difference between formoterol and salmeterol studies is that in the formoterol [ 67 ] study the main outcome parameter was the incidence of exacerbations whereas the salmeterol studies mainly focused on lung function and asthma symptoms. Most studies allowed a constant dose of theophylline but not oral steroid use ( Table 3, see Additional file 1 ). Six out of the 12 studies excluded patients having previous exacerbations (generally during previous month). Only 2 studies lasted one year [ 67 , 68 ], whereas most studies lasted at least 24 weeks. Most reports did not identify whether the study were performed by respiratory specialists or general practitioners. All studies were financially supported by pharmaceutical companies. Lung function and asthma symptoms Formoterol – studies included in this systematic review The addition of formoterol was compared with the increase (4-fold) in the dose of inhaled budesonide (from 200 μg/d to 800 μg/d) in patients with moderate to severe symptomatic chronic asthma [ 67 ]. The patients (n = 852) in this study had a FEV 1 of at least 50% of predicted (mean 75–76%) with an increase in FEV 1 ≥15% after inhalation of terbutaline. Addition of formoterol was superior to the increase in steroid dose in increasing FEV 1 and morning PEF (Figure 3A ; Table 5, see Additional file 1 ). Similarly, addition of formoterol was equal or superior to the 4-fold increase in ICS dose in reducing day- or night-time symptom scores or rescue medication use ( Table 6, see Additional file 1 ). Most importantly, the effect of formoterol was sustained over the one-year treatment period. In this study, no statistical comparison was made between the low-dose budesonide + formoterol and high dose budesonide groups. Figure 3 Formoterol add-on study showing forced expiratory volume in one second (FEV 1 ) (panel A, from ref 64 with permission) and the estimated yearly rates (no. patients/year) of severe asthma exacerbations in the different treatment groups of the study (panel B). For estimated yearly rate of exacerbations, the P-values given were formoterol vs placebo P = 0.01 and lower vs higher dose of budesonide P < 0.001. Another study [ 69 ] compared the addition of formoterol (4.5 μg bid) to a small dose of budesonide (160 μg/d) in single inhaler (Symbicort ® ) with an increased dose of budesonide (400 μg/d) in adults with mild to moderate asthma (mean FEV 1 81–82%) not fully controlled on low doses of ICS alone. The increase in mean morning and evening PEF was significantly higher for budesonide/formoterol compared with budesonide alone. In addition, the percentage of symptom-free days and asthma control days were significantly improved in the budesonide/formoterol group. Budesonide and formoterol decreased the relative risk of an asthma exacerbation by 26% as compared with higher dose budesonide alone. The results of the formoterol study [ 67 ] on the benefits of addition of formoterol were confirmed in patients with mild asthma (mean FEV 1 86–87% of predicted and using approximately 1 rescue inhalation per day) [ 68 ]. In this study, the addition of formoterol was superior to doubling the dose of budesonide in increasing FEV 1 and morning PEF in the patients already treated with a low dose of ICS, but not in steroid-naïve patients ( Table 5 ), or in reducing the percentage of days with symptoms, number of rescue inhalations or nights with awakenings in the patients with mild persistent asthma already treated with low doses of ICS ( Table 6, see Additional file 1 ). A subgroup of the patients participating in the formoterol study [ 67 ] was analysed for asthma quality of life parameters using the Asthma Quality of Life Questionnaire (AQLQ) [ 79 ]. Following randomisation there was a significant increase in the AQLQ score only in the group with higher budesonide + formoterol group. Although the patterns of mean responses for AQLQ scores and for the clinical variables were very similar, correlations between change in AQLQ scores and change in clinical measures over the randomized period were only weak to moderate (maximum r = 0.51). The data confirm that the benefit from the addition of formoterol is sustained. However, instead of improving pulmonary function parameters patients are usually more interested in how their normal everyday life and activities are limited by the disease. The analysis of AQLQ parameters and their comparison with the clinical data in that analysis also suggest that if only pulmonary function parameters are to be analysed, the benefits of addition of LABA to the treatment may be over-estimated. Also, it should be noted that no correlation has been found between measures of pulmonary function and daytime asthma symptoms [ 80 ]. Formoterol – other literature As compared with the abovementioned three studies, similar superiority of addition of formoterol on morning PEF, rescue medication use and asthma symptoms were reported in an open randomised parallel-group study comparing the addition of formoterol to the low-dose BDP with 2-fold higher dose of BDP in patients suffering from symptomatic asthma, despite the use of inhaled BDP [ 81 ]. Salmeterol – studies included in this systematic review Addition of salmeterol as compared with the increase in the dose of ICS BDP or FP has been studied in 9 randomised parallel group studies with 3651 patients with moderate to severe persistent asthma ( Tables 3 and 4, see Additional file 1 ). Addition of salmeterol improved FEV 1 better than increasing the dose of ICS 2–4-fold in 5 studies (analysed in 6 studies) and mean morning PEF in 7 studies (analysed in 9 studies), respectively ( Table 5, see Additional file 1 ). Similarly, addition of salmeterol was significantly better than the increase in the dose of ICS in increasing the number of days or nights without symptoms or without rescue medication or reducing day- or night-time symptom score as well as daytime or night-time rescue medication use in most studies ( Table 6, see Additional file 1 ). However, although addition of salmeterol seems to be superior to increased dose of ICS, a statistically significant difference was not always reached ( Tables 5 and 6, see Additional file 1 ) in the single studies when FEV 1 , morning PEF, asthma symptom scores or rescue medication use were analysed. Another feature typical of these studies is that the results favour the addition of salmeterol more at early time points and this difference is reduced as the study proceeds. Salmeterol – other literature Most of the studies mentioned above, (except ref [ 72 ]), have recently been analysed in a meta-analysis [ 13 ]. In addition, the published meta-analysis included 1 study (n = 488) that remains unpublished at the present. At baseline these patients (n = 3685, aged ≥12) used BDP 200 – 400 – 1000 μg/d or FP 200 – 500 μg/d. The addition of salmeterol to those doses was compared with increasing the dose of BDP or FP up to 2–2.5-fold. The mean FEV 1 was <75% in most studies included in the meta-analysis and a reversibility of ≥10–15% in PEF or FEV 1 after inhalation of short-acting bronchodilator was required for inclusion in all but three studies. In patients receiving salmeterol the morning PEF was 22–27 L/min greater and FEV 1 was 0.10 – 0.08 L greater after three to six months of treatment, compared to the response to increased steroids. Similarly, the mean percentage of days and nights without symptoms was increased 12–15% and 5%, respectively, as well as the mean percentage of days and nights without need for rescue treatment increased 17–20% and 8–9%, respectively. Effect of LABA on asthmatic inflammation The results of the above mentioned studies favour the addition of a LABA instead of increasing the dose of ICS in patients not adequately controlled with low to moderate doses of ICS. However, there have been concerns that regular use of inhaled β 2 -agonists may mask an increase in the underlying airway inflammation in asthma. Also, some proinflammatory effects have been described for β 2 -agonists such as delay of constitutive eosinophil apoptosis [ 82 ] or reversal of corticosteroid-induced apoptosis [ 83 ]. Furthermore, development of tolerance to their protective effects against various asthma-provoking stimuli has been reported. There is some disagreement whether the addition of formoterol or salmeterol changes the level of pulmonary inflammation in patients already treated with inhaled glucocorticoids or whether they may even mask the inflammation. Three studies [ 84 - 86 ] do not indicate any significant increase in the inflammatory indices following addition of formoterol or salmeterol, whereas treatment of asthma with salmeterol with concomitant steroid tapering has been shown to increase the numbers of eosinophils in sputum [ 87 ]. Formoterol – studies included in this systematic review In a randomised, double-blind and parallel-group study (n = 61) with similar inclusion and exclusion criteria than in the formoterol add-on study [ 67 ], the effect of adding formoterol (12 μg bid) to a low dose of budesonide (200 μg/d) was compared with a higher dose of budesonide (800 μg/d) for 1 year after a run-in with budesonide (1600 μg/d) for 4-wk [ 84 ]. Budesonide (1600 μg/d) during run-in significantly reduced median sputum eosinophils. No significant changes in the proportion of eosinophils, other inflammatory cells, or ECP levels in sputum were observed over the ensuing one year treatment with formoterol + budesonide (200 μg/d) or higher dose budesonide (800 μg/d). Clinical asthma control was not significantly different between both groups. Salmeterol – other literature In a small study (n = 9) with asthma patients using regular inhaled glucocorticoids and inhaled salbutamol for symptom relief, the addition of salmeterol for 8 weeks was studied in a double-blind crossover placebo-controlled protocol [ 86 ]. Bronchoalveolar lavage (BAL) cell profile, albumin and tryptase levels, percentages of CD4 + and CD8 + lymphocytes and lymphocyte activation as assessed as proportions of lymphocytes expressing HLA-DR were measured in BAL samples before and after treatment. There were no significant changes after salmeterol treatment. In another double-blind, parallel-group, placebo-controlled study [ 85 ] the effect of addition of salmeterol (50 μg bd) or fluticasone (200 μg/d) for 12 weeks was studied in 45 symptomatic patients with asthma who were receiving ICS (range 100–500 μg/d). Bronchial biopsies and BAL were analysed before and after the treatment. After treatment with salmeterol there was no deterioration of airway inflammation, as assessed by mast cell, lymphocyte, or macrophage numbers in BAL or biopsies, but a significant fall in EG1-positive eosinophils in the lamina propria was found, which was not seen after treatment with FP. The only cellular effect of added FP was a decrease in BAL lymphocyte activation as assessed as proportions of lymphocytes expressing HLA-DR. There was a concurrent improvement in clinical status, more marked with salmeterol than with increased ICS. These two studies thus suggest that adding salmeterol to ICS is not associated with increased airway inflammation. In another study in 13 asthmatic individuals requiring ≥1500 μg ICS daily, the steroid sparing and "masking" effects of salmeterol versus placebo were studied in a randomised, placebo-controlled, double-blind and crossover trial [ 87 ]. Subjects were re-stabilised on their original dose of ICS for 4 wk before crossover to the alternative treatment. Corticosteroid doses were reduced weekly until criteria were met for an exacerbation or the corticosteroid was fully withdrawn. Mean ICS dose was reduced significantly more (87%) during salmeterol treatment, than with placebo (69%). Sputum eosinophils increased before exacerbation, despite stable symptoms, FEV 1 and PEF. In the week before clinical exacerbation, sputum eosinophil counts were higher in the salmeterol-treatment arm as compared with placebo, whereas there were no differences in PC 20 or serum ECP. Five subjects showed >10% sputum eosinophilia before exacerbation during salmeterol treatment, compared to two receiving placebo. This suggests that the use of salmeterol allowed subjects to tolerate a greater degree of inflammation without increased symptoms or reduced lung function. Thus, during progressive reduction of ICS the bronchodilator and symptom-relieving effects of salmeterol may mask increasing inflammation and delay awareness of worsening asthma. These findings strengthen guideline recommendations that LABA should not be described as sole anti-asthma medication and that they should be used as "add-on" therapy rather than for steroid tapering purposes. The effect of addition of salmeterol (50 μg bd), FP (200 μg/d) or placebo for 3 months on airway wall vascular remodelling has been studied in 45 symptomatic patients with asthma who were receiving treatment with ICS (range 400–1000 μg/d) [ 88 ]. Bronchial biopsies were analysed before and after treatment. There was a decrease in the density of vessels of lamina propria after treatment only in the salmeterol group compared to baseline. There was no significant change within the FP or placebo groups and no treatment was associated with increased airway wall vascularity. Asthma exacerbations If there were a marked masking of pulmonary inflammation by LABA, one would expect to see an increase in the number and severity of asthma exacerbations during their long-term use. There is some difficulty in comparing the different studies done with formoterol and salmeterol as the definition of exacerbation varies. In formoterol studies [ 67 , 68 ] a severe exacerbation was defined as need for treatment with oral corticosteroids, as judged by the investigator, or hospital admission or emergency treatment for worsening of asthma or a decrease in morning PEF >25%–30% from baseline on two consecutive days. In contrast, in the salmeterol "add-on" studies the exacerbation was not defined at all or was more loosely defined for example as "a clinical exacerbation", "any worsening of asthma symptoms requiring a change in prescribed therapy, other than increased use of rescue medication" or "any asthma event that required treatment with oral or parenteral steroids". Formoterol – studies included in this systematic review In the formoterol study [ 67 ] the main outcome parameter was the rate of exacerbations during combination therapy. The results show that the 4-fold increase in the dose of budesonide reduced the rates of severe and mild exacerbations by 49% and 37%, respectively, whereas addition of formoterol to the lower dose of budesonide reduced the rates of severe and mild exacerbations by 26% and 40%, respectively. Patients treated with formoterol and the higher dose of budesonide had the greatest reductions, 63% and 62%, respectively (Figure 3B ; Table 7, see Additional file 1 ). This suggests that if frequent asthma exacerbations are a major problem, increasing the dose of ICS may help to reduce the number of exacerbations. The results of the formoterol study [ 67 ] as well as the salmeterol meta-analysis [ 13 ] suggest that addition of LABA has divergent effects on asthma control: it is superior to the increased steroid dose in improving lung function, but is equal or less efficient in reducing exacerbations (Figure 3AB ). The data also suggest that to achieve a better control of asthma exacerbations, the dose of ICS should be increased 4-fold. When 425 exacerbations of the formoterol study [ 67 ] were analysed [ 89 ], the use of higher dose of ICS or the use of formoterol was shown not to affect the pattern of change in PEF values or in symptoms during asthma exacerbation (Figure 4B ). Figure 4 A. Change in supplemental salbutamol use before and after exacerbation in patients treated with fluticasone and salmeterol combination or with high-dose fluticasone (with permission from ref 90), B. Change in morning PEF (percent fall from day -14) over the 14 d before and 14 d after an exacerbation in relation to treatment as analyzed from a subgroup of a FACET study (with permission from ref 89). In contrast to that described in moderate to severe asthma, in the other formoterol study [ 68 ] addition of formoterol (6 μg bid) to either the lower (200 μg/d) or higher (400 μg/d) dose of budesonide in patients suffering from mainly mild asthma reduced the risk of the first asthma exacerbation by 43% (RR = 0.57, 95% CI 0.46–0.72). There was also a significant 52% reduction in the rate of severe exacerbations (RR = 0.48; 95% CI 0.39–0.59). In addition, significant improvement was observed for the rate of severe exacerbations (RR = 0.58, 95% CI 0.44–0.76). Thus, the data suggest that there may be a difference in the effect of ICS and formoterol on the exacerbations between mild and moderate to severe asthma so that in mild asthma addition of LABA may be more efficient in preventing exacerbations, whereas in moderate to severe asthma increasing the dose of ICS may be more efficient ( Table 7, see Additional file 1 ). However, the formoterol studies [ 67 , 68 ] are not fully comparable in that way that in the other study [ 67 ] the increase in the dose of budesonide was 4-fold whereas in the other study [ 68 ] it was 2-fold. Another study [ 69 ] compared the addition of formoterol (4.5 mg/d) to a small dose of budesonide (160 μg/d) in single inhaler (Symbicort ® ) with an increased dose of budesonide (400 μg/d) in adults with mild to moderate asthma (mean FEV 1 81–82%) not fully controlled on low doses of ICS alone. Budesonide/formoterol combination significantly decreased the relative risk of an asthma exacerbation by 26% as compared with higher dose budesonide alone. In contrast, the estimated risk of having a severe exacerbation was 6% lower in patients treated with budesonide/formoterol compared with those receiving budesonide alone, but this was not statistically significant. Salmeterol – studies included in this systematic review Only two studies [ 70 , 78 ] of those included in this systematic review reported the actual monthly or annual rates for moderate or severe exacerbations. In those studies there were no significant differences in the yearly rate of exacerbations or percentages of patients experiencing at least exacerbation. The other studies generally reported the percentages of patients experiencing at least one exacerbation (Table 7). In salmeterol studies, the data were presented mostly in a form, which did not allow us to calculate the yearly rate of exacerbations. Salmeterol – other literature In the salmeterol studies lasting 3–6 months the numbers of patients with exacerbations were analysed. The meta-analysis [ 13 ] revealed that fewer patients experienced any exacerbation with salmeterol (difference 2.7%), and the proportion of patients with moderate or severe exacerbations was also lower (difference 2.4%). Thus, to prevent one exacerbation 37–41 patients should be treated with salmeterol instead of increasing the dose of ICS. Rather than indicating salmeterol being superior, the result suggests that there is no increased risk for exacerbations with the use of salmeterol. Unfortunately, in most salmeterol studies the severity and/or yearly incidence of exacerbations was not analysed. As one patient can experience more than one asthma exacerbation during the study, the parameter used in the salmeterol studies (proportion of patients experiencing an exacerbation) may not reflect the actual number of exacerbations. Another factor that may affect our interpretation of the effect of these therapies on asthma exacerbations is that in 6 of the 12 LABA studies, patients could be withdrawn from the study if they experienced >1–5 exacerbations ( Table 7, see Additional file 1 ). This may underestimate the total incidence of exacerbations, as those patients experiencing several exacerbations were excluded from analysis. However, these are the patients the "add-on" therapies are most frequently prescribed. Recently, the exacerbation rates and clinical measures of asthma worsening were assessed in an analysis combining results from two double-blind studies (n = 925) comparing addition of salmeterol to low-dose-FP with increasing the dose of FP 2.5-fold [ 90 ]. The addition of salmeterol resulted in a significantly lower rate (0.23 vs. 0.39 per patient per year) of exacerbations compared with higher dose FP. Salmeterol combined with low-dose FP was significantly more protective than 2.5-fold higher dose of FP in preventing asthma exacerbations, as assessed by the time to first exacerbation. In both groups clinical indicators of worsening of asthma showed parallel changes before asthma exacerbation, and greater improvements in morning PEF, supplemental salbutamol use and asthma symptom score were observed after exacerbation with salmeterol compared with higher dose FP (Figure 4A ). Thus, the ability to detect deteriorating asthma and the severity of exacerbation is not negatively affected by salmeterol. Adverse effects of LABA The addition of LABA to the treatment regimen usually results in a slight increase in those pharmacologically predictable adverse events such as tremor and tachycardia. However, generally these do not lead to the discontinuation of the treatment. In the formoterol studies [ 67 - 69 ], no significant differences were reported on the adverse effects between the groups, but no detailed data was presented. Also, in the salmeterol studies [ 70 - 78 ], the incidence of adverse events was very low and generally was not different between the treatment groups. Although LABA appear to be generally very safe, one should not forget that they are generally not suitable for patients with symptomatic coronary heart disease or hyperthyroidism and may provoke more severe adverse events such as supraventricular tachycardias, atrial fibrillation and extrasystoles. Rarely hypersensitivity reactions and painful muscular cramps may occur. Also one should note that the "add-on" studies included in this review are not originally planned and powered to detect significant differences in the adverse effects. Adding a leukotriene receptor antagonist (LTRA) Rationale Cysteinyl leukotriene receptor-antagonists (LTRA), such as montelukast, pranlukast and zafirlukast, are a new class of asthma medication, whose role in the stepwise management of asthma has not yet been fully established. Leukotriene antagonists blunt the obstructive response and have weak anti-inflammatory activity. In some studies corticosteroids are not very effective inhibitors of cysteinyl leukotriene pathways, at least when assessed by their inability to reduce cysteinyl leukotriene concentrations [ 91 , 92 ] and thus combination of these therapeutic classes may offer some benefit. Montelukast – studies included in this systematic review We identified one randomised, double-blind, parallel-group 16 week study (Jadad score 3) comparing the addition of montelukast (10 mg/d) to budesonide (800 μg/d) with doubling the dose of budesonide (1600 μg/d) in patients inadequately controlled on inhaled budesonide (800 μg/d, n = 448) [ 93 ]. The inclusion criteria were: patients (aged 15–75 years) who were not optimally controlled as judged by the investigators in spite of a regular ICS (600–1200 μg/d for BDP, budesonide, TAA, flunisolide or 300–800 μg/d for FP). Patients were required to have FEV 1 ≥50% predicted at visits 1 and 3, with a ≥12% bronchodilator response and symptoms requiring β-agonist treatment of at least 1 puff/day during the last 2 weeks of the run in period (total 4 weeks). Both groups showed progressive improvement in several measures of asthma control compared with baseline. Mean morning PEF improved similarly in the last 10 weeks of treatment compared with baseline in both the montelukast + budesonide group and in the double dose budesonide group (33.5 vs 30.1 L/min). The improvement in montelukast + budesonide group was faster as the mean morning PEF was significantly higher during days 1–3 after start of treatment in this group as compared with the double dose budesonide group (20.1 vs 9.6 L/min) (Figure 5 ). Both groups showed similar improvements with respect to rescue β 2 -agonist use, mean daytime symptom score, nocturnal awakenings, exacerbations, asthma free days, peripheral blood eosinophil counts, and asthma specific quality of life. The authors conclude that addition of montelukast to ICS offers comparable asthma control to doubling the dose of ICS. However, it needs to be remembered that, in most cases, to obtain a statistically significant improvement in asthma control at least a 4-fold increase in the dose of ICS is needed (see above). Figure 5 Effect of addition of montelukast (10 mg/d) or doubling the dose of ICS on morning peak expiratory flow (AM PEF) over 12 week treatment period in patients not adequately controlled by budesonide 800 μg/d (solid line = montelukast + budesonide 800 mg/d, dashed line = budesonide 1600 μg/d) (with permission from ref 93). Montelukast – other literature A large (n = 639) study [ 94 ] recruited patients with asthma not optimally controlled by ICS (stable dose equivalent to budesonide 400–1600 μg/d). The patients were required to have FEV 1 ≥55%, a bronchodilator response greater than 12%, symptoms and rescue β 2 agonist use of at least 1 puff/day. The mean FEV 1 at baseline was 81% predicted. The patients were randomised to obtain either montelukast (10 mg/d) or placebo in a double-blind manner. The ICS dose remained constant throughout the study. The primary efficacy end point was the percentage of asthma exacerbation days. The major advantage of this study is that this study adopted several different definitions for asthma exacerbation days from previously published other studies, making comparison to other studies more easy. The median percentage of asthma exacerbation days was 35% lower (3.1% vs 4.8%, p = 0.03) and the median percentage of asthma free days was 56% higher (66.1% vs 42.3%, p = 0.001) in the montelukast group than in the placebo group. Thus, the NNT with montelukast to avoid one exacerbation day was 13, and the NNT to avoid one day not free of asthma – that is, to gain an asthma free day – was 10. Patients receiving concomitant treatment with montelukast had significantly less (25.6% vs 32.2%, p = 0.01) nocturnal awakenings, and significantly greater reductions in β 2 -agonist use (17.26% v 4.92%, p = 0.05, baseline use was 3.2–3.3 puffs/day), and morning PEF (16.86 L/min vs 11.30 L/min, p = 0.05, baseline 365–373 L/min). No significant difference was found in asthma specific quality of life or in morning FEV 1 . The results of this study suggest that although the effect of montelukast on endpoints such as morning PEF, FEV 1 and rescue β 2 -agonist use are only small or modest, addition of montelukast may produce a significant improvement of asthma control by reducing the number of asthma exacerbation days. In another study with patients (n = 642) with symptomatic persistent asthma despite the treatment with BDP (400 μg/d), addition of montelukast (10 mg/d), improved morning FEV 1 and PEF, asthma symptom score and the percentage of asthma exacerbation free days better than placebo during 16 week treatment period [ 95 ]. The increase in morning FEV 1 was approximately 140 mL and in morning PEF 10 L/min. There was a tendency towards reduced rescue medication use with the combination therapy, but the reduction was only 0.2 puffs/day. Addition of montelukast to ICS seemed to prevent the increase in the number of peripheral blood eosinophils seen in other treatment groups. In an atypical "add-on" study (randomised double-blind, placebo-controlled and crossover trial), addition of montelukast (10 mg/d) was compared with placebo in patients with asthma (n = 72) and symptoms despite treatment with ICS and additional therapy [ 96 ]. Most of the patients used several different types of combination therapy, except leukotriene antagonists, at baseline. The inclusion criteria were defined as "any patient with physician diagnosis of asthma in whom the recruiting physician felt a trial of montelukast was indicated for continued asthma symptoms despite other anti-asthma therapy". A current worsening of asthma requiring oral corticosteroid treatment, or worsening in the preceding month were both exclusion criteria, but did not exclude any of those referred for inclusion in the trial. In this setting corresponding to a typical hospital outpatient clinic, addition of montelukast did not result in any significant change in symptom scores, rescue inhaled β 2 -agonist use, or morning or evening PEF. When treatment response was defined as a 15% or greater increase in mean PEF recordings, there were four responders to montelukast and seven responders to placebo. Although several points in this study may be criticised (loose inclusion criteria, small sample size, short 2 week treatment period, no wash-out period, encapsulation of the tablets, exacerbations not analysed as end-point), the results suggest that the effects of montelukast are not as evident in unselected population than in the more clearly defined patients included in other trials [ 93 - 95 ]. The additional anti-inflammatory activity obtained by adding montelukast to the treatment regimen has been assessed in three randomised, double-blind, cross-over studies lasting 10 days–8 weeks. In one study [ 97 ], addition of montelukast (10 mg/d) to salmeterol (50 μg bid) and fluticasone (250 μg bid) combination was compared with placebo in patients with mild-moderate asthma for 3 weeks. Compared with salmeterol/fluticasone run-in period, adding montelukast was better (p < 0.05) than placebo for inflammatory markers such as AMP-threshold, recovery, exhaled NO, and blood eosinophils but not for lung function. In another study [ 98 ], addition of montelukast for 8 weeks to FP (100 μg bid) was compared with placebo in patients with mild asthma. There were no differences in FEV 1 or histamine PC 20 between the two treatment regimens. There was no difference in the efficacy of either treatment in decreasing T cell, CD45RO+, mast cell or activated eosinophil numbers in bronchial biopsies. In a third study [ 99 ], the addition of montelukast (10 mg/d) to budesonide (400 μg/d) for 10 days to steroid-naïve patients with asthma was reported not to produce any additional anti-inflammatory benefit when compared with budesonide alone in reducing airway hyperresponsiveness or sputum eosinophilia. Zafirlukast – other studies Addition of high-dose zafirlukast (80 mg b.i.d.: 4-fold greater than the approved dose) improved asthma control better than placebo in patients (n = 368) on high-dose ICS (1000 – 4000 μg/d) [ 100 ]. Compared with placebo, addition of zafirlukast improved morning and evening PEF and reduced daytime symptom score and rescue medication use [ 100 ]. According to a recent meta-analysis [ 101 , 102 ], in symptomatic asthmatic adults, addition of zafirlukast (80 mg bid) to ICS did not reduce the risk of an exacerbation requiring systemic steroids after 12 weeks of treatment, compared to double dose ICS [RR = 1.08; 95% CI 0.47, 2.50]. There were no differences in any other measure of outcome. Higher doses of zafirlukast than currently licensed were associated with increased risk of liver enzyme elevation. Conclusions on adding a LTRA According to recent meta-analyses (12 adult studies and 1 in children) [ 101 , 102 ], leukotriene antagonists (zafirlukast or pranlukast at 2–4 times the licensed dose) combined with ICS (300–2000 μg/d BDP equivalent) reduce the number of patients with exacerbations that require systemic corticosteroids, compared to ICS alone [RR = 0.34; 95% CI 0.13, 0.88]. This equates to 20 patients (95% CI 1,100) treated to prevent one needing systemic corticosteroids. There was no difference in side effects [ 101 , 102 ]. The addition of licensed doses of LTRA to ICS resulted in a non-significant reduction in the risk of exacerbations requiring systemic steroids (two trials, RR 0.61, 95% CI 0.36, 1.05). This systematic review did not include the recent study comparing the addition of montelukast to double-dose ICS [ 93 ]. As that systematic review did not include any data of LTRA drugs at currently licensed doses compared with high dose ICS, the author came to a conclusion that the addition of LTRA to ICS may modestly improve asthma control compared with ICS alone but this strategy cannot be recommended as a substitute for increasing the dose of ICS [ 101 ]. However, based on one relatively large trial [ 93 ], the evidence suggests that addition of montelukast may be equal to doubling the dose of ICS. However, one might criticise this conclusion as this study [ 93 ] lacked placebo arm, ie. it is possible that increasing (doubling) the dose of ICS does not produce any real improvement in asthma control as compared with lower ICS dose and thus the result showing non-inferiority to double dose ICS might mean no effect at all. Thus, more data is needed to compare the efficacy of LTRA at currently licensed doses with increasing the dose of ICS. Adding theophylline Rationale Although theophylline has traditionally been classified as a bronchodilator, its ability to control chronic asthma is greater than can be explained by its relatively small degree of bronchodilator activity. In fact, theophylline has immunomodulatory, anti-inflammatory and bronchoprotective effects that may contribute to its efficacy as an anti-asthma drug [ 103 ]. There is some evidence that addition of theophylline to ICS treatment improves pulmonary function and asthma symptoms [ 104 ], although all studies have not been able to confirm this result [ 105 ]. Theophylline – studies included in this systematic review The addition of theophylline has been compared with doubling the dose of ICS (BDP and budesonide; 400 μg/d → 800 μg/d) in two separate studies with 195 patients with symptomatic asthma for 6 to 12 weeks [ 106 , 107 ]. Theophylline was used at relatively low doses, the mean serum theophylline concentrations were 8.7 and 10.1 mg/L in these studies. In the study (Jadad score 4) of Evans and coworkers [ 106 ] addition of low-dose theophylline to budesonide (400 μg/d) was compared with doubling the dose of budesonide (800 μg/d) in a randomised double-blind trial for 3 months. Patients (n = 62) were required to have FEV 1 predicted normal ≥50%, bronchodilator response of at least 15% and to have symptoms despite the use of ICS (equivalent to budesonide dose of 800–1000 μg/d). The overall treatment effect of addition of theophylline was superior to double-dose budesonide in improving FVC and FEV 1 (Figure 6 ), although at single timepoints there were no significant differences between the treatments. There was no significant difference between the treatments in improving home PEF recordings or reducing β 2 -agonist use or symptom scores. There was no difference in the occurrence of possibly drug-related adverse effects between the groups. The statistical power of this study was calculated to detect significant changes over baseline, but not to detect differences (superiority) or non-inferiority between the treatments. Figure 6 Mean (+- SE) change in FEV 1 in 31 asthma patients treated with high-dose budesonide (1600 μg/d) and 31 patients given low-dose budesonide (800 μg/d) and theophylline (with permission from ref 106). A randomised, double-blind parallel-group study (Jadad score 3) by Ukena and coworkers [ 107 ] compared the addition of theophylline to low dose BDP (400 μg/d) with double-dose BDP (800 μg/d) for 6 weeks. Patients (n = 133) were required to have FEV 1 50–85% predicted normal and a documented reversibility of at least 15% of FEV 1 over baseline and to be not controlled by BDP (400 μg/d) or equivalent. The sample size of this study was powered to detect equivalence. No significant differences were found between the high-dose BDP and low-dose BDP plus theophylline groups in outcomes such as morning or evening PEF, PEF variability, FEV 1 , daytime or nighttime symptom scores or rescue medication use. Both treatments were well tolerated. Lim et al . [ 108 ] recruited asthmatic patients that were symptomatic while being treated with low dose inhaled steroids (400 μg BDP, 200 μg FP or 400 μg BDP daily). Patients (n = 155) were required to have PEF ≥50% of the predicted normal with at least 15% variability in PEF. The patients were randomised to treatment either with low dose BDP (400 μg/d) alone, theophylline plus BDP (400 μg/d) or high-dose BDP (1000 μg/d) for six months in a double-blind trial (Jadad score 5). No significant differences were found between any of the treatment groups in morning PEF, evening PEF, PEF variability, rescue β 2 -agonist use, symptom scores or in the number of exacerbations. Of note is that there were no difference between the low dose BDP alone and high dose BDP groups in any of the parameters. This study was powered to detect superiority of theophylline plus BDP as compared with high-dose BDP. There were no significant differences between the treatment groups for any of the commonly reported adverse effects. The results of this study suggest that when the benefit of an "add-on" therapy is evaluated as compared with double-dose inhaled steroid, additional group using low-dose steroid alone should be included to see whether even the doubling of the dose of steroid produces any benefit to the patient. Conclusions on the addition of theophylline Taken together, the results from two relatively small studies suggest that addition of low-dose theophylline may be equal to doubling the dose of ICS in the treatment of asthma not adequately controlled by low dose of ICS. However, one needs to remember that the effect of doubling the dose of ICS on asthma control is generally small or negligible (see above). Furthermore, a placebo group should be included in these studies to see whether an improvement in asthma control is obtained by doubling the dose of ICS. Thus, more data is needed to confirm the present results. Use of theophylline at concentrations at the lower limit or slightly below the recommended therapeutic range may help to limit the adverse effects. Comparison between LTRA, theophylline and LABA as add-on options Montelukast versus salmeterol – studies included in this systematic review Combination of fluticasone (100 μg bid) and salmeterol (50 μg bid) in a single inhaler has recently been shown to provide more effective asthma control than montelukast (10 mg daily) combined with FP (100 μg bid) in a 12 weeks study (randomised, double-blind, double-dummy, Jadad score 3) in patients (n = 447) whose symptoms were suboptimally controlled by ICS only [ 109 ]. The inclusion criteria were FEV 1 between 50% and 80% predicted normal, and at least 1 additional sign of inadequate asthma control during the 7 preceding days. Salmeterol/FP combination was superior to montelukast/FP in improving morning PEF (24.9 vs 13.0 L/min), evening PEF (18.9 vs 9.6 L/min), FEV 1 (0.34 vs 0.20 L) and shortness of breath symptom score (-0.56 vs -0.40) as well as increasing the percentage of days without rescue medication (26.3 vs 19.1%). In contrast, there was no significant difference in outcomes such as chest tightness, wheeze and overall symptom scores. Asthma exacerbation rates were significantly (P = 0.031) lower in the FP + salmeterol group (2%) than in the FP+ montelukast group (6%). Adverse event profiles were reported to be similar. A similar study [ 110 ] comparing the efficacy of combination of FP (100 μg bid) and salmeterol (50 μg bid) in a single inhaler with combination of montelukast (10 mg daily) and FP (100 μg bid) in a 12 weeks study (randomised, double-blind, double-dummy, Jadad score 4) in patients (n = 725) whose symptoms were suboptimally controlled by ICS (BDP, budesonide, flunisolide 400–1000 μg/d or FP 200–500 μg/day) only. The inclusion criteria were FEV 1 above 50% and at least 15% bronchodilator response, and asthma symptoms at least at 4/7 days during run-in. Salmeterol/FP combination was superior to montelukast/FP in improving morning PEF (36 vs 19 L/min), evening PEF (29 vs 14 L/min), FEV 1 (0.26 vs 0.17 L), percentage of symptom-free days (42.9 vs 31.5%), percentage of symptom-free nights (46.5 vs 41.1%) as well as increasing the percentage of days without rescue medication (47.9 vs 46%). In contrast, there was no significant difference in percentage of rescue free nights. The number of patients experiencing at least one asthma exacerbation (any severity) was significantly (P < 0.05) lower in the FP + salmeterol group (9.6%) than in the FP+ montelukast group (14.6%). The percentage of patients who had at least one asthma exacerbation of either moderate or severe intensity was 4.8% in the salmeterol + FP group and 8.4% in the montelukast + FP group, but this difference did not reach statistical significance. The time to the first exacerbation was significantly (P < 0.05) longer in the salmeterol + FP group than in the montelukast + FP group. Adverse event profiles were reported to be similar. Another very similar study [ 111 ] was designed to demonstrate the non-inferiority of combination of montelukast (10 mg daily) and FP (100 μg bid in dry powder inhaler) as compared with combination of FP (100 μg bid in dry powder inhaler) and salmeterol (50 μg bid; metered dose inhaler) on asthma exacerbations. This 48 weeks study (randomised, double-blind, double-dummy, Jadad score 5) included patients (n = 1490) whose symptoms were suboptimally controlled by ICS (equivalent to BDP 200–1000 μg/d). The inclusion criteria were FEV 1 50–90% predicted and at least 12% bronchodilator response, short-acting β 2 -agonist use of one puff/day or more and asthma symptoms. Salmeterol/FP combination was superior to montelukast/FP in improving morning PEF (34.6 vs 17.7 L/min), FEV 1 (0.19 vs 0.11 L). In contrast, there was no significant difference in nocturnal awakenings and asthma specific quality of life score. The percentage of patients experiencing at least one asthma exacerbation (any severity) was shown to be similar in the FP + salmeterol group (19.1%) than in the FP+ montelukast group (20.1%). Also there was no difference in the time to the first exacerbation between the salmeterol + FP and the montelukast + FP groups. Peripheral blood eosinophils were reported to be reduced significantly more in the montelukast + FP group (-0.04 × 10 3 /μl) than in the salmeterol + FP group (-0.01 × 10 3 /μl). Interestingly more serious adverse events were reported in the salmeterol + FP group. In another randomised, double-blind, double-dummy, parallel-group study (Jadad score 3) in patients (n = 948) with symptomatic asthma despite treatment with ICS, addition of montelukast (10 mg daily) was compared with addition of salmeterol (50 μg bid) for 12 weeks [ 112 ]. Patients were required to have symptoms despite the constant dose of ICS (any brand at any dose) and FEV 1 between 50% and 80% predicted and at least 12% bronchodilator response. Treatment with salmeterol resulted in significantly greater improvements from baseline compared with montelukast for most efficacy measurements, including morning PEF (35.0 vs 21.7 L/min), percentage of symptom-free days (24 v 16%) and percentage of rescue-free days (27 vs 20%). Also total supplemental salbutamol use (-1.90 vs -1.66 puffs per day) and nighttime awakenings per week (-1.42 vs -1.32) decreased significantly more with salmeterol than with montelukast. Six percent of patients in the salmeterol group experienced a total of 27 asthma exacerbations compared with 5% of patients in the montelukast group who experienced 24 asthma exacerbations during the 12 weeks treatment period. However, the patients experiencing an asthma exacerbation were withdrawn from the study. Thus, annualised incidences of exacerbations cannot be compared [ 112 ]. The safety profiles of the two treatments were reported to be similar. Taken together, addition of salmeterol seems to produce better improvement of asthma control when lung function is assessed than addition of montelukast in patients with asthma suboptimally controlled by small to moderate doses of ICS. However, in one long-term study [ 111 ] addition of montelukast to fluticasone was shown to be non-inferior to addition of salmeterol when the percentage of patients with at least one asthma exacerbation was used as the primary endpoint. Whereas addition of salmeterol may produce a better improvement in lung function, addition of montelukast may provide additional anti-inflammatory efficacy to ICS that is reflected in a long-term efficacy on asthma exacerbations. A factor that may produce a selection bias in these studies [ 109 - 111 ] is that a positive response to bronchodilator was required for inclusion. In fact, the reported mean improvements in FEV 1 in response to β 2 -agonist were 23–24% [ 109 ], 27.0–27.4% [ 110 ] and 18.4–18.8% [ 111 ] in the single studies. This may produce a selection bias favouring long-acting β 2 -agonist. However, one needs to remember that many of those studies done with leukotriene receptor antagonist to prove their efficacy in the treatment of asthma have been performed with patients displaying a significant response to β 2 -agonist. Another factor that might be considered to produce bias is that all the above three studies that report salmeterol to be better have been sponsored by the producer of salmeterol and that study reporting the non-inferiority of montelukast as compared with salmeterol has been sponsored by producer of montelukast. Montelukast versus salmeterol – other literature In addition to the normal clinical endpoints, the effects of addition of salmeterol (50 μg bid) or montelukast (10 mg/d) to the treatment regimen were analysed on AMP bronchial challenge, blood eosinophil counts and exhaled NO in a placebo-controlled, double-dummy, crossover study in patients (n = 20) with persistent asthma not controlled with ICS [ 113 ]. For the provocative concentration of AMP causing a 20% fall in FEV 1 , compared to placebo, there were significant differences with the first and last doses of montelukast as well as the first but not the last dose of salmeterol, thus indicating the development of some tolerance with salmeterol. Only montelukast produced a significant, albeit trivial, suppression of blood eosinophil count. There were significant improvements with the first doses of salmeterol for all parameters of lung function. After 2 weeks of treatment, there were significant improvements with both drugs on rescue bronchodilator requirement and morning PEF. There were no significant differences between drugs for any endpoints except blood eosinophils. Thus, the results suggest some anti-inflammatory activity for montelukast when used as an "add-on" therapy. Salmeterol versus zafirlukast – studies included in this systematic review In a randomised, double-blind, double-dummy parallel-group trial (Jadad score 3) addition of zafirlukast (20 mg bid) was compared with the addition of salmeterol (50 μg bid via MDI) for 4 weeks in adult and adolescent patients (n = 429) with persistent asthma [ 114 ]. Patients were required to have FEV 1 percentage predicted normal between 50 and 70% with or without asthma symptoms, or FEV 1 of 70.1% to 80% of predicted normal values and symptoms or requirement for rescue β 2 -agonist use ≥4 puffs/day or diurnal PEF-variation of more than 20% at two days during 6 days run-in. Both inhaled salmeterol and oral zafirlukast resulted in within-group improvements from baseline in measures of pulmonary function (morning and evening PEF and FEV 1 ), asthma symptoms, and supplemental salbutamol use. Salmeterol treatment resulted in significantly greater improvements from baseline compared with zafirlukast for most efficacy measurements, including morning PEF (28.8 vs 13.0 L/min), evening PEF (21.8 vs 11.2 L/min), combined patient-rated symptom scores for all symptoms (-35 vs 21%), daytime albuterol use (41 vs 25%) and night-time salbutamol use (42% vs 16%). Also, statistically significant differences favouring the addition of salmeterol were noted on patient-rated symptom scores for shortness of breath and chest tightness, percentage of symptom-free days, sleep symptoms, nighttime awakenings and percentage of days and nights with no albuterol use. There was no difference between the groups in symptom score for wheezing. Interestingly, the difference between salmeterol and zafirlukast was clear at week 1, but not at 4 weeks when the effect on FEV 1 was analysed. One factor that may affect the results of this study is that there may be a randomisation bias as the proportions of patients using FP or TAA were not similar in the salmeterol and zafirlukast groups. This study was funded by the producer of salmeterol. Salmeterol versus zafirlukast – other literature As a part of the above study [ 114 ], a randomised, double-blind, double-dummy parallel-group trial comparing the addition of zafirlukast (20 mg b.i.d) with the addition of salmeterol (50 μg bid) for 4 weeks in patients (n = 289) with persistent asthma, 80% of whom were on a concurrent ICS regimen has been published [ 115 ]. Both inhaled salmeterol and oral zafirlukast resulted in within-group improvements from baseline in measures of pulmonary function (morning and evening PEF and FEV 1 ), asthma symptoms, and supplemental salbutamol use. Salmeterol treatment resulted in significantly greater improvements from baseline compared with zafirlukast for most efficacy measurements, including morning PEF (29.6 vs 13.0 L/min), percentage of symptom-free days (22.2% vs 8.8%) and percentage of days and nights with no supplemental albuterol use (30.5% vs. 11.3%). Formoterol versus zafirlukast versus theophylline – other literature An open, randomised Turkish study [ 116 ] recruited patients with moderate persistent asthma having symptoms despite the use of moderate to high doses of ICS. The patients were required to have a FEV 1 reversibility of at least 15%. Patients (n = 64) were randomised to three different treatments budesonide (800 μg/d) plus formoterol (9 μg bid), budesonide (800 μg/d) plus zafirlukast (20 mg bid) or budesonide (800 μg/d) plus sustained-release theophylline (400 mg/d) for three months. After three months there were no between group differences in endpoints such as morning and evening PEF, PEF variability, FEV 1 , daytime or nighttime symptom scores and rescue terbutaline use. However, the addition of formoterol produced earlier improvements compared with the two other groups in criteria such as PEF variability, day- and night-time asthma symptom scores and supplemental terbutaline use. Patients in budesonide plus zafirlukast group experienced most adverse effects, but no statistical analysis was presented. The authors conclude that in patients who still have symptoms despite the treatment with ICS, the addition of any of these medications to the treatment is a logical approach and may be chosen. Conclusions on the comparisons between LABA, LTRA and theophylline as add-on options LABA (salmeterol) seem to have superior efficacy as add-on therapy in persistent asthma not controlled by low to moderate doses of ICS as compared with LTRA (montelukast; four studies or zafirlukast; one study). More studies comparing the different add-on options are needed as well as studies with longer duration as the current evidence is mostly limited to follow-up period of 3 months. Compliance and treatment strategies When assessing a patient with persistent asthma who is not adequately controlled by low to moderate doses of ICS: • It is important to find out whether the patient is using the prescribed medication correctly. Poor compliance in asthma patients treated with ICS is a very common reason for treatment failure. Compliance with ICS is often less than 50% [ 117 , 118 ]. Oral asthma therapies may result in better compliance [ 119 ]. • Secondly, it is important to check whether the inhalation technique is adequate. Problems with the inhalation techniques are very common, especially among children and the elderly [ 120 ]. Good patient education, especially if it is self-management oriented improves health outcomes in adults with asthma [ 121 ]. • Thirdly, it is important to search for possible environmental factors, such as changes in home and working environment, hobbies and pets. If asthma exacerbations are the dominant problem, guided self-management of asthma has been proven to be an efficient treatment strategy. In a Cochrane review [ 121 ] self-management of asthma was compared with usual care in 22 studies. Self-management reduced hospital admissions (odds ratio; OR 0.58, 95% confidence interval; CI 0.38 to 0.88), emergency room visits (OR 0.71; 95% CI 0.57–0.90), unscheduled visits to the doctor (OR 0.57; 95% CI 0.40 to 0.82), days off from work or school (OR 0.55; 95% CI 0.38 to 0.79) and nocturnal asthma (OR 0.53; 95% CI 0.39 to 0.72). Conclusions Addition of formoterol or salmeterol seems to be superior as compared with the increase in the dose of the ICS in improving lung function, controlling asthma symptoms and reducing the use of rescue bronchodilator treatment. By increasing (doubling) the dose of the ICS the clinical improvement is likely to be of small magnitude. However, if frequent exacerbations are the major problem, increasing the dose of ICS may significantly help to reduce the number of exacerbations. By avoiding doses above 1000 – 1500 μg/d (budesonide and BDP) or 500 – 750 μg/d (FP) the risk of systemic adverse effects remains low. However, it should be noted that the evidence on the superiority of LABA is limited to symptomatic patients with mild to severe persistent asthma currently treated with low to moderate doses of ICS and presenting with a significant bronchodilator response. Also, addition of the LTRA montelukast or zafirlukast may improve asthma control in patients remaining symptomatic with ICS and addition of montelukast may be equal to double-dose ICS. Addition of LABA (salmeterol) seems to produce better asthma control as compared with a LTRA (montelukast or zafirlukast) whereas the long-term efficacy of LTRA (montelukast) on asthma exacerbations may be equal to LABA (salmeterol). There is evidence that addition of low-dose theophylline to the treatment regimen may be equal to doubling of the dose of ICS. However, more studies are needed to better clarify the role of leukotriene antagonists and theophylline as "add on"-therapies. For patients with inappropriate inhalation technique the value of LTRA or theophylline are especially worth considering. More studies are now needed to compare between different add-on therapies and to explore the effect of more than one add-on therapy in patients with more severe asthma as well as in those having symptoms but not significant bronchodilator response. Another issue not addressed by these studies of large patient groups are the different responses of patients to the different add-on therapies. This needs to be studied by comparing add-on treatments in the same patients, but these studies are difficult and prolonged. In the future it may be possible to predict factors that predict the value of a particular add-on therapy in a particular patient, but the currently published studies unfortunately provide no guidance. Abbreviations ACTH: corticotrophin, AMP: adenosine monophosphate, AQLQ: asthma quality of life questionnaire, BAL: bronchoalveolar lavage, BDP: beclomethasone dipropionate, ECP: eosinophil cationic protein, FEF 50 : forced expiratory flow when 50% of vital capacity has been exhaled, FENO: exhaled nitric oxide, FEV 1 : forced expiratory volume in one second, FP: fluticasone propionate, FVC: forced vital capacity, HFA: hydrofluoroalkane-134a formulation, HPA: hypothalamic-pituitary-adrenal, ICS: inhaled corticosteroid, LABA: long-acting β 2 -agonist, LTRA: leukotriene receptor antagonist, MDI: metered dose inhaler, NNH: number needed to harm, NNT: number needed to treat, PC 20 : provocative concentration causing a 20% fall in FEV 1 , PD 20 : provocative dose causing a 20% fall in FEV 1 , PEF: peak expiratory flow, TAA: triamcinolone acetonide Authors' contributions HK carried out the literature searches, evaluated the studies, conceived the review and drafted the manuscript. AL, EM and PJB participated in the design and writing of the review. All authors read and approved the final manuscript. Supplementary Material Additional File 1 Tables 1–7-Kankaanranta.doc contains tables 1–7 of this review. Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC528858.xml
523845	Which Risk Factors Matter to Whom?	null	There is a much-quoted saying, attributed to the epidemiologist Geoffrey Rose: “A large number of people exposed to a small risk may generate many more cases than a small number exposed to a very high risk.” This is true for many individual risk factors such as salt intake (linked to high blood pressure and cardiovascular disease) and speeding on the highway (linked to injuries and accidents). Does it apply to many other global health risks? The study by Anthony Rodgers and colleagues suggests that it does. To develop effective health policies, one must understand the existing health risks and disease burdens. On a worldwide scale, this is a tough challenge. The Global Burden of Disease Database, maintained by the World Health Organization (WHO), collects data from countries around the world on risk factors such as tobacco, malnutrition, childhood abuse, unsafe sex, childbirth, and cholesterol levels, as well as on disease burdens, for example depression, blindness, and diarrhea. A large group of scientists from all over the world has developed a framework to analyze these data. To compare different risks or burdens, they calculate disability-adjusted life-years, or DALYs—the number of healthy life years lost because of a particular disease or risk factor. Tobacco is a major player in the global burden of disease (Photo: Bill Branson) Rodgers and colleagues used data from the WHO database for 26 risk factors and from 14 epidemiological subregions of the world to calculate the proportion of risk-factor-attributable disease burden in different population subgroups defined by age, sex, and exposure level. For being underweight in childhood, for example—the leading risk factor for global loss of healthy life—they found that only 35% of the disease burden occurred in severely underweight children, the rest occurred in those only moderately underweight. The relative risks for the moderately underweight are much lower, but the number of children in that category is so large that the total attributable burden amounted to almost two-thirds of the total global burden of disease for that risk factor. The analysis confirms—and extends to a global level—previous research showing that many major health risks are important across the range of exposure levels, not just among individuals exposed to high levels of risk. It also points to risk factors that are particularly prevalent among specific populations and age groups, and for which highly targeted interventions could be effective. Despite numerous caveats and limitations of studies like this one, such analyses are essential aids in guiding the distribution of limited funds to lower the burden of life years lost to premature death and disability.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC523845.xml
550673	Single cell studies of the cell cycle and some models	Analysis of growth and division often involves measurements made on cell populations, which tend to average data. The value of single cell analysis needs to be appreciated, and models based on findings from single cells should be taken into greater consideration in our understanding of the way in which cell size and division are co-ordinated. Examples are given of some single cell analyses in mammalian cells, yeast and other microorganisms. There is also a short discussion on how far the results are in accord with simple models.	Introduction What is the point of single cell studies of the cell cycle? The simple answer is that they provide extra information that is not available from studies of cell populations. Without them a cell biologist can be misled. It is easiest for me to start with the theme of the extensive results on single cells of the fission yeast Schizosaccharomyces pombe with which I have worked since the mid-1950s. It was then a fairly obscure organism for physiological studies though it had a good genetic background found by U. Leupold in Bern [ 1 ]. Since then it has flourished and quite large international meetings are now devoted entirely to it. For those unfamiliar with it, it is like a scaled-up bacterial rod with division at a medial septum, unlike budding yeasts. One the early results on its growth came from a single cell study by Bayne-Jones and Adolph [ 2 ]. Here I need to make a small digression about references. They will be given in this article but there are much longer accounts of nearly all the topics in my recent 100-page review [ 3 ]. When I took up fission yeast in the mid-fifties, I used a new microscopic technique, which gave by optical interferometry the total dry mass of single growing cells as well as their volume [ 4 ]. Volume increased, approximately in an exponential curve, through the first three quarters of the cycle but then stayed constant for the last quarter between mitosis and division. But total dry mass increased approximately linearly through the whole cycle. This was the first demonstration of linear growth, and I was surprised. Early synchrony techniques by induction This period of the fifties was when attention in this field was largely focused on the successful synchronisation of Tetrahymena and Chlorella by periodic changes in their environment. Good synchronous cultures would mean that powerful biochemical techniques, often enzyme activity assays at that time, could be applied in a cell cycle context. In the next 15 years, induction synchrony was somewhat improved but the cell cycles were always and inevitably distorted. Methods were also developed to select out a fraction of an asynchronous culture in one stage of the cycle and grow it up separately (for example," membrane elution", where cells growing on a membrane come away at division). They produce less distortion but a much lower yield than induction. Because of what can be measured in synchronous cultures, they are the natural choice for the molecular biologist. But it is as well to remember their limitations. The distortions after induction have been mentioned, but even with selection synchrony there are problems. The main one is that they are, in practice, not all that synchronous. The selected cells come from more than a very narrow region of the cycle. Some of the variation can be reduced by a correction for asynchrony [ 5 ] but there is still cell-to-cell variation in cycle stage and this can obscure the fine detail of the cycle. Single cell measurements may help here. Single cell analysis in yeast Returning to single cell analyses of fission yeast, volume growth was followed in finer detail by Mitchison and Nurse [ 6 ]. One part of this analysis, on films taken previously by Fantes [ 7 ], showed that increase in volume was not a simple exponential during the growth phase in the first three quarters of the cycle but rather two linear segments with a rate change point (RCP) between them. The position of the RCP showed a large cell-to-cell variation. An important moral here is that these two linear segments vanished into an apparent exponential increase in a "well synchronised" culture made by selection. Such a culture scarcely showed the plateau in growth during the last quarter of the cycle. This distinction between single cells and synchronous cultures does of course depend on the frequency and accuracy of the data points. If the points have too much scatter, the fine detail of the single cell linear patterns is lost. There is also a second RCP at the end of the cycle. A much more detailed analysis of populations of single cells followed on films was made by Sveiczer et al. [ 8 ] on fission yeast. A plot of extension growth against birth size has a strong negative slope. So also does a plot of cycle time against birth size. This has important implications for the definitions of "size control", discussed in that paper. Problems of single cell analysis Single cell studies have their problems. We have been lucky in using yeasts that are not apparently affected by growing on warm agar pads under a coverslip. They show "balanced growth", a property in which there is no change in extensive properties between successive cycles [ 9 ] and that should always be checked. Useful deductions can often be made with unbalanced growth but it will be a distortion of the normal cycle. The cells also have to keep still or be followed, a problem discussed below. We have not found ways of sticking yeast to glass (e.g. with lectins) that permit "normal" growth. Cells may also need a continuous supply of fresh medium, probably for oxygenation. Various types of microscopic mounting chambers have been described in the last 50 years or so, e.g. [ 10 ], but few seem to have been stringently tested. Many experimental studies on cell growth kinetics can be tedious; single cell studies are no exception. Here, however, modern automation is beginning to have very promising prospects. Anyone who has spent a day on a yeast film re-focusing the microscope every 5 min will welcome auto-focusing devices that are now available. Analysis has also become much easier with electronic imaging followed by image analysis programmes, and perhaps presentation on spreadsheets. It is now possible to have a programme that requires some hand work in the initial setting up under the microscope but will then run automatically, measuring cell length and diameter. This has been done for fifty or more single cells of fission yeast – a long way from the early days of using a ruler to measure the length of yeast cells on projected film images. Another point that should be raised here is that the new technology could profitably be applied to the growth of Escherichia coli . The limitations of synchronous cultures in hiding the fine detail of increases in volume or area could well mean that single cell studies might reveal more than an exponential increase. There might even be something like the two linear patterns that were popular models in earlier work with this bacterium [ 11 ]. What to measure Volume and area of a rod-shaped organism are two of the parameters that can be measured in single growing cells. So is dry mass by interferometry. But there others, of which one of the most interesting is the use of the Cartesian diver, which was originally developed some fifty years ago at the Carlsberg Laboratory in Copenhagen. It requires technical skills and very tightly controlled temperature in water baths, but it is exquisitely sensitive. It can be used in at least two ways. One is as a diver balance, which measures "reduced weight" or weight in water. Providing there are not major changes in chemical composition, this is proportional to total dry mass. It was used on single cells of Amoeba proteus in an important classic paper by Prescott [ 12 ] mentioned below. It can also be used with minute divers as a respirometer. Hamburger [ 13 ] measured oxygen uptake in Acanthamoeba a nd CO 2 production in fission yeast (Hamburger et al. [ 14 ]), in both cases over several cell cycles starting with single cells – a remarkable achievement. In both cases, the results were elegant linear patterns with an RCP at division. Another interesting single cell method was the colorimetric enzyme assay of single yeast cells in microdrops [ 15 ]. This might have been developed with promise, but was not followed up, partly perhaps because the results differed from similar assays in synchronous cultures. One of the advantages of single cell work with yeasts is that they stay still on an agar pad so they can be followed for a couple of cycles before overlapping spoils the image. This is not true of many mammalian cells, which move around on the substrate. One solution to this problem comes in the work on fibroblasts (mouse L cells) described in Zetterberg [ 16 ], Killander and Zetterberg [ 17 ], Zetterberg and Killander, [ 16 ]. These are part of an impressive body of work initiated using optical machinery gathered by Trigvar Caspersson, along with a great deal of skill and hard work. In one set of experiments on single cells [ 17 ], they made a measurement of the dry mass of single cells by interferometry and then placed it in the cycle by following it as it moved about until it divided. The difference in timing between the measurement and cell division gave the timing in the cycle. A second set of experiments used frequency analysis to set the cycle stages. This is a method widely used to determine G1, S and G2 in flow cytometry but is less suitable for the slow and imprecise doublings in something like dry mass. I therefore regard the single cell analyses as more reliable and they are not the same as those from the second method. What are needed now are techniques that combine the subtlety and precision of single cell measurements with the new techniques of automation. A promising start was made by Zicha and Dunn [ 19 ], and the development is being actively pursued elsewhere. Organisms which tend to be forgotten about these days are those lower eukaryotes that make poor material for molecular biologists because of inadequate genetic backgrounds. One important set of results are those from the early pioneer work of Prescott [ 12 ] on Amoeba proteus mentioned above. The results showed that the increase of single cell "dry mass" fell in a reverse exponential, with a rapid increase at the start of the cycle falling to zero towards the end. This, of course, is lethal for anyone who believes that a rising exponential is the paradigm for the cell cycle. Tetrahymena pyriformis has a long and distinguished history in the cell cycle with its early induction synchrony. But in the 1960s there was a burst of studies on selected single cells or small groups. The growth patterns were often not well defined but it seems that absolute measurements of volume and of respiration rate were a better fit to linear growth (Prescott, [ 20 ]). Such analyses might now be checked using some of the semi-automated procedures referred to above. Growth in syncytia Physarum polycephalum is a myxomycete of considerable importance in some earlier work on cell cycle control. It is effectively a big single multinucleate cell with complete natural synchrony in nuclear division. It does not show exponential increase in macromolecular synthesis. For instance, there are two peaks in the rate of protein synthesis, one in the S period and the other in G2 (Mittermayer et al , [ 21 ]). General conclusion It would appear that there are no universal patterns of growth in these lower eukaryotes. Models My title makes mention of "some models". Let me be clear that there are two quite different types of cell cycle models. One type includes detailed mathematical and molecular models dealing with discrete periodic events like mitosis (e.g. [ 22 ]). These are complex and can illustrate the relations between many components of a network at the event, on reasonable assumptions. They are important aids in understanding the events and are a fairly recent development in the cell cycle world. There are certain limitations at present. With mitosis, the models have problems with the starting event (a size control?), with location in cellular compartments, and with the final mechanical events. However, such models will certainly develop. However, what I am concerned with here are much earlier and much simpler models, not of periodic events in the cycle like DNA synthesis, but of continuous growth. Here the two dominant models were, for simplicity, an exponential pattern of increase and a linear one. My own view [ 3 ] of the earlier experiments is that, on the whole, they favour linear increase but it was also clear that some patterns, e.g. volume in fission yeast, are more complex. Linear increases with rate change points have certainly survived in fission yeast where there are no exponential increases (Table 1 in [ 3 ]) and this has revived for me an old hypothesis of "gene dosage". What, for instance, happens to synthesis rates between G1 and G2? But one thing is clear – that a single unifying dream of exponential synthesis is not in accord with the facts. It is really useless to wave Occam's Razor around. The end of his razor blade is "without necessity". In all reasonable judgements, the necessity is there. Beyond that is prejudice. Figure 1 Modes of growth in cell length of wild-type and wee1 mutant cells of fhe fission yeast Schizosaccharomyces pombe , after Sweiczer et al. (1996) Figure 2 Growth through one cycle of individual Amoeba, after Prescott (1976)	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC550673.xml
538281	Lower rate of genomic variation identified in the trans-membrane domain of monoamine sub-class of Human G-Protein Coupled Receptors: The Human GPCR-DB Database	Background We have surveyed, compiled and annotated nucleotide variations in 338 human 7-transmembrane receptors (G-protein coupled receptors). In a sample of 32 chromosomes from a Nordic population, we attempted to determine the allele frequencies of 80 non-synonymous SNPs, and found 20 novel polymorphic markers. GPCR receptors of physiological and clinical importance were prioritized for statistical analysis. Natural variation and rare mutation information were merged and presented online in the Human GPCR-DB database . Results The average number of SNPs per 1000 bases of exonic sequence was found to be twice the average number of SNPs per Kilobase of intronic regions (2.2 versus 1.0). Of the 338 genes, 111 were single exon genes, that is, were intronless. The average number of exonic-SNPs per single-exon gene was 3.5 (n = 395) while that for multi-exon genes was 0.8 (n = 1176). The average number of variations within the different protein domain (N-terminus, internal- and external-loops, trans-membrane region, C-terminus) indicates a lower rate of variation in the trans-membrane region of Monoamine GPCRs, as compared to Chemokine- and Peptide-receptor sub-classes of GPCRs. Conclusions Single-exon GPCRs on average have approximately three times the number of SNPs as compared to GPCRs with introns. Among various functional classes of GPCRs, Monoamine GPRCs have lower number of natural variations within the trans-membrane domain indicating evolutionary selection against non-synonymous changes within the membrane-localizing domain of this sub-class of GPCRs.	Background The 7TM (7 trans-membrane domain proteins) genes, also known as the hetero-trimeric GTP-binding protein (G protein)-coupled receptors (GPCRs), are members of a large family of genes with an estimated 700 members in the human genome [ 1 ]. These receptors are plasma membrane-bound and have evolved to respond to a large number of extracellular and chemical signals. Upon interaction with their ligands, GPCRs act through the G proteins in signaling pathways that influence physiological functions. All GPCRs, in spite of great diversity in sequence composition, share a common protein structure. An N-terminal extracellular domain of variable length is followed by seven hydrophobic transmembrane-helices, connected by three intracellular (IL) and three extracellular (EL) loops, which then terminates in a C-terminal intracellular domain [ 2 ]. The functional and structural role of the different domains has been elucidated by systematic point mutations and crystal structure analysis for many of the human GPCR proteins. Several studies have collectively analyzed the occurrence, and importance of coding GPCR SNPs [ 3 - 5 ] and also the relevance and importance of mutations within these genes for the pharmaceutical industry [ 6 ]. The functional significance of thousands of point mutations has been described by a large number of investigations as evident at NCBI's PubMed Central. Mutation databases dedicated to GPCR mutations are currently available online like GPCR DB [ 7 ] and tinyGRAP [ 8 ]. Although an extensive collection of mutations is available at these sources, the distribution of these mutations and variations within the gene or peptide, along with common SNPs is not easily accessible or evident. The SNP databases in public domain (for example: NCBI's dbSNP) have highlighted all non-synonymous SNPs (nsSNPs). Also HGVBase has further classified the location of the amino acid within the encoded proteins to more accurately predict the detrimental effects of a change in peptide sequence. From a pharmacogenetics viewpoint, the information about natural variations within GPCR transcripts and peptides, with allele frequency and validation data and disease association, is an important, yet currently unavailable, public resource. A database with functional promoter SNP, allele frequency, peptide variation information, population and haplotype information presented in a graphically accessible format would facilitate pharmacogenomics research related to GPCR proteins. HUMAN GPCR-DB aims to provide such a public resource. Also the rate of false positive SNPs determined experimentally in GPCR genes is reported to be relatively high [ 5 ]. For typical case-control association studies, prevailing designs favor highly polymorphic loci as against loci where the frequency of the minor allele is below 10%. Therefore more nsSNPs need to be validated and frequencies determined across ethnically diverse populations. We have designed genotyping assays and attempted to validate a number of GPCR nsSNPs for which no validation information was available on public databases, and deposited the validation information at HUMAN GPCR-DB. We have also collected published literature SNPs and added to our online database. Several recent studies have focused on the subset of nsSNPs that most likely influence phenotype [ 9 - 13 ]. Comparatively, fewer attempts have been made on predicting and validating functional promoter SNPs [ 14 ]. As a part of a parallel work, we have developed a streamlined bioinformatics and wet-lab analysis methods to identify putative functional promoter SNPs with up to 70% probability of influencing gene expression. By applying this analysis package to all of the 338 genes in our database, we have highlighted, putative functional promoter SNPs. HUMAN GPCR-DB also attempts to merge SNP and other variations from published articles from PubMed and online SNP databases to facilitate direct identification of the functional significance of a natural variation. Results A total of 427 non-redundant Human GPCR peptides were obtained from Swissprot of which 338 had Swissprot ID and the remaining had only TrEMBL identifications. Also, 89 of the 427 GPCR were classified as olfactory receptors. While all of the 427 entries were included in the HUMAN GPCR-DB database, only non-olfactory (i.e. 338) were considered for further analysis of gene structure, alternative transcripts, SNPs, and protein variations. Although the TrEMBL entries have also been included, the data displayed for these entries would be more accurately presented in future updates of the database. The statistical calculations for the genomic SNPs are based on the 338 non-olfactory entries, while the data for nsSNPs encoding peptide variations is based on 222 entries for which there are documented evidence for a 7-TM domain structure. Transcript information for each gene was used for calculating average exonic and intronic SNPs. For genes with multiple transcript variants, the longest transcript was selected. Average number of SNPs per exon (n = 1511) of the 338 genes was one, while the average number of SNPs per intron (n = 1174) was nine. However, the average number of SNPs per 1000 bases of exonic sequence was twice the average number of SNPs per kilobase of intronic regions (2.2 versus 1.0). Of the 338 genes, 111 were single exon genes, that is, were intronless. The average number of exonic-SNPs per single-exon gene was 3.5 (n = 395) while that for multi-exon genes was 0.8 (n = 1176). This observation is in agreement with earlier observation based on a smaller number of GPCRs, where, compared to intronless GPCRs, exons in genes with introns on average had fewer SNPs [ 15 ]. Of the 1511 SNPs from the exons of 338 GPCR transcripts, 816 SNPs were coding SNPs, among which 392 were nsSNP; 211 of which had no validation information in either of the source databases, i.e. NCBI/dbSNP and HGVBase. The number of validated and non-validated SNPs is shown in Table 1 . By excluding genes with 'probable', 'putative', 'precursor' and other ambiguous terms as a part of their description (as designated by Ensembl), we reduced the number of SNPs from 211 to 123 non-validated, nsSNPs which were considered as our list of prime candidates for the wet-lab validation process (Table 2 ). As a part of our first stage validation process, we designed assays for 80 of the 123 SNPs in GPCR genes of interest based on our better understanding of their role in human disease and physiology. Finally, of the 80 assayed SNPs, 20 were found to be polymorphic in our Nordic population sample consisting of DNA from 16 un-related healthy individuals. Of the 20 polymorphic markers 12 had a minor allele frequency higher than 10%. Table 1 Distribution of validated and non-validated SNPs. Synonymous SNPs Non-synonymous SNPs Total SNPs Validated SNPs 212 181 393 Non-validated SNPs 212 211 423 Total 424 392 816 The number of synonymous and non-synonymous SNPs and those with validation information, as deposited at NCBI/dbSNP, during 2003–4. Table 2 Number of GPCRs in validated and non-validated categories. Nr. of GPCR genes Total nsSNPS Validated nsSNPs Non-validated nsSNPs This study 338 (genes classified as GPCRs) 392 181 211 (123 were considered for validation). 80 of 123 assayed. 20 were polymorphic 222 (bearing evidence for 7-TM domains) 283 112 (Added 120 from [16]) 171 101 (classified in 3 sub-groups – Monoamine, Peptide and Chemokine). 182 53 (Added 83 from [16]) = 136 (used in Table 3). The initial 338 GPCRs were selected as per annotation by Ensembl. Support for structural evidence of 222 GPCRs was obtained from SWISSPROT. Classification of GPCRs was obtained from GPCRDB , Ensembl , International Union of Pharmacology . Table 2 shows the number of GPCRs categorized by different criteria, and SNPs categorized in the two groups of validate and non-validated nsSNPs. Of the 338 genes, 222 had a documented GPCR structure as described by SWISSPROT/TrEMBL database. The total number of nsSNPs in this subset of 222 proteins was 283, of which 112 had validation information (leaving 171 with no validation information). To these we added 120 SNPs from Pubmed reports [ 14 ]. The identity of the 120 variations from published reports was verified to be SNPs and not rare mutations. Therefore a significant proportion of 'disease causing', rare variations were eliminated since they were reported from rare family based disease cases. These 120 SNPs are represented in the HUMAN GPCR-DB as 'rs-missing' since dbSNP records for many of these were not found. As future updates from dbSNP assign rs-IDs to the new SNPs, our database would update the records likewise. According to the International Union of Pharmacology , 101 GPCRs from 338 were categorized either as 'peptide receptors' (n = 47) or 'chemokine receptors' (n = 54) or 'monoamine receptors' (n = 36). The distribution of nsSNPs across the 5 structural and functional domains (N-terminus, external loops, trans-membrane, internal loops and C-terminus) of these 101 GPCRs was calculated (Table 3 ). These 101 GPCRs have in total 182 nsSNPs, of which 53 SNPs have validation information in the major public databases. To these 53 SNPs we added 86 SNPs from published PubMed sources [ 14 ], bringing the total number of validated SNPs, used for this analysis, to 136 SNP. The 20 SNPs validated in this study were not included for this analysis since we wanted to analyze publicly available data only, at this time. The distribution of nsSNP numbers was compared between individual groups (monoamines-receptors only, or chemokine-receptors only or peptide-receptors only) and in various combinations with other two groups (monoamines plus chemokines or peptides plus monoamine, etc). None of the groups of receptors deviated from the mean of the three groups together, in any significant way. The N-terminus and external-loop SNPs were then combined in one group, and C-terminus and internal-loop SNPs in another group, and compared with nsSNPs in TM region. The nsSNP distribution in the three domains approached significance (Pearson's p-value 0.06) in the Monoamine sub-group of GPCRs. We then calculated the average number of nsSNP per 1000 bases of each of the 5 domains. The average number of nsSNPs in the TM region of Monoamine receptors (two SNPs per kilobase) was half of the average for each of the other groups (four or five nsSNPs per kilobase). This difference was not observed for any of the other four (N-term, e- and i-loops and C-term) structural domains of the peptides (Table 3 ). Table 3 SNP distribution in peptide domains. Peptide domain Genes N-term e-loop TM i-loop c-term p-value N-term + e-loop TM C-term + i-loop p-value Monoamine + Peptide + Chemokines 101 20 (3.7) 17 (3.6) 43 (3.0) 29 (3.5) 27 (4.5) 37 (3.7) 43 (3.0) 56 (4.0) Monoamine Only 36 7 (5.7) 6 (3.9) 11 (1.9) 16 (3.3) 18 (3.9) 0.26 13 (4.7) 11 (1.9) 34 (3.5) 0.06 Peptide Only 47 10 (2.9) 8 (3.9) 25 (4.2) 11 (4.3) 15 (5.7) 0.89 18 (3.3) 25 (4.2) 26 (5.0) 0.79 Chemokine Only 18 3 (4.0) 3 (2.8) 7 (2.5) 2 (2.5) 4 (4.5) 0.87 6 (3.3) 7 (2.5) 6 (3.5) 0.72 Chemokine + Monoamine 54 10 (5.1) 9 (3.4) 18 (2.1) 18 (3.2) 12 (4.1) 0.93 19 (4.1) 18 (2.1) 30 (3.5) 0.78 Peptide + Monoamine 83 17 (3.7) 14 (3.9) 36 (3.2) 27 (3.6) 23 (4.9) 0.94 31 (3.8) 36 (3.2) 50 (4.2) 0.96 Chemokine + Peptide 65 13 (3.1) 11 (3.5) 32 (3.7) 13 (3.8) 19 (5.4) 0.87 24 (3.3) 32 (3.7) 32 (4.6) 0.71 Total number of nsSNP in various domains of 3 subgroups of GPCR proteins. Abbreviations: N-term : N terminus; C-term : C terminus; i-loop : internal loops; e-loop : external loops; TM: trans-membrane. The numbers in the brackets are the average number of SNPs per 1000 base pairs of a specific domain. The Fisher's Exact p-values were calculated using a 5 × 2 contingency table at . Contingency tables of 2 × 3 were constructed at . P-values of below or close to 0.05 are considered significant. We compiled together the functional properties of peptide variations from published records along with the knowledge of the location and the two alleles of a nsSNP in our database. Searching PubMed records, we found 38 nsSNPs located in the precise position, and substituting the same amino acid, as those studied for functional analysis shown in Table 4 [See additional file 1 ]. Database interface and layout The HUMAN GPCR-DB is currently online . This database allows for 3 alternative queries, either Ensembl gene ID, or Swissprot/TrEMBL ID or part of the gene name or description. Resulting hits are displayed along with total number of coding SNPs and protein variations. These links in turn display a graphic representation of the SNP locations within exons (or peptide) and the query gene. The exons are drawn in proportion to the largest exon, while the introns are of fixed length. The SNP list provides allele information, flanking sequences (for assay development and strand verification, etc) and links for validation information and source databases. The promoter information is drawn in a similar manner, with mouse conserved regions indicated with green bars underneath and SNPs within conserved regions marked with a symbol 'M' in the SNP full-list. SNPs predicted to influence protein binding according to our prediction model are marked 'T' in the SNP full-list. The link for protein variations displays a window with SNP, marked in red arrows, and mutation distribution across 3 regions of the peptide; N-terminus, C-terminus and trans-membrane and loop regions. Association with diseases and the corresponding PubMed ID are displayed as a popup menu following the link under 'disease' column. Discussion We have constructed a database, which combines mutation information with validated SNP information from publicly available sources. We have then attempted to validate and determine the frequencies of 80 of the 123 non-synonymous SNPs for which no validation information was available publicly. Proportion of true polymorphic loci was 20%, in agreement with reported expectations from several studies [ 5 ]. The statistical approach for the analysis of distribution of natural variations in GPCRs, presented here is borrowed from two recent studies [ 15 , 16 ]. While in the first of these studies [ 15 ] 64 GPCR genes were sequenced in 82 individuals of divergent ethnic backgrounds and resulting frequency distribution of nsSNPs were compared with non-GPCR genes, the later study [ 16 ] analyzed differences in distribution of published and publicly available nsSNP in 62 GPCR genes, across the 5 peptide domains. For our current report we analyzed 222 GPCR genes with over 200 nsSNPs (283 snSNPs available on public databases, and 120 nsSNP from PubMed records) [ 16 ]. Transcripts lacking introns had on average higher density of SNPs (2-fold) than those with introns, in agreement with an earlier published report [ 15 ]. The distribution of the nsSNP in the 5 different peptide domains (N-term, e-loops, trans-membrane, i-loops and C-term) was found not to be different between any of the ligand specific sub-groups of GPCR proteins, namely peptide receptors, monoamine receptors and chemokine receptors. We reasoned that the evolutionary constraints on outer- and inner-loops along with the N-term and C-term regions would be related to the function of these regions while the constraints on the trans-membrane regions might be related to their structure. We, therefore compared the distribution of validated nsSNP in the 3 major peptide domains (e-loops + N-term = region 1; trans-membrane = region 2; i-loops + C-term = region 3). We observed a difference in nsSNP distribution across the three regions, which approached significance (Pearson's p-value = 0.06). There was a two-fold decrease in frequency of occurrence of nsSNP in the TM region of monoamine receptors as compared to the other sub-groups of receptors. This indicates that there might perhaps be a functional selection against variations, acting on TM domains of monoamine receptors, which is less selective on the TM domains of peptide receptors, and chemokine receptor GPCRs. A recent study reported differences between the distribution of 'disease causing' and 'non-disease causing' variations in different sub-groups of GPCR family members [ 16 ]. Our study excluded rare mutations, which were known to be associated with disease, and therefore a similar comparison was not possible. Although HUMAN GPCR-DB database does obtain the bulk of the information and data from Ensembl and dbSNP, it is not merely a subset of these major databases. While Ensembl provides sequence and genetic variation information, it provides SNP validation information obtained from public sources, which may include, as shown in several published studies, up to 50% false positives. NCBI's dbSNP provides validation-, submitter- and method-information, yet rates of false positives have proven to be high. These databases harbor information of genetic variations for all coding and non-coding regions of the human genome. The HUMAN GPCR-DB, in addition to providing this set of information, provides in-house validation and assay-information for non-validated nsSNPs. HUMAN GPCR-DB provides natural variation, mutation, promoter- and peptide-variation information along with gene structure and peptide 7-TM structure information and SNP validation information of a focused group of clinically important genes. Transcriptional regulatory regions on the 5'-FR of human genes encode short sequences which serve as targets for binding of transcription factors (TFs). Eukaryotic TFs tolerate considerable sequence variation in their target sites and recent works in bioinformatics [ 17 - 19 ] have developed reliable methods to model the DNA binding specificity of individual TFs [ 20 ]. Currently the most successful approach to overcome this information gap is based on the assumption that gene sequences conserved between species (here Human and Mouse) would most likely mediate biological function [ 21 - 25 ]. Our recent study (our manuscript, 2004) describes a method for the detection and validation of functionally important SNPs in the 5'-flanking regions of human. The rate of successful detection of SNPs influencing TFBS using our method is approximately 70%. This prediction algorithm has been used to highlight SNPs in 5' flanking regions of the GPCR in the HUMAN GPCR-DB genes to facilitate selection and study of functionally important promoter SNPs. The knowledge of functional promoter SNPs would help us study disease related GPCR in more details. The functional domains of human GPCRs have over the passed decade been dissected by systematically mutating the peptide sequence [ 8 ]. A collection of mutations and their disease significance and influence on the function of the protein together with common variations within the human population would facilitate our understanding of the variations and disease association. HUMAN GPCR-DB attempts to merge SNP and mutation information along with disease information in easily accessible and user-friendly manner. Although direct links to disease databases would be included in the next release, the existing information about the source publication can be helpful in obtaining the relevant details about the mutations. Current and future updates Of the 123 nsSNPs without validation information, we have currently validated 80 SNPs. Future updates would include information about the remaining nsSNPs, and any additional which are reported by public databases. We have also collected 120 published SNPs, which have as yet not been deposited at public databases, or are in the process of being deposited. We would have a complete update of SNP data for every new release of NCBI and HGVBase SNP tables. New GPCR identification and characterization, or changes in existing GPCR genes or proteins information would be updated once a year from Ensembl and SwissProt Databases. PubMed references would be updated monthly or as often as necessary to complete the mutation coverage of the 222 GPCR proteins. Haplotype information and genetic association studies for available SNPs along with published records on functional promoter SNPs would be added. A valuable addition would be to indicate variation frequencies in ethnically diverse populations. We are currently adding such information about the allele frequencies and populations in the database and would be provided in future updates. Conclusions Single-exon GPCRs on average have approximately three times the number of SNPs as compared to GPCRs with introns. Among various functional classes of GPCRs, Monoamine GPRCs have lower number of natural variations within the trans-membrane domain indicating evolutionary selection against non-synonymous changes within membrane localizing domain. The HUMAN GPCR-DB compiles SNPs and mutations in one database. Using a recently developed method for identification of functionally important SNPs in the 5'-flanking regions of human, with approximately 70% success rate, the database highlights such SNPs to facilitate selection and study of functionally important promoter SNPs. Methods The list of Human GPCR genes was compiled by collecting gene names from several different sources and subsequently the list was updated by removing duplicates and entries with incomplete information like peptide fragments, partial sequences and hypothetical proteins. The Ensembl MART genome server database was queried for GPCR family members. The Gene Ontology server Amigo was queried with the search term 'GO:0004930', which describes the GPCR group of genes. The list of Swiss-Prot and TrEMBL entries were fetched from GPCRDB [ 7 ], ensemble, International Union of Pharmacology . All the lists were merged and redundancies removed and Ensembl gene numbers (ENSG) were obtained for all of the genes from Ensembl genome server. The final list consisted of total of 427 genes with unique ENSG numbers. Of these 427 genes, 89 genes belonged to the sub-family of olfactory genes. Also of the 427 genes, 222 had demonstrable or convincing evidence for GPCR domain structure and sequence information in Ensembl and SwissProt databases. The gene and transcript map information were obtained from Ensembl databases 'homo_sapien_core_25_34d' and 'ensemble_mart_25_1', released in September 2004. The tables for gene mapping, SNP mapping and Human-Mouse alignment were obtained from ensemble, dbSNP, HGVBASE and UCSC and installed locally. For the chromosomal and genomic location of SNPs and validation information, NCBI's dbSNP tables 'snp', 'snpcontigloc', and 'snpcontiglocusid' and Ensembl's tables 'ContigHit' and 'locus' were used. Information about the location of the SNP within protein domains was obtained from Swissprot using bioperl modules for accessing protein features. HGVBase release version 14 was used for obtaining HGVBASE SNP identities and validation and frequency information. The flanking sequence information was obtained from both Ensembl's RefSNP table and by downloading sequence flat files from dSNP (ds_flat_chr'1–22. X, Y'.fa); For mapping Transcription Factor Binding Sites, the TFBS perl programming system [ 26 ] was used. This program applies position weight matrices (PWM) to DNA sequences to generate mathematical probability for the binding of a TF, based on the earlier described thermodynamics of binding energy [ 27 - 29 ]. Recent reviews and articles describe methods related to PWM and the bioinformatics of regulatory site prediction [ 18 , 30 ]. For determining human-mouse conserved regions, global best alignments files were downloaded from UCSC. SNPs in 5' flanking sequences were analyzed according to the differences in the absolute bind score derived from the matrices for each TF. A total of 78 factors from vertebrate class were used, hosted at the TFBS database, JASPAR [ 31 ]. MySQL™ version 4.0 with ActiveState™ Komodo version 2.3 as perl programming IDE and bioperl modules version 1.2 were used for database development. The web server technology used was Apache™ 2.0 with PHP 4.0, as supplied by NuSphere™ version 3.0. Non-synonymous SNPs were identified from all the SNPs, after the construction of the GPCR SNP database based on data acquired directly from dbSNP and HGVBASE. Validation of SNPs was carried out by DynaMetrix Inc., UK, using the DASH platform [32]. The allele frequency validation was performed on DNA samples from 16 anonymous individuals of Nordic descent, with the Institutional Review Board Approval KI 02-544. Abbreviations 7TM: 7 transmembrane; GPCR:G-protein Coupled Receptors; SNP: single Nucleotide Polymorphism; nsSNP : non-synonymous SNPs; cSNP : coding SNP; rSNP : regulatory SNPs; N-term : N terminus; C-term : C terminus; e-loop : external loops; i-loop : internal loops. Authors' contributions SM-T did all the coding, analysis, manuscript preparation and reviewer correspondence. AJB contributed genotyping information and validated a number of SNPs. CW provided running costs and assistance with writing the manuscript. All authors read and approved the final manuscript. Supplementary Material Additional File 1 Table 4 A list of natural non-synonymous variations and mutations with references to articles describing the phenotype associated with the variation. Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC538281.xml
523851	Aging is associated with increased collagen type IV accumulation in the basal lamina of human cerebral microvessels	Background Microvascular alterations contribute to the development of stroke and vascular dementia. The goal of this study was to evaluate age and hypertension related changes of the basal lamina in cerebral microvessels of individuals, who died from non-cerebral causes. Results We examined 27 human brains: 11 young and 16 old patients. Old patients were divided into two subgroups, those with hypertension (n = 8) and those without hypertension (n = 8). Basal lamina changes of the cerebral microvessels were determined in the putamen using antibodies against collagen type IV and by quantitative analysis of vessel number, total stained area of collagen, thickness of the vessel wall and lumen, and relative staining intensity using immunofluorescence. The total number of collagen positive vessels per microscopic field was reduced in old compared to young subjects (12.0+/-0.6 vs. 15.1+/-1.2, p = 0.02). The relative collagen content per vessel (1.01+/-0.06 vs. 0.76+/-0.05, p = 0.01) and the relative collagen intensity (233.1+/-4.5 vs. 167.8+/-10.6, p < 0.0001) shown by immunofluorescence were higher in the older compared to the younger patients with a consecutive reduction of the lumen / wall ratio (1.29+/-0.05 vs. 3.29+/-0.15, p < 0.0001). No differences were observed for these parameters between old hypertensive and non-hypertensive patients. Conclusions The present data show age-related changes of the cerebral microvessels in sections of human putamen for the first time. Due to the accumulation of collagen, microvessels thicken and show a reduction in their lumen. Besides this, the number of vessels decreases. These findings might represent a precondition for the development of vascular cognitive impairment. However, hypertension was not proven to modulate these changes.	Background Aging is associated with a deterioration of cognitive function including a decrease in the ability to process and store new information [ 1 ]. Processes that might negatively affect cognitive function during aging are manifold. Among them, the cerebral vascular system has a major impact on brain function. Craigie first described that the density of cerebral microvessels may correlate with functional activity [ 2 ]. However, morphological studies of the microvessels have been inconsistent up to now. Meier-Ruge and coworkers reported an increase in capillary density in older individuals [ 3 ], whereas others had shown a reduction [ 4 ]. One important reason for the different findings in microvascular changes might be the heterogeneity of the examined brain regions between, but also within the studies. Nevertheless changes in the cerebral microvessels have a major impact on secondary pathophysiological changes, like reduced cerebral blood flow (CBF) [ 5 , 6 ], and a decrease in doppler sonographic blood flow velocity in old age [ 7 ]. In addition, microvascular alterations are responsible for reduced cerebral metabolic rates for oxygen and cerebral glucose utilization, which is observed with increasing age [ 8 , 9 ] leading to an impaired transport of nutrients which in turn impairs neuronal function. Additional factors such as chronic hypertension may accelerate the progression of age-related capillary changes [ 10 ]. Although changes of cerebral vessels with aging and hypertension have been reported, there are presently no consistent data on the microvascular basal lamina. The aim of our study was an evaluation of age-related changes on cerebral microvessels and on the vascular extracellular matrix in humans and the possible impact of chronic hypertension using several different immunohistochemical methods for the detection of collagen type IV. Results Determination of collagen type IV in microvessels by immunohistochemistry Old patients (OP) showed 12.0 +/- 0.6 vessels per microscopic field, and young patients (YP) exhibited 15.1 ± 1.2 vessels per microscopic field (p = 0.02). The total area of collagen type IV positive vessels did not differ between OP and YP (11.0 ± 0.5 vs. 11.4 ± 0.6, n.s.). Therefore the calculated content of collagen type IV per vessel was higher in OP than in YP (1.01 ± 0.06 vs. 0.76 ± 0.05, p = 0.01). No differences were observed between old non hypertensive patients (ONHP) and old hypertensive patients (OHP) (vessels per microscopic field: 11.3 ± 1.1 in ONHP vs. 12.7 ± 0.7 in OHP, n.s.; area of collagen type IV: 12.1 ± 0.8 in ONHP vs. 10.6 ± 0.9 in OHP, n.s.; calculated content of collagen type IV per vessel: 1.11 ± 0.1 in ONHP vs. 0.91 ± 0.07 in OHP, n.s.). Thickness of vessel wall and lumen Microvessel wall thickness, inner lumen and ratio lumen/wall thickness were statistically different in OP vs. YP. The vessel wall was thicker in old than in young patients (3.14 ± 0.10 μm vs. 1.62 ± 0.06 μm, p < 0.0001). In addition the vessel lumen was reduced in the old group compared to the young group (4.00 ± 0.14 μm vs. 5.24 ± 0.13 μm, p < 0.0001, Figure 1 and 2 ). Therefore the ratio between the thickness of the vessel lumen and the vessel wall was lowered in the OP group compared to YP (1.29 ± 0.05 vs. 3.29 ± 0.15, p < 0.0001). The comparison of the same parameters in ONHP vs. OHP showed no significant distinctions (thickness of vessel wall: 3.13 ± 0.12 μm in ONHP vs. 3.16 ± 0.18 μm in OHP, n.s.; vessel lumen: 4.00 ± 0.26 μm in ONHP vs. 4.00 ± 0.14 μm in OHP, n.s.; ratio: 1.28 ± 0.07 in ONHP vs. 1.30 ± 0.08 in OHP, n.s.) Relative collagen type IV intensity in microvessels determined by CLSM Analysis of the relative intensity of collagen type-IV positive vessels was performed by confocal laser scanning microscopy. The relative content of collagen type IV in the microvessel wall was higher in OP than in YP (233.1 ± 4.5 vs. 167.8 ± 10.6, p < 0.0001). Again, the intensity in non-hypertensive old persons showed no difference compared to hypertensive old patients (230.7 ± 4.2 vs. 234.3 ± 6.7, n.s., Figure 3 ). Discussion In the present study, we investigated age- and hypertension-related alterations in the vascular extracellular matrix and the basal lamina in human brains. Our main finding is the age-related change of the basal lamina component collagen type IV. In old as compared to young individuals we found a significant decrease of the vessel number containing collagen type IV, a thickening of the vessel wall and narrowing of the vessel lumen and an increase in collagen type IV content per vessel. These changes in extracellular matrix proteins were demonstrated by immunohistochemistry as well as by confocal laser scanning microscopy. However, we were unable to establish significant changes of the microvessels in hypertensive old persons compared to normotensive old patients. These results are in good accordance with previous studies that have shown basal lamina thickening in experimental studies [ 10 - 12 ]. Research on basal lamina changes in humans has mostly been focused on their relationship to neurodegeneration. In Alzheimer's patients Kalaria and coworkers found a 55% increase of collagen type IV content in cerebral microvessels in comparison to age-matched controls [ 13 ]. Farkas and colleagues described collagen accumulation in the basal lamina of Parkinson's disease [ 14 ]. Data about changes in the cerebral microvasculature in normal aging, however, are scarce. One study examined cerebral microvessels in human neocortex, yet failed to demonstrate an age-related thickening of the basal lamina [ 15 ], however one study was able to show a decrease of microvascular density by age in the hypothalamus [ 4 ], interestingly also this study failed to show hypertension related changes. One explanation for the decreased microvascular density in the putamen might be a reduced neoangiogenesis in the aging human brain, this hypothesis should be examined in further studies. One reason for the inconclusive findings in human cerebral microvessels might be the local origin of the samples, as up to now rarefaction was only found in the deep grey matter. The microvasculature is organized differently in the basal ganglia than in the cortex. The basal ganglia microvasculature has the geometry of a tree-like vascular bed, in contrast the cortical microvascular networks are rather organized like a grid structure [ 16 ]. Due to this reason the deep grey matter might be more vulnerable to metabolic changes in age, with reduced capabilities for neoangiogenesis, as this might compensatory occur in cortical areas. Therefore differential patterns of microvascular changes within the brain might occur, resulting in controversial results of microvascular density in the aging brain. Another explanation seems to be a methodological one. As the region of interest in putaminal sections can be exactly defined from section to section, analyses of cortical areas might be more problematic resulting in an imprecise definition of comparable areas [ 15 ]. The reasons for the alterations of vascular extracellular matrix proteins in aging are largely unknown. Not only age is associated with an increase of extracellular matrix proteins, as these changes were observed in several diseases, like hypertension, brain tumors [ 17 ], HIV-encephalopathy [ 18 ], Alzheimer's [ 13 ] or Parkinson's disease [ 14 ]. One explanation for the extracellular matrix accumulation might be the reduction of the proteolytic systems activity. The matrix metalloproteinases (MMP) and the natural tissue endogenous inhibitors (TIMP), as well as the plasminogen/plasmin system are involved in the regulation of ECM metabolism [ 19 , 20 ], and changes in these proteases activity may contribute to vascular remodelling in age by modulating the extracellular matrix components. Therefore a reduction of these proteases might result in a decreased turnover of the basal lamina with a consecutive increase of these components. The reduction of MMP activity by age was recently shown in an experimental study [ 21 ] and in humans by antihypertensive treatment [ 22 ], however studies about the role of these proteases in the aging human brain are lacking and the impact has to be evaluated in further studies. The strength of our study is the combination of several different methods for the detection of basal lamina changes. Even if the study by Abernethy and colleagues [ 4 ] was the first one, that showed age related changes in the deep grey matter, their study has some limitations. First they used a nonspecific staining technique with alkaline phosphatase, and second no detailed morphometric analysis of the vessels or changes of the basal lamina were performed. On the other hand, previously published studies most often used only one method for the determination of microvascular changes. For example the analyses of the vessel wall and lumen was widely used as the only parameter. [ 23 - 25 ] Unquestionably this method has a subjective approach and therefore we see our results of the increased vessel wall/lumen ratio in aging as a confirmation of previous results. But in contrast to these studies our approach employed an additional variety of complementary methods to examine the increase of basal membrane components: number of vessels, total stained area of collagen type IV, relative collagen type IV content per vessel and relative immunofluorescence by CLSM, all indicating into the same direction. Some methodological issues have to be discussed, starting with the unexpected lack of a difference between old normotensive and old hypertensive patients. One explanation might be due to silent hypertension in the ONHP group, as well as treated hypertension in the OHP group. We tried to minimize this problem due to carefully study of the case records in all patients. In addition in our study hypertension was defined as a history of hypertension, rather than the actual blood pressure in hospital, as these patients with severe diseases might not had representative blood pressure values in their last days or weeks of life than decades before. On the other hand another study failed to show the impact of hypertension on microvascular densitiy in a neuropathological study of the human hypothalamus [ 4 ]. This unexpected lack of hypertension related changes in the microvessels should be regarded as preliminary, as neither the duration of hypertension nor the effectiveness of treatment was considered. Conclusions The present data show age-related changes of the cerebral microvessels in sections of human putamen for the first time. Due to the accumulation of collagen, microvessels thicken and show a reduction in their lumen. Besides this, the number of vessels decreases. These findings might represent a precondition for the development of vascular cognitive impairment. Methods The study was performed on 27 post-mortem human brain samples from the putamen, which were taken from autopsy. The clinical diagnoses were confirmed by routine pathology and are shown in the table. Two groups of subjects were compared: first young patients, all without a history of hypertension (YP; n = 11, mean age 38.8 ± 6.8 years), and old patients (OP, n = 16, mean age 73.9 ± 4.1 years). The old patients were divided into two subgroups, those without a history of hypertension (ONHP, n = 8, mean age 73.1 ± 4.9 years), and those with a history of hypertension (OHP, n = 8, mean age 74.6 ± 3.4 years). There were no significant differences in age between OHP and ONHP and in sex between YP and OP, as well as between ONHP and OHP (see Table 1 ). The putamen either of the right or left side were removed completely and fixated in paraffin. We chose the putamen region, as it is easily to define and vascular changes and strokes are predominantly located in this area. The blocks were cut cross sectional in the same anterior-posterior direction resulting in axial sections with a thickness of 10 μm. The sections were deparaffinized and immersed at 37°C in 0.4% Pepsin (Sigma, Germany) in 0.01 N HCl for one hour. Collagen IV-positive vessels were stained with a monoclonal mouse anti-collagen-IV antibody (Sigma, Germany). Each section was incubated with 150 μl of the primary antibody solution (at a concentration of 1:200) for two hours at 37°C followed by incubation with biotinylated secondary antibody against mouse IgG for 30 minutes at 37°C (Vector Laboratories). Vectastain ABC reagent was added for 30 minutes at 37°C. Chromogen (AEC Kit Biomeda Corp.) was used to develop the peroxidase signal. Negative and positive controls were routinely performed in each staining experiment. The same procedure was used for immunofluorescence staining. Instead of using the Vectastain ABC kit containing avidin, avidin marked FITC (Dianova, Hamburg, Germany) was added for 30 minutes at a dilution of 1:100. The number of peroxidase stained vessels was determined with the aid of a computerized video imaging system at a magnification of ×100 (Optimas Version 6.5 from Media Cybernetics, Silver Spring, USA). Only vessels smaller than 30 μm were included. Total area of collagen IV positive vessels in the sections was analyzed using the same system. Results are presented in arbitrary units. To obtain the relative amount of collagen type IV per vessel the area of collagen type IV was divided by the number of stained vessels per microscopic field ([collagen type IV/microscopic field]/[vessels/microscopic field]) The size of the observed microscopic field was 150 × 200 μm. To estimate microvessel hypertrophy, the ratio between the diameter of vessel lumen and vessel wall, respectively, was calculated semiquantitatively with the help of a second computerized video imaging system (Medmo, Homburg, Germany). Twenty entire cross-sectional microvessels from the putamen were randomly selected at a magnification of ×400. To calculate the wall to lumen ratio, average distances of vessel wall and vessel lumen were selected. Fluorescence intensity measurements of microvessel-associated FITC anti-mouse IgG against the anti-collagen antibody were performed with confocal laser scanning microscopy (CLSM, Leica, Heidelberg, Germany). All measurements were performed with the same pinhole size, brightness and contrast, zoom, and laser time. Each vessel was scanned in the z plane (10 scans per 1 μm), and a summed image was calculated. Also, a summed image was obtained from the background area to normalize the local intensity to the background. The normalized intensity is expressed as mean ± SEM for each microvessel using a scale from 0 to 255 arbitrary units (U). The technique was adopted from Hamann et al [ 26 ]. Twenty randomly selected microvessels each of 7.5 to 30 μm in diameter of the putamen were measured in each specimen. Statistical analysis Data are presented as mean +/- standard error of mean. Statistical evaluations were performed using t-test. Authors' contributions OU carried out the immunohistochemical experiments, ML performed the statistical analysis and drafted the manuscript. JH participated in the design of the study and collected the brain specimens. AD participated in the study design. GFH supervised the thesis, and participated in its design and coordination.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC523851.xml
535941	Comparison of nested PCR and real time PCR of Herpesvirus infections of central nervous system in HIV patients	Background Molecular detection of herpesviruses DNA is considered as the reference standard assay for diagnosis of central nervous system infections. In this study nested PCR and real time PCR techniques for detection of Herpes simplex virus type 1 (HSV-1), Cytomegalovirus (CMV) and Epstein-Barr virus (EBV) in cerebrospinal fluid of HIV patients were compared. Methods Forty-six, 85 and 145 samples previously resulted positive for HSV-1, CMV and EBV by nested PCR and 150 randomly chosen negative samples among 1181 collected in the period 1996–2003 were retrospectively reassessed in duplicate by real time PCR and nested PCR. Results Samples giving positive results for CMV, HSV-1 and EBV with nested PCR were positive also with real time PCR. One of the negative samples resulted positive for HSV and one for EBV. Real time PCR showed comparable sensitivity and specificity vs nested PCR. Conclusion Real time PCR proved to be a suitable method for diagnosis of herpesvirus infections in CNS, showing comparable sensitivity and being less time consuming than nested PCR.	Background Opportunistic infections as well as tumors and vascular and metabolic disorders are common in HIV-infected patients [ 1 - 3 ]. Generally, opportunistic viral infections are caused by a broad spectrum of different species with similar clinical patterns, especially those affecting the central nervous system (CNS), where differential diagnosis requires simultaneous screening of a wide range of different viruses [ 4 , 5 ]. Moreover, immunodeficiency induced by HIV infection favours reactivation of herpesviruses which could cause important diseases by themselves [ 6 ]. Although the rate of CNS complications is relatively low, if compared with the high prevalence of herpesviruses in the population, these viruses represent the most important pathogens associated with viral encephalitis and meningitis [ 7 , 8 ], being cytomegalovirus (CMV) the most frequently identified virus in HIV-positive patients, followed by Epstein-Barr virus (EBV) and Herpes simplex virus type 1 (HSV-1) [ 9 ]. CMV, that is often responsible of asymptomatic infections in immunocompetent host, is able to cause serious manifestations such as retinitis, pneumonia and encephalitis in presence of an alteration of immunoresponse, while the recovery of Epstein-Barr virus (EBV) in cerebrospinal fluid (CSF) seems to be a prognostic sign for the development of cerebral tumors in patients with AIDS [ 10 , 11 ]. HSV-1 is the most commonly detected virus in diagnostic laboratory, being cause of a variety of clinical symptoms in different anatomical sites such as skin, lips, oral cavity and, especially in immunocompromised patients, CNS [ 12 ]. One-step or nested polymerase chain reaction (PCR) has rapidly replaced immunological assays based on virus specific Ig antibodies in CSF for laboratory diagnosis of Herpesvirus infections, even if serological methods are considered an additional tool for defining clinical diagnosis. Although nested PCR is considered the method of choice in terms of specificity [ 9 , 13 , 14 ], some additional aspects should be considered. In the last years, introduction of real time PCR has markedly increased the ease and the speed in the virology laboratory due to the relevant technology that permits rapid temperature cycling within a close system. Considering the importance of relationship between viral load in CSF and severity and outcome of disease, an additional advantage of Real Time PCR is the capability to perform simultaneous qualitative and quantitative analysis. Here is reported our experience gained in the diagnosis of herpesvirus infections of the CNS in HIV patients by means of nested PCR and Real Time PCR, which has been recently applied in our laboratory. Methods CSF samples collection A total of 1181 CSF samples collected in the period 1996 – 2003 from HIV patients attending at Luigi Sacco Teaching Hospital of Milan (Italy) affected by acute encephalitis or meningitis or encephalopathy or other neurological syndromes were considered. Particularly, they consisted of 684, 954 and 933 CSF samples previously tested for HSV-1, CMV and EBV, respectively, by means of nested PCR. Of these, all the positive samples and 150 negative samples randomly chosen were retested by means of nested PCR and real time PCR. Each sample was re-extracted and run in duplicate. CSF positive and negative samples were stored at -80°C until analysis. Clinical data of patients with positive samples were available only in the 20% of the total. These patients generally recorded meningoencephalitis signs such as central motor or sensory alterations, consciousness loss, seizures defects. Nucleic acid extraction Spin-column based QIAamp Mini Kit (Quiagen, Hilden, Germany) protocol extraction for CSF was used as indicated by the manufacturer. This procedure allowed for the rapid purification of DNA from 200 μL of CSF and comprised four successive steps carried out using QIAamp Spin Columns in a standard microcentrifuge. Purified DNA was concentrated at a final volume of 20 μL. Nested PCR Nested PCR was carried out in a 50 μl mixture containing 40 μl of first amplification mix (outer primers, buffer, dNTP)-(Amplimedical SpA-Bioline Division-Italy), 2U/μl Taq DNA polymerase (Roche Diagnostics-Germany) and 5 μl of purified DNA. Primer pairs selecting for glycoprotein D gene of HSV-1 [ 15 ], for the late protein gp58 of CMV [ 16 ], and for Bam HI-W region of EBV [ 17 ] are shown in Table 1 . After an initial 2 min denaturation at 94°C, 35 cycles of 94°C for 30 sec, 55°C for 30 sec and 72°C for 30 sec were carried out, followed by a 5 min extension at 72°C using a thermal cycler (Gene Amp PCR System 2400-Applied Biosystem – Monza – Italy). The reaction mixture for the second amplification round was the same as for the first one, except for the "inner" primers used instead of the "outer" primers. In the second amplification round 44 μl of amplification mix and 1 μl of the first amplification round PCR product were used. The thermal cycling was repeated as for the first amplification round but using 30 cycles after the initial 2 min denaturation. Each amplification run contained a negative control, consisting of water and a positive plasmidial control. Analysis for the PCR products was performed by means of 4 % agarose gel electrophoresis followed by visualization with ethidium bromide (0.4 μg/mL) staining and UV illumination to confirm the expected products. Real Time PCR For diagnostic real time PCR Taq polymerase RT PCR Kit (Amplimedical SpA-Bioline Division-Turin Italy) was used. Target regions for HSV and CMV were the same as in nested PCR, while EBNA-1 gene was amplified for EBV [ 9 ] as shown in Table 2 .. The RT PCR was performed in 25 μl mixture containing 20 μl of amplification mix (buffer, dNTPs, Taq gold polymerase, Rox passive fluorocrome, primers and MGB Eclipse probe) and 5 μl of purified DNA. The amplification program included an initial decontamination with uracile N'-glycosilase at 50°C for 2 min, followed by denaturation at 95°C for 10 min and 45 two steps of 15 sec at 95°C and 1 min at 60°C. The RT PCR products were detected by measuring fluorescence with passive reference dye in Sequence Detection System ABI Prism 7000 (Applied Biosystem). Control threshold (C t ) values were calculated by determining the point at which the fluorescence exceeded a background limit of 0.04. Each analytical session comprised also a negative control (distilled water). Quantification was carried out by analysing four positive plasmidial standards at 10 2 , 10 3 , 10 4 and 10 5 copies/reaction. The standards were obtained by cloning the target amplification product in a plasmid, which was transformed and cultured in Escherichia coli . Plasmidic DNA was purified with a commercial kit (Qiagen) and its concentration determined spectrophotometrically. Then, plasmidic DNA was serially diluted in a stabilizing buffer to the final desired concentration. Amounts of copies/mL in each sample were determined by means of a quantification software (Amplimedical), by considering an extraction recovery of 80%. Results Nested PCR Of 954 CSFs previously examined for CMV by means of nested PCR, 85 samples resulted positive. Among the 684 CSFs tested for HSV-1, 46 samples were found positive, while, 145 of 933 CSFs tested resulted positive for EBV. Reassessment of these positives and of 150 negative samples confirmed results previously obtained with the same method, with the exception of one HSV-1 positive sample, which resulted negative when retested, as shown in Table 3 . Real Time PCR Results from real time PCR are reported in Table 3 . All the samples for which nested PCR gave positive results were confirmed by Real Time-PCR for CMV, HSV-1 and EBV. Of the 150 samples resulted negative by nested PCR, 148 were negative, while 1 sample resulted positive for HSV and 1 for EBV. The positive sample giving negative result when reassessed by nested PCR was negative also by real time PCR. Sensitivity and specificity By considering nested PCR as gold standard [ 9 , 13 ], sensitivity and specificity of real time PCR are described in Table 4 . Comparable sensitivity (100%) and specificity (99–100 %) were found for real time PCR in respect to nested PCR. Discussion Introduction of PCR into routine diagnostic has rapidly gained a pivotal role for diagnosis of a wide range of diseases, supplanting, in many cases, other methods, such as the classical serodiagnosis. This is particularly true for diagnosis of herpes virus infections in immunocompromised patients, where diminished or suppressed virus-specific antibody responses do not reflect possible reactivated herpes mediated aethiologies [ 18 ]. Real time PCR has represented a further step forward, since it allows for quantitative detection of target DNA in a single sample over a large range, remaining possible qualitative detection. Even if contradictory results have been found [ 19 - 21 ], quantification of DNA could represent an important issue to evaluate the severity and outcome of herpesvirus encephalitis, and it may be also used to monitor the success of antiviral therapy. This strategy has already been used for monitoring of patients at risk for CMV infections, when viral load kinetic patterns are used to identify patients who are more likely to have recurrence of CMV disease after the initiation of therapy, as well as to identify patients needing treatment [ 22 - 24 ]. Moreover, in these cases, use of a highly sensitive assay could be of crucial value. In the last years several real-time PCR methods have been developed for detection of herpesviruses in different biological specimens [ 25 - 27 ]. Real-time PCR has been well recognized to offer several advantages over nested PCR other than allowing quantification of viral load: it reduces the risk of amplicon contamination, being a close-system, is a safer laboratory protocol by not using ethidium bromide, and it allows a notable reduction of time required for response. In the present work, we compared a real time PCR panel for detection of herpesvirus DNA in CSF with methods employing nested PCR. Our results indicate an overall agreement between the two methods, as reported by other authors [ 9 , 27 , 28 ]. Differences between the two assays were observed for HSV-1 and EBV analysis, where one negative sample was found positive by real time PCR. Since extraction panels and primers for HSV and CMV were uniform for the both types of assays, inhibitors present in the DNA preparations could not explain the different results obtained for these samples. Moreover, being nested PCR generally considered as the gold standard for diagnosis of herpes virus in CSF, we tried to use similar primers, chemistry and amplification conditions in order to limit differences for a better comparison of the performance of the two methods. Thus, discrepancies could be likely attributed to the different detection of amplification products, although real time PCR has been reported to be as sensitive as nested PCR [ 29 ]. This suggestion is also supported by the fact that the two patients with CSF negative for HSV-1 and EBV with nested PCR showed clinical syndromes compatible with viral encephalitis and clinically improved with antiviral treatment (data not shown). These data seem to suggest that real time PCR could be more sensitive than nested PCR. Since the limit of detection is generally calculated by using plasmidial DNA, it may be possible that, although molecular sensitivity is reported to be similar for nested PCR and real-time PCR [ 27 , 28 ], some differences may occur for biological specimens. The sample classified positive for HSV-1, which resulted negative when retested with both methods, was one of the oldest in our collection, dating 1996, and it might have degraded over the 7 years storage. From this point of view, development of standardized quality controls might be very helpful [ 30 ]. Conclusions Data obtained in this study confirms the validity of real-time PCR method for detection of herpesvirus DNA in CSF specimens of HIV patients, being sensitive, rapid and quantitative. Since specific and rapid diagnosis is the main target in the case of CNS infections, real time PCR could be considered the method of choice, due to its high specificity, sensitivity and rapidity, once proper quality controls will be available. Competing interest The company Amplimedical SpA-Bioline Division-Italy will pay the article processing fees for the manuscript. Authors' contributions LD conceived of the study and participated in its design and coordination, AL carried out real time PCR, EDV participated in data analysis and drafted the manuscript, GG performed statistical analysis, RB carried out nested PCR, MRG participated in design and coordination of the study. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535941.xml
546011	Doxorubicin increases the effectiveness of Apo2L/TRAIL for tumor growth inhibition of prostate cancer xenografts	Background Prostate cancer is a significant health problem among American men. Treatment strategies for androgen-independent cancer are currently not available. Tumor necrosis factor-related apoptosis-inducing ligand (Apo2L/TRAIL) is a death receptor ligand that can induce apoptosis in a variety of cancer cell lines, including androgen-independent PC3 prostate carcinoma cells. In vitro , TRAIL-mediated apoptosis of prostate cancer cell lines can be enhanced by doxorubicin and correlates with the downregulation of the anti-apoptotic protein c-FLIP. This study evaluated the effects of doxorubicin on c-FLIP expression and tumor growth in combination with Apo2L/TRAIL in a xenograft model. Methods In vitro cytotoxic effects of TRAIL were measured using a MTS-based viability assay. For in vivo studies, PC3 prostate carcinoma cells were grown subcutaneously in athymic nude mice and tumor growth was measured following treatment with doxorubicin and/or Apo2L/TRAIL. c-FLIP expression was determined by western blot analysis. Apoptosis in xenografts was detected using TUNEL. Statistical analysis was performed using the student t-test. Results In vitro experiments show that PC3 cells are partially susceptible to Apo2L/TRAIL and that susceptibility is enhanced by doxorubicin. In mice, doxorubicin did not significantly affect the growth of PC3 xenografts but reduced c-FLIP expression in tumors. Expression of c-FLIP in mouse heart was decreased only at the high doxorubicin concentration (8 mg/kg). Combination of doxorubicin with Apo2L/TRAIL resulted in more apoptotic cell death and tumor growth inhibition than Apo2L/TRAIL alone. Conclusions Combination of doxorubicin and Apo2L/TRAIL is more effective in growth inhibition of PC3 xenografts in vivo than either agent alone and could present a novel treatment strategy against hormone-refractory prostate cancer. The intracellular mechanism by which doxorubicin enhances the effect of Apo2L/TRAIL on PC3 xenografts may be by reducing expression of c-FLIP.	Background Prostate cancer is a significant health problem among American men. This year about 230,110 men will be diagnosed with the disease and about 30,000 will likely die in this country alone [ 1 ]. Treatment options are limited and are associated with significant morbidity and mortality [ 2 ]. Localized cancer is treated with radical prostatectomy, brachy- or cryotherapy, and external beam radiation while cancer that has escaped the prostatic capsule is generally treated by androgen ablation. However, the eventual development of an androgen-independent phenotype leads to incurable disease, indicating the need for better treatment strategies. Experimental approaches include delivery of oncolytic viruses, immunomodulatory molecules, p53 and p21, enzymes that metabolize prodrugs and agents that can induce apoptosis [ 3 ]. One agent that has received considerable attention as a novel apoptosis-inducing agent is tumor necrosis factor-related apoptosis inducing ligand (TRAIL/Apo2L). TRAIL is a type II membrane protein that can induce apoptosis by binding to death domain containing receptors DR4 and DR5 [ 4 ]. Unlike other death receptor ligands such as TNF and FasL, which cause septic shock and hepatotoxicity, respectively, TRAIL is tolerated well in mice and non-human primates [ 5 - 7 ]. TRAIL induces apoptosis in a variety of cell lines in vitro and in vivo . Apoptosis is initiated by binding to receptors DR4/DR5 (TRAIL-R1, 2), which is followed by assembly of the death inducing signaling complex and activation of caspase-8. Subsequent activation of caspases-3/7 (in a mitochondria-dependent or independent fashion) leads to execution of apoptosis [ 4 ]. Active caspase-8 tetramers are generated by cis- and transcatalytic cleavage from pro-caspase-8 homodimers [ 8 ]. These cleavage steps are inhibited by heterodimer formation of caspase-8 with c-FLIP. The apoptosis inhibitor c-FLIP is structurally similar to caspase-8 but lacks the cysteine residue essential for catalytic activity. A strong correlation between c-FLIP expression and malignant potential has been observed in carcinomas of the colon and liver as well as melanomas [ 9 - 11 ]. In addition, a high c-FLIP/caspase-8 ratio has been associated with resistance to death receptor-mediated apoptosis [ 11 - 13 ]. Thus downregulation of c-FLIP is a desirable strategy to enhance the apoptotic response to death receptor ligands. We have previously shown that prostate cancer cells are relatively resistant to recombinant TRAIL but can be sensitized by pretreatment with the chemotherapeutic agent doxorubicin [ 7 , 14 ]. The enhanced susceptibility of prostate cancer cells was independent of androgen-phenotype or p53 status but correlated with doxorubicin-mediated downregulation of c-FLIP expression. In this study we extended those observations to an in vivo model using xenografts of PC3 prostate carcinoma cells. Our results show that doxorubicin reduces c-FLIP expression in xenografts and that combination of doxorubicin with TRAIL is more effective in tumor growth inhibition than either agent alone. Methods Cells and reagents PC3 cells were purchased from the ATCC and were maintained in RPMI1640 supplemented with 10% FBS at 37°C with 5% CO 2 . Apo2L/TRAIL was generously provided by Genentech Inc., San Francisco CA. KillerTRAIL was purchased from Alexis, San Diego, CA. Doxorubicin was obtained from the MUSC pharmacy. The CellTiter Aqueous One Solution Cell Proliferation Assay and DeadEndTUNEL kits were purchased from Promega, Madison WI. The c-FLIP antibody NF-6 was kindly provided by Dr. Marcus Peter, University of Chicago. The Dave-2 (c-FLIP) antibody and anti-actin were purchased from Alexis, San Diego, CA and from Sigma, St. Louis, MO, respectively. Supersignal DuraWest was obtained from Pierce Biotechnology Inc., Rockford, IL. MTS viability assay Cells were seeded into 96-well plates at 1 × 10 4 cells/well, incubated overnight and subsequently treated with doxorubicin and/or Killer-TRAIL/Apo2L/TRAIL. The MTS [3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium] reagent was added 24 hours after initiation of treatment and plates read at an absorbance of 490 nm one to two hours later using a Vmax kinetic microplate reader (Molecular Devices, Sunnyvale, CA). All treatments were performed in triplicate. Background absorbance was determined by incubating media with substrate alone and subtracting the values from wells containing cells. Percent cytotoxicity was calculated as follows: % cytotoxicity = 1 - [(OD of experimental/OD of control) × 100]. Experiments were repeated three times with similar results. There were no discrepancies between results obtained in the MTS assay and visual assessment of the cells prior to adding the MTS reagent. Animal experiments Athymic male nude mice (3–4 weeks old) were purchased from Harlan, Indianapolis IN and were housed under pathogen-free conditions according to Medical University of South Carolina animal care guidelines. All animal experiments were reviewed and approved by the Institutional Animal Care and Use Committee at MUSC. PC3 cells (4 × 10 6 ) were injected subcutaneously into the flanks of mice. Animals bearing tumors were randomly assigned to treatment groups (five or six mice per group) and treatment initated when xenografts reached volumes of about 100 mm 3 . Tumors were measured using digital calipers and volume calculated using the formula: Volume = Width 2 × Length × 0.52, where width represents the shorter dimension of the tumor. Treatments were administered as indicated using vehicle (PBS containing 0.1% BSA), doxorubicin (2–8 mg/kg), Apo2L/TRAIL (500 μg/animal), or a combination of 4 mg/kg doxorubicin followed by 500 μg Apo2L/TRAIL. Doxorubicin was administered systemically whereas Apo2L/TRAIL was given either intra-tumorally (Fig 4 ) or systemically (Fig 5 ). All treatments were given once. Mice were monitored daily for signs of adverse effects (listlessness and scruffy apparance). Treatments seemed to be well tolerated. The mean ± SEM was calculated for each data point. Differences between treatment groups were analyzed by the student t-test. Differences were considered significant when P < 0.05. Western blotting Following doxorubicin treatment, mice were sacrificed, tissue removed and immediately frozen in liquid nitrogen. Protein was prepared in RIPA buffer containing freshly added mammalian protease inhibitor cocktail. Lysates were stored at -70°C and were centrifuged (20,000 × g) prior to performing protein assays on the supernatant. Protein (50 μg) was separated on 4–12% Bis/Tris NuPage gels in MES buffer and transferred to nitrocellulose for 60–90 minutes at 30 V. After transfer and blocking (5% milk), blots were probed with Dave-2 (1:500 in 0.5 % milk) or NF-6 (1:5 in TBS-Tween) overnight at room temperature. Following three washes with TBS-Tween, membranes were incubated with the anti-mouse (1:5,000) or anti-rat (1:15,000) HRP-conjugated secondary antibodies for 1 hour at room temperature in 5% milk in TBS-Tween. Membranes were washed three times in TBS-Tween followed by chemiluminescent detection of the secondary conjugates with DuraWest Supersignal. Membranes were reprobed with anti-actin (1:2000) to ensure equal loading. TUNEL staining Tumors were excised and fixed in 10% formalin and processed by standard procedures. Sections were analysed for apoptotic cells using the DeadEND Tunel kit according to the manufacturers instructions. Stained slides were examined for TUNEL positive cells using a Zeiss Axiovert 200 microscope (20×). TUNEL positive cells from 4 fields/slide were counted by two investigators to calculate the mean ± SEM. P-values were calculated using the student's t-test. Results Comparison of KillerTRAIL and Apo2L/TRAIL in vitro We have previously shown that the TRAIL apoptotic response in prostate cancer cells can be enhanced by doxorubicin [ 14 , 15 ]. These studies were conducted using KillerTRAIL, a commercially available form of the protein that is crosslinked for maximal activity and contains a histidine tag. Concerns about hepatotoxicity were raised when a polyhistidine tagged recombinant version of TRAIL induced apoptosis in human hepatocytes [ 16 ]. A subsequent study revealed that different recombinant versions of TRAIL vary widely in their biochemical properties and their potential to cause toxicity. TRAIL containing a histidine tag (Apo2L/TRAIL.His) contained less Zinc, had a less ordered conformation and was more heterogeneous than TRAIL without a tag (Apo2L/TRAIL.0) [ 17 ]. In contrast to Apo2L/TRAIL.His, Apo2L/TRAIL.0 was non-toxic to human hepatocytes. Thus non-tagged Apo2L/TRAIL.0 (referred to as Apo2L/TRAIL) is the preferable form to use in preclinical studies. Initially we compared the susceptibility of PC3 prostate carcinoma cells to KillerTRAIL and Apo2L/TRAIL in parallel assays in vitro . As shown in Figure 1A , at 1000 ng/ml KillerTRAIL resulted in about 50% cytotoxicity whereas less than 20% cytotoxicity was obtained with Apo2L/TRAIL. This difference may stem from the histidine tag of KillerTRAIL. In the presence of doxorubicin, KillerTRAIL and Apo2L/TRAIL were about equally effective (Fig 1B–D ). Doxorubicin lowered the concentration requirement for TRAIL with near maximal killing achieved at 10 ng/ml ligand. Effect of doxorubicin on growth and c-FLIP expression in vivo To determine an appropriate dose of doxorubicin for the in vivo experiments, mice bearing PC3 xenografts were injected with 2, 4 or 8 mg/kg doxorubicin and tumor volume was measured over time (Figure 2 ). A dose of 2 mg/kg did not affect tumor growth while higher dosages delayed tumor growth initially (p < 0.05 at days 18 and 22). However, no statictically significant differences were detected between untreated and doxorubicin treated groups at later time points. Previously, we have examined possible targets of doxorubicin that may be responsible for increasing the susceptibility of prostate cancer cells to TRAIL in vitro . These included TRAIL receptors, Bax, Bcl-2, Bcl-xl, and c-FLIP [ 15 ]. In PC3 cells, the only change observed in response to doxorubicin was a decrease in c-FLIP that correlated with onset and magnitude of caspase-8 activation and apoptosis. To investigate whether doxorubicin would have a similar effect on c-FLIP in vivo , protein from PC3 xenografts was isolated for western blotting 24 hours after systemic delivery of the drug. As shown in Figure 3A , we found that either 4 mg/kg or 8 mg/kg doxorubicin significantly reduced levels of c-FLIP in PC3 xenografts. The antibody used for the detection of c-FLIP (NF-6) is human-specific and thus detects only c-FLIP of PC3 origin. We also investigated the effect of doxorubicin on c-FLIP in the heart from the same mice. The heart was chosen because it expresses high levels of c-FLIP [ 18 ]. In addition, c-FLIP-deficient mice do not survive past day 10.5 of embryogenesis due to impaired heart development [ 19 ], indicating that c-FLIP plays a critical role in this organ. Protein isolated from the heart was analyzed with the rat monoclonal antibody Dave-2 that detects both mouse and human c-FLIP, although all c-FLIP detected should be exclusively of mouse origin, since PC3 cells are grown subcutaneously. Two major proteins, each migrating somewhat slower than the human c-FLIP isoforms, were detected with the Dave-2 antibody (Fig. 3B ). The signal of the lower band (c-FLIP S ) was not diminished following doxorubicin treatment, whereas the signal of the upper band (c-FLIP L ) was slightly reduced following administration of 8 mg/kg doxorubicin. Since 4 mg/kg doxorubicin reduced c-FLIP in PC3 xenografts as effectively as 8 mg/kg without affecting endogenous mouse c-FLIP, this dose was chosen for combination therapy. Effectiveness of Apo2L/TRAIL doxorubicin combination therapy in vivo To determine apoptosis in vivo following treatments, mice bearing bilateral PC3 xenografts were injected systemically with vehicle or 4 mg/kg doxorubicin, followed by intratumoral injection of vehicle (left tumor) or Apo2L/TRAIL (right tumor). Twenty-four hours after Apo2L/TRAIL injection, tumors were harvested and analysed for apoptosis by TUNEL staining (Fig. 4 ). There was no significant difference in TUNEL-positive cells in control and doxorubicin treated tumors. Treatment with Apo2L/TRAIL resulted in 70% more apoptotic cells than in vehicle treated tumors. However, combination treatment of doxorubicin and Apo2L/TRAIL yielded the most TUNEL positive cells (278% compared to the control) and was statistically different from all other treatment groups. Next, we determined whether this pattern would be reflected in tumor growth of PC3 xenografts. Animals received systemic admininstration of doxorubicin followed by systemic injection of TRAIL after 24 hours, when c-FLIP levels were expected to have decreased. Tumors in animals injected with doxorubicin reached volumes of approximately 1000 mm 3 within 26 days after treatment initiation (1025 ± 138 mm 3 ), which was not significantly different from the control group (1025 ± 118 mm 3 , p = 0.41). Tumors in animals treated with Apo2L/TRAIL were reduced compared to either untreated or doxorubicin treated groups (833.5 ± 150 mm 3 , p = 0.03). However, tumors that had been exposed to both doxorubicin and Apo2L/TRAIL were significantly smaller (224 ± 145 mm 3 , p < 0.001) and remained smaller at day 33 (490 ± 197 mm 3 ) when the experiment was terminated. Analysis of tumor volume over time suggests that a single round of combination treatment delays growth until day 26 (Fig 5C ). After tumors escape this delay they resume growth at rates similar to the other groups. Discussion In this study we have shown that combination of Apo2L/TRAIL and doxorubicin is more effective in retarding tumor growth of PC3 prostate carcinoma xenografts than either agent alone. Doxorubicin is an anthracycline, which intercalates into DNA thereby activating DNA repair pathways and elevating levels of p53. PC3 cells are p53 -/- , which may explain why doxorubicin as a single agent was relatively uneffective against tumor growth inhibition [ 20 ]. However doxorubicin was able to reduce expression of the anti-apoptotic protein c-FLIP in vivo . We have previously shown that sequential treatment of doxorubicin followed by TRAIL resulted in cell death in vitro [ 15 ]. Our data suggest that downregulation of c-FLIP is an important step in vivo that enhances TRAIL-induced apoptosis as evidenced by increased TUNEL staining in tumors treated with combination therapy. Other agents that reduce c-FLIP and increase death ligand mediated apoptosis include cycloheximide, 9-nitrocamptothecin, cisplatin, and the proteasome inhibitor PS-341 [ 21 - 24 ]. These agents affect multiple cellular responses suggesting that downregulation of c-FLIP may not be the sole factor by which death ligand-induced apoptosis is enhanced. However, selective reduction of c-FLIP using RNA interference or anti-sense technology is sufficient to sensitize human cell lines, including Du145 prostate carcinoma cells to death receptor ligands, indicating that c-FLIP is a major provider of resistance in this apoptotic pathway [ 25 , 26 ]. Tumors in mice that received Apo2L/TRAIL alone were about 20% smaller than those of the control group. Thus the growth inhibitory effect of Apo2L/TRAIL on PC3 xenografts in vivo was reflective of the results obtained in vitro . In contrast to single agent therapy, combination treatment with doxorubicin and TRAIL resulted in tumors that had 80% less volume than those in the control group. Tumors that received combination therapy continued to grow slowly, indicating that complete regression of PC3 xenografts was not achieved using a single treatment. In a recent study, PC3 xenografts were treated with irradiation followed by TRAIL, which resulted in complete growth inhibiton following three weekly rounds of therapy [ 27 ]. This indicates that multiple treatments may be neccessary to achieve a complete response. We observed that a single administration of 4 mg/kg doxorubicin reduced cFLIP protein in PC3 xenografts but not mouse heart. One possibility is that cFLIP expression is differentially affected in normal versus malignant cells. If so, then doxorubicin would preferentially lower the apoptotic threshold in cancer cells and facilitate the selective elimination of these cells. Alternatively, the difference may be species specific. When grown in mice, human xenografts of various origins have successfully been treated by TRAIL in combination with other agents [ 5 , 24 , 27 - 29 ]. In contrast, combination of doxorubicin and TRAIL in vitro can induce cell death in normal human breast, mesothelial, or prostate epithelial cells, [ 14 , 30 , 31 ], which suggests that combination therapy may lack specificity when applied to humans. Could combination therapy in mice be non-toxic because endogenous mouse c-FLIP remains unaffected by chemotherapy? To avoid the possible complication of toxicity the effect of chemotherapy on c-FLIP expression in human patients should be carefully evaluated before considering systemic combination of chemotherapy and Apo2L/TRAIL. Conclusions We found that combination of doxorubicin chemotherapy and Apo2L/TRAIL is more effective in tumor growth inhibition than either agent alone, indicating that this may represent a novel treatment strategy against prostate cancer. One of the mechanisms by which doxorubicin may enhance Apo2L/TRAIL apoptosis in PC3 xenografts is reduced expression of the anti-apoptotic protein c-FLIP. Future studies are needed to further investigate the effect of chemotherapy on c-FLIP in human patients and to optimize the treatment schedule to achieve complete tumor regression. Competing interests The author(s) declare that they have no competing interests Authors' contribution AZ carried out the animal experiments using doxorubicin and Apo2L/TRAIL and TUNEL staining. JM carried out the animal experiments using doxorubicin. CVJ carried out viability assays and western blot analysis, conceived of the study, participated in its design and coordination, and prepared the manuscript. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC546011.xml
538256	Expression of leukemia inhibitory factor (LIF) and its receptor gp190 in human liver and in cultured human liver myofibroblasts. Cloning of new isoforms of LIF mRNA	Background The cytokine leukemia inhibitory factor (LIF) mediates its biological effects through binding to its high affinity receptor made of the low-affinity LIF receptor subunit gp190 (LIF-R) and the gp130 subunit. LIF exerts several important effects in the liver, however, data on liver expression of LIF are scarce. The aim of this study was to examine the expression of LIF and LIF-R in human liver. Results LIF expression, analyzed by immunohistochemistry, was barely detectable in normal liver but was strong within cirrhotic fibrous septa and was found in spindle-shaped cells compatible with myofibroblasts. Accordingly, cultured human liver myofibroblasts expressed high levels of LIF as shown by ELISA and Northern blot. Biological assay demonstrated that myofibroblast-derived LIF was fully active. RT-PCR showed expression of the LIF-D and M isoforms, and also of low levels of new variants of LIF-D and LIF-M resulting from deletion of exon 2 through alternative splicing. LIF receptor expression was detected mainly as a continuous sinusoidal staining that was enhanced in cirrhotic liver, suggestive of endothelial cell and/or hepatocyte labeling. Immunohistochemistry, flow cytometry and STAT-3 phosphorylation assays did not provide evidence for LIF receptor expression by myofibroblasts themselves. LIF secretion by cultured myofibroblasts was down regulated by the addition of interleukin-4. Conclusions We show for the first time the expression of LIF in human liver myofibroblasts, as well as of two new isoforms of LIF mRNA. Expression of LIF by myofibroblasts and of its receptor by adjacent cells suggests a potential LIF paracrine loop in human liver that may play a role in the regulation of intra-hepatic inflammation.	Background Leukemia inhibitory factor (LIF) belongs to the interleukin (IL)-6 family of cytokines, together with IL-11, ciliary neurotrophic factor, cardiotrophin-1, oncostatin M and neurotrophin-1/B cell stimulating factor-3. LIF is widely expressed in tissues and in many isolated cells. LIF expression is commonly up-regulated during inflammation. Nevertheless, its role seems to be complex as both pro- and anti-inflammatory properties have been described for that cytokine. Although LIF, like IL-6, is able to drive a significant acute-phase reaction in non-human primates [ 1 ], this has been questioned in humans [ 2 ]. LIF exerts its biological activities through its binding to a hetero-oligomeric receptor complex between the low-affinity LIF receptor subunit gp190 and the signal-transducing subunit gp130. The gp130 subunit is common to all members of the IL-6 family. Several isoforms of LIF consecutive to alternative splicing have been described. The second and third exons are common to all isoforms, whereas there are 3 alternate first exons – D, M, and T. The fate of the mature LIF molecule is highly dependent on exon 1 usage; thus, the human LIF-D transcript encodes a secreted protein that is biologically active and can signalize via the LIF receptor. The human LIF-M transcript does not contain any in-frame AUG, but it is known to be translated into both secreted and intracellular proteins [ 3 ]. The secreted LIF-M protein can also be found sequestered in the extracellular matrix where it is biologically active [ 4 ]. Finally, the first exon from the human LIF-T, which does not contain any in-frame AUG, is responsible for the synthesis of an intracellular protein with a leucine zipper motif that might function as a transcription factor [ 5 ]. As outlined above, LIF is potentially involved in liver physiology and pathophysiology; however, data on liver expression of LIF are scarce. LIF expression was not detected in normal rat liver but it was highly induced following partial hepatectomy, mainly in non- parenchymal cells [ 6 ], suggesting its involvement in liver regeneration. To our knowledge, the expression of LIF has not been described in human liver. Therefore, the aim of this study was to examine the expression of LIF and of its specific receptor gp190 in human liver. Results obtained with immunostaining of liver sections led us to examine LIF expression by cultured liver myofibroblasts, cells that play a major role in liver fibrogenesis. Results LIF expression Human liver tissues were examined for LIF expression by immunohistochemistry. In normal liver, a faint but consistent LIF expression was detected in the stroma of portal tracts (Fig. 1A ). No signal was observed along sinusoids. In fibrotic liver tissues, an intense expression of LIF was seen along fibrous septa which is consistent with the presence of myofibroblasts (Fig. 1B ). Staining adjacent sections with LIF antibody and with an antibody to alpha-smooth muscle actin (that labels myofibroblasts) suggested a large degree of colocalization (Figs. 1C,1D ). Part of the LIF staining also appeared to be extracellular. There was no difference in the type of staining whatever the etiology of liver fibrosis. No labeling was found when the LIF antibody was replaced by a species-matched control antibody. Figure 1 Immunohistochemical analysis of LIF expression in normal and cirrhotic human liver. ( a ): LIF expression is seen in normal liver in the stroma of portal tracts (arrows); ( b ): LIF is strongly expressed in fibrotic septa in cirrhotic liver (arrows); ( c ) and ( d ): consecutive sections of a cirrhotic liver analyzed for LIF ( c ) or alpha-smooth muscle actin ( d ) expression. No labeling was seen when the antibodies were replaced by a species-matched control antibody. Analysis of total RNA from cultured human liver myofibroblasts by Northern blot revealed a single 4.5 kb transcript (Fig. 2A ). RT-PCR experiments, described in more detail later, demonstrated the expression of both D and M isoforms of LIF (Fig. 2B ). When cell supernatants were tested with an ELISA assay specific for human LIF, levels ranged between 800 and 8 000 ng/ml in different isolates. In order to make sure that this corresponded to biologically active LIF, the supernatants were tested for their ability to promote the growth of the LIF-dependent cell line BaF3, stably transfected with the human gp190 and gp130 isoforms. As shown in Fig. 3 , myofibroblasts supernatants efficiently stimulated the growth of these cells in a dose-dependent fashion, confirming that biologically active LIF was effectively produced. Furthermore, the effect on BaF3 transfectants growth was abolished in the presence of the blocking gp190 LIF receptor antibody 12D3. The results of the ELISA combined with the 100 fold inhibition of biological activity, seen after anti-gp190 addition, further confirmed that most of the BaF3 growth-promoting activity produced by cultured myofibroblasts is likely to be LIF. Figure 2 Detection of LIF transcripts in cultured human liver myofibroblasts. ( A ): Northern blot. Total RNA from cultured human liver myofibroblasts was hybridized with a cDNA probe to human LIF. A single 4.5 kb band was observed; ( B ) and ( C ): RT-PCR. Total RNA was subjected to reverse transcription then to PCR with the hLIF-D3/hLIF-N4 ( B ) or with the hLIF-M3/hLIF-N4 primers ( C ). Figure 3 Biological activity of myofibroblast-derived LIF. BaF3 cells stably transfected with the gp130 and the gp190 subunits were exposed to dilutions of recombinant human LIF (starting concentration: 4 ng/ml) (open circles), culture medium (filled squares), myofibroblast conditioned medium alone (filled circles) or together with the blocking anti-gp190 antibody 12D3 at 20 μg/ml (filled triangles). Cell growth was monitored with a colorimetric assay. The figure shows the mean ± SD of 3 experiments performed in duplicate (SD are not always visible due to their small size). As shown in Figure 4 , when cells were incubated with graduated amount of recombinant human IL-4, the constitutive LIF secretion was dose-dependently reduced, demonstrating that this production may be regulated in vivo . Figure 4 Regulation of LIF secretion by interleukin-4. Confluent cultures of human liver myofibroblasts were cultured in the presence of the indicated concentrations of IL-4, for 48 h in serum-free medium. LIF was measured by ELISA in the supernatant and the results were normalized according to the DNA content of the monolayer (mean ± SD of 3 experiments). The effect of IL-4 was highly significant, as assessed by ANOVA (p = 0.001). Cloning of new isoforms of LIF mRNA In order to test whether myofibroblasts transcribed all the alternatively spliced D, M or T first exons, a first set of RT-PCR experiments was carried out using the forward primers chosen in the alternative D, M or T first exons (hLIF-D3, hLIF-M3 and hLIF-T5), and a common reverse primer chosen in exon 3 (hLIF-N4) (Table 1 and Fig. 5 ). As shown in Fig. 2B , PCR with D- or M-specific primers was positive. Moreover, it always yielded a second, shorter, PCR product in addition to the expected amplified product (Fig. 2B ). Similar results were obtained with other primer sets specific for either LIF-D (hLIF-D) or LIF-M (h-LIFM5) combined with hLIF-3N (data not shown), which strengthened the previous observation. No amplification products were obtained with the T primer. Then, we designed a reverse primer within exon 2 (hLIF-2N) that was used in conjunction with the forward hLIF-D and hLIF-M2 primers. In that case, we detected only a product of the expected size for both D and M PCRs (not shown). Table 1 Primers used for PCR Primer Sequence (): 5'> 3' Orientation Ref. hLIF-D ATAATGAAGGTCTTGGCGGCAG Forward HLIF-D3 AAA CTGCAG GCATCTGAGGTTTCCTCCAA Forward hLIF-M2 CTGGAAGCGTGTGGTCTG Forward HLIF-M3 AAA CTGCAG CTGGAAGCGTGTGGTCTG Forward hLIF-M5 TA GAATTC TGGAAGCGTGTGGTG Forward [3] hLIF-T5 AT GAATTC TGTCACCTTTCACTTTCCT Forward [3] hLIF-2N AATAAAGAGGGCATTGGCAC Reverse hLIF-3N TTCTGGTCCCGGGTGATGTT Reverse [3] HLIF-N4 GC TCTAGA GAAGGCCTGGGCCAACAC Reverse () Bases in italics refer to restriction sites. Figure 5 Sequence of LIF-D, M and T isoforms. Exons D, M and T are the 3 alternate first exons. Primers used for PCR are underlined. Primers hLIF-M2, M3 and M5 cover the same sequence but differ because of the presence or the absence of restriction sites. The sizes of the additional products obtained with the hLIF-N4 primer were shorter by about 200 bp, which is the exact size of exon 2, raising therefore the possibility that the shorter PCR products were derived from a hitherto not described mRNA species where exon 2 was deleted through alternative splicing. In order to investigate this possibility, the short D and M fragments were cloned into a plasmid and sequenced. Sequencing indeed revealed that the first exon (either D or M) was directly spliced to the third one resulting in new, short transcripts that we have designated s-LIF-D and s-LIF-M. The existence of these alternate transcripts could be observed in several hepatocellular carcinoma cell lines (HepG2, HuH7, Hep3B) and in the HEK293 cell line, derived from embryonic human kidney (Fig. 6A ). They were also expressed in normal human liver samples (Fig. 6B ) as well as in cirrhotic ones (Fig. 6C ). Figure 6 RT-PCR analysis of LIF-M expression in various cell lines and in human liver. ( A ): LIF-M expression was analyzed with the hLIF-M2 and hLIF-3N primers: Line 1, human liver myofibroblasts; Line 2, HepG2; Line 3, Hep3B; Line 4, HuH7; Line 5, HEK293. Product sizes are shown in bp; ( B ): normal human liver samples. LIF-D expression was analyzed with the hLIF-D3 and hLIF-N4 primers in 4 different samples. The same samples also expressed LIF-M (not shown). Product sizes are shown in bp; ( C ): diseased human liver samples. In that case, LIF-M expression was analyzed with the hLIF-M3 and hLIF-4N primers in 4 cases of cirrhotic liver. The same samples also expressed LIF-D (not shown). Product sizes are shown in bp; ( D ): semi-quantitation of LIF-D and s-LIF-D expression in a human liver myofibroblasts sample. LIF-D expression was analyzed with the hLIF-D3 and hLIF-N4 primers. The left part shows the migration pattern of the PCR-amplified products with the number of cycles above and the size of the products indicated by arrows, in bp. The graph on the right shows the signal quantification. Similar results were obtained with LIF-M. The relative abundance of the variant transcripts relative to the classical transcripts was studied using a semi-quantitative RT-PCR method, where PCR was carried out for varying cycles numbers. As can be seen in Fig. 6D , expression of the s-LIF-D transcript lagged several cycles behind that of the long transcript. Similar results were obtained with the s-LIF-M transcript (not shown). LIF receptor expression The expression of the gp190 subunit by liver cells was then examined by immunohistochemistry. Five different antibodies, directed against separate epitopes, were used and yielded similar results. In normal liver tissue, LIF receptor (LIF-R) expression was detected as a continuous sinusoidal staining and in the stroma of portal tracts (Fig. 7A ). In the cirrhotic liver, the sinusoidal staining was enhanced, whereas a very faint staining was observed in fibrous septa (Fig. 7B ). Staining adjacent sections with LIF receptor antibody and with an antibody to CD31 (endothelial cells in the cirrhotic liver were labeled) showed a large degree of colocalization (Figs 7C,7D ). Figure 7 Detection of LIF receptor by immunohistochemistry. ( a ): LIF-R expression in normal liver is observed in portal tracts (arrows) as well as along sinusoids (arrowheads); ( b ): Sinusoidal staining is highly increased in cirrhotic liver (arrows); ( c ) and ( d ): consecutive sections of a cirrhotic liver analyzed for LIF-R ( c ) or CD31 ( d ) expression. No labeling was seen when the antibodies were replaced by a species-matched control antibody. In a subsequent step, cultured human liver myofibroblasts were examined for their membrane expression of gp190 using flow cytometry. Adherent cells were released by action of EDTA and subjected to anti-gp190 labeling. No detectable levels of gp190 were observed with any of the 5 antibodies, although gp130 expression could be detected with the B-R3 antibody. In order to detect a low-level expression of functional LIF-R, myofibroblasts were exposed for 15 minutes to 10 ng/ml recombinant LIF; then, STAT-3 phosphorylation was examined by Western blot. No consistent effects were seen in 7 separate experiments. When a very weak signal was occasionally seen, it was not inhibited by 2 separate blocking antibodies to LIF-R (data not shown). Finally, production of soluble receptor was never detected in myofibroblast supernatants either. Discussion In this study, we demonstrate for the first time that LIF is expressed at low levels in normal human liver, whereas it is greatly increased in fibrotic liver, in a localization consistent with that of activated myofibroblasts. The slightly diffuse staining is suggestive of extracellular matrix deposition consistent with the expression of the M-type isoform of LIF. Experiments using cultured human liver myofibroblasts confirmed that these cells secreted extremely high levels of LIF in the range of 0.1–1 μg/10 6 cells/48 h. These levels are similar to those produced by activated lymphocytes, a classic source of LIF, and suggest that liver myofibroblasts may be a major source of LIF during chronic liver diseases. Our results are in agreement with data obtained in the rat showing that non-parenchymal cells, possibly activated stellate cells ( i.e. , myofibroblasts), express LIF [ 6 ]. Another study also reported an increased expression of LIF in peri-ductular cells, following bile duct ligation in IL-6 knock-out mice [ 7 ]; this location likely qualifies those cells as myofibroblasts. LIF expression by liver myofibroblasts is also reminiscent of its expression by kidney mesangial cells, a close relative to liver myofibroblasts, that we have previously reported [ 8 ]. On the other hand, and in contrast with mesangial cells [ 9 ], liver myofibroblasts do not appear to express cell surface LIF-specific gp190 receptor subunit. This is based on results obtained from immunohistochemistry, flow cytometry, as well as functional experiments. This indicates that LIF cannot exert an autocrine effect on liver myofibroblasts. However, we show that myofibroblasts express the IL-6 family common transducing subunit gp130. In this regard, others have shown that human liver myofibroblasts are responsive to oncostatin-M [ 10 ], indicating the presence of its functional alternative receptor consisting of gp130 and the specific OSMRβ chain. Nonetheless, LIF receptor expression was detected by immunohistochemistry in human liver, in a peri-sinusoidal location. Similar results were obtained with 5 different antibodies directed to several epitopes of gp190. The pattern of continuous sinusoidal staining and the colocalization experiments are in favor of an expression in sinusoidal endothelial cells. However, we can not exclude staining of the sinusoidal domain of hepatocytes. In any case, these data indicate that cells close to LIF-producing myofibroblasts express LIF receptors and could thus respond to LIF in a paracrine fashion. This study led to the discovery of new LIF transcripts resulting from a direct splicing of exon 1 to exon 3. This was observed for both LIF-D and LIF-M. Those transcripts were present at much lower levels than full-length transcripts, as suggested by RT-PCR and by the fact that they do not appear on Northern blot; thus, their biological relevance can be questioned. Whether s-LIF-D or s-LIF-M transcripts are translated also remains hypothetical. In the case of s-LIF-D, initiation at the AUG within exon D would result in a reading-frame shift following the 6 th amino-acid (aa) and a termination at aa 88, the resulting protein bearing no homology with LIF. There are, however, several in-frame CUG codons within exon 3. Initiation at CUG 113 would result in the synthesis of a 125 aa polypeptide, recapitulating the sequence of the C-terminal part of LIF. Similar considerations apply to s-LIF-M that, in any case, does not contain an initiating AUG in exon 1. It should be emphasized that the lack of an AUG codon does not preclude the translation of the classical forms of LIF-M or LIF-T [ 3 , 5 ]. More experiments are needed to know whether these new transcripts are translated. LIF secretion was dose-dependently decreased by IL-4, a known inhibitor of LIF secretion in other cell types [ 11 , 12 ]. IL-4 is also known to up-regulate collagen synthesis in human liver myofibroblasts and could thus be a pro-fibrogenic mediator in the liver [ 13 ]. Whether LIF expression is relevant to liver fibrogenesis needs to be assessed. LIF could affect extracellular matrix remodeling since it regulates the expression of several matrix proteinases and their inhibitors in various cell types [ 14 , 15 ]. In addition, LIF could play a role in the pathophysiology of chronic liver diseases through action on endothelial cells and on hepatocytes. Regarding endothelial cells, and depending on the model, both pro-angiogenic [ 16 ] and anti-angiogenic effects [ 17 ] have been described. Especially interesting is the demonstration that LIF can stimulate the adhesion of neutrophils to endothelial cells [ 18 ]; indeed, neutrophils are involved in the pathogenesis of liver diseases such as alcoholic liver disease. As already mentioned, the effects of LIF on human hepatocytes are still being debated [ 2 ]. Conclusions For the first time, we show the expression of LIF in human liver myofibroblasts, as well as of two new isoforms of mRNA. Hepatic stellate cells and activated myofibroblasts have already been shown to synthesize a number of mediators involved in the control of inflammation, such as monocyte chemotactic-1 protein [ 19 ], or platelet-activating factor [ 20 ]. Expression of LIF by myofibroblasts and of its receptor by adjacent cells suggest a potential LIF paracrine loop in human liver that may play a role in the regulation of intra-hepatic inflammation and reinforces the concept of a major role of liver myofibroblasts in the regulation of intra-hepatic inflammation [ 21 ]. Methods Tissue samples Histologically normal/subnormal liver samples were obtained from macroscopically normal location in hepatectomy specimens, taken at a distance from a focal nodular hyperplasia (n = 5); a hemangioma (n = 1); or a colon cancer metastasis (n = 1). Cirrhotic specimens (n = 11) were obtained from patients undergoing liver transplantation for cirrhosis with associated hepatocellular carcinoma. In 10 out of 11 cases, the patients underwent liver transplantation. The cirrhosis etiologies were viral hepatitis C (n = 4); viral hepatitis B + D (n = 2); alcoholic (n = 4); or a combination of viral hepatitis B + C + alcoholic (n = 1). Tissue sampling and processing A portion of fresh tissue samples was routinely formalin-fixed and paraffin-embedded for diagnosis and a portion immediately frozen in liquid nitrogen-cooled isopentane and stored at -80°C. Five μm-thick serial frozen sections of each sample were air-dried on Super Frost/Plus slides (Menzel Glaser, Germany) and processed for immunohistochemistry. The procedures were in accordance with the European guidelines for the use of human tissues. Materials Culture medium and additives, recombinant human epidermal growth factor (EGF) and Moloney Murine Leukemia Virus reverse transcriptase were from Gibco-BRL (Life Technologies, Cergy-Pontoise, France). Taq polymerase and the pGEM-Teasy plasmid were from Promega (Madison, WI). The Qiagen RNeasy minikit was from Qiagen (Courtaboeuf, France). The [α 32 P]dCTP, Hybond N + membrane, ECL reagent, and the Ready-to-go DNA labeling kit were from Amersham (Les Ulis, France). Ultrahyb solution was from Ambion (Austin, TX). Recombinant human IL-4 was a gift from Schering-Plough (Kenilworth, NJ). Anti-gp130 mAb B-R3 was from Diaclone (Besançon, France), anti-gp80 mAb M91 was from Coulter-Immunotech (Marseille, France), anti-phospho-STAT-3 (Tyr705) was from Cell Signaling Technology (Beverly, MA). All other chemicals were from Sigma (St Quentin Fallavier, France). Cell culture Human hepatic myofibroblasts were obtained from explants of non-tumoral liver resected during partial hepatectomy and characterized as previously described [ 22 , 23 ]. Myofibroblasts were routinely grown in DMEM containing 5% fetal calf serum, 5% pooled human AB serum and 5 ng/ml EGF. For studies of LIF secretion, cells were grown to confluence, made quiescent in serum and EGF-free Waymouth medium for 2 days and then exposed to agonists for 2 days. The results were normalized according to the DNA content of the monolayer [ 24 ]. Detection of LIF in culture supernatants ELISA Human LIF was measured using an ELISA based on two specific monoclonal antibodies, exactly as described previously [ 25 ]. A standard curve was obtained with recombinant glycosylated CHO-derived human LIF. The detection limit of the assay is 20 pg/ml, and LIF can be quantified at concentrations up to 1.2 ng/ml, without sample dilution. This ELISA is not sensitive to soluble receptor binding to the LIF molecule. Biological assay The Ba/F3 proliferation assays were performed, as described previously [ 26 ], using the Ba/F3 gp190 + gp130 transfectant cell line which expresses the two human LIF receptor chains (gp190 and gp130) and responds to all cytokines sharing gp190. LIF-dependent Ba/F3 cells were washed three times with RPMI to remove LIF which is required to maintain the cell line; then, cells (5 × 10 3 per well, in 50 μl, in duplicates) were incubated in the presence of 50 μl of three-fold dilutions of cytokines or supernatant, as indicated. After three days at 37°C, 0.015 ml of a 5 mg/ml solution of 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide (Sigma, Saint-Quentin Fallavier, France), in PBS, was added to each well. After 4 hours at 37°C, 0.11 ml of a mixture of 95 % isopropanol + 5 % formic acid was added to the wells, and the absorbance values were read at 570 nm, in a Titertek Multiskan microplate reader (Labsystems, Les Ullis, France). The blank consisted of eight wells containing the cells incubated with the Ba/F3 culture medium without any added cytokine. Detection of LIF mRNA by Northern blot Total RNA was isolated using the Qiagen RNeasy minikit. For Northern blot, 2 μg RNA were separated on a 1.0% agarose gel containing ethidium bromide in MOPS buffer. Running buffer and gel contained 0.2 M formaldehyde. The RNAs were transferred onto a Hybond N + membrane by downward capillary transfer in running buffer. Examination of the stained membrane under UV light was used to confirm the quality of loading and transfer. The probe used was a 730 bp cDNA containing the whole coding sequence of human LIF [ 27 ]. Probes were labeled with [α 32 P]dCTP, by random priming using the Ready-to-go kit. Hybridization was performed using the Ultrahyb solution. The blots were washed in stringent conditions (0.1X SSC, 0.1% SDS at 65°C). RT-PCR and cloning One μg of total RNA was reverse-transcribed using MMLV-RT. An aliquot was used for PCR. Thirty five cycles were performed, each consisting of 94°C, 30 s; 60°C, 30 s; and 72°C, 30 s. PCR was performed in 50 μl of a reaction buffer containing 50 mM KCl, 10 mM Tris-HCl (pH 9.0), 1% Triton X-100, 1.5 mM MgCl 2 , 0.4 mM dNTPs, 0.2 mM primers, and 1.25 units of Taq polymerase. Then, an aliquot of the reaction was analyzed by agarose gel electrophoresis. The primers used are listed in Table 1 and are also positioned on the LIF sequence in Figure 5 . When indicated, PCR products were directly cloned in the pGEM-Teasy plasmid and sequenced on both strands (Genome Express, Meylan, France). Detection of LIF and LIF receptor expression Antibodies and immunoperoxidase histochemistry A commercially available polyclonal antibody against human LIF (R&D Systems, Minneapolis, Minnesota, USA), and different monoclonal antibodies against gp190, previously described [ 28 ], were used at concentrations optimised on control tissues. For colocalization experiments, mouse monoclonal antibodies against α-smooth muscle actin (Dako A/S, Glostrup, Denmark), and CD 31 (Dako) were used. For immunohistochemistry, frozen sections were incubated with the antibodies diluted in phosphate-buffered saline, pH 7.4, containing 4% bovine serum albumin. After washing, the epitopes were detected with the Envision + system HRP detection and revealed with liquid diaminobenzidine (Dako). As negative control, we used either a clarified mouse myeloma ascites (Cappel Research Products, Durham, USA) or a rabbit non-immune immunoglobulin fraction (Dako), at the same concentration as the respective antibodies. Sections were examined with a Zeiss Axioplan 2 microscope (Carl Zeiss Microscopy, Jena, Germany). Images were acquired with an AxioCam camera (Carl Zeiss Vision, Hallbergmoos, Germany) by means of the AxioVision image processing and analysis system (Carl Zeiss Vision). Flow cytometry For each staining, 2 × 10 5 cells were incubated for 30 min at 4°C with saturating concentrations (10 μg/ml) of the indicated antibody in 0.1 ml of PBS supplemented with 1 % bovine serum albumin (BSA) and 0.1 % human polyclonal IgG (w/v, both from Sigma). Then, cells were washed twice with the same buffer and incubated for 30 min at 4°C with the FITC-conjugated goat anti-mouse IgG. After washing with PBS, the cells were resuspended in 0.14 ml of PBS containing 1% formaldehyde (v/v) and analysed by flow cytometry with a three color FACScalibur flow cytometer (Becton-Dickinson, Mountain View, CA) equipped with the CellQuest software. Control stainings used the second antibody only. ELISA (soluble receptor) The sandwich ELISA assay for soluble gp190 measurement has already been described [ 28 ]. It uses mAb 6G8 as the capture mAb, and biotinylated 10B2 mAb as the tracing mAb. Both mAb recognize distinct epitopes specific to the ectodomain of gp190. The assay has a detection limit of 0.5 ng/ml. Immunodetection of phosphorylated STAT-3 Confluent cultures of myofibroblasts were left for 2 days in serum-free medium, and subsequently exposed to recombinant human LIF for 15 minutes [ 29 ]. Then, cells were lyzed in modified RIPA buffer in the presence of protease and phosphatase inhibitors, as described [ 30 ]. Identical amounts of proteins were analyzed by Western blot with an antibody against phospho-STAT-3. The blots were stripped and rehybridized with an antibody against total STAT-3. Authors' contributions TH performed most of the cell culture and RT-PCR experiments and cloned the new LIF variants. AD and NS performed the immunohistochemistry experiments and prepared the corresponding figures. JLT provided the monoclonal antibodies to LIF-R and participated in the design of the experiments showing the secretion of active LIF. SD performed the LIF ELISA assays, the biological activity testing and flow cytometry experiments. VN performed the experiments looking at STAT-3 phosphorylation. JFB provided the human liver samples. JFM was involved in the coordination of the project and in the critical reading of the manuscript. JR conceived the study and was the main coordinator and responsible for the redaction of the manuscript. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC538256.xml
517504	Heterologous expression of plant virus genes that suppress post-transcriptional gene silencing results in suppression of RNA interference in Drosophila cells	Background RNA interference (RNAi) in animals and post-transcriptional gene silencing (PTGS) in plants are related phenomena whose functions include the developmental regulation of gene expression and protection from transposable elements and viruses. Plant viruses respond by expressing suppressor proteins that interfere with the PTGS system. Results Here we demonstrate that both transient and constitutive expression of the Tobacco etch virus HC-Pro silencing suppressor protein, which inhibits the maintenance of PTGS in plants, prevents dsRNA-induced RNAi of a lac Z gene in cultured Drosophila cells. Northern blot analysis of the RNA present in Drosophila cells showed that HC-Pro prevented degradation of lacZ RNA during RNAi but that there was accumulation of the short (23nt) RNA species associated with RNAi. A mutant HC-Pro that does not suppress PTGS in plants also does not affect RNAi in Drosophila . Similarly, the Cucumber mosaic virus 2b protein, which inhibits the systemic spread of PTGS in plants, does not suppress RNAi in Drosophila cells. In addition, we have used the Drosophila system to demonstrate that the 16K cysteine-rich protein of Tobacco rattle virus , which previously had no known function, is a silencing suppressor protein. Conclusion These results indicate that at least part of the process of RNAi in Drosophila and PTGS in plants is conserved, and that plant virus silencing suppressor proteins may be useful tools to investigate the mechanism of RNAi.	Background RNA interference (RNAi) is a process in which the introduction of dsRNAs into cells leads to inactivation of expression of genes containing homologous sequences by sequence-specific degradation of mRNA (for reviews see [ 1 , 2 ]). RNAi has been observed in a variety of organisms, including fruit fly, nematodes, zebrafish, mice and humans [ 3 - 7 ], and is mechanistically similar to post-transcriptional gene silencing (PTGS) in plants [ 8 ] and quelling in fungi [ 9 ]. Genetic studies have identified a number of the proteins that are involved in these processes in Neurospora crassa [ 10 - 12 ], Caenorhabditis elegans [ 13 - 15 ] and Arabidopsis thaliana [ 16 - 18 ]. dsRNA has been shown to induce RNAi in cultured Drosophila cells [ 19 - 22 ] and biochemical studies of this system have revealed the involvement of two distinct activities; an RNA-induced silencing complex (RISC), containing both nuclease activity and RNA that carries out enzymatic degradation of target RNA [ 19 ], and an RNase III-like protein (Dicer) that is involved in production of short (22 nucleotide) guide RNAs from dsRNA as an early step in the RNAi process [ 22 ]. PTGS in plants can operate as a defence mechanism against virus infection (for reviews see [ 8 , 23 ]) and numerous plant viruses encode silencing suppressors, which are thought to have developed as a response to the plant PTGS system [ 24 ]. Plant virus-encoded silencing suppressors may target different components of the PTGS system. The HC-Pro protein, encoded by potyviruses such as Tobacco etch virus (TEV), was initially characterised as a determinant of virus movement in plants [ 25 ] and subsequently was shown to inhibit silencing of transgenes in transformed plants [ 26 ]. HC-Pro antagonizes silencing in all tissues [ 26 ] and appears to target a step involved in the maintenance of silencing [ 27 ]. In contrast, the 2b protein encoded by Cucumber mosaic virus (CMV), a cucumovirus, interferes with the systemic spread of the silencing signal, preventing initiation of silencing in newly emerging tissues [ 28 ]. A survey of a small number of other different plant viruses showed that the comovirus Cowpea mosaic virus , the geminivirus African cassava mosaic virus , the potexvirus Narcissus mosaic virus, the tobamovirus TMV, the sobemovirus Rice yellow mottle virus (RYMV), the tombusvirus Tomato bushy stunt virus (TBSV) and the tobravirus Tobacco rattle virus (TRV) also were able to suppress GFP silencing [ 24 ]. Although in this study the potexvirus PVX did not suppress silencing, using a different assay these authors showed that PVX is in fact able to prevent systemic silencing [ 29 ]. Thus, it seems probable that many plant viruses encode proteins that allow them to evade or inhibit PTGS in certain plant species, and that different suppressor proteins target different parts of the PTGS pathway. An amenable system for studying these suppressor proteins would be an aid in determining the molecular basis of their action. Recently it has been demonstrated that an insect virus, flock house virus (FHV) encodes a protein, 2b, which acts as a suppressor of PTGS in plants [ 30 ]. We have investigated the effects on RNAi in cultured Drosophila cells of expression of some silencing suppressors from plant viruses. Results Effects of transient expression of the TEV HC-Pro and CMV 2b proteins on RNAi in Drosophila cells The TEV HC-Pro or CMV 2b proteins were transiently expressed in Drosophila DS2 cells and their effect on dsRNA-mediated silencing of a lac Z gene examined using a previously described system [ 20 ]. Expression of the CMV 2b protein could be detected by immunoblotting in lysates of cells transfected with pMT-2b and induced by addition of CuSO4 to the growth medium (Fig 1A , lane 4). No cross-reacting protein was detected in similarly induced cells not transfected with pMT-2b (Fig 1A , lane 2) or in control cells, or cells transfected with pMT-2b that were not induced (Fig 1A , lanes 1 and 3. In the absence of an antibody to TEV HC-Pro we used an expression plasmid (pMT-HC-Pro/K) that expressed a mutated HC-Pro (TEV K) containing an insertion of 3 amino acids in the central region of the protein [ 31 ] as a negative control to verify that any suppressor activity was due to an effect of the HC-Pro protein. This mutant version of HC-Pro was shown previously to be defective in suppression of PTGS in plants [ 32 ]. Northern blot analysis of RNA from cells transfected with pMT-HC-Pro or pMT-HC-Pro/K showed that transcript from both plasmids accumulated in the cells (Figure 1B , lanes 2 and 3). β-galactosidase activity could be detected by staining in ~80% of Drosophila cells when they were transfected with a lacZ expression plasmid (pMT/V5-His/lacZ) (Table 1 ). ~17% of cells stained for β-galactosidase when they were co-transfected with pMT/V5-His/lacZ and dsRNA corresponding to the first ~500nt of the lac Z gene demonstrating induction of RNAi. No significant decrease in the number of cells stained was observed when cells were transfected with pMT/V5-His/ lacZ and dsRNA derived from the green fluorescent protein gene indicating that the silencing was specific. Co-transfection of Drosophila cells with pMT/V5-His/lacZ, lacZ -specific dsRNA and an HC-Pro expression vector (pMT-HC-Pro) resulted in staining of ~46% of cells, indicating that RNAi was being suppressed. Transfection of cells with the mutant pMT-HC-Pro-K along with pMT/V5-His/lacZ and dsRNA resulted in no suppression of RNAi. This indicates that a mutation affecting the ability of HC-Pro to suppress PTGS in plants also affects suppression of RNAi in Drosophila cells. Transfection of cells with pMT-2b along with pMT/V5-His/lacZ and dsRNA resulted in no increase in the percentage of transfected cells staining for β-galactosidase activity compared to cells transfected only with pMT/V5-His/lacZ and dsRNA. RNAi in stable cell lines expressing TEV HC-Pro or CMV 2b Stable cell lines expressing HC-Pro (DS2-HC-Pro), or CMV 2b (DS2-2b) were produced in order to improve the efficiency of the RNAi suppression assay by reducing the number of different nucleic acid molecules needed for co-transfection. Immunoblotting of lysates of DS2-2b cells confirmed that the 2b protein accumulated in these cells (Fig 2A , lane 8) and this protein was not produced when the cells were not treated with CuSO4 (Fig 1A , lane 7). Northern blot analysis of RNA from the DS2-HC-Pro cells showed that an HC-Pro-specific transcript accumulated in these cells (Fig 1C , lane 1). An unrelated cell line (DS2-scAb S20) expressing a recombinant antibody [ 33 ] (Reavy et al ., 2000) was used as a control in order to eliminate the possibility that stable transformation of the cells could interfere with the RNAi mechanism. The transfection efficiency of all transformed lines with pMT/V5-His/lacZ was lower than that of the control DS2 cells (Table 1 ); possibly the transformed cells are more recalcitrant to transfection than the control cells or the additional copies of the metallothionein promoter are saturating the induction factors. Nonetheless, RNAi was strongly induced in the DS2-scAb S20 cells as co-transfection with pMT/V5-His/lacZ and dsRNA reduced the number of cells staining for β-galactosidase activity to only 6.5% compared to ~41% when transfected with pMT/V5-His/lacZ alone (Table 1 ). Transfection of DS2-HC-Pro cells with pMT/V5-His/lacZ resulted in ~42% of cells expressing β-galactosidase. This was reduced only to ~32% when transfection was carried out using pMT/V5-His/lacZ and dsRNA, showing that RNAi was significantly inhibited in this cell line. In contrast, no suppression of RNAi was observed in the DS2-2b line (expressing the CMV 2b protein) when transfected with pMT/V5-His/lacZ and dsRNA, as the percentage of stained cells was similar to that in DS2-scAb S20 cells transfected with pMT/V5-His/lacZ and dsRNA. Similar results were obtained when the DS2-2b cells were grown in the presence of CuSO4 to induce expression of the virus protein before transfection with pMT/V5-His/lacZ and dsRNA, indicating that the CMV 2b protein could suppress neither the initiation nor the maintenance of the dsRNA-induced silencing in the Drosophila cells. Identification of silencing suppression in Drosophila cells by another plant virus protein We were interested to determine if the Drosophila cell system could be used as a screen for RNAi suppression effects caused by other virus proteins. We chose to examine the potential RNAi suppression activity of a protein from the tobravirus TRV. This virus is able to suppress transgene silencing in plants [ 24 ] but the specific viral protein responsible for this activity has not been identified. TRV, like the other tobraviruses has a bipartite, positive strand RNA genome [ 34 ], however, the larger RNA (RNA1) can infect plants systemically in the absence of RNA2 to produce what is known as an NM-infection. This occurs frequently in particular cultivars of potato and is often associated with increased symptom severity. Clearly, therefore, RNA1 encodes all the functions necessary for virus multiplication including, possibly, suppression of PTGS/host defence. The one protein encoded by RNA1 without an assigned function is a 16K cysteine-rich protein which was, thus, investigated as a candidate silencing suppressor protein. One characteristic of plant virus silencing suppressor proteins is often significant enhancement of disease symptoms when they are over-expressed from a viral vector [ 26 ]. Similar results were obtained from preliminary experiments showing that expression of the TRV 16K gene from a PVX vector did increase the severity of symptoms in infected plants. Inoculation of Nicotiana benthamiana plants with PVX alone initially induced vein chlorosis and systemic leaf curling, although the plants continued to grow. Inoculation with PVX carrying the 16K gene caused similar initial symptoms but led to tip necrosis and death of the plants [ 35 ]. The TRV 16K protein was expressed in Drosophila cells after transfection of cells with the expression plasmid pMT-16K, and the 16K protein could be detected by western blotting when the cells were induced with CuSO4 (Fig 2 , lane 4) but not when cells were not induced (Fig 2 , lane 3). A stably-transformed cell line (DS2-16k) containing pMT-16k behaved in a similar way (Fig 2 , lanes 7, 8). No cross-reacting protein was detected in non-transfected cells (Fig 2 , lanes 1,2). Co-transfection of pMT/V5-His/lacZ with dsRNA and pMT-16K resulted in ~47% of transfected cells staining blue (Table 1 ) compared to ~18% when transfected with pMT/V5-His/lacZ and dsRNA. The 16K protein therefore partially suppressed RNAi in Drosophila cells. Detection of lacZ gene transcripts Northern blot analysis of the RNA present in transfected cells confirmed that HC-Pro was effective in preventing cytoplasmic degradation of the lacZ transcript. RNA with the expected size of the lacZ transcript could not be detected in extracts of DS2 cells that were transfected with pMT/V5-His/lacZ and dsRNA (Fig. 3 , lane 2). However, intact lac Z RNA was present in extracts of DS2-HC-Pro cells that were transfected with pMT/V5-His/lacZ regardless of whether the cells were co-transfected with dsRNA (Fig 3 , lanes 3 & 4). Similarly, the lacZ transcript was intact in DS2-16K cells after transfection with pMT/V5-His/lacZ and dsRNA (Fig 3 , lane 6). The amounts of lacZ transcript in the DS2-Hc-Pro and DS2-16K cells were less after transfection with pMT/V5-His/lacZ and dsRNA than after transfection with pMT/V5-His/lacZ alone indicating partial suppression of RNAi. No lacZ transcript was detected in DS2-2b cells after transfection with pMT/V5-His/lacZ and dsRNA (Fig 3 , lane 10) but the lacZ transcript could be detected in DS2-2b cells transfected with pMT/V5-His/lacZ alone Fig 3 , lane 8). Detection of siRNAs Suppression of silencing by viral proteins in plants is often associated with an inhibition of the production of small, 21 to 25 nucleotide RNAs that may be analogous to the 22 nucleotide guide RNAs identified as part of the Drosophila RISC complex [ 22 , 27 , 36 ]. We were able to identify short RNAs specific to the region of transfected dsRNA in extracts of Drosophila cells transfected with pMT/V5-His/lacZ and dsRNA (Fig 4 , lane 2). A lot of larger RNA species were also detected presumably as a result of degradation of the input dsRNA. Unfortunately it is not possible to probe for the presence of short RNAs outwith the region of input dsRNA, as RNAi is not transitive in Drosophila [ 37 ]. Expression of the plant viral HC-Pro suppressor protein in the DS2-HC-Pro cells did not prevent the accumulation of these small RNAs when transfected with pMT/V5-His/lacZ and dsRNA (Fig 4 , lane 4). Co-transfection of Drosophila cells with pMT/V5-His/lacZ, dsRNA and pMT-16K also resulted in production of small RNA species (Fig 4 , lane 8) even though the 16K protein also partially suppresses silencing in the Drosophila cells. Possibly the inefficiency of the transfection procedure and the failure of the suppressors to completely suppress RNAi in these experiments masks any visible effect by the suppressors on the accumulation of these molecules. Similarly, small RNAs were detected in the DS2-2b cells after transfection with pMT/V5-His/lacZ and dsRNA (Fig 4 , lane 6). The dsRNA preparations used to induce RNAi were examined to determine if small RNA species of a similar size to siRNAs were present and were the species detected in figure 4 . No short RNA species of 21–25 nucleotides were observed in the dsRNA preparation used to transfect the cells indicating that the species observed in the transfected cells were produced as a result of cellular processing (Fig 5 , lane 3). Some larger products were observed and these are likely to have arisen as a result of premature terminations during the dsRNA synthesis reaction. Discussion Suppression of RNAi in Drosophila cells by some plant virus proteins indicates that at least part of the processes of RNAi and PTGS is conserved between plants and Drosophila. TEV HC-Pro is one of a family of proteins that suppress PTGS in plants and the CMV 2b protein has a similar ability [ 26 ]. These proteins are thought to target different components of the PTGS system, as the CMV 2b protein interferes with the spread of a silencing signal after initial induction of silencing, thus, preventing silencing from initiating in newly emerging leaves. The potyvirus HC-Pro protein, however, interferes with the maintenance of silencing in all tissues [ 26 , 27 ]. These differences were reflected in the Drosophila cell system where HC-Pro could suppress RNAi but the 2b gene apparently was ineffective. As there is no spread of a silencing signal in cultured Drosophila cells, the failure of the CMV 2b protein to act in this system is not unexpected. It was also significant that a mutant version of HC-Pro (K) that is defective in suppression of gene silencing [ 32 ] but is effective in proteolytic cleavage [ 31 ] was also unable to suppress RNAi in Drosophila cells. This strongly supports the idea that the TEV HC-Pro protein targets the same component of the plant PTGS system and the Drosophila RNAi system. Our demonstration that the TRV 16K protein also is able to interfere with RNAi in Drosophila cells, leads us to suggest that many other plant virus silencing suppressor proteins are likely to be functional in this system. SiRNAs were detected in our cell lines expressing HC-Pro and the TRV 16K protein even though suppression of silencing was observed. This is not totally unexpected for a number of possible reasons. Firstly, suppression of RNAi is partial in our cultures indicating that some RNAi, and presumably siRNA production, does occur in some of the cells. We have observed with cells expressing the lacZ gene alone that there is considerable variation in the amount of staining of individual cells indicating differences in the amount of gene expression in individual cells. There is no reason to suppose that this variability in gene expression is limited to the lacZ gene and it is possible that low levels of the suppressors may fail to inhibit silencing in some of the cells within a culture. Secondly, the experimental protocol requires a time delay between transfection of the cells with the lacZ plasmid and dsRNA before induction of expression of the suppressor proteins in order to allow the cells to recover from the transfection procedure. Some production of siRNAs from the dsRNA may possibly occur in cells during this lag period before the suppressors are induced. For these reasons the usefulness of the Drosophila RNAi system for studying the mechanisms of action of plant virus suppressor proteins may be limited. Mutants of Arabidopsis that are impaired in PTGS have an increased susceptibility to CMV, showing that PTGS can operate as a defensive system that targets viruses [ 18 ]. Plant virus-encoded silencing suppressors are thought to have developed as a response to this defensive aspect of the plant PTGS system, and subsequent studies have shown that, as would be expected, a wide variety of plant viruses encode silencing suppressors [ 24 ]. The suppressors encoded by these viruses are unrelated in amino acid sequence, and are likely to act at different points in the silencing process, making them ideal probes to investigate the silencing machinery. We anticipate that using different suppressors from a wide range of viruses may permit a detailed examination of the biochemical process of RNAi in Drosophila and possibly other organisms as well as allowing detailed characterisation of the mode of action of the virus suppressor proteins. Furthermore, it is becoming apparent that RNAi or PTGS may play a role in processes other than defence against foreign RNAs. Transformation of plants with TEV HC-Pro or rgs-CaM, a plant-encoded PTGS-suppressor protein related to calmodulin, interrupted normal development and led to the formation of differentiated tumours at the stem/root junction [ 38 ]. The Arabidopsis gene CARPEL FACTORY is related to Drosophila Dicer and is involved in plant development and fertility [ 39 ], and the EGO-1 gene of C. elegans also appears to have a role in both RNAi and germ-line development 15]. Components of the RNAi mechanism including a homolog of Dicer are also involved in synthesis of short temporal RNAs that regulate developmental timing in C. elegans [ 40 - 42 ]. Intervention with plant virus silencing suppressors may therefore have significant utility in determining the involvement of RNAi in development and differentiation in plants and animals, and possibly in manipulating these processes. Conclusions These results indicate that at least part of the process of RNAi in Drosophila and PTGS in plants is conserved, and that plant virus silencing suppressor proteins may be useful tools to investigate the mechanism of RNAi. Methods Plasmid constructions A region (nucleotides 1055–2449) of the TEV genome containing the HC-Pro sequence was amplified by reverse transcription – polymerase chain reaction (RT-PCR) using as a template RNA from an infected plant, and primers HC-Pro-1 (5'-CCGGTACCATGAGCGACAAATCAATCTCTGAGGC-3') and HC-Pro-2 (5'-GGCTCGAGCTACACATCTCGGTTCATCCCTCC-3'). The primers add an ATG codon to the start of the open reading frame and the HC-Pro gene was cloned as a Kpn I- Xho I fragment into pMT/V5-HisA (Invitrogen) to produce plasmid pMT-HC-Pro. The same primers were also used to clone a mutant HC-Pro gene (TEV-K; [ 31 ]) from transgenic plants to produce plasmid pMT-HC-Pro/K. The 2b gene was amplified from RNA2 of CMV isolate Fny using primers 392 (5'-GAACCATGGAATTGAACGTAGGTGC) and 393 (5'-GGGTACCTCAGAAAGCACCTTCCGCC) and cloned into pGemT (Promega) before subcloning into pMT-V5-HisA. The TRV 16K gene was amplified from RNA1 of TRV isolate PpK20 using primers 433 (5'-TCATCATGACGTGTGTACTCAAGGG-3') and 434 (5'-AAGGTACCATCAAAAAGCAAACG-3') to insert Bsp HI and Kpn I sites upstream and downstream, respectively, of the gene. The PCR product was cloned into pMT/V5-HisA to produce plasmid pMT-16K. The nucleotide sequences of the cloned PCR inserts were confirmed by sequencing. dsRNA synthesis cDNA corresponding to ~500 bp of the 5' end of the lac Z gene was amplified using pcDNA3.1/HisB/lacZ (Invitrogen) as a template and primers lacZ -1 (5'- TAATACGACTCACTATAG GGAGACCCAAGCTGGCTAGC-3') and lacZ -2 (5'- TAATACGACTCACTATAG GGCAAACGGCGGATTGACCG-3'). Both primers contain T7 RNA polymerase promoter sequences (shown underlined). The PCR product was used to direct synthesis of dsRNA using T7 RNA polymerase (Invitrogen) after which the DNA template was removed by DNase digestion. Cell culture and transfection DS2 cells, DES expression medium and the lacZ expression plasmid pMT/V5-His/lacZ were supplied as part of the Drosophila Expression System (Invitrogen) and cells were grown according to the manufacturer's instructions. Cells were grown in 60 mm dishes and transfected by calcium phosphate co-precipitation with various mixtures of 10 μg each of pMT/V5-His/lacZ and the suppressor expression plasmid DNAs and 5 μg dsRNA. In control transfections, 10 μg of an empty expression plasmid (pMT/V5-HisB) replaced the suppressor expression plasmid. After transfection the cells were washed twice in DES medium and grown for eight hours before expression of proteins was induced by addition of CuSO 4 to a final concentration of 500 μM. Stably transformed lines expressing the HC-Pro, 2b or a6k genes were established by co-transfection of cells with pMT-HC-Pro or pMT-2b and pCo-Hygro (Invitrogen) followed by selection of transformed cells in medium containing hygromycin. Cells were stained 48 hrs after transfection to detect lacZ gene expression using a β-Gal Staining Kit (Invitrogen). Four randomly selected fields of view each containing ~100 cells were selected in each of duplicate plates and the number of cells staining blue was counted. Immunoblotting Cell lysates were analysed by polyacrylamide gel electrophoresis and separated proteins were transferred to nitocellulose using a carbonate buffer [ 43 ]. The blots were probed with an anti-CMV 2b [ 44 ] or anti 16K antibody [ 45 ] followed by a goat anti-rabbit alkaline phosphatase conjugate. Blots were developed using SigmaFast NBT/BCIP substrate (Sigma, Poole, UK). Northern blot analysis of RNA from Drosophila cells RNA was extracted from Drosophila cells using TriPure Isolation Reagent (Boehringer). Samples of RNA were separated by electrophoresis on formaldehyde/agarose gels, transferred to nylon membrane and probed with digoxigenin-labelled DNA probes corresponding to ~500nts at the 5' end of the lacZ gene or to the HC-Pro cDNA described above, as appropriate and essentially as described [ 46 ]. The probes were made by PCR using primers lacZ-3 (5'- GGAGACCCAAGCTGGCTAGC-3') and lacZ-4 (5'- GGCAAACGGCGGATTGACCG-3') for the lacZ probe, and HC-Pro-1 and HC-Pro-2 for the HC-Pro probe. Digoxigenin-labeled RNA Molecular Weight Marker II (Roche Molecular Biochemicals) was run as a size marker. For detection of short (~23nt) RNA species total RNA preparations from Drosophila cell cultures were fractionated by chromatography using sepharose CL-2B agarose (Sigma) in Micro Bio-Spin columns (Bio Rad), following the manufacturers instructions. ~20 μg of the short RNA species were separated by electrophoresis in a 15% polyacrylamide gel containing 7 M Urea and electroblotted as described by Llave et al. [ 27 ]. After electrophoresis the gel was stained with ethidium bromide and photographed under ultra-violet light before the RNA species were transferred to Hybond N+ membrane (Amersham) by electroblotting. For the detection of siRNAs, a digoxigenin-labelled RNA probe complementary to the 5' region of the lacZ gene as described above was synthesised from a PCR fragment containing the gene fragment downstream of a T7 RNA polymerase promoter, and hybridised to the blots according to the manufacturer instructions (Roche Diagnostics). Induction and detection of siRNAs to GFP or to 35S promoter sequences in agroinfiltrated plant tissue was performed as described by Canto et al . [ 47 ]. Competing interests None declared. Authors' contributions BR conceived of the study, and participated in its design, carried out the Drosophila expression experiments, northern blots and drafted the manuscript. SD carried out the immunoassays. TC carried out some of the siRNA assays. SM participated in the design of the study, carried out some of the siRNA assays and contributed to the manuscript. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517504.xml
555737	CyProQuant-PCR: a real time RT-PCR technique for profiling human cytokines, based on external RNA standards, readily automatable for clinical use	Background Real-time PCR is becoming a common tool for detecting and quantifying expression profiling of selected genes. Cytokines mRNA quantification is widely used in immunological research to dissect the early steps of immune responses or pathophysiological pathways. It is also growing to be of clinical relevancy to immuno-monitoring and evaluation of the disease status of patients. The techniques currently used for "absolute quantification" of cytokine mRNA are based on a DNA standard curve and do not take into account the critical impact of RT efficiency. Results To overcome this pitfall, we designed a strategy using external RNA as standard in the RT-PCR. Use of synthetic RNA standards, by comparison with the corresponding DNA standard, showed significant variations in the yield of retro-transcription depending the target amplified and the experiment. We then developed primers to be used under one single experimental condition for the specific amplification of human IL-1β, IL-4, IL-10, IL-12p40, IL-13, IL-15, IL-18, IFN-γ, MIF, TGF-β1 and TNF-α mRNA. We showed that the beta-2 microglobulin (β2-MG) gene was suitable for data normalisation since the level of β2-MG transcripts in naïve PBMC varied less than 5 times between individuals and was not affected by LPS or PHA stimulation. The technique, we named CyProQuant-PCR (Cytokine Profiling Quantitative PCR) was validated using a kinetic measurement of cytokine transcripts under in vitro stimulation of human PBMC by lipopolysaccharide (LPS) or Staphylococcus aureus strain Cowan (SAC). Results obtained show that CyProQuant-PCR is powerful enough to precociously detect slight cytokine induction. Finally, having demonstrated the reproducibility of the method, it was applied to malaria patients and asymptomatic controls for the quantification of TGF-β1 transcripts and showed an increased capacity of cells from malaria patients to accumulate TGF-β1 mRNA in response to LPS. Conclusion The real-time RT-PCR technique based on a RNA standard curve, CyProQuant-PCR, outlined here, allows for a genuine absolute quantification and a simultaneous analysis of a large panel of human cytokine mRNA. It represents a potent and attractive tool for immunomonitoring, lending itself readily to automation and with a high throughput. This opens the possibility of an easy and reliable cytokine profiling for clinical applications.	Background Cytokines are a family of low-molecular weight proteins secreted by various cell types, with pleiotropic functions and constitute a tightly regulated network that plays a central role in the immune system. Cytokines, classified into different groups such as interleukins (IL), interferons (IFN), colony-stimulating factors (CSF), tumour necrosis factors (TNF), tumour growth factors (TGF) and chemokines are implicated in the differentiation, proliferation, migration and effector functions of immune cells. Interacting one with the others, they have polarizing effects on the target cells and are pivotal in tuning immune responses [ 1 ]. Therefore, it is rather the make-up of cytokines milieu that influences the immune response rather than the action of a single cytokine. Numerous studies indicate that the clinical and/or immunological status depends on the balance between pro-inflammatory cytokines and their regulatory counterparts [ 2 ]. Thus, cytokine profiling should be achieved through analysis of simultaneous quantification of a pattern of cytokines including pro and anti-inflammatory cytokines [ 2 , 3 ]. Moreover, recent reports have highlighted the need for clinical immuno-monitoring of patients to adapt treatment or prevent relapses [ 4 - 6 ]. Thus, analysis of the cytokine pattern is central not only in the definition of the immunological status of patients but also in the study of the pathophysiological pathways as well as the cellular subpopulations involved [ 7 , 8 ]. Cytokines are often produced locally so that the concentration of circulating cytokines in the plasma is usually low. Their half-life and turnover may vary complicating the delineation of informative cytokine profiles. Although transcription of messenger RNA is not strictly correlated to protein secretion and activity, detection of cytokine RNA by real time PCR is now considered a reference technique for analysis of small-size samples with high sensitivity [ 9 ]. It can be used on its own or to validate and complement information obtained with other techniques such as micro-arrays [ 10 , 11 ]. The already available techniques, which offer a so-called "absolute quantification" of the target cytokine mRNA, achieve quantification by reference to an external standard curve based on serial dilutions of a known amount of the corresponding cDNA [ 12 ]. Moreover, to allow for comparison between experiments, data are normalized by reference to an internal standard, which is an endogenous gene for which the number of copy per cell is supposed constant under different experimental conditions [ 13 , 14 ]. The term of "absolute" quantification is not completely appropriate since these techniques neither control for the variable efficiency of the RT step nor take it into account in their measurements [ 15 , 16 ]. In the present study, we first show that the efficiency of the RT step depends on the target mRNA and on the experiments and that these variations have critical impact on the reliability of mRNA quantification. To overcome this, we describe here CyProQuant-PCR, a new technique for absolute measurement of cytokine mRNA based on an external RNA standard curve. Primer pairs have been designed for allowing amplification of a set of cytokine mRNA using the same conditions both in terms of thermocycling parameters and master mix components, a prerequisite for multiple cytokine mRNA measurements with high throughput. In the present paper, we describe i) the construction of the synthetic RNA standard, ii) the primer pairs specific for the following human cytokines IL-1β, IL-4, IL-10, IL-12p40, IL-13, IL-15, IL-18, IFN-γ, and for the tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein, zeta polypeptide (YWHAZ), the β2-microglobulin (β2-MG) and the ubiquitin-C (UBC) to be used as internal standards and iii) the conditions for efficient real time amplification of multiple cytokine specific mRNA. The technique was validated using in vitro stimulated PBMC and its intra and inter-experimental variability were assessed. Finally, CyProQuant-PCR was used to quantify TGF-β1 transcripts in small blood samples from children with acute Plasmodium falciparum malaria. Results Primer design and validation Primer pairs were designed from published genomic sequences using Primer Express software (Applied Biosystems), except for the UBC and the YWHAZ genes for which the primers had already been described [ 17 ]. When possible, the following criteria were applied. The percent of G+C content was kept in the 20–80% range and runs of an identical nucleotide were avoided. The five nucleotides at the 3' end had no more than two G and/or C bases and the melting temperature was kept between 58 and 60°C. Among the primers proposed by Primer Express software, we selected forward and reverse primers amplifying a product spanning one intron and not leading to the amplification of pseudogenes or other related genes to secure primer specificity for target cDNA. The specificity of the amplification was assessed for each gene by electrophoresis and dissociation curve analysis as shown in Figure 1A and 1B . The absence of any contaminating bands corresponding to genomic DNA amplification on agarose gel as well as the presence of a unique peak on the dissociation curve validated the specificity of the primer pair for the target cDNA. PCR products were systematically sequenced after cloning and showed more than 98% identity with the expected sequence (data not shown). Primer concentrations were optimised to determine the minimum primer concentrations giving the lowest threshold cycle (C T ) and the maximum signal-to-noise fluorescence ratio (ΔR n ) while minimising non-specific amplification (data not shown). Among the final concentrations of 50, 300 and 900 nM tested for each primer, the optimal final concentration was set at 900 nM for every primer. The primer pairs are shown in Table I . Standard curve design Generation of the RNA standard curve To generate the standard RNA corresponding to each target sequence, gene specific primers were fused in their 5' end to the RNA polymerase T7 promoter sequence. The PCR performed with these modified primers pairs lead to a larger PCR product with the T7 promoter sequences upstream and downstream from the specific amplicon. The in vitro transcription gave a synthetic RNA, which was assessed for its integrity and clonality by electrophoresis (data not shown). The molecular mass of each RNA standard was calculated on the basis of its sequence and solutions ranging from 10 1 to 10 12 copies of standard RNA were made. These serial diluted solutions were reverse transcribed and the cDNA amplified in duplicate to generate a standard curve by plotting the threshold cycle (C T ) against the logarithmic value of the starting RNA copy number for each dilution. Figure 1C shows an example of these curves for β2-MG. Every curve generated a dynamic range of a least 6 orders of magnitude. This allowed for a reliable and reproducible quantification of cellular mRNA sample. Comparison between CyProQuant-PCR and DNA standard based approach CyProQuant-PCR was compared to the classical approach based on a DNA standard curve for the quantification of TNF-α, IL-1β and IFN-γ transcripts. DNA templates for RNA transcription of TNF-α, IL-1β and IFN-γ were used to generate a range of concentration from 10 to 10 12 copies/μl. We thus disposed of DNA and RNA ranges stemming from the same sequence. In parallel, a range of cDNA was also generated from the RNA standards. Ranges of RNA standard were reverse transcribed and then amplified in parallel to the cDNA and DNA ranges by real time PCR under the same conditions (same mix, same final volume, same PCR plate). The slope of the standard curves generated by these three ranges as well as the corresponding efficiency are shown in Table 2 . The data show that the RT-PCR efficiencies from CyProQuant-PCR are lower than the PCR efficiencies from cDNA or DNA ranges for the three genes. Since the PCR efficiencies of the cDNA and DNA ranges are similar for each gene, this discrepancy is due to the RT efficiency, which is lower than 100% (from 75% for IFN-γ to 98% for TNF-α). In addition, data show that RT efficiency displays intergenic variation with a lower efficiency for IFN-γ compared to the two other cytokines (Table 2 ). In order to show that these intergenic differences were not due to interassay variation, we compared amplification efficiencies of TNF-α, β2-MG, MIF and UBC from i) a single cellular RNA sample, ii) a pooled RNA sample containing standard RNA for each of the targets and iii) a pooled DNA sample corresponding to the cDNA obtained from the pooled RNA sample containing standard RNA for each of the targets. Results obtained are summarised in Table 3 and show that RNA amplification efficiencies vary from one gene to the other but are not affected by the molecular context since they do not differ between the single cellular RNA sample and the pooled RNA sample. Moreover, since the amplification efficiencies of the pooled DNA sample have been normalised to 100%, the difference of amplification efficiency between the RNA samples and the pooled DNA corresponds to differences in RT efficiency. Altogether, these data demonstrate that RT efficiency displays intergenic variations that are not due to interassay differences. Impact of RT efficiency variability on transcripts quantification reliability RT efficiency has critical impact on transcript quantification as shown in Table 4 . Indeed, if we compare CyProQuant-PCR or DNA standard based approach for quantification of a known number of TNF-α, IL-1β and IFN-γ transcripts, results show that DNA standard based approach underestimates of around two logs the number of transcripts. Moreover, this underestimation varies according to the transcript studied and the experiment. CyProQuant-PCR is thus more precise and more reliable than the classically used DNA-based approaches. Taken together, these results support the use of RNA as external standard for reliable and reproducible quantification of transcripts. RT-PCR efficiency is similar for sample and RNA standard Since RT-PCR efficiency varies, absolute mRNA quantification can only be reliably obtained if the external RNA standard and the cellular RNA are retro-transcribed and amplified with the same efficiency. This was secured by comparing the standard curves obtained after amplification of a range of 10X serial dilution of cellular RNA and β2-MG external RNA standard. The slopes obtained were of -3,307 and -3,305 for the cellular RNA and for the external RNA standard, respectively. This corresponds to efficiencies of 100,6% and 100,7% respectively (data not shown). Endogenous standard, normalization and reliability of the technique The choice of a stable expressed endogenous gene to be used as an internal standard is a prerequisite for accurate RT-PCR expression profiling. This has to be adapted to the clinical situation and the tissue of origin of the samples. Since our purpose was to establish a technique to be used with peripheral blood leukocytes, we tested three genes reported in stable amounts in leukocytes: the β2-MG, the UBC and the YWHAZ [ 17 ]. We compared the stability of the amount of transcript under different conditions of activation. Total cellular RNA was extracted on two different days from the same two aliquots of PBMC stimulated for 3 hours by LPS. Thus, reverse transcription was realised on RNA extracted from the same number of cells. The three endogenous gene transcripts were amplified by CyProQuant-PCR in the same plate in duplicate using the same PCR master mix. Table 5 shows the mean of the results obtained for the two extractions. The data show that β2-MG transcripts are stable with less than 12% of variation whereas UBC and YWHAZ transcript levels showed up to 47% variability depending on the in vitro conditions. Moreover, based on the approach recently described by Pachot et al. [ 18 ], we assessed the inter-individual variability of the β2-MG basal level using whole blood samples from 6 different healthy donors. All CT values were within 19 and 21,86 cycles, which corresponds to a variation of less than 5 times in gene expression for an overall RT-PCR reaction efficiency of 1,81 (slope = -3,867). This is considered as acceptable for a reference gene [ 18 ]. β2-MG was thus chosen as an internal standard gene for future experiments. The same conclusions were drawn using PHA (phytohemagglutinin) as a stimulant (data not shown). In addition, these data validate the reliability and reproducibility of our RNA extraction protocol. Validation of the CyProQuant-PCR technique Human TNF-a transcript quantification: comparison of CyProQuant-PCR to TaqMan ® Since TaqMan ® technology developed by Applied Biosystems is considered as the "gold standard", we compared CyProQuant-PCR to TaqMan ® for the quantification of TNF-α transcripts in isolated monocytes stimulated for 6 hours with LPS. Total RNA was extracted from 10 6 cells and 10X serial dilutions were prepared and reverse transcribed. The resulting cDNA were amplified in duplicate on the same plate in the same thermocycling conditions using either the TaqMan ® commercial kit (Applied Biosystems) or CyProQuant-PCR primers with SYBR Green PCR master mix (Applied Biosystems). Figure 2 shows the curves obtained using the two techniques on the same samples. This indicates that CyProQuant-PCR is as good as TaqMan ® in terms of sensitivity and even more efficient: 105% (slope -3,198) for CyProQuant-PCR versus 80 % (slope -3,908) for TaqMan ® . Quantification of TNF-α and MIF by CyProQuant-PCR and ELISA To validate our approach, we measured the levels of TNF-α and MIF transcripts and secreted proteins by CyProQuant-PCR and ELISA respectively. PBMC from healthy individuals were stimulated by LPS or SAC for 3, 9 and 18 hours. Figure 3 shows an increase of TNF-α transcripts after 3 hours of stimulation that precedes protein secretion. In contrast, MIF transcripts were constitutively present at the steady state and LPS or SAC stimulation did not significantly modify the level of transcripts although it induces the release of substantial amounts of MIF proteins. This is in agreement with reports showing that MIF exists within cells under homeostasis both as a preformed protein ready to be secreted and as a messenger RNA, which can then be rapidly translated in the absence of induced transcription [ 19 ]. Early cytokine profiling of PBMC from healthy donors stimulated with LPS or SAC To validate the panel of primers designed for CyProQuant-PCR, we assessed the early cytokine response as well as the kinetics of expression for PBMC from two healthy donors under stimulation with LPS or SAC. Figure 4 illustrates CyProQuant-PCR detection of IL-1β, IL-4, IL-10, IL-12p40, IL-13, IL-15, IL-18, MIF, TGF-β1 and TNF-α transcripts for one donor. No significant increase in IL-15, IL-18, MIF or TGF-β1 transcripts was detected. Both stimulants induce a similar kinetics of transcript accumulation for IL-1β and IL-10. In contrast, IL-4 and IL-13 transcripts peaked very early (2 hours) under LPS stimulation but not under SAC stimulation. IL-12p40 transcripts also accumulated earlier under LPS stimulation compared to SAC but was detected later on. The amount of TNF-α transcript was sustained under SAC stimulation compared to LPS. Detection of the corresponding proteins for TNF-α, IL-10 and IL-12p40 confirmed the mRNA profile observed (data not shown). The amplitude of the response differed somewhat between the two donors but the kinetics obtained for each cytokine were similar (data not shown). Application: Use of CyProQuant-PCR to quantify TGF-β1 transcripts in malaria patients Before the application to clinical samples, the experimental reproducibility of CyProQuant-PCR was evaluated using a range of TGF-β1 and β2-MG RNA standards. Inter-experimental variability was assessed from eight independent experiments. The coefficients of variation were satisfactory whatever the starting copy number of RNA standard, ranging from 0,39 to 1,07% and from 0,91 to 1,2% for TGF-β1 and β2-MG respectively (Table 6 ). TGF-β1 transcripts were then quantified after LPS stimulation of PBMC from malaria patients and asymptomatic controls. Figure 5 shows a significant difference in the amount of TGF-β1 transcripts between patients and asymptomatic controls only after LPS stimulation (p = 0,036). Discussion In this paper we have described a new technique, CyProQuant-PCR, for absolute quantitative profiling of human cytokine mRNA using real time RT-PCR. Although real time PCR is now becoming a popular technique, it still requires improvement for proper quantification. All the techniques for absolute quantification available so far use a DNA standard curve assuming that RT efficiency is constant and approaches 100% [ 12 ]. We show here that RT is not 100% efficient, and more importantly, that its efficiency changes from one gene to another and from one experiment to another. We thus designed primers and standard RNA to be used in such a way that all cytokine mRNA of interest as well as three housekeeping genes could be amplified using the same conditions (thermocycling parameters and buffer). This technique is rapid, reliable and reproducible. When compared to the TaqMan ® commercial kit, which represents the "gold standard", CyProQuant-PCR was as sensitive but less expensive and more flexible. Indeed, for a given gene, the design of the probe might be quite difficult and, with judicious selection of primer pairs, comparable sensitivity can be achieved with the use of SYBR Green [ 20 ]. A major reason for not using an RNA standard curve is its poor stability due to its sensitivity to RNase degradation. In our hands, we did not find any detectable degradation when keeping the standard in concentrated aliquots (stock solution at 1000 μg/mL) in RNase-free water at minus 80°C and avoiding freeze-thawing cycles. However, for a greater stability, we incorporated modified dNTP, 2'-Fluoro-dCTP and 2'-Fluoro-dUTP, which have been reported to decrease the sensitivity of the in vitro transcribed RNA to specific RNases [ 21 , 22 ], and tested the effect of this incorporation on the RT-PCR efficiency. We did not find significant difference in the efficiencies (100% for the non-modified IL-4 standard versus 104% for the 2'-Fluoro-standard) (data not shown). The methodology described here was easily and successfully applied to the quantification of several cytokine genes. The quantification is reliable on 7 to 8 logs with a sensitivity ranging from 1000 to 100 copies depending the cytokine. It was first used to determine the magnitude and the kinetics of early induction of cytokines mRNA upon PBMC stimulation using bacterial derived materials. This showed that CyProQuant-PCR is powerful enough to detect early on modest cytokine induction such as IL-4. Such a tool is useful to decipher the kinetics of cytokine response involved in physiopathological pathways but also as a read-out to measure minor immune responses to specific antigens [ 23 ]. We finally applied CyProQuant-PCR to measure the level of TGF-β1 transcripts in asymptomatic controls and malaria patients after LPS stimulation. Interestingly, we observed that cells from malaria patients have a significant higher capacity to respond to LPS compared to controls. The increased TGF-β1 transcript accumulation by patients' cells, suggests that an inadequate production of TGF-β1 might play a role in malaria pathogenesis as already proposed [ 24 ]. Further work is needed to elaborate on this finding. This first study provided the proof that CyProQuant-PCR is readily applicable to small clinical samples from paediatric cases. This opens the possibility to further quantify the cytokine imbalance associated with malaria pathogenesis and generate a disease cytokine signature, a prerequisite for novel therapeutic interventions targeting cytokine gene expression. Areas of application include both infectious and non-infectious diseases, as well as chronic inflammatory diseases such as rheumatoid arthritis or sarcoidosis and acute diseases such as sepsis or malaria. Beside its importance in patient immuno-monitoring, cytokine profiling is also a major tool to study specific immune response against antigens for design and testing of immuno-modulatory drugs or vaccines. Conclusion In conclusion, we provide here CyProQuant-PCR, a simple technique for genuine absolute quantification of cytokine mRNA using SYBR Green ® which is as sensitive as the TaqMan ® technique. Because the parameters of amplification are identical for all the cytokines developed, CyProQuant-PCR is readily automatable notably for 384-well plates and might allow multiple cytokine profiling of samples of very limited size at a relatively high throughput. CyProQuant-PCR opens the possibility to use cytokine mRNA measurements in clinical studies not only for an increased knowledge but also to help clinicians in patients' stratification and treatment decision. Methods Healthy human PBMC: isolation and in vitro stimulation Blood was collected from healthy donors at the French Blood Bank. Peripheral blood mononuclear cells (PBMC) were isolated by density separation over Ficoll Hypaque and washed two times in RPMI 1640 (Gibco BRL, Invitrogen, Cergy Pontoise, France). Cells were re-suspended in RPMI 1640 (Gibco BRL, Invitrogen, Cergy Pontoise, France) supplemented with 2 mM glutamine (Gibco BRL, Invitrogen, Cergy Pontoise, France) and 10% AB + human serum (French Blood Bank) at 2.10 6 cells/mL and either used directly for RNA extraction or cultured in duplicate with or without LPS (10 ng/mL, E. Coli O111: B7, Sigma, L'Isle d'Abeau Chesnes, France) or SAC (0,0075%, PanSorbine Cells, Calbiochem, La Jolla, CA, USA). After incubation, cells were washed with PBS and re-suspended in RNA-PLUS (Q-Biogene, Illkirch, France) for RNA isolation. Malaria patients and asymptomatic controls Twenty children admitted during the high malaria transmission season of 2001 to the emergency room at the Department of Child Health, Korle-Bu Teaching Hospital, Ghana were included. Five asymptomatic controls matched to patients for age, residence location and time of sample collection were enrolled. The general inclusion and exclusion criteria were as described by Kurtzhals et al. [ 25 ]. Parents or guardians signed informed consent forms. The study received ethical clearance from The Ethics and Protocol Review Committee at the university of Ghana Medical School and the Ministry of Health. Total cellular RNA was extracted from PBMC recovered from 500 μL of blood following supplier's instructions (RNA PLUS, Q-Biogene, Illkirch, France) after 22 hours of incubation at 37°C, 5% CO2 with or without LPS (10 ng/mL, E. Coli O111: B7, Sigma, L'Isle d'Abeau Chesnes, France). RNA isolation RNA was extracted following supplier's instructions, re-suspended in 60 μL of RNase-free water (Ambion, Huntingdon, UK) and quantified spectrophotometrically at 260 nm. Primers Oligonucleotide primers were synthesized at Eurogentec (Saraing, Belgium). To validate primers, a pool of cDNA from healthy human PBMC stimulated for 6 and 12 hours with LPS (10 ng/mL, E. Coli O111: B7, Sigma, L'Isle d'Abeau Chesnes, France) and PHA-L (10 μg/mL, Sigma-Aldrich, Lyon, France) was used. Analysis of the amplicons was assessed by 4% agarose gel electrophoresis and dissociation curve studies using Dissociation Curve Software (Applied Biosystems, Foster City, CA, USA). PCR products were cloned into pCR2.1 vector using Original TA cloning kit (InVitrogen, Cergy Pontoise, France) and sequenced (Genome Express, Meylan, France). Construction of the external DNA and RNA standards PCR products generated by each primer pairs were column-purified (Nucleospin, Macherey-Nagel, Hoerdt, France) and quantified spectrophotometrically at 260 nm. The molecular weight of the standard DNA was calculated by N487-[(N-1)175] were N is the number of bases composing the standard DNA. Stock solution of 10 12 copies of standard DNA /3,85 μL were made in Tris-EDTA buffer (Ambion, Huntingdon, UK), split in single-use aliquots and stored at -80°C in safe-lock tubes. External DNA standard range was made extemporaneously by 1:10 serial dilutions in water. Gene specific primers were fused on their 5' end to the sequence of the RNA polymerase T7 promoter to generate modified primers. These primers were used to amplify a gene specific PCR product flanked by transcription initiation sites. Five hundred nanograms of this construct were in vitro transcribed (MegaShortScript, Ambion, Huntingdon, UK). The standard RNA generated was purified (MegaClear, Ambion, Huntingdon, UK) and loaded on a 4% agarose gel for electrophoresis. The concentration of the standard RNA was determined spectrophotometrically at 260 nm. The molecular weight of the transcript was calculated by N500-[(N-1)175] were N is the number of bases composing the standard RNA. Stock solution of 10 12 copies of standard RNA /3,85 μL were made in RNA storage solution (Ambion, Huntingdon, UK), split in single-use aliquots and stored at -80°C in safe-lock tubes. External RNA standard range was made extemporaneously by 1:10 serial dilutions in water. Reverse transcription For CyProQuant-PCR assays, 100 ng of total cellular RNA from PBMC and serial dilution of external RNA standard were reverse transcribed simultaneously in a parallel procedure using Reverse Transcription TaqMan reagents (Applied Biosystems, Foster City, CA, USA) on a MasterCycler Gradient (Eppendorf, Le Pecq, France). The final volumes were set at 100 μL for the cellular RNA samples and 50 μL for the external RNA standard range. The thermocycling parameters were as follows: 25°C, 10 min.; 48°C, 60 min. and 95°C, 5 min. cDNA were immediately used for PCR amplification. Real-time RT-PCR and quantification of transcripts Reverse-transcribed standard RNA and cellular RNA were amplified simultaneously on the same PCR plate on an ABI Prism 7700 (Applied Biosystems, Foster City, CA, USA). An aliquot of 5 μL of the RT reaction was amplified in duplicate in a final volume of 30 μL of SYBR Green PCR Master mix (Applied Biosystems, Foster City, CA, USA). Thermocycling conditions were 50°C for 2 min., 95°C for 10 min. and 40 cycles of [95°C/15 sec.; 60°C, 1 min]. The sample target RNA copy numbers were calculated using SDS 1.9 Software (Applied Biosystems, Foster City, CA, USA). The baseline fluorescence was set manually to correct for differences in initial cDNA concentration and the threshold was positioned at a fluorescence level that was 10 times higher than the background signal. Target mRNA copy numbers in cellular samples were calculated based on a standard curve generated by SDS 1.9 Software (Applied Biosystems, Foster City, CA, USA) by plotting cycles at threshold (C T ) against the logarithmic values of the starting RNA standard copy number. ELISA tests TNF-α and MIF secreted proteins were quantified by sandwich ELISA following supplier's instructions (Bio-Source, Clinisciences, Montrouge, France and R&D, Lille, France respectively). Results are expressed as pg/mL for one million living cells. Statistical analysis Tests for significance were done using Stata software (Stata Corporation, College Station, Texas, 77845 USA) by Kruskal-Wallis rank test. Patent application Results disclosed in this manuscript have been protected in French patent application FR0408645. Authors' contributions PB developed the entire technique. IV did the in vitro stimulation experiments. DJ gave technical assistance. SL helped in RNA extraction of malaria samples. JCB introduced PB to molecular biology techniques and provided critical advices. BDA designed and conducted the study that yielded the malaria samples. OMP revised the manuscript and supported the work. CB conceived the strategy and coordinated the study. PB and CB drafted the manuscript. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555737.xml
529429	Blinding Trachoma: A Disease of Poverty	Trachoma accounts for 15% of blindness worldwide, affecting the world's poorest communities. How can the disease be controlled?	Trachoma is almost exclusively a disease of poor families and communities living in developing countries. It accounts for 15% of blindness worldwide—around 6 million people [ 1 ]. Although the disease is avoidable, it continues to blind. With so few voices speaking out on behalf of people affected by trachoma, it remains a neglected public health issue. Epidemiology Trachoma, a chronic keratoconjunctivitis, is caused by episodes of infection with Chlamydia trachomatis , an obligate intracellular bacterium. Only serovars A, B, Ba, and C are implicated in trachoma. Trachoma is the second leading cause of blindness worldwide [ 1 ]. According to the World Health Organization, currently 84 million people, mostly children, have active disease, and another 7.6 million people have trichiasis—a stage of trachoma in which the upper eyelid turns inward and one or more eyelashes rub against the eyeball [ 2 ]. An estimated 10% of the world's population lives in endemic areas and is at risk of developing trachoma. Global loss of productivity related to impaired vision and blindness from trachoma is thought to be as high as $US 5.3 billion annually [ 3 ]. More than 55 countries have been identified as endemic for trachoma, most of them in Africa and Asia ( Figure 1 ) [ 4 ]. Figure 1 The Worldwide Distribution of Trachoma (Map: Silvio Mariotti/WHO) Humankind has known trachoma since antiquity. Ibn-e-Isa, an Arab physician, was the first person to describe the different stages of trachoma and noted trichiasis as one of its sequelae. So prevalent was the disease not so long ago that trachoma hospitals were established in many parts of Europe and America [ 5 ]. The disease then declined dramatically in what is now called the developed world, mainly because of socioeconomic development [ 4 ]. Transmission occurs from eye to eye via hands, clothing, and other fomites. Flies have been identified as a major vector for the infection's spread [ 6 ]. The presence of open latrines favors the vector population ( Figure 2 ). Factors associated with trachoma include the extent to which the water supply is limited, the distance from the water source, the amount of water used for washing purposes, and overcrowding [ 7 ]. One case-control study in a Gambian village compared water use in 18 families having one or more active trachoma cases among the children with that in 16 trachoma-free families in the same village. The families with trachoma were found to use significantly less water per person per day for washing children than did the control group [ 8 ]. Figure 2 A Typical Community in Which Trachoma Is Endemic Some of the factors linked with the continued presence of the disease in affected communities are lack of access to water, overcrowding, lack of facial hygiene, eye-seeking bazaar flies, and open latrines. (Illustration: Aslam Bashir, Aga Khan University) The disease tends to cluster in certain communities within a village and certain families within a neighborhood. Women, especially in rural areas, are affected twice as often as men [ 9 ]. Clinical Manifestations and Grading Trachoma initially presents in childhood as red eye—itching, redness, and pain. The essential lesion is a trachomatous follicle (lymphoid cell aggregate) occurring typically in the upper tarsal conjunctiva ( Figure 3 ). The roughened appearance of the upper tarsal conjunctiva gives the disease its name (trachoma is the Greek word for “rough”). Trachomatous involvement of the cornea manifests itself initially as superficial keratitis. At a later stage, pannus formation (new vessel growth) may occur over the margin of the cornea, usually limited to the upper half. Figure 3 WHO Simplified Grading System: A Guide for the Assessment of Trachoma (Photos: from [ 11 ], with permission from WHO) In a subgroup of individuals, fibrosis occurs because of repeated infections, resulting in scarring of the conjunctiva (scarring trachoma). In scarring trachoma, the upper eyelid is shortened and distorted (entropion) and the lashes abrade the eye (trichiasis). Blindness results from corneal opacification, which is related to the degree of entropion or trichiasis [ 10 ]. Based on the presence or absence of some of the key signs of the disease, WHO has developed a simplified grading system for the assessment of trachoma ( Figure 3 ) [ 11 , 12 ]. The system can easily be used by non-specialists, after appropriate training, for the assessment of disease at the community level. Herbert's pits (healed follicles in the superior limbus) and Arlt's line (a horizontal scar on the upper tarsal conjunctiva) are two other classical features of the disease. Managing Trachoma: The SAFE Strategy WHO currently recommends the “SAFE” strategy for the management of trachoma: Surgery for trichiasis, Antibiotics for active disease, Facial hygiene, Environmental improvement to reduce the transmission of the disease [ 13 , 14 , 15 ]. Surgery. People with trachomatous trichiasis are at risk of blindness, and so treating these people is the first priority for the SAFE strategy. An evidence-based review of the SAFE strategy found that trichiasis surgery can alleviate discomfort and improve vision, though the evidence is less clear on whether such surgery prevents corneal opacification [ 14 ]. The review authors suggested that a protective effect of surgery against opacification is likely. There are different types of surgical procedures to correct trachomatous trichiasis [ 16 ]. Their high costs and the lack of surgical expertise in endemic regions, however, restrict the use of many of these as public health interventions. On the basis of a controlled trial by Reacher and colleagues [ 16 ], WHO recommends the bilamellar tarsal rotation procedure as the preferred technique; it is easy to perform and easy to learn [ 17 ]. Surgical effectiveness is defined in terms of recurrence of trichiasis; in the controlled trial, bilamellar tarsal rotation produced a recurrence rate of around 20% at follow-up 9–21 months after surgery, while other procedures saw 60% of patients with recurrence of trichiasis in the same period [ 16 ]. In several countries, different levels of health staff, including nurses and ophthalmic assistants, have been trained to perform the bilamellar tarsal rotation procedure. In addition to recurrence, there are other problems with the surgical approach to managing trachoma. It cannot correct all the complications, such as dry eyes. Even more important, and the main obstacle to preventing blindness from trachoma, is the low rate of uptake of surgery by communities with trachomatous trichiasis [ 14 ]. Barriers to uptake include distance to travel to surgery, perceived cost of the operation, child care duties, and lack of awareness about the treatment [ 14 ]. In Tanzania, less than a fifth of women with trichiasis opted for surgery, even when it was offered for free [ 18 ]. Offering surgery at the community level, rather than in distant medical facilities, is one strategy that could reduce travel times and costs and increase uptake. A cluster randomized controlled trial of village-based surgery versus health-center-based surgery in Gambia found a significantly higher uptake rate with the village-based service [ 19 ]. Antibiotics. The use of antibiotics aims to treat active infection and eliminate the reservoir. WHO currently recommends two regimens for the treatment of trachoma in endemic regions. These are 1% topical tetracycline ointment (twice daily for six weeks) or a single dose of oral azithromycin (1 g in adults and 20 mg/kg in children) [ 20 ]. Although antibiotics are a cornerstone of the SAFE strategy, clinical trials of antibiotics versus control (no treatment, placebo, or vitamin tablets) have produced conflicting results and are difficult to pool because of their heterogeneity. A recent Cochrane systematic review concluded “there is some evidence that antibiotics reduce active trachoma but results are not consistent and cannot be pooled” [ 20 ]. The review also found that “oral treatment is neither more nor less effective than topical treatment” [ 20 ]. Several questions remain about the use of antibiotics, such as who should receive them and how often. Lietman and colleagues have developed a mathematical model of frequency of treatment that uses available epidemiological data from a variety of countries [ 21 ]. Based on their model, they recommend that in areas where trachoma is moderately prevalent (less than 35% of children with active infection), it should be treated annually, but in hyperendemic areas (more than 50% of children with active infection), it should be treated biannually. Such models, however, need to be validated by well-designed clinical trials. Facial hygiene. Good facial hygiene aims to reduce transmission, the risk of autoinfection in a community, and the risk of attracting flies [ 13 , 15 ]. Many cross-sectional surveys have shown that children with clean faces are less likely to have trachoma, and are less likely to have severe trachoma [ 14 ]. A recent study in Mali found dirtiness of the face to be the most important risk factor associated with trachoma [ 22 ]. A Cochrane systematic review found evidence that face washing combined with topical tetracycline can be effective in reducing severe active trachoma [ 23 ]. However, the evidence does not support a beneficial effect of face washing alone or in combination with topical tetracycline in reducing non-severe active trachoma [ 23 ]. Interventions aimed at promoting facial hygiene have not yielded expected results in all settings, as behavioral change is not always readily achievable. Environmental improvement. This component of the SAFE strategy also aims to reduce transmission of trachoma by eliminating or reducing its risk factors, some of which are ubiquitous while others are specific to a region. Improving access to water is a key element. Other measures, such as provision of latrines to reduce the fly population, have also been found effective in reducing transmission [ 6 ]. Such environmental improvements will also provide other health benefits to a community, such as reduction in the incidence of diarrhea. As mentioned previously, there is an important association between water and trachoma—though the association is not a simple one. The distance to the water source constrains the amount of water used for hygiene practices. Improving access to water on its own, however, may not be enough. In the case-control study in Gambia, families with trachoma used less water per person per day for washing children than families without the disease, regardless of the amount of water available [ 8 ]. In other words, interventions aimed at increasing the availability of water should also promote its appropriate use. Getting community “buy in” for these interventions is important. Why Is Trachoma So Neglected? Trachoma is a disease of poor, underprivileged, and socioeconomically disadvantaged communities. It affects people who have little or no say in public decision making [ 24 ]. Investing in trachoma may sometimes mean compromising on other important issues. Many countries in which trachoma is endemic are also marred by regional conflicts, civil wars, and widespread corruption. Scarce resources are being spent on arms and debt servicing. These countries often lack the political commitment needed to fight against the disease. In addition, there is a lack of commitment by international donors. Still, there is some room for optimism, given WHO's vision of the global elimination of trachoma by the year 2020 and the efforts of the International Trachoma Initiative and other non-governmental organizations. The implementation of the SAFE strategy to eliminate blinding trachoma has already proven effective in several countries [ 25 ]. Many countries have already started trying to eliminate trachoma themselves. Future Directions Although the new initiatives in trachoma control are encouraging, trachoma elimination programs clearly need to be extended to many more countries. In addition, there are three crucial steps that still need to be undertaken if blindness from trachoma is to be eliminated. First, there needs to be more emphasis on the “F” and “E” components of the SAFE strategy. Antibiotics and surgery alone will not eliminate trachoma; work also needs to be done to eliminate the risk factors and decrease the transmission of the disease in affected communities. Such primary prevention is more likely to have a sustainable impact but requires a prolonged effort and investment [ 15 , 24 ]. Because elimination of trachoma requires improvement in education and hygiene practices, improved accessibility to water, and economic development of endemic regions, collaboration among departments and ministries is vital. An example of such collaboration is the recent involvement of the Water Supply and Sanitation Collaborative Council ( www.wsscc.org ) in trachoma control efforts. Similar partnerships need to be strengthened [ 25 ]. Socioeconomic development must be at the heart of control efforts—trachoma was eradicated from much of the developed world even before the advent of antibiotic programs for trachoma, and much of this eradication was attributable to socioeconomic development [ 26 ]. Second, research into different aspects of the disease should continue. Work on a vaccine for trachoma, although not successful thus far, should receive more attention [ 10 ]. Future research should look at risk factors for trachoma in diverse communities and at barriers to implementation of the SAFE strategy. Third, awareness about the disease and the SAFE strategy need to be promoted globally. At the same time, local, cost-effective solutions to trachoma need to be encouraged. Provision of pit latrines to reduce fly populations is just one such measure [ 6 ]. Unless these steps are taken, trachoma will continue to be a major cause of blindness in communities in the developing world [ 24 ].	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529429.xml
544835	Isolated low high density lipoprotein-cholesterol (HDL-C): implications of global risk reduction. Case report and systematic scientific review	Background The importance of low high-density lipoprotein cholesterol (HDL-C), elevated non HDL-C (as part of the metabolic syndrome, prediabetes, and type 2 diabetes mellitus), and an isolated low HDL-C is rapidly emerging. The antiatherosclerotic roles of reverse cholesterol transport and the pleiotropic antioxidant – anti-inflammatory mechanistic effects of HDL-C are undergoing rapid exponential growth. Case presentation In 1997 a 53-year-old Caucasian male presented with a lipoprotein profile of many years duration with an isolated low HDL-C and uric acid levels in the upper quintile of normal. He developed an acute myocardial infarction involving the right coronary artery and had percutaneous transluminal coronary angioplasty with stenting of this lesion. He also demonstrated a non-critical non-flow limiting lesion of the proximal left anterior descending coronary artery at the time of this evaluation. Following a program of global risk reduction this patient has done well over the past 7 years and remains free of any clinical signs and symptoms of atherosclerosis. His HDL-C and uric acid levels are currently in the normal physiological range. Conclusion Low HDL-C and isolated low HDL-C constitute an important risk factor for atherosclerosis. Therapies that lead to a return to normal physiologic range of HDL-C may result in the delay of atherosclerotic progression.	Case presentation MRH, a 53-year-old Caucasian male (physician) developed an acute inferior myocardial infarction (MI) associated with bradycardia and occasional PVCs. Emergency medication included aspirin, nitroglycerin and a bolus of TPA. The cardiology team preformed PCTA at the site of near complete blockage of the right coronary artery with successful stent placement. At this time a non-critical 40% lesion located in the proximal left anterior descending coronary artery was noted, which was not manipulated. The patient was discharged following 24 hours of stable monitoring. Past Medical History Relapsing fever 1971 full recovery, spontaneous left pneumothorax times two (1982–83), lumbar fusion back surgery 1985, and Herpes Simplex encephalitis 1989 with full recovery. Family History Mother with CVA (cerebellar) age 58 full recovery. Died of Hodgkin's lymphoma 64. Brother with type 1 diabetes mellitus with onset at age 29 (known PAD and aorto-femoral bypass age 49) died in sleep age 51. Father with CVA (vertebrobasilar) age 75 with full recovery, COPD, died in sleep while recovering from TIA and pneumonia age 84. Grandparents lived to their 80s and died of old age. Social History High stress family physician who seldom drank alcohol and smoked a pipe occasionally. Blood pressure at times of high stress would elevate to 140/85–88 and return to 120–130s/ 70–75 at times of non-stress in the office. He was physically active with no dedicated exercise program Laboratory Values Five months prior to MI and reflective of numerous metabolic profiles over the preceding decades. Total cholesterol 198 mg/dL Triglycerides 154 mg/dL HDL-C 34 mg/dL. HDL-C (1970–1973 32 mg/dL and 34 mg/dL) LDL-C calculated 120 mg/dL Non HDL-C = (198-34) = 164 Total Chol/HDL ratio = 6.2 > than 5 and is high Uric acid 6.5 mg/dL Blood sugar non-fasting 102 mg/dL Homocysteine first week post MI fasting: 28 mcmol/L LFTs, electrolytes, calcium and phosphorus, serum iron, renal function, and CBC were all in normal range. Patient started a program reflecting the global risk reduction approach described in the RAAS acronym (table 1 ) and is currently taking an angiotensin receptor blocker, aspirin, beta blocker, folic acid, and a statin. Patient was intolerant of ACE inhibitor therapy due to cough and fatigue and has been unable to tolerate niacin on numerous attempts both pre and post MI due to incapacitating headaches. Table 1 The RAAS acronym: global risk reduction R Reductase inhibitors (HMG-CoA). Decreasing modified LDL-cholesterol, i.e. oxidized, acetylated LDL-cholesterol. Decreasing triglycerides and increasing HDL-cholesterol Improving endothelial cell dysfunction. Restoring the abnormal Lipoprotein fractions. Thus, decreasing the redox and oxidative stress to the arterial vessel wall and myocardium. Redox stress reduction. A AngII inhibition or blockade: ACEi-prils. ARBS-sartans. Both inhibiting the effect of angiotensin-II locally as well as systemically. Affecting hemodynamic stress through their antihypertensive effect as well as the deleterious effects of angiotensin II on cells at the local level – injurious stimuli -decreasing the stimulus for O 2 . production. Decreasing the A-FLIGHT toxicities. Plus the direct-indirect antioxidant effect within the arterial vessel wall and capillary. Antioxidant effects. Aspirin antiplatelet, anti-inflammatory effect. Adrenergic (non-selective blockade) in addition to its blockade of Prorenin→Renin Amlodipine with its calcium channel blocking antihypertensive effect, in addition to its direct antioxidant effects. Redox stress reduction. A Aggressive control of diabetes to HbA 1c of less than 7. (This usually requires combination therapy with the use of: Insulin secretagogues, insulin sensitizers (thiazolidinediones), biguanides, alpha-glucosidase inhibitors, and ultimately exogenous insulin.) Decreasing modified LDL cholesterol, i.e. glycated – glycoxidated LDL cholesterol. Improving endothelial cell dysfunction. Also decreasing glucotoxicity and the oxidative – redox stress to the intima and pancreatic islet. Aggressive control of blood pressure , which usually requires combination therapy, including thiazide diuretics to attain JNC 7 guidelines. Aggressive control of dyslipidemias , which frequently requires combination therapy (especially in the metabolic syndrome and T2DM), including TLC, statins, fibrates, selective cholesterol inhibitors such as ezetimibe, and niacin Aggressive control of Hcy with folic acid with its associated additional positive effect on re-coupling the eNOS reaction by restoring the activity of the BH4 cofactor to run the eNOS reaction and once again produce eNO. Redox stress reduction. S Statins. Improving plaque stability (pleiotropic effects) independent of cholesterol lowering. Improving endothelial cell dysfunction. Plus, the direct – indirect antioxidant anti-inflammatory effects within the islet and the arterial vessel wall promoting stabilization of the unstable, vulnerable islet and the arterial vessel wall. Style: Lifestyle modification: lose weight, exercise, and change eating habits. Stop Smoking Redox stress reduction Current Laboratory Values 2004: Total cholesterol: 138 mg/dL Triglycerides: 94 mg/dL HDL-C: 45 mg/dL LDL-C calculated: 74 mg/dL Non HDL-C: (138-45) = 93 Total Chol/HDL ratio = 3.0 Uric acid: 6.5 mg/dL Blood sugar: Fasting 80 mg/dL, 2 hour post prandial 118 mg/dL Homocysteine: 7.2 mcmol/L Lp(a): 4.2 mg/dL in normal range immediate post MI and again at this time: 4.3 mg/dL. hs-CRP: 0.7 mg/L. LFTs, electrolytes, calcium and phosphorus, serum iron, renal function, and CBC are all in normal range. This patient has done well over the past seven years and remains free of any clinical signs and symptoms of cardiovascular disease. While this patient will always remain a CHD risk, his current laboratory values remain in a normal physiological range. As noted above his HDL-C and uric acid levels are currently in the normal physiological range and his hs-CRP remains in the second quartile. Comment According to Framingham risk scores associated with the National Cholesterol Education Program Adult Treatment Panel III (NCEP ATP III) guidelines [ 1 ], few would have recommended any treatments other than therapeutic lifestyle changes (TLC) and possibly niacin, which our patient was intolerant both pre and post event in 1997. If we score this patient according to the estimate of 10-year risk for men he gets 6 points for age 53, 2 points for total cholesterol 160–199 age 53, 3 points for being a pipe smoker, 2 points for HDL being < 40 mg/dL, and 1 point for systolic blood pressure 140–159 untreated. This totals 14 points and results in an estimated 10-year risk for men of 16%, which is less than the 20% recommended for more aggressive therapy. Even if we apply the NCEP ATP III guidelines of having two plus risk factors: Male sex, hypertension, smoking, and low HDL-C with a 10 risk < or = to 20% we obtain the following recommendations: LDL-C goal < 130 mg/dL, initiation of TLC if LDL-C is = or > 130 mg/dL, consideration of drug therapy if LDL-C is > or = to 130 mg/dL after three months of TLC. It is important to note that our patient had a LDL-C of 120 mg/dL prior to his event. Even if we look at the non HDL-C levels, which are allowed to be 30 mg/dL higher than LDL-C goals we have a patient with a non HDL-C of only 164. MRH became a CHD risk patient within a short period of time of 5 months. Discussion The importance of low HDL-C and cardiovascular disease associated with the lipid triad (Low HDL-C, elevated triglycerides, and increased small dense LDL-C) found in the metabolic syndrome (metS) and overt type 2 diabetes mellitus (T2DM) and a contributing factor to the elevated non HDL-C discussed in the current NCEP ATP III guidelines or the patients with isolated low HDL-C is rapidly evolving. The accelerated atherosclerosis (atheroscleropathy) associated with the metS and T2DM has been previously reviewed and is definitely a serious problem associated with the current epidemic of obesity – diabesity and T2DM [ 2 - 4 ]. Both isolated low HDL-C and elevated non HDL-C (total cholesterol minus HDL-C) levels are difficult to get to known NCEP ATP III recommendations and this task usually requires combination therapy. These therapies consist of therapeutic life style changes and pharmacotherapy including statins, fibrates, selective cholesterol inhibitors such as ezetimibe, and niacin in addition to a global risk reduction of all non HDL-C existing risk factors (table 2 ) [ 5 ]. Table 2 Effects of drugs on HDL-C levels DRUG PERCENT INCRESE Nicotinic acid (niacin) 15% – 35% Fibrates 10% – 15% Estrogens 10% – 15% Statins Coupled Dual Effect Associated with potent LDL-C reduction, which make the statins "shine" 5% – 10% Alpha blockers 10% – 20% Alcohol (in moderation) 10% Ezetimibe 3% In this case report a focus on isolated low HDL-C is appropriate. This case report demonstrates a marked improvement of all lipid parameters including his low HDL-C. However, this marked improvement is not always as simple as this case and therefore, both the patient and the clinician need to be very patient, as well as, creative in order to achieve global risk reduction [ 5 ]. Isolated low HDL-C In 1977 the Tromso Heart Study demonstrated that CAD patients have HDL-C levels 35% lower than controls and those patients with low HDL-C are three times more likely to develop CAD than those with elevated LDL-C [ 6 ]. These early views certainly support the concept that an isolated low HDL-C is a common antecedent of clinical CHD, as well as being important in accelerating the progression of atherosclerosis. The inverse relation of HDL-C to CHD events has been widely discussed since the original publication of data from the Framingham study (1986) [ 7 , 8 ]. Castelli WP et al. were able to show an inverse association of high HDL-C and low coronary risk was as statically as strong as the direct association of high LDL-C and high coronary risk in a cohort of men and women age 40–82 followed for 12 years who were free from CAD at study entry. At any level of cholesterol low HDL-C increases the rate of CHD [ 1 ] The NCEP ATP III guidelines clearly defines a level < 40 mg/dL as an independent risk factor for CHD [ 1 ]. Raising HDL-C is not a target for either primary or secondary prevention at this time, however its importance as a tertiary target is rapidly emerging. Michael Miller has stated: "Low HDL-C is the most common lipoprotein abnormality in patients with CHD and is predictive of CHD events, even when total cholesterol levels are normal" [ 9 ]. Goldbourt U et al. , found that the prevalence of isolated low HDL-C as a risk factor for CHD mortality to be present in one out of six or 16.6 % while studying a 21-year follow up of 8000 men [ 10 ]. Furthermore, they found that an excess CHD risk associated with isolated low HDL-C appeared particularly increased in men with diabetes mellitus, whose death rate was 65% higher than in diabetics with HDL-C > 0.9 mmol/L or 36 mg/dL. There are at least eight secondary causes for low HDL-C (table 3 ) and at least seven drugs that have a positive effect on raising HDL-C (table 2 ). As demonstrated in our case report, the beneficial effects of raising HDL-C with statin therapy and a program of global risk reduction have been positive in preventing the progression of atherosclerosis and recurrent acute coronary syndromes (table 4 ). Table 3 Secondary causes of low HDL-C 1. Elevated triglycerides. (Component of metS) End stage renal disease Hypothyroidism [also increased total Chol/HDL-C ratio. 2. Obesity and Overweight. – [waist measurement] – (Component of metS) [Visceral obesity in particular] For every 3 kg. (7 lbs.) weight loss HDL-C increased 1 mg/dL. 3. Prediabetes and overt Type 2 Diabetes Mellitus. (Component of metS) 4. Physical inactivity (lifestyle choice). 5. Smoking (lifestyle choice). 6. Very high carbohydrate intakes > 50–60% of energy (lifestyle choice). [Especially Fructose Containing Soft Drinks.] 7. Metabolic Syndrome: As potent a risk factor as smoking. 8. Drugs, such as beta-blockers, anabolic steroids, and progestational agents. Table 4 Beneficial effects of HDL-C REVERSE CHOLESTEROL TRANSPORT Accepts cholesterol from the macrophage and tissues and transports it back to the liver for disposal in the bile (figure 1). Acts a an apoprotein donor to the other lipoproteins ANTIOXIDANT Antioxidant activity (through intimal paraoxonase, and redox -sensitive methionine residues of apo A-1) Increases eNOS and endothelial nitric oxide ANTIINFLAMMATORY Downregulates adhesion molecule expression on endothelium: (I-CAM, V-CAM and MCP-1) Inhibits neutrophil degranulation ANTITHROMBOTIC Antithrombotic activity via its ability to block TxA 2 and potentiates activity of proteins: C and S. Stimulates prostacyclin production (antithrombotic and vasodilitory). ENDOTHELIAL PROTECTION PROTERTIES Acts as an endothelial mitogen and inhibits endothelial cell apoptosis: This would help to decrease the incidence of plaque erosion and promote plaque stabilization Stimulates endothelial nitric oxide (eNO and its enzyme eNOS) and prostacyclin production with vasodilatation, antioxidant, and anti-inflammatory properties. HDL-C is synthesized in the intestine and liver and is extremely important in reverse cholesterol transport from the tissues to the liver for disposal. It works in conjunction with the ABCA1 cholesterol transporter within the intimal macrophages (figure 1 ). Figure 1 Reverse Cholesterol Transport. This figure demonstrates the process of reverse cholesterol transport. It begins in the arterial vessel wall and with the assistance of the ATP binding cassette transporter A-1 (ABCA-1) and in collaboration with the Apo A-1 protein attached to the outer shell of the nascent HDL-C lipoprotein particle free cholesterol is internalized within the HDL-C lipoprotein particle. The enzyme lecithin cholesterol acyltransferase (LCAT) esterifies free cholesterol (FC) via a lipidation process and internalizes it within the HDL-3, which matures to a larger HDL-2 lipoprotein particle. From this point in time the HDL-3 and 2 particles can enter the hepatic cycle via the Scavenger Receptor B-1 and subsequently excreted in the bile. The alternative pathway is for the larger HDL-C apoA-1 lipoprotein particles to undergo a transference of the cholesterol esters through an exchange process with triglycerides via cholesterol ester transfer protein (CETP) to the ApoB-100 lipoprotein particles and enter the liver for further metabolism via the low density lipoprotein receptor (LDLR) to be subsequently excreted in the bile. This important dual interaction of HDL-C and the ABCA1 transporter is of great importance and recently we have learned that certain gene polymorphism of ABCA1 transporter may have a profound effect on HDL-C in addition to the well known abnormality of Tangier disease [ 11 ]. Probst MC et al. , have even set aside a website to list all of the known ABCA1 gene polymorphisms [ 11 ]. Oxidative stress and the reductive stress (redox stress) associated with overt T2DM and multiple risk factors associated with accelerated atherosclerosis may result in a damaging effect to HDL-C and interfere with the ABCA1 transporter in reverse cholesterol transport via a mechanism of oxidation and nitration of tyrosine residues on the apo A-1 lipoprotein outer shell of HDL-C lipoprotein [ 12 ]. This biochemical alteration of the apo A-1 lipoprotein could disable the process of reverse cholesterol transport and aggravate an underlying isolated low HDL-C level. Lifestyle changes that are important in raising low HDL-C consist of smoking cessation, weight loss, exercise, and the use of alcohol in moderation. HDL-C has numerous positive effects on the endothelium and arterial vessel wall, which decrease non-diabetic atherosclerosis and the accelerated atherosclerosis – atheroscleropathy associated with metS and overt T2DM (table 4 ). Emerging novel risk markers of atherosclerosis NCEP ATP III allows the clinician to factor in the additional risks associated with novel, emerging risk markers such as our patients elevated homocysteine. Other risk markers would be the lipid markers: Elevated triglyceride, remnant lipoproteins, lipoprotein (a), an abnormal TC/HDL-C ratio, small dense LDL particles, HDL subspecies, and apolipoprotein A and B. The non lipid markers would include: An elevated glucose, inflammatory markers (elevated hs-CRP and the emerging importance of the various interleukins and in particular IL-6, which is the driving force behind hs-CRP elevation), coagulation markers (elevated PAI-1, Lp(a), and fibrinogen), the emerging roles of matrix metalloproteinases (MMPs) and of course the established risk marker of an elevated homocysteine. It is interesting to note that Qujeq D et al., noted a negative correlation between total homocysteine and HDL-C levels (p < 0.05, r = 0.93) in a study evaluating 126 patients (67 male and 59 females, aged 29–73 mean of 48.65 +/- 5.79) with unequivocal changes of acute myocardial infarction in the electrocardiogram as compared to 135 normal healthy controls, while noting a positive correlation between total homocysteine and LDL-C levels (p < 0.05, r = 0.98) [ 13 ] The reader may note that the patients' uric acid level was quite elevated prior to his acute coronary event and that this level returned to a very normal level following global risk reduction and aggressive therapy for his multiple risk factors in addition to his isolated low HDL-C. Although not a considered a risk factor or even an emerging, novel risk marker, uric acid may be a quite sensitive marker of underlying redox and oxidative stress. Uric acid levels greater than 4 mg/dL may be considered a red flag in those patients, such as our case report, with high risk for CHD [ 14 ]. The Atherosclerotic Kitchen Sink When viewing the sources for atherosclerosis it is important to note that there are two routes for accumulation of atherogenic lipoproteins (input and outflow) within the arterial intima and subsequent remodeling of the arterial vessel wall. The atherosclerosis equation: Lipoprotein Accumulation (retention) in the arterial vessel wall = Lipoprotein in - lipoprotein out. L-A avw = L-in - L-out. L-in, would equal the net lipoproteins derived from the GI tract (absorption) plus that synthesized by the liver. Lipoprotein out is strictly via reverse cholesterol transport to the liver and secretion via bile into the gut. L-in is primarily the beta lipoproteins or apolipoprotein B containing lipoprotein particles, whereas L-out depends primarily on the alpha lipoproteins, apolipoprotein A or HDL-C. The beta lipoproteins are atherogenic and the alpha lipoproteins are antiatherogenic. From this analogy one can see can see why non HDL-C was so important in the recent NCEP ATPIII guidelines: non HDL-C = total Cholesterol – HDL-C (reflecting the total atherogenic burden). This is also why the recent global (52 countries) INTERHART study found the ApoB/ApoA-1 ratio (the ratio of atherogenic lipoproteins to non atherogenic lipoproteins) to be the best predictor of CHD (odds ratio of 3.25 for top verses lowest quintile) as compared to the other eight other risk factors (table 5 ) [ 15 ]. Table 5 Nine risk factors account for up to 90 % of MIS worldwide in both sexes, all ages, and in all regions RISK FACTOR ODDS RATIO Abnormal lipids: ApoB/ApoA-1 3.25 Smoking 2.87 Diabetes 2.37 Hypertension 1.91 Abdominal obesity Reason for such a high OR: This could aggravate smoking, diabetes, hypertension, obesity, alcohol abuse and even nutrition (eating aggressively) 1.12 Psychosocial Factors 2.67 Alcohol use 0.91 Physical Activity 0.86 Consumption of fruits and vegetables 0.70 L-Aavw = ApoB/ApoA-1 ratio of the INTERHART study L-in, would be comparable to the faucet (GI tract and Liver) delivering the atherogenic apoB lipoproteins. While the kitchen sink would represent the accumulation of atherogenic lipoproteins within the arterial vessel wall or L-Aavw. In a like manner, the DRAIN would represent L-out or HDL-C or apoA-1 lipoproteins. From this analogy it can easily be seen that if there is inadequate HDL-C or apoA-1 the atherogenic kitchen sink will overflow and result in acute coronary syndromes as happened in our case report (figure 2 ). Figure 2 The Atherosclerotic Kitchen Sink. This image portrays the importance of the HDL-C drain in maintaining a certain level of atherogenic lipoproteins within the arterial vessel wall to prevent accumulation and the undesirable possibility of an acute event with overflow or acute coronary syndromes. This simple analogy of homeostasis points to an important concept: That being the frequent need for combination therapy in order to control the various components of the atherogenic lipoprofile. Isolated low HDL-C is certainly a red flag regarding the development of atherosclerosis and CHD and additionally the elevation of low HDL-C levels may have a DRANO-LIKE effect to open a clogged drain in an atherosclerotic arterial vessel wall. Conclusion While the treatment of isolated HDL-C may seem overwhelming at times, it will be rewarding for both the clinician and the patient as demonstrated by the our case study. This patient has done well for seven years and it is anticipated he will continue to do well with his laboratory values now in a sustained, normal physiologic range. Additional tests by nuclear magnetic resonance spectroscopy (NMR LipoProfile) would assist us in knowing the LDL particle number (LDL-P) and would assist us in even more aggressive therapy. In addition to his current goals he has met, he should have an LDL-P under 1000 micromol/L and small LDL-P under 700 micromol/L. Even though we have discussed LDL-C from a quantity perspective, due to an isolated low HDL-C, we should additionally be aware that there exists and equally important role for the quality of HDL-C [ 12 ]. Recently, the Apo A-1 Milano and Apo A-1 Paris have resulted in a marked increase in research interest for the HDL-C lipoprotein particle and its future manipulation [ 16 ]. In the near future we may be utilizing gene transfer utilizing variations of the Milano and Paris forms, as well as the newer apoA-1 mimetics such as L-4F [ 17 ]. Recently there has been increased interest in CETP inhibitors and Phase II studies are underway with torcetrapib and the combination of torcetrapib and atorvastatin [ 18 ]. Additional attention to the PPAR agonists and atherosclerosis and the liver X receptor alpha (LXR alpha) agonists is being employed at the present and the positive dual effects on HDL-C and atherosclerosis is being actively investigated. This dual agonism of PPAR alpha, gamma, and possible delta, as well as the dual effects of PPAR alpha and LXR alpha are quite exciting and we will learn a great deal regarding their effects on atherosclerosis and HDL-C in the near future [ 19 ]. Recently John Snow, M.D. (1813–1858), a legendary figure in the field of epidemiology, of London, England was honored [ 20 ]. He hypothesized that Cholera was transmitted by water rather than miasma (bad air). He suspected the water from the Broad Street pump was the source of the disease and subsequently had the pump handle removed in 1854 (150 years ago) [ 21 ]. Could low HDL-C be the "pump handle" of atherosclerosis and CHD? List of abbreviations ABCA-1: ATP binding cassette transporter A-1 BMI: body mass index CHD: coronary heart disease CAD: coronary artery disease hs-CRP: highly sensitive C reactive protein T2DM: type 2 diabetes mellitus metS: metabolic syndrome HDL-C: high density lipoprotein cholesterol LDL-C: low density lipoprotein cholesterol LFTs: liver function tests PCTA: percutaneous transluminal coronary angioplasty VLDL-C: very low density lipoprotein cholesterol TC: total cholesterol TLC: therapeutic lifestyle changes TPA: tissue plasminogen activator Competing interests The author(s) declare that they have no competing interests. Author contribution MRH conceived the idea to write this manuscript. MRH and SCT wrote, and edited this manuscript together.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544835.xml
368160	Taste Perception: Cracking the Code	Our sense of taste begins with taste buds and ends in the brain. Researchers are beginning to unravel the mechanisms and connections that lie in between	The ability to taste food is a life-and-death matter. Failure to recognise food with a high enough caloric content could mean a slow death from malnutrition. Failure to detect a poison could result in near-instant expiration. And now, as researchers begin to understand some of the nuts and bolts of taste perception, it seems that the sense of taste may also have more subtle effects on health. The Basics of Taste At the front line of the taste sensory system are the taste buds—onion-shaped structures on the tongue and elsewhere in the mouth ( Figure 1 ). Up to 100 taste receptor cells—epithelial cells with some neuronal properties—are arranged in each taste bud. In the tongue, the taste buds are innervated by the chorda tympani (a branch of the facial nerve) and the glossopharyngeal nerve. These nerves carry the taste messages to the brain. Figure 1 A Taste Bud in a Mouse This taste bud was taken from a transgenic mouse in which the marker green fluorescent protein is being driven by the T1R3 promoter; 20%–30% of the cells in the taste bud are expressing T1R3. (Photograph courtesy of Sami Damak, Mount Sinai School of Medicine, New York, New York, United States.) Taste is the sense by which the chemical qualities of food in the mouth are distinguished by the brain, based on information provided by the taste buds. Quality or ‘basic taste’, explains Bernd Lindemann, now retired but an active taste researcher in Germany for many years, is a psychophysical term. Large numbers of people describe different tastants and then statistical analyses are used to define the important tastes. ‘The number of taste qualities has varied over the years’, says Lindemann. ‘We are now settling at around five, though I would not be surprised if some additional qualities pop up’. The five qualities that Lindemann refers to are salty, sour, bitter, sweet, and umami, the last being the Japanese term for a savoury sensation. Salty and sour detection is needed to control salt and acid balance. Bitter detection warns of foods containing poisons—many of the poisonous compounds produced by plants for defence are bitter. The quality sweet provides a guide to calorie-rich foods. And umami (the taste of the amino acid glutamate) may flag up protein-rich foods. Our sense of taste has a simple goal, explains Lindemann: ‘Food is already in the mouth. We just have to decide whether to swallow or spit it out. It's an extremely important decision, but it can be made based on a few taste qualities’. From Physiology to Molecular Biology Taste has been actively researched for many decades. During the 20th century, electrophysiologists and other researchers worked hard to understand this seemingly simple sense system. Then, in 1991, the first olfactory receptors were described. These proteins, which are exposed on the surface of cells in the nose, bind to volatile chemicals and allow us to detect smells. This landmark discovery, in part, encouraged many established taste researchers to investigate the molecular aspects of taste. The olfaction results also enticed researchers from other disciplines into the taste field, including collaborators Charles Zuker (University of California, San Diego [UCSD], La Jolla, California, United States) and Nick Ryba (National Institute of Dental and Craniofacial Research [NIDCR], Bethesda, Maryland, United States). About six years ago, explains Zuker, who previously worked on other sensory systems in flies, ‘there was a disconnect between our understanding of sensations in the case of photoreception, mechanoreception, touch, and so on and what we knew about taste’. There was evidence, says Ryba, that a class of protein receptors called G-protein-coupled receptors (GPCRs) were involved in sweet and bitter taste, ‘but the receptors weren't known, so we started to look for them …. These molecules are intrinsically interesting, but more importantly, they provide tools with which we can dissect out how taste works’. Bitter, Sweet, and Umami Receptors The bitter receptors fell first to the onslaught of the UCSD–NIDCR team and other molecular biologists. In 1999, the ability to taste propylthiouracil, a bitter tasting compound, had been linked to a locus on human Chromosome 5p15. Reasoning that this variation might be due to alterations in the coding sequence for a bitter receptor, the UCSD–NIDCR researchers used the draft of the human genome to search for sequences that resembled GPCRs on Chromosome 5p15. ‘That was how we found T2R1, the first bitter receptor, and, subsequently, a whole family of T2Rs’, says Zuker. Researchers want to know: how is taste coded? All these receptors, says Zuker, are coexpressed in bitter taste receptor cells, a result that contradicts other research showing that different bitter-responsive cells react to different bitter molecules. ‘To me’, says Zuker, ‘it makes sense that all the bitter receptors would be expressed in each bitter taste cell. We just need to know if something is bitter to avoid death’, not the exact identity of the bitter tastant. The sweet receptor story started in 1999 with the identification of two putative mammalian taste receptors, GPCRs now known as T1R1 and T1R2. In early 2001, four groups reported an association between the mouse Sac locus, which determines the ability of mice to detect saccharin, and T1R3, a third member of the T1R family. The UCSD–NIDCR team subsequently showed that the T1R2 and T1R3 heterodimer (a complex of one T1R2 and one T1R3 molecule) forms a broadly tuned sweet receptor, responsive to natural sugars and artificial sweeteners, and that a homodimer of two T1R3 molecules forms a low-affinity sugar receptor that responds to high concentrations of natural sugars only. All sweet detection, concludes Zuker, is via the T1R2 and T1R3 receptors. And umami? A truncated glutamate receptor was identified as an umami receptor by researchers at the University of Miami (Florida, United States) School of Medicine in 2000. Zuker, however, believes that the one and only umami receptor is a heterodimer of T1R1 and T1R3. In October 2003, Zuker and his coworkers reported that mice in which either T1R1 or T1R3 has been knocked out show no preference for monosodium glutamate (MSG), an umami tastant. However, other researchers reported in August 2003 that T1R3 knockouts retain some preference for MSG. ‘We believe this is due either to the truncated glutamate receptor or another unknown receptor’, says lead author Sami Damak (Mount Sinai School of Medicine, New York, New York, United States). Damak says he does not know why the two sets of T1R3 knockout mice behaved differently, but the UCSD–NIDCR researchers suggest that the residual response to MSG seen by Damak et al. is a response to the sodium content of MSG. Damak is not alone, however, in thinking there may be more than one umami receptor (and additional sweet receptors). Commenting on these recent discoveries, taste expert Linda Bartoshuk (Yale University School of Medicine, New Haven, Connecticut, United States) says that ‘it is lovely to see all these details, especially as they confirm what we already believed conceptually’. For example, she says, it is no surprise that there are many bitter receptors but probably only one sweet receptor. ‘There are so many poisons and it makes perfect sense to have lots of receptors feeding into a common transduction pathway. Sweet is a different problem. In nature, there are many molecules with structures similar to sugar that we must not eat because we cannot metabolise them. So I would have predicted one or at most a few highly specific sweet receptors’. What about Salty and Sour Receptors? The salty and sour receptors may be very different from the GPCRs involved in bitter, sweet, and umami perception, which bind complex molecules on the outside of the cell and transmit a signal into the cell. For salty and sour perception, the taste cell only needs to detect simple ions. One way to do this may be to use ion channels—proteins that form a channel through which specific inorganic ions can diffuse. Changes in cellular ion concentrations could then be detected and transmitted to the nervous system. Physiologist John DeSimone (Virginia Commonwealth University, Richmond, Virginia, United States) says there are at least two ion channel receptors for salt in rodent taste receptor cells. The first of these is the epithelial sodium channel, a widely expressed channel that can be blocked specifically with the drug amiloride. In rats, says DeSimone, only 75% of the nerve response to salt can be blocked by amiloride, so there is probably a second receptor. This, he says, seems to be a generalist salt receptor—the amiloride-sensitive channel only responds to sodium chloride—and may be the more important receptor in people. Sour tastants are acids, often found in spoiled or unripe food. DeSimone's current idea is that strong acids enter taste cells through a proton channel (probably a known channel present on other cell types) while weak acids, like acetic acid (vinegar), enter as neutral molecules and then dissociate to lower intracellular pH. DeSimone believes that he has identified the proton channel involved in sour taste as well as an ion channel that could be the second salt receptor, and he plans to do knockout experiments on both. If these channels are essential elsewhere in the body, as DeSimone suspects, to avoid lethality he will need to construct conditional knockouts in which the channel is lost only in the taste receptor cells. Zuker, meanwhile, is not convinced that the current ion channel candidates for salt and sour perception are correct. And, he says, GPCRs could also be involved in these modalities. ‘There is a precedent for that’, he claims, noting that extracellular calcium is sensed by a GPCR. Taste-Coding With many taste receptors now identified, researchers are turning to a long-standing question in taste perception: how is taste coded? When we eat, our tongue is bombarded with tastants. How is their detection and transduction of information organised so that the appropriate response is elicited? Taste physiologist Sue Kinnamon (Colorado State University, Fort Collins, Colorado, United States) explains the two theories of taste-coding. In the ‘labelled-line’ model, sweet-sensitive cells, for example, are hooked up to sweet-sensitive nerve fibres that go to the brain and code sweet. If you stimulate that pathway, says Kinnamon, ‘you should elicit the appropriate behavioural response without any input from other cell types’. In the ‘cross-fibre’ model, the pattern of activity over many receptors codes taste. This model predicts that taste receptor cells are broadly tuned, responding to many tastants. Support for this theory, says Kinnamon, comes from electrical recordings from receptor cells and from nerves innervating the taste buds that show that one cell can respond to more than one taste quality. Zuker and Ryba's recent work strongly suggests that taste-coding for bitter, sweet, and umami fits the labelled-line model in the periphery of the taste system. Their expression data show that receptors for these qualities are expressed in distinct populations of taste cells. In addition, in early 2003, they reported that, as in other sensory systems, a single signalling pathway involving the ion channel TRPM5 and PLCβ2, a phospholipase that produces a TRPM5 activator, lies downstream of the bitter, sweet, and umami receptors. When the UCSD–NIDCR researchers took PLCβ2 knockout mice, which did not respond to bitter, sweet, or umami, and engineered them so that PLCβ2 was only expressed in bitter receptor-expressing cells, only the ability to respond to bitter tastants was regained. These data, says Zuker, support the labelled-line model. The latest data supporting the labelled-line model came last October when Zuker and colleagues described mice in which a non-taste receptor—a modified κ-opioid receptor that can only be activated by a synthetic ligand—was expressed only in cells expressing T1R2, sweet-responsive cells. The mice were attracted to the synthetic ligand, which they normally ignore, indicating that dedicated pathways mediate attractive behaviours. The researchers plan similar experiments to see whether the same is true for aversive behaviours. Even with all these molecular data, the cross-fibre model of taste-coding still has its supporters—just how many depends on whom one talks to. Both Damak and Kinnamon, for example, believe that there is at least some involvement of cross-fibre patterning even in the taste receptor cells. But, says neurobiologist and olfaction expert Lawrence C. Katz (Duke University, Durham, North Carolina, United States), ‘the onus is now on people who believe otherwise [than the labelled-line model] to provide compelling proof for the cross-fibre theory because now, at least at the periphery, the evidence is compelling for a labelled line for bitter, sweet, and umami’. Bartoshuk also says the debate is decided in favour of the labelled-line model in the periphery. The crossfibre model is an interesting historical footnote, she comments. Whether this putative link between taste perception and health can be confirmed and whether it will be possible to manipulate food preferences to improve health remain to be seen. However, it seems certain that, as in the past five years, the next five years will see large advances in our knowledge of many aspects of taste, a fascinating and important sensory system. What Next—and Why Study Taste Anyway? The periphery of the taste sensory system has yielded many of its secrets, but relatively little is known about the transduction pathways in taste, how taste cells talk to the nervous system, or about events further downstream in the brain. How are signals from taste receptors integrated with those from olfactory receptors to form a representation of complex food flavours, for example? With their expanding molecular toolbox, researchers can now delve deeper into these aspects of taste perception. This may tell us not only about taste but about how the nervous system in general is put together, says Ryba. But understanding taste is not just an academic exercise. It has practical uses too. DeSimone suggests that by understanding salt receptors, it may be possible to design artificial ligands to help people lower their salt intake. As Kinnamon succinctly puts it, ‘Can you imagine eating potato chips and not having the salty component?’ An artificial salt receptor ligand could make salt-free foods a palatable option for people with high blood pressure. Lindemann also sees a great future in artificial ligands for taste receptors. The sense of taste is partly lost in elderly people, he says, so better tastants—effectively ‘chemical spectacles’—might give them back their pleasure of eating and thereby improve their quality of life. Finally, some aspects of taste may be inextricably tied up with general health, says Bartoshuk. Many people who can taste propylthiouracil are also ‘supertasters’—they have more fungiform papillae, structures containing taste buds, on their tongues than non-tasters ( Figure 2 ). Supertasters find vegetables bitter—particularly brassicas, like Brussel sprouts—so they tend to eat fewer vegetables as part of their regular diet than non-tasters. ‘Being a supertaster affects your taste preferences, your diet, and ultimately your health’, claims Bartoshuk. Figure 2 Non-Taster or Supertaster? (A) Top surface of the tongue of a non-taster. (B) Tongue of a supertaster. The small circles are fungiform papillae, each of which contains about six taste buds.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC368160.xml
550667	Assessment of atrial regional and global electromechanical function by tissue velocity echocardiography: a feasibility study on healthy individuals	Background The appropriate evaluation of atrial electrical function is only possible by means of invasive electrophysiology techniques, which are expensive and therefore not suitable for widespread use. Mechanical atrial function is mainly determined from atrial volumes and volume-derived indices that are load-dependent, time-consuming and difficult to reproduce because they are observer-dependent. Aims To assess the feasibility of tissue velocity echocardiography (TVE) to evaluate atrial electromechanical function in young, healthy volunteers. Subjects and methods We studied 37 healthy individuals: 28 men and nine women with a mean age of 29 years (range 20–47). Standard two-dimensional (2-D) and Doppler echocardiograms with superimposed TVE images were performed. Standard echocardiographic images were digitized during three consecutive cardiac cycles in cine-loop format for off-line analysis. Several indices of regional atrial electrical and mechanical function were derived from both 2-D and TVE modalities. Results Some TVE-derived variables indirectly reflected the atrial electrical activation that follows the known activation process as revealed by invasive electrophysiology. Regionally, the atrium shows an upward movement of its walls at the region near the atrio-ventricular ring with a reduction of this movement towards the upper levels of the atrial walls. The atrial mechanical function as assessed by several TVE-derived indices was quite similar in all left atrium (LA) walls. However, all such indices were higher in the right (RA) than the LA. There were no correlations between the 2-D- and TVE-derived variables expressing atrial mechanical function. Values of measurement error and repeatability were good for atrial mechanical function, but only acceptable for atrial electrical function. Conclusion TVE may provide a simple, easy to obtain, reproducible, repeatable and potentially clinically useful tool for quantifying atrial electromechanical function.	Introduction The enlargement of left atrial (LA) diameter is associated with cardiovascular disease and is a risk factor for atrial fibrillation, stroke and death. [ 1 - 6 ] LA function reliably predicts exercise capacity in patients with recent myocardial infarction[ 7 ] or non-ischemic dilated cardiomyopathy[ 8 ] and differs in patients with ischemic and dilated cardiomyopathy.[ 9 ] Moreover, LA volume is an independent prognostic factor in several subsets of patients. [ 10 - 12 ] Although commonly used, LA size assessed by M-mode echocardiography does not correlate well with LA volumes, so several methods to estimate LA volumes have been developed.[ 13 , 14 ] The LA reservoir, conduit and pump functions may be estimated from volume measurements. [ 15 - 19 ] However, the reliability and clinical usefulness of those methods have been poorly studied. Pulsed-wave Doppler interrogation of the blood flow velocity during atrial contraction, the peak mitral inflow A wave, and its velocity time integral have also been used as surrogate markers of atrial function. [ 20 - 22 ] These variables represent the diastolic properties of the LV [ 23 - 25 ] and do not accurately reflect atrial mechanical properties. Tissue velocity echocardiography (TVE) has now been developed as a valuable tool for the evaluation of left and right ventricular systolic and diastolic functions.[ 26 , 27 ] Furthermore, this technique has also been used to assess the regional functions of the left and right atrium.[ 28 , 29 ] Although atrial anatomy was described more than a century ago, a new interest in atrial anatomy and its relations with atrial electromechanical function has only recently emerged.[ 30 , 31 ] Conventionally, atrial electrical function has been evaluated from resting electrocardiography (ECG), and more accurately by invasive electrophysiology techniques. [ 32 - 34 ] The rapid development of these invasive techniques has improved not only diagnostic capabilities,[ 35 , 36 ] but also our understanding of how the electrical impulse spreads through atrial tissues,[ 37 , 38 ] and has led to improvements in the treatment of supra-ventricular tachy-arrhythmias. [ 38 - 40 ] However, the invasive nature and the high costs of these procedures limit their widespread use and repeatability. Therefore, the development of noninvasive, safe, accurate and repeatable methods that might provide similar information is necessary. We aimed here to find simple and repeatable methods to assess both electrical and mechanical regional atrial functions by means of TVE. Methods Population We studied 37 healthy individuals: 28 men and nine women with a mean age of 29 years (range 20–47). The individuals were recruited from among hospital employees, cardiovascular technicians and medical students. None showed symptoms of cardiovascular disorders or were receiving pharmacological cardiovascular agents. All had normal standard two-dimensional (2-D) and Doppler echocardiograms. All subjects were on sinus rhythm and none hade A-V or intra-ventricular conduction defects. The Ethical Committee at the Karolinska University Hospital, Huddinge, approved the study. All individuals received written information and gave informed consent. Echocardiography A standard 2-D and Doppler echocardiogram with superimposed TVE images was performed using a 3.5 MHz transducer with commercially available equipment (System FiVe™, GE Vingmed, Horton, Norway). Standard parasternal short- and long-axis views as well as apical 2-, 3- and 4-chamber views acquired at expiratory apnea with at least 90 frames per second were digitized during three consecutive cardiac cycles in cine-loop format for off-line analysis. Off-line analyses All echocardiographic images were analyzed off-line using software (Echopac™ 6.3.4, GE Vingmed) for the calculation of standard 2-D and Doppler echocardiography as well as for the analysis of TVE variables. Standard 2-D and Doppler echocardiography Measurements of the left ventricular (LV) function comprised septum and posterior wall thickness; LV end-systolic and diastolic dimensions; LV fractional shortening, and LV ejection fraction (LVEF) according to international standards.[ 41 ] Measurements of atrial function comprised left atrial (LA) diameter measured from the parasternal long axis; right atrium (RA) and LA long and short axes; LA and RA maximal volume; LA and RA minimal volume, and RA and LA volumes at the beginning of the P-wave measured from the apical 4- and 2-chamber views. LA and RA ejection fractions were measured according to the formula: (maximal volume-minimal volume)/maximal volume. LA and RA active emptying values were calculated as (volume at P-wave-minimal volume)/volume at P-wave.[ 8 , 9 , 42 ] Tissue velocity echocardiography The RV and LV long axis functions were assessed from apical views. Six basal LV segments were identified as follows: the RV free wall; the LV postero-septal wall, and the LV lateral wall from the apical 4-chamber view; the LV inferior and anterior walls from the apical 2-chamber view, and the LV posterior wall from the apical 3-chamber view. A sample volume was positioned at the base of each ventricular wall excluding the A-V plane during the entire heart cycle to obtain a tissue velocity profile during three consecutive cardiac cycles. Both systolic and diastolic phases of the velocity profile were considered and the following parameters were analyzed (upper part of Fig. 1 ): peak systolic velocity (PSV, in centimeters per second), measured at the peak velocity during the ejection period; peak velocity at early diastole ( E'-wave, in centimeters per second), measured at the peak velocity at early diastole, and peak velocity at late diastole (A'-wave, in centimeters per second), measured at the peak velocity at late diastole. The atrio-ventricular myocardial wall displacement (A'-V' disp., in millimeters) in the long axis was obtained by automated temporal integration of the PSV of the basal segments during the ejection period. Figure 1 Assessment of atrial and ventricular mechanical function. The upper panel shows the systolic and diastolic velocities (a) and the A-V place displacement (b) measured at the basal level of the inter ventricular septum. The lower panel shows the atrial velocity (c), atrial displacement (d), atrial strain rate (e) and atrial strain (f) measured at the inter atrial septum below the mitral ring. The different atrial walls were identified from the same apical views as follows: the right atrial wall (RA), the inter-atrial septum (IAS), and the left atrial lateral wall (LA-Lat) from the apical 4-chamber view; the left atrial inferior wall (LA-Inf) and the left atrial anterior wall (LA-Ant) from the apical 2-chamber view; and the left atrial posterior wall (LA-Post) from the apical 3-chamber view. Each atrial wall was studied at low and mid levels, placing a 2 mm sample volume at low atrial walls excluding the A-V plane during the entire cardiac cycle and at the mid portion of each atrial wall. The regional electromechanical function at each atrial wall was studied by the following time intervals (Figure 2 ). The PA-start interval (P-Aa' start) was defined as the time between the beginning of the P-wave on the monitor's ECG to the start of the A' wave on the TVE-curve profile. The PA-peak interval (P-Aa' peak) was the time between the beginning of the P-wave on the monitor's ECG to the peak of the A' wave on the TVE-curve profile. The A-wave duration (Aa'-dur.) was the time from the beginning to the end of the A'-wave on the TVE-curve profile. The total electromechanical activity (TEMA) was the time between the beginnings of the P-wave on the monitor ECG to the end of the A' wave on the TVE-curve profile. Figure 2 Assessment of some time intervals and Aa' wave velocity at the low level of the inter atrial septum The regional mechanical function of each atrial wall was assessed by the peak velocity during atrial contraction (Aa' peak vel.), the atrial displacement occurring during atrial contraction (Aa' disp.) and the ratio of atrial displacement measured at atrial level to the total LV myocardial displacement measured at ventricular level (Aa' cont.) (lower part of Figure 1 ). In addition, strain rate (Aa' SR) and strain (Aa' S) were assessed in each low atrial wall using a sample volume of 12 mm. Statistical analyses Data are presented as means ± standard deviations (SD). Analysis of variance (ANOVA) with repeated measures was used to test statistical significance of the studied variables at different atrial and ventricular walls. When ANOVA showed statistically significant differences among atrial and ventricular walls, post hoc analysis with Bonferroni's test was performed to assess differences among those walls. Correlation coefficients were calculated to assess the relationship among several markers of atrial mechanical function. The inter- and intra-observer repeatability and measurement errors for variables reflecting the atrial electromechanical function were assessed by the coefficient of variation and by the British Standards Institution method, the value below which the difference between two measurements will lie with a probability of 0.95. P < 0.05 was considered statistically significant. Results All demographic features and measures of standard 2-D and pulsed-wave Doppler echocardiography data are shown in Table 1 . Of interest, no differences were found between measures of RA and LA functions, as assessed by short or long axes, or among volumes and volume-derived indices. TVE-derived variables assessing the RV and LV long-axis systolic and diastolic functions are shown in Table 2 . No significant differences among LV walls were found for any index of systolic and diastolic function. TVE-derived variables obtained from the RV free wall were significantly different from each LV wall. Table 1 Demographic features and resting echocardiographic data. Numbers are means ± SD. Age, years 29 ± 7 Gender (M/F) 28/9 Height, cm 175 ± 8 Weight, kg 76 ± 14 Heart rate, bpm 66 ± 12 P-Q time, ms 166 ± 16 LA diameter, mm/m 2 18.9 ± 1.5 Septal wall thickness, mm 9.6 ± 1.1 Posterior wall thickness, mm 9.5 ± 1.2 LV end diastolic diameter, mm/m 2 26.5 ± 2.3 LV fractional shortening, % 35 ± 6 LV ejection fraction, % 72 ± 8 E-wave, cm 90 ± 17 A-wave, cm 56 ± 12 E/A ratio 1.68 ± 0.45 LA long axis, mm/m 2 26 ± 2 RA long axis, mm/m 2 25 ± 3 LA short axis, mm/m 2 21 ± 3 RA short axis, mm/m 2 22 ± 2 LA maximal volume, ml/m 2 29 ± 5 RA maximal volume, ml/m 2 31 ± 7 LA minimal volume, ml/m 2 15 ± 3 RA minimal volume, ml/m 2 17 ± 4 LA P-wave volume, ml/m 2 18 ± 4 RA P-wave volume, ml/m 2 19 ± 5 LA ejection fraction, % 49 ± 9 RA ejection fraction, % 46 ± 10 LA active emptying, % 17 ± 7 RA active emptying, % 15 ± 9 Table 2 Systolic and diastolic myocardial velocities measured at different right and left ventricular walls Variable Ventricular walls RV Post-sep Lateral Inferior Anterior Posterior PSV, cm/s 10.5 ± 1.3 6.8 ± 0.9 8.5 ± 1.8 7.5 ± 1.1 7.9 ± 1.6 7.6 ± 1.3 E'-wave, cm/s 10.2 ± 2.3 9.9 ± 1.1 12.2 ± 1.5 10.6 ± 2.1 10.1 ± 2.1 12.4 ± 1.9 A'-wave, cm/s 8.4 ± 2.8 5.9 ± 1.3 4.7 ± 1.5 6.3 ± 2.0 4.7 ± 1.7 6.1 ± 2.1 E'/A' ratio 1.2 ± 0.1 1.7 ± 0.1 2.6 ± 0.1 1.7 ± 0.2 2.1 ± 0.2 2.0 ± 0.1 A'-V' disp., mm 21.5 ± 3.5 13.6 ± 1.5 13.7 ± 2.0 15.3 ± 1.6 13.7 ± 1.9 15.4 ± 1.8 Atrial disp., mm 5.8 ± 2.2 4.1 ± 1.3 2.5 ± 0.8 3.8 ± 1.5 2.9 ± 0.9 3.0 ± 1.1 Atrial cont., % 27 ± 8 30 ± 9 19 ± 7 25 ± 9 22 ± 8 20 ± 6 Abbreviations: cont., contribution; disp., displacement; PSV, peak systolic velocity Table 3 shows several time intervals. The PA-start interval (P-Aa' start) was longer at low atrial levels in each atrial wall than at the mid atrial level and shorter at the RA than for all LA walls (Fig. 3 ). Some statistical significant differences among different LA walls were also found. The PA-peak interval (P-Aa' peak) was similar at low and mid atrial levels in almost all atrial walls with exceptions in the inferior and posterior LA walls. This interval was shorter for the IAS, inferior and posterior LA walls than for the lateral and anterior LA walls at mid and low levels (Fig. 4 ). The A-wave duration (Aa'-dur.) was shorter at the low atrial level than at the mid atrial level in each atrial wall, but not in the inferior and posterior LA walls. The total electromechanical activity (TEMA) was similar in all RA and LA atrial walls measured at low and mid levels, and no differences were found between any of the LA walls. Table 3 Time intervals expressed in milliseconds measured at low and mid atrial levels in the myocardial walls of the RA and LA. Variables Level RA IAS LA-Lat LA-Inf LA-Ant LA-Post P* P-Aa' start Low 51 ± 11 59 ± 9 69 ± 11 62 ± 10 70 ± 10 62 ± 11 < 0.001 (ms) Mid 38 ± 9 47 ± 8 57 ± 9 51 ± 11 59 ± 10 52 ± 11 < 0.001 P < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 P-Aa' peak Low 117 ± 22 108 ± 14 119 ± 14 108 ± 12 123 ± 14 107 ± 11 <0.001 (ms) Mid 110 ± 20 104 ± 15 115 ± 15 98 ± 17 123 ± 16 99 ± 17 < 0.01 P 0.06 0.04 0.1 < 0.001 0.9 < 0.001 Aa'-dur. Low 135 ± 16 120 ± 16 106 ± 10 118 ± 13 113 ± 12 112 ± 12 < 0.001 (ms) Mid 145 ± 19 127 ± 16 117 ± 10 115 ± 12 121 ± 16 115 ± 12 < 0.01 P < 0.001 < 0.001 < 0.001 0.1 < 0.001 0.07 TEMA Low 186 ± 17 179 ± 18 175 ± 15 179 ± 15 183 ± 14 174 ± 12 0.07 (ms) Mid 183 ± 21 174 ± 16 173 ± 15 175 ± 15 180 ± 19 167 ± 17 0.2 P 0.3 0.1 0.5 0.1 0.3 0.06 Abbreviations : Aa'-dur., duration of the A wave; IAS, inter-atrial septum; LA-Ant, left atrial anterior wall; LA-Inf, inferior left atrial wall; LA-Lat, left lateral atrial wall; LA-Post, left posterior atrial wall; ms, milliseconds; P , by paired t test; P* , by analysis of variance; P-Aa' start, time from the beginning of the P-wave to the start of the A-wave; P-Aa' peak, time from the beginning of the P-wave to the peak of the A-wave; RA, right atrial wall; TEMA, total electromechanical activity. P-Aa' start Low : RA vs all LA-walls ( P < 0.001), IAS and LA-Inf vs LA-Lat ( P < 0.001) P-Aa' start Mid : RA vs all LA-walls ( P < 0.01), IAS vs LA-Lat and LA-Ant ( P < 0.001); LA-Inf vs LA-Lat and LA-Inf ( P < 0.001); LA-Post vs LA-Ant ( P < 0.01) P-A'a peak Low and Mid : IAS, LA-Inf and LA-Post vs LA-Lat and LA-Ant ( P < 0.001 for all comparisons) Aa' dur. Low : RA, IAS and LA-Inf vs LA-Lat ( P < 0.001) Aa' dur. Mid : RA vs all walls ( P < 0.001), IAS vs LA-Lat, LA-Inf and LA-Post ( P < 001) TEMA Low : RA vs LA-Lat, LA-Inf and LA-Post ( P < 0.001). No differences among all LA-walls. TEMA Mid : RA vs LA-Lat, LA-Inf, LA-Post ( P < 0.001), LA-Ant vs LA-Inf and LA-Post ( P < 0.001) Figure 3 Assessment of the duration of the PA-start interval in all atrial walls. Comparisons were done with ANOVA with repeated measures and the Bonferroni's test. Figure 4 Assessment of the duration of the PA-peak interval in all atrial walls. Comparisons were done with ANOVA with repeated measures and the Bonferroni's test. Table 4 shows several velocities and velocity-derived variables: The peak velocity during atrial contraction (Aa' peak vel.) was higher at low than mid levels in each atrial wall, but no significant differences were found between any LA walls. This variable was higher in RA than in all LA walls. Similar results were found for the atrial displacement occurring during atrial contraction (Aa' disp.) and the ratio of atrial displacement measured at atrial level to the total LV myocardial displacement measured at ventricular level (Aa' cont.). The strain rate (Aa' SR) was higher in the RA than in all LA-walls, and lower in the IAS than in the lateral and posterior LA walls. The strain (Aa' S) was higher in the RA than in all LA-walls, and no differences were found between LA walls. Table 4 Myocardial velocity and velocity-derived variables measured at right and left atrial myocardial walls. Variables Level RA IAS LA-Lat LA-Inf LA-Ant LA-Post P* Aa' peak vel. Low 8.1 ± 2.7 6.3 ± 1.4 6.2 ± 1.7 6.9 ± 1.8 6.3 ± 1.8 6.8 ± 1.8 NS (cm/s) Mid 6.9 ± 2.4 5.2 ± 1.5 5.7 ± 1.4 5.1 ± 1.7 5.7 ± 1.8 5.4 ± 1.7 NS P < 0.001 < 0.001 < 0.001 < 0.001 < 0.01 < 0.001 Aa' disp. Low 6.7 ± 2.3 4.5 ± 0.9 3.7 ± 1.0 4.2 ± 1.0 4.1 ± 1.2 4.0 ± 0.9 NS (mm) Mid 5.8 ± 2.5 3.2 ± 0.9 3.4 ± 0.9 2.9 ± 0.7 3.1 ± 1.3 3.0 ± 0.8 NS P < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 Aa' cont. Low 31 ± 10 33 ± 8 29 ± 10 28 ± 7 31 ± 12 26 ± 6 NS (%) Mid 27 ± 11 24 ± 7 25 ± 9 19 ± 6 23 ± 11 19 ± 5 NS P < 0.001 < 0.001 0.02 < 0.001 < 0.001 < 0.001 Aa' SR (s- 1 ) -4.9 ± 0.8 -2.7 ± 0.7 -3.7 ± 0.9 -3.2 ± 0.8 -3.3 ± 1.0 -3.7 ± 1.0 0.001 Aa' S (%) 29 ± 7 14 ± 5 15 ± 4 16 ± 5 15 ± 6 16 ± 4 NS Abbreviations : Aa' cont., atrial contribution; Aa' disp., atrial displacement; Aa' peak vel., atrial A-wave peak velocity; Aa' S, atrial strain; Aa' SR, atrial strain rate: P , paired t test; P* , analysis of variance; otherwise as in Table 3a. Aa' peak vel Low : RA vs IAS, LA-Lat and LA-Ant ( P < 0.001). No differences were found between any LA walls. Aa' peak vel Mid : RA vs IAS, LA-Inf, LA-Post and LA-Ant (P < 0.001). No differences were found between any LA walls. Aa' disp. Low and Mid : RA vs all walls (P < 0.001). No differences were found between any LA-walls. Aa' cont. Low and Mid : No differences were found between walls Aa' SR : RA vs all LA walls ( P < 0.001), IAS vs LA-Lat and LA-Post ( P < 0.001) Aa' S : RA vs all LA walls ( P < 0.001). No differences were found between any LA walls. Tables 5 and 6 show the correlation coefficients between 2-D-derived and TVE-derived variables of LA and RA global mechanical function. There were no correlations between 2-D- and TVE-derived variables, apart from modest correlations between LA diameter and LA displacement, between RA long axis diameter and RA displacements, and between RA ejection fraction and strain rate. Table 5 Correlation coefficients between 2-D- and TVE-derived variables of left global atrial mechanical function. Individual values of the inter-atrial septum, and the inferior, anterior, lateral and posterior LA walls were averaged. LA-Aa' peak vel. LA-Aa' disp. LA-Aa' cont. LA-Aa' SR LA-Aa' S LA-Diameter 0.33 0.57* 0.43 0.2 0.21 LA-Long axis 0.00 0.17 0.02 -0.21 -0.08 LA-Short axis -0.2 0.02 -0.08 -0.16 -0.07 LA-Area -0.12 0.01 -0.13 -0.12 0.12 LA-P-wave volume -0.15 0.02 -0.08 -0.14 -0.04 LA-Maximal volume 0.07 0.05 -0.1 -0.03 -0.21 LA-Minimal volume 0.2 0.15 0.13 -0.13 -0.21 LA-Ejection fraction -0.27 -0.23 -0.31 -0.03 0.19 LA-Active emptying -0.35 -0.22 -0.31 -0.03 0.19 Abbreviations: LA. Left atria; otherwise as in Table 3. * P < 0.01 Table 6 Correlation coefficients between 2-D- and TVE-derived variables of right global atrial mechanical function. Individual values of the inter-atrial septum, and the inferior, anterior, lateral and posterior LA walls were averaged. RA-Aa' peak vel. RA-Aa' disp. RA-Aa' cont. RA-Aa' SR RA-Aa' S RA-Long axis 0.35 0.58* 0.53 0.26 -0.07 RA-Short axis 0.06 0.30 0.20 -0.00 -0,14 RA-Area 0.10 0.31 0.20 -0.00 -0.14 RA-P-wave volume 0.09 0.31 0.21 0.08 -0.15 RA-Maximal volume -0.06 0.16 0.12 -0.04 -0.26 RA-Minimal volume 0.10 0.24 0.25 0.14 -0.18 RA-Ejection fraction -0.33 -0.24 -0.35 -0.41* -0.09 RA-Active emptying 0.07 0.27 0.12 -0.08 -0.08 Abbreviations: RA, right atrial; otherwise as in Table 3. * P < 0.01 The inter- and intra-observer measurement error and repeatability, as expressed by the British Standards Institution guidelines and coefficients of variation are presented in Table 7 . The PA-start interval and the PA-peak interval , which mainly express atrial electrical function showed the largest inter- and intra-observer measurement errors and variability. However, the A-wave duration and the total electromechanical activity , which express a combination of the atrial electrical and mechanical functions, had better values of measurement error and repeatability. The same was true for all the TVE-derived variables that express regional and global atrial mechanical function. Table 7 Assessment of inter- and intra-observer measurement error and repeatability according to the British Standards Institution guidelines and coefficients of variation Inter-observer Intra-observer Variable BSI CV (%) BSI CV (%) P-Aa' start, ms 37 24 28 19 P-Aa' peak, ms 51 16 47 14 Aa' duration, ms 32 8.8 25 7.5 TEMA, ms 53 9.7 45 8.2 Aa' peak velocity, cm/s 2.14 10.1 1.78 8.7 Aa' displacement, mm 1.81 12.3 1.77 11.8 Aa' SR -0.79 9.4 -0.71 7.8 Aa' S, % 5.4 9.6 4.5 9.1 LA maximal volume, mL 22 18.4 18 14.7 Abbreviations: BSI, British Standards Institution; CV, coefficient of variation; otherwise as in Table 3. Discussion The main new findings of this study of healthy young individuals are as follows. (1) Some TVE-derived variables indirectly reflect the atrial electrical activation that follows the known activation process as revealed by invasive electrophysiology. (2) The regional and global atrial mechanical function is explained by an upward movement of the atrial walls at the region near the A-V ring with a continuous reduction of this movement towards the upper levels of atrial walls. (3) The atrial mechanical function is quite similar in all LA walls; however, all indices of mechanical function were higher in the RA than in the LA. (4) There were no correlations between the 2-D- and TVE-derived variables expressing atrial mechanical function. (5) Values of measurement error and repeatability were good for atrial mechanical function, but only acceptable for electrical function. Atrial electrical activation, as assessed by the PA-start interval , began at the RA and followed through the IAS, to the inferior and posterior LA walls. This is the known normal electrical activation process, as obtained by invasive electrophysiology techniques.[ 32 , 37 ] In the present study, there were no statistical significant differences in the PA-start interval between IAS and the inferior and posterior LA walls, indicating that the activation process could indistinctly occur through any of these walls, as demonstrated by the presence of preferential conduction pathways nearby the IAS, the posterior LA wall and the coronary sinus.[ 33 , 34 , 37 ]. In a recent study, using M-mode color tissue Doppler registrations of the tricuspide and mitral rings, an abnormal time interval from the onset of P wave until the backward motion of the left atrio-ventricular ring was used to indirectly detect abnormal atrial electromechanical coupling in patients with paroxysmal atrial fibrillation.[ 43 ] The relation between atrial anatomy and its mechanical function has been poorly studied. The present study showed that all atrial walls actively moved upwards from the region of the A-V ring at late diastole, with a reduction of this movement towards the upper parts, thus empting the atria and contributing towards the last part of filling of the LV. This longitudinal movement of the atrial walls is probably related to the longitudinal endocardial muscular fibers along the walls of the LA and RA. The more pronounced longitudinal movement in the RA may be explained in part by the larger pectinate muscles in the RA, but also by the lower pressures in the heart's right side. To what extent circumferential contraction of the atrial muscle fibers might contribute to atrial mechanical function is unknown. Anatomically, the large amount of circumferential muscle fibers present in the vestibules of the RA and LA[ 30 , 31 , 44 ] might imply some kind of circumferential or radial contraction of the atria. However, no movement of the posterior LA wall at late diastole can be observed by conventional M-mode echocardiography. Other circumferential fibers, such as Bachman's bundle located at the subepicardium joining the RA and LA, seem to play a critical role for electrical impulse spreading[ 37 ] rather than in circumferential atrial contraction. The assessment of circumferential atrial mechanical function by conventional echocardiography and TVE remains elusive. No correlations were found between 2-D- and TVE-derived variables of atrial mechanical function, as was also found in a previous study[ 29 ]. Although 2-D-derived variables measure volumes and volume-derived indices that might indicate some kind of atrial mechanical force, it was surprising to find no correlations between the variables obtained by the two different techniques. This might indicate that the velocities and the displacements registered from all atrial walls by TVE are less dependent on volume loading conditions than 2-D-derived variables and therefore could be used as reliable measurement of pure atrial mechanical contraction or inotropism. In fact, Donald et al. showed that LA function assessed by TVE was relatively independent of LV function.[ 45 ] It should also be considered that movements of the heart not related to atrial contraction might partly contribute to the velocities and displacements registered from all atrial walls. Therefore, 2-D- and TVE-derived variables might not be used interchangeably to assess atrial mechanical function. Some measures of atrial electrical function, for example the PA-start interval and the PA-peak interval , had only fair measures of repeatability and measurement error. However, most of the TVE-derived variables expressing atrial mechanical function had good values of repeatability and measurement error. Assessing atrial mechanical function by measuring volumes is time-consuming and depends on age, gender, and body surface area[ 14 , 19 ] In addition, atrial volume indices are also dependent on loading conditions[ 46 , 47 ] and are not necessarily more reproducible than TVE-derived variables. Possible clinical implications The identification of an abnormal electrical activation process could be of interest in some patients with atrial fibrillation or other supra-ventricular tachy-arrhythmias, in whom the premature atrial contraction acting as a triggering factor could be aggravated by local delayed conduction (reviewed in[ 48 , 49 ]). Further refinement of the TVE technique are necessary not only to identify the mechanical activation atrial sequence during normal sinus rhythm, but also to identify the origin and the activation sequence of supra-ventricular ectopic beats and in patients with RA, IAS or bi-atrial pacing. Thus, TVE could be an excellent adjunct to invasive electrophysiological techniques in selecting adequate patients and in the evaluation of atrial electromechanical consequences of RA, IAS or bi-atrial pacing. The assessment of pure mechanical atrial function by means of atrial wall movements may give more concrete clues about the recovery process of atrial electromechanical function after conversion for atrial fibrillation and flutter and can give additional pathophysiological insights on the thromboembolic process that occur in some of those patients.[ 50 ] TVE-derived parameters may also give additional pathophysiological information on the process of atrial electromechanical remodeling that occurs in patients with sustained supra-ventricular tachy-arrhythmias.[ 51 ] Several studies have shown the independent prognostic value of atrial function measurements in subsets of patients.[ 6 , 11 , 12 ] TVE-derived variables of atrial mechanical function may have an additional role for facilitating the assessment of atrial function and consequently in the process of risk stratification. Study limitations The results of the present study refer only to a group of young healthy individuals and the values for each of the studied variables are, therefore, only applicable to that population group. As discussed, the measures of atrial electrical function showed only fair values of repeatability and measurement error. There were two reasons: the image acquisition rate (less than 100 frames per second) means an implicit measurement error of 10 ms; it was also difficult to identify the beginning of the P-wave in the ECG from the monitor in the echocardiography machine. Improving temporal resolution by image acquisition at more than 200 frames per second, and improving and adjusting the ECG quality in the present equipment may help solve or decrease this problem. The velocities and displacements registered by atrial walls do not only represent the process of atrial contraction, but also the translational movement of the heart. Until now, no appropriate algorithms that correctly deal with this problem have been found. Presently, it is not possible by means of TVE to simultaneously record the electromechanical function of all atrial walls in one heartbeat. The development of three-dimensional TVE may help resolve this difficulty. Conclusion TVE is a noninvasive bedside tool that requires further refinements to provide reproducible, repeatable and potentially clinically useful data on atrial electromechanical function in health and disease.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC550667.xml
548955	Facts from Text—Is Text Mining Ready to Deliver?	The mining of information from scientific literature using computational tools has tremendous potential for knowledge discovery, but how close are we to realizing this potential?	Biological databases offer access to formalized facts about many aspects of biology—genes and gene products, protein structure, metabolic pathways, diseases, organisms, and so on. These databases are becoming increasingly important to researchers. The information that populates databases is generated by research teams and is usually published in peer-reviewed journals. As part of the publication process, some authors deposit data into a database but, more often, it is extracted from the published literature and deposited into the databases by human curators, a painstaking process. Research literature and scientific databases fulfil different needs. Literature provides ideas and new hypotheses, but is not constrained to provide facts in formats suitable for use in databases. By contrast, databases efficiently provide large quantities of data and information in a standardised schema representing a predefined interpretation of the data. While the acceptance of a paper can enforce the submission of data to a central data repository, such as EMBL ( www.ebi.ac.uk/embl/ ) or ArrayExpress ( www.ebi.ac.uk/arrayexpress/ ), nobody receives credit for the submission of a fact to a database without an associated publication. As long as this practice continues, curation will be necessary to add the (re)formalised facts to biological databases. Given that publications are not about to be replaced with routine deposition of data into databases, is it possible to develop software tools to support the work of the curator? Could we automatically analyse new scientific publications routinely to extract facts, which could then be inserted into scientific databases? Could we tag gene and protein names, as well as other terms in the document, so that they are easier to recognise? How can we use controlled vocabularies and ontologies to identify biological concepts and phenomena? Fortunately, there are many groups that are now seeking to answer these questions, precisely with a view to extracting facts from text. Part of the motivation for this effort in text mining technology is the inexorable rise in the amount of published literature ( Figure 1 ). This massive growth, coupled with the current inefficiencies in transferring facts into other data resources, leads to the unfortunate state that biological databases tend to be incomplete (for example, DNA sequences without known function in genetic databases), and there are inconsistencies between databases and literature. Figure 1 Medline Article Deluge This figure shows the exploding number of articles available from Medline over the past 65 years (data retrieved from the SRS server at the European Bioinformatics Institute; www.ebi.ac.uk/ ). In 2003, about 560,000 articles were added to Medline, and from 2000 to 2003, 2 million articles. (Articles already registered for 2005 are given as well.) In theory, text mining is the perfect solution to transforming factual knowledge from publications into database entries. But computational linguists have not yet developed tools that can analyse more than 30% of English sentences correctly and transform them into a structured formal representation [ 1 , 2 ]. We can analyse part of a sentence, such as a subphrase describing a protein–protein interaction or part of a sentence containing a gene and a protein name, but we always run into Zipf's law whenever we write down the rules for how the extraction is done ( Figure 2 ) [ 3 ]. A small number of patterns describe a reasonable portion of protein–protein interactions, gene names, or mutations, but many of those entities are described by a pattern of words that's only ever used once. Even if we could collect them all—which is impossible—we can't stop new phrases from being used. Figure 2 Zipf's Law Zipf's eponymous law is illustrated by the analysis of 30,000 Medline abstracts (4,952,878 occurrences of words; 144,841 different words). Frequent terms account for a large portion of the text, but a large fraction of terms appear at a low frequency and often only once (69,782 words appear only once). Zipf was a linguistic professor at Harvard University [ 3 ]. Curators—The Gold Standard Hand-curated data is precise, because the curator is trained to inspect literature and databases, select only high-quality data, and reformat the facts according to the schema of the database. In addition, curators select citations from the text as evidence for the identified fact, and those citations are also added to the database. Curators read and interpret the text at the same time, and if they don't understand the meaning of a sentence, they can go back and pick a new strategy to analyse it—they can even call the authors to iron out any ambiguities. Curators can also cope with the high variability of language described by Zipf's law. At present, no computer-based system comes close to matching these capabilities. In particular, it is difficult to convert all the curators' domain knowledge into a structured training set for the purposes of machine learning approaches. Curators fulfil a second important task: they know how to define standards for data consistency, in particular, the most relevant terminology, which has led to the design of standardised ontologies and controlled vocabularies (see Box 1 for an explanation of these and related terms). Examples of these include Gene Ontology (GO; www.geneontology.org/ ), Unified Medical Language System ( www.nlm.nih.gov/research/umls/ ), and MedDRA ( www.meddramsso.com/NewWeb2003/index.htm ) [ 4 ]. These terminological resources help to relate entries in bioinformatics databases to concepts mentioned in scientific publications and to link related information in databases using different schemas. Text miners would love such standards to be used in text, but there is an understandable reluctance to impose and use standards that might limit the expressiveness of natural language. Box 1. Glossary Controlled vocabulary: A set of terms, to standardise input to a database. F-measure: A statistic that is used to score the success of NE recognition by text mining tools. The F-measure is an average parameter based on precision (how many of the entities found by the tool are correct identifications of an entity) and recall (how many of the entities existing in the text did the tool find). Machine learning: The technology and study of algorithms through which machines (computers) can “learn”, or automatically improve their systems through data gathered in the past (experience). Ontology: A set of terms with clear semantics (language), clear motivations for distinction between the terms, and strict rules for how the terms relate to each other. Curation and Text Mining—In Partnership The problem with curation of data is that it is time consuming and costly, and therefore has to focus on the most relevant facts. This compromises the completeness of the curated data, and curation teams are doomed to stay behind the latest publications. So, is it possible for curation and text mining to work together for rapid retrieval and analysis of facts with precise postprocessing and standardisation of the extracted information? There are several software tools that perform well in the identification of standardised terms from the literature. Examples include Textpresso and Whatizit [ 5 , 6 , 7 , 8 ]. Extensive term lists come from the Human Genome Organization ( www.gene.ucl.ac.uk/hugo ; 20,000 gene and protein names), GO (almost 20,000 terms), Uniprot/Swiss-Prot ( www.ebi.uniprot.org/index.shtml ; about 200,000 terms), and other databases. In addition, terms describing diseases, syndromes, and drugs are available from the Unified Medical Language System. Altogether, about 500,000 terms constitute the basis of domain knowledge in life sciences. To gain some perspective of this figure: an average individual handles 2,000 to 20,000 terms in his or her daily language, and Merriam-Webster's Collegiate Dictionary provides definitions for 225,000 terms ( www.merriam-webstercollegiate.com/ ). The identification of all terms by a text mining system still sets challenging demands. All variants of a term have to be taken into account, including syntactical variants and synonyms. In the case of ambiguities, relevant findings have to be distinguished from other findings—a process referred to as disambiguation. Depending on the curation task, it might therefore be advantageous to select only part of the terminological resources and thus restrict the domain of the terminology to the curators' needs ( Figure 3 ). Figure 3 GOAnnotator The illustrated software tool brings together data from text mining and from databases to support curators in the GO annotation of proteins (Couto FM, Lee V, Dimmer E, Camon E, Apweiler R, et al., unpublished data). Here a protein is shown in conjunction with the GO terms that have been gathered from various databases and attributed to the protein through electronic annotation. Both are evaluated against similar GO terms extracted from text documents. The curator looks into the evidence and decides whether any of the GO terms extracted from the documents should be assigned to the protein. Available text mining solutions are concerned with named entity (NE) recognition (entities are, for example, proteins, species, and cell lines), with identification of relationships between NEs (such as protein interactions), and with the classification of text subphrases according to annotation schemata in general (thyroid receptor is a thyroid hormone receptor) [ 9 , 10 , 11 , 12 , 13 , 14 , 15 ]. Whilst the identification of a curation team's terminology in the scientific text under scrutiny is immensely valuable, there is still a long way to go before this becomes routine. Some Immediate Challenges Not all terms used in the literature (NEs) can actually be found in some kind of database (perhaps because of an author error, or an alternative name for an entity adopted by the community). Text mining methods therefore have to detect new terms and map the term to known terminology [ 16 ]. If several mappings are possible, the correct version has to be selected (disambiguation). Over the past several years text mining research teams have presented various approaches that train a software tool to locate representations of gene or protein names (for example, BioCreative, www.pdg.cnb.uam.es/BioLINK/BioCreative.eval.html , and JNLPBA, www.genisis.ch/~natlang/JNLPBA04/ ) [ 17 , 18 ]. These tools are scored with a statistic known as the F-measure, with the best methods scoring about 0.85. At the level of 0.85, curators still tend to be unhappy. However, analyses have shown that this score is in the range of curator–curator variation (unpublished data, measured as part of the project work for [ 19 ]), which suggests that such methods produce useful results. Additional information-extraction methods have been proposed, for example, for the documentation of mutations in specific genes and for the extraction of the subcellular location of proteins [ 11 , 13 ]. An even larger number of tools focus on the identification of appropriate terminology for the annotation of genes (GO terms) [ 7 ]. The evaluation of their usefulness depends on the demands of the user groups. Finally, another way to support curation teams would be to provide information-retrieval methods to guide the team members towards documents containing relevant information. For example, in 2002, the participants in the Knowledge Discovery and Data-Mining Challenge Cup ( www.cs.cornell.edu/projects/kddcup/ ) had to select documents from a given corpus that contained relevant experimental results about Drosophila [ 20 ]. How Can Publishers Contribute? For all automated information-extraction methods, it is obvious that access to literature is crucial. Electronic access has, of course, already had a huge impact, but the structure and organisation of manuscripts could also be improved. For example, semantic tags could be integrated into the text. The markup would not appear on web pages or when the document is printed, but it would help software to deal with semantic aspects of the document. Inserting tags, for example, to mark protein names would allow retrieval software to find documents about proteins even if they look like common English words, such as “you” or “and”. Retrieval engines currently often ignore such terms. In addition, explicit tags would enable text mining methods, for example, when looking for protein–protein interactions, to use the correct semantic interpretation. Text mining systems already available today, such as Whatizit, can integrate semantic tags during submission, which have to be verified by the author. Text mining is ready to deliver tools whereby information is passed back to the authors about the proper use of terminology within their documents. If the use of a term raises conflicts or ambiguities or if the use of a term is wrong, the author is asked to provide feedback. The curation effort is resolved at the earliest possible time-point. Author, publisher, reviewer, and reader profit from consistent information representation, which leads to better dissemination of documents and journals and easily offsets the additional cost in the generation of an article. Publishers and authors have to agree on standards though. Is Text Mining Ready to Deliver? Text mining solutions have found their way into daily work, wherever fast and precise extraction of details from a large volume of text is needed. We have to keep in mind, however, that any text mining tool, just like other bioinformatics resources, will only be suitable for a limited number of tasks. For example, the same text may serve curators from different communities who extract different types of facts, depending on their domain knowledge. Furthermore, different communities have different expectations for accuracy. For example, curators dealing with a small set of proteins prefer tools with high recall, whereas curators dealing with a large number of proteins prefer tools with high precision. Although text mining cannot dissect English sentences completely, and cannot extract the meaning and put the facts into a database, text mining tools are becoming increasingly used and valued. Text mining is ready to deliver handling of complex terminology and nomenclature as a mature service. It is only a matter of time and effort before we are able to extract facts automatically. The consequences are likely to be profound. Not only will we have a more effective approach for the mining of knowledge from the literature, our approach to the publication process itself might change. If a fact is clear enough for automatic extraction, it could be reported in a fact database instead of a publication. As methods improve, authors will see more and more of their text being analysed and formalised in a database. If appropriate quality control is provided, and if authors receive due credit for their deposition of facts into databases, we might well see a shift towards original papers describing new creative ideas and visions rather than just listing facts.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548955.xml
368174	Neandertals Likely Kept Their Genes to Themselves	xx	Scientists searching for clues to our origins have long relied on studying fossils to piece together our evolutionary history. Now, with the tools of molecular genetics, they can reach beyond morphological evidence to retrieve traces of DNA preserved in the remnants of bone. And in these ancient DNA sequences, they're finding bits and pieces of the evolutionary record. Over the course of evolution, changes in DNA sequences accumulate at a predictable rate. These mutations can reveal not only how closely related we are but also when evolutionary lineages diverged. Identifying both a typical range of genetic variation and rate of mutation for a given species or population, for example, can serve as a frame of reference for analyzing DNA sequences from other species or populations. Most molecular anthropologists use DNA found in mitochondria—intracellular structures that convert food into energy—to reconstruct human evolution. Distinct from nuclear DNA, mitochondrial DNA (mtDNA) exists in the cytoplasm of a fertilized egg and is passed on only through the maternal lineage. An ongoing debate about human origins has revolved around the theory that Homo sapiens and Homo Neanderthalensis interbred, since the two species coexisted. Neandertals lived roughly 150,000 to 30,000 years ago, toward the end of the Pleistocene era, and inhabited Europe, parts of Asia, and the Middle East. Modern-day humans arose between 100,000 and 200,000 years ago. Recently, an international multidisciplinary team of scientists led by Svante Pbo of the Max Planck Institute for Evolutionary Anthropology have analyzed the largest sample of Neandertal and early human remains to date and conclude that Neandertals could not have made a significant genetic contribution to early modern humans. Part of the challenge of resolving the human–Neandertal interbreeding issue stems from the fact that so many fossil samples—of both early humans and more archaic humans—are contaminated with the DNA of the contemporary humans who have handled them. So even if a Neandertal sample contained a “real” (or endogenous) DNA sequence resembling early humans—which would indicate intimacy between the two groups—it might be considered contaminated. When Pääbo and colleagues looked for modern DNA, they found it in every sample they examined: in the Neandertal and early human fossils—and even in cave bear teeth. To circumvent this problem, they looked only for Neandertal mtDNA as evidence of interbreeding. Since it is easy to distinguish modern human mtDNA sequences from the four Neandertal mtDNA samples that have been sequenced so far, the researchers decided to determine whether Neandertal-like mtDNA could be found in other Neandertal fossils as well as in early human remains. Neandertal skull from La Chapelle aux Saints As these fossils are precious commodities, Pbo's group applied a technique developed in their lab that uses amino acid content as a measure of extractable endogenous DNA and requires removing just 10 mg of bone from a specimen rather than much larger pieces of bone. Of 24 Neandertal and 40 early modern human fossils analyzed, they found four Neandertal and five early human specimens that passed the amino acid test. These fossils included samples classified as “transitional” between the two groups and represented a wide distribution across Europe, where the two groups would likely have encountered one another. When they analyzed these samples for Neandertal mtDNA, they found mtDNA sequences that are absent in contemporary human mtDNA genes but quite similar to those found in the four previously sequenced Neandertals. They found no Neandertal-like mtDNA in the early human samples. While the authors explain that it's impossible to definitively conclude that no genetic flow occurred between early humans and Neandertals given the limited number of early human fossils available, they point out that even fossil samples considered as anatomically transitional between modern humans and Neandertals failed to show evidence of mtDNA exchange. Thus, Pääbo and colleagues conclude, while it's possible that Neandertals made a small contribution to the genetic makeup of contemporary humans, the evidence cannot support the possibility of a large contribution.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC368174.xml
528723	Parallel Chemical Genetic and Genome-Wide RNAi Screens Identify Cytokinesis Inhibitors and Targets	Cytokinesis involves temporally and spatially coordinated action of the cell cycle and cytoskeletal and membrane systems to achieve separation of daughter cells. To dissect cytokinesis mechanisms it would be useful to have a complete catalog of the proteins involved, and small molecule tools for specifically inhibiting them with tight temporal control. Finding active small molecules by cell-based screening entails the difficult step of identifying their targets. We performed parallel chemical genetic and genome-wide RNA interference screens in Drosophila cells, identifying 50 small molecule inhibitors of cytokinesis and 214 genes important for cytokinesis, including a new protein in the Aurora B pathway (Borr). By comparing small molecule and RNAi phenotypes, we identified a small molecule that inhibits the Aurora B kinase pathway. Our protein list provides a starting point for systematic dissection of cytokinesis, a direction that will be greatly facilitated by also having diverse small molecule inhibitors, which we have identified. Dissection of the Aurora B pathway, where we found a new gene and a specific small molecule inhibitor, should benefit particularly. Our study shows that parallel RNA interference and small molecule screening is a generally useful approach to identifying active small molecules and their target pathways.	Introduction Small molecule inhibitors are useful tools for studying dynamic biological processes. Compared to mutations and RNA interference (RNAi), cell-permeable small molecules allow inhibition of protein function with precise temporal control, and may also spur development of new therapeutics. One approach to finding useful small molecules is phenotypic screening, in which cells are treated with small molecules from a library and scored for inhibition of the process of interest. The rate-limiting step in this approach is identifying the cellular targets of active small molecules. Traditionally, the targets of small molecules have been identified by methods based on physical affinity, for example, affinity chromatography ( Harding et al. 1989 ). These require chemical modification of the small molecule and suffer the limitation that irrelevant proteins will bind in addition to the authentic target. A complementary method is to use information on the biological activity of the small molecule to identify the cellular pathway it perturbs. In some cases an educated guess can be successful ( Mayer et al. 1999 ), but to be generally useful, the biological activity of a small molecule would need to be systematically compared to the effect of perturbing different cellular pathways. Currently, the most general method for systematically perturbing pathways and collecting phenotypic information is RNAi, which can be used to inhibit protein function on a genome-wide basis. Here, we develop a parallel screening strategy for finding small molecules that inhibit the biological process of cytokinesis, the genes required for this process, and, by cross-comparison of phenotypes, information on the protein targets of the small molecules. Cytokinesis is the final step of cell division, when the daughter cells are physically separated by constriction of a cleavage furrow. Complex spatiotemporal coordination of several cell systems, including the microtubule and actin cytoskeletons, the cell cycle engine, and vesicle trafficking, is required for furrow positioning, assembly, ingression, and eventual cell separation. Some important components of the cleavage furrow, for example, actin ( Schroeder 1973 ), Myosin ( Mabuchi and Okuno 1977 ), and Anillin ( Oegema et al. 2000 ), have been identified, as well as signaling systems that position and regulate the furrow, such as Aurora B ( Carmena and Earnshaw 2003 ) and Polo ( Carmena et al. 1998 ) kinases and Rho family GTPases ( Prokopenko et al. 2000 ). Since many of these proteins also play roles in additional cellular processes, analysis of their function in cytokinesis by genetic methods can be difficult, and small molecule tools would be useful. To date, only three cytokinesis proteins, actin, Myosin II, and Aurora B kinase, have been targeted with small molecules, but even this limited set has been very useful. For example, the actin inhibitor cytochalasin was used to discover the central role of actin in cytokinesis (reviewed in Peterson and Mitchison 2002 ), and the Myosin II inhibitor, blebbistatin, provided insight into the coordination of different processes in cytokinesis ( Straight et al. 2003 ). Small molecule inhibitors of Aurora B kinase have recently been reported, but their effect on cytokinesis has yet to be investigated in detail ( Ditchfield et al. 2003 ; Hauf et al. 2003 ). Although several key cytokinesis proteins are known, we lack a complete list of proteins required for cytokinesis in any organism. In a genome-wide study, 98 proteins were reported to localize to the bud neck, the site of cytokinesis in Saccharomyces cerevisiae ( Huh et al. 2003 ), but their functional roles have not yet been systematically investigated. A proteomic screen that identified many components of mammalian midbodies, organelle-like remnants of the cleavage furrow, was reported recently ( Skop et al. 2004 ) and several small-scale RNAi studies have been conducted in Drosophila ( Somma et al. 2002 ; Goshima and Vale 2003 ; Kiger et al. 2003 ; Rogers et al. 2003 ). These screens identified genes required for cytokinesis, but did not assay an entire genome. Genome-wide RNAi screens have been carried out in Caenorhabditis elegans and in Drosophila cells, but they did not focus on genes required for cytokinesis ( Kamath et al. 2003 ; Boutros et al. 2004 ). Results/Discussion Parallel Screening Protocols To identify all genes required for cytokinesis, and small molecules that target their products, we developed an assay for both comprehensive functional genomic and large-scale chemical genetic screens in cultured Drosophila cells. We chose this system because of both the availability of genome-wide RNAi resources ( Boutros et al. 2004 ) and the ease of RNAi in Drosophila cells. Drosophila cells can take up long pieces of double-stranded RNA (dsRNA) from the culture medium and process them into small interfering RNAs without triggering an interferon response, in contrast to mammalian cells ( Clemens et al. 2000 ; Elbashir et al. 2001 ). Furthermore, use of a single targeting dsRNA is efficient for the knockdown of a specific gene in each experiment. Therefore, we were able to screen a library of existing dsRNAs with an average length of 408 bp ( Hild et al. 2003 ) to functionally test nearly all Drosophila genes for roles in cytokinesis. Cells that undergo mitosis normally, but fail cytokinesis, acquire two nuclei. This phenotype is a specific and irreversible consequence of cytokinesis failure, and can be scored by automated fluorescence microscopy. Drosophila Kc 167 cells were cultured together with either gene-specific dsRNAs or discrete small molecules in optical-bottom 384-well plates. In total, we screened 19,470 dsRNAs covering more than 90% of the annotated genome in triplicate at the Drosophila RNAi Screening Center ( http://www.flyrnai.org ) and over 51,000 small molecules at the Institute of Chemistry and Cell Biology ( http://iccb.med.harvard.edu ). The cells were incubated for 4 d in the presence of dsRNAs to allow for depletion and turnover of targeted gene products, or 2 d for small molecules to permit each cell to complete at least one cell cycle. After fixation, cells were stained with amine-reactive tetramethylrhodamine-NHS ester to visualize total cytoplasm and Hoechst dye to visualize DNA. In the RNAi screen, microtubules were visualized by immunofluorescence. Cells were imaged by automated fluorescence microscopy, and assay wells containing a high frequency of binucleate cells were identified by a combination of automated image analysis and visual inspection. Small Molecule Screening Results From approximately 51,000 small molecules that included a mixture of commercial “drug-like” molecules, natural product extracts, and natural-product-like libraries synthesized at the Institute of Chemistry and Cell Biology ( http://iccb.med.harvard.edu , we identified 50 small molecule inhibitors of cytokinesis, and selected 25 of the most potent and readily available for further analysis ( Table S1 ). This structurally diverse group, which we named binucleines 1–25, included 12 small molecules from commercial libraries, ten known bioactives, and three natural product extracts. We screened at a nominal concentration of 12.5 μg/ml (approximately 25 μM) and retested the effect of our 25 small molecule inhibitors on Drosophila tissue culture cells at three different concentrations: 100 μM, 30 μM, and 10 μM ( Table S1 ). To determine cross-reactivity with other species, we also assayed cytokinesis inhibition in HeLa (64%, 16/25 active) and BSC-1 (52%, 13/25 active) tissue culture cells as well as growth inhibition in drug-sensitive S. cerevisiae (48%, 12/25 active) (see Table S1 ). Since most compounds currently known to inhibit cytokinesis are natural product actin binders, we tested if the small molecule inhibitors affected actin polymerization. Binucleines 4, 6, 24, and 25 inhibited pyrene–actin polymerization in a pure protein assay (data not shown). Binucleines 24 and 25 are the actin binders cytochalasin D and jasplakinolide, which were included in our small molecule collection as control compounds. Binucleine 4 is a natural product extract from Ircinia ramosa, which contains swinholide A, a known actin binder, as its active ingredient (F. C. Schroeder and J. Clardy, personal communication). To learn more about the cellular targets of the remaining compounds, we proceeded with our plan to systematically compare small molecule and RNAi phenotypes. Genome-Wide RNAi Screening Results We identified dsRNAs corresponding to 214 genes with phenotypes important for cytokinesis ( Table S2 ). Only dsRNAs that resulted in a binucleate phenotype in at least two of the three replicate screens were summarized in our final results ( Table S3 ). These genes resulted in either a strong, medium, or weak increase in frequency of binucleate cells ( Figure 1 ) and represented a diverse range of predicted cellular functions ( Figure 2 ; Table S4 ), reflecting the complexity of cytokinesis. Of the RNAi phenotypes, 20% identified genes previously directly implicated ( Table S5 and references therein) or involved in processes associated with cytokinesis. Eleven of the strong phenotypes identified such genes, including two copies of actin (Act57B and Act5C), Myosin heavy chain (zipper), Anillin (scraps), a formin (diaphanous), Rho GTPase (Rho1) and its known guanine nucleotide exchange factor (pebble) and GTPase-activating protein (RacGAP50C), a kinesin (pavarotti), Citron kinase (CG10522), Aurora B kinase (ial), and a PRC1 homolog (fascetto) . We discovered one new gene essential for cytokinesis ( CG4454, see discussion below), increasing the number of specific essential proteins confirmed by RNAi to thirteen (the twelve genes listed above and INCENP; see Table S5 ). Although required for cytokinesis ( Adams et al. 2001 ) and successfully resynthesized for later experiments, INCENP was not identified in our screen because of failure in INCENP dsRNA synthesis. Figure 1 Distribution of Active Small Molecules and Genes Targeted by RNAi Identified by Penetrance of Binucleate Phenotype and by Phenotypic Classes (A and B) Penetrance of binucleate phenotype for small molecules (A) and RNAi hits (B). (A) For the small molecules, 24% (6/25) were strong (s), 44% (11/25) medium (m), and 32% (8/25) weak (w). (B) For the RNAi hits 6% (13/2114) were strong (s), 20% (43/214) medium (m), and 74% (158/214) weak (w). In a weakly penetrant phenotype, the binucleate level was increased by more than 1.25-fold relative to the two neighboring wells in at least two experiments. In a medium penetrance phenotype, the binucleate level was above 4%, or four times as high as the neighboring wells. In a strongly penetrant phenotype, the binucleate level was above 15%. The average binucleate level in controls was approximately 1%. (C and D) Phenotypic classes for small molecules (C) and genes targeted by dsRNAs (D). (C) For the small molecules, 40% (10/25) were binucleate (b; Figure 2 A), 8% (2/25) binucleate with large, diffuse DNA (d; Figure 2 B), 28% (7/25) binucleate with low cell count (lc; Figure 2 C), and 24% (6/25) binucleate with microtubule extensions (MT; Figure 2 D). (D) For the RNAi hits, 51% (109/214) were binucleate (b; Figure 2 A), 2% (5/214) binucleate with large, diffuse DNA (d; Figure 2 B), 29% (62/214) binucleate with low cell count (lc; Figure 2 C), and 12% (25/214) binucleate with microtubule extensions (MT; Figure 2 D). In addition, 5% (10/214) were binucleate with low cell count and microtubule extensions, and 1% (3/214) were binucleate with low cell count and large, diffuse DNA. Figure 2 Predicted Functional Annotations of 214 Genes Associated with RNAi Binucleate Phenotypes Functional groups were assigned using Gene Ontology information presented in FlyBase or the literature (see Table S4 ). Genes involved in processes associated with cytokinesis are shown in shades of yellow, nucleic acid and protein synthesis and degradation in shades of red, and uncharacterized genes in shades of blue. The uncharacterized genes encode protein sequences that predict recognizable domains (“putative domain”), no recognizable domains (“no recognized domain”), or new gene predictions from the reannotation of the Drosophila genome used as the basis of the dsRNA library (“new annotation”). Cytokinesis is a complex, multistep process, unlikely to be regulated and executed by only thirteen proteins, suggesting that many other proteins with less stringent requirements are also involved. Our screen identified 201 genes with a loss-of-function phenotype of a medium or weakly penetrant cytokinesis failure. Of these genes, 54 were predicted to encode proteins with unknown function, including 25 that were targeted with dsRNA on the basis of new gene model predictions ( Hild et al. 2003 ). The remaining genes had a variety of predicted functions, including cell cycle regulation and vesicle transport. Cytokinesis is known to require insertion of new plasma membrane ( Finger and White 2002 ), consistent with our identification of genes involved in vesicle transport (12 genes). We were surprised, however, that this group included most of the components of the coatomer complex COPI (5/7 COPI subunits). The COPI complex is thought to be involved in retrograde transport from Golgi apparatus to endoplasmic reticulum, and its role in cytokinesis remains to be elucidated. An unexpected functional group, identified mostly with weak phenotypes, included genes involved in nucleic acid and protein synthesis and degradation, including a large number of ribosomal proteins. We subjected nine of these genes to further analysis. In seven of nine cases, filamentous actin staining was very weak, while other proteins such as tubulin and Myosin were of normal abundance, suggesting that the phenotype may result from low-level synthesis of actin or other cortical components (data not shown). Since a library of a single long dsRNA per gene was used in this screen, it is conceivable that some phenotypes are due to off-target effects. The approximately 400-nt dsRNAs are processed into smaller small interfering RNAs, and if appropriately processed, could cross-hybridize partially or completely with identical sequences in mRNA corresponding to other genes ( Bartel 2004 ; Tijsterman and Plasterk 2004 ). Of 214 dsRNAs identified in our screen, 57 had a potential 21-nt overlap with other genes (see Table S3 ). In the majority of these cases (39/57), full-length dsRNA corresponding to the potential cross-match gene did not itself score. Some related genes, for example, the five copies of actin identified in our screen, show high homology and are therefore expected to contain overlapping dsRNA sequences. Comparison of RNAi Screen to Other Screens Since one of our goals was to create an inventory of all genes required for cytokinesis, it is important to evaluate the success rate of the genome-wide RNAi screen with respect to other published screens and to the cytokinesis literature in general. Four small-scale screens have examined the role of specific genes in cytokinesis in Drosophila cells ( Somma et al. 2002 ; Goshima and Vale 2003 ; Kiger et al. 2003 ; Rogers et al. 2003 ). Results from our genome-wide screen correlate well with data from the four smaller screens and other experiments, indicating that the field is converging on a consensus of genes absolutely required for cytokinesis (see Table S5 ). Ten genes, reported elsewhere with RNAi binucleate phenotypes in Drosophila cells, did not score in our screen ( INCENP [ Adams et al. 2001 ]; syx1A [ Somma et al. 2002 ]; profilin , aip1 [CG10724], and capt [ Rogers et al. 2003 ]; and kst, Toll, Toll-4, bazooka, and kekkon [ Kiger et al. 2003 ]). The dsRNA targeting these genes, apart from INCENP, passed quality control, suggesting alternative explanations for the differences between various RNAi experiments. Three of the smaller screens were carried out in Drosophila S2 cells ( Somma et al. 2002 ; Goshima and Vale 2003 ; Rogers et al. 2003 ), which may differentially express or require certain proteins. The timing of RNAi experiments may also contribute to differences that were observed. We exposed cells to dsRNAs for 4 d, balancing sufficient depletion with potentially detrimental effects of prolonged culture and exposure of cells to dsRNA. With an average cell cycle of 24 h, 4 d may be too short to completely deplete very stable proteins. For example, depletion of the Myosin II regulatory light chain (spaghetti squash) resulted in a weak phenotype, whereas depletion of its complex partner encoded by zipper, the Myosin II heavy chain, resulted in a much higher frequency of binucleates. This illustrates the importance of identifying medium and weak phenotypes. Genes with weaker binucleate phenotypes may also be significant because depletion of cytokinesis proteins with multiple functions during the cell cycle can cause arrest prior to cytokinesis, diminishing the likelihood of detecting the phenotype in unsynchronized cells. Overlap between our screen and a recently published proteomic analysis of the midbody ( Skop et al. 2004 ) highlights the importance of identifying genes with weaker binucleate phenotypes. For example, the Arp2/3 complex was not thought to play a role in cytokinesis in metazoans, but components of this complex were identified in both approaches. Eventually, a combination of different methods will result in a definitive list of all proteins involved in cytokinesis. Systematic Comparisons between RNAi and Small Molecule Phenotypes Following our strategy to systematically compare chemical genetic and functional genomic data, we classified the data from both screens into four phenotypic groups (summarized in Figure 1 C and 1 D): (1) binucleate cell phenotype only ( Figure 3 A), or a combination of binucleate cells with an additional phenotype of (2) large, diffuse DNA ( Figure 3 B), (3) low total cell count ( Figure 3 C), or (4) microtubule extensions ( Figure 3 D). The initial classification into four phenotypic groups was based on the raw screening data, where our parameters were whole cell, DNA, or tubulin staining. While these phenotypic classes are useful for global analysis and preliminary characterization, there were too many genes in each group to allow meaningful comparisons between small molecule and RNAi phenotypes. Therefore, we selected 40 genes and 25 small molecules for more detailed analysis. To ascertain specific defects in cytokinesis, we determined by immunolocalization the behavior of 15 proteins involved in cytokinesis. Our bank of reagents included antibodies to proteins that are normally found in the cleavage furrow such as actin (phalloidin), Anillin, Myosin II, and the septin protein Peanut; proteins involved in the regulation of cytokinesis such as Aurora B ( Giet and Glover 2001 ), RhoA, Pebble ( Prokopenko et al. 1999 ), and Polo kinase ( Tavares et al. 1996 ); proteins involved in other aspects of cytokinesis like Diaphanous, Lava-lamp ( Sisson et al. 2000 ), and Pavarotti; and proteins that report on the stage of cytokinesis or the state of the cell cycle such as Lamin ( Risau et al. 1981 ), phospho-Histone H3, and tubulin. As a specific example of this detailed analysis, the phenotypes for Aurora B kinase and CG4454 are discussed below. Figure 3 Phenotypic Classes The phenotypic classes are (A) binucleate ( CG10522 RNAi) and binucleate with (B) large, diffuse DNA ( aurora B RNAi), (C) low cell count ( RpS18 RNAi), or (D) microtubule extensions ( Act5C RNAi). In (A), (B), and (C), the cytoplasm (tetramethylrhodamine stain) of Kc 167 cells is shown in red and DNA in green. In (D), tubulin is shown in red and DNA in green. See Table S2 for full classification. Small Molecules Can Result in Additional Phenotypes We identified phenotypes common to both datasets, but the detailed phenotypic analyses did not match exactly, with more phenotypic subclasses distinguished with the small molecules. Two considerations may account for the existence of additional phenotypic categories for small molecules. One is timing. During our detailed secondary analysis, we added small molecules to cells for variable amounts of time (3 h to 48 h). When cells were exposed to a drug for a short time, we were able to analyze localization of furrow components immediately after cytokinesis failure. These phenotypes became less apparent upon longer exposure because the long delay gave cells the opportunity to disassemble residual furrow structures. When cells were exposed to dsRNAs for days, long, variable delays between cytokinesis failure and fixation may have obscured interesting phenotypes, which could be revealed by subsequent real-time imaging experiments ( Goshima and Vale 2003 ). The other consideration is potential gain-of-function effects of small molecules. For example, a natural product extract from Cowania mexicana containing a cucurbitacin (M. Fujita and J. Clardy, personal communication) caused clusters of filamentous actin to accumulate in interphase cells, in addition to completely blocking cytokinesis. Since no dsRNA caused this phenotype, we suspect it is a gain-of-function effect of the small molecule, whose mechanism we will pursue. Two Sub-Phenotypes Correlate in Both Small Molecule and RNAi Datasets Systematic comparison between the phenotypic categories based on detailed immunofluorescence analysis of both screens did, however, allow us to connect small molecules to two specific pathways involved in cytokinesis, namely actin cortex integrity and the Aurora B pathway. Both dsRNAs and small molecules that weakened the actin cortex caused microtubule-rich extensions to protrude from interphase cells as well as failure of cytokinesis ( Figure 4 ). These included dsRNAs targeted against several actin genes (see Table S2 ) and three natural product small molecules known to target actin that were present in our small molecule collection (cytochalasin D, jasplakinolide, and swinholide A [from Ircinia ramosa extract]). This sub-phenotype represents a portion of the genes identified as “binucleate with microtubule extensions” shown in Figure 3 D. Figure 4 Kc 167 Cells Exposed to dsRNA Targeting Act5C or to Cytochalasin D The cells were exposed to dsRNA targeting Act5C for 4 d (A) or to cytochalasin D at 5 μM for 48 h (B). Tubulin is shown in red, DNA in green. A second phenotypic class exhibited a high incidence of both mitosis and cytokinesis defects, a sub-phenotype of the category “binucleate with diffuse DNA” (see Figure 3 B). Mitosis was abnormal, with malformed spindles and misaligned chromosomes, resulting in large, diffuse arrangements of DNA in binucleate cells ( Figure 5 ). Individual depletion of any of three proteins encoded by aurora B , INCENP, and CG4454 , or addition of one small molecule N′ -[1-(3-chloro-4-fluorophenyl)-4-cyano-1H-pyrazol-5-yl]- N , N -dimethyliminoformamide (binucleine 2; Figure 5 ), caused this phenotype. Figure 5 Kc 167 Cells Untreated or Exposed to aurora B dsRNA, borr (CG4454) dsRNA, or Binucleine 2 TMR-stained cells were untreated, or treated with dsRNA for 4 d or binucleine 2 (50 μM) for 2 d. TMR is shown in red, DNA in green. The chemical structure of binucleine 2 is also shown. CG4454 RNAi Phenotype and Localization Matches Chromosomal Passenger Proteins Aurora B, INCENP ( Adams et al. 2001 ), and Survivin ( Wheatley et al. 2001 ) form the chromosomal passenger complex, which also includes CSC-1 in C. elegans ( Romano et al. 2003 ) and Borealin/Dasra B in humans ( Gassmann et al. 2004 ; Sampath et al. 2004 ). Aurora B kinase plays a number of roles during mitosis ( Carmena and Earnshaw 2003 ), including phosphorylating Histone H3 on Ser-10 ( Giet and Glover 2001 ) and detecting errors in chromosome attachment in mitosis ( Lampson et al. 2004 ), and performs an essential, but poorly understood, function in cytokinesis. Chromosomal passenger proteins localize to the inner centromere during mitosis and move to the interzonal microtubules, the cleavage furrow, and eventually the midbody during cytokinesis. Because the sequences that targeted CG4454 and aurora B both had 21-bp overlaps with other genes in the dsRNA collection we screened (see Table S3 ), we remade dsRNA targeting different areas of these two genes and observed no change in phenotype. Since RNAi depletion of the new gene we discovered in our screen, CG4454, resulted in the same phenotype as depletion of aurora B and INCENP, we hypothesized that it could be a new member of the chromosomal passenger complex. We constructed green fluorescent protein (GFP) fusion proteins to both C- and N-termini of CG4454. CG4454-GFP exhibited the signature localization of a passenger protein and co-localized with Aurora B throughout mitosis and cytokinesis ( Figure 6 ), suggesting that it might be complexed to Aurora B. RNAi depletion of CG4454 or aurora B resulted in an absence of phosphorylated Histone H3 on mitotic chromosomes ( Figure 7 , bottom row), further supporting the participation of CG4454 in the chromosomal passenger complex. Although CG4454 amino acid sequence reveals a remote similarity with Borealin/Dasra B ( Gassmann et al. 2004 ), it is unclear at this point whether CG4454 is its Drosophila homolog. Unlike CG4454, RNAi depletion of Borealin does not significantly reduce Histone H3 phosphorylation ( Gassmann et al. 2004 ). It might not be possible to confirm whether CG4454 and Borealin are related until structural information becomes available. However, to prevent further confusion in naming conventions, we have decided to tentatively name CG4454 Borealin-related (Borr). Figure 6 Kc 167 Cells Transfected with Borr-GFP In the top row, cells in metaphase, anaphase, and cytokinesis are shown. Borr-GFP is shown in green, tubulin in red, and DNA in blue. The bottom row shows cells in metaphase and cytokinesis. Borr-GFP is shown in green, Aurora B in red, and DNA in blue. Figure 7 Kc 167 Cells Untreated or Exposed to aurora B dsRNA, borr (CG4454) dsRNA, or Binucleine 2 INCENP-stained cells in the top row were untreated or treated with aurora B dsRNA for 5 d, borr (CG4454) dsRNA for 3 d, or binucleine 2 (20 μM) for 4 h. Phospho-Histone H3 stained cells in the bottom row were untreated or treated with dsRNA for 4 d or binucleine 2 (20 μM) for 4 h. White arrows indicate absence of phospho-Histone H3 staining in the failed mitotic figures. Detailed Comparison of Binucleine 2 and Aurora B Complex Phenotypes We compared the phenotypes caused by RNAi depletion of aurora B and borr to treatment of cells with binucleine 2 using immunofluorescence. The phenotypes were very similar, as judged by perturbation of localization or expression of 14 of the 15 markers used in our detailed analysis ( Figure S1 ), suggesting that the two genes and binucleine 2 perturb a similar step in cytokinesis. The only difference we observed was localization of the chromosomal passenger protein INCENP ( Figure 7 , top row). No INCENP staining at any cell site was detected in borr -depleted cells, suggesting that Borr is required for INCENP localization. This phenotype was also observed in Borealin-depleted cells ( Gassmann et al. 2004 ). In contrast, we observed INCENP accumulations in binucleine 2–treated and aurora B –depleted cells. INCENP localizes to the chromosome arms during prometaphase in aurora B –depleted cells, which is consistent with reported observations ( Adams et al. 2001 ). In cells exposed to binucleine 2, INCENP aggregated ( Figure 7 , top row), but did not appear to co-localize with Aurora B or DNA. Given its effect on INCENP localization, binucleine 2 might be a useful tool to study the localization and movement of the Aurora B complex during mitosis and cytokinesis, since the factors that regulate these processes remain obscure. In total, binucleine 2 shares phenotypes with aurora B RNAi and affects localization of INCENP, a member of the Aurora B complex, suggesting that binucleine 2 targets the Aurora pathway. Small molecules can target and inhibit protein activity directly, whereas dsRNAs target destruction of mRNA. This difference in mechanism between small molecule inhibition and RNAi could account for the variation in INCENP localization we observed. The specific activity of binucleine 2, however, is very highly related to its structure. We assessed the effect of several similar compounds and found that none were more active than binucleine 2, while most had very little activity ( Figure S2 ). To test whether binucleine 2 inhibits Aurora B kinase function, we monitored Histone H3 phosphorylation on Ser-10 in mitotic cells ( Giet and Glover 2001 ). When cells were exposed to binucleine 2, phospho-Histone H3 was absent on chromosomes in mitotic cells ( Figure 7 , bottom row). To get a quantitative measure of both the concentration of binucleine 2 required and the speed of its action, we assayed about 10,000 cells per time point and concentration for phospho-Histone H3 staining by immunofluorescence ( Figure 8 ). We were unable to detect phospho-Histone H3 in cells treated with 25 μM or 100 μM binucleine 2 for only 30 min, while the percentage of cells exhibiting phospho-Histone H3 staining decreased over time in cells treated with binucleine 2 at 1 μM and 5 μM ( Figure 8 ). While binucleine 2 inhibits Aurora B–dependent phosphorylation, it is not a general kinase inhibitor. Binucleine 2 did not inhibit cyclin-dependent-kinase-dependent entry into mitosis and had no effect on bulk phosphorylation activity in a Drosophila cell extract (data not shown). Altogether, phenotypic similarities between loss-of-function for Aurora B and binucleine 2 strongly suggest that binucleine 2 targets a protein involved in the Aurora B pathway. Several small molecule inhibitors of Aurora kinases have been reported, although their chemical scaffolds are different from binucleine 2. These small molecules are not active in fly cells, while binucleine 2 is inactive in mammalian systems (data not shown). Aurora kinase levels are elevated in some tumors, making these proteins a potential target for cancer therapy. Interestingly, lower binucleine 2 concentration or shorter Aurora B RNAi treatment favors binucleate formation, while higher drug concentration and longer incubation for RNAi results in a relative increase in large cells with diffuse DNA. Thus, the cytokinesis function of the Aurora pathway may be more sensitive to inhibition than its mitosis function. This observation may be important for understanding the response of cancer cells to Aurora inhibitors now entering clinical trials ( Harrington et al. 2004 ). Figure 8 Time- and Concentration-Dependence of Binucleine 2 Kc 167 cells were treated with 1 μM, 5 μM, 25 μM, or 100 μM binucleine 2. Phospho-Histone H3 staining was assessed at different time points. Binucleine 2 at 100 nM and 300 nM was also tested and showed no effect (data not shown). Conclusion In summary, our parallel screening approach succeeded in identifying new proteins involved in cytokinesis, and new small molecules that inhibit it. We identified 214 proteins important for cytokinesis, including 25 previously uncharacterized predicted proteins. Depletion of one new gene, borr, had a profound effect on cytokinesis. Borr exhibits the signature localization of a chromosomal passenger protein and co-localizes with Aurora B kinase throughout the cell cycle. We also uncovered a potential role of the COPI coatomer complex in cytokinesis. By comparative phenotypic analysis we were able to show that one class of small molecules targets actin cortex integrity, and another the Aurora B pathway. A third class of small molecules, whose phenotype has no RNAi counterpart, presumably causes gain-of-function effects. Traditional methods like affinity chromatography and enzyme inhibition assays will be required to describe the precise biochemical mechanisms of these new cytokinesis inhibitors, but the information already gained from comparative screening will focus this work and allow rapid confirmation or invalidation of candidate biochemical targets. The problem of target identification has been one of the main barriers to more widespread use of phenotype-based screening in drug discovery. As functional genomic data and systematic RNAi resources become widely available for human cells, parallel screening approaches like the one we describe could be used to discover leads for therapeutic drugs as well as research reagents. Materials and Methods Small molecule screen 20,000 Drosophila Kc 167 cells in 40 μl of medium (Schneider's Drosophila Medium [GIBCO, San Diego, California, United States] supplemented with 10% heat-inactivated fetal bovine serum [HyClone, South Logan, Utah, United States] and penicillin/streptomycin [Cellgro, Mediatech, Herdon, Virginia, United States]) were added to each well using a MultiDrop 384 (Thermo Electron, Waltham, Massachusetts, United States) liquid dispenser and incubated at 24 °C overnight. Then, 100 nl of compound stocks dissolved in DMSO at approximately 10 mg/ml was added using the pin transfer robot at the Institute of Chemistry and Cell Biology at Harvard Medical School ( http://iccb.med.harvard.edu ). Cells were incubated at 24 °C for 48 h. All fixation, staining, and washing steps were carried out using a MultiDrop liquid dispenser and 24-channel wand (V&P Scientific, San Diego, California, United States) for liquid removal. Cells were fixed and permeabilized in 40 μl of 100 mM Pipes/KOH (pH 6.8), 10 mM EGTA, 1 mM MgCl 2 , 3.7% formaldehyde, and 0.2% TritonX-100 for 15 min and washed in 50 μl of PBS. The cytoplasm was stained with 40 μl of 0.5 μg/ml NHS-tetramethylrhodamine (TMR, 5-[and-6]-carboxytetramethylrhodamine, succinimidyl ester C-1171, Molecular Probes, Eugene, Oregon, United States) in PBS for 15 min. Subsequently, 40 μl of 5 μg/ml Hoechst 33342 (Sigma, St. Louis, Missouri, United States) in TBST (TBS with 1% TritonX-100) was added for 30 min. This step stains the DNA and quenches excess NHS ester to ensure uniform TMR staining. Cells were washed twice with 40 μl of TBST and sealed with aluminum seals (Costar 6570, Corning, Corning, New York, United States) for image acquisition. Pyrene–actin assay Pyrene-labeled actin (2 μM, final concentration; 80 μl, final volume) was added to 10 mM HEPES (pH 7.7), 2 mM MgCl 2 , 100 μM CaCl 2 , 100 mM KCl, 5 mM EGTA, 200 μM ATP, and 10 μM small molecule. Pyrene–actin polymerization was followed by fluorescence spectroscopy over 45 min. An increase in fluorescence indicates actin polymerization. Adapted from Peterson et al. (2001) . RNAi screen dsRNAs were aliquoted into black, clear-bottom 384-well plates (Costar 3712, Corning) at the Drosophila RNAi Screening Center at Harvard Medical School ( http://www.flyrnai.org ). Each well contained 5 μl of approximately 0.05 μg/μl dsRNA in water. 10,000 Drosophila Kc 167 cells in 10 μl of serum-free Schneiders's Drosophila Medium were added to each well containing dsRNA using a MultiDrop liquid dispenser. After 1 h of incubation at room temperature, 30 μl of medium (Schneider's Drosophila Medium supplemented with 10% heat-inactivated fetal bovine serum and penicillin/streptomycin) was added. The plates were sealed or placed in a humidified chamber and incubated for 4 d at 24 °C. Fixation, TMR, and Hoechst staining were carried out as described above for the small molecule screen. Cells were then blocked in 40 μl of AbDil (TBST with 2% BSA) for 30 min and stained overnight at 4 °C with 20 μl of 1:250 monoclonal anti-tubulin (DM1α, Sigma) and 2 μg/ml Alexa 488 goat anti-mouse antibody (Molecular Probes) in AbDil. Cells were washed twice with 40 μl of TBST and sealed with aluminum seals (Costar 6570) for image acquisition. In order to identify weak hits reliably, the RNAi screen was carried out in triplicate on three separate occasions. Image acquisition Plates from the RNAi and small molecule screens were imaged using a Universal Imaging (Downingtown, Pennsylvania, Unites States) AutoScope or a Universal Imaging Discovery-1. The AutoScope is a Nikon (Tokyo, Japan) TE300 inverted fluorescence microscope with filter wheel (Lamda10-2, Sutter Instruments, Novato, California, United Stats), x-y stage (Prior H107N300), piezoelectric-motorized objective holder (P-723.10, Physik Instruments, Downingtown, Pennsylvania, United States), and a CCD camera (OrcaER, Hamamatsu, Hamamatsu City, Japan). MetaMorph software (Universal Imaging) running the Screen Acquisition drop-in allowed coordination of software-based autofocusing, movement between wells, imaging, and image evaluation. Two images per well were acquired in each of two (small molecule screen) or three (RNAi screen) channels using a 20x objective with 2 × 2 binning. Scoring of images Given the need for greater accuracy in an annotation screen, the RNAi images were initially scored by visual inspection. We looked at two images per well from two independent datasets (datasets 2 and 3, approximately 85,000 images) and noted wells with elevated binucleate levels. To determine the percentage of binucleate cells per image, we used the Integrated Morphometry feature in the MetaMorph software to count the number of nuclei per image and then manually counted the number of binucleate cells. We collected two images per assay well and report the level of binucleates per well as an average of both images. Because the level of binucleates can vary depending on the location of the well in the assay plate or the position of the assay plate in a stack of plates, we decided to compare each proposed hit well to its two neighboring wells to prevent false positives or negatives due to local variations. If a neighboring well also exhibited a phenotype, we chose the next neighbor for our analysis. Although we performed the screen in triplicate, we were only able to apply this analysis to two datasets because the cells in the third screen were too clustered to use automated cell counting. It was possible, however, to estimate the binucleate level in the third dataset by visual inspection. We only scored a phenotype if it repeated in at least two experiments, and the vast majority of phenotypes repeated in all three datasets. In a weakly penetrant phenotype the binucleate level was increased by more than 1.25-fold relative to the average of both neighboring wells. In a medium penetrance phenotype the binucleate level was above 4%, or four times as high as the neighboring wells. In a strongly penetrant phenotype the binucleate level was above 15%. Borr-GFP cloning and transfection For the C-terminal fusion protein, Borr (CG4454) cDNA (LD36125) was cloned into the EcoRI and KpnI sites of pEGFP-C1 (Clontech, Palo Alto, California, United States), cut with NheI and KpnI, and ligated into pPacPL ( Krasnow et al. 1989 ). For the N-terminal fusion, CG4454 digested with SpeI and HindIII, pEGFP-N1 (Clontech) digested with NotI and HindIII, and pPacPL digested with SpeI and NotI were ligated in a triple ligation reaction. Kc 167 cells were transfected with these constructs using Insect GeneJuice transfection reagent (Novagen, Madison, Wisconsin, United States) according to the manufacturer's instructions and were used for live cell imaging and immunofluorescence 6–7 d after transfection. Immunofluorescence analysis Cells were exposed to dsRNA or small molecules, fixed, and stained as described in the screening protocols. Cells were stained with TRITC-labeled phalloidin (Sigma) to visualize actin or antibodies to the following proteins (data not shown): Anillin, Aurora B (a gift from D. Glover), Diaphanous, INCENP (a gift from W. Earnshaw), Lamin (a gift from H. Saumweber), Lava-lamp, Myosin II, Pavarotti, Peanut, Pebble (a gift from H. Bellen), phospho-Histone H3 (Upstate Biotechnology, Lake Placid, New York, United States), Polo (a gift from D. Glover), Rho1 (from the Developmental Studies Hybridoma Bank) and tubulin (DM1α, Sigma). Synthesis of N′ -[1-(3-chloro-4-fluorophenyl)-4-cyano-1H-pyrazol-5-yl]- N , N -dimethyl iminoformamide (binucleine 2) Since binucleine 2 is no longer available commercially, we resynthesized it (see Figure S3 ): 3-chloro-4-fluorophenylhydrazine hydrochloride (compound 1 in Figure S3 ) (500 mg, 2.5 mmol, Alfa Aesar, Karlsruhe, Germany) and ethoxymethylenemalononitrile (compound 2 in Figure S3 ) (305 mg, 2.5 mmol, Sigma-Aldrich, St. Louis, Missouri, United States) were refluxed in 3 ml of ethanol for 4 h. The resulting pyrazol (compound 3 in Figure S3 ) was partially purified by recrystallization from ethanol. Pyrazol (140 mg, 0.5 mmol) and N,N -dimethylformamide dimethyl acetal (150 μl, 1 mmol, Aldrich) were refluxed in ethanol for 1 h. The product (binucleine 2) (compound 4 in Figure S3 ) was recrystallized from ethanol. 1 H NMR (500 MHz, (CD 3 ) 2 SO) δ 8.29 (s, 1 H), 8.06 (dd, J 1 = 2.7 Hz, J 2 = 6.8 Hz, 1 H), 8.03 (s, 1 H), 7.84–7.81 (m, 1 H), 7.55 (t, J = 9.2 Hz, 1 H), 3.14 (s, 3 H), 3.00 (s, 3 H). ESI-MS calculated for C 13 H 11 ClFN 5 291, [M + H] + found 292. Dose response of binucleine 2. Kc 167 cells were treated with 100 nM, 300 nM, 1μM, 5μM, 25μM, or 100 μM of binucleine 2 and fixed after 15 min, 30 min, 45 min, 1 h, 1.5 h, 2 h, 2.5 h, 3 h, 3.5 h, or 4h. After staining with phospho-Histone H3 antibody (Upstate), tubulin (DM1α, Sigma), and DNA and then imaging, cells with phospho-Histone H3 staining were counted. The total number of cells was counted using the Integrated Morphometry feature in the MetaMorph software, and the percentage of cells with H3 staining was calculated and is plotted in Figure 8 . Approximately 4,000–5,000 cells per experiment were assessed in two separate experiments for each time point. Supporting Information Figure S1 Examples of Detailed Secondary Analysis Using Immunofluorescence Cells were untreated or treated with binucleine 2 (100 μM, 48 h) or dsRNA corresponding to aurora B or borr (CG4454) . Cells were stained with TRITC-labeled phalloidin to visualize actin, or antibodies against Anillin, tubulin, Lava-lamp, or Lamin. Lamin-stained cells were treated with binucleine 2 (100 μM) for 24 h. (4.5 MB TIF). Click here for additional data file. Figure S2 Structure Activity Relationships for Binucleine 2 (49KB DOC). Click here for additional data file. Figure S3 Synthesis of Binucleine 2 (32 KB DOC). Click here for additional data file. Table S1 Small Molecule Phenotypes and Structures Kc 167 cells were exposed to small molecules at 100 μM, 30 μM, or 10 μM for 48 h. In a weakly penetrant phenotype (w), the binucleate level was increased by at least 1.25-fold above background. In a medium penetrance phenotype (m), the binucleate level was above 4%, and in a strongly penetrant phenotype (s), the binucleate level was above 15%, while the average binucleate level was approximately 1%. In the binucleate phenotype column, “binucleate” indicates binucleate cells only, “diffuse DNA,” binucleate cells with large, diffuse DNA, “lc,” binucleate cells with low cell count, and “MT ext,” binucleate cells with microtubule extensions. HeLa and BSC-1 cells were exposed to small molecules at 30 μM for 24 h. Growth inhibition in drug-sensitive S. cerevisiae RDY98 (Mat a, erg6ΔTRP1 cg , pdr1ΔKAN, pdr3ΔHIS5+, ade2, trp1, his3, leu2, ura3, can1) was measured at a small molecule concentration of 250 μM after an overnight exposure. (159 KB DOC). Click here for additional data file. Table S2 List of Targeted Genes Identified by Binucleate Cells in the RNAi Screen The “DRSC dsRNA ID” is an internal dsRNA ID number. In the potency column, “s” represents strong, “m,” medium and “w,” weak penetrance of the binucleate cell phenotype. In the phenotypic classification column, “binucleate” indicates binucleate cells only, “diffuse DNA,” binucleate cells with large, diffuse DNA, “lc,” binucleate cells with low cell count, and “MT ext,” binucleate cells with microtubule extensions. Six genes were independently identified in multiple wells, either scored twice— CG10522, cycA, Pp4-19C, RpL32, Tra1, and Ubi-63E —or three times— crn . (141 KB DOC). Click here for additional data file. Table S3 Information about Genes with Binucleate Phenotypes and Quantitative Analysis of Binucleate Phenotypes Gene names, FlyBase IDs, Gene Ontology annotations, forward and reverse primers, and amplicon lengths are shown. “HFA amplicon” and “DRSC dsRNA ID” are internal dsRNA identifiers. The number of potential secondary targets based on 21 nucleotide fragments is the number of genes that have at least one length of 21 bp or more with matching sequence of 21 bp or more of this amplicon. The criteria used in this analysis are such that it may be prone to false positives for secondary targets. The binucleate percentage per well, the relative increase in binucleates relative to the neighboring wells (1.25 = 25% increase), cell number per well, and relative increase or decrease in cell number relative to the neighboring wells are shown for datasets 1 and 2. Annotations for dataset 3 are only shown when they help to define a particular phenotype. (97KB XLS). Click here for additional data file. Table S4 Genes That Scored in the RNAi Screen Sorted by Assigned Functional Groups Functional groups are based on the predicted function as reported by FlyBase (new annotation excluded). (90 KB DOC). Click here for additional data file. Table S5 Genes Reported to Be Involved in Cytokinesis and Genes That Resulted in Strong and Medium RNAi Phenotypes (181 KB DOC). Click here for additional data file.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC528723.xml
551592	Anaphylatoxin C3a receptors in asthma	The complement system forms the central core of innate immunity but also mediates a variety of inflammatory responses. Anaphylatoxin C3a, which is generated as a byproduct of complement activation, has long been known to activate mast cells, basophils and eosinophils and to cause smooth muscle contraction. However, the role of C3a in the pathogenesis of allergic asthma remains unclear. In this review, we examine the role of C3a in promoting asthma. Following allergen challenge, C3a is generated in the lung of subjects with asthma but not healthy subjects. Furthermore, deficiency in C3a generation or in G protein coupled receptor for C3a abrogates allergen-induced responses in murine models of pulmonary inflammation and airway hyperresponsiveness. In addition, inhibition of complement activation or administration of small molecule inhibitors of C3a receptor after sensitization but before allergen challenge inhibits airway responses. At a cellular level, C3a stimulates robust mast cell degranulation that is greatly enhanced following cell-cell contact with airway smooth muscle (ASM) cells. Therefore, C3a likely plays an important role in asthma primarily by regulating mast cell-ASM cell interaction.	Role of complement system in the development of asthma Asthma, a complex airway inflammatory disease, is characterized by bronchoconstriction, airway hyperresponsiveness (AHR) and remodeling. Current consensus suggests that T H 2 cytokine producing T cells, mast cells and ASM cells play central roles in the pathogenesis of asthma [ 1 - 7 ]. The complement system forms an important part of innate immunity against bacteria and other pathogens. As a system of 'pattern recognition molecules', foreign surface antigens and immune complexes initiate a proteolytic pathway leading to the formation of a lytic membrane attack complex. The anaphylatoxins C3a and C5a are released as byproducts of complement activation and modulate innate immunity. Accordingly, C5a is involved in a number of inflammatory diseases such as immune-complex-mediated lung injury and sepsis [ 8 , 9 ]. A role for C3a in innate or adaptive immunity, however, has only been recently recognized [ 10 ]. C3a levels are elevated in bronchoalveolar lavage (BAL) fluid after segmental allergen challenge in asthmatic but not healthy subjects [ 11 - 14 ]. Furthermore, plasma C3a is also elevated in acute exacerbations of asthma [ 11 ]. Additionally, single nucleotide polymorphisms in C3 and C3a receptor genes increases susceptibility to asthma [ 15 ]. Collectively, these findings suggest that C3a and the cognate G protein coupled receptor (C3aR) may play a role in the development of airway hyperresponsiveness (AHR) and inflammation. C3a receptors in models of Airway Hyperresponsiveness Studies with animal models provided compelling evidence for C3aR activation in the development of AHR and inflammation. Humbles et al., [ 12 ], showed that C3aR (-/-) mice in BALB/c strain are protected from AHR in response to aerosolized ovalbumin challenge following intraperitoneal sensitization with ovalbumin [ 12 ]. However, C3aR (-/-) mice developed normal airway inflammatory response with no difference in T H 2 cytokine production and eosinophil recruitment in BAL when compared to wild-type mice. Additionally, guinea pigs with a natural defect in C3aR expression were also protected from AHR in response ovalbumin to challenge with no effect on airway inflammation [ 16 ]. These initial findings suggested that C3a modulates AHR, perhaps, via a direct action on airway smooth muscle cells [ 12 , 17 , 18 ]. Recent studies using C3aR (-/-) mice provided new insights on the role of C3a on both AHR and airway inflammation [ 19 ]. When sensitized intraperitoneally with extracts of Aspergillus fumigatous and challenged intranasally with allergen, these mice experienced substantial decreases in both AHR and airway eosinophilia relative to wild-type mice. Furthermore, BAL levels of T H 2 cytokines (IL-4, IL-5, IL-13), IgE titres and mucous production were all significantly reduced in C3aR (-/-) mice. Allergen-challenged C3 (-/-) mice also display diminished AHR, lung eosinophilia and T H 2 cytokine production when compared to wild-type mice [ 20 ]. These findings support a role of C3a receptors in the development of AHR and inflammation. However, the effect of C3aR on different phases of AHR models may depend on the nature of the allergen, method of sensitization and the strain of mice used. C3a generation in asthma Increased level of C3a in BAL of subjects with asthma implies a potential role for this apaphylatoxin in promoting airway inflammation. However, the cells responsible for C3a generation and the airway effector cells stimulated by C3a remain unknown. Plausibly, antibody generated during antigen sensitization may interact with allergen to activate the classical complement pathway. Additionally, airway epithelial cells and pulmonary macrophages secrete both C3 and several components of the alternate pathway of complement (factors B, H, and I and properdin) [ 21 - 23 ]. Thus, activation of alternative or the lectin pathway on the allergen may also lead to the generation of C3a. It is noteworthy that house dust mite protease, allergenic extracts of Aspergillus fumigatous and mast cell tryptase also activate the complement pathway directly [ 13 , 24 - 26 ]. Thus, combination of different pathways likely generates C3a in the airway of individuals with asthma (Figure 1 ). Figure 1 Model for C3a generation in individuals with asthma . C3 may be secreted from pulmonary resident cells (e.g. epithelial cells and macrophages) or derived from plasma leakage. Antibody (IgG) present in the serum of sensitized individual can form a complex with allergen to activate complement via the classical pathway. Proteases derived from allergen or released from activated mast cells are able to cleave C3 to generate C3a. Activation of alternative or the lectin pathway on the allergen together with factors B, H, and I and properdin released resident cells may generate C3a. C3a has little effect on allergen sensitization in models of AHR Both antigen-presenting cells (APCs) and activated T cells express C3aR [ 27 - 30 ], raising the possibility C3a may regulate sensitization phase of allergic asthma. Kawamoto et al [ 31 ], recently used wild-type and C3aR-/- mice to characterize the immune response to C3a. Convincingly, C3aR deficiency had little effect on T H 2 cytokine response to intraperitoneal ovalbumin sensitization. Furthermore, C3a had no effect on T H 2 cytokine production in response to T cell receptor ligation. Further, Taube et al., [ 32 ] showed that administration of complement inhibitor in mice after sensitization but before allergen challenge prevented the development of AHR and blocked T H 2 cytokine production and lung inflammation. Additionally, a small molecule antagonist of C3a receptor, when administered after sensitization but before challenge also caused significant inhibition of airway inflammation [ 33 ]. These findings suggest that the effect of C3a on the development of allergic AHR may not involve modulation of the sensitization phase of the disease. Relationship between C3aR and FcεRI in mast cell activation in asthma Mast cells appear to play a pivotal role in the development of AHR and inflammation [ 34 ]. The ability of allergen to cross-link high affinity IgE receptors (FcεRI) on mast cells to induce degranulation and leukotriene generation is well documented [ 35 , 36 ]. Surprisingly, the role of C3a in mast cell activation remains controversial and appears to depend on the mast cells subtype. For example, murine bone marrow-derived mast cells and a rat basophilic leukemia, RBL-2H3 cells, which have been used extensively as mast cell models, do not express C3a receptors [ 37 ]. In contrast, C3a receptors are expressed in human CD34 + -derived primary mast cell cultures [ 38 , 39 ], human mast cell lines HMC-1 [ 40 , 41 ] and LAD 2 [ 39 ] as well as murine pulmonary mast cells (Thangam, B and Ali, H, unpublished data). Interestingly, C3a is one of the most potent mast cell chemoattractants known [ 42 , 43 ]. C3a also induces robust mast cell degranulation [ 38 , 39 ] and leukotriene C 4 generation (Thangam, B and Ali, H, unpublished data). These findings suggest that allergen induces mast cell degranulation by at least two mechanisms: cross-linking of FcεRI and via C3a generation following complement activation by allergen protease (Figure 2 ). Mast cell proteases also activate the complement pathway to generate C3a [ 26 ]. Therefore, C3a generation following FcεRI aggregation may amplify mast cell mediator release (Figure 2 ). Figure 2 Proposed interaction between FcεRI and C3aR leading to mast cell activation . Allergen cross-links FcεRI on mast cells to induce degranulation. Allergen can also activate complement pathway (see Fig. 1) to generate C3a, which in turn activates its cognate G protein coupled receptors on mast cells to induce degranulation. Mast cell proteases also activates complement cascade to generate C3a. This C3a may serve to amplify mast cell mediator release. Mast cell-ASM interaction in asthma Recent studies with immunohistological analysis of bronchial biopsy specimens from subjects with asthma and those from patients with eosinophilic bronchitis provided important insight on the role of mast cell-ASM cell interaction in the development of AHR in asthma [ 4 , 44 , 45 ]. Asthma and eosinophilic bronchitis are characterized by similar inflammatory infiltrates in the submucosa of the lower airway. However, ASM infiltration by mast cells is a feature of asthma and not eosinophilic bronchitis. This difference in mast cell recruitment in asthma is associated with AHR, which is absent in in eosiniphilic bronchitis [ 6 ]. Furthermore, degranulated mast cells are detected in greater number in ASM bundles of patients who died from asthma when compared to non-asthmatic control [ 46 ]. Based on these findings, new hypothesis suggests that increased mast cell recruitment and interaction with ASM may promote release of mast cell-derived mediators that modulate resident airway cell function is asthma [ 4 , 5 , 44 ]. ASM is not only a contractile tissue that responds to mast cell-derived mediators in asthma, but also modulates mast cell function and airway inflammation. ASM cells express stem cell factor (SCF), which induce mast cell chemotaxis, survival and differentiation [ 47 , 48 ]. Interleukin-1β, tumor necrosis factor (TNF) and T H 2 cytokines IL-4 and IL-13 derived cytokines also stimulate ASM to express a large number of chemokines and cytokines [ 49 - 52 ]. Thus, activated ASM cells secrete chemokines and cytokines that may recruit and retain mast cells into the ASM. C3a receptors and mast cell-ASM cell interaction C3a has long been recognized as an agent that evokes force generation in smooth muscle. In guinea pigs, C3a-induced contraction of lung parenchyma may involve indirect effects of histamine and arachidonic acid metabolites [ 53 ]. In mice, C3a does not cause shortening of isolated tracheal strips [ 10 ]. Furthermore, C3a fails to induce AHR after intratracheal instillation in naïve mice [ 10 ]. In contrast, in mice immunized with house dust mite, subsequent intratracheal administration of C3a stimulates both AHR and airway inflammation [ 10 ]. These findings suggest that C3a-induced AHR and bronchoconstriction requires enhanced infiltration and activation of inflammatory cells, likely mast cells. Recently, investigations showed that human mast cells but not human or murine ASM express C3aR [ 54 ]. Interestingly, incubation of mast cells with human ASM cells, but not its culture supernatant, significantly enhanced C3a-induced mast cell degranulation. Although stem cell factor (SCF) and its receptor c-kit are constitutively expressed on ASM cells and mast cells respectively, neutralizing antibodies to SCF and c-kit failed to inhibit ASM cell-mediated enhancement of mast cell degranulation. Dexamethasone-treated ASM cells however normally express cell surface SCF but were significantly less effective in enhancing C3a-induced mast cell degranulation when compared to untreated cells. Collectively, these findings suggest that cell-cell interaction between ASM cells and mast cells, via a SCF-c-kit independent but dexamethasone-sensitive mechanism, enhances C3a-induced mast cell degranulation, which likely regulates ASM function and may contribute to the pathogenesis of asthma. While mast cells and ASM cell interaction plays a role in AHR, airway inflammation in asthma is strongly linked to T H 2 lymphocyte and their cytokines IL-4, IL-5 and IL-13. These cytokines play key roles in the recruitment and activation of eosinophil, mucous production and IgE synthesis. Allergen challenge of sensitized C3 (-/-) and C3aR (-/-) mice decreased production of T H 2 cytokines in BAL and substantially reduced recruitment of T cells, eosinophils and neutrophils in lung tissue [ 19 , 20 ]. Furthermore, inhibition of complement activation or administration of C3aR antagonist during the effector phase of asthma substantially inhibited airway inflammation [ 32 , 33 ]. These findings suggest activation of C3aR is required for T H 2 effector function in murine model of allergen-induced inflammation. Accordingly, in human mast cells, C3a stimulates the production of MCP-1, RANTES [ 39 ], IL-8 and IL-13 (Thangam, B and Ali, H, unpublished data)-cytokines and chemokines are responsible for the recruitment of T lymphocytes, eosinophils and neutrophils into the airway. Further, C3aR are expressed on basophils, eosinophils and bronchial epithelial cells [ 18 , 54 - 57 ]. Thus, interaction of a number of inflammatory and resident cells likely regulate C3a-dependent T H 2 cytokine and chemokine production in asthma (Figure 3 ). Figure 3 Model for the role C3a in AHR and airway inflammation in asthma . C3a generated in individuals with asthma (see Fig. 1) induces mast degranulation (Fig. 2) to promote ASM force generation. Chemokines and cytokines expressed by ASM recruit and retain mast cells into the ASM layer resulting in further smooth muscle dysfunction. T H 2 cytokines and chemokines generated from mast cells (and possibly eosinophils and bronchial epithelial cells) regulate AHR and airway inflammation. Conclusion Accumulating evidence suggests that C3a may play an important role in the pathogenesis of asthma. In murine models of allergic AHR and inflammation, inhibition of complement activation or small molecule antagonists of C3a receptor after sensitization but before allergen challenge inhibits airway responses. Furthermore, cell-cell interaction between ASM cells and mast cells enhances C3a-induced mast cell degranulation, which likely regulates ASM function, thus contributing to the pathogenesis of asthma. Further investigations on cellular and molecular mechanisms by which C3a modules mast cell-ASM interactions may offer novel therapeutic approaches to the treatment of asthma and airway inflammation. List of Abbreviations used C3aR, C3a receptor; AHR, airway hyperresponsiveness: ASM, airway smooth muscle; BAL, bronchoalveolar lavage. Competing interests The author(s) declare that they have no competing interests.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC551592.xml
546005	Context-dependent selection of visuomotor maps	Background Behavior results from the integration of ongoing sensory signals and contextual information in various forms, such as past experience, expectations, current goals, etc. Thus, the response to a specific stimulus, say the ringing of a doorbell, varies depending on whether you are at home or in someone else's house. What is the neural basis of this flexibility? What mechanism is capable of selecting, in a context-dependent way an adequate response to a given stimulus? One possibility is based on a nonlinear neural representation in which context information regulates the gain of stimulus-evoked responses. Here I explore the properties of this mechanism. Results By means of three hypothetical visuomotor tasks, I study a class of neural network models in which any one of several possible stimulus-response maps or rules can be selected according to context. The underlying mechanism based on gain modulation has three key features: (1) modulating the sensory responses is equivalent to switching on or off different subpopulations of neurons, (2) context does not need to be represented continuously, although this is advantageous for generalization, and (3) context-dependent selection is independent of the discriminability of the stimuli. In all cases, the contextual cues can quickly turn on or off a sensory-motor map, effectively changing the functional connectivity between inputs and outputs in the networks. Conclusions The modulation of sensory-triggered activity by proprioceptive signals such as eye or head position is regarded as a general mechanism for performing coordinate transformations in vision. The present results generalize this mechanism to situations where the modulatory quantity and the input-output relationships that it selects are arbitrary. The model predicts that sensory responses that are nonlinearly modulated by arbitrary context signals should be found in behavioral situations that involve choosing or switching between multiple sensory-motor maps. Because any relevant circumstancial information can be part of the context, this mechanism may partly explain the complex and rich behavioral repertoire of higher organisms.	Background The concept of a direct, one-to-one association between a sensory stimulus and a motor response has been strongly influential in neuroscience [ 1 ]. Such associations may be quite complex; for instance, monkeys can learn visuomotor mappings based on arbitrary rules [ 2 - 4 ]. But from a mechanistic point of view, it is their flexibility which is remarkable. Humans and other mammals react to a given stimulus in drastically different ways depending on the context [ 1 , 5 - 7 ]. What is the neural basis for this? How do current goals, recent events, and other environmental circumstances gate or route immediate sensory signals to generate an adequate action? Gain control is a common mechanism by which neurons integrate information from multiple modalities or sources [ 8 , 9 ]. Gain-modulated neurons typically have a sensory receptive field, but in addition, their overall excitability depends on some other modulatory parameter. A classic example are the neurons in parietal area 7a, whose activity can be described by the product of a gain factor, which is a function of the gaze angle, and the response profile of the visual receptive field [ 10 , 11 ]. That is, gaze direction determines the amplitude of their stimulus-dependent responses. According to theoretical studies, gain-modulated responses are useful for performing a class of mathematical operations known as coordinate transformations [ 12 - 16 ]. For example, by combining multiple eye-centered inputs that are gain modulated by gaze direction, a downstream neuron can generate a response that depends on the location of a stimulus relative to the body [ 12 - 14 ]. Experimental studies have reported gain changes due to a wide range of proprioceptive signals, such as gaze direction [ 10 , 11 , 17 ], eye and head velocity [ 18 ] and arm position [ 19 , 20 ]. Modulations relevant to attention-centered [ 21 - 23 ] or object-centered representations [ 24 , 25 ] have also been documented. Interestingly, all of these examples deal with the same problem – spatial localization – but the computations that can be effectively carried out through gain-modulated responses are much more general [ 13 , 16 , 26 ]. In particular, here I show that modulating the activity of a population of neurons is equivalent to turning on and off different subsets of neurons. Thus, the modulation can be thought of as a switch that can activate one of many possible sensory networks, each instantiating a different sensory-motor map. Crucially, the modulatory signal itself does not have to provide any spatial information; it can be a recent instruction or some other aspect of the current behavioral context. Examples of choices between multiple sensory-motor maps determined in a context-dependent manner include speaking in one language or another, and the ability of musicians to interpret a musical score depending on the clef and key signature at the beginning of each stave. But the same principles also apply in more simplified settings, such as behavioral tasks where a given stimulus is arbitrarily associated with two or more motor responses, depending on a separate instruction [ 4 , 27 - 29 ]. For instance, the shape of a fixation point may be used to indicate whether the correct movement should be a saccade toward a spot of light or an antisaccade away from it [ 30 ]. What all of these cases have in common is a functional reconnection between visual and motor networks that must occur very quickly and without explicit spatial guidance from the context information. Using theoretical and computer-simulation methods, I show that this type of functional switching can be achieved through contextual modulation regardless of how the context is encoded – whether continuously or discontinuously – and independently of the discriminability of the stimuli. The results are presented using neural network models of hypothetical behavioral tasks similar to those used in experiments with awake monkeys. A report with a different example was published previously [ 31 ]. Results All model networks discussed below have the same general, two-layer architecture [ 14 - 16 ]. A first layer of gain-modulated (GM) neurons drives a second layer of output or motor neurons through a set of feedforward connections, with each GM unit projecting to all output units. In each trial of a task, the GM neurons are activated by the sensory and context signals, and a motor response is generated by the output neurons (see Methods). Each model proceeds in three steps. First, the GM and the desired output responses are specified according to the task. Then, synaptic weights are found that, across all stimulus and context combinations, make the driven output responses as close as possible to the desired ones. Finally, the network is tested in multiple trials in which the GM neurons drive the output units. Model performance is measured by comparing the resulting, driven pattern of motor activity in each trial with the desired, pre-specified one. The first task, with only two contexts, serves to illustrate the analogy between gain modulation and a switch. Switching between saccades and antisaccades In the antisaccade task, a stimulus appears briefly at position x along the horizontal and the subject responds by making an eye movement (Fig. 1 ). There are two possible contexts or conditions. In the first one, the movement should be to the location where the stimulus appeared, x ; in the second one, the movement should be to the mirror-symmetric point, - x . Both condition and stimulus location vary across trials. The color of the fixation spot (or any other arbitrary cue) may serve to indicate which condition applies in each trial [ 30 ]. Examples of model GM responses chosen for this task are shown in Fig. 2 . These neurons simply respond to visual stimuli presented at different locations; however, they are also sensitive to the context. Each graph shows the mean firing rate of one unit as a function of x , with one curve for each of the conditions (red and green traces). These tuning curves are bell-shaped because Gaussian functions were used to define them (see Methods). Because context affects the gain of the responses, for any given cell, the two curves differ only in their amplitudes. The context that produces the highest gain is the preferred one. The maximum and minimum gains for each neuron are model parameters that can be between 0 and 1. The four GM units in Fig. 2 illustrate various degrees of modulation. The case of full modulation (maximum gain = 1, minimum gain = 0) depicted in Fig. 2a corresponds to a neuron that is switched on and off by context: in its preferred condition it is highly active, whereas in its non-preferred condition it is fully suppressed. First consider what happens if the first layer of a model network is composed of two populations of such switching neurons. One population is active in context 1 and the other in context 2. This is illustrated in Fig. 3a . The rectangle encloses the responses of all model neurons (60 GM and 25 output units) in a single trial of the antisaccade task. The firing rates of the GM neurons are in color. The two populations (red and green) have opposite context preferences but identical sets of sensory tuning functions. The black dots are the responses of the driven output neurons. Their center of mass (Equation 19), which in this case is the same as the location of the peak, is interpreted as the target location for an impending saccade. The network performs accurately in the four trials shown in the column, since the encoded movement location is equal to x for saccades (context 1) and to - x for antisaccades (context 2). It is easy to see why such a network can implement two entirely independent sensory-motor maps: each population has its own set of synaptic connections driving the downstream motor neurons, and the maps are kept separate because the two populations are never active at the same time. Figure 3d shows the corresponding matrix of synaptic connections. To interpret this figure, notice that GM units 1–30 are the ones that prefer context 1 (red dots in Fig. 3a ), whereas units 31–60 prefer context 2 (green dots in Fig. 3a ). Preferred stimulus locations are arranged in increasing order for both populations. Units 1–30 generate direct saccades, so their connections are aligned with the motor neurons; that is, GM neuron 1 excites output neuron 1 most strongly, GM neuron 2 excites output neuron 2 most strongly, etc. Thus, in context 1, stimuli to the right generate movements to the right. In contrast, the GM units that generate antisaccades are connected in the reverse order: GM neuron 31 excites output neuron 25 most strongly, GM neuron 32 excites output neuron 24 most strongly, and so on. Thus, in context 2, stimuli to the right result in movements to the left. The model correctly produces saccades in context 1 and antisaccades in context 2. Furthermore, this scheme for switching sensory-motor maps as a function of context would also work for any two maps driven by the two populations. This model switches maps successfully because the GM neurons are themselves switched on and off by context, so this case is trivial. However, the main result in this section is that a network of partially modulated GM neurons has exactly the same functionality. The more rigorous statement is this: for a discrete number of contexts and everything else being equal, a network of partially modulated neurons can generate the same mean downstream responses as a network of switching neurons. Figure 3 illustrates this equivalence: identical output activity profiles are generated when all GM neurons are fully suppressed in their non-preferred context (Fig. 3a ), when all are partially modulated by the same amount (Fig. 3b ), and when the modulation varies randomly across cells (Fig. 3c ). These three cases require different sets of synaptic connections between GM and output layers, but this is simply because the GM responses vary across cases. In particular, note the dark blue diagonal bands in Figs. 3e,3f , compared to Fig. 3d . They correspond to negative weights needed to subtract out activity that is irrelevant to a particular context. For instance, in the direct saccade trials of Fig. 3b , the responses of the antisaccade-preferring neurons should be cancelled, and viceversa. The new negative weights combined with larger positive weights achieve this. The key point is that, under relatively mild conditions, partial and full modulation lead to the exact same repertoire of switchable sensory-motor maps (the difference lies in their accuracy, as discussed below). The formal proof is presented in Appendix A. This result is interesting because it provides an intuitive interpretation of gain modulated activity: modulations that may seem small at the single-unit level may produce drastically different output responses due to their collective effects, the result being as if different sensory populations had been turned on and off. Partial versus maximum gain modulation The equivalence between networks of neurons that switch across contexts and networks with partial modulation is subject to an important condition and a qualification. The key condition is that the modulation factors that determine the gain of all the neurons with similar stimulus selectivities must be linearly independent across contexts (Appendix A). In practice, one way to achieve this is to include all relevant combinations of sensory and contextual preferences. For instance, if there are two neurons that respond maximally when x = 5, the condition is fulfilled for that pair if one neuron prefers context 1 and the other context 2. As long as this independence constraint is satisfied, there is great flexibility in the actual amount of modulation; it does not need to be 100%, as with a full switch. The qualification, however, is also critical, because a network of partially modulated GM neurons is not exactly the same as one composed of switching neurons: in most functionally relevant cases, partially modulated neurons are effectively noisier. In general, variability plays an important role in the performance of these networks. No fluctuations were included in the simulations of Fig. 3 , so performance was virtually perfect. But the magnitude of the error between correct and encoded movement directions increases depending on the amount of noise that is added to the GM responses, and as the difference between the minimum and maximum gains diminishes, the impact of noise typically goes up. This is shown analytically in Appendix C and is illustrated in Fig. 4 . Two measures of noise sensitivity are plotted in Fig. 4 . The first one is the standard deviation of a single output response across trials with identical stimulus and context. This number, σ R , quantifies the variability of single neurons. Figure 4a plots σ R as a function of γ , which is the minimum gain of the GM neurons (the maximum is 1). When γ = 0, the GM neurons are fully suppressed in their non-preferred context; when γ = 1, the GM responses are identical in both contexts. The three curves are for three levels of noise. Their order shows that, as expected, higher noise in the input layer always produces higher variability in the output. For each data point, the synaptic weights were set so that the average firing rates of the output neurons, as functions of stimulus location and context, were always the same (Appendix B). Thus, for all γ values, the average profile of motor responses for x = -15 and x = 10 looked exactly like those in Fig. 3 . The monotonically increasing curves in Fig. 4a indicate that the variability of the output rates goes up with γ , as predicted theoretically (Appendix C). The second measure of noise sensitivity is σ CM , which estimates the error between the desired movement location and the center of mass of the output population, which is considered the encoded movement location (Equations 19, 20). Thus, σ CM quantifies the variability of the network. Figure 4b shows that σ CM also increases with γ , reaching a saturation level. This error saturates because, in contrast to the individual neuron responses, the encoded movement location is restricted to a limited range of values, so its variance cannot grow above a certain limit. Figures 4c and 4d show the same measures of variability but when the synaptic weights are computed using the standard, optimal algorithm (see Methods). For each value of γ , the optimal algorithm considers both the mean and the variance of the output responses [ 32 , 33 ], striking a balance between them that, overall, minimizes the average squared difference between the driven and the desired output rates (Equation 11). Therefore, in Figs. 4c,4d , the mean output responses are not quite the same for all data points; in particular, for x = -15 and x = 10 there are small differences compared to the curves in Fig. 3 (data not shown). This method markedly reduces the variability of the individual output neurons relative to the case where only the mean values are considered. It also produces a modest decrease in σ CM (compare Figs. 4b and 4d ). However, it does not change the main effect: the error in the encoded location still grows monotonically with γ . Note that, as explained in Appendix C, γ > 0 does not always produce higher variance in the output, compared to γ = 0. For instance, if the sensory-motor maps in the two contexts are the same, the optimal strategy is to activate both populations of GM neurons simultaneously, i.e., to use γ = 1. This is simply because the average of two noisy responses with equal means is better than either of them. In general, however, switching is relevant precisely when the sensory-motor maps are different, as in Figs. 3 and 4 , in which case weaker modulation (higher γ ) results in higher output variability. In conclusion, as the modulation becomes weaker, the performance of the network typically becomes less accurate, even though the average output responses may be close or identical to those obtained with maximum modulation. In Fig. 4 , this becomes more of a problem when the minimum gain γ is above 0.6 or so, at which point σ CM is about twice that observed with full modulation. These results were obtained using the same γ for all GM neurons, but almost identical curves were produced when γ varied randomly across cells and the results were plotted against its average value. Continuous vs discontinuous context representations The possible contexts encountered by an organism could be numerous and diverse, so it is not clear how the brain might encode them. There are at least two distinct ways: as separate, discrete states, or as points along a smooth, continuous space. What would be the difference in terms of the functionality of the remapping networks studied here? This is investigated next, using a generalization of the antisaccade task referred to as the scaling task. The scaling task is very much like the antisaccade task, except with more contexts. The subject's response should be an eye movement toward a location determined by the position of the stimulus, x , and a scale factor, y ; the movement should be toward the point xy . When y = 1, the movement is simply a saccade toward x ; when y = -1, the movement is an antisaccade toward - x ; when y = 0.5, the movement should be to a point halfway between fixation and the location of the stimulus, and so on. To begin with, five possible conditions are considered, corresponding to scales of -1, -0.5, 0, 0.5 and 1. Figure 5 shows the responses of four GM units in this task plotted as functions of the position of the stimulus along the horizontal. A family of five curves, one per condition or scale factor, is drawn for each unit. As in the previous task, the shape of these curves is constant across conditions, because of the multiplicative interaction between stimulus- and context-dependent factors. The neurons in Figs. 5a,5b encode the context in a discontinuous way, because the order in which they prefer the five scales was set randomly (see Methods). Thus, for each unit, the order of the colors in Figs. 5a,5b is random. In contrast, the neurons in Figs. 5c,5d encode context smoothly; their response amplitudes decrease progressively as the current scale y differs from each cell's preferred scale. All units in the figure have approximately the same minimum gain, γ ≈ 0.5. Differences between these two coding strategies can be observed in Fig. 6 . This figure shows the performance of two versions of the network model, each with 900 GM cells, in four trials of the scaling task. In the first version, illustrated in Figs. 6a-6d , context is encoded discontinuously, as in Figs. 5a,5b . The GM firing rates are color-coded, ordered according to their preferred stimulus locations (x-axis) and preferred scales (y-axis). In each trial, the GM rates form a band of activity centered on the location of the stimulus. The most intense responses are somewhat clustered, although high firing rates are scattered throughout the band. The band occurs because the responses vary smoothly as functions of stimulus location, and the scatter in the y-direction is due to the random order in which each neuron prefers the contexts; such scatter would be present even without noise. The output neurons have profiles of activity (black traces) with the highest peak located near the intended movement target. The small wiggles and secondary bumps are due to noise. The performance of the network is accurate, however: the encoded movement is close to the intended one for all combinations of stimulus location and scale factor (Figs. 6a-6d , red vs black vertical lines). The second version of the model, illustrated in Figs. 6e-6h , is almost identical to the first, except that context is encoded continuously, as in Figs. 5c,5d . Now the the activation pattern that emerges is clearly localized, centered on the current stimulus and context values. Performance is similar for the two networks, both having σ CM ≈ 0.6. Figures 7a,7b evaluate the performance of these two models across a wider range of parameters. The graphs show σ CM as a function of the number of GM neurons for three levels of noise. In all cases, the error decreases approximately as – a sign that noise is what limits the accuracy of the system. This is consistent with the virtually perfect performance obtained with zero noise. With the five selected contexts, results are almost identical for the continuous and discontinous cases. Robustness and generalization There are two aspects of these networks that could vary depending on how context is encoded. The first is their robustness. In addition to random variations in the GM responses, there could be fluctuations in other elements of the circuits, in particular, the synaptic connections. Thus, a key question is how finely-tuned these connections need to be in order to obtain accurate performance. The answer: not very much. The networks tolerate considerable alterations in synaptic connectivity. This is illustrated in Figs. 7c,7d , which show σ CM as a function of the number of GM neurons in networks in which the connections were corrupted. For these plots, the connections were first set to their optimal values, as in the standard simulations, but then 25% of them, chosen randomly, were set to zero. To generate the same range of output firing rates, all remaining connections were divided by 0.75, but no further adjustments were made. Performance was then tested. Compared to the results with unaltered weights (Figs. 7a,7b ), performance is evidently worse, but the disruption is not catastrophic; in particular, the error still goes down with network size. The increase in error is most evident when the noise is relatively low. Random weight deletion was used for these simulations because it is a rather extreme form of weight corruption, but other manipulations generated similar results. The second important issue about these networks is their capacity to generalize. So far, the models have been tested with the same stimuli and contexts used to set the connections, but what happens when new stimuli or contexts are presented? This is where partial modulation and a smooth organization of response properties make a difference. First consider the model in which scale is encoded discontinuously. Its performance in generalization is shown in Fig. 7e . For this graph, only 8 stimulus locations, in combination with the 5 possible scales, were used to calculate the synaptic weights. That is, only 8 evenly-distributed values of x were used in Equation 3, giving a total of 40 combinations of stimulus and context. However, the network was tested with all 151 combinations of 31 stimulus locations (between -15 and +15) and 5 scales. Accuracy is practically the same as in the original simulations (Fig. 7a ), where the 31 stimuli and 5 scales were used both for setting the connections and evaluating performance. The same scales had to be used in both cases because, given the discontinuous encoding, the gain factors for other scales could not be interpolated or inferred. In contrast, in the continuous case, generalization can be tested in both the sensory and modulatory dimensions; the GM responses can be obtained for any combination of stimulus location and scale, because both quantities are defined analytically (Equations 3 and 6). Results are shown in Fig. 7f . For this graph, 8 stimulus locations and 8 scale factors were used to set the connections. The network was then tested on 31 stimulus locations and 31 scales within their respective ranges. Performance is slightly better than in the standard condition in which identical combinations of 31 stimulus locations and 5 scales were used throughout (Fig. 7b ). In summary, this task requires somewhat more complex GM neurons than the antisaccade task, because there are more contexts. In the discontinuous case, the basic intuition for why the model works is the same as in the previous task: with the proviso that they are effectively noisier, partially modulated neurons are equivalent to switching neurons, which can trivially establish independent sensory-motor maps. However, the key advantage of a continuous neural representation of context over a discontinuous one is that it allows a network to perform accurately on combinations of stimulus and context that have not been explicitly encountered before. By its very definition, such continuous encoding requires partial modulation. Therefore, although partial modulation is typically detrimental for switching between discrete contexts (relative to full switching), it is highly advantageous when context is parameterized by a continuous variable, because it serves to generalize. Remapping based on ambiguous stimuli In the scaling task, all stimuli and contexts are unambiguous, but in many real-life situations and experimental paradigms, motor actions are preceded by perceptual processes that involve the interpretation or analysis of sensory information. That is, specific actions (e.g., pressing a left or right button) are often based on ambiguous information (e.g., whether on average a group of flickering dots moved to the left or to the right). In theory, switching between maps should be independent of the perceptual component of a task (Appendix A). To investigate this, consider the orientation discrimination task illustrated in Fig. 8 . In each trial, a bar is presented and the subject must determine whether it is tilted to the left or to the right. The judgement is indicated by making an eye movement either to a left or a right target. Discrimination difficulty varies depending on orientation angle x . The task is most difficult when x is near 0° and the bar is nearly vertical, but it becomes easier as x approaches ± 45°. This is also a remapping task because the association between bar orientation and correct target is not unique: the color of the fixation spot determines whether left and right targets correspond to bars tilted to the left ( x < 0) and to the right ( x > 0), respectively, or viceversa. There is also a no-go condition, which gives a total of three. The GM cells in this case are tuned to stimulus orientation. The response curves are not shown, but have a single peak, as in Figs. 5a,5b – the difference is that the sensory variable is orientation, which varies from -90° to +90°, and that there are only three conditions, three values of y (see Methods). The order in which each GM cell prefers the three contexts is set randomly, so context is encoded discontinuously. The responses of the model output units are shown in Figs. 9a-9h . In no-go trials (Figs. 9g,9h ), all neurons fire near their baseline rates, as prescribed (Equation 9). Thus, in this condition the network ignores the stimuli. In go trials, however, the profile of output responses has peaks at -10 and +10, which are the only two target locations in this task. In contrast to the activity profiles seen in previous tasks, here there never is a unique peak, even with zero noise (Figs. 9a,9c,9e ). Instead, the relative amplitude of the two peaks varies as a function of bar orientation. The difference in the amplitudes of the two hills of activity decreases as the bar becomes more vertical, thus reflecting the difficulty of the task. Without any noise, the largest peak is always located at the correct target, but with noise the amplitudes vary across trials and errors are produced (Fig. 9d ). To quantify the performance of the network in this case, the generated movement was set equal to the location of the tallest hill of activity. This always corresponded to one or the other target location, +10 or -10, so each trial could be scored as either correct or incorrect. The assumption here is that a profile of activity with two peaks, as in Figs. 9c,9d , can be converted into a profile with a single peak, such that the smaller hill of activity is erased. Networks with recurrent connections organized in a center-surround fashion can do just that [ 26 , 34 - 36 ]. So, if such lateral interactions were added to the output layer of the network, only the largest hill of activity would remain. Equivalently, the responses of the output neurons could serve as inputs to an additional, third layer that performed the single-target selection [ 34 ]. Either way, given that this is a plausible operation, it is reasonable to simply consider the location of the largest peak to determine the evoked movement. Based on this criterion, the performance of the network is shown in Figs. 9i,9j , which plot the probability or fraction of movements to the target on the right as a function of stimulus orientation x . These are essentially neurometric curves – psychometric curves computed from neuronal responses – and indeed have the sigmoidal shape that is characteristic of many psychophysical measurements. Figure 9i shows the results for condition 1, in which bars with x > 0 correspond to movements to the right; Fig. 9j shows the results for condition 2, in which the association is reversed and bars with x > 0 correspond to movements to the left. The gray curves are best fits to the simulation data points. The fits have two parameters, the center point, or bias (indicated by dashed lines), and a second parameter that determines the steepness at the center point and is inversely proportional to the discrimination threshold (see Methods). Without noise, performance is virtually perfect (not shown), in which case the bias and threshold are zero and the neurometric curve becomes a step function. However, both quantities increase in magnitude as noise is increased, producing the observed sigmoidal curves. The presence of a bias might be surprising, given the symmetry of the network. However, the bias depends on the number of trials used to estimate the probabilities. If each orientation were tested an infinite number of times, the data points in Figs. 9i,9j would line up perfectly along continuous curves. The discrimination thresholds of those curves would not be significantly different from those shown, but their biases would be zero. With finite samples, a bias in the neurometric curve is inevitable. Figures 10a,10b show the bias and discrimination threshold as functions of network size for three levels of noise. Both quantities decrease with network size, so in this sense, the network is just as effective as that for the scaling task. Because large numbers of trials were used, the bias is about an order of magnitude smaller than the threshold. Figures 10c,10d plot the results when the synaptic connections in the network are corrupted by deleting 25% of them at random, as in Figs. 7c,7d . This manipulation leaves the discrimination threshold virtually unchanged, but increases the bias by about an order of magnitude, making it comparable to the threshold. This bias is a true limitation of the network; it does not decrease with more trials. Figures 10e,10f show performance during generalization, as in Fig. 7e . In this case, only the two extreme orientations, -8° and +8°, were used to set the connections (in combination with the three possible conditions). The network was then tested on the standard set of 64 orientations. A true bias also appears in this case. It stays lower than the threshold, which remains essentially unchanged. In summary, although the ambiguity of the sensory information is reflected in the motor responses, it does not interfere with the context-dependent selection mechanism. Discussion Gain modulation as a switch The above results demonstrate that contextual modulation could serve to select one of many associations or maps between sensory stimuli and motor responses. Indeed, a key insight is that modulating the gain of a neural population is, in a sense, equivalent to flipping a switch that turns on or off specific subpopulations of neurons. This explains why networks of GM neurons can generate large changes in downstream responses – even all-or-none changes, as in go vs no-go conditions (Figs. 9a-9h ) – although their own activity may vary rather subtly. In this framework there is a distinction between the selection process and the sensory representations. The capacity to switch depends on the collection of gain factors, whereas the space of possible functions of the stimulus that can be computed downstream is determined primarily by the sensory tuning curves (Appendix A). A weaker modulation typically increases the sensitivity to noise of the resulting motor responses (Appendix C), but otherwise, partial modulation can achieve the same sensory-motor map selection as maximal, all-or-none modulation. This is why the mechanism works across a large variety of tasks and representations that involve some type of switch. In a landmark paper, Pouget and Sejnowski [ 13 ] studied the capacity of GM networks for coordinate transformations using the concept of basis functions. A group of functions of x form a basis set when any arbitrary function of x can be computed as a linear superposition of those functions in the group; sines and cosines of are a well known example. The function of x typically associated with a neuron's response is its tuning curve – its firing rate measured as a function of x . Pouget and Sejnowski showed that, starting with two networks that form separate basis sets for x and y , a network of GM neurons comprising all possible combinations (i.e., pairwise products) of those two sets would form a basis set for functions that depend simultaneously on x and y . This means that any function of x and y can be computed from the resulting GM responses. This was a crucial result, because it provided a rationale for generating such a combined representation. However, it assumed that both the sensory and modulatory variables are continuous and that, taken independently, the sets of x - and y-dependent tuning curves both form true basis sets. The present results relax some of these assumptions and provide a complementary point of view. When the modulatory quantity y varies discretely, each of its values corresponds to computing a different function of the stimulus x . Furthermore, the x - dependent tuning curves determine what functions of x or sensory-motor maps can be computed downstream, but there is no requirement for them to form a strict basis set. As mentioned, the discontinuous case fits better with the idea of switching between various possible maps, as if separate populations of neurons were turned on and off. This approach also highlights two important characteristics of these networks, that the modulation factors need to be nonlinear functions of context (Appendix A), and that the sensitivity to noise depends on the magnitude of the modulation (Appendix C). Relation to other models An important property of networks of GM neurons is that the output units read out the correct maps using a simple procedure, a weighted sum [ 13 - 15 ]. As a consequence, the overall strategy of these networks can be described as follows: the input data are first projected onto a high-dimensional space, and the responses in this space are then combined through much simpler downstream units that compute the final result – in the present case, x and y are the inputs and the high-dimensional space is composed of the GM responses. Interestingly, such expansion into an appropriate set of basis functions [ 13 , 37 ] is the central idea of many other, apparently unrelated models. For instance, this scheme is a powerful technique for tackling difficult classification and regression problems using connectionist models [ 33 ]. It also works for calculating non-trivial functions of time using spiking neurons [ 38 ]. This strategy might constitute a general principle for neural computation [ 37 ]. In addition, these networks are capable of generalizing to new stimuli and are quite resistant to changes in the connectivity matrix, so they don't require exceedingly precise fine-tuning. The problem of high dimensionality A crucial requirement for the above scheme of projecting the data onto a suitable set of basis responses is to cover all relevant combinations of sensory stimuli and modulatory signals in the GM array [ 13 - 15 ]. It is the potentially large number of such stimulus-context combinations that may pose a challenge for these networks, a problem sometimes referred to as the curse of dimensionality [ 33 ]. In terms of the antisaccade task, for example, the context could be signaled by the shape or color of the fixation spot, the background illumination of the screen, a sound, or simply by past events, as would happen if the experiment ran in blocks of saccade and antisaccade trials. Each one of these potential cues would need to have a similar modulatory effect on the sensory responses, and it is not clear how the brain could establish all the necessary connections for this. Part of the problem is that we don't know how many independent dimensions there are – independence being the crucial property. For instance, the model for the antisaccade task has two contexts and requires two populations of switching neurons. More neurons are needed to deal with the version of the task that has five scales or contexts, but the number of necessary neurons does not keep growing endlessly; if the modulatory terms are chosen appropriately, a relatively small number of neurons can generalize to any scale, in effect generating an infinite number of sensory-motor maps. Of course, the key is that these are not independent, so the network can generalize. Thus, the scheme might work with realistic numbers of neurons if the number of independent context dimensions is not exceedingly large, but estimating this number is challenging. Another possibility is to have a relatively small number of available gain modulation patterns controlled by an additional preprocessing mechanism that would link them to the current relevant cue (the color of the fixation spot, its shape, the background illumination, etc.), a sort of intermediate switchboard between possible contexts and possible gain changes. Attention has some features that fit this description – it can select or favor one stimulus over another, it can act across modalities, and it can produce changes in gain [ 21 - 23 , 39 ]. No specific proposals in this direction have been outlined yet, neither theoretically nor experimentally, but this speculative idea deserves further investigation. Is exact multiplication needed? A key ingredient of the general, two-layer model is that the GM neurons must combine sensory and modulatory dependencies, f ( x ) and g ( y ), nonlinearly [ 13 - 15 ]. Results of two manipulations elaborate on this. First, when f and g were added (Equation 15) instead of multiplied, all transformations failed completely, as expected [ 13 ]. Second, when the sensory- and context-dependent terms were combined using other nonlinear functions (a sigmoid function, a rectification operation or a power-law; see Equations 16–18), accuracy remained approximately the same in all tasks. Results are shown in Table 1 , which compares the performance of networks that implemented different types of stimulus-context interactions but were otherwise identical. This shows that the exact form of the nonlinearity used to combine f and g is not crucial for these models. However, in some cases a multiplication allows the synaptic connections to be learned through simple Hebbian mechanisms [ 14 , 15 ], so it may be advantageous for learning. At least under some conditions, neurons combine their inputs in a way that is very nearly multiplicative [ 11 , 21 - 23 ]. Perhaps they do so when multiplication provides a specific computational advantage. Mixed sensory-motor activity In the model for the orientation discrimination task, the level of activity of the output neurons reflects not only the evoked movement but also the difficulty of the sensory process. This is consistent with the observation that, during sensory discrimination tasks, neuronal responses in many motor areas carry information about the stimulus [ 40 - 42 ]. This activity is often interpreted as related to a decision-making process. In the discrimination model, the responses of the neurons encoding the movement toward one of the targets increased in proportion to the strength of the sensory signal linked with that target (Figs. 9a-9f ), as observed experimentally [ 40 - 42 ]. The model was not designed to do this. It simply could not generate single, separate peaks of activity for two nearby orientations on the basis of a single feedforward step; an additional layer or additional lateral connections would be required for that. Nevertheless, when such selection mechanism is assumed to operate, remapping proceeds accurately, even when the strength of the sensory signal varies. According to the model, sensory and motor information should be expected to be mixed together when distinct, non-overlapping responses (e.g., movement to the left or to the right) are generated on the basis of small changes in a stimulus feature that varies continuously, as orientation did in this task. Responses that depend on multiple cues In the present framework, if sensory responses were modulated by multiple environmental cues, the responses of downstream neurons could be made conditional on highly specific contextual situations (see ref. 16). Therefore, this mechanism may also explain the capability of some neurons to drastically change their response properties in a context-dependent way. Two prominent examples are hippocampal place cells, whose place fields can be fully reconfigured depending on multiple cues [ 43 , 44 ], and parietal visual neurons, which become selective for color only when behavioral context dictates that color is relevant [ 45 ]. In many tasks, two or more inputs are combined into conditional statements – 'if X and Y then Z '. The switching property of GM networks is useful in these situations as well. The study of abstract rule representation by Wallis and Miller [ 29 ] is a good example. In their paradigm, the decision to hold or release a lever depends on an initial cue and on two pictures. The cue indicates which of two rules, 'same' or 'different', is applied to the pictures. If the rule is 'same', the lever is released when the two pictures are identical but not when they are different; if the rule is 'different' the situation reverses, the lever is released when the two pictures are different but not when they are identical. To execute the proper motor action, two conditions must be checked. With the framework presented here, it is straightforward to build a model for that task; all it requires is a neural population that encodes the similarity of the pictures (i.e., is selective for matching vs non-matching pairs) and is gain modulated by the rule. Although the exact form of the modulation, for instance, whether it is close to multiplicative, is hard to infer from their data, the findings of Wallis and Miller [ 29 ] are generally consistent with the types of responses predicted by the model. Experimental predictions Other experimental studies also include results that are consistent with gain interactions between multiple sensory cues [ 2 , 27 , 28 , 30 ] or with gain changes due to expected reward [ 46 ]. Interpreting these data is problematic, however, because those experiments were not designed to test whether changes in context generate changes in gain. The tasks described here, or similar paradigms, may be simplified to eight or so stimuli and two or three conditions, generating stimulus sets that would be within the range of current neurophysiological techniques with awake monkeys. The key is to be able to construct full response curves (Figs. 2 , 5 ), so that neuronal activity across contexts can be compared for several stimuli – not only for two, as is often done. This is because the models make three basic predictions that can only be tested with multiple stimuli and conditions: the responses should have mixed dependencies on stimulus and context, the mixing should be nonlinear, and the neurons should behave approximately as a basis-function set, in the sense that a weighted sum of their responses should approximate an arbitrary function of stimulus and context extremely well [ 31 , 47 ]. Ideally, the nonlinear mixture will show up through multiplicative changes in gain, as in Figs. 2 and 5 , where the context-dependent variations in firing intensity respect stimulus selectivity. This could certainly happen [ 11 , 21 - 23 ], especially for some individual neurons, but other nonlinearities are possible [ 19 , 47 , 48 ] and might work equally well. A key observation is that context can include widely different types of circumstancial information, such as expected reward, motivation, fear or social environment [ 1 , 5 - 7 ]. Therefore, given the versatility of the models discussed here, a broader implication of the present work is the possibility that, as a basis for adaptive behavior, the brain systematically creates sensory responses that are nonlinearly mixed with numerous types of contextual signals. Conclusions The framework discussed here demonstrates how to make a neural network adaptable to various environmental contingencies, labeled here simply as context. To achieve this flexibility, context must influence the ongoing sensory activity in a nonlinear way. This strategy was illustrated with tasks akin to those used in neurophysiological experiments with awake monkeys, but is generally applicable to the problem of executing a sensory-evoked action only when a specific set of conditions are satisfied. The mechanism works because changing the gain of multiple neurons is, in a sense, equivalent to flipping a switch that turns on and off different groups of neurons. Its main disadvantage is that all relevant combinations of stimulus and context must be covered, which may require a large number of units. On the upside, however, the switching functionality is insensitive to the quality or content of the sensory signals, is robust to changes in connectivity, and places minimal restrictions on how context is encoded. Future experiments should better characterize how cortical neurons integrate sensory and contextual information. Methods Neuronal responses The GM responses depend on a sensory feature x , which may represent stimulus location or stimulus orientation, and on a context signal y . For each model GM cell, these quantities are combined through a product of two factors, f and g . The former determines the sensory tuning curve of the neuron and the latter its gain or amplitude as a function of context y . The mean firing rate r j of GM unit j is thus written as r j = r max f j ( x ) g j ( y ) + B , (1) where f j and g j vary between 0 and 1, B is a baseline firing rate equal to 4 spikes/s and r max = 35 spikes/s. The specific functions used for f j and g j depend on the task and are described below. However, note that because these two terms are combined through a multiplication, changes in context alter the overall responsiveness of a cell, but not its selectivity, which is the defining feature of gain modulation [ 8 , 9 ]. To include neuronal variability, Gaussian noise is added to all GM responses in each trial of a task. The noise is multiplicative; that is, the variance of the noise for unit j is equal to α r j , where α is a constant. Qualitatively similar results are obtained with additive instead of multiplicative noise. The output neurons form a population code that represents the location of an impending eye movement. Each firing rate is determined by a weighted sum of the GM rates, where the weights correspond to synaptic connections. The mean rate of output neuron i is where w ij is the synaptic weight from GM neuron j to output neuron i . The output rates should encode the location of the movement to be made in each trial, so their profile of activity should have a single peak indicating the location that should be reached. The synaptic connections that achieve the correct sensory-motor alignment are found through an optimal algorithm, which is described further below. Equation 2 is used when the GM neurons drive the output neurons. But for each task, there is also an intended or desired response for each output unit. This is denoted as F i , and is a function of the stimulus and the context. Thus, R i and F i refer to the same postsynaptic neuron, but one is the actual driven response and the other is the intended response. The functions f , g and F vary across tasks, as described next. Parameters for the antisaccade task In this task (Fig. 1 ), stimuli appear at a location x in two possible contexts, labeled y = 1 and y = -1. The response should be a movement toward the location equal to xy . The firing rates depend on the following functions. The tuning curves are Gaussians, with the preferred stimulus location a j between -25 and +25 and σ f = 4 (in Figs. 2 , 3 ) or σ f = 6 (everywhere else). Because there are two conditions, the modulatory functions g j take only 2 values, 1 and γ , where γ is the minimum gain. Crucially, one half of the GM neurons have g j ( y = 1) = 1 and g j ( y = -1) = γ , whereas the other half have the opposite context preference, g j ( y = 1) = γ and g j ( y = -1) = 1. The only exceptions are Figs. 3c,3f , in which the gain factors g j for each neuron were chosen randomly from uniform distributions. In this task, the desired response of output neuron i is where c i is the preferred movement location of unit i and σ F = 4. Therefore, the output profile of activity (obtained by plotting F i vs c i ) should be a Gaussian centered at the intended target location, x or - x , depending on the context. Parameters for the scaling task The scaling task is identical to the antisaccade task, except that the context y can take many values. The tuning functions f j , are the same (Equation 3) and the output responses again encode the location given by xy (Equation 4). The gain factors depend on which of two possible representations is used. When context is encoded discontinuously, five scales are used, y = - 1, -0.5, 0, 0.5 or 1, so the modulatory functions g j must take five values; these are g j ( y ) = {1, 0.9, 0.75, 0.65, 0.5}. (5) Crucially, they are assigned randomly to each of the 5 conditions, with a new random permutation for each GM unit. As a final step, the g j values are jittered by small, random amounts (see Figs 5a,5b ). On the other hand, when context is encoded continuously, each neuron is assigned a preferred scale b j between -1.4 and +1.4, and its gain is a Gaussian function of y , with σ g = 0.3. Note that the minimum gain is 0.5 in both cases. Parameters for the orientation discrimination task In this task (Fig. 8 ), x is the orientation of a bar and varies between -8° and +8°, where x = 0° corresponds to vertical. The discrimination can occur in two ways: either left and right targets correspond to left- and right-tilted bars, respectively ( y = 1), or viceversa ( y = 2). In addition, there is a no-go condition ( y = 3), for a total of three contexts. The orientation tuning curves are given by cosine functions, where a j is now a preferred orientation between -90° and +90°. The modulation functions g j are generated as in the discontinuous version of the scaling task, except with three values, g j ( y ) = {1, 0.75, 0.5}. (8) The order in which each GM neuron prefers the three contexts is random. The responses of the motor neurons are given by with y = 3 being the no-go condition and σ F = 4. Here, M ( x , y ) is the correct movement location, either -10 or +10, when orientation x is presented in condition y . Specifically, for y = 1, M = -10 if x < 0 and M = +10 if x > 0; and for y = 2, M = -10 if x > 0 and M = +10 if x < 0. In no-go trials, all output responses should stay at the baseline level, B . Simulation results in the orientation discrimination task are presented in terms of the probability of generating a movement toward the target on the right, P R ( x ), which is a function of orientation. Those results are fitted to the curve where erf is the error function. This expression has two parameters: a e , which is the center point, or bias, and b e , which is inversely proportional to the maximum slope. The discrimination threshold is defined as one half of the difference between the values of x that give P R = 0.75 and P R = 0.25; for Equation 10 it is equal to b e erfinv(1/2), where erfinv is the inverse of the error function. Calculation of synaptic weights The synaptic weights are chosen so that, on average, the driven and desired responses of the output neurons are as close as possible. This means that must be minimized. As in Equation 2, w ij is the connection from GM neuron j to output neuron i . The angle brackets indicate an average over all values of x and y and over multiple trials. The optimal connections are found by taking the derivative of the above expression with respect to w pq , setting the result equal to zero, and solving for the connections. The result is where C kj ≡ < r k r j > (13) L kj ≡ < F k ( x , y ) r j >. (14) Equations 12–14 are the recipe for setting the connections. Here C -1 is the inverse of the correlation matrix C defined above. This inverse (or the pseudo-inverse) is found numerically. To calculate the averages defined above, the GM rates for all values of x and y are needed. These are found by evaluating Equation 1 plus the noise term for each GM neuron. When the noise is uncorrelated across neurons, as in the simulations, it only contributes to the diagonal of C . Because the variance of the noise is equal to α times the mean response, noise adds an amount α < r i > to element C ii of the correlation matrix. Except for this, all averages C kj and L kj are obtained from the mean input responses given by Equation 1 and the corresponding F i functions of the output neurons. Having specified the parameters of the network (number of GM and output units, tuning and gain functions, stimulus-movement association), the procedure for setting the synaptic weights is run only once. Afterward, the connections are not adjusted any further. Other response functions Equation 1 is based on an exact multiplication between f j ( x ) and g j ( y ). The effects of other possible interactions between stimulus and context are investigated using four alternative expressions in place of Equation 1. First, a linear combination of sensory and context signals, Then, three nonlinear interactions. The first one is based on rectification, r j = r max [ f j ( x ) + g j ( y ) - 1] + + B , (16) where [ x ] + = max{0, x }. The second one uses a sigmoid function, The sigmoid is widely used in artificial neural networks [ 12 ] and has two parameters, a s and b s . The third nonlinear interaction is based on a power law, and has two parameters too. This type of expression approximates some of the gain effects observed experimentally [ 49 ]. The free parameters in these expressions are adjusted so that the resulting firing rates are as close as possible to those given by Equation 1. All else is as in the original simulations. Outline of the simulations Having specified a task, the tuning and gain curves of the GM neurons, and the network connections, the model is tested in a series of trials of the task. Each trial consists of the following steps: (1) specifying the stimulus and context, x and y , (2) generating all GM responses (Equation 1), (3) calculating the driven, output responses (Equation 2), and (4) determining the encoded movement M out by using the center of mass of the motor activity profile (Equation 19). Finally, the encoded movement is compared to the movement M desired that should have been performed given x and y – their difference is the error in that particular trial. The encoded movement M out is equated with the center of mass of the output population, where c i is the preferred target location of output unit i . The root-mean-square average of the motor error is used to quantify performance over multiple trials, where only go trials are included in the calculation. On average, the encoded movement is very near the desired one, < M out - M desired > ≈ 0. Thus, σ CM is the standard deviation of the motor error, and measures the accuracy of the output population as a whole. In all tasks, 25 output units are used, with c i uniformly spaced between -25 and 25. Preferred stimulus values a j and preferred context values b j are first distributed uniformly and then jittered by small, random amounts. All simulations were performed using Matlab (The Mathworks, Natick, MA). The source code is available on request. Appendix A This section shows that, with a finite number of contexts, gain modulation is functionally equivalent to a switch. More precisely, for a discrete number of contexts and everything else being equal, a network of partially modulated neurons can generate the same mean downstream responses as a network of switching neurons. Consider M populations or groups of sensory neurons with identical sets of tuning functions f j ( x ). There are N neurons in each population, so index j runs from 1 to N . These populations project to a postsynaptic neuron through synaptic connections , where the superscript indicates the presynaptic population of origin. Thus, is the synaptic weight from neuron j in group p to the postsynaptic unit. The sensory neurons are gain modulated, so the mean response of unit j in population p is given by where x and y label the stimulus and the context, as before. Next, assume that there are M possible contexts, so y can take integer values from 1 to M . Therefore, the gain factors can be expressed as three-dimensional arrays, and the presynaptic firing rates can be rewritten as Here, corresponds to the gain of unit j in population p during context k . With this notation, the response of the downstream neuron during context k becomes where the sums are over all populations and all units in each population. Note that, for each index j , the coefficient in front of the tuning function is given by the product of an M -dimensional vector of weights times an M × M matrix of gain factors. The essential idea is to compare the response of the postsynaptic unit under two conditions: when only one input population is active in any particular context (and all others are fully suppressed), and when the populations are only partially suppressed, with different combinations of gain factors for each context. For this, the hat symbol ^ is used to label all quantities obtained in the former case, with switching neurons; that is, the hat means 'obtained with full modulation'. Full modulation occurs when the matrix of gain factors for all the units with index j is equal to the identity matrix, According to this expression, for population 1, the gain is 1 when y = 1 and is 0 otherwise; for population 2, the gain is 1 when y = 2 and is 0 otherwise, and so forth. Substituting this expression in equation 23 gives the firing rate of the downstream unit when driven by switching neurons, Again, the hat simply indicates that the quantity was obtained with maximally modulated input neurons. In this case, ( x , y = k ) implements a different function of x for each context value, such that the function expressed in context 1 depends only on the weights from the first population of switching neurons, , the function expressed in context 2 depends only on the weights from the second population, , and so on. This is the situation depicted in Figs. 3a,3d . On the other hand, the postsynaptic response driven by partially modulated units is simply as in Equation 23, where the absence of a hat means 'obtained with partial modulation'. Under what conditions is the output response driven by partially modulated neurons, R ( x , y = k ), equal to the response obtained with full modulation, ( x , y = k )? Compare the right hand sides of Equations 23 and 25; for them to be the same, the coefficients in front of the tuning functions must be equal; that is, This condition is satisfied if the weights with partial modulation are set equal to where h j is the inverse of the matrix of gain factors g j ; that is, . Therefore, the key constraint here is that the gain factors in the partial modulation case must have linearly independent values across contexts, so that the inverses exist; in other words, the matrices g j must have full rank. An important consequence of this is that for M > 2, the gain of each neuron as a function of context ( in Equation 21) must be nonlinear. Equation 27 is the key result. It provides a recipe for going from a network of switching neurons to a network of partially modulated neurons (given equal numbers and types of tuning functions f j ( x )). For the recipe to apply, the gain factors in the latter must have the appropriate inverses, but otherwise they are arbitrary. Because each possible function that a network can generate corresponds to a different matrix of synaptic connections, this implies that all the possible functions of x that the output can implement with fully switching neurons can be replicated with partial gain modulation. This statement is exact when there is no noise; with noise it applies to the average downstream responses. Notice that this result is independent of the tuning functions f j ( x ). These determine the possible functions of x that can be generated downstream – that is, the available sensory-motor maps – but have no effect on how these are switched or selected. Finally, the result is also valid if the postsynaptic response is equal, not simply to the weighted sum of GM responses, but to an arbitrary function of that sum. Appendix B To illustrate the result in Appendix A, consider a simple case with two populations and two contexts, as in Figs. 1 , 2 , 3 , 4 . With full modulation, the response of the downstream neuron is in context 1, and in context 2. This is simply Equation 25 for M = 2. Context turns one sensory population on and another off. Now, how can we obtain the same downstream responses, as functions of x , when the GM neurons are partially modulated? First, suppose that the modulation matrices are The gain factors can only take two values, 1 for the preferred context, and 1 > γ ≥ 0 for the non-preferred one; the full-modulation case is recovered when γ = 0. For simplicity, these factors are the same for all units in each population, so there is no variation across index j . This matrix was used in Fig. 3 , with γ = 0 and γ = 0.5 for the left and middle columns, respectively, and in Figs. 4a,4b . Its inverse is Next, substitute into the transformation rule found earlier, Equation 27; the result is With these synaptic weights, the downstream responses driven by the partially modulated neurons (Equation 23 with p = 1, 2 and the gain factors in Equation 30) become identical to the rates driven by the switching neurons (Equations 28,29). This is the linear transformation used in Figs. 4a,4b . Appendix C Using partial instead of full modulation to switch between maps does come at a price: the variability of the postsynaptic response typically increases. This can be seen as follows. If additive noise is included in the input firing rates, the response of neuron j in population p becomes where is a random fluctuation for unit j in population p during context k . The variance across trials of this random variable is denoted as and is the same for all GM neurons. The downstream neuron has the same mean response as before (Equation 23), but now it has a variance, which is equal to Here, the angle brackets indicate an average over trials, which affects the noise terms only. To go from the second to the third line above, the key is to assume that the fluctuations are independent across neurons, such that . The next step is to compare the variance of the postsynaptic unit when driven by the switching neurons and by the regular, partially modulated GM neurons. For simplicity, consider the same 2 × 2 case as in Appendix B, where the modulation is parameterized by γ . The variance of the postsynaptic response driven by switching neurons is exactly as in Equation 35, but with and p = 1,2. This must be compared to the variance obtained with partial modulation for the same mean postsynaptic responses. The synaptic weights that achieve this are given by Equations 32 and 33; substituting those into Equation 35 gives This is the variance of the postsynaptic response driven by an array of partially modulated GM neurons as a function of the variance obtained when the response is driven by fully modulated, switching neurons. Here, a depends on the weights , but is not a function of γ , where b is a constant. Note that a is a measure of the overlap between the sets of connections from the two populations. The dependence of on the weights is such that a ≤ 2 . Equation 36 shows that, although the average postsynaptic response is the same function of x and y for all γ , its variability changes with γ . A similar result is obtained when the variance of the input firing rates is proportional to their mean. In that case, with = 1. A calculation analogous to the one just described leads to where a is the same as in Equation 37, except with a different proportionality constant. In this case, it is still true that a ≤ 2 . This expression was used to generate the continuous lines in Fig. 4a . For this, was simply the variance in the postsynaptic firing rate found from the simulations with γ = 0, and b was chosen to generate the best fit to the rest of the simulation data points. Equations 36 and 39 do not always increase monotonically with γ . This depends on a , which is a measure of the similarity between the sensory-motor maps established in the two contexts. For instance, when the two maps are the same, for all j , and a attains its maximum value, 2 . In that case, the variance with partially modulated neurons is always smaller than with switching neurons. This makes sense: if the maps in the two contexts are the same, it is always better to have the two populations active at the same time, as this reduces the noise. According to the analysis, when a = 2 and γ = 1, Equation 36 gives . The variance is divided by 2 because noise is additive and there are two active populations doing the very same thing. In contrast, when the two maps are different, their respective synaptic weights are also different, and a is either positive but much smaller than 2 , or negative. Then, might have a minimum for some intermediate value of γ , or may increase monotonically, which is what happens in Figs. 3 and 4 , with saccades vs antisaccades. List of abbreviations GM, gain-modulated.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC546005.xml
517712	A robust, low- to medium-throughput prnp genotyping system in sheep	Background In many countries breeding programs for resistance to scrapie in sheep are established. Therefore, the demand on genotyping capacities of the polymorphisms of the prion protein gene ( prnp ) relevant to presently known disease associations and EU regulations is steadily increasing. Most published typing methods are not well suited for routine typing of large sample numbers in smaller service laboratories for different reasons: they require partly manual data processing, sophisticated and sensitive protocols, high efforts regarding time and manpower, multiple step reactions or substantial hardware investments. To overcome these drawbacks, we developed a prnp typing method that is based on a `multiplex amplification refractory mutation system' (ARMS) reaction. Methods In this study we combined the amplification refractory mutation system (ARMS) with standard fluorescent based fragment length analyses method to develop a prnp genotyping method (PRNP ARMS). Results By optimised primer design it was possible to type the 4 relevant single nucleotide polymorphisms (SNPs) in the prnp simultaneously in one multiplex reaction. Automated fragment length analysis enabled automated allele designation. Suitability of the PRNP ARMS for routine application was proven by typing samples with known genotypes and larger sample numbers from half-sib families. Conclusion The ARMS PRNP typing method established in this study is universally suited for a broad range of typing projects with different requirements. It provides an efficient and inexpensive diagnostic mutation analysis that will improve the quality of prnp genotyping compared with other low-cost methods. It can be implemented by most molecular genetic laboratories using standard equipment.	Background Scrapie is a contagious prion disease [ 1 ] of sheep and goat. In contrast to BSE, there is no evidence for a transmission to humans. Nevertheless, BSE is experimentally transmittable to sheep and the resulting disease can not be distinguished from scrapie [ 2 , 3 ]. Even though BSE has not been found in farmed sheep yet, the possibility that BSE occurs in "scrapie" diseased sheep can not be excluded. Therefore, scrapie control programs were implemented in many countries. Since no vaccine or therapeutic means is currently available, these programs rely on selective breeding for scrapie resistance. Susceptibility to scrapie is largely controlled by three polymorphic amino acid positions (136, 154, 171) of the ovine prion protein gene ( prnp ) [ 4 ] and reliable genotyping of the corresponding DNA polymorphism is required as a basis for selection decisions. For determination of prnp alleles or haplotypes, there are several typing techniques applied today. Direct sequencing covering the region of exon 2 encoding amino acid positions 136, 154 and 171 is the most accurate method, that enables the typing of additional and detection of hitherto unknown polymorphisms. Other methods are classical PCR-restriction fragment length polymorphism (RFLP)-typing [ 5 , 6 ], DNA strand conformation polymorphism detection, denaturing gradient gel electrophoresis (DGGE) [ 7 ], PCR-single-strand conformational polymorphism (PCR-SSCP) [ 8 ], hybridisation with allele-specific oligonucleotides [ 9 , 10 ], primer extension [ 11 ], and so on. For high-throughput genotyping matrix assisted laser desporption/ionisation – time of flight (MALDI-TOF) systems (e.g. for bovine prnp [ 12 ] and the LGC-web page [ 13 ]), taqman ® (Applied Biosystems, CA) [ 14 ], and pyrosequencing (Biotage AB, Uppsala, Sweden) are used, but methodologically details of these methods are not very well documented in the literature. The amplification refractory mutation system (ARMS) method is well established [ 15 ] and has been applied not only to single nucleotide polymorphism (SNP) detection but also to haplotype determination [ 16 ]. In the present work we combined the use of fluorescence labelled oligonucleotides with ARMS technology for the simultaneous detection of 4 SNPs without using common primers. Based on this PRNP ARMS method we developed a cost-effective, well-reliable and low- to high-throughput technology for prnp typing that can easily be adopted by a wide range of other laboratories. Methods DNA-Samples EDTA-blood or Typifix ® -tissue samples were collected from half-sib families from three common sheep breeds in Bavaria. All lambs were born in the years 2001 and 2002. DNA was isolated by using the E.Z.N.A. Blood DNA Mini Kit (#12-3482-03, PEQLAB Biotechnologie GmbH, Erlangen, Germany) or the NucleoSpin ® Multi-96 Tissue kit (MACHEREY-NAGEL GmbH & Co. KG, Düren, Germany), respectively. PRNP-ARMS Design of Primers An overview of primer locations is given in Fig. 1 . Initially, allele-specific primers were designed that differed only at the 3'-nucleotide (given in bold letters in Table 1 ). Using these primers it was not possible to obtain reliable discrimination of alleles. Especially the simultaneous use of primers for alleles Q-171 and R-171 resulted in cross-amplification products. Therefore, a number of additional primers were tested to optimise allele discrimination. Destabilizing mismatches were introduced at the first or second penultimate base in addition to the allele-specific base at the 3' termini of the primers (ARMS principle, indicated by small letters in Table 1 ). The sense primers specific for the alleles at amino acid position 136 were labelled with two different fluorochromes, 4, 7, 2', 7'-tetrabetweenchloro-6-carboxyfluorescein (TET) and 6-carboxyfluorescein (FAM) (Table 1 , Fig. 1 ). That allows the detection and discrimination of the alleles by standard gel electrophoresis on a fluorescent sequencer. Pieces of neutral sequence [ 17 ] were added to the 5'-end (underlined in Table 1 ) of the unlabeled antisense primers for amino acid positions 154 and 171. These enable the differentiation of the alleles by different lengths. Mismatches between primers were deliberately introduced at the 5' region in these neutral sequences to avoid jumping amplification products. For a delineation of primer sequences with a reference sequence (Fig. 3 ). PCR-reaction For the ARMS reaction haplotypes were amplified from 2 μl of DNA solution with standard buffer conditions, 1.5 mM MgCl 2 , dNTP's (25 nM each), and 0.75 units of HotStar-taq polymerase (Qiagen, Hilden, Germany) in a final volume of 10 μl on a t-gradient 96-well thermocycler (Biometra, Göttingen, Germany). Primer concentrations were 10 nmol each. Cycling was for 15 min at 95°C, [0.5 min at 94°C, 1 min at 62°C, 1 min at 74°C]33× without final extension. Fragment analysis Fragment lengths of ARMS amplification products were analysed on an ABI PRISM ® 310 genetic analyser (Applied Biosystems, Foster City, CA) with 5 sec injection time, 15 kV, 60° and 18 min runtime using 36 cm capillaries. Allele designations were generated automatically using the Genotyper ® software (V. 2.5, Applied Biosystems) as described [ 18 ]. The data were imported into an Microsoft Access database. Standard DNA samples representing different alleles were included in the assay to control each PCR mix and reaction. As an additional control a sql-query was used to flag all genotypes transferred to the database that were not compatible with the 15 known standard genotypes (e.g. peaks from FAM in the range of 153 bp) for retyping. Comparative sequencing Part of prnp exon 3 coding for amino acids 47 to 246 of the prion protein was amplified in 10 μl reaction volume using 2 μl of genomic DNA, 0.05 μM of each primer (5'-tcc tgg agg caa ccg cta tc-3' and 5'-gga gga tca cag gag ggg aag-3'), 50 μM of each dNTP, 0.5 units of HotStar-taq polymerase (Qiagen, Hilden, Germany), and reaction buffer containing 1.5 mM MgCl 2 . Cycling-conditions on a Biometra T gradient 96-well thermocycler were: 15 min at 95°C, [0.5 min at 95°C, 1 min at 60°C, 0,5 min at 72°C]35× and 10 min, 60°C final extension. Before direct sequencing PCR-reactions were purified using the 96-well MultiScreen-PCR plates (Millipore, Bedford, MA). Sequencing was performed using BigDye V 2.0 terminator cycle sequencing kit (Applied Biosystems, Foster City, CA). Sequencing analysis was run on an ABI PRISM ® 310 genetic analyser. Results Development of the PRNP ARMS method We have established a multiplex `amplification refractory mutation system' (ARMS) based method for the identification of prnp 136-154-171 genotypes in sheep. The PRNP ARMS method was developed using standard DNAs representing all five main haplotypes. In a first stage, different primer pairs were tested for allele specific amplification using an anneal-temperature gradient. By redesign of primers for amino-acid positions 154 and 171 and introducing deliberate mismatches close to the 3' end an optimised primer set was established (Table 1 ) allowing multiplex, allele-specific amplification. "False positive fragments" (peaks above the background signal generated by primers complementary to alleles that were not present in the corresponding sample) were not observed using the optimised primer set. Nevertheless, the average peaks heights differed between alleles. For example the average peak heights of the allele Q-154 was 1.5 times higher than that of R-154. The ratio of peak heights in heterozygous individuals differed from 0.5 to 2.5 fold (compare Fig. 2 ). These were observed when different PCR assays with low number of samples and small volumes of master-mix are compared. The peak height differences depend mainly on the relative primer concentrations and can be avoided e.g. by using premixed primer solutions containing all ARMS primers. For automated "base calling" the minimum peak height was set to 100 (in practice most samples gave peak heights well above 1000). Proof-of-principle To provide a proof-of-principle the same 140 sheep DNAs were analysed by direct PCR sequencing and by PRNP ARMS. Typing results were identical with both methods. In a second step, 420 sheep from half-sib families and 20 samples from the international sheep and goat DNA typing comparison test 2003 of the International Society of Animal Genetics were analysed by the PRNP ARMS method. All genotypes followed the rules of Mendelian inheritance and no deviation from the main 5 haplotype patterns was observed. Allele frequencies derived from these animals from three different Bavarian breeds are shown in Table 2 . Paternity of all lambs was checked with a set of 9 microsatellites (data not shown). Typical electropherograms representing different haplotype combinations are shown in Fig. 2 . Allele length determination was highly reproducible (Table 3 ). The range of individual peaks from each individual allele was well below +/-1 bp allowing unambiguous allele-calling and automatic generation of results tables using the genotyper software. Furthermore, the PRNP ARMS reaction worked fairly stable with different DNA qualities. The quality and concentration of the DNAs that were isolated from tissue varied widely (5 – 50 ng/μl, various degrees of fragmentation). Failed PCR reactions (e.g. one primer missing) were identified by analysing the standard samples. Failed individual samples (e.g. due to low DNA concentrations or failure of the fragment analyses run) were easily identified by missing or very low and out-of-range peaks. Nevertheless, the proportion of samples that had to be retyped was well below 1%. Discussion The PRNP ARMS genotyping method has several advantages. It is based on a one step reaction using competitive allele discrimination. The fragments can be analysed on an automated sequencer that allows data to be directly transferred to a database. The reaction proved to be highly specific and no false negative or positive results were observed yet. Furthermore, it is robust with respect to variations of DNA quality and, since no further purification or reaction is necessary, the costs for consumables are low. The ARMS method allows the determination of partial (136-154 and 136-171, compare Fig. 1 ) prnp haplotypes, thereby facilitating the detection of 'complex' prnp genotypes that do not match one of the 15 standard genotype patterns. If necessary, it can easily be extended to complete haplotype determination by adding additional labelled forward primers for position 154 (for haplotype 154-171). All methods currently available for SNP typing are inherent sensitive to nucleotide changes within the primer attachment or restriction sites. The resulting mismatches can cause inconclusive typing results or null alleles, as frequently observed with microsatellite loci. Therefore, it is crucial to be aware of this phenomenon and to take measures to reduce the risk of mistyping. The primers for PRNP ARMS were carefully checked against a prnp in-house database containing 16 published and 11 unpublished (mainly from rare German and Spanish breeds) alleles. No known polymorphism interfere with the attachment of the selected ARMS primers. Nevertheless, the recently described rare polymorphism K-171 [ 19 ] with unknown effect to scrapie resistance is not included in the standard PRNP ARMS set (that gives Q-171 as typing result for this allele) and requires one additional specific oligonucleotide. This oligonucleotide should be added if the respective breeds (mainly hair breeds [ 19 ]) are genotyped. When further polymorphisms associated with resistance to scrapie will be identified appropriate oligonucleotides can easily be added. Unlabelled additional oligonucleotides can be used when SNPs upstream of the amino acid position 136 shall be typed. Nevertheless, the inclusion of positions downstream of position 136 would require additional labelled oligonucleotides. A third fluorescent dye might be useful especially when the lengths of the fragments interfere with on of the other alleles. The PRNP ARMS method is highly flexible with respect to scale and instrumentation. It is easily adoptable from low- to medium-throughput typing system using identical reaction conditions. In principle, all primer combinations can even be performed as single reactions that can be analysed on agarose gels (data not shown). Nevertheless, the reliability of genotyping results improves when competitive reaction conditions and commonly available DNA sequencing machines are used. Since all reactions can be performed in microtiter plates, the ARMS reaction can easily be adopted for a pipetting robot. Together with the one step reaction and the automated data transfer, the risk of cross-contamination or interchanging of samples is minimized. The runtime of 18 min results in 25 min analyses time per sample on the ABI 310 and could be optimised by using shorter capillaries. When using the standard run conditions on the smallest, single capillary sequencer the throughput is limited to about 55 samples per working day but can be scaled up to more than 5000 samples per day by using a 96-well capillary sequencer. That should be sufficient even for large, existing breeding programs. PRNP ARMS can be multiplexed with other typing systems (microsatellite or SNP), since it is analysed using standard fragment analysis technique. Moreover, it is suitable for further applications as e.g. allele frequency estimation from pooled DNA [ 20 ] for investigating rare breeds. Conclusions An easy and robust one-step prnp typing method has been established that is universally suited for a broad range of typing projects with different requirements. This typing method was developed by optimising ARMS primers and combining these with standard fragment length analyses. The method provides an efficient and inexpensive diagnostic mutation analysis that can be implemented by most molecular genetic laboratories using standard equipment. It should contribute to reliable and economic genotyping of the ovine prnp by the many smaller labs throughout Europe. Competing interests The authors declare that they have no competing interests. Authors' contributions J.B. designed the project and protocols involved, performed analysis of results and drafted this manuscript. J.S. conceived the ARMS reaction, fragment analysis and DNA sequencing. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517712.xml
545054	Can the concepts of depression and quality of life be integrated using a time perspective?	Background Little is understood about the conceptual relationship of depression and quality of life (QoL). Judgments concerning both, implicitly or explicitly, involve a time perspective. The aim of this study was to test de Leval's theoretical model linking depression and QoL with a time perspective. The model predicts that changes in cognitions about one's past, present and future QoL, will be associated with changes in depressive symptomatology. Methods Eighteen psychiatric in-patients with a clinically confirmed diagnosis of depression were assessed on commencing treatment and 12 weeks later. QoL was assessed by the Schedule for Evaluation of Individual Quality of Life (SEIQoL), depression by the Beck Depression Inventory (BDI-II) and hopelessness by the Beck Hopelessness Scale (BHS). Time perspective was incorporated by asking QoL questions about the past, present and future. Results Depression and hopelessness were associated with a poorer present QoL. Depression lowered present QoL but did not alter future QoL, as these remained consistently high whether participants were depressed or recovering. However, depressed individuals had a larger gap between their actual present QoL and future (aspired to) QoL. Changes in QoL were influenced by depression and hopelessness. Contrary to the model, perception of "past" QoL was not affected by depression or hopelessness. Conclusions de Leval's model was largely confirmed. Thus depression and hopelessness influence a person's present and future QoL. The analysis of a temporal horizon was helpful in understanding the link between depression and QoL.	Background Assessment of quality of life (QoL) has become increasingly important in health care, particularly as an evaluative method to measure outcomes of the impact of disease and interventions. To date it is unclear how research on QoL relates to other psychological constructs such as depression and anxiety. Many clinical studies assess a number of related psychosocial dimensions but without a theoretical basis for the unique contribution of each. On an intuitive level, QoL and depression can appear as opposing phenomena – crudely representing all the positive and negative aspects of well-being. Poor QoL is sometimes seen as a consequence of depression [ 1 - 3 ]. On the other hand, poor QoL may also be a precursor to depression. In other formulations, depression can be seen as a component of QoL. Whatever the implicit models of their interrelationships, there has been little theoretical attention or research to understand the relationship between depression and QoL. A theoretical approach developed by de Leval [ 4 , 5 , 8 ] tries to capture and highlight possible relationships between depression and QoL in a "three-time-dimension" theory. This theory links depression and QoL on a timeline of the past-present-future. Time can be perceived objectively and subjectively. It consists of three dimensions: past, present and future. The presence of psychopathology, e.g. depression has been found to influence time perception. For instance, individuals who are depressed have been reported as finding that time passes more slowly [ 11 ]. In comparison to others, depressed individuals also have a temporal focus, which is less future directed and more focused on the past. The "three-time-dimension" theory describes the dislocated temporal horizon of the depressed patient. It situates both depression and quality of life as part of a continuum in time rather than as independent phenomena. De Leval proposes that, for the depressed individual, time passes slowly, the present is dissociated from the past and the potential for the future is lost or viewed with hopelessness [ 4 ]. During the course of their depression, individuals want to go back to their past when things were perceived as better. This searching for the past becomes the individual's future. According to de Leval, depressed individuals have two pasts: the actual past when they were well and the past as a position they wish to regain or aspire to, i.e. the 'therapeutic future', when things were better than they are in the present. In the proposed model by de Leval, in accordance with DSM-III-R criteria, depression is referred to as "ill-being". Ill-being is the current or present state of the patient experiencing depression. De Leval uses the term depression for "phenomenological-depression", i.e., depression as perceived by the individual in question and it is placed on the past-present timeline. She proposes that this "phenomenological-depression" is related to the perception of a gap between a healthy past and a present ill-being. The greater the gap between past and present, the greater the phenomenological-depression. In de Leval's theory, QoL is perceived as the gap between actual experience and future aspirations. Whereas depression is placed on the past-present timeline, QoL is placed along the timeline using present and future. QoL according to this model is defined as being "the appropriateness of future aspirations to the present" or "the making present of the future". The larger the gap, the lower the QoL of the individual (Figure 1 ). Although not included explicitly in de Leval's theory, the concept of hopelessness as described by Beck [ 2 ] is worth considering given its timeline focus on negative evaluations of the future. Figure 1 Illustration of de Leval's model In a cross-sectional study, de Leval [ 5 ] examined the three-time-dimension theory in a group of 110 clinically depressed psychiatric patients. They completed a 30-item questionnaire – the Three-Time-dimension Synoptic Scale (3TSS – French version) developed by the author [ 6 ]. Questions were chosen to reflect the content of existing mental health scales and each question asked about their feelings now, their feelings in the past (before being depressed) and what they wanted to feel in the future. Findings indicated preliminary support for the theory. No other study has to our knowledge tested Leval's theory empirically. The present study sought to advance the understanding about the conceptual relationship of depression and QoL by empirically testing de Leval's model in a longitudinal study. It also sought to assess the model using a previously developed individualised system for assessing QoL. If de Leval's proposals concerning the centrality of the temporal assessment are correct, then any QoL assessment instrument which measures the present can be adapted to measure the past and the future as aspired to. In this study, the model was operationalised as follows: scores on standardised depression measures were taken to demonstrate an individual's level of depression, i.e., 'present ill-being'. QoL was the gap between an individual's present status (' present ill-being ' in the case of depression) and anticipated status (i.e ' therapeutic future ') and was measured by the discrepancy between present and future actual QoL scores. The level of phenomenological depression – i.e. the gap between a person's perceived past and present " ill-being " was measured by the discrepancy between the past and present actual QoL scores. These gaps are referred to as time comparisons gaps. The full model is displayed in Figure 1 . In addition to components of de Leval's model, aspirational QoL was measured and assessed for all three time periods: past, present and future. This was included to provide further information about the gap between where an individual is and where he/she would like to be. These gaps are referred to as preference comparisons gaps. It was hypothesized that the greater the gap between actual and aspirational scores at any time (past/present/future), the worse the QoL and vice versa. This reflects Calmans definition of QoL. He defines QoL as the gap between actual QoL and preferred QoL [ 3 ]. As an additional measure of an individual's perception of the future hopelessness was included . The model was tested at two time points: when individuals were clinically depressed (time 1) and approximately three months later when their progress towards recovery could be estimated (time 2). The following hypotheses were tested in order to validate de Leval's model and additional components: Hypothesis 1: The size of the gap between actual and past QoL scores is larger when patients are depressed, in comparison to when they are less/not depressed. Hypothesis 1a: The therapeutic future is the view of a healthy past. Therefore scores for past and future QoL are equal. Hypothesis 2: The change in depression scores (present "ill-being") from time 1 to time 2 predicts the change in the gap between actual and aspirational QoL scores. Hypothesis 3: The change in hopelessness scores from time 1 to time 2 predicts the change in the gap between actual and aspirational QoL scores. Hypothesis 4: A reduction in depression narrows the gap between the past and present, and enhances QoL by narrowing the gap between present and future QoL. Methods Consecutive psychiatric in-patients (N = 27) were approached to participate in a longitudinal study of depression. Study inclusion criteria were age ≥ 18 years, diagnosis of moderate to severe uni-polar depression (DSM-IV-criteria), a moderate or severe depression rating on the Beck Depression Inventory II [ 1 ], no cognitive disabilities (Mini Mental State Exam, [ 7 ]). Exclusion criteria were diagnosis of severe psychopathology or borderline personality disorder, more than two previous episodes of depression that were of more than three months duration and history of prolonged substance abuse or use of psychotropic medication, which would impede participation in a research interview. Patients were recruited as soon after admission as was deemed appropriate by medical staff. Instruments Quality of life (QoL) QoL was assessed by the Schedule for the Evaluation of Individual Quality of Life (SEIQoL) [ 12 ], a well-established method of assessing QoL which incorporates the value system of the individual participant. To reduce interview demands for depressed participants, the shorter direct weighting (SEIQoL-DW) procedure was applied [ 9 ]. In the first step of the standard SEIQoL procedure, participants are asked to nominate the five areas of their life (cues) that are most important to them. In the second step, participants rate their current status or level of functioning on each cue. Ratings are against a vertical axis anchored from "worst possibly" to "best possibly". The final assessment step in SEIQoL involves quantifying the relative importance (weight) of each cue to the judgment of participants' overall QoL. This is obtained by using a weighting disk, consisting of five interlocking coloured disks that can be rotated around a central point to form a pie chart. These disks are labelled with the five nominated cues and are adjusted by participants until they are satisfied that the proportion of the pie chart displayed by each life area accurately reflects the relative importance they attach to these life areas. The SEIQoL Index score is calculated by multiplying each cue weight with the relevant level of functioning. These five scores then are summed. The scale of the SEIQoL Index score ranges from worst possible (0) to best possible (100). This was done for each time anchor and point. Time perspective To measure the temporal nature of the gap in terms of perceptions of their actual and aspirational past, present and future QoL according to the Leval's theory and Calman's definition of QoL, participants were asked to rate the status of each nominated SEIQoL cue (i.e. how were you doing?) based on the following questions: Past: - Where were you before you got depressed? (actual past) - Where would you have liked to have been? (aspirational past) Present: - Where are you now? (actual present) - Where would you like to be? (aspirational present) Future: - Where do you expect to be in a year’s time? (actual future) - Where would you like to be in a year’s time? (aspirational future) Thus scores were taken for actual present, aspirational present, actual past, aspirational past and actual future, aspirational future. Depression The Beck Depression Inventory (BDI), a 21 item self-report instrument for measuring the severity of depression in adults, was developed for the assessment of symptoms corresponding to criteria for diagnosing depressive disorders based on the DSM-IV. In this study an updated version is used (BDI-II) [ 1 ]. The BDI-II requires about 10 minutes to complete and has a two week time-frame including today. Items are rated on a four point scale ranging from 0 to 3, with higher scores reflecting more severe depression. A cut-off score of 20 was used. The psychometric properties have been well established [ 1 ]. Hopelessness The Hopelessness Scale [ 2 ] is a 20 item self-report instrument designed to assess pessimistic expectations. Items are rated true (1) or false (0), with a higher score indicating hopelessness. Internal consistency (split-half reliability) exceeds .90 in a range of samples and concurrent validity has been established [ 2 ]. Results Of 27 patients approached, 24 gave written consent and took part in the study (89% response rate). Over the three month follow-up period, six patients dropped out (75% follow-up rate). Information on the complete sample of 18 patients was analysed. Sociodemographic and clinical details are presented in table 1 . Table 1 Sociodemographic and clinical profile of recruited sample (N = 24) Mean ± SD % Min Max Gender (% male) 50 Age (years) 36.7 ± 14.4 18 72 Marital status Married/partner 42 Single 54 Separated 4 Occupation: Employed 50 Unemployed/retired 16 Student 21 Homemaker 13 Residence (% urban) 71 Time since hospital admission (days) 5 ± 2.7 1 9 No. of previous episodes of depression 0.7 ± 0.7 0 2 Duration of previous episodes of depression (weeks) 2.6 ± 3.6 <1 12 Baseline scores of the depression measure (BDI-II) showed high levels of severity, which were reduced significantly after three months (p < .001, table 2 ). While at baseline all 18 patients scored above 20, only six patients remained above this threshold at three months follow-up. The mean hopelessness score (BHS) of 10 at baseline dropped significantly by 5 (p < .002). Table 2 Profile of mental health at baseline and three month follow-up (N = 18) Baseline three months Mean SD Mean SD t p Depression BDI-II 34.0 9.5 15.0 11.4 4.8 <.001 Hopelessness BHS 10.0 5.8 5.0 4.9 3.2 .002 The five most commonly mentioned cues for the SEIQoL at time 1 were: mental health, family of origin, work marriage/relationship, friends and leisure. There were no new cues introduced at time 2. This is consistent with previous research on cue profiles across varying populations. The actual present level of overall QoL was very low at baseline (time 1, see table 3 ). This increased significantly over three months. Aspirational level of present QoL (where one would like to be now) was significantly higher at both baseline and three months then the actual present level. QoL did not change significantly over time (p = .102). Table 3 Quality of life (QoL) over time and actual and aspirational "gaps" at baseline and three months measured by SEIQoL Index Score Actual t p Aspirational t p t p Present Baseline 35.4 ± 19.1 88.2 ± 11.4 10.5 <.001 3.2 .002 1.3 .102 three months 61.1 ± 22.8 88.5 ± 8.5 5.5 <.001 Past Baseline 70.2 ± 13.7 87.0 ± 10.5 4.8 <.001 .2 .411 1.5 .07 three months 68.7 ± 18.8 83.7 ± 8.4 4.4 <.001 Future Baseline 66.8 ± 21.0 90.8 ± 10.5 5.1 <.001 .3 .362 .2 .396 three months 74.4 ± 23.6 91.5 ± 7.5 3.6 .001 Actual past level of overall QoL did not change from baseline to three months later. Aspirational past QoL (where one would have liked to be then) was significantly higher at baseline and three months than actual past and this did not change over time (table 3 ). Actual future QoL did not change significantly over the three months follow-up. Actual future QoL was significantly lower at baseline and three months than the aspirational future QoL. Aspirational future QoL was consistently high over time. The gap between actual and aspirational QoL is presented in table 4 . The biggest gap was perceived between baseline actual present level of QoL and aspirational present QoL. This large gap was significantly reduced after three months. This confirms hypothesis 1. Other gaps remained unchanged (figure 2 ). Correlations between the preference comparison QoL gaps for present, past and future assessments with the severity of depression (d) showed significant high correlations for the present (p) and future (f) gap at baseline (1) and three months (2) (r pd1 = .468, r fd1 = .475, both p < .05, r pd2 = .815, r fd2 = .730, both p < .01). In addition, correlations between the QoL preference comparison gaps for present, past and future assessment and hopelessness (h) were statistically significant at baseline for the present (p) gap and at three months (2) for present (p) and future (f) gaps (r ph1 = .567, r ph2 = .586, r fh2 = .422, all p < .05). There was no difference between actual past and actual future QoL scores at baseline and three months (p = .475, p = .528). This confirms hypothesis 1a. Table 4 Differences between actual and aspirational QoL "gaps" at baseline and three months as measured by the SEIQoL Index Score Time Baseline actual/aspirational gap (M ± SD) Three months actual/aspirational gap (M ± SD) F Time Present 52.7 ± 21.0 27.6 ± 21.2 9.21 .007 Past 16.8 ± 14.7 14.9 ± 14.9 .154 .699 Future 23.9 ± 19.6 17.0 ± 17.9 2.86 .109 Figure 2 Illustration of the preference comparison gaps (actual/aspirational QoL) for the three time points (past, present and future) at baseline and three months. Δ represents the change from baseline to three months for the time anchor "present" p < .001 Multiple regression was conducted to investigate whether depression and hopelessness contribute to the change in the size of the gap of actual and aspirational QoL over time. Changes in depression and hopelessness scores over time were entered as independent variables in a multiple regression, while the change in gap scores was entered as the dependent variable. A significant amount of variance was explained by changes in the present and future gaps for actual and aspirational QoL (52.4%, 59,9% respectively both p < 0.01). Change in depression contributed to change (reduction) in the gap between present actual and aspirational Qol (β =.676). Change in hopelessness (β =.589) was the significant contributing factor for change (reduction) in the gap between actual and aspirational future QoL (table 5 ). This confirms hypothesis 2 and 3 in parts. Table 5 Multiple regression analyses to explain the change in the actual/aspirational gap by change in Beck Depression Inventory (BDI) and Beck Hopelessness Scale (BHS) change scores from time 1 to time 2. Δ in actual/aspirational Gap Variable name 1 R 2 p-value β p Present .524 .004 Δ BDI 12 .676 .020 Δ BHS 12 .063 .813 Past .043 .720 Δ BDI 12 .053 .888 Δ BHS 12 .165 .661 Future .599 .001 Δ BDI 12 .230 .352 Δ BHS 12 .589 .026 1 Change of Beck Depression Inventory (BDI) and change of Beck Hopelessness Scale (BHS) were entered as predicting variables in these analyses In table 6 , the differences of present and past (phenomenological-depression) and present and future QoL (quality of life) are assessed according to de Leval's model. As already reported, baseline actual present QoL scores were considerably lower than actual past scores (35.4 ± 19.1 vs. 70.2 ± 13.7, p <.001). This significant gap (34.8 ± 22.5) between present and past QoL was significantly reduced over three months to a gap of 7.6 ± 30.9 (p = .011). At the three month follow-up, there was no statistical difference in the gap between actual present and past QoL perception (figure 3 ). The gap between present and future QoL was large and significant at baseline (31.3 ± 27.2, p < .001). Differences remained significant at the 3 month follow-up (13.3 ± 13.5, p = .001) with higher QoL levels for the future. However, the reduction in the gap over time was also significant (p = .010). Table 6 Testing de Leval's model: time gap comparison for actual QoL scores measured by SEIQoL Index Score QoL at Baseline t p QoL at three months t p Past 70.2 ± 13.7 68.7 ± 18.8 6.5 p < .001 1.0 .310 Present 35.4 ± 19.1 61.1 ± 22.8 Present 35.4 ± 19.1 61.1 ± 22.8 4.8 p < .001 4.2 .001 Future 66.8 ± 21.0 74.4 ± 23.6 Figure 3 Illustration of the time comparison gaps (past-present/present-future QoL; shaded areas) for the three time points (past, present and future) at baseline and three months. Δ represents the change from baseline to three months for the time anchor "past" and "future" p < .001; ns ... not significant To investigate how depression and hopelessness may influence the change in the size of the gaps proposed by de Leval, multiple regression analysis was conducted. Change in depression and hopelessness scores over time were entered as independent variables in a multiple regression, while the change in the gap over time, was entered as the dependent variable. Changes in the size of the two time comparison gaps (past/present; present/future) were not influenced by change in depression or hopelessness scores (R 2 = .281, p = .09; R 2 = .162, p = .266). This partly disconfirms hypothesis 4 Discussion In this study the relationship of QoL and depression was investigated in two ways. First the preference comparison gaps were assessed, i.e. where patients perceive themselves to be and where they wish to be. This was done for all three time dimensions (past, present and future). Secondly, the time comparison gap was assessed, i.e. where patients see themselves now in comparison to the past, and in comparison to the future. Both approaches were carried out on the basis of de Leval's model linking depression to all three time dimensions. Our findings are consistent with previous results, showing that people use their current affective state as a basis for making judgments of how happy and satisfied they are with their lives. A depressed person will usually see his or her well-being, social functioning and living conditions as worse than they appear to an independent observer or to patients themselves after recovery – the so called "affective fallacy" [ 10 ]. Clinically depressed patients were assessed at two time points before and three months after admission to a psychiatric hospital. Clinical measures (depression and hopelessness) showed significant improvements for all 18 patients, indicating successful treatment. Alongside the reduction in depression, actual present QoL improved significantly over time. In contrast, actual past QoL did not change over time. Successful treatment of depression had no influence of the individual's perception of the past. In addition, there was no significant change in actual future QoL, i.e. how patients realistically estimated their QoL to be one year from now. Thus assessments of the past and of the expected future did not change when patients recovered from a depressive episode. Aspirational QoL was the same between the three time dimensions and remained constant over time. There was always a preference comparisons gap over time – independent of depression or recovering depression. Our findings suggest the larger the present preference comparison gap, the greater the depression. While the preference comparison gap of the past did not change over time with the decrease of depression, there was a trend for the future preference comparison gap that might become statistically significant with a larger sample size. However, aspirations were not affected by depression, since they remained the same for past, present and future from baseline in hospital to three months later. Overall these findings may indicate that a reduction in the present preference gap should be the main goal for therapeutic actions for depressed patients. An achievable target may be to keep the same preference comparison gap between and over time points to maintain a healthy homeostasis and a good QoL (figure 2 ). Interestingly, the mean level of aspirational QoL was not at the top of the possible scale (100). This may reflect an individual "QoL set-point" as discussed by Seligman [ 14 ]. He proposes that an individual might be genetically predetermined, to a particular set-point, to which he/she might return after both ups and downs. In therapy then the goal might be to help an individual actively work to return to his/her set point as soon as possible. Successful treatment of depression, indicated by a significant change of BDI-II and BHS scores, was accompanied by a significant reduction in the gap between present actual and aspirational QoL. A reduction in the gap between actual present QoL with the aspirational QoL ("where one would like to be right here and now") reflects a more positive view of the present and is accompanied by significant improvements in QoL. Depression was consistently correlated with the time comparison gap between present and future QoL assessments. However, depression was more highly associated with the present and future preference comparison gap at three months. This suggests, that depression may be more highly associated with the actual/aspirational gap when people are less depressed. In contrast, hopelessness was consistently correlated with the preference comparison future gaps and only at baseline with the gap of present actual/aspirational QoL. The gap between the actual past and aspirational past QoL was not explained by either depression or hopelessness. Change in depression or hopelessness was important in reducing the preference comparison gap for present and future QoL assessments. Change in depression was important in reducing the size of the present gap. Change in hopelessness significantly contributed to reducing the size of the future gap. Given that present "ill-being" is seen as depression, while hopelessness is future orientated and directed, this finding makes sense. This has implications for therapy: to improve the QoL of patients with depression, relief from depression must be provided. Hopelessness needs to be tackled in addition to achieving a realistic future QoL perspective. To a large extend this study confirms de Leval's theory. The assessment of de Leval's model: present and future QoL (quality of life) and past and present (phenomenological-depression) showed interesting results. The theoretical time comparison gap between a "healthy past" and present "ill-being" was confirmed by the significant difference between patients' perceptions of actual present and future QoL, at baseline. Over time, following appropriate treatment, this time comparison gap should close and QoL should improve, if the treatment is considered to be successful. The significant reduction of the past-present gap over time, and the loss in significance in size of this gap three months later, further confirms de Leval's model. This reduction of the time comparison gap between the present "ill-being" and a "healthy past" resulted in an improvement in QoL. The reduction was achieved by a change in the actual view of the present, leaving the view of the past unchanged (figure 3 ). According to de Leval's model, recovery from depression is achieved when the appraisal of present and past QoL are very close, i.e. a very narrow gap. This means that a reduction of the gap between present and future QoL should improve current QoL. Although this study showed a decrease in depression over time by standard measures of depression and the size of the time comparison gap was significantly reduced between the two time points, the size of the gap did remain significant between the present and future at three months follow up. This finding may either indicate that the patients were still depressed at three months, or this part of the model could not be confirmed (figure 3 ). de Leval argues that therapeutic future is the view of a healthy past. Therefore one would expect similar scores for past and future QoL. This was supported by our findings for both time points. The change in the size of the time comparison gaps was not predicted by the change in either hopelessness or depression scores over time. This suggests that the change in time comparison gap's may be influenced by other variables, not assessed in this study. Depression lowered a person's actual QoL, but not their aspirational QoL. This was consistent whether a person was depressed or improving and for each of the time dimensions (past-present-future). The recollection of the past, either actual or aspirational did not change for the depressed or recovering person, neither did their view of the future. Thus for this patient group, the perception of the past and the future was not altered by depression status, which might be linked to individual QoL set points. Limitations Since we tested the model in a small and specialised sample, the findings are not generalisable. However, our study did test de Leval's model in a clinically depressed sample which is the focus of the most likely benefit of understanding the interplay of the two phenomena of depression and quality of life. It is also a challenging group to recruit, engage with and maintain in a research project. Conclusion This study showed that depression influenced individual QoL by lowering the person's actual QoL. Thus depression was associated with a larger gap between current reality (actual perceptions) and patient aspirations and realistic expectations. Aspirations for past, present and future QoL remained the same over time. However, patients actual appraisal of their present QoL improved with successful treatment of depression, reflected by a closure of the present preference gap. de Leval's model was largely confirmed. Thus depression and hopelessness influence a person's present and future QoL. The analysis of a temporal horizon was very helpful in understanding the link between depression and QoL. Therapeutic interventions can be considered as closing the gap between a person's present QoL and their present aspirations and realistic expectations for the future. This could be achieved, according to the data presented here, by changing the view of the present and not necessarily by changing the level of expected future QoL. Knowing the persons aspirational QoL level may be a helpful guide for the therapist in that it provides important information about a depressed person's position in the progress towards what he or she considers to be recovery. Authors' contributions Margaret Moore contributed to conception and design, acquisition, analysis and interpretation of the data and revised the article. Stefan Höfer drafted the article and contributed to the analysis and interpretation of the data. Hannah McGee contributed to the conception and design and interpretation of the data and revised the article. Lena Ring contributed to the interpretation and revised the article. All authors have given final approval of the version to be published	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545054.xml
521075	Comparative genomics of cyclin-dependent kinases suggest co-evolution of the RNAP II C-terminal domain and CTD-directed CDKs	Background Cyclin-dependent kinases (CDKs) are a large family of proteins that function in a variety of key regulatory pathways in eukaryotic cells, including control over the cell cycle and gene transcription. Among the most important and broadly studied of these roles is reversible phosphorylation of the C-terminal domain (CTD) of RNA polymerase II, part of a complex array of CTD/protein interactions that coordinate the RNAP II transcription cycle. The RNAP CTD is strongly conserved in some groups of eukaryotes, but highly degenerate or absent in others; the reasons for these differences in stabilizing selection on CTD structure are not clear. Given the importance of reversible phosphorylation for CTD-based transcription, the distribution and evolutionary history of CDKs may be a key to understanding differences in constraints on CTD structure; however, the origins and evolutionary relationships of CTD kinases have not been investigated thoroughly. Moreover, although the functions of most CDKs are reasonably well studied in mammals and yeasts, very little is known from most other eukaryotes. Results Here we identify 123 CDK family members from animals, plants, yeasts, and four protists from which genome sequences have been completed, and 10 additional CDKs from incomplete genome sequences of organisms with known CTD sequences. Comparative genomic and phylogenetic analyses suggest that cell-cycle CDKs are present in all organisms sampled in this study. In contrast, no clear orthologs of transcription-related CDKs are identified in the most putatively ancestral eukaryotes, Trypanosoma or Giardia . Kinases involved in CTD phosphorylation, CDK7, CDK8 and CDK9, all are recovered as well-supported and distinct orthologous families, but their relationships to each other and other CDKs are not well-resolved. Significantly, clear orthologs of CDK7 and CDK8 are restricted to only those organisms belonging to groups in which the RNAP II CTD is strongly conserved. Conclusions The apparent origins of CDK7 and CDK8, or at least their conservation as clearly recognizable orthologous families, correlate with strong stabilizing selection on RNAP II CTD structure. This suggests co-evolution of the CTD and these CTD-directed CDKs. This observation is consistent with the hypothesis that CDK7 and CDK8 originated at about the same time that the CTD was canalized as the staging platform RNAP II transcription. Alternatively, extensive CTD phosphorylation may occur in only a subset of eukaryotes and, when present, this interaction results in greater stabilizing selection on both CTD and CDK sequences. Overall, our results suggest that transcription-related kinases originated after cell-cycle related CDKs, and became more evolutionarily and functionally diverse as transcriptional complexity increased.	Background Cyclin-dependent kinases (CDKs) belong to a large protein family with 13 members described so far in human cells including CDKs1-11, along with PCTAIRE and PFTAIRE kinases named after conserved domain sequences [ 1 ]. These kinases are essential for cell cycle progression, and also are involved in control of transcription, DNA repair and post-mitotic cellular process [ 2 - 4 ]. Generally, CDKs1-6, PCTAIRE and PFTAIRE have been linked to cell cycle regulation, and CDKs7, 8 and 9 to control of RNA polymerase II (RNAP II) transcription [ 4 - 8 ]. The functions of CDKs10 and 11 have not been defined clearly, but recent research implicates them in coordination of transcription and RNA-processing [ 9 - 13 ]. Among the most important and broadly studied roles of CDKs in transcription is the reversible phosphorylation of the C-terminal domain (CTD) of the largest subunit (RPB1) of RNAP II. The CTD consists of multiple repeats of an evolutionarily conserved heptapeptide with the consensus sequence Tyr 1 -Ser 2 -Pro 3 -Thr 4 -Ser 5 -Pro 6 -Ser 7 [ 14 ]. The number of repeats varies among different organisms, ranging from 26–27 in yeast to 52 in mammals [ 15 , 16 ] with 8 repeats in yeast and 28 repeats in human cells required for viability [ 15 , 17 , 18 ]. Both biochemical and genetic evidence places the CTD in a central position in the 'mRNA factory,' where it functions as a platform for interactions with processing factors and other transcription-related proteins [ 19 , 20 ]. More than a passive scaffold, reversible phosphorylation of the CTD regulates the cycling of RNAP II between a hypophosphorylated (IIO) form, which is competent to enter the preinitiation complex, and a hyperphosphorylated (IIA) form capable of processive transcript elongation [ 21 ]. Throughout this cycle the CTD binds essential transcription-related proteins that help to regulate gene expression, promote efficient elongation, and effectively couple transcription to pre-mRNA processing [ 19 - 24 ]. To date at least five of the CDKs (CDK1, 2, 7, 8 and 9) have been shown to phosphorylate the CTD in vitro ; they all have been referred to as 'CTD kinases' [ 25 - 28 ]. Both CDK7 and CDK8 are found tightly associated with the pre-initiation complex and are involved in transcriptional regulation [ 29 ]. The CDK9 subunit of P-TEFb (positive transcription elongation factor b) induces hyper-phosphorylation of the CTD and stimulates elongation. Unlike CDKs 7, 8 and 9, which have demonstrated interactions with the CTD in vivo , CDK1 and CDK2 are primarily cell-cycle related kinases [ 4 ]. CDK2 has been characterized functionally only human and Drosophila in mammals and its role in Tat-dependent HIV-1 transcription is still unclear [ 27 , 28 ]. Although phosphorylation of yeast RNAP II by CDK1 (CDC2) can inhibit transcription in vitro , the role of the CDK1 in mRNA synthesis in vivo is not, as yet, clearly understood. It has been proposed as a candidate for mitotic RNAP II inactivation by inhibition of CDK7 CTD-kinase activity [ 26 ]. In animals and yeasts, interactions between the CTD and CTD-specific kinases have become a focal point of biochemical and genetic investigations of RNAP II transcription and transcription-linked mRNA processing [ 25 , 26 , 30 ]. However, the ancestry and evolutionary relationships among CTD kinases have not been investigated thoroughly. Evolutionary analyses of the RNAP II CTD show that canonical CTD heptads are strongly conserved only in a subset of eukaryotic groups. In evolutionary trees based on RPB1 sequences, all eukaryotic groups in which the CTD is strongly conserved appear to be descended from a single common ancestor (descendents of this ancestor have been referred to as the "CTD-clade") [ 31 ]. The reasons for differential conservation of the CTD have not been clarified, nor have evolutionary correlations been established between strong conservation of CTD structure and the presence of essential CTD/protein interactions. In addition, although the functions of various CDKs are reasonably well characterized in mammals and yeasts, very little is known for most other eukaryotes, and the overall evolution of CDKs has been investigated only in animals and yeasts [ 32 ]. Therefore, a comparative evolutionary study also can provide clues as to which CDK orthologs, and presumably CDK functions, are present over a broad range of eukaryotic diversity. Here we present a comparative genomic analysis of CDKs, using complete genomes from members of the "CTD clade" (animals, plants, yeasts and Microsporidia), as well as from other diverse eukaryotic organisms lacking a canonical CTD ( Trypanosoma , Plasmodium and Giardia ), to explore the evolutionary relationships between the CTD and CTD kinases. We also provide a phylogenetic distribution of CDKs from a wide range of organisms, suggesting new hypotheses regarding the emergence and evolution of different members of the CDK family. Results We identified 133 CDK family members, 123 from animals, plants, yeasts, and four protists from which genome sequences have been completed, and 10 additional CDKs from incomplete genome sequences of organisms with known CTD sequences (Table 1 ). Although all of sequences are included in our supplemental phylogenetic analysis ( additional file 1 ), only 101 of them are included in the major phylogenetic analysis (Fig. 1 ); a large plant-specific amplification of CDK9-like kinases (the phylogenetic weight of these sequences disrupts the CDK9 sub-clade) and sequences from incomplete genomes are excluded (see Fig. 1 and additional file 1 legends for further explanation). The nomenclature for kinases from Arabidopsis followed Joubès et al. (2000) and Vandepoele et al. (2002) [ 33 , 34 ] (Table 1 ). The catalytic core base, Gly-rich motif and T-loop, required for characterized CDK function, appear to be conserved across all defined and putative kinase sequences analyzed ( additional file 2 ). The 50% majority rule consensus tree of 4,000 likelihood trees, sampled from the posterior probability distribution from Bayesian phylogenetic inference, is shown in Figure 1 . This tree provides strong support for grouping a number of previously uncharacterized CDKs, from a variety of organisms, with defined CDKs from animals and yeast. Overall, however, very little support is found for relationships among different CDK orthologous groups. Table 1 CDK-related kinases used in this study. Species Genes gi number Abbreviations Trypanosoma brucei Cdc2-related kinase2 397162 TbCrk2 Cdc2-related kinase3 397365 TbCrk3 Cdc2-related kinase6 23392965 TbCrk6 Cdc2-like kinase 10458 TbCdc2L Leishmania major Cdc2-related kinase 1 9857049 LmCrk1 Cdc2-related kinase 3 15526337 LmCrk3 Giardia lamblia Cdc2-like1 29248279 GlCdc2L1 Cdc2-like2 29245850 GlCdc2L2 Cdc2-like3 29250990 GlCdc2L3 Cdc2-like4 29249431 GlCdc2L4 CAKlike 29249713 GlCAKlike Cryptosporidium parvum Cdc2-like kinase 3329529 CpCdc2L Plasmodium falciparum MO15-related kinase 23507945 PfMrk PK5 23619490 PfPk5 PK6 23618947 PfPk6 Crk1 23510162 PfCrk1 Crk3 23509994 PfCrk3 Crk4 23957709 PfCrk4 Dictyostelium discoideum Cdc2 kinase 167686 DdCdc2 Cdc2-related protein 167696 DdCrp Cdk7 1705721 DdCdk7 Cdk8 15778146 DdCdk8 Cdk9-like kinase 28828850 DdCdk9L Entamoeba histolytica Cdc2 kinase 543971 EhCdc2 Guillardia theta Cdc2 kinase 13812042 GtCdc2 Saccharomyces cerevisiae Cdc28 115915 ScCdc28 Pho85 295932 ScPho85 Kin28 1199540 ScKin28 Cdk8/Srb10 2131219 ScSrb10 Ctk1 486235 ScCtk1 Bur1 218486 ScBur1 Cak1 1480663 ScCak1 Schizosaccharomyces pombe Cdc2 173359 SpCdc2 PhoA 19075421 SpPhoA Mcs6 19113141 SpMcs6 Cdk8/Srb10 7493197 SpSrb10 AC2F3.15 19115305 SpAC2F3.15 Cdk9 32363142 SpCdk9 Csk1 299548 SpCsk1 BC18H10.5 3006177 SpBC18H10.5 Encephalitozoon cuniculi Cdc2-related kinaseA 19173516 EcCrkA Cdc2-related kinaseB 19069621 EcCrkB Cdc2-related kinaseC 19171093 EcCrkC Cdc2-related kinaseD 19074929 EcCrkD Cdc2-related kinaseE 19173349 EcCrkE Cdk7 like kinase 19068706 EcCdk7 Drosophila melanogaster Cdk1 115921 DmCdk1 Cdc2c 7708 DmCdk2 Cdk4 1523997 DmCdk4 Cdk5 1523999 DmCdk5 Cdk7 1336061 DmCdk7 Cdk8 1718193 DmCdk8 Cdk9 24658274 DmCdk9 Dcdrk 541654 DmDcdrk CG6800 23171908 DmCG6800 Pitslre 1524005 DmPitslre CG7597 24668136 DmCG7597 EiP63E 1524003 DmEip63E Caenorhabditis elegans K03E5.3 3158523 CeK03E5.3 Cdk1 5001728 CeCdk1 Cdk4 21902501 CeCdk4 Cdk5 5001732 CeCdk5 Zc123.4 21913082 CeZc123.4 Pctaire1 5001730 CePctaire1 Cdc2-like kinase5 7494824 CeB0385.1 Cdk7 5031478 CeCdk7 Cdk8 32563668 CeCdk8 Cdk9 17507939 CeCdk9 B0495.2 2499649 CeB0495.2 Zc504.3 897712 CeZc504.3 H01G02.2 7504821 CeH01G02.2 Homo sapiens Cdk1 115922 HsCdk1 Cdk2 29849 HsCdk2 Cdk3 4557439 HsCdk3 Cdk4 33304135 HsCdk4 Cdk5 7434324 HsCdk5 Cdk6 21885467 HsCdk6 Pctaire1 13623189 HsPctaire1 Pctaire2 21542571 HsPctaire2 Pctaire3 30583437 HsPctaire3 Pftaire1 6912584 HsPftaire1 Cdk7 13529020 HsCdk7 Cdk8 1000491 HsCdk8 Cdk9 12805029 HsCdk9 Cdk10 6226784 HsCdk10 Cdk11 16357492 HsCdk11 Cdc2-Like kinase5 10443222 HsCdc2L5 Cdc2-related kinase with RS domain 7107392 HsCrkRS Cell cycle related kinase 23344742 HsCCRK Oryza sativa CdkA.1 20343 OsCdkA.1 CdkA.2 266410 OsCdkA.2 CdkB2.1 7489567 OsCdkB2.1 CdkB1.1 34907628 OsCdkB1.1 R2 231707 OsCdk7 CdkE 12039362 OsCdkE CdkC.1 31442141 OsCdkC.1 OJ991113_30.14 38344237 OsCAD41330 B1015E06.16 34903661 OsB1015E06.16 P0560B06.11 34914693 OsP0560B06.11 P0453E05.113 28460677 OsP0453E05.113 P0450A04.129 34899281 OsP0450A04.129 P0498H04.21 42408343 OsP0498H04.21 P0435E12.11 46390990 OsP0435E12.11 P0482D04.8 34907029 OsP0482D04.8 OJ1562.H01.5 38424086 Os1562.H01.5 Arabidopsis thaliana CdkA1 30693081 AtCdkA.1 CdkB1.1 30694007 AtCdkB1.1 CdkB1.2 42569740 AtCdkB1.2 CdkB2.1 30699181 AtCdkB2.1 CdkB2.2 18394928 AtCdkB2.2 CAK1 15235518 AtCdkF CAK2 15147864 AtCdkD.3 CAK3 15147866 AtCdkD.1 CAK4 20521156 AtCdkD.2 CdkE 10177042 AtCdkE CdkC.1 30698081 AtCdkC.1 CdkC.2 11346412 AtCdkC.2 F12B7.13 17065202 AtF12B7.13 K9H21.7 17064770 AtK9H21.7 K9L2.5 15241455 AtK9L2.5 T22H22.5 25405751 AtT22H22.5 T12H1.1 15229881 AtT12H1.1 K16E14.2 26449318 AtK16E14.2 F21B7.1 7488248 AtF21B7.1 AT4g22940 15235867 At4g22940 F8L10.9 15219169 AtF8L10.9 F26A9.10 42572067 AtF26A9.10 AT4g10010 30681286 At4g10010 F14J9.26 18391043 AtF14J9.26 F6A14.22 15221833 AtF6A14.22 F1M20.1 25406336 AtF1M20.1 AAF21469.1 6649591 AtAAF21469.1 T4P13.34 42570106 AtT4P13.34 Note: The sequences in bold are the additional sequences from incomplete genomes and uncharacterized CDK9 like-kinases from Arabidopsis and Oryza included in supplemental phylogenetic tree (additional file 1). Figure 1 Unrooted 50% majority consensus tree from 4,000 ML trees sampled from the Bayesian posterior probability distribution. Support values are shown above the internode from Bayesian inference/distance bootstrap respectively. Only values above 50% are reported and values under 50% are indicated by (-). 100% values are indicated by (+). CDK names in blue are from organisms that fall into the "CTD-clade" in RPB1 phylogenetic analyses (see Fig. 2); and those in red are from groups in which the CTD is not strongly conserved. Inferred groups of CTD-directed CDKs 7, 8 and 9 are shown in bold. A large group of unidentified CDKs from Arabidopsis and Oryza , which appear to represent a plant-specific amplification of CTK9, were excluded from this analysis to determine whether identified plant CDK9s show a specific phylogenetic affinity to either the BUR1 or CTK1 subgroup. All identified plant sequences are included in an expanded analysis shown in additional file 1. In this unrooted tree the highly diversified cell-cycle kinases defined in humans, CDKs1-6, fall into a large cluster with 69% Bayesian support. This grouping includes CDKs from all organisms examined in the study. Among these putative cell-cycle CDKs, some plant and protistan kinases can be assigned with reasonable confidence to specific CDK groups. For example, apparent orthologs of human CDK1 are found in other animals ( Drosophila and Caenorhabditis ), yeasts, both plants ( Arabidopsis and Oryza ), Encephalitozoon and Giardia (Fig. 1 ). Likewise, putative orthologs of CDK5 were identified in all organisms examined, except for the two plants (Fig. 1 ). A number of other sequences, such as TbCrk2 and 3 from Trypanosoma , cluster with cell-cycle kinases but not clearly with any specific CDK family. Significantly, and consistent with the results of Liu and Kipreos (2000) [ 32 ], CDK5 and PCTAIRE-like kinases from fungi and animals form a strongly supported group, indicating their close relationship (Fig. 1 ). In contrast to cell-cycle kinases, our phylogenetic results failed to identify a clear ortholog of any transcription-related CDKs from two of the complete genomes examined, Trypanosoma brucei and Giardia lamblia . This includes strongly supported clades of presumed orthologs of human CDKs7-11 respectively. A well-defined CDK7 family is recovered, including sequences from yeasts, the microsporidian, plants, and animals. These are the primary groups that make up the "CTD-clade," in which the RNAP II CTD is invariably conserved (Fig. 2 ). CDK7 shows an interesting sister relationship to HsCCRK from human and apparent orthologs from Drosophila , Caeorhabditis and Arabidopsis . In Arabidopsis , four possible CDK7 orthologs were found, as reported previously by Shimotohno and colleagues (2003) [ 35 ]; however, AtCdkF (CAK1) is quite divergent from the core CDK7 family and related specifically to HsCCRK in our analyses. PfMRK from Plasmodium , suggested previously to be a CDK7 [ 36 ], does not fall within the well-defined CDK7 group, but clusters with another Plasmodium kinase. The a priori hypothesis that PfMRK belongs in the core CDK7 group is strongly rejected with our data set in a likelihood paired-sites test. Figure 2 Hypothesis of RNA polymerase II evolution inferred from phylogenetic analyses of RPB1 sequences conserved regions A-H. The tree displayed, after Stiller and Cook [60] had the highest likelihood of all trees sampled from the posterior probability distribution in 10 6 generations of Bayesian inference. Organisms with genomes included in this study are in larger/bold font, and whether each of the three primary CTD kinases (CDKs7,8,9) are present in this genome, as inferred from global phylogenetic analyses and distributions of CDK kinases (see Fig. 1), are indicated next to the name. Names in blue indicate the presence of repeated heptads at the RPB1 C-terminus, which includes several from protistan organisms that do not conform to the consensus sequence or known structural requirements of the canonical CTD [60]; names in red have no tandem-heptapeptide structure whatsoever. The node supporting a "CTD-clade," in which the consensus sequence and repetitive structure of the CTD are invariably conserved, occurred in 98% of the 8000 trees sampled from the Bayesian posterior probability distribution. See references 31 and 60 for a more complete phylogenetic treatment of the origin and conservation of the CTD. Likewise, GlCAKlike (gi: 292497120) has been proposed as a CDK7 from Giardia , based on nearest sequence similarity to Kin28 in a more limited comparison to CDK sequences from fission yeast [ 38 ]. In our expanded analyses of CDKs from 11 completed genomes, we find no evidence supporting an orthologous relationship to CDK7 for this, or any Giardia sequence. The a priori hypothesis that GlCAKlike belongs in the core CDK7 group also is strongly rejected in a likelihood paired-sites test. A robust CDK8 family is recovered with strong support values in both distance bootstrap and Bayesian inference. Like CDK7, this family includes putative orthologs only from members of the "CTD-clade," specifically yeasts, animals and plants. Although the microsporidian Encephalitozoon is a member of the RNAP II "CTD clade," TBlastN searches of the complete genome of Encephalitozoon found six CDKs but none show a phylogenetic affinity to CDK8. A CDK9 grouping also is supported as monophyletic with representative CDKs from yeasts, Encephalitozoon , animals, plants and Plasmodium . This group is divided into two well-defined sub-clades. One of them consists of BUR1 from yeast along with CDK9 orthologs from animals; the other contains CTK1 from yeast, CDC2L5 and CrkRS from human, and apparent orthologs from Drosophila and Caenorhabditis , both plants, and Plasmodium . A putative CDK9 also is found in Encephalitozoon , but falls at the base of the larger CDK9 grouping and does not associate clearly with either subgroup (Fig. 1 ). Plants also contain a large number of putative CDKs that show strong phylogenetic affinity to CDK9 ( additional file 1 ). These kinases appear to represent a plant-specific amplification of CDK9, although their functions have not been determined experimentally. Human CDK10 and CDK11 group with apparent orthologs from other animals, plants, fission yeast, and PfCRK1 from Plasmodium . Once again, no kinases from either Trypanosoma or Giardia show any phylogenetic affinity to this group. Discussion A suggestion of co-evolution between the RNAP II CTD and CTD kinases At least five CDKs have the capacity to phosphorylate RNAP II CTD repeats in vitro or in vivo [ 25 - 28 ]. Kinases that appear to be related closely to CDK1, which can phosphorylate the CTD in vitro , are present in all organisms sampled; however, it is not clear that CDK1 interacts with the CTD in vivo or is essential for CTD function. CDK2 was found only in human and Drosophila and, based on our analyses from a diverse group of eukaryotes, appears to be derived from within a larger CDK1 family. In any case, according to its restricted phylogenetic distribution, possible CTD/CDK2 interactions cannot explain the conservation of CTD structure in diverse members of the "CTD-clade." Evolutionary investigations of RPB1 sequences show that canonical CTD heptads are conserved strongly in only a subset of eukaryotic groups, all apparently descended from a single common ancestor [ 31 ]. This "CTD-clade" is composed of animals, plants, fungi, and related protistan groups, including microsporidians, chytridiomycetes, choanoflagellates and slime molds (Fig. 2 ). A handful of organisms that do not fall inside the "CTD-clade" do have tandemly repeated C-terminal heptads. For example, RPB1 from Plasmodium falciparum contains a short set of seven tandem C-terminal repeats. Based on codon usage and comparative alignment with sequences from other Plasmodium species, these heptads are best explained by a recent tandem duplication of a single heptad motif in P. falciparum or its immediate ancestor [ 31 ]. No other apicomplexan RPB1 contains tandemly repeated heptads, nor does the nearest evolutionary relative of the apicomplexans (Fig. 2 ). Although vestigial or convergent heptad repeats are found in a few organisms scattered across the eukaryotic evolutionary tree, strong stabilizing selection on CTD structure appears to be restricted to those eukaryotic lineages found in the "CTD-clade" (Fig. 2 ). In our analyses of CDKs, members of this "CTD-clade" are precisely the same eukaryotes to which clear orthologs of CDK7 and CDK8 are restricted. When sequences recovered from additional but incomplete eukaryotic genomes are included in phylogenetic analyses, distribution of these two kinases remains tightly correlated with strong conservation of canonical CTD repeats (see additional file 1 ). Moreover, unlike CDK1, the primary characterized function of both of these kinases is to mediate RNAP IIA/IIO cycling through reversible phosphorylation of CTD residues [ 19 - 24 ]. Taken together, these findings suggest that the RNAP II CTD has undergone a co-evolutionary process with CDK7 and CDK8. If phylogenetic results based on CDK and RPB1 sequences reflect evolutionary history, the inference of a "CTD-clade" in both sets of analyses suggests that CDK7 and 8 originated as part of a major shift in the mechanics of RNAP II transcription in the ancestor of the "CTD-clade" [ 31 ]. It was in that ancestor that reversible phosphorylation of the CTD became a central organizing principle for regulating the transcription cycle, and laid the foundation for more complicated mechanisms of transcriptional control in these organisms. Such a profound shift in the mechanics of RNAP II transcription would explain why the CTD is conserved so strongly in members of the CTD-clade, but not in many other eukaryotic lineages [ 31 ]. In this scenario, other known or putative CTD kinases (certainly CDK1 and apparently CDK9) originated before canalization of a CTD-based RNAP II transcription cycle, and were adapted later as CTD kinases. It also is possible that the co-evolution inferred from comparisons of the phylogenetic distribution of RPB1 and CDKs7/8 does not reflect the pattern of evolutionary history but, instead, results from functional constraints driven by CTD/CDK interactions. Both GlCAKlike from Giardia and PfMrk from Plasmodium have been suggested previously to be orthologs of CDK7 [ 36 , 38 ]; these hypotheses are rejected strongly by our phylogenetic analyses. Assuming these kinases really are CDK7s, then their failure to cluster with other orthologs must be due to phylogenetic artifacts, frequently referred to as "long-branch attraction" [ 39 ], that can be common when rates of evolution vary dramatically among sequences. The large amounts of sequence divergence of PfMrk and GlCAKlike from other CDK7s, along with a complete degeneration of the CTD in Giardia species and apicomplexans as a group, are unlikely to coincidental. It is possible that those organisms retaining a RNAP II transcription cycle mediated by CDK7 and 8 kinase activity form distinct clades, in both RPB1 and kinase derived trees, because both sets of proteins share parallel modes of evolution driven by their physical interactions. In this case, the observation of co-evolution between the CTD and CTD-directed kinases need not have a phylogenetic basis, only a functional one. Most putative CDKs from Giardia and Trypanosoma , and several from Plasmodium , do not associate strongly with any established CDK family. It is reasonable to assume that at least some of these kinases are orthologs of defined CDK groups, but have diverged to the point that they are not recognizable using sequence-based phylogenetic methods. Although such a scenario may have disturbing implications for the use of these methods across broad evolutionary distances, particularly when functional interactions among sequences are unknown or poorly understood, it cannot be ruled out as an explanation for our observations. Analyses of additional genomes from diverse eukaryotes are required, both to verify our observations of co-evolution between the CTD and CTD-directed kinases, and to determine its bases. General evolutionary trends in the CDK family Kinases from protistan organisms In an effort to understand the broader evolutionary history of CDKs, three deep-branching protists with complete genomes, Plasmodium falciparum , Trypanosoma brucei and Giardia lamblia , were included in our study. Our Blast searches detected 15 putative kinases from these protists; six from Plasmodium falciparum , four from Trypanosoma brucei and five from Giardia lamblia (Table 1 ). The phylogenetic positions and orthologous relationships of these kinases generally are not well defined by phylogenetic analyses (Fig. 1 ). Four of them (GlCdc2L3, GlCAKlike, TbCrk6 and PfCrk4), along with two microsporidian kinases (EcCrkB and EcCrkD) branched close to ScCak1 and SpCsk1, cyclin-activating kinases from yeasts. All of these sequences are highly divergent, and it is difficult to determine, whether their branching positions are due to a phylogenetic artifact or a phylogenetic relationship. As noted above, GlCAKlike kinase has been proposed as a Giardia CDK7 ortholog based on JTT distance data [ 38 ], a relationship not supported by our broader phylogenetic analyses. Moreover, there are no experimental data reported on the functions of any of these kinases. Other putative protistan CDKs, GlCdc2L4, PfMrk and PfPk6, scatter among CDKs from other organisms, but with no statistical confidence for any implied relationship. Our most strongly supported results indicate that six of these kinases (TbCdc2L, TbCrk2 and 3, PfPk5, GlCdc2L1 and L2) belong to cell-cycle related kinase families CDK1 and CDK5. In particular, PfPk5 is well-supported as an ortholog of CDK5. In addition, two kinases from Plasmodium (PfCrk1 and PfCrk3) appear to be transcription-related kinases, PfCrk1 groups with the CDK10/11 family, and PfCrk3 with CDK9. The phylogenetic distribution of protistan kinases indicate that cell-cycle related kinases are present, or at least their functions are more strongly conserved (see discussion above regarding CTD/CDK co-evolution), in a more diverse array of eukaryotes than are transcription-related kinases. This pattern also is seen in a more widely-sampled analyses including CDKs from a number of organisms with incompletely sequenced genomes, including Dictyostelium discoideum that has a canonical RNAP II CTD, and Leishmania major , Cryptosporidium parvum and Entamoeba histolytica , which all lack a CTD (see additional file 1 ). Thus, the overall results suggest that cell-cycle related kinases are more ancient than transcription-related kinases, and probably ancestral to them, and that their core functions are more similar across the broad diversity of eukaryotic lineages. It will be interesting to see whether these preliminary hypotheses are supported as more genomes are sequenced completely, particularly from diverse protistan organisms. Cell-cycle related kinases Our analyses support well-defined groups for cell-cycle kinases CDK1, CDK4/6 and CDK5. An ortholog of either CDK1 or CDK5 is found in all of the organisms in our study, and these two families appear to be closely related. TbCrk3 was proposed as a functional homolog of CDK1 in Trypanosoma [ 40 ]; here it groups among cell-cycle kinases, but is not specifically related to CDK1. CDK4/6 appears to be present only in human, Drosophila and C. elegans . The CDK5 family has undergone expansion in metazoans, including PFTAIRE and PCTAIRE kinases, and putative orthologs of CDK5 are detected in Plasmodium , Trypanosoma and Giardia . Interestingly, no CDK from plants associates strongly with the CDK5 group, while the CdkB-type kinases, which are specific to plants, branch as sister to a broader CDK1/CDK5 clade. Our overall results suggest that cell-cycle kinases have undergone extensive and independent evolutionary diversification in different eukaryotic lineages, and it may be difficult to classify many of them based on orthologous relationships in phylogenetic analyses. It may be that functional homologies, once established experimentally, will prove to be more consistent criteria for designating CDK groups. The CDK7 family Clear orthologs of CDK7 from animals, plants, yeasts and Microsporidian are strongly supported as a core family, with CDK-activating kinase from Arabidopsis (AtCdkF), and its apparent orthologs from animals, branching as a sister group. In addition to their role as CTD kinases, members of the CDK7 family in plants, animals and fission yeast can function as a CDK-activating kinase (CAK) [ 41 , 42 ]. Unlike animals and yeast, however, four CDK7-like of CAKs were isolated from Arabidopsis [ 35 ]. AtCdkF (AtCAK1), which groups with human CCRK and apparent orthologs from Drosophila and Caenorhabiditis , exhibits only CAK activity but no CTD kinase activity. Consistent with the phylogenetic relationships recovered in our analysis, human CCRK and other animal orthologs were recently shown to have CAK activity [ 43 ]. In contrast, AtCdkD3 (AtCAK2) and AtCdkD2 (AtCAK4) display both CAK and CTD kinase activity and, along with a single CDK7 from rice, are included in a strongly supported CDK7 clade. Interestingly, and despite its high sequence similarity to AtCdkD3, no kinase activity was reported from AtCdkD1 (AtCAK3) [ 35 ]. Apparently CAKs in Arabidopsis have diversified substantially, and may be regulated in different ways from those in yeast, animals, and even rice. ScCAK1 and SpCSK1 from yeasts also have CAK activity; however, despite their functional similarity to kinases in the CCRK group, they do not group with animal or plant CAKs (Fig. 1 ). Interestingly, in the single most likely tree recovered in our expanded Bayesian analysis of 133 sequences, ScCAK1 and SpCsk1 group with other CAKs in the sister clade to CDK7 ( additional file 1 ); however, there is no support for this placement in the Bayesian probability distribution. ScCAK1 and SpCSK1 sequences are highly divergent from all CDKs, and the regulation of CAK activity in yeast is very different from that of animals and plants [ 42 , 44 ]. Thus, alternative lines of evidence may be required to determine whether there is any specific evolutionary relationship among all CAKs. The CDK 8 family CDK 8 (SRB10 in yeast) is a component of the multi-subunit Mediator complex, which transduces signals from cis regulatory elements to RNAP II; it is proposed to inhibit transcription initiation by phosphorylation of the CTD. CDK8/SRB10 and its partner cyclin C/SRB11, together with SRB8 and SRB9, form a specific sub-module that is variably associated with the RNAP II holoenzyme, and potentially with the free mediator complex [ 45 ]. Apparent orthologs of CDK8 form a well-defined group, including sequences from plants, animals and yeasts. Interestingly, although a member of the CTD clade (Fig. 2 and note that all microsporidian RPB1 genes isolated to date encode a CTD), no ortholog of CDK8 was identified from Encephalitozoon . Our further blast results (unpublished data) failed to identify any of the units of the CDK8/SRB10 (SRBs8-11) sub-module in the Microsporidia suggesting a loss of CDK8/SRB10 unit from these highly reduced parasites. Although the CDK8/SRB10 sub-module has been implicated in negative regulation of transcription by phosphorylation of TFIIH, leading to the inhibition of the TFIIH CTD kinase and transcription [ 46 ], the exact mechanism still is unclear. Recent research shows that the Mediator containing this sub-module is isolated only in free form, not associated with RNAP II. In contrast, Mediator lacking this sub-module associates with the polymerase [ 47 ]. There also is experimental evidence that negative Mediator-RNAP II regulation by the SRB8-11 sub-module is evolutionarily conserved from yeast to humans [ 47 ]. Therefore, the absence of identifiable components of the SRB8-11 sub-module in Encephalitozoon suggests CDK8/SRB10 function is absent from the Microsporidia. The loss of CDK8 from Microsporidia, along with absolute conservation of CDKs7 and 9 in all members of the "CTD-clade" (Figs. 1 and 2 ) implies that interactions between the CTD and Mediator complex are less strongly entrained into essential RNAP II function, than are those regulated by TFIIH and P-TEFB kinase activity. The CDK 9 family CDK9 is a component of the P-TEFb complex, which is a positive-acting RNAP II transcription elongation factor [ 48 , 49 ]. Research has focused on P-TEFb from animals and budding yeast. A definitive yeast homolog of animal P-TEFb has not yet been determined from functional studies, but two candidates have emerged: the BUR1 complex and the CTDK-I complex [ 26 ]. Based on our blast and phylogenetic analyses, BUR1 and CTK1 (subunit of CTDK-I complex) are found in two distinct but related kinase groups, each with orthologs from other eukaryotes. BUR1 is identified as the specific ortholog of CDK9 from metazoans, budding yeast and probably the Microsporidia. Unexpectedly, the CDC2-like5 kinases and CrkRS from animals are highly supported as orthologs of CTK1 from yeasts. Although their functions are not yet clear [ 50 ], our results suggest that human CDC2-like5 kinases and CrkRS have CDK9 function. Recent analyses of CrkRS (CDC2-related kinase with an RS-rich domain) suggest that it has CTD kinase activity and helps to link transcription directly to intron splicing [ 51 ]. This CTK1 clade also contains putative CDK9 (CdkC) kinases from plants and as well as a CDC2-like kinase from Plasmodium (PfCRK3). The latter is the only apparent ortholog of a CTD-directed kinase (CDKs 7, 8 or 9) identified in our analyses from any organism outside the "CTD-clade." It remains to be determined whether PfCRK3 possesses the P-TEFb function of CTK1, since it is the only protistan sequence present in either CDK9 sub-group, and the RNAP II CTD has not been conserved in apicomplexans or their closest relatives (Fig. 2 ). In addition to the two previously identified copies of CDK9 (CdkC1 and CdkC2) from Arabidopsis , and one from Oryza (CdkC1) [ 33 , 34 ], our Blast searches also retrieved a large group of CDK9-like sequences (14 from Arabidopsis and 8 from Oryza ) (Table 1 ). These kinases are annotated as "Cdc2-like" in databases and some of them also were identified in previous analyses of CDK evolution [ 38 ]. With one exception (Os1562.H01.5), all of these kinases group in a single cluster, with 100% support, and as sister to previously identified CDK9s of Arabidopsis and Oryza ( additional file 1 ). Os1562.H01.5 (Gi: 38424086) from Oryza is extremely similar to OsCdkC1 and very likely a second copy of CdkC (CDK9) from Oryza . There is no evidence of biological functions for these kinases as yet, but our results indicate that they are part of a large CDK9 complex specific to plants. The CDK10/11 family In this group, orthologs of CDK10 are found only in human and Drosophila , while CDK11 occurs in human, Drosophila and Caenorhabditis . Three putative CDK11 orthologs were found in plants (two from Arabidopsis and one from Oryza ). CDK10 has been implicated in the regulation of the G2/M phase of the cell cycle [ 52 ], but a cyclin partner has yet to be defined. Only one protein associated with CDK10, ETS2 transcription factor, has been identified so far, suggesting a link to transcription [ 9 ]. CDK11 associates with cyclin L as a partner, and is a proposed component of a signaling pathway that helps to coordinate transcription and RNA-processing events [ 10 - 13 ]. The close relationship between the CDK10 and CDK11 may reflect evolutionary and/or mechanistic similarities, but neither kinase family has been well characterized functionally. In addition, BC18H10 from S. pombe and PfCRK1 from Plasmodium show close relationships to the CDK10/11 family, but no function has yet been determined for these kinases either. Conclusions The apparent co-evolution between the CTD and certain CTD-specific kinases suggests an explanation for strong stabilizing selection on CTD structure in some eukaryotes, and its complete degeneration in others. Based on the genomes examined in this study, either the origins of CDK7 and CDK8 in an unknown ancestor of the "CTD-clade," or the canalization of reversible phosphorylation of the CTD in some eukaryotic groups but not others, could account for the variation seen in RPB1 C-terminal structure. In either case, once thoroughly "locked" into RNAP II function, the CTD must have recruited other transcription and processing related proteins into a growing machinery of the "transcriptosome" [ 53 ]. Our results suggest that was the case for several CDKs that clearly predate the canalization of CTD-based RNAP II transcription; further genomic analyses are underway to look for other protein-protein interactions that could be responsible for strong evolutionary conservation of the CTD in members of the "CTD-clade." This work also provides a new perspective on the overall evolution CDKs and evolutionary relationships among kinase families. Our combined genomic and phylogenetic analyses suggest that transcription-related kinases originated later than cell cycle-related CDKs. Finally, our results point to potential functions for a variety of previously uncharacterized kinases, based on their apparent orthologous relationships to defined CDKs. Additional completed genomes, particularly those from broadly diverse protists (especially non-parasitic forms), will be critical to address these questions further. Such comparative analyses will be invaluable in helping to guide experimental studies, which ultimately are required to verify the functional properties of each putative CDK. Methods Identification and alignment of protein sequences Representatives of all previously identified CDKs from budding yeast and human were obtained from Genbank, and used as probes in TBlastN and PSI-Blast [ 54 ] against the National Center for Biotechnology Information (NCBI), and additional specific complete genome databases, with an absolute cut-off of E<0.001. To confirm the identities of putative CDKs detected by the TBlastN, each identified sequence was used as a query in reciprocal Blast searches, to verify that it retrieved the original query sequences, and global sequence alignments were performed to confirm putative homologies to CDKs, according to the CDC-related kinase characterized motifs that use CDK2 as the model [ 55 ]. Initially, a number of inferred protein sequences were grouped into six subsets according to clear similarities to specific CDK family orthologs. These subgroups first were aligned in CLUSTAL X [ 56 ], and the resulting sub-alignments then were aligned with each other and adjusted through visual inspection and comparison to the kinase alignment of Liu and Kipreos (2000) [ 32 ]. Regions that could not be aligned reliably were excluded from subsequent phylogenetic analysis. The resulting alignment included 233 positions including gaps (See additional data file 2 and 3 for the original and final aligned matrices used in this study). Phylogenetic analysis Maximum-likelihood (ML) estimates of substitution parameters were made with the program TREEPUZZLE-50 [ 57 ] assuming a mixed model for variation among sites, with one category for invariable sites and a four-category discrete approximation to Γ-distribution, and the JTT weighting matrix for probability of change among amino acids. Further analyses were performed in MrBayes 3.0 b4 [ 58 ] using metropolis-coupled Markov chain Monte Carlo analysis. Four simultaneous Markov chains were run, also under an invariant + Γ rate model and a JTT substitution matrix. Four chains, one heated, were run for 500,000 generations, beginning with random a priori trees. Trees were sampled from the posterior probability distribution every 100 generations. The empirical burn-in required for likelihoods to converge was less than 100,000 generations; an additional 400,000 generations were run and the first 100,000 were excluded from analysis of Bayesian posterior probabilities. Thus, a total of 4,000 trees were examined to determine the 50% majority-rule consensus tree and Bayesian support values. In addition, 1000 distance (PROTDIST + NEIGHBOR) bootstrap replicates were performed in PHYLIP 3.573 [ 59 ], also using a JTT substitution model. Several a priori alternative hypotheses regarding CDK7 evolution were compared by KHT likelihood paired-sites tests [ 37 ]. Trees were constrained to require PfMRK from Plasmodium or GlCAKlike from Giardia , which previously have been characterized as a CDK7 orthologs [ 36 , 38 ], to group with the well-defined CDK7 clade. All most parsimonious trees retaining these constrained relationships were tested against the fully resolved Bayesian consensus tree to determine whether the a priori hypotheses of orthologous relationships to CDK7 were significantly worse than the Bayesian consensus tree. Authors' contributions ZG was primarily responsible for database searching and assembly of CDK genes. ZG and JWS performed phylogenetic analysis. ZG drafted the manuscript and figures and JWS contributed editorial revisions. All authors read and approved the final manuscript. Supplementary Material Additional File 1 The single most likely tree, with branch lengths, recovered from 16,000 ML trees in the posterior probability distributions of four separate iterations of Bayesian inference. Thirty-two additional sequences were added to this analysis, and are indicated in bold in Table 1. They represent CDKs identified in incomplete genomes of organisms from which CTD structure is known, as well as a large amplification of apparent plant-specific orthologs of CDK9 from Arabidopsis and Oryza . The phylogenetic weight of these latter plant sequences disrupts inferred relationships among CDK9 orthologs as shown in Fig. 1. Support values are from Bayesian inference and only values above 50% are shown. As in Fig. 1, CDK names in red are from groups in which the CTD is not strongly conserved, those in blue from members of the "CTD-clade." Inferred groups of CTD-directed CDKs 7, 8 and 9 are shown on the tree. Click here for file Additional File 2 Original protein sequence alignment. Click here for file Additional File 3 Edited protein sequence alignment. Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC521075.xml
533868	ESTIMA, a tool for EST management in a multi-project environment	Background Single-pass, partial sequencing of complementary DNA (cDNA) libraries generates thousands of chromatograms that are processed into high quality expressed sequence tags (ESTs), and then assembled into contigs representative of putative genes. Usually, to be of value, ESTs and contigs must be associated with meaningful annotations, and made available to end-users. Results A web application, Expressed Sequence Tag Information Management and Annotation (ESTIMA), has been created to meet the EST annotation and data management requirements of multiple high-throughput EST sequencing projects. It is anchored on individual ESTs and organized around different properties of ESTs including chromatograms, base-calling quality scores, structure of assembled transcripts, and multiple sources of comparison to infer functional annotation, Gene Ontology associations, and cDNA library information. ESTIMA consists of a relational database schema and a set of interactive query interfaces. These are integrated with a suite of web-based tools that allow a user to query and retrieve information. Further, query results are interconnected among the various EST properties. ESTIMA has several unique features. Users may run their own EST processing pipeline, search against preferred reference genomes, and use any clustering and assembly algorithm. The ESTIMA database schema is very flexible and accepts output from any EST processing and assembly pipeline. ESTIMA has been used for the management of EST projects of many species, including honeybee ( Apis mellifera ), cattle ( Bos taurus ), songbird ( Taeniopygia guttata ), corn rootworm ( Diabrotica vergifera ), catfish ( Ictalurus punctatus , Ictalurus furcatus ), and apple ( Malus x domestica ). The entire resource may be downloaded and used as is, or readily adapted to fit the unique needs of other cDNA sequencing projects. Conclusions The scripts used to create the ESTIMA interface are freely available to academic users in an archived format from . The entity-relationship (E-R) diagrams and the programs used to generate the Oracle database tables are also available. We have also provided detailed installation instructions and a tutorial at the same website. Presently the chromatograms, EST databases and their annotations have been made available for cattle and honeybee brain EST projects. Non-academic users need to contact the W.M. Keck Center for Functional and Comparative Genomics, University of Illinois at Urbana-Champaign, Urbana, IL, for licensing information.	Background Expressed sequence tag (EST) collections represent partial descriptions of the transcribed portions of genomes. They are generated from single-pass cDNA library sequencing that is often carried out by small or mid-size sequencing centers and research groups. The research may be aimed at transcriptome expression analysis using microarray, RT-PCR, or hybridization techniques. An increasing number of research centers are involved in sequencing multiple cDNA libraries. This is manifest by the exponential growth in the number of ESTs deposited to dbEST at NCBI (National Center for Biotechnology Information), and as of the May 21, 2004 update, that number is at over 21.2 million ESTs from 686 species. Small sequencing centers are faced with searching for cost-effective and convenient tools for EST management and querying, visualization, public and private user access, and functional classification. EST processing pipelines that transform raw chromatograms into high-quality filtered sequences, and software to cluster and assemble them into contigs are convenient to implement, and various tools are now available to researchers (StackPACK [ 1 ], ESTWeb [ 2 ], ESTAP [ 3 ], PipeOnline2.0 [ 4 ]). The TIGR gene indices [ 5 ], for example, are generated by performing an all-against-all pairwise similarity search of the ESTs, and then clustered via single-node transitive closure. The clusters are fed into CAP3 [ 6 ] for assembly. We have used the following assembly protocol for in-house sequencing projects such as cattle [ 7 ] and honeybee [ 8 ]. High quality sequences are pooled and run through BlastClust [ 9 ] to form clusters of similar sequences. Then the BlastClust output is run through CAP3. Commercial software (such as Paracel [ 10 ]) that is targeted for bioinformatics centers equipped with high-performance computing machines is also available. We describe ESTIMA (EST Information Management and Annotation) software that provides a database schema for management of raw and annotated ESTs, and is coupled with a suite of custom web-based tools that facilitate searching various aspects of ESTs and contigs, visualization, pairwise searching by BLAST [ 9 ], and functional classification based on the controlled vocabulary defined by the Gene Ontology (GO) Consortium [ 11 ]. ESTIMA accepts assembled sequences from any EST processing and clustering software, and has been equipped with password protection so that project researchers can use the ESTIMA tool confidentially with academic or industry collaborators. To form an association between a GO term and an EST, an organism that has already been annotated with the GO terms is selected. The ESTs are searched against this reference organism's annotated sequences, and the resulting alignments are used to ascribe putative function. The ESTIMA database schema provides users the flexibility to use any reference database that has GO term association, and any number of external databases, such as NCBI's non-redundant (nr) database or EMBL's Swissprot collections, to annotate ESTs and their contigs. This flexibility in the schema is critical given the increasing number of sequenced genomes, and specialized reference databases that researchers can download and integrate with existing annotations. Other database schemas that house EST project and analysis data have been reported [ 3 , 12 ]. But the inherent flexibility of multiple EST project management that ESTIMA affords, by allowing users to create multiple instances of the PROJECT schema within the GENOME schema (see section on databases), and allowing any number of reference genomes to be added, is unique to ESTIMA. The inputs to ESTIMA are the chromatograms, raw and processed ESTs, clusters and assembled EST contigs from any source, reference database annotations of ESTs, and contigs. All of these get loaded into the tables generated from the ESTIMA database schema and loading scripts. The raw and processed data, with their annotations, are made available to users through the ESTIMA query interfaces, for exporting, visualization, and further research. Implementation ESTIMA is composed of three major components: an ODBC-compliant database, loading scripts and a web application. Figure 1 shows the relationship of these components. Databases The heart of the system is a pair of database schemas. Figure 2 shows an ER diagram describing both schemas. The first schema, GENOME is common to all projects in the installation. It houses tables containing the GO structure, gene association information, annotation, and project security. GO terms are stored in two tables. Each GO term has a record in the Term table. The Term2Term table contains one record for each edge in the directional graph linking the terms. The graph can then be searched by starting with a term identifier (ID), finding all child IDs, then finding all records where each of the child terms appear as a parent. In this way, the entire graph can be searched with one call per tier below the original term. Term_Seq_Count contains the pre-calculated counts of ESTs associated with each term, and all the terms' child terms. Pre-calculation saves time at execution, and is handled by a Perl script that is rerun each time the GO tables are updated. The Gene_Association and Annotation table contains information that links the ESTs to the reference genomes. The Gene_Association table has one record for each reference gene that is associated with an EST or contig. The reference gene is then related to a GO term in the Term table. GENOME also contains the Blast_All table that holds information regarding the association between the ESTs and any project-specific DNA sequences. The ESTs are searched against sequences of interest, such as NCBI's nr, and the results are parsed and loaded into the Blast_All table. In addition to the common data, each project has its own schema that contains information such as DNA sequences, contig assemblies and paths to the chromatogram files. The PROJECT schema is centered on the Project table that contains a simple link between the other PROJECT tables and the GENOME schema. This organization has proven useful when the contents of the PROJECT schema are to be transferred directly from a sequencing facility to a Laboratory Information Management System (LIMS). Each project may have several different contig assemblies deriving from different assembly techniques. The assemblies are all housed in the Assembly_Project, Contig, Consensus_Link and Singlet tables. Each assembly has a single ID that is used by the web application to determine which is "live". After a new assembly is completed and loaded, it can be taken live by simply changing a single number in the web applications configuration file. Finally, the sequence data itself is stored in the DNA_Sequence table, while related GenBank accession numbers are stored in their own table. Although accession numbers are not required by the web application, if they are available, then the web application will accept them where a sequence ID would be used. Loading The second major component of the ESTIMA system is a series of Perl and JAVA applications used to parse and load data into the database. Because each project will have its own needs, the loading system was not automated. Instead, several dozen separate scripts and applications have been developed to allow researchers to manipulate and analyze large data sets using standard analysis applications. Web application The third component of ESTIMA is the web-application that queries and reports the EST information to the end-users. The ESTIMA web application is organized around seven points of entry into the system – the start screen and six query applications. Each query application interfaces with the databases, and allow users to query raw and annotated ESTs and contigs. ESTIMA supports password-protection, multiple-projects, and multiple libraries within a project. This is achieved with an XML configuration file with project-specific information that is called by the various applications when project-related information is needed. Figure 3 shows the relationship between the components of the web application. Their implementation is discussed in detail below. The start screen The start screen provides a convenient point of reference to the system. The page contains information about each library (genomic, cDNA, etc) in the project. All of the elements of the start page are contained within the configuration file. GO Browser The GO Browser (Figure 4 ) allows a researcher to start with a GO term, and find all ESTs associated with the term, and all of the descendent terms. The browser has a term search option that locates terms based on user-defined strings; for example, 'DNA%' will locate all GO terms starting with DNA. Once a term has been identified, the browser will provide a map of the GO tree both above and below any term, as well as a count of the ESTs associated with each of the terms displayed. For each term the browser also provides the option of either downloading the sequences of all ESTs and contigs associated with the term in a single FASTA file, or download a spreadsheet of the EST identifiers, associated GO terms and information about the linkage between the two. Alternately, the spreadsheet can be viewed as a web page. In this form the sequence identifiers will link to the appropriate ESTIMA page, ESTs to the Sequence ID page and contigs to the contig viewer. Further, the reference sequence used to form the association will link to the appropriate external web site. The GO Browser output can be filtered so that only ESTs from a single library are displayed. Sequence ID ESTs can be accessed directly from the sequence ID interface. More commonly, results from queries on annotations or contigs are dynamically linked to the EST sequence information. From these screens users can access chromatograms, and both raw and filtered sequences. Gene association Information about the association between the ESTs and the reference genome sequences can be accessed through these pages. BlastUI ESTIMA provides a Blast User Interface that allows FASTA-formatted sequences to be searched against the EST libraries. The libraries that are available for BLAST [ 9 ] can be defined for each project. We usually allow at a minimum, the raw sequences, the clean sequences, and the unique assembled sequences (contigs plus singletons). This allows researchers to rapidly identify those ESTs associated with any particular sequence of interest. The data sets available for BLAST searches can be easily extended for specific projects. For example, we make the Baylor University Honeybee Genome assemblies [ 13 ] available to the honeybee site [ 8 ]. Annotations The Annotation page is an optional page, the presence or absence of which is controlled from a configuration file. This page allows additional BLAST-derived annotations, beyond GO annotations, to be displayed and queried in a number of different ways. The web application creates an interface that allows users to query the Blast_All table directly. For example, songbird ESTs [ 14 ] were searched against the nr from NCBI, Swiss-Prot, and chicken database from TIGR. A researcher can use this to find all the ESTs and contigs in the songbird collection that are associated with any term in the sequence description. The term RNA binding, for instance, returns 38 hits, of which 19 are to nr proteins, 11 to Swiss-Prot, and 8 to TIGR chicken. Contig Viewer The Contig Viewer provides an image of the assembled contigs showing the relationship of the contig with the ESTs it contains as members. It provides the contig's consensus sequence, and links to each of the member ESTs. The web application maintains a common look and feel to all the pages within a project. This is implemented by having a single block of HTML stored within a configuration file. As each page is rendered, the same HTML is called to generate the header block of the page. Each project has its own HTML, so the look of each project's web site can be customized. The inter-connectivity of the different ESTIMA modules allows researchers to engage in a "free form" exploration of the data. Users can query the GO Browser, for example, to find a contig associated with a term of interest, drill down to see the structure of the contig, and then, if desired, drill down to get specific information on each EST in the contig. Just as easily, the users could take a sequence of interest from their research, and similarity-search it against the assembled ESTs. Then they could drill down on an EST that aligned with their sequence, and see the additional information about the sequence, including any annotations. Chromatograms ESTIMA allows users to view the actual chromatograms of the ESTs. Chromatogram files are stored in a file system visible to the web server, while file paths are stored in the database. When a user requests a chromatogram, Phred [ 15 , 16 ] is called to convert the chromatogram to SCF format and this is sent to Traceviewer [ 17 ] that displays the trace in the user's web browser. Results ESTIMA is independent of an EST processing pipeline ESTIMA is unlinked from the backend EST processing pipeline, clustering, and assembly of ESTs. It serves as a stand-alone web application that allows users to store, access, research, and visualize the raw and annotated ESTs and contigs, including GO annotation. The output from a sequencing center's EST processing pipeline (base-called high-quality ESTs, assembled contigs, BLAST results against a reference genome) gets loaded into databases, and serves as input to ESTIMA. (The W.M. Keck Center will gladly share the source code to its pipeline with any interested academic institution). Additionally, researchers may choose to use any reference genome to annotate their sequences. ESTIMA comes packaged with a flexible database schema that supports the linking of sequences to GO terms, and other user-supplied sequences. The flexibility of the ESTIMA database schema becomes more relevant with the increasing number of sequenced genomes. ESTIMA provides a flexible, password-protected, multi-project environment to researchers. It facilitates analysis of an unlimited number of ESTs and contigs linked to GO and non-GO annotations, and the download of annotated sequences related to any GO term. ESTIMA comes with an implementation of a GO browser that allows users to view the entire child term tree for any term, conveniently from a single query interface. There can be multiple installations of ESTIMA ESTIMA is designed to be a stand-alone application. Each installation of the web application has all system dependent information in its own configuration files, including the information needed to connect to the databases. We maintain two instances of the interface, a development and a production copy. As new projects are introduced, they can be tested, and any interface modifications that are needed can be perfected on the development machine, concurrent with the creation of the production version of the databases. The production database can be examined using the development web application and any errors corrected before taking the system "live" to the production version. ESTIMA maintains multiple projects and supports multiple libraries in a project ESTIMA is designed to accept new projects. As new EST sequencing projects are finished, they can be easily added to an existing ESTIMA installation. A new schema is created for the new project and the loading scripts are used to populate the database. All that is required to activate the web application is the addition of a block of XML code to the configuration file, and a connection string and some HTML to the system configuration file. XML tags within a configuration file control project- specific issues such as whether the data is password protected, or which BLAST databases can be accessed from the Blast User Interface. There are three public ESTIMA projects currently administered through the W.M. Keck Center for Comparative and Functional Genomics at the University of Illinois. The bee-ESTdb [ 18 ] is a resource created from a normalized unidirectional cDNA library from which 21,408 cDNA clones were partially sequenced. These sequences were assembled into 8,966 putatively unique sequences. The contigs were then tested for similarity to sequences in the Drosophila genome, and based on these similarities the sequences were tentatively assigned one or more molecular functions and biological processes. Likewise, BOVEST, the cattle EST database [ 19 ], contains 17,452 cDNAs from a bovine placenta library and 6,144 from a spleen library, all of which were annotated against the human UniGene [ 20 ]. The songbird project contains 14,461 sequences from a songbird brain library [ 14 ] that were assembled into 8,526 unique sequences. ESTIMA allows multiple libraries within each project. Information about each library is stored in the configuration file, and the interface elements dynamically generate the start page and the library filters within the GO Browser. Practical examples that demonstrate research utility of ESTIMA The presence of multiple projects in ESTIMA allows for efficient cross-tissue or cross-species homology searches. For example, a mouse brain EST may be used to interrogate the honeybee brain or songbird brain library to test the hypothesis that the gene is expressed as a brain-specific transcript. Thus, a mouse brain transcript, NCBI accession number BM875176, similar to human tubulin alpha-1 chain protein may be used to do a TBLASTX against the honeybee brain assembled ESTs from the BlastUI interface in ESTIMA. The BLAST results retrieve a significant hit, Contig2466, to the honeybee brain database (Figure 5 ). The contig identifier is hyperlinked to the Sequence ID interface in ESTIMA from where the consensus sequence may be downloaded. The chromatograms for the ESTs that make up the contig, may also be checked for quality from the same interface in ESTIMA as shown in Figure 5 . Both mouse and honeybee brain sequences, may then be used to do a deeper phylogenetic search with a BLASTX against non-redundant protein database to test the tissue-specificity hypothesis. ESTIMA projects, as compared to other public web-applications such as TIGR gene indices [ 5 ], allow access to singlets from the EST assemblies, and chromatogram retrieval. These singlets would include rare, novel transcripts and divergent homologs that are increasingly the sole motivation for a research project. Since ESTIMA includes only high quality sequences in the databases, users may search for and download these novel transcripts, and also efficiently implement a homology search strategy using the web-application. Another strength of ESTIMA is in facilitating chromatogram and contig viewing from a common interface (Sequence ID). Any contig may be displayed and chromatograms of the member ESTs checked for errors in base-calling that may result in a premature stop-codon, or frameshift indels. Thus, ESTIMA is a valuable genome research tool. Conclusions ESTIMA is a full-featured web-application and database, designed to simplify exploring and sharing EST libraries and databases. It can be easily adapted to a wide variety of system configurations, and back-end database engines. Our installation of ESTIMA easily supports three public projects, with five different EST libraries, and additionally a growing number of private projects. Availability and requirements ESTIMA is available free to academic users at . Under 'Downloadable Software' section of the web page, detailed installation instructions and a user manual have been included as well. ESTIMA is still in active development. New features are constantly being developed to meet the changing needs of the research projects that use it. Further, new projects are being added to our ESTIMA installation. The system has been written to facilitate its own change, and as such, researchers should find it approachable with a good working knowledge of Perl, SQL, and HTML. System requirements ESTIMA requires Perl and a database. All communication with the database is handled through Perl DBI, which is extensible to any ODBC compliant database. The use of Perl DBI and ODBC allow the databases to reside on separate servers from the web interface. Although many databases may be used, in practice, there are several complex joins in the code that could result in slow performance on large EST sets unless a well-optimized database was selected. ESTIMA requires certain additional Perl modules, specifically BIO, CGI, DBI, and GD. Bioperl, the BIO module [ 21 ] is used extensively. All BLAST and FASTA file parsing, as well as all references to sequence objects in the user interface are handled with BIO methods. GD [ 22 ] is used to generate the Portable Network Graphics (PNG) files displayed in the contig viewer. We have been using ESTIMA throughout its development. Our schemas are housed in an Oracle 8I database on a Silicon Graphics Origin 2000 16 × 250 MHz machine running IRIX 6.5.20. The web application, including the user requested BLAST jobs are run on a Sun Microsystems SunFire 280 R with dual 700 SPARK V9 CPUs. Authors' contributions GG and LL developed the database schemas. CGK, LL and LR developed the initial web application prototype. HAL managed project development and contributed to design concepts. RL modified the web application and optimized the code. All authors read and approved the manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC533868.xml
514569	The ECOS-16 questionnaire for the evaluation of health related quality of life in post-menopausal women with osteoporosis	Background The aim of this study is to validate the questionnaire ECOS-16 (Assessment of health related quality of life in osteoporosis) for the evaluation of health related quality of life (HRQoL) in post-menopausal women with osteoporosis. Methods An observational, prospective and multi-centre study was carried out among post-menopausal women with osteoporosis in primary care centres and hospital outpatient clinics. All patients attended 2 visits: at baseline and at 6 months. In addition, the subgroup of outpatients attended another visit a month after the baseline to assess the test-retest reliability. The psychometric properties of the questionnaire were evaluated in terms of feasibility, validity (content validity and construct validity) and internal consistency in baseline, and in terms of test-retest reliability and responsiveness to change in visit at month and visit at 6 months, respectively. In all visits, ECOS-16, EUROQoL-5D (EQ-5D) and four 7-point items about health status (general health status, back pain, limitation in daily activities and emotional status) were administered, whereas only outpatients were given MINI-OQLQ (Mini Osteoporosis Quality of Life Questionnaire), besides all clinical variables; and sociodemographic variables at baseline. Results 316 women were consecutively included, 212 from primary care centres and 104 from hospital outpatient clinics. Feasibility : 94.3% of patients answered all items of the questionnaire. The mean administration time was 12.3 minutes. Validity : factor analysis suggested that the questionnaire was unidimensional. In the multivariate analysis, patients with vertebral fractures, co-morbidity and a lower education level showed to have worse HRQoL. Moderate to high correlations were found between the ECOS-16 score and the other health status questionnaires (0.47–0.82). Reliability : internal consistency (Cronbach's α) was 0.92 and test-retest reliability (ICC) was 0.80. Responsiveness to change : ECOS-16 scores increased according to change perceived by the patient, as well as the effect size (ranges between 1.35 to 0.43), the greater the perception of change in patients' general health status, the greater the changes in patients' scores. The Minimal Clinically Important Difference (MCID) suggested a change of 0.5 points in the ECOS-16 score, representing the least improvement in general health status due to their osteoporosis: "slightly better". Conclusion ECOS-16 has been proven preliminarily to have good psychometric properties, so that it can be potentially a useful tool to evaluate HRQoL of post-menopausal women with osteoporosis in research and routine clinical practice.	Background Osteoporosis is characterised by low bone mass and a deterioration in bone tissue micro-architecture, leading to increasing bone weakness and consequently risk of fracture. The most common clinical complications of osteoporosis are hip fracture, vertebral deformity and wrist fracture. According to bone densitometry values, osteoporosis affects approximately 2 million women in Spain [ 1 , 2 ]. The most frequent symptom of osteoporosis is low back pain resulting from vertebral fractures. This pain can have a considerable impact on the ability to carry out usual activities of daily living. Patients are unable to work normally, are limited in their social and leisure activities, and may be severely affected emotionally [ 3 ]. To date, clinical trials on osteoporosis have been based on outcomes measured by imaging tests. But these measurements do not adequately reflect the extent to which the patient is affected in their usual daily activities, and are not appropriate to assess patients' disability and symptoms [ 4 ]. Nevertheless, recently some specific Health Related Quality of Life (HRQoL) questionnaires such as OPAQ (Osteoporosis Assessment Questionnaire) have been used as the main outcome in clinical studies on osteoporosis [ 5 ]. Several generic HRQoL questionnaires, such as the Sickness Impact Profile (SIP), SF-36 or the Nottingham Health Profile (NHP), have been used more frequently to assess the impact of osteoporosis on HRQoL [ 6 ]. These questionnaires are applicable to any population and disease, thereby enabling comparison between subjects suffering from different diseases. However, they have serious limitations given the fact that they fail to explore in detail the specific aspects of osteoporosis. For instance, some studies have shown that certain aspects, such as the fear of falling and suffering a bone fracture, the inability to adequately carry out domestic tasks, the ability to dress oneself adequately without help and despair about an uncertain future are all stressful for these patients [ 7 ]. These items are not included in generic questionnaires and their omission could lead to an incomplete or biased evaluation of HRQoL of patients with osteoporosis. Disease specific questionnaires for osteoporosis are available, such as the Osteoporosis Quality of Life Questionnaire (OQLQ) [ 8 ] or the Quality of Life Questionnaire of the European Foundation for Osteoporosis (QUALEFFO) [ 9 ]. However, their limited applicability due to their length and time for administration have restricted their use to clinical trials and highlighted the need for the development of questionnaires which are easier to administer in routine clinical practice. In order to expand their use in clinical practice, it is necessary to develop valid, short, easy to administer and comprehensible questionnaires. For this reason, a specific short form HRQoL questionnaire for women with osteoporosis was developed [ 10 ]. Its items were obtained from the Spanish versions of the OQLQ and QUALEFFO questionnaires, and were then reduced by using the Rasch analysis to obtain a total of 16 items; 12 from the QUALEFFO and 4 from the OQLQ. These questionnaires were selected because they were the only questionnaires already validated in Spain. More information regarding the development process of the ECOS-16 has been published elsewhere [ 11 , 12 ]. The aim of the present study is to evaluate the psychometric properties of the ECOS-16 in post-menopausal women with osteoporosis. Methods 316 post-menopausal women with primary osteoporosis attended in Primary Care Centres or in outpatient clinics were included in the study. Diagnosis was confirmed by a Bone Mineral Density (BMD) using Dual Energy X-Ray Absorptiometry (DEXA). Furthermore, outpatients should have at least one prevalent vertebral fracture confirmed by Genant's radiological criteria due to osteoporosis, a requirement which was not essential in patients from Primary Care Centres. 212 patients from 49 Primary Care Centres and 104 patients attended in outpatient clinics from 14 hospitals were consecutively selected and evaluated from March 2000 to August 2001. All patients attended two visits, a baseline and a follow-up visit 6 months after the inclusion. In order to evaluate test-retest reliability, outpatients were also attended in another follow-up visit one month after the baseline. All patients received the study information and gave their informed consent. Study design An observational, prospective and multi-centre study was carried out for the validation of the ECOS-16 in post-menopausal women with vertebral fractures due to osteoporosis in conditions of clinical practice. At the baseline, feasibility together with the content and construct validity of the ECOS-16 were evaluated. At the visit after 6 months, responsiveness to change with regard to ECOS-16 was evaluated. Outpatients also attended another visit a month after the inclusion so that the test-retest reliability of the ECOS-16 could be evaluated. In the baseline, data on the patients' sociodemographic characteristics (age, education level) and clinical variables (weight, height, body mass index, age at onset of menopause, BMD, presence and site of vertebral and non-vertebral fractures, concomitant chronic diseases and received treatment) were collected as well as the ECOS-16, the EUROQoL-5D and four 7-point items which refer to general health status, back pain, limitation in daily activities and emotional status. These items were used in a previous study and showed its validity [ 10 ]. Outpatients were also administered the Spanish version of the MINI-OQLQ questionnaire [ 3 ]. In the two follow-up visits, any modifications to the specific treatment for osteoporosis prescribed in the baseline, the number of concomitant treatments, and patients' withdrawals causes were registered. Moreover, the ECOS-16, EUROQoL-5D and the four 7-point change items were again administered. Outpatients were again administered the MINI-OQLQ questionnaire. In the present study, the EQ-5D and MINI-OQLQ questionnaires were used in order to assess the validity of the ECOS-16. Health related quality of life questionnaires ECOS-16 The ECOS-16 (Please see additional file 1 ([appendix]) was developed with the aim of measuring HRQoL in postmenopausal women with osteoporosis. It is based on the combination of two disease-specific HRQoL questionnaires for women with osteoporosis: the Osteoporosis Quality of Life Questionnaire (OQLQ) [ 8 ] and the Quality of Life Questionnaire of the European Foundation for Osteoporosis (QUALEFFO) [ 9 ]. The development process consisted in five phases: Phase I-Search for common structures; Phase II-Independent OQLQ and QUALEFFO item reduction using Rasch analysis; Phase III-OQLQ and QUALEFFO item equating; Phase IV-Quantitative reduction of equated items; Phase V-Qualitative reduction [ 11 ]. This newly questionnaire consists of 12 items from the QUALEFFO and 4 from the OQLQ (see Annex 1). All items have five possible response options, although the response options differ from one item to another. The 16 items in the new questionnaire are divided qualitatively into four dimensions. The nature of the four dimensions also suggests that they can be further combined to produce two summary scores that would include Physical Function and Pain in one Physical score, and another one that would include Fear of Illness and Psychosocial Function in a Mental score. These two summary scores could, in turn, be combined to provide an overall score for the questionnaire. However, although the 16 items can be classified qualitatively into four dimensions, this is an unidimensional questionnaire, according to the quantitative analysis [ 11 ]. The score of each item ranges from 1 to 5. ECOS-16 generates a single summary score obtained from the arithmetic mean of the answered items, so the total score ranges from 1 (best HRQoL) to 5 (worst HRQoL). The time frame for the questionnaire was one week. All items have the same weight on the overall questionnaire score and the overall score is calculated as the mean score of all the response items. It is a self-administered questionnaire, apart from some special cases (eyesight difficulties or illiteracy) where it was acceptable for the questionnaire to be administered by health care personnel or experienced interviewers. EUROQoL-5D The EUROQoL-5D (EQ-5D) is a generic HRQoL questionnaire. The EQ-5D consists of two parts: a descriptive system and a Visual Analogue Scale (VAS) [ 13 ]. The descriptive system contains 5 health status dimensions: Mobility, Self-Care, Usual activities, Pain/Discomfort and Anxiety/Depression. These dimensions are always presented in the same order, each one with 3 degrees of severity: no problems, some or moderate problems, and extreme problems, given a value of 1, 2 and 3, respectively. For each dimension, the respondent should mark the degree of severity which best describes their actual health status. The VAS of the EQ-5D is a vertical scale divided into millimetres along a 20 centimetres long thermometer where the two ends are labelled "worst imaginable health state" and "best imaginable health state" with a score of 0 and 100, respectively. The respondent should mark the point on the thermometer which, in their opinion, best describes their actual overall health status. MINI-OQLQ The MINI-OQLQ is a specific HRQoL questionnaire for women with vertebral fractures due to osteoporosis. The MINI-OQLQ is based on a selection of the two highest impact items from each of the five domains of the Osteoporosis Quality of Life Questionnaire (OQLQ). The MINI-OQLQ is, therefore, composed of 10 items grouped into the same five HRQoL dimensions (symptoms, physical function, activities of daily living, emotional function and leisure) [ 3 ]. Each item has seven response options ranging from 1 (worse HRQoL) to 7 (better HRQoL). The scoring is obtained per dimension, by calculating the mean score of the response items for each dimension, so that the higher the score in each dimension, the better the resulting HRQoL. 7-point change items (general health status, back pain, limitation in daily activities and emotional status) due to osteoporosis Changes in four health status items were assessed through four different items regarding the change in the patient's overall health status: the change in the patient's general health status due to osteoporosis, the change in the patient's back pain, the change in the patient's limitation in daily activities and the change in the patient's emotional status, all of them due to osteoporosis, with reference to the baseline. The items have seven possible response options, ranging from "Much better" to "Much worse" and including the category "More or less the same". These items were designed to be self-administered and validated in previous studies [ 10 , 14 ]. Statistical analysis Double data entry was carried out with a subsequent validation to guarantee the quality and consistency of the data. A statistical significance level of p < 0.05 was used in all statistical tests performed. The statistical program SPSS ® for Windows version 10.0 (SPSS, Inc., Chicago, Illinois) was used to carry out the entire data analysis. Previously the statistical analysis, a Kolmogorov-Smirnov test was conducted to assess the distribution of the variables in order to use a parametric or non-parametric tests. With the aim to describe the study sample characteristics and to evaluate differences among patients with and without vertebral fractures a descriptive and comparative analysis was done on patients' sociodemographic characteristics (age, education level) and clinical variables (Body Mass Index (BMI), number of years with menopause, presence or absence of non-vertebral fractures, concomitant diseases and received treatment during the previous year) according to vertebral fracture presence. In order to compare the two groups, the chi-squared test was used, given that the variables were categorical, and the Bonferroni correction for multiple tests. The feasibility of the ECOS-16 was analysed on the basis of missing data, time and method of administration. In order to evaluate the missing data, the number and percentage of missing response items in the whole questionnaire as well as the number and percentage of patients who failed to respond to any of the questionnaire's items were both calculated. The time spent administering the questionnaire was evaluated according to the method of administration (self-administered or administered by an interviewer). The floor (percentage of patients with the lowest score) and ceiling (percentage of patients with the highest score) effect were calculated for each one of the ECOS-16 items and for the overall score. An exploratory factor analysis was used to assess the content validity of the ECOS-16, using the scores obtained during the baseline by all patients for all 16 questionnaire items. Factors were extracted using principal-axis factoring method and varimax rotations. The adequacy of the factor analysis was assessed with the Kaiser-Meyer-Olkin measure and the Bartlett's test of sphericity [ 15 ]. To evaluate the construct validity, the correlations between the ECOS-16 scores and the patients' sociodemographic and clinical characteristics (bivariate analysis) were analysed. Pearson's correlation coefficient was used for continuous variables and the analysis of variance (ANOVA) for categorical variables. A multivariate analysis was also carried out taking the ECOS-16 score as a dependent variable and the sociodemographic and clinical variables that were significant in the bivariant analysis as independent variables, so that any possible confounding factors could be controlled. Secondly, the relationship between the ECOS-16 and the EQ-5D scores, the four 7-point items (general health status, back pain, limitation in daily activities and emotional status) and the MINI-OQLQ (only outpatients) were all analysed using Spearman's correlation coefficient, apart from the VAS which was analysed using Pearson's correlation coefficient. Higher correlations were expected between dimensions that measure the same HRQoL aspects. Because the lowest score in ECOS-16, in EQ-5D and in 7-point general health status item represent the best HRQoL, the expected correlations among their dimensions would be positive. The opposite applies to the dimensions of MINI-OQLQ and the other three 7-point items (back pain, limitation in daily activities and emotional status), where the lowest score represent the worst HRQoL and, therefore, the expected correlations between ECOS-16 would be negative. The reliability of the ECOS-16 was evaluated in terms of internal consistency and test-retest reliability. Internal consistency was calculated by Cronbach's α coefficient using the baseline scores of all questionnaire items. Test-retest reliability was evaluated only for outpatients who did not perceive a change in their general health status due to osteoporosis after a month, as shown by the change in 7-point general health status item (response category: 'More or less the same'). The Intraclass Correlation Coefficient (ICC) between the scores for both visits was used for this analysis. The hypothesis that the standard psychometric recommendations for Cronbach's α and ICC were greater than or equal to 0,7 was taken as a starting point for both internal consistency and test-retest reliability [ 16 ]. Longitudinal validity of the ECOS-16 was evaluated by analysing the correlations among changes registered in the ECOS-16, and changes in the EQ-5D, changes in the MINI-OQLQ and changes in the four 7-point change items from baseline to visit at 6 months. For this purpose the Spearman's correlation coefficient was used. The expected correlations are the same as in the construct validity. To assess responsiveness to change, first of all, attention has been drawn to whether the questionnaire detects the changes which are perceived by patients between the baseline and the visit at 6 months. In order to do so, the Student's t-test for paired data was used. To assess the magnitude of changes, the effect size was calculated, thus it was establish that the changes in the patients' scores increased at the same time that the changes perceived by patients. In order to calculate the effect size, the change in 7-point general health status item is used [ 17 ]. The Minimal Clinically Important Difference (MCID) has been defined as the smallest difference between the scores in a questionnaire that the patient perceives to be beneficial [ 14 ]. The MCID was calculated for those patients who, at visit at 6 months, declared changes "slightly better" in the general health status item (difference between the scores from baseline and the visit at 6 months). Results Table 1 shows the patients' sociodemographic and clinical characteristics evaluated according to the presence of vertebral fractures. Table 1 Patients' sociodemographic and clinical characteristics according to the presence of vertebral fracture With vertebral fracture Without vertebral fracture Total Age ‡ ≤ 65 years 42 (32.8%) 92 (51.4%) 134 (43.6%) > 65 years 86 (67.2%) 87 (48.6%) 173 (56.4%) Total 128 (100.0%) 179 (100.0%) 307 (100.0%) Education level No formal education 35 (28.2%) 46 (25.7%) 81 (26.7%) Primary school 78 (62.9%) 101 (56.4%) 179 (59.1%) Secondary school 11 (8.9%) 27 (15.1%) 38 (12.5%) University --- 5 (2.8%) 5 (1.7%) Total 124 (100.0%) 179 (100.0%) 303 (100.0%) BMI † ≤ 30 94 (73,4%) 153 (83,6%) 247 (79,4%) > 30 34 (26,6%) 30 (16,4%) 64 (20,6%) Total 128 (100,0%) 183 (100,0%) 311 (100,0%) Years with menopause ‡ ≤ 20 years 59 (46.8%) 115 (65.3%) 174 (57.6%) > 20 years 67 (53.2%) 61 (34.7%) 128 (42.4%) Total 126 (100.0%) 176 (100.0%) 302 (100.0%) Non-vertebral fractures Presence 24 (18.7%) 21 (11.4%) 45 (14.4%) Absence 104 (81.3%) 163 (88.6%) 267 (85.6%) Total 128 (100.0%) 184 (100.0%) 312 (100.0%) Concomitant diseases ‡ Presence 81 (62.8%) 144 (79.1%) 225 (72.3%) Absence 48 (37.2%) 38 (20.9%) 86 (27.7%) Total 129 (100.0%) 182 (100.0%) 311 (100.0%) Received treatment In the previous year ‡ Yes 92 (70.2%) 71 (38.4%) 163 (51.6%) No 39 (29.8%) 114 (61.6%) 153 (48.4%) Total 131 (100.0%) 185 (100.0%) 316 (100.0%) † p < 0.05 ‡ p < 0.007 Statistically significant differences (p < 0.01) were found for age, concomitant diseases (patient-reported), and length of time with menopause, it being the case that those patients with vertebral fractures were older and had had menopause for a longer time. In addition, those patients with vertebral fracture were those that had received more treatments during the previous year (p < 0.01) and also had a greater BMI (p < 0.05). However, those patients without a vertebral fracture presented concomitant diseases more frequently (p < 0.01). Only 14.4% of patients showed some type of non vertebral fracture and about 70% had completed at least primary education level. After the Bonferroni correction, the analysis that were still significant (p < 0,007) were age, years with menopause, concomitant disease and previous treatment. At baseline, all osteoporotic patients were prescribed or changed their treatment by the physicians, some having changed his previous treatment but others having received no treatment before. The mean (SD) administration time of the questionnaire was 12.3 (7.8) minutes for all patients, with a median of 10 minutes. In 55.1% of cases, the questionnaire was self-administered, the remainder being administered by health care personnel (due to eyesight difficulties or illiteracy). No statistically significant differences were observed in administration times according to the type of administration. The 99.7% of the patients answered all items of the questionnaire. Only one patient failed to respond to any item. A third of the items had a floor effect greater than or equal to 20%, 38.3% in item "Do you have problems with dressing?". The maximum ceiling effect was observed in 55.7% of the patients in item "How often have you had back pain in the last week?". In the remaining items, the ceiling effect was lower, ranging from 1.3% to 20.3%, the highest being in item "Are you afraid of getting a fracture?". Two patients (0.7%) recorded the highest scores for all the items (ceiling effect of the overall score). In the factor analysis, the Kaiser-Meyer-Olkin measure was 0.916 indicating a good sampling adequacy. The Bartlett's test of sphericity (p < 0.001) made it possible to accept the identity of the matrix correlations for the ECOS-16 items, thus indicating the suitability of the factor analysis. Table 2 shows the relationship between the patients' sociodemographic and clinical characteristics and the ECOS-16 score. In the bivariate analysis was observed a worse HRQoL in women with greater BMI (p < 0.05), a lower education level (p < 0.01) and concomitant chronic diseases (p < 0.05). Table 2 ECOS-16 scores according to patients' clinical and sociodemographic characteristics N ECOS-16 mean (SD) score Age ≤ 65 years 134 2.79 (0.77) > 65 years 173 2.92 (0.82) Total 307 2.87 (0.80) Education level ‡ No formal education 81 3.15 (0.73) Primary school 179 2.84 (0.81) Secondary school 38 2.46 (0.63) University 5 1.94 (0.58) Total 303 2.86 (0.80) BMI † ≤ 30 247 2.81 (0.79) > 30 64 3.06 (0.81) Total 311 2.86 (0.80) Years with menopause ≤ 20 years 174 2.80 (0.78) > 20 years 128 2.95 (0.83) Total 302 2.86 (0.81) Vertebral fractures Presence 131 2.94 (0.83) Absence 185 2.81 (0.78) Total 316 2.86 (0.80) Non-vertebral fractures Presence 45 2.88 (0.71) Absence 267 2.87 (0.81) Total 312 2.87 (0.80) Lumbar BMD * 298 0.024 Lumbar T-score * 308 -0.004 Neck BMD * 257 -0.042 Neck T-score * 264 0.090 Concomitant diseases † Presence 225 2.91 (0.81) Absence 86 2.71 (0.72) Total 311 2.85 (0.79) Received treatment in the previous year Yes 163 2.87 (0.77) No 153 2.86 (0.83) Total 316 2.86 (0.80) *Pearson's correlation coefficient † p < 0.05 ‡ p < 0.01 A multivariate analysis was carried out to identify patients' characteristics that were related to the ECOS-16 score (the variables entered into the multivariate analysis were age, education level, BMI, years with menopause, non-vertebral fractures, concomitant diseases and received treatment in previous years, all were codified as in table 1 ). The results showed that the variables of education level, number of concomitant diseases and the presence of vertebral fractures were related to the ECOS-16 score. Nevertheless, the percentage of variance explained by the variables included in the multivariate model was low, 11.1%. Table 3 shows the relationship between the ECOS-16 score and each one of the EQ-5D's dimensions, the four 7-point items and the MINI-OQLQ. All the EQ-5D's dimensions showed a statistically significant correlation with the ECOS-16 score, the dimensions with the greatest correlation being 'Mobility', 'Self-Care' and 'Pain/Discomfort'. The VAS was also statistically significant, with a Pearson's correlation coefficient of 0.61. The four 7-point items showed high correlations (greater than 0.7) with the ECOS-16 score, the item 'Limitation in daily activities' being the highest correlated item (0.82). The MINI-OQLQ dimensions showed moderate but statistically significant correlations (range: 0.47–0.73) with the ECOS-16, the 'Symptoms' and 'Leisure' dimensions having the highest correlations (0.71 and 0.74, respectively). Table 3 Correlations between the ECOS-16 scores and the EUROQoL-5D, the four 7-point items scores and the MINI-OQLQ N Spearman's correlation EUROQol-5D Mobility 315 0.643 Self-care 315 0.623 Usual activities 315 0.594 Pain/Discomfort 314 0.608 Anxiety/Depression 315 0.555 VAS a 312 -0.610 General health status 316 0.712 Back pain 313 -0.741 Limitation in daily activities 313 -0.822 Emotional status 313 -0.791 MINI-OQLQ b Symptoms 104 -0.711 Physical function 104 -0.670 Activities of daily living 104 -0.455 Emotional function 104 -0.473 Leisure 104 -0.736 a Pearson's Correlation Coefficient; b Outpatients only; All correlations are statistically significant at the 0.001 level The observed correlations between the changes among ECOS-16 questionnaire and the changes among the dimensions of the HRQoL questionnaires administered: EQ-5D, MINI-OQLQ and the four 7-point change items were similar to those of the construct validity (range: 0.33–0.76). All these correlations were also statistically significant. The internal consistency of the ECOS-16 was very high, with a Cronbach's α coefficient of 0.92. Test-retest reliability was analysed for 44 outpatients who declared that their general health status due to osteoporosis had not changed after a month, with an Intraclass Correlation Coefficient of 0.80 and a mean (SD) score change of 0.2 (0.5) points. Figure 1 shows the ECOS-16 patient scores for baseline and at visit at 6 months, as well as the effect size (ES) according to the changes in 7-point general health status item perceived by the patient after 6 months. The greater the perception of change in patients' general health status, the greater the changes in patients' scores. The same happened with the effect size as the patients who declared a 'much better' change in their general health status due to their osteoporosis at visit at 6 months had an effect size of 1.35, compared to those patients who declared a 'quite better' change having an ES of 1.23. The Minimal Clinically Important Difference (MCID) represented a mean change (SD) in the ECOS-16 score of 0.69 points, taking the category representing the least improvement in general health status due to their osteoporosis: 'slightly better'. Figure 1 ECOS-16 scores according to perceived changes in general health status Discussion The measurement of HRQoL has attracted increasing attention as a clinically relevant outcome of research and clinical practice. HRQoL questionnaires reflect the impact of health care interventions on health aspects such as physical, mental and social well-being. However, either in clinical research and in practice, a lengthy questionnaire is problematic for both the health care personnel and the patient. Shortish measure attempt to minimize time and effort as well as to increase patient interest [ 18 ]. Thus, shortish questionnaires need to be sufficiently psychometrically robust, proving that they are truly measuring what they set out to (validity), that they measure in a reliable way (reliability) and that they are capable of detecting real changes in perceived health status among patients with osteoporosis (responsiveness to change). The ECOS-16 originates from the reduction of two validated and widely used HRQoL questionnaires in osteoporosis patients with vertebral fracture [ 11 ]. However, in the present study, the ECOS-16 was administered to as many patients with fracture as those without, in order to establish its general applicability to patients with osteoporosis. Although the ECOS-16 requires a short time to be administered, self-administration was not possible in a high percentage of patients who needed help from health care personnel under conditions of usual clinical practice. In the future, it would be necessary to assess whether scores obtained through the questionnaire are maintained once the administration method changes, which would give more consistency to the questionnaire. Almost all the patients responded to all the questionnaire items. The presence of a floor effect for one third of the questionnaire items could lead to the conclusion that the questionnaire detects changes only when the severe disease status occurs. However, according to a previous qualitative division [ 12 ], the floor effect is concentrated both in the items belong to physical and psychosocial dimensions as for the mobility and self-care EQ-5D dimensions -the most conceptually equivalent- leading to think that the study sample has a certain clinical stability. As some previous studies, this study shows that the variables usually used to evaluate patients with osteoporosis, such as Bone Mineral Density (BMD) and the presence of vertebral fractures, have low or no correlation with HRQoL scores. This finding is not new [ 19 , 20 ] and suggests that HRQoL scores could be influenced by other factors such as personal, clinical and sociodemographic characteristics. In this study, the fact that patient's education level is the most significantly correlated variable draws particular attention. However, this observation has already been made in previous studies in patients with musculoskeletal problems [ 21 , 22 ]. The presence of concomitant diseases and/or a higher Body Mass Index (BMI) is scarcely correlated with HRQoL. This finding is interesting when taking into account studies evaluating HRQoL in patients with osteoporosis, whether they be descriptive or experimental in its design. The outcomes between compared groups could be underestimated if potential characteristic differences, particularly the education level, are not monitored. In this respect, the ECOS-16 apparently remains scientifically robust in its ability to discriminate among different education levels. In this study, the fact that the presence of vertebral fractures does not have a negative effect on HRQoL (bivariate analysis) deserves special attention. There seems to be some discrepancy regarding this issue in literature. Several studies have shown that HRQoL progressively deteriorates in relation to the presence and number of vertebral fractures [ 23 , 24 ]. However, other studies in the same area failed to find such a relationship [ 25 ], or have only found it when vertebral deformity is severe, while failing to find a relationship with any other fractures [ 26 , 27 ]. Recently, a significant number of studies have highlighted the importance of the site of the vertebral fracture and its effect on HRQoL. In this regard, it seems that the site of vertebral fracture has a much greater effect on HRQoL than the presence and number of vertebral fractures [ 28 - 30 ]. This difference could be explained by the relative rigidity of the thorax column in relation to the lumbar column, in the sense that mobility is more restricted when a lumbar rather than a thoracic region fracture occurs [ 31 ]. Moreover, lumbar column deformities have probably a greater impact on postural stability than alterations to the thoracic column. When the severity and site of the fracture is taken into account, fractured vertebrae in the transitional thoracolumbar region have a negative impact on HRQoL, at a Genant's degree greater than 1 [ 32 ]. Nevertheless, prospective studies addressing this issue should be conducted in the future assessing the impact of the time of the fracture and the site. Although in the present study, the bivariate analysis does not discriminate between patients with and without vertebral fracture, the multivariate model shows that the presence of fractures is indeed significant even if the percentage of explained variation is small. This is not surprising given the fact that such analysis is dealing with prevalent fractures in a short term observational study [ 28 ]. The inclusion of the time spent since the fracture occurred could improve the model, since it is well known that pain and disability due to a fracture progressively diminish over time [ 33 ]. However, among other health problems, a small percentage of explained variance was also found [ 33 ]. Nevertheless, the variables were entered into the multiple regression analysis dichotomously, as in table 1 , an this may reduce the likelihood of finding a relationship. Therefore, it is likely that administering a HRQoL questionnaire in conjunction with the analysis of clinical variables could provide a better overall picture of the osteoporosis impact on patients. The results obtained for internal consistency and test-retest reliability showed high levels of homogeneity among questionnaire items and good reproducibility over time. Moreover, the outcomes, have also been shown that the new questionnaire effectively detects changes in patients' perceived health status due to osteoporosis. Expressing responsiveness to change is an important characteristic of the new instrument, one which will doubtless allow its use in clinical research. The high correlation between the ECOS-16 and generic (EQ-5D) and specific (MINI-OQLQ) HRQoL questionnaires corroborates this hypothesis. It also highlights the new questionnaire's validity by demonstrating that it measures concepts which are closely related to already validated HRQoL questionnaires [ 3 , 13 ]. The mean change in score per question corresponding to the effect size in general and to the MCID in particular is consistent, in terms that the larger the change assessed by the ECOS-16, the larger the effect size. Moreover, MCID is consistent with the results of other HRQoL questionnaires, and it is useful to compare the magnitude of changes detected between them. MCID will also be useful in the planning of new trials, as sample size depends on the magnitude of the difference investigators consider clinically important and are not willing to risk failing to detect [ 34 ]. The potential limitations of this study are mainly due to it being unable to rely on certain variables which have shown a clear influence on HRQoL. In particular, the "time spent since the fracture occurred" [ 35 ] was not analysed even though the objective of the study were women with established osteoporosis. The inclusion of prevalent fractures and exclusion of the incidence fractures means that a smaller variability among the patients in this study was established and, possibly, it also means that vertebral fractures had less influence on HRQoL. A further possible limitation is the limited sample size, which was relatively low to obtain statistical significance for more than one fracture site and for a certain fracture severity. This was a prospective, observational study with a limited follow-up time, but under conditions of usual clinical practice it served to prove the short-term good responsiveness of the questionnaire and test its remaining psychometric properties. Nevertheless long term follow-up studies will be necessary in the future. The low education level of the study sample must also be taken into account. It is consistent with previous studies [ 8 , 9 ], and as in other chronic diseases, education level has an impact on the prevalence of osteoporosis [ 36 ] and on the preferences of patients [ 37 ]. Furthermore, the impact of osteoporosis on HRQoL, the site of the vertebral fracture, coupled with the time spent since its onset, are extremely important variables which must be taken into account, since traditional clinical variables -i.e. bone densitometry- do not have a remarkable relationship with HRQoL [ 38 ]. Conclusions In conclusion, the ECOS-16 is a HRQoL questionnaire which is short, easy to administer (although some women need aid) and with adequate preliminary psychometric properties. This makes the ECOS-16 potentially very useful during routine clinical practice or/and research for the treatment and follow-up of post-menopausal women with osteoporosis. Nevertheless, its actual potential must be proven in future clinical trials in order to recommend its use in research and clinical practice. Supplementary Material Additional File 1 ECOS Appendix1.doc, the ECOS-16 questionnaire Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC514569.xml
555535	Soilborne wheat mosaic virus (SBWMV) 19K protein belongs to a class of cysteine rich proteins that suppress RNA silencing	Amino acid sequence analyses indicate that the Soilborne wheat mosaic virus (SBWMV) 19K protein is a cysteine-rich protein (CRP) and shares sequence homology with CRPs derived from furo-, hordei-, peclu- and tobraviruses. Since the hordei- and pecluvirus CRPs were shown to be pathogenesis factors and/or suppressors of RNA silencing, experiments were conducted to determine if the SBWMV 19K CRP has similar activities. The SBWMV 19K CRP was introduced into the Potato virus X (PVX) viral vector and inoculated to tobacco plants. The SBWMV 19K CRP aggravated PVX-induced symptoms and restored green fluorescent protein (GFP) expression to GFP silenced tissues. These observations indicate that the SBWMV 19K CRP is a pathogenicity determinant and a suppressor of RNA silencing.	Background Viruses survive in their hosts either by evading or countering host defenses. Viral evasion is a passive mechanism by which viruses overwhelm host defenses, or invade organs or cells where the host defenses cannot reach them. The ability of a virus to counter host defenses requires an active mechanism to either bypass or disarm the host machinery. Viruses invading vertebrate hosts produce virokines and viroceptors which interact with immune response molecules to inhibit or modulate their anti-viral activities [ 1 , 2 ]. Recent studies have shown many viruses infecting a wide range of eukaryotic hosts encode proteins that suppress the RNA silencing, anti-viral defense response [ 3 - 6 ]. Silencing suppressors encoded by viruses limit degradation of viral RNAs by the RNA silencing machinery. Among plant viruses, some silencing suppressor proteins also affect symptom development and increase virus titer. The Cucumber mosaic virus (CMV) 2b, the Tobacco etch virus (TEV) HC-Pro, and the Tomato bushy stunt virus (TBSV) P19 [ 7 - 10 ] proteins are among the best studied silencing suppressors that are also pathogenicity determinants. The TBSV P19 protein was unique because it affects disease severity in a host specific manner [ 11 , 12 ]. Little is known about the evolution and phylogenetic relationships of silencing suppressor proteins. In particular, viruses belonging to the genera Furo-, Hordei-, Tobra-, Peclu-, Beny-, Carla -, and Pomovirus encode small cysteine-rich proteins (CRPs) near the 3' ends of their genomes, and some have been identified as both silencing suppressor proteins and pathogenicity factors. For example, the Barley stripe mosaic virus (BSMV; a hordeivirus) gamma b protein and the Peanut clump virus (PCV; a pecluvirus) 15K protein suppress RNA silencing, modulate symptom severity, and systemic virus accumulation [ 13 - 16 ]. The Tobacco rattle virus (TRV; a tobravirus) 16K CRP has been described as a pathogenicity factor and suppresses RNA silencing [ 17 ]. In complementation studies, the Soilborne wheat mosaic virus (SBWMV; a furovirus) 19K CRP, the BSMV gamma b protein, and the CMV 2b (which is not a CRP) protein functionally replaced the 16K CRP of TRV [ 15 ]. Since deletion of the TRV 16K CRP ORF reduced virus accumulation in plants, functional replacement by these heterologous viral ORFs indicates that these CRPs share some common function. Characterizing the functional similarities among these CRPs is crucial to understanding their evolutionary relationship. Until now the phylogenetic relationships among these CRPs are unclear [ 18 ]. This study was undertaken to characterize the SBWMV 19K CRP. SBWMV is a bipartite RNA virus and is the type member for the genus Furovirus [ 19 ]. RNA1 encodes the viral replicase and putative viral movement protein (MP). The viral replicase is encoded by a single large open reading frame (ORF) and is phylogenetically related to the Tobacco mosaic virus (TMV) replicase [ 20 ]. The 3' proximal ORF of RNA1 encodes a 37K MP that shares sequence similarity with the dianthovirus MP [ 21 , 22 ]. SBWMV RNA2 encodes four proteins. The 5' proximal ORF of RNA2 encodes a 25K protein from a nonAUG start codon [ 23 ] and its role in virus infection is unknown. The coat protein (CP) ORF has an opal translational termination codon and readthrough of this codon produces a large 84K protein [ 23 ]. The CP readthrough domain (RT) is required for plasmodiophorid transmission of the virus [ 24 ]. The 3' proximal ORF of RNA2 encodes a 19K CRP. To gain insight into the role of the SBWMV 19K CRP in virus infection, amino acid sequence comparisons were conducted to determine the relatedness of the SBWMV 19K CRP to other viral CRPs. The Potato virus X (PVX) infectious clone was used to express the SBWMV CRP and to study its role in virus pathogenicity and suppressing RNA silencing. Results SBWMV 19K protein is a conserved CRP The Pfam Protein Families Database reports a family of CRPs with similar sequences which includes proteins from BSMV, PSLV, PCV and SBWMV (Pfam 04521.5). Since there are viruses not included in the Pfam report that encode CRPs, this study was undertaken to determine if there is a larger CRP family containing related viral proteins. Further examination in this study reveals that the CRPs encoded by all known hordei-, peclu- and furoviruses share significant sequence similarity (Fig. 1 ). Efforts to find similarity between these proteins and CRPs encoded by pomo-, beny- and potyviruses were not successful. Whether these other plant viral CRPs are also suppressors of silencing can not be concluded at this point for two reasons: insufficient study and only weak sequence similarity relationships. Sequences of CRPs that affect virus replication and are encoded by members of other virus genera were also determined to be unrelated [ 25 ]. Figure 1 Amino acid sequence alignment of the CRPs encoded by furo-, peclu-, tobra- and hordeiviruses. The positions of amino acids are numbered above the alignment. The secondary structure prediction is shown directly above the alignment. Cys and His residues are bold uppercase letters. The leucines of leucine zippers are in bold face. The placement of residues that differ from Pfam are underlined. Vertical bars at the bottom represent where the Pfam family starts and stops. The genus for each virus is indicated on the right of the sequence. Abbreviations and accession numbers for the 33 aligned viruses are used (those displayed are underlined): LyRSV , Lychnis ringspot virus gi_1107721; CWMV-2 , Chinese wheat mosaic virus gi_14270345; CWMV, gi_9635448; OGSV Oat golden stripe virus , gi_9635452; SBWMV-NE88 gi_9632360; SBWMV-NE gi_1449160; SBWMV OKl-1 , gi_1085914; SBWMV-NY, gi_21630062; SBCMV-Ozz, Soilborne cereal mosaic virus gi_12053756; SBCMV-Fra, gi_9635249; SBCMV-O , gi_6580881; SBCMV-G, gi_6580877; SBCMV-C, gi_6580873; JSBWMV, Japanese soilborne wheat mosaic virus gi_7634693; SCSV , Sorghum chlorotic spot virus gi_21427644; PSLV , Poa semilatent virus gi_321642; BSMV-PV43 , Barley stripe mosaic virus gi_19744921; BSMV-RUS, gi_94465; BSMV-JT, gi_808712; BSMV-ND18, gi_1589671; PCV , Peanut clump virus gi_20178597; IPCV , Indian peanut clump virus gi_30018260; TRV-PpK20, Tobacco rattle virus , gi_20522121; TRV-ORY gi_2852339; TRV-Pp085 gi_42733086; TRV-PSG, gi_112699; TRV-PLB, gi_465018; TRV-CAN, gi_1857116; TRV-FL, gi_3033549; TRV-RSTK, gi_6983830; TRV-TCM , gi_112701; PepRSV , Pepper ringspot virus , gi_20178602; PEBV , Pea early browning virus , gi_9632342. The SBWMV 19K protein is a CRP because it contains nine Cys residues [ 20 ]. Seven of these Cys residues are conserved in all furovirus proteins and are located in the N-terminal half of the protein. Five of these residues are within the block of sequences designated as protein family Pfam04521.5 and three of the conserved Cys residues are also conserved in the hordeiviral and pecluviral proteins. A selection from this alignment was corrected for several misplacements of short peptide sequences and is shown in Figure 1 . The alignment represents the entire length of these proteins, although the termini are aligned with less confidence than the core. Examination of the tobraviral CRP sequences revealed sufficient similarity to justify their alignment with the Pfam04521.5 sequences. The alignment resulted in a significance score between 6 and 7, suggesting that the tobraviral proteins belong to this family. The multiple sequence alignment of 33 CRPs from furo-, tobra-, peclu-, and hordeiviruses (Fig. 1 ) revealed three absolutely conserved residues: Cys70, Cys112, and His116 (numbering based on the aligned sequences). Gly113 was conserved in all viruses (except TRV-CAN) and is contained within a Cys-Gly-Xaa-Xaa-His motif in which one of the two Xaa residues is Lys or Arg. There is a Cys residue at position 7, 8 or 9 which is conserved in all except PCV and IPCV (pecluvirus) amino acid sequences. Alignment of the N-terminus is not exact since the PCV and IPCV proteins are N-terminally truncated. Within the N-terminal half, there are additional positions containing Cys residues that are conserved for some but not all viruses. For example, Cys9 is conserved among hordei-, tobra-, and some furoviruses; Cys at position 32 and 33 is conserved among all but pecluviruses; Cys36 is conserved among hordei- and furoviruses; Cys45 is conserved among furo and tobraviruses; Cys76 is conserved among furo and tobraviruses (except for SCSV; the pecluvirus PCV, but not IPCV, also has Cys76); Cys80 is conserved among all viruses except PeRSV and PEBV. Lys at position 52 and Arg at position 54 or 55 (Lys-Xaa-Arg or Lys-Xaa-Xaa-Arg) are conserved among all except PSLV. Gly at position 77 is conserved among all except tobraviruses. The secondary structure prediction derived from the multiple sequence alignment is a long helical region extending from or slightly beyond the Cys-Gly-Xaa-Xaa-His motif to within 20 residues of the C-terminus. The furoviral proteins have spacings of conserved Leu residues from positions 89 to 106 consistent with a leucine zipper structure (which was not apparent in the original Pfam 04521.5). The N-terminal halves of the aligned amino acid sequences, containing most of the Cys residues, have a mixture of extended, helical and loop predicted structures. The pecluviruses PCV and IPCV, and the hordeiviruses BSMV, LyRSV, and PSLV proteins contain a Ser-Lys-Leu sequence at the C-terminus. This tripeptide was shown for PCV to be a peroxisomal targeting signal [ 16 ]. This signal is not present in CRPs of furo- or tobraviruses. SBWMV 19K CRP aggravates PVX-induced symptoms The tobravirus and hordeivirus CRPs have been described as pathogenicity determinants that regulate symptom severity in infected plants [ 15 ]. Since the SBWMV 19K protein is a similar CRP, experiments were conducted to determine if it also has an effect on symptom expression. The SBWMV 19K ORF was inserted into the PVX genome and PVX.19K infectious transcripts were used to inoculate N. benthamiana, N. clevelandii, C. quinoa , and C. amaranticolor leaves (Fig. 2 ). As a control, plants were also inoculated with PVX.GFP, which has the green fluorescent protein (GFP) gene inserted into the viral genome. The spread of PVX.GFP expression was monitored using a handheld UV lamp to monitor GFP expression and verify systemic virus accumulation (data not shown). Figure 2 Plants infected with PVX.GFP or PVX.19K at 21 dpi. (A) N. benthamiana plants infected with PVX.GFP (left) and PVX.19K (right). (B, D) PVX.19K-infected N. benthamiana and N. clevelandii plants, respectively, at 21 dpi show systemic necrosis. (C) PVX.GFP-infected N. clevelandii plants. (E, F) C. quinoa and C. amaranticolor leaves infected with PVX.19K (left both panels) and PVX.GFP (right in both panels). Symptoms were first observed in plants inoculated with PVX.GFP and PVX.19K between 10 and 14 dpi. By 21 dpi, systemic necrosis was evident in N. benthamiana and N. clevelandii plants inoculated with PVX.19K (Fig. 2A, B and 2D ) while PVX.GFP infected plants showed systemic mosaic symptoms (Fig. 2A and 2C ). N. benthamiana plants infected with PVX.19K were clearly stunted in comparison to plants infected with PVX.GFP (Fig. 2A ). The PVX.19K infected N. clevelandii leaves collapsed by 21 dpi (Fig. 2D ). Immunoblot and northern analyses were conducted to verify PVX accumulation in the upper leaves of N. benthamiana plants. Immunoblot analysis was conducted using anti-PVX CP serum. High levels of PVX CP was detected in plants that were systemically infected with PVX.GFP (Fig. 3A lanes 1–4) and PVX.19K (Fig. 3A lanes 5–8). The SBWMV 19K CRP had no obvious effect on PVX accumulation in upper noninoculated leaves. Viral RNA accumulation was analyzed by northern blot and high levels of genomic RNA was detected in the upper leaves of PVX.GFP (Figure 3B lanes 2–4) and PVX.19K (Fig. 3B lanes 5–8) inoculated plants. Thus, the SBWMV 19K CRP did not seem to have a deleterious effect on PVX accumulation. RT-PCR was used to verify that the SBWMV 19K ORF was maintained in the PVX genome in systemically infected plants. RNA samples taken from the upper leaves of N. benthamiana plants which were used for northern analysis, were also used in RT-PCR reactions to verify the presence of the SBWMV 19K ORF in the PVX genome. In all samples it appeared that the SBWMV 19K CRP was stably maintained in the PVX genome (data not shown). Figure 3 Immunoblot and northern analyses of the PVX infected N. benthamiana plants. (A) Immunoblot analysis conducted using PVX CP antiserum show similar levels of PVX.GFP virus (lanes 1–4) and PVX.19K virus (lanes 5–8). Lane 9 contains extract of non inoculated plants. (B) Northern analysis of RNA isolated from a healthy plant (lane 1), upper noninoculated leaves of PVX.GFP infected plants (lanes 2 – 4) and upper noninoculated leaves of PVX.19K infected plants (lanes 5 – 8). Blots were probed with a GFP sequence probe. The bottom image is the ethidium bromide stained gel showing ribosomal RNAs. Abbrev.: g, genomic RNA. PVX.19K produced large necrotic lesions in the C. quinoa and C. amaranticolor leaves. Local lesions were detected in plants inoculated with PVX.GFP or PVX.19K between 5 and 7 dpi. PVX.19K-inoculated C. quinoa plants showed severe necrotic local lesions (Fig. 2E ). The necrotic lesions gradually merged and the infected tissue eventually collapsed. PVX.19K-inoculated C. amaranticolor leaves showed enlarged chlorotic lesions advancing to necrotic lesions over time (Fig. 2F ). PVX.GFP-inoculated C. quinoa leaves showed small chlorotic and necrotic local lesions while PVX.GFP-inoculated C. amaranticolor leaves showed mild flecks (Fig. 2F ). Association of PVX.GFP with the local lesions was verified using a hand held UV lamp (data not shown). SBWMV 19K CRP is a suppressor of RNA silencing In this study we employed a widely used "reversal of silencing assay" to determine if the SBWMV 19K CRP is a suppressor of RNA silencing in plants [ 28 ]. In this assay, GFP-expression in the 16C transgenic N. benthamiana plants (Fig. 4B ) was silenced by infiltrating young leaves with a suspension of Agrobacterium expressing GFP. The progression of GFP silencing was viewed first locally and then systemically using a hand held UV lamp. Within two weeks, the spread of GFP silencing was viewed systemically (Fig. 4C ) and by three weeks, the only visible fluorescence is red fluorescence due to chlorophyll (Fig. 4D ). At this time, the silenced plants were inoculated with PVX.19K. As PVX.19K viruses spread locally and then systemically, there was no change in GFP expression in the inoculated leaves or in the upper leaves (Fig. 4E ). However, GFP expression was observed in the emerging leaves (Fig. 4F – H ). The SBWMV 19K CRP prevented RNA silencing only in emerging leaves where RNA silencing had not developed prior to virus infection. As a control, plants were also inoculated with PVX.GUS following infiltration with Agrobacterium . There was no evidence of GFP expression in the inoculated, mature, or new emerging leaves. The silencing phenotype was unaffected by PVX.GUS. Figure 4 Evidence for RNA silencing suppression by the SBWMV 19K CRP. (A) nontransgenic N. benthamiana under a UV lamp exhibits red fluorescence due to chlorophyll. (B) GFP-transgenic N. benthamiana (line 16C) exhibits green fluorescence under a UV lamp. (C) GFP was systemically silenced in the 16C transgenic N. benthamiana following infiltration with Agrobacterium . Here in the upper most leaves GFP silencing is vein centric. Systemic GFP silencing is detected initially within 2 weeks. (D) Within 3 weeks, GFP expression is completely silenced in the upper leaves. (E) GFP silenced plant inoculated with PVX.GUS. Emerging tissues of the infected plant remain silenced. (F, G, and H) GFP expression was observed in the emerging tissues of plants that were inoculated with PVX.19K. (I) Northern analyses of total RNAs from nontransgenic tissues (lanes 1, 2) and GFP transgenic tissues (lanes 4 – 7) probed with a labeled GFP sequence probe. Lane 3 is blank. Lanes under the northern blot show ribosomal RNAs on an ethidium bromide stained gel. (J) Northern analysis of total RNAs from 16C plants infiltrated with Agrobacterium containing GFP constructs and probed with a labeled GFP sequence probe. Lanes 1–4 are RNA samples taken from plants that were also inoculated with PVX.19K. Lanes 5–8 are RNA samples taken from plants inoculated with PVX.GUS. Lanes under the northern blot show ribosomal RNAs. Northern analyses was conducted to confirm RNA silencing in the upper leaves of Agrobacterium -infiltrated leaves and in the plants inoculated with PVX.GUS (Fig. 4I and 4J ). GFP specific RNAs were detected in transgenic plants (Figure 4I lanes 4–7) and emerging leaves of plants injected with Agrobacterium and inoculated with PVX.19K (Figure 4J lanes 1–4). GFP specific RNAs were not detected in untreated nontransgenic plants (Figure 4I lanes 1–3) or in plants that were injected with Agrobacterium and inoculated with PVX.GUS (lanes 4–8). RNA samples collected from non silenced and silenced plants were also tested by Northern analysis to confirm the systemic accumulation of PVX.GUS or PVX.19K (data not shown). Since, GFP expression was restored in plants systemically infected with PVX.19K but remained silenced in plants inoculated with PVX.GUS, it is likely that the SBWMV 19K ORF is a suppressor of RNA silencing. Discussion Many viruses encode proteins that suppress RNA silencing but the phylogenetic relatedness of these proteins is poorly understood. In this study, one class of viral CRPs, which were described as suppressors of RNA silencing and/or viral pathogenicity determinants, were shown to be phylogenetically related. These CRPs have a conserved Cys-Gly-Xaa-Xaa-His motif in which one of the two Xaa residues is Lys or Arg. The N-terminus has several conserved Cys residues that likely comprise a zinc finger motif. In fact, the ability of the gamma b protein of BSMV to bind Zn(II) was recently demonstrated [ 25 ]. Prior to 1999, SBWMV, BNYVV, PCV, and PMTV belonged to the genus Furovirus . As sequence data from different furoviruses have become available, it became clear that there are significant differences in the genome organization of these viruses, and therefore furovirus classification was revised in 1999 [ 19 ]. The genus Furovirus now consists of viruses similar in genome organization to SBWMV [ 29 ]. These viruses are bipartite and have a single MP that is phylogenetically related to the tobamovirus and dianthovirus MPs [ 20 , 22 ]. BNYVV, PCV, and PMTV were reclassified into the genera Benyvirus, Pecluvirus , and Pomovirus , respectively, for two reasons [ 19 , 29 ]. First, the MPs of these viruses are phylogenetically distinct from SBWMV. BNYVV, PCV, and PMTV each possess a cluster of three slightly overlapping ORFs known as the "triple gene block", which has been shown for BNYVV [ 30 ] to mediate viral cell-to-cell movement. Second, benyviruses and pomoviruses differ from furoviruses in the number of genome segments. BNYVV has four or five genome segments while PMTV has three genome segments [ 31 ]. Pecluviruses like furoviruses have two genome segments, thus the primary difference between these virus genera is the MP ORFs [ 32 ]. This is significant because the initial amino acid sequence comparisons of CRPs from furo-, hordei-, tobra-, and carlaviruses included BNYVV as the type-member of the genus Furovirus and concluded that these small CRPs were unrelated [ 33 ]. Reclassification of the BNYVV as a member of the genus Benyvirus and inclusion of new members into the genus Furovirus led us to reexamine the relatedness of the viral CRPs. Based on the most recently defined taxonomic structure, the current amino acid sequence comparison presented in Figure 1 indicates that the CRPs derived from viruses of the genera Furo-, Hordei-, Peclu -, and Tobravirus are phylogenetically related. On the other hand, these proteins are so different from CRPs encoded by Pomo -, Beny- and Carlaviruses that the latter ones could not be included in the alignment (Fig 1 ). The present study shows that the SBWMV 19K CRP, when expressed from the PVX genome, functions as a pathogenesis factor and a suppressor of RNA silencing. The SBWMV 19K CRP, when it was expressed from the PVX genome, induced systemic necrosis on Nicotiana benthamiana, N. clevelandii, C. quinoa , and C. amaranticolor . These symptoms are distinct from the symptoms associated with PVX infection in these hosts, and from symptoms induced by SBWMV in its natural hosts. In systemic hosts, both PVX and SBWMV typically cause mosaic symptoms that range from mild to severe. In C. quinoa and C. amaranticolor both PVX and SBWMV cause mild chlorosis. Severe necrosis and ultimate collapse of the tissue has been reported for other unrelated viral proteins that are pathogenicity factors and suppressors of RNA silencing. This include the Poa semilatent virus (PSLV) gamma b, TBSV P19, Tobacco etch virus HC-Pro, and the Rice yellow mottle virus P1 proteins[ 7 , 11 , 14 , 34 ]. When we introduced the SBWMV 19K ORF into the TBSV vector and inoculated it to N. benthamiama, N. tabacum, C. quinoa , and C. amaranticolor (data not shown) plants, the SBWMV 19K CRP did not have any effect on symptomology (data not shown). However, it was reported previously that protein expression levels from the TBSV vector might be too low to test the effects of heterologous proteins on symptom severity [ 35 ]. Since an antibody to the SBWMV 19K CRP is unavailable, the levels of protein expression from PVX or TBSV vectors could not be analyzed to determine if gene dosage or protein expression levels contribute to symptom severity. In a related study, the SBWMV 19K and the BSMV gamma b CRPs could substitute for the TRV 16K CRP within the TRV genome, promoting virus replication and systemic accumulation [ 15 ]. The ability of the SBWMV 19K and the BSMV gamma b CRPs to induce severe symptoms when expressed from the PVX genome is reminiscent of phenomena described in relation to viral synergisms. The best studied viral synergism is between Tobacco etch virus (TEV) and PVX in which the TEV HC-Pro protein enhances accumulation and disease severity of PVX [ 34 ]. HC-Pro promotes infection of PVX by suppressing the anti-viral RNA silencing defense mechanism that would normally act on PVX to reduce virus infection. HC-Pro has the ability to increase PVX accumulation in the same way the SBWMV 19K CRP and the BSMV gamma b proteins were shown previously to enhance accumulation of TRV in infected plants [ 15 ]. Conclusion The phylogenetic relatedness of the hordei-, peclu-, and furovirus CRPs is further substantiated by evidence that these proteins are all capable of suppressing RNA silencing in emerging leaves. This was demonstrated in the present and related studies using the same reversal of silencing assay used in this study. The SBWMV 19K CRP, the BSMV and PSLV gamma b CRPs, and the PCV 15K CRPs were each unable to change GFP expression in leaves where GFP was silenced prior to virus infection. However in each case, GFP expression occurred in newly emerging leaves [ 14 , 16 ]. Thus, members of this family of CRPs similarly act on the RNA silencing machinery to block spread of the silencing signal into newly emerging leaves. In each case, the silencing suppressor activities of these CRPs have been compared to CMV and potyviruses in preventing onset of RNA silencing in new growth [ 14 , 16 ]. While there is no evidence that the hordei-, peclu- and furovirus CRPs are related to the CMV or potyvirus silencing suppressors, it seems that the mode of action might be conserved among diverse viruses. Methods Amino acid sequence comparisons Related protein sequences were identified and retrieved from the NCBI data bank using PSI-BLAST. A PSI-BLAST search was launched with the amino acid sequence of the 19K CRP of Chinese wheat mosaic virus (CWMV, a furovirus). A similar search began with the amino acid sequence of BSMV gamma b, a sequence recovered in the CWMV search. Both searches converged at the second iteration and retrieved the same set of 22 sequences. This set contained CRPs derived from furo-, peclu- and hordeiviruses and contained the conserved P18 PFAM domain ("protein family"URL reference [ 36 ]. A preliminary alignment of the retrieved proteins sequences was performed using the multiple sequence alignment mode of ClustalX. These twenty two furovirus and hordeivirus sequences were aligned using ClustalX alignments suggested in the BLAST outputs and PFAM. The tobraviral CRPs were not recovered by the above procedure, but upon manual inspection, appeared to have Cys residues in a linear arrangement that was similar to the set of 22 proteins. Eleven tobraviral protein sequences, exclusive members of a conserved domain in the Conserved Domain database were aligned using ClustalX [ 37 ]. This tobraviral amino acid sequence alignment and the alignment of the 22 amino acid sequences sequences were assembled by ClustalX in profile mode, followed by manual adjustment. Amino acid sequences of aligned furo- and hordeiviral proteins were aligned with tobraviral amino acid sequences in profile mode of ClustalX (a total of thirty three sequences were aligned). A total of 33 amino acid sequences were aligned. In all cases, adjustments to the alignments were made using Se-Al [ 38 ]. Significance scores for the alignment of the two groups of protein sequences were calculated as previously described, using a structural conservation matrix, SCM2, for scoring [ 39 ]. Plasmids and bacterial strain All plasmids were used to transform Escherichia coli strain JM109 [ 40 ]. The plasmids pPVX.GFP is an infectious viral clone and contains a bacteriophage T7 promoter [ 39 ]. The pPVX.GFP plasmid contains the PVX genome and the GFP adjacent to a duplicated CP subgenomic promoter. The plasmid pHST2-14 contains the TBSV genome and a mutation in the TBSV P19 ORF eliminating expression of a protein that suppresses RNA silencing [ 10 , 42 ]. The plasmid pTBSV.GFP contains GFP inserted into the TBSV genome replacing the viral CP ORF [ 10 ]. The SBWMV 19K CRP ORF was inserted into the PVX.GFP genome, replacing the GFP ORF. The 19K CRP ORF was reverse transcribed and PCR amplified from purified SBWMV RNA using a forward primer (GCG GGG ATC GAT ATG TCT ACT GTT GGT TTC CAC) containing added sequences encoding a Cla I restriction site (underlined) and a reverse primer (CGC GTC GAC TCA CAA AGA GGA TAT CTT CTT TGG C) containing sequences encoding a Sal I restriction site (underlined). PCR products and pPVX.GFP plasmids were digested with Cla I and Sal I and then were ligated to prepare pPVX.19K. In vitro transcription and plant inoculations In vitro transcription reactions contained: 0.25 μg of linearized DNA, 5 μl of 5X T7 transcription buffer, 1.0 μl of 0.1 M DTT, 0.5 μl of SUPERase·In™ ribonuclease inhibitor (20 U/ μl) (Ambion, Austin, TX), 2.5 μl of an NTP mixture containing 5 mM ATP, CTP, UTP, and GTP (Pharmacia-Pfizer, Mississauga, Ontario, Canada), 0.7 μl of T7 polymerase (Ambion), and nuclease-free water to a final volume of 25 μl. The reactions were incubated for one and a half hour at 37°C [ 10 ]. Nicotiana benthamiana, N. clevelandii, Chenopodium quinoa , and C. amaranticolor plants were inoculated with infectious transcripts to study disease severity. Four plants, two leaves per plant, were inoculated in each experiment. Experiments were repeated at least three times. Ten μl of undiluted PVX.GFP or PVX.19K transcripts were rub-inoculated to each plant. The transgenic N. benthamiana line 16C was used to study RNA silencing. This line is homozygous for the GFP transgene at a single locus [ 44 ]. Plants were inoculated with transcripts following infiltration with Agrobacterium (see below). Agrobacterium infiltration of leaves Agrobacterium tumefaciens strain C58C1 (pCH32) carrying a binary plasmid expressing GFP from a Cauliflower mosaic virus (CaMV) 35S promoter was used to silence GFP expression in N. benthamiana line 16C. Agrobacterium cultures were grown overnight at 28°C in 5 ml of L-broth medium containing 5 μg/ml of tetracycline and 50 μg/ml of kanamycin. This 5 ml culture was used to inoculate 50 ml L-broth and grown overnight in medium containing 5 μg/ml tetracycline, 50 μg/ml kanamycin, 10 mM MES, and 20 μM acetosyringone. Cultures of Agrobacterium containing GFP were pelleted by centrifugation and resuspended in a solution containing 10 mM MgCl 2 , 10 mM MES, and 150 μM acetosyringone. The final concentration of Agrobacterium was 0.5 OD 600 . The suspension was left at room temperature for 2–3 hours and then loaded into a 2 ml syringe. The syringe was used to infiltrate the suspension into the underside of the leaf. Visualization of GFP A hand-held model B-100 BLAK-RAY long wave ultraviolet lamp (Ultraviolet Products, Upland, CA) was used to monitor GFP expression in 16C plants infiltrated with Agrobacterium and in PVX.GFP inoculated plants. GFP fluorescence was recorded with a Sony Digital Still Camera model DSC-F717 (Sony Corporation of America, New York City, New York). In all plants analyzed, GFP expression was monitored every 3 days for up to 21 days post inoculation (dpi) or post infiltration with Agrobacterium . Immunoblot analysis Immunoblot analyses were conducted according to [ 40 ]. Total protein from uninfected and infected N. benthamiana leaves was extracted in 1:10 (w/v) grinding buffer (100 mM Tris-HCl pH 7.50, 10 mM KCl, 5 mM MgCl 2 , 400 mM sucrose, 10% glycerol, and 10 mM β-mercaptoethanol). Extracts were centrifuged at 10,000 g for 10 min. Equal volumes of supernatants and protein loading buffer (2 % SDS, 0.1 M dithiothreitol, 50 mM Tris-HCl pH 6.8, 0.1% bromophenol blue, and 10 % glycerol) were mixed and boiled for 5 min. SDS-PAGE was carried out for 1 h at 200 V using 30 μl of each sample and 12.5% SDS -PAGE and the Biorad Mini-Protean 3 system (Biorad Laboratories, Hercules, CA). Proteins were transferred to PVDF membranes (Amersham Biosciences Corp., Piscataway, NJ) at 4°C overnight using protein transfer buffer (39 mM glycine, 48 mM Tris base, 0.037% SDS, and 20% methanol, pH 8.3) and a BioRad Trans-Blot system (BioRad Laboratories). Immunoblot analyses were conducted using the ECL-Plus Western Blotting Detection Kit (Amersham Biosciences Corp.). PVX CP antiserum (1:200) (Agdia, Elkhart, IN) was used. Northern analysis Northern analyses were conducted according to [ 40 ]. For analyses of PVX infected plants and GFP expressing transgenic plants, a radiolabeled DNA probe was prepared using Rediprime II Random Prime Labeling System (Amersham Biosciences Corp.). Labeling was conducted using PCR products corresponding to either the GFP or PVX CP ORFs. For detection of TBSV.GFP and TBSV.19K in infected plant extracts, a DNA probe was labeled with digoxigenin (DIG). TBSV.GFP plasmids were digested with Nco I and Sal I and a 614 nt fragment was gel eluted and labeled using Dig High Prime kit (Roche Applied Science Inc. Indianapolis, IN). The CSPD DIG Luminescence Detection Kit (Roche Applied Science Inc.) was used for chemiluminescence detection of DIG-labeled probes. Special thanks to Wenping Qui at Southwest Missouri State University for assistance with studies using TBSV to express the SBWMV 19k. The p26SBE-2 plasmid was obtained from Kay Scheets at Oklahoma State University and contains the 26S ribosomal RNA gene in pBluescript. This plasmid was used to prepare a DNA probe for membrane detection of rRNA [45]. The p26SBE-2 plasmid was digested with Bam HI and Eco RI and a 1 kb fragment corresponding to the 26S rRNA was recovered and labeled using the Dig High Prime DNA labeling system (Roche Applied Science Inc.). Competing interests The author(s) declare that they have no competing interests. Authors' contributions Jeannie Te did all cloning, plant inoculation experiments, gene silencing experiments. Ulrich Melcher did the amino acid sequence alignments and phylogenetic comparisons. Amanda Howard did some gene silencing experiments, photography. Jeanmarie Verchot-Lubicz conceived the study, did some molecular analysis, and wrote the paper. Special thanks to Wenpiny Qiu at Southwest Missouri State University for assistance with studies using TBSV to express the SBWMV 19k.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555535.xml
406390	Ras and Gpa2 Mediate One Branch of a Redundant Glucose Signaling Pathway in Yeast	Addition of glucose to starved yeast cells elicits a dramatic restructuring of the transcriptional and metabolic state of the cell. While many components of the signaling network responsible for this response have been identified, a comprehensive view of this network is lacking. We have used global analysis of gene expression to assess the roles of the small GTP-binding proteins, Ras2 and Gpa2, in mediating the transcriptional response to glucose. We find that 90% of the transcriptional changes in the cell attendant on glucose addition are recapitulated by activation of Ras2 or Gpa2. In addition, we find that protein kinase A (PKA) mediates all of the Ras2 and Gpa2 transcriptional effects. However, we also find that most of the transcriptional effects of glucose addition to wild-type cells are retained in strains containing a PKA unresponsive to changes in cAMP levels. Thus, most glucose-responsive genes are regulated redundantly by a Ras/PKA-dependent pathway and by one or more PKA-independent pathways. Computational analysis extracted RRPE/PAC as the major response element for Ras and glucose regulation and revealed additional response elements mediating glucose and Ras regulation. These studies provide a paradigm for extracting the topology of signal transduction pathways from expression data.	Introduction Complex intracellular networks inform a cell's developmental and growth decisions in response to external nutrients or signaling molecules. Defining the topology of such networks has generally relied on combinations of genetic epistasis and biochemical techniques to establish the linear order of components that convey information on the presence of a particular stimulus. Generally, only one or a few endpoints, such as enhanced transcription of a responsive gene, are monitored in gauging the output of a pathway. More recently, global transcriptional analysis has allowed reseachers to capture the entire transcriptional output of a signaling process and assess the consequence of eliminating individual components of the signaling network on the entire response ( Fambrough et al. 1999 ; Roberts et al. 2000 ). This approach has the potential to extract a complete description of a network from a relatively limited set of experimental perturbations. We have used global transcriptional analysis to dissect the signaling network activated by glucose addition to yeast cells, with an emphasis on the role of the small GTP-binding proteins, Ras2 and Gpa2, in that signaling process. Addition of glucose to yeast cells growing on a nonfermentable carbon source induces a dramatic restructuring of the metabolic and transcriptional state of the cell ( Johnston and Carlson 1992 ). At the metabolic level, the cell becomes reprogrammed for fermentative rather than oxidative growth. This involves the inactivation and repression of gluconeogenic enzymes and mitochondrially based oxidative phosphorylation processes and the induction of glycolytic enzymes. In addition, since yeast cells extract energy more efficiently from fermentable carbon sources, they are able to grow more rapidly and thus require an increase in the capacity for mass accumulation. This translates primarily into a need for increased protein synthetic capacity with an attendant increased production of ribosome components and other elements of the translational apparatus. The dramatic change in the metabolic activity and protein synthesis capacity attendant on glucose addition to starved cells is accompanied, and driven in part, by a reprogramming of the transcriptional state of the cell ( Johnston and Carlson 1992 ; DeRisi et al. 1997 ; Johnston 1999 ). Cells respond to glucose addition by repressing genes involved in the use of alternative carbon sources and in oxidative phosphorylation and by upregulating glucose-specific transport systems and glycolytic enzymes. Substantial work on glucose regulation of genes required for metabolism of alternate carbon sources, sometimes referred to as carbon catabolite repression, has identified a number of components of the network responsible for this repression and defined their interconnections ( Gancedo 1998 ). For instance, an AMP-stimulated kinase, Snf1/Snf4, inactivates a repressor, Mig1, thereby allowing transcription of genes normally repressed in the presence of readily fermentable carbon sources, and upregulates Cat8, an activator of gluconeogenic genes ( Carlson 1999 ). In addition, a number of transcriptional activators, such as Hap2/3/4, Adr1, etc., required for transcription of glucose-repressible genes, are inactivated by growth on fermentable carbon sources. Transcriptional upregulation of hexose transporters occurs by a glucose-induced degradation of Rgt1, a repressor of a number of glucose-induced genes ( Johnston 1999 ). The mechanism by which glucose regulates genes needed for increased translational capacity is less clear, although Rap1 and, more recently, Sfp1 and Fhl1 have been implicated as activators responsible for increased expression of growth-related genes in response to glucose ( Warner 1999 ; Jorgensen et al. 2002 ; Lee et al. 2002 ; Fingerman et al. 2003 ). However, it is not well defined whether the signal for such upregulation is the increased energy output or the presence of glucose per se. The small GTP-binding proteins, Ras1 and Ras2, play a role in the cell's adaptation to glucose by coupling cyclic AMP (cAMP) production to the presence of glucose in the medium ( Broach and Deschenes 1990 ; Tatchell 1993 ; Thevelein 1994 ). As in other organisms, yeast Ras proteins can transmit a regulatory signal by shuttling between an inactive GDP-bound form and an active GTP-bound form. In yeast, the GTP-bound Ras proteins stimulate adenylyl cyclase, encoded by CYR1 , to yield an increase in intracellular cAMP levels ( Toda et al. 1985 ). Addition of glucose to starved cells or cells growing on a nonfermentable carbon source yields within minutes a significant increase in intracellular cAMP concentrations, which rapidly decline to a level somewhat higher than that in prestimulated cells. This cAMP response to glucose is dependent on Ras. cAMP functions in yeast to liberate the yeast cAMP-dependent protein kinase A (PKA) catalytic subunit, encoded redundantly by TPK1, TPK2, and TPK3, from inhibition by the regulatory subunit encoded by BCY1 ( Toda et al. 1987 ). Active PKA can phosphorylate a number of proteins involved in transcription, energy metabolism, cell cycle progression, and accumulation of glycogen and trehalose ( Broach and Deschenes 1990 ; Tatchell 1993 ; Thevelein 1994 ; Boy-Marcotte et al. 1998 ; Smith et al. 1998 ; Stanhill et al. 1999 ). Several epistasis experiments have suggested that in some cases Ras may also function upstream of a MAP kinase cascade in yeast, primarily to direct pseudohyphal growth under conditions of nutrient limitation ( Mosch et al. 1996 , 1999 ). GPA2, a member of the Gα family of heterotrimeric G proteins, also regulates cAMP levels through a pathway parallel to Ras ( Nakafuku et al. 1988 ). Gpa2 associates with a protein, encoded by GPR1, which is structurally related to seven-transmembrane, G-protein–coupled receptors and whose ligand may be fermentable sugars ( Yun et al. 1997 ; Xue et al. 1998 ; Lorenz et al. 2000 ). Several lines of evidence suggest that Gpa2 activates adenylyl cyclase in a Ras-independent fashion. Overexpression of Gpa2 yields increased cAMP levels in the cell and an activated allele of Gpa2, even in a ras2 background, induces phenotypes associated with activated PKA, such as heat-shock sensitivity, repression of Msn2/4-dependent transcription, induction of pseudohyphal development, and loss of cellular stores of glycogen and trehalose ( Nakafuku et al. 1988 ; Lorenz and Heitman 1997 ). Reciprocally, gpa2 is synthetically lethal with ras2 , a phenotype that is reversed by inactivation of PDE2, the major cAMP phosphodiesterase in the cell ( Kubler et al. 1997 ; Xue et al. 1998 ). Whether Gpa2 functions solely to modulate PKA or has other signaling functions has not been resolved. To address the role of Ras and Gpa2 in reconfiguring the yeast cell's transcriptional framework in response to glucose and to define the signaling network associated with glucose signaling, we examined the global transcriptional response of cells to glucose and compared the response to that of cells following induction of activated alleles of these two G proteins. The results of this analysis indicate that the vast majority of the transcriptional remodeling the cell undergoes in response to glucose addition can be recapitulated by induction of Ras2 or Gpa2. However, much of this change can also be accomplished in the absence of signaling through cAMP. This indicates that glucose signaling of transcriptional reorganization proceeds through redundant, overlapping pathways, only one of which is regulated by Ras2 or Gpa2. Results Activation of Ras2 or Gpa2 Recapitulates Most Glucose-Induced Transcriptional Changes In order to examine the role of Ras2 and Gpa2 in effecting transcriptional changes in the cell in response to glucose, we measured the global transcriptional response of yeast cells immediately following induction of an activated allele of RAS2 or GPA2 (designated RAS2* and GPA2* in the figures) and compared that to the changes following glucose addition to glycerol-grown cells. To focus on signaling events, rather than the transcriptional consequences of metabolic changes in the cell, we examined the transcriptional response as it changed immediately following addition of glucose. Similarly, to examine the effects of Ras2 or Gpa2 activation, we constructed gal1 strains that carried an activated form of RAS2 or GPA2 under control of the galactose-inducible GAL10 promoter. Since gal1 strains cannot metabolize galactose, addition of galactose resulted in induction of the activated RAS2 or GPA2 allele and a small number of other galactose-inducible genes, but resulted in no changes in the metabolic state of the cell. In a parallel set of experiments, we examined the transcriptional changes in response to glucose addition of yeast cells containing a PKA that is unresponsive to intracellular cAMP levels. The mutant PKA, referred to as tpk-w , lacks the regulatory subunit and two of the redundant catalytic subunits, with the third catalytic subunit crippled in its activity ( Cameron et al. 1988 ). As a consequence, such strains possess constitutive, low-level PKA activity that is unresponsive to changes in cAMP levels in the cell. Thus, changes in cellular behavior dependent on modulation of PKA activity should be abrogated in this strain. The results of both sets of experiments are available in Table S1 . The results of comparing Ras2 activation to glucose addition, provided in Figure 1 , indicate that most of the transcriptional changes in the cell immediately following addition of glucose to glycerol-grown cells are recapitulated by activation of Ras2. Prior to initiation of the experiment (during growth on glycerol), the expression pattern of all genes in the wild-type strain (W303 gal1 ) closely resembled that of the strain carrying the inducible activated Ras allele (W303 gal1 GAL10 p -RAS2 G19V ), with only 0.4% of the genes exhibiting greater than 3-fold differences in absolute expression levels ( Figure 1 A; r = 0.96). This reflects the isogenicity of the strains and indicates that the inducible RAS2 G19V allele is not expressed under these conditions. Figure 1 Glucose Stimulation and Ras2 or Gpa2 Activation Yield Similar Transcriptional Responses (A–E) Expression levels (represented as absolute intensity values from Affymetrix hybridization scans) of individual yeast genes (points) plotted for two different strains and conditions. Dotted red lines indicate 2-fold difference boundary. (A) Strain Y2864 (Wt) prior to glucose addition versus Y2866 (GAL-RAS2 * ) prior to galactose addition. (B) Strain Y2864 prior to glucose addition versus Y2876 (GAL-GPA2 * ) prior to galactose addition. (C) Strain Y2864 20 min after glucose addition versus 0 min after addition. (D) Strain Y2866 60 min after galactose addition versus Y2864 20 min after glucose addition. (E) Strain Y2876 60 min after galactose addition versus Y2864 20 min after glucose addition. Values are in log 10 . (F and G) Induction ratios (mRNA level at 60 min/mRNA level at 0 min) of genes in Y2866 (F) and Y2876 (G) versus induction ratios (mRNA level at 20 min/mRNA level at 0 min) for the same genes in Y2864. Values are in log 2 . Addition of glucose to wild-type cells yields a substantial and rapid change in the transcriptional profile of the cell. By 20 min postaddition, 22% of all genes changed expression by greater than 3-fold and 41% changed expression by 2-fold, with essentially the same number of genes increasing as decreasing ( Figure 1 C). This dramatic change in the transcriptional profile was substantially recapitulated by activation of Ras. By 60 min postinduction, the profile of gene expression in the activated strain closely resembled that of the wild-type strain stimulated with glucose ( Figure 1 D; r = 0.94). Of those genes exhibiting a change in expression levels of at least 3-fold following addition of glucose, greater than 92% of those showed at least a 2-fold change in the same direction following activation of Ras2 ( Figure 1 F). Thus, since glucose yields activation of Ras2 and since Ras2 activation yields changes in transcription that are substantially similar to those observed following addition of glucose, we conclude that a major portion of the glucose signaling pathway regulating transcription can proceed through cAMP via Ras2. Similar results emerge from analysis of expression changes following activation of Gpa2. Only 0.8% of all genes showed a greater than 3-fold difference in absolute expression levels between the wild-type strain and the strain carrying the inducible activated allele of Gpa2 during growth on glycerol ( Figure 1 B; r = 0.97). The pattern of expression at 1 h following activation of Gpa2 strongly resembles that at 20 min following addition of glucose to wild-type cells ( Figure 1 F; r = 0.93). However, the response following activation of Gpa2 under these conditions is not as robust as that following activation of Ras2 or addition of glucose. While the overall magnitude of the Ras2-induced response is essentially equivalent to that obtained by glucose addition, the overall magnitude of the Gpa2-induced response is only half that of the glucose-induced changes ( Figure 1 F and 1 G). Nonetheless, although somewhat muted, the pattern of transcriptional change induced by Gpa2 closely resembles that induced by glucose. These results are consistent with the hypothesis that the major role of Gpa2 in the cell is modulation of cAMP in response to the presence of a fermentable carbon source. Redundant Signaling Pathways Control Glucose- Regulated Genes To analyze the pattern of transcriptional response to glucose addition and cAMP induction, we used a partitional clustering algorithm to group genes on the basis of their behavior over all 32 samples analyzed ( Heyer et al. 1999 ). Prior to clustering, the expression levels of each gene over the 32 samples were normalized by subtracting from each value the average expression of that gene over all experiments and dividing by the standard deviation of the expression values. This procedure emphasizes the pattern of response of each gene over the experiments, rather than the absolute levels of response. This process yielded 144 clusters ranging in size from seven to 506 members each. By hierarchical clustering ( Eisen et al. 1998 ), these clusters were further organized into groups on the basis of the similarity of their patterns, yielding eight major classes exhibiting significant change in some respect over the course of the experiments. These classes, encompassing approximately 50% of all genes, are summarized in Table 1 , and the corresponding pattern of expression is shown in Figure 2 . The list of genes in each class is provided in Table S3 . Figure 2 Expression Patterns of Clustered Genes Diagrams show the patterns of expression of genes in the classes (Roman numerals) listed in Table 1 , which were clustered as described in Materials and Methods . Each line represents the average expression level of all genes in that cluster during the time course (20-min intervals over 1 h) in the strain and condition indicated. Absolute intensity values were normalized for each gene over all 32 conditions examined by subtracting the average expression level for that gene over the all conditions and dividing by the standard deviation for that gene. Thus, expression values ( y -axis units) are represented as the standard deviations of each time point from the average expression value for each gene over the entire set of experiments. Error bars indicate the standard deviation in expression values of all genes in the cluster at the indicated timepoint. Abbreviations: Wt + Glu, glucose addition to strain Y2864; Wt + Gal, galactose addition to strain Y2864; RAS2 * + Gal, galactose addition to strain Y2866; GPA2 * + Gal, galactose addition to strain Y2876; tpk-w + Glu, glucose addition to strain Y2872. Table 1 Functional Enrichment among Genes Clustered by Response to Glucose and Ras Activation Gene expression data from 32 experiments representing time course data with five strains were clustered as described in Materials and Methods. The number of genes in each supercluster (Class) is indicated, and the pattern of expression of the members of the group in strains Y2864 (wild type [WT]), Y2866 (Ras), Y2876 (Gpa2), and Y2872 ( tpk-w ) following addition of glucose (Glu) or galactose (Gal) is indicated (I, induced; R, repressed; –, no change). Genes involved in the indicated processes under Functional Association were enriched in the indicated class. The numbers of genes in the functional category present in the class and in the genome are indicated, along with the −log 10 P that the enrichment is random. Functional enrichment and logP values were determined using the Gene Ontology (GO) term finder in the SGD Web site ( http://www.yeastgenome.org/) In general, glucose addition yielded a rapid change in expression of genes, which remained unchanged or tended back to starting conditions at later times. We interpret this behavior to indicate that the initial response, seen at the 20 min timepoint, generally represents the response of genes to the signal initiated by addition of glucose. The later deviation from that initial response represents either adaptation of the signaling process or readjustment of expression as a consequence of the change in metabolism of the cell. In contrast, gene expression in response to activation of Ras2 or Gpa2 generally showed a lag of 20 min, followed by a monotonic change in expression over the remainder of the experiment. This is consistent with the expectation that the effects of induction of Ras2 or Gpa2 can be seen only after the new activated protein is transcribed and translated. Further, since under these conditions no significant changes in metabolism occur, the change in expression is due solely to activation of the signaling pathway. This reinforces the notion that the initial response of the cell to glucose is a signaling response, since the pattern of this monotonic change at later times, following activation of Ras2 or Gpa2, generally matches the initial response of those genes to glucose addition. If those genes induced by glucose and by activation of Ras2 are regulated by glucose solely through the Ras2–Gpa2–cAMP pathway, then we would anticipate that glucose-induced transcriptional alteration would be abrogated in a tpk-w strain. This is the case for a subset of glucose-affected genes (classes II and VI), indicating the existence of a glucose signaling pathway that relies solely on the Ras signaling pathway. Inversely, a subset of genes is activated (or repressed) by glucose in both the wild-type and tpk-w strains but is unaffected by activation of Ras2 or Gpa2, indicating the existence of a Ras2-independent glucose signaling pathway (class III). However, the vast majority of genes that respond to glucose are affected by Ras2 activation and also respond in the tpk-w background (classes I and V). This suggests that the majority of glucose-responsive genes are regulated by redundant pathways, one of which requires Ras2 and the other one(s) of which is Ras2 independent. Thus, the major transcriptional response of glucose addition diverges prior to activation of Ras2, but converges before gene activation. This is elaborated further in the Discussion. Ras and Gpa2 Signal Exclusively through PKA To assess the extent to which the effects on transcription of Ras2 activation are mediated by PKA, we examined the pattern of expression following activation of Ras2 in tpk-w cells compared to that in Tpk + cells. For those genes whose induction or repression by Ras2 is exerted through PKA, the tpk-w mutations would be expected to abrogate that response. In Figure 3 we plot the change in expression of each gene 60 min after galactose addition to the GAL10 p -RAS2 V19 tpk-w strain versus the change in expression of each gene 60 min after galactose addition to the GAL10 p -RAS2 V19 strain. As evident, almost all genes fail to respond to Ras2 activation in the tpk-w background. Of the 789 genes (out of 4,037 analyzed) in this experiment whose expression increased by more than 2-fold at 60 min following addition of galactose to the GAL10 p -RAS2 V19 strain, only 16 (2%) also showed increased expression through activation of Ras2 in the tpk-w background. Similarly, of the 1,121 genes whose expression decreased by more than 2-fold following activation of Ras2 in a wild-type background, only five (0.5%) also showed decreased expression in the tpk-w background. Repetition of these experiments using cDNA microarrays and direct Northern blot analysis of candidate genes failed to confirm that expression of any gene was altered by Ras induction in a tpk-w background (data not shown). Thus, we conclude that the entirety of the transcriptional response to Ras2 activation is mediated through PKA. Figure 3 Ras and Gpa2 Affect Transcription Exclusively through PKA (Top) Induction ratios (mRNA level at 60 min/mRNA level at 0 min) of genes in strain Y2873 ( y -axis) versus induction ratios (mRNA level at 60 min/mRNA level at 0 min) for the same genes in strain Y2866. Values are in log 2 . (Bottom) Similar analysis for strain Y2897 ( y -axis) versus strain Y2876. The results are similar for Gpa2 activation. As noted above, the response to Gpa2 activation is not as robust as that to Ras2 activation, and, as noted in Figure 3 , the attenuation of the response to Gpa2 induction in a tpk-w strain is not as obvious as that seen with Ras2. Of the 444 genes in this experiment whose expression increased 2-fold or more in response to Gpa2 activation in a wild-type background, 75 (17%) also showed increased expression in the tpk-w background. Similarly, of the 831 genes whose expression decreased by 2-fold or more, 24 (3%) also showed decreased expression in the tpk-w background. However, multiple replicates of this experiment using cDNA microarrays failed to identify any gene consistently altered in transcription by Gpa2 in a tpk-w background. Thus, as with Ras, the vast majority, if not all, of Gpa2-responsive genes are regulated exclusively through PKA. Gpr1 Is Required for Efficient Glucose Response GPR1 encodes a protein structurally related to seven-transmembrane, G-protein–coupled receptors, and both biochemical and genetic evidence suggests it regulates Gpa2 activity in response to glucose ( Xue et al. 1998 ; Kraakman et al. 1999 ; Lorenz et al. 2000 ). Accordingly, to assess the role of Gpr1 in the cell's transcriptional response to glucose, we examined the global transcriptional pattern of isogenic GPR1 and gpr1 strains at 20-min intervals following glucose addition to glycerol-grown cells. Further, to assess the extent to which Gpr1-mediated signaling was processed through PKA, we performed a similar time course experiment with isogenic GPR1 tpk-w and gpr1 tpk-w strains. The full set of data is available in Table S2 . In both experiments we found that the overall transcriptional response (both induction and repression) was attenuated, although not eliminated, in the gpr1 strain relative to the GPR1 strain. For instance, for those genes whose expression changed by more than 50% following glucose addition to the GPR1 TPK strain, the average induction or repression ratio in the gpr1 strain was approximately half that in the GPR1 strain. K-means clustering of normalized data confirmed this general view ( Figure 4 ). For instance, cluster 1, which included 470 genes highly enriched in those involved in ribosome biosynthesis, exhibited on average induced expression in the GPR1 TPK strain following glucose addition, but no induction in the gpr1 TPK1 strain. Similar results were observed for genes in cluster 8, and induction of genes in clusters 4 and 6 was attenuated in the gpr1 strain compared to that in the GPR1 strain. Thus, these results are consistent with the hypothesis that Gpr1 participates in glucose signaling, but is not the sole mediator of that signaling. Figure 4 Loss of Gpr1 Diminishes the Glucose Response Diagrams show the patterns of expression of genes in clusters based on time course changes (20-min intervals over 1 h) in gene expression following glucose addition to the indicated strains ( GPR1 , Y2092; gpr1 , Y3159; GPR1 tpk-w , Y2857; gpr1 tpk-w , Y3077). For clustering, absolute intensity values were normalized for each gene over all 12 conditions examined by subtracting the average expression level for that gene over all conditions and dividing by the standard deviation for that gene, but the plotted expression values ( y -axis units) represent the average of the absolute intensity of expression (converted to log 2 ) of all the genes in the cluster at the indicated timepoint. Error bars indicate the standard deviation in expression values of all genes in the cluster at the indicated timepoint. The number of genes in each cluster and any highly enriched function group (including the p value) are indicated in each graph. The time course data from the tpk-w strain suggest that Gpr1 might affect multiple glucose signaling pathways. If a Gpr1-initiated signal were transmitted solely through PKA, then the pattern of gene expression following glucose addition to the gpr1 tpk-w strain would be essentially identical to that observed in the GPR1 tpk-w strain. While the correlation between the expression patterns of gpr1 tpk-w and GPR1 tpk-w ( r = 0.73) is higher than that between gpr1 TPK and GPR1 TPK ( r = 0.65), the patterns of expression of gpr1 tpk-w and GPR1 tpk-w, as highlighted by the cluster analysis, are similar but notably distinct (particularly in clusters 2, 4, 6, and 7). Thus, these results could suggest that Gpr1 impinges on both PKA-dependent and PKA-indepen-dent signaling pathways. Alternatively, the steady-state differences between gpr1 and GPR1 strains at the onset of the experiment could render the strains differentially responsive to glucose. This issue could be resolved by appropriate conditional alleles in GPR1 and TPK . Ras, Gpa2, and Glucose Induce Genes in Mass Accumulation and Repress Genes in Respiration and Mitochondrial Function We have addressed the nature of the genes regulated by glucose and Ras2 in two different but related ways. First, we asked how those genes that have been annotated as performing related functions behave on average over the set of experiments. Second, we have determined whether genes performing a common function are significantly overrepresented in any cluster of coexpressed genes. Both approaches give essentially the same results. In Figure 5 , we present the average level of expression of all the genes associated with the indicated function (as annotated by the Munich Information Center for Protein Sequences [MIPS] program) relative to that at time 0 in the wild-type strain. As evident, genes required for translation are upregulated by glucose and activation of Ras2 or Gpa2. This includes genes for RNA polymerase I and III subunits, cytoplasmic tRNA synthetases, rRNA and tRNA processing enzymes, translation initiation factors, and, to a slightly lesser degree, ribosomal proteins. Similarly, genes for these functional categories are highly enriched in those clusters in which expression increases following addition of glucose to wild-type or tpk-w cells or following activation of Ras2 or Gpa2 (see Table 1 ). Thus, a major portion of the transcriptional restructuring following glucose addition is directed toward enhancement of the translational machinery. Somewhat surprisingly, though, this is induced not solely by increased metabolism, but at least in part by a direct response to a signaling circuit, which is mediated at least in part by Ras2. Figure 5 Functional Analysis of Glucose- and Ras-Induced Expression Changes The average expression levels of genes grouped by the functional category listed on the right in the indicated strains over the 1-h time course are indicated by color (red, induced; green, repressed; yellow, unchanged). Values are relative to the expression level in strain Y2864 prior to glucose addition. The Functional Classification Catalog was obtained from MIPS at http://mips.gsf.de/proj/yeast/CYGD/db/index.html . Functional group analysis was performed using the ratio of vector magnitudes ( Kuruvilla et al. 2002 ). The computer source code was derived from http://www-schreiber.chem.harvard.edu . Strains: Y2864 (WT), Y2872 (tpk-w), Y2866 (RAS2), Y2873 (RAS2 tpk-w), Y2876 (GPA2), Y2897 (GPA2 tpk-w) . On the other side of the coin, genes involved in oxidative respiration, including components of the TCA cycle, oxidative phosphorylation apparatus, and ubiquinone (CoQ) synthesis, and all the genes required solely for gluconeogenesis are significantly downregulated both by glucose addition and by activation of Ras or Gpa2. These functional categories of genes are significantly overrepresented in that class of coexpressed genes that are downregulated in all conditions tested (class V). Thus, Ras2-dependent and Ras2-independent repression pathways redundantly regulate the restructuring associated with conversion from respiration to fermentation. Several groups of genes appear to be regulated by glucose exclusively through a PKA-dependent pathway. These are genes repressed by Ras2 or Gpa2 and by glucose in the wild-type strain, but not in the tpk-w strain (class VI), and include those involved in carbohydrate storage (trehalose and glycogen) and, to a large extent, in ubiquinone synthesis. A number of genes exhibit induction by glucose in an exclusively Ras2-dependent fashion and include genes involved in ribosome biogenesis. Reciprocally, a number of genes exhibit induction by glucose in a completely Ras-independent fashion. As noted in Figure 2 , expression of members of class III increases monotonically following glucose addition, in contrast to the pattern seen with genes in other induction classes, in which an initial rapid increase in expression following glucose addition is followed by an immediate stabilization or downshift. This may indicate that these genes are upregulated as a consequence of the metabolic changes or growth acceleration attendant on glucose addition. The enrichment of genes involved in DNA replication in this category is consistent with this hypothesis. Identification of Potential Transcription Factors Mediating the Response to Ras2 Activation We have used a number of computational approaches to identify potential regulatory sequences and regulatory factors responsible for changes in gene expression in response to glucose and/or Ras2 activation. All of these approaches are based on the assumption that genes exhibiting a common expression pattern over all the experiments are more likely to share a common regulatory sequence or respond to a common transcription factor (see Supporting Information). Several motifs (RRPE, PAC) and transcription factor-binding sites (Sfp1, Rap1, Fhl1) are associated with the class of genes induced by glucose through both a Ras-dependent and a Ras-independent pathway. Rap1- and Fhl1-binding sites have previously been associated with ribosomal protein genes ( Lieb et al. 2001 ; Lee et al. 2002 ), and the enrichment of these sites in this class represents the high proportion of ribosomal protein genes in the clusters comprising this class. Similarly, the RRPE and PAC motifs have been associated with genes encoding elements of the translational machinery and with genes that are upregulated following overexpression of Sfp1 ( Hughes et al. 2000 ; Wade et al. 2001 ; Jorgensen et al. 2002 ). Thus, these three transcription factors and their associated motifs are potential loci through which glucose and/or Ras activates transcription of translation-related genes. To evaluate whether the predicted motifs mediate Ras-activated transcription, we inserted each motif upstream of a reporter gene lacking any other upstream activation sequence (UAS) and then introduced the individual constructs into strains containing the inducible RAS2* or GPA2* alleles. As a positive control, we examined expression of the RPS18B promoter/enhancer region when it was fused to the reporter construct. As evident in Table 2 , activation of Ras2 or Gpa2 resulted in a 3-fold increase in expression of the reporter construct, consistent with the observation that expression of this gene increased following induction of either RAS2* or GPA2* in our genome-wide expression analysis. Having confirmed the ability of this system to detect Ras-responsive promoters, we examined the ability of the Rap1-binding site or the RRPE or PAC element to enhance transcription in response to activation of the Ras pathway. As noted in Tables 2 , 3 , and 4 , both the Rap1-binding site and the RRPE element yielded strong enhancer activity, especially when present in multiple copies. In contrast, the PAC element exhibited no enhancer activity. Further, the Rap1 enhancer activity increased modestly but consistently in glucose versus glycerol medium and following activation of Ras2 or Gpa2. Activation of Ras2 or Gpa2 also consistently yielded increased expression driven by the RRPE element. Finally, an MCB element provided modest enhancer activity that was further stimulated by growth on glucose but not by activation of Ras2 or Gpa2. This is consistent with the expression pattern of genes in the cluster in which the MCB motif is enriched. Table 2 Functional Analysis of Motifs: Potential Activator Elements Strains Y2864 (wild type [WT]), Y2866 (GAL-RAS) , and Y2876 (GAL-GPA2) were transformed with plasmid TBA23 (Vector), RPS18B (fusion of the promoter of RPS18B to lacZ ), or TBA23 DNA into which a 20-bp sequence spanning the indicated motif was inserted (2× and 3× indicate that two or three copies, respectively, of the motif oligonucleotide were present in the vector) and grown in SC media with either 5% glycerol (Gly), 2% galactose (Gal), or 2% glucose (Glu) as a carbon source. β-Galactosidase assays were performed on samples from three separate transformants, and the average specific activities (Miller units/OD 600 ) of the three samples are presented. Individual values differed from the mean by less than 10% for all measurements. NA, not available. Test sequences, derived from the indicated promoters, were TATGTGGTGTACGGATATGA (RAP1), TTCCGAAAATTTTCATTGGC (RRPE), GGGATGAGATGAGATGAGAT (PAC), and ACAAAAGACGCGTGAACTAA (MCB) Table 3 Functional Analysis of Motifs: Potential Repressor Elements A 20-bp sequence corresponding to the indicated motif from the indicated gene was inserted into plasmids consisting of TBA30, which were then transformed into strains and grown as described in Table 2 . Assays were performed in triplicate and all values differed from the mean by less than 10%. Values are β-galactosidase specific activities, with fold repression relative to the vector grown under similar conditions indicated in parentheses. Sequences used were GAACCTCGGCGGCAAAAATA (CAT8), GAAATATCCCTTAAAACTTC (SSE2), TTGTTACAGCCGCCCGTGGC (PDR10), and GAGGCAGCTTCCCTTCTGAT (FOX2) . See Table 2 caption for abbreviations Table 4 Functional Analysis of Motifs: PDR10 Element Is Not Ume6 Dependent Plasmids were transformed into the indicated ResGen (Invitrogen) wild-type (WT) and deletion strains, and transformants were assayed for β-galactosidase activity after growth on SC medium plus glucose. Fold repression is indicated in parentheses Several motifs were identified as correlated with repression by glucose and by Ras2 or Gpa2. These included binding sites for Rpn4, Ume6, Hap2/3/4, and Msn2/4 as well as several sequences of unknown association. We tested several of these motifs for their ability to mediate glucose- or Ras-induced transcriptional repression by inserting them between the CYC1 UAS and the promoter of a CYC1 - lacZ reporter construct and examining expression under different growth conditions. Most of the known elements manifested modest repression activity that was not enhanced by growth on glucose or by Ras or Gpa2 activation. However, multiple copies of an Ume6-like element from PDR10 elicited strong glucose-enhanced repression activity. As evident from Table 3 , the element caused 5- to 10-fold repression when cells were grown in glycerol and 500-fold repression when cells were grown in glucose. While this element exhibits some similarity to a Ume6-binding site, it does not mediate repression by Ume6. As noted in Table 4 , deletion of UME6 (or RPN4, MIG1, MIG2, MSN4, PHD1, RGM1, STD1, RIM101, SFL1, or NRG1; data not shown) did not alleviate the repressive effects of this element, although this deletion eliminated repression effected by a known Ume6-binding site from CAT8 . Repression by the PDR10 site was alleviated by deletion of TUP1 or SSN6 . Thus, this element likely functions by recruiting the Tup1/Ssn6 repressor complex to the promoter through a specific DNA binding factor intermediate. Given the lack of correspondence between the sequence of the element and known regulatory motifs, the element likely represents a novel glucose repression mechanism. Discussion Defining the Glucose Signal Transduction Pathway Transcriptional regulation by glucose has been examined extensively by genetic and biochemical analyses of specific glucose-repressible and glucose-inducible genes as well as by global transcriptional analysis ( DeRisi et al. 1997 ; Lutfiyya et al. 1998 ; Hughes et al. 2000 ; Wade et al. 2001 ; Jorgensen et al. 2002 ). These studies have highlighted pathways involved in connecting the presence of glucose with changes in the transcription state of the cell, particularly those pathways mediated by the Snf1/4 kinase and the Grr1 ubiquitin ligase ( Carlson 1999 ; Johnston 1999 ). Similarly, previous studies have demonstrated that the Ras/PKA pathway responds to glucose addition and affects gene expression, implicating Ras/PKA as a mediator of the cell's response to glucose. However, the overall topology of the glucose signaling network in yeast and the extent to which these different branches contribute and interconnect have not been previously addressed. The approach described in this report, following an earlier conceptual framework ( Roberts et al. 2000 ), provides a means of developing systematically a comprehensive topological map of the glucose signal network. Thus, this report is a first step in defining such a network. In this study, we have shown that most of the changes in transcription attendant on glucose addition can be recapitulated by activation of Ras2 or Gpa2. Thus, most of the glucose-induced changes in gene expression can be mediated by Ras2 and Gpa2. This is surprising since most transcriptional responses to glucose, particularly glucose repression, have been associated with Ras-independent mechanisms ( Gancedo 1998 ). In fact, though, since most of the glucose-induced transcriptional changes are also observed in a strain lacking a cAMP-responsive PKA, most of the glucose effects can also be mediated by a Ras/PKA-independent pathway. Thus, a minimal topology for the signaling pathway for modifying transcription in response to glucose comprises (1) redundant signaling pathways for repression and induction of the majority of genes, (2) a Ras/PKA-independent branch, and (3) a branch that is solely mediated by Ras/PKA ( Figure 6 ). Whether the redundant pathways reconverge at specific transcription factors or at the promoters of genes themselves remains to be determined. In addition, the relative contributions of known glucose regulatory circuits to the Ras-independent pathways, such as those mediated by Snf1/4 and Grr1, have not been determined. Studies similar to those described here for Ras are currently in progress with other contributing pathways. Figure 6 The Role of Ras and Gpa2 in Glucose Regulation of Transcription Diagram of information flow in glucose signal of transcription as deduced from global analysis of expression of genes in the strains used in this study. The number of genes regulated by each branch of the pathway, the nature of the regulation (red, induction; green, repression), and some of the functional categories of genes enriched in each branch are indicated. A redundant pathway for glucose signaling is consistent with previous observations suggesting that while activation of Ras/PKA elicits substantial changes in growth and carbohydrate metabolism in the cell, most of those changes can be effected even in the absence of an active Ras/PKA pathway. Cameron et al. (1988 ) constructed and analyzed tpk-w strains of yeast, which, as noted above, contain a PKA that is unresponsive to changes in cAMP levels. The authors found that tpk-w strains not only reverse all the phenotypes of bcy1 strains, but also regain the ability to respond to glucose depletion and readdition (glycogen accumulation, sporulation, etc.) in a timely and appropriate manner. Thus, the authors concluded that, while Ras/PKA could affect the cell's growth response to nutrients, one or more cAMP-indepen-dent pathways regulate the cell's response to nutrient availability. Under circumstances in which the cAMP signaling pathway is maintained at a moderate but constant level, this additional pathway(s) is sufficient for normal nutrient regulation. The presence of redundant glucose signaling in yeast could explain these earlier results. Most of the changes in transcription measured in these experiments likely result from the activity of a signal transduction pathway responsive to glucose, rather than from indirect effects due to changes in growth rate or metabolism. We saw the same global response whether the induction protocol was galactose addition in a gal1 background or addition of the gratuitous inducer β-estridiol to a strain with Ras2 or Gpa2 activation driven by a lexA-ER-VP16 chimeric transcription factor ( Louvion et al. 1993 ). Thus, the method of induction does not influence the results, ruling out any metabolic influences on the response. In addition, the glucose-induced transcription effects are observed early, likely prior to substantial reprogramming of the metabolic machinery of the cell. Transcriptional responses to glucose addition at later timepoints are often in opposite polarity to those at early timepoints, which suggests that the cell adapts its transcriptional response to the new conditions and emphasizes the importance of kinetic analysis in order to capture the structure of the signaling network under initial conditions. Several patterns of expression are not explained in a straightforward manner by the network depicted in Figure 6 . For instance, genes in class IV are induced by activation of either Ras2 or Gpa2 and by glucose addition to wild-type cells, but are repressed by glucose addition to tpk-w cells. Genes of class VII show the inverse behavior. One possible explanation is that the Ras-dependent and Ras-independent pathways have opposite effects on expression of these sets of genes. Alternatively, the physiology of the tpk-w cells may be significantly different than that of wild-type cells under initial conditions, such that the baseline expression of some genes at time 0 is significantly different in the two strains. In fact, the transcriptional profile of the tpk-w strain at time 0 is significantly different from that of the isogenic wild-type strain. This latter explanation may account for the behavior of genes in class VIII, which exhibit repression only by glucose addition to tpk-w cells. The behavior of class VIII genes may also suggest that some transcription factor activity or promoter activity is saturable, an hypothesis explored in more depth elsewhere ( Lin et al. 2003 ). Ras and Gpa Signal Exclusively through PKA We used epistasis analysis to define the functional topology of the Ras2 and Gpa2 branch of the glucose signaling pathway. That is, we examined the transcriptional consequences of activating Ras2 or Gpa2 in a background lacking a cAMP-responsive PKA. Since the readout of this experiment is the entire transcriptome of the cell, we can determine whether any gene is regulated by Ras in a PKA-independent fashion without knowing a priori what that gene might be. Our results demonstrate that all transcriptional effects of Ras2 and of Gpa2 are mediated by PKA. Previous studies have suggested that in certain strains Ras2 can activate the filamentous growth MAP kinase pathway ( Mosch et al. 1996 , 1999 ). Our results clearly indicate that in the strain examined grown under the conditions described, no such connection between Ras and the MAP kinase pathway exists. Further, identical epistasis experiments performed with diploid Σ1278 strains yielded the same result (data not shown). Thus, while Ras exerts PKA-independent effects on the yeast cell, all the transcriptional effects of Ras proceed through PKA. Substantial information has accumulated to suggest that, like Ras2, activated Gpa2 stimulates adenylyl cyclase, leading to an increase in cellular cAMP levels ( Kubler et al. 1997 ; Lorenz and Heitman 1997 ), although recent evidence suggests that Gpa2 might activate PKA directly (J. P. Hirsch, personal communication). Genetic epistasis data to date indicate that to activate PKA, Ras2 and Gpa2 proteins act in redundant parallel pathways, rather than in sequential steps in the same pathway ( Xue et al. 1998 ). However, whether activation of PKA is the sole activity of Gpa2 is not known. Our results indicate that, like Ras2, all of the transcriptional effects of Gpa2 are mediated by PKA. Consistent with that conclusion, we do not detect any group of genes whose expression is altered by activation of Gpa2 and is not also similarly altered by activation of Ras2. We do note that the intensity of transcriptional response following activation of Gpa2 is approximately half that seen following activation of Ras, suggesting that while both proteins function in similar roles, they have quantitatively different effects. Potential Transcriptional Network Various computational approaches identified a number of sequence motifs and transcription factors through which glucose and Ras2 or Gpa2 might be modulating transcription. The presence of the previously identified RRPE and PAC motifs is strongly correlated with genes induced following Ras2 activation. The pattern of genes induced by Ras2 closely resembles that of genes induced by increased expression of the Sfp1 transcription factor ( Jorgensen et al. 2002 ). We find that RRPE acts as a strong enhancer element in reporter gene constructs and that its enhancer activity is increased following activation of Ras2. Sfp1 contains several PKA consensus phosphorylation sites. However, evidence that Sfp1 acts directly through PAC/RRPE or that Sfp1 is the locus of PKA-induced activation is not yet available. Our studies also returned a strong correlation between genes induced by Ras2 and those containing Rap1-binding sites in their promoters, confirming the previously identified role of Rap1 in mediating PKA regulation of ribosomal protein gene expression ( Klein and Struhl 1994 ; Neuman-Silberberg et al. 1995 ). Recent results suggest that Rap1 binding to promoter sites serves to recruit the histone acetyl transferase Esa1 and that Rap1 binding is constitutive, but Esa1 recruitment is modulated by growth conditions ( Reid et al. 2000 ; Rohde and Cardenas 2003 ). Thus, PKA may affect the interaction of Rap1 and Esa1, an hypothesis currently under investigation. Our computational studies confirmed the presence of a number of motifs associated with glucose regulation and PKA, including the STRE element as well as binding sites for Hap2/3/4, Ume6, and Rpn4. Recent data have shown that PKA directly affects the nuclear localization of Msn2, one of the transcription factors that acts through STRE, but that PKA does so through a mechanism independent of the one responsive to environmental stress ( Gorner et al. 2002 ). Thus, the convergence of the glucose signal and the stress response signal on this transcription factor could account in part for the overlap of transcriptional response of the cell to glucose depletion and other forms of environmental stress ( Gasch et al. 2000 ; Causton et al. 2001 ). We also identified a motif associated with genes repressed by glucose through Ras-dependent and Ras-independent pathways. This motif provokes repression in reporter constructs that is substantially enhanced in growth on high levels of glucose, although the repression does not appear to be altered by PKA activation. While the motif bears resemblance to both Ume6 and Rpn4, it does not mediate repression by either factor, since deletion of either gene does not alleviate glucose-dependent repression by the motif. Thus, we have identified a novel glucose regulatory motif through these computational approaches. Further analysis of the many other motifs identified in this study could yield additional novel regulatory elements. Materials and Methods Strains All strains used in this study were derived from W303–1B and are listed in Table 5 . tpk-w alleles were isolated as described by Cameron et al. (1988 ) and confirmed by sequencing and retransformation of the mutant tpk2 allele. Construction of the galactose-inducible RAS2 G19V allele has been described by Fedor-Chaiken et al. (1990 ). The activated allele of GPA2 ( GPA2 Q300L ) was placed under the control of the GAL10 promoter (plasmid B2364), digested with ClaI, and integrated into the LEU2 locus of Y2864 and Y2895 to obtain strains Y2876 and Y2897. Yeast Consortium Deletion Strains created in the BY4742 background ( MAT α his3 Δ leu2 Δ lys2 Δ ura3 Δ) were obtained from Research Genetics (Invitrogen, Carlsbad, California, United States). The ho Δ strain was used as a wild-type control . Table 5 Strains Used in This Study a All strains used in this study were derived from Y2092 (W303-1B) Cell Growth Cells were streaked on YEPD plates and grown for 2–3 d at 30°C. Fresh colonies were inoculated into synthetic complete (SC) medium supplemented with 3% glycerol as the only carbon source. Cells were grown at 30°C and shaken at 200 rpm to an OD 600 of 0.25 (budding index, approximately 20%), at which time an aliquot of cells was removed as the time-0 control. Glucose or galactose was then added to 2% in the remaining culture and aliquots (40 ml) of cells were collected at 20, 40, and 60 min following sugar addition. Cells were mixed with 100 ml of prechilled water and quickly spun down by centrifugation at 2,500 rpm for 3 min at 4°C. RNA Isolation, Labeling, and Hybridization Cell pellets were lysed in TRI reagent (Molecular Research Center, Cincinnati, Ohio, United States) by vortexing with glass beads for 3 min. After a 5-min incubation at room temperature, 0.2 ml of chloroform per 1 ml of TRI reagent was added and mixed well with the homogenate. After centrifugation at 14,000 rpm for 15 min at 4°C, the upper aqueous phase was removed and precipitated with equal volume of isopropanol. RNA pellets were washed with 75% ethanol, air-dried, and dissolved in water. mRNA was purified from the total RNA with oligotex (Qiagen, Valencia, California, United States). First-strand cDNA was synthesized from mRNA using HPLC-purified T7-(dT) 24 primer (Genset, San Diego, California, United States) and SuperScript II RT (Invitrogen). Second-strand cDNA was synthesized using DNA ligase (10 U), DNA polymerase I (40 U), and RNase H (2U) from Invitrogen. Biotin-labeled cRNA was made with a BioArray HighYield RNA transcript labeling kit (Enzo Diagnostics, Farmingdale, New York, United States) and purified using an RNeasy mini-kit (Qiagen). The cRNA was fragmented, mixed with control cRNA cocktail, and hybridized to yeast genome S98 array (Affymetrix, Santa Clara, California, United States) for 16 h in a 45°C oven rotating at 60 rpm. The probe arrays were washed and stained using the GeneChip Fluidics station 400 (Affymetrix) and scanned at 570 nm with the Agilent GeneArray scanner (Affymetrix). For each experiment, we examined multiple timepoints, and for samples of significant interest we performed the experiment in triplicate. For initial analysis, we used MicroArray Suite 5.0 software (Affymetrix) to determine whether the hybridization signal for a gene was reliable and incorporated in our analysis only those measurements that were judged present, which generally included greater than 90% of the gene measurements in any one sample, with greater than 75% of all genes yielding reliable values over all the experiments. We also eliminated from our initial analysis those genes that were induced more than 3-fold in the gal1 strains by addition of galactose (25–30 genes, depending on the experiment). All experiments were normalized to the same total signal intensity. Data for all experiments can be obtained from Tables S1 , S2 , and S3 or at http://www.molbio.princeton.edu/labs/broach/microarray.htm . Computational Methods Expression clustering and motif discovery Partitional clustering of gene expression data was performed using the Qtclust algorithm ( Heyer et al. 1999 ), which creates a partitioning of genes into nonoverlapping clusters. Not all genes are assigned to clusters, as the members of each cluster are guaranteed to have a minimal intergene Pearson correlation (in our case, 0.75). In order to identify putative transcription factor-binding sites, the members of each cluster were used to search for common DNA sequence motifs in their 5′ upstream region using the AlignACE algorithm ( Tavazoie et al. 1999 ). For each cluster, three independent motif searches were performed. The resulting pool of approximately 5,000 motifs contained significant redundancy, as many known binding sites were identified multiple times. Using a motif similarity measure in the CompareACE algorithm ( Hughes et al. 2000 ), we clustered all the motifs into a largely nonredundant set of 251 members. In order to obtain a more “coarse-grained” view of genome-wide expression patterns, the original 144 clusters were combined by hierarchical clustering ( Eisen et al. 1998 ) of their mean expression profiles, yielding the eight classes discussed in the paper. Known transcription factor binding sites We assembled a set of weight matrices corresponding to 45 well-characteri z ed Saccharomyces cerevisiae transcription factors. These matrices were constructed from a mix of experimentally determined binding sites, augmented with extensive expression and chromatin IP-derived data ( Lee et al. 2002 ). To this list, we added three weight matrices (PAC, RRPE, A/T_repeat), which had strong computational evidence for being real transcription factor-binding sites. Statistical analysis To determine the statistical significance of functional enrichments in expression clusters, we used the hypergeometric distribution to quantify the chance probability of obtaining the observed overlap between an expression cluster and any of the 200 functional categories defined in the MIPS database ( Tavazoie et al. 1999 ; Mewes et al. 2002 ). The hypergeometric distribution was also used to quantify the probability of obtaining the observed overlap between expression clusters and the set of 300 genes with the highest-scoring occurrences of a motif in their 5′ upstream region. Reporter Gene Analysis Oligonucleotides containing motif sequences from selected promoters were cloned into the XhoI site of the CYC1 - lacZ reporter vectors pTBA23 and pTBA30, as described previously ( Mead et al. 1996 ). The former vector contains the CYC1 promoter and UAS, with the XhoI site residing between them, and the latter vector contains only the CYC1 promoter. Assays were performed on three separate transformants for each construct, grown as indicated. Results of β-galactosidase assays differed by less than 10% for triplicate measurements ( Gailus-Durner et al. 1997 ). Supporting Information Table S1 Gene Expression Patterns Following Glucose Addition and Ras2 and Gpa2 Activation Strains were as follows: Wild-type = Y2864 ( MATα ade2-1 can1-100 his3-11,15 leu2-3,112 trp1-1 ura3-1 gal1::HIS3 ); tpk-w = Y2872 ( MATα ade2-1 can1-100 his3-11,15 leu2-3,112 trp1-1 ura3-1 gal1::HIS3 tpk1::URA3 tpk2 V218G tpk3::KAN bcy1::LEU2 ); RAS* = Y2866 ( MATα ade2-1 can1-100 his3-11,15 leu2-3,112 trp1-1 ura3-1 gal1::HIS3 TRP1-GAL10-RAS2 V19 ); RAS tpk-w = Y2873 ( MATα ade2-1 can1-100 his3-11,15 leu2-3,112 trp1-1 ura3-1 gal1::HIS3 TRP1-GAL10-RAS2 V19 tpk1::URA3 tpk2 V218G tpk3::KAN bcy1::LEU2 ); GPA2* = Y2876 ( MATα ade2-1 can1-100 his3-11,15 leu2-3,112 trp1-1 ura3-1 gal1::HIS3 LEU2-GAL10-GPA2 Q300L ) GPA tpk-w = Y2897 ( MATα ade2-1 can1-100 his3-11,15 leu2-3,112 trp1-1 ura3-1 gal1::HIS3 LEU2-GAL10-GPA2 Q300L tpk1::HIS3 tpk2 V218G tpk3::TRP1 bcy1::hisG ) . Experimental conditions were as follows. Cells were grown in SC medium plus 3% glycerol to A 600 = 0.3. Glucose or galactose was added to 2%, and samples were removed at 0, 20, 40 and 60 min following the addition. Microarrays were performed as follows. RNA was isolated and labeled as described in Materials and Methods and hybridized to Affymetrix yeast genome S98 arrays. Data presentation is as follows. The first five columns provide the gene name (if known), the Saccharomyces Genome Database (SGD) gene designation, the MIPS functional category, and the function of the gene product, and the Affymetrix probe was set for that gene. For each time sample, the first column provides the normalized intensity values and the second column provides the determination from the MicroArray Suite 5.0 software as to whether the value was significant (P), insignificant (A), or indeterminate (M). The table is in a tab-delimited text format. (1.99 MB TXT). Click here for additional data file. Table S2 Gene Expression Patterns Following Glucose Addition to gpr1 and gpr1 tpkw Strains Strains were as follows: Y2092 = MATα ade2-1 can1-100 his3-11,15 leu2-3,112 trp1-1 ura3-1 ; Y3159 = MATα ade2-1 can1-100 his3-11,15 leu2-3,112 trp1-1 ura3-1 gpr1::hphMX ; Y2857 = MATα ade2-1 can1-100 his3-11,15 leu2-3,112 trp1-1 ura3-1tpk1::HIS3 tpk2 V218G tpk3::TRP1 bcy1::LEU2 ; Y3077 = MATα ade2-1 can1-100 his3-11,15 leu2-3,112 trp1-1 ura3-1tpk1::HIS3 tpk2 V218G tpk3::TRP1 bcy1::LEU2 gpr1::hphMX . Experimental conditions were as follows. Cells were grown in SC medium plus 3% glycerol to A 600 = 0.3. Glucose was added to 2%, and samples were removed at 0, 20, 40 and 60 min following the addition. Microarrays were performed as follows. Reference samples (RNA from 0 timepoint in each experiment) were labeled with Cy3, and each test sample (RNA from subsequent timepoints) was labeled with Cy5, mixed with the corresponding reference sample, and hybridized to cDNA microarrays printed in-house. Data presentation is as follows. Values are ratios of RNA levels for each gene at the indicated timepoint relative to the level for that gene at time 0 in that particular experiment. (477 KB TXT). Click here for additional data file. Table S3 Members of Gene Expression Classes The set of genes, specified by their SGD designation, comprising each of the gene expression classes listed in Table 1 and diagrammed in Figure 2 , is listed for each class. The table is a tab-delimited text file. (29 KB TXT). Click here for additional data file.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC406390.xml
524482	Control and maintenance of mammalian cell size: Response	A response to Cooper S: Control and maintenance of mammalian cell size. BMC Cell Biol 2004, 5 :35	Stephen Cooper was kind enough to send to us an original draft of his paper that now appears in BMC Cell Biology [ 1 ]. Although we have exchanged a number of e-mails with him in attempt to clarify points of confusion and disagreement, he continues to tilt at windmills and attack straw men. This is discouraging. His paper claims to be an analysis of our paper that was published in Journal of Biology last year [ 2 ]. Unfortunately, in much of his paper, he challenges our answers to questions that we did not address, conclusions that we did not draw, and arguments that we did not make. There are too many examples of this to deal with them all. One problem is semantic. He uses the term "cell growth" ambiguously, to mean both cell enlargement and cell number increase, which is confusing when discussing both cell growth and cell proliferation. In the Background, for example, he writes "How do cells maintain a constant cell size and cell size distribution during extended cell growth?". There is more important confusion in the phrase "linear growth". He uses it to mean the addition of an equal amount of mass at each stage of the cell cycle, and he claims that we use it in our paper in the same way. In our paper, however, we clearly defined linear growth to describe our observation that Schwann cells, blocked in S phase with aphidicolin, added a constant amount of volume and mass per cell over time, independent of their size. This confusion leads him to claim erroneously, in his Abstract and elsewhere, that we proposed that mammalian cells grow linearly during the division cycle; we do not believe this, and we did not test it or discuss it in our paper. Much of his paper is based on the premise that we were trying to understand how cell growth changes through the cell cycle. In fact, we have never addressed this question, in the paper or elsewhere. For this reason, much of Cooper's paper is not relevant to ours. Cooper criticizes individual experiments in our paper, but this too is almost always based on unnecessary misunderstanding. He accuses us, for example, of "an egregious error" in studying protein synthesis in Schwann cells that were not synchronized and therefore in all phases of the cell cycle. In fact, however, the cells were all arrested at the start of S phase with aphidicolin, as pointed out in both the text and figure legend. Despite its length, Cooper's paper never comes to grips with either the two main findings in our paper or the points of the experiments described in it. Unlike yeast cells ( S. pombe ) blocked in S phase by a mutation, which grow faster as they enlarge [ 3 ], we found that Schwann cells blocked in S phase with aphidicolin continue to grow at the same rate as they enlarge, adding a constant amount of volume and protein each day, independent of their size, We argued, as have others [ 4 ], that if big and little cells grow at the same rate (at the same point in the cell cycle), they do not need a cell-size checkpoint to maintain a constant distribution of sizes as they proliferate, unlike the situation for yeast cells. A second important difference from yeast cells that we found was that Schwann cells shifted from serum-free medium to serum-containing medium took 5–6 divisions and more than a week to attain the larger size of cells continuously proliferating in serum. This is what one would predict for cells that do not have a cell-size checkpoint and where little cells grow at the same rate as big cells at the same point in the cell cycle [ 2 , 4 ]. By contrast, when similar shift-up experiments are performed with yeast cells, the cells attain their new larger size within one cell cycle when shifted from a nutrient-poor culture medium to a richer medium [ 5 ]. We concluded that, if Schwann cells have cell-size checkpoints, they are very different form those that operate in yeast cells. Cooper also ignores our earlier findings that Schwann cell size at division depends simply on how fast the cells are growing and how fast they progress through the cell cycle and that both of these rates depend on the concentrations of extracellular signals that can regulate the two rates independently [ 6 ]. We found that GGF, for example, stimulated cell-cycle progression in these cells but not cell growth, whereas IGF-1 stimulated cell growth and synergized with GGF to stimulate cell-cycle progression. When IGF-1 concentration (and therefore cell growth) was held constant, an increase in the concentration of GGF drove the cells through the cycle faster; with less time to grow, the cells were smaller in high GGF compared to low GGF, at all stages of the cycle. These findings do not fit easily with Cooper's model that cell mass is the driving force of the cell cycle in all cells. Cooper's model for how cell growth and cell division can be coordinated is one version of a cell-size checkpoint model, in which progression through the cell cycle is somehow linked to cell size. Such models have been widely accepted in the cell-cycle field to explain how proliferating cells maintain their appropriate size over time [ 7 ]. Whereas the evidence for cell-size checkpoints in single-cell organisms is strong, the evidence for them in animal cells is weak, despite Cooper's arguments to the contrary. Our studies suggest that cultured Schwann cells (and we suspect many other animal cells) do not need, and probably do not have, such cell-size checkpoints to coordinate their growth and division. This difference between single-cell organisms and animal cells is not surprising given their very different life styles: in bacteria and yeasts, cell growth and proliferation are controlled mainly by nutrients, whereas in animals, they are controlled mainly by signals from other cells.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524482.xml
555938	Quantitative inference of dynamic regulatory pathways via microarray data	Background The cellular signaling pathway (network) is one of the main topics of organismic investigations. The intracellular interactions between genes in a signaling pathway are considered as the foundation of functional genomics. Thus, what genes and how much they influence each other through transcriptional binding or physical interactions are essential problems. Under the synchronous measures of gene expression via a microarray chip, an amount of dynamic information is embedded and remains to be discovered. Using a systematically dynamic modeling approach, we explore the causal relationship among genes in cellular signaling pathways from the system biology approach. Results In this study, a second-order dynamic model is developed to describe the regulatory mechanism of a target gene from the upstream causality point of view. From the expression profile and dynamic model of a target gene, we can estimate its upstream regulatory function. According to this upstream regulatory function, we would deduce the upstream regulatory genes with their regulatory abilities and activation delays, and then link up a regulatory pathway. Iteratively, these regulatory genes are considered as target genes to trace back their upstream regulatory genes. Then we could construct the regulatory pathway (or network) to the genome wide. In short, we can infer the genetic regulatory pathways from gene-expression profiles quantitatively, which can confirm some doubted paths or seek some unknown paths in a regulatory pathway (network). Finally, the proposed approach is validated by randomly reshuffling the time order of microarray data. Conclusion We focus our algorithm on the inference of regulatory abilities of the identified causal genes, and how much delay before they regulate the downstream genes. With this information, a regulatory pathway would be built up using microarray data. In the present study, two signaling pathways, i.e. circadian regulatory pathway in Arabidopsis thaliana and metabolic shift pathway from fermentation to respiration in yeast Saccharomyces cerevisiae , are reconstructed using microarray data to evaluate the performance of our proposed method. In the circadian regulatory pathway, we identified mainly the interactions between the biological clock and the photoperiodic genes consistent with the known regulatory mechanisms. We also discovered the now less-known regulations between crytochrome and phytochrome. In the metabolic shift pathway, the casual relationship of enzymatic genes could be detected properly.	Background Biological phenomena at different organismic levels have revealed some sophisticated systematic architectures of cellular and physiological activities implicitly. These architectures were built upon the biochemical processes before the emergence of proteome and transcriptome [ 1 - 3 ]. Under the molecular machinery, the biochemical processes are mostly interpreted as frameworks of connectivity between biochemical compounds and proteins, which are synthesized from genes to function as transcription factors binding to regulatory sites of other genes, as enzymes catalyzing metabolic reactions, or as components of signal transduction pathways [ 4 - 6 ]. This implies that, in order to understand the molecular mechanism of genes in the control of intracellular or intercellular processes, the scope should be broadened from DNA sequences coding for proteins to the systems of genetic regulatory pathways determining which genes are expressed, when and where in the organism and to which extent [ 7 ]. In the experience of engineering field, the systematic architecture and dynamic model could investigate the characteristics of signaling regulatory pathways [ 8 ]. Therefore, how to construct the dynamic model of a signaling pathway from the system structure point of view might be the first key to the door of system biology. Most biological phenomena directly or indirectly influenced by genes such as metabolism, stress response, and cell cycle are well studied on the molecular basis. Thus, identification of a signal transduction pathway could be traced back to the genetic regulatory level. The rapid advances of genome sequencing and DNA microarray technology make possible the quantitative analysis of signaling pathway besides the qualitative analysis. More particularly, the embedded time-course feature of microarray data would promote the system analysis of signal regulatory pathways as well, which is very mature in the field of engineering. In addition to northern blots and reverse transcription-polymerase chain reaction (RT-PCR), which study a small number of genes in a single assay, the transcriptome analysis has, via DNA microarray technology [ 9 ], managed to achieve high-throughput monitoring of the almost genome-wide mRNA expression levels in living cells or tissues. Two types of available microarrays, the spotted cDNA and in situ synthesized oligonucletide [ 10 ] chips, which permit the spatiotemporal expression levels of genes to be rapidly measured in a massively parallel way, are used in different experimental requirements and stocked in the databases on net, such as Stanford Microarray Database (SMD) [ 11 ], Gene Expression Omnibus(GEO) [ 12 ] in NCBI, and ArrayExpress [ 13 ] in EBI. Microarray experiments are now routinely used to collect large-scale time series data that facilitate quantitative genetic regulatory analysis while qualitative discussion is the traditional thinking [ 14 - 17 ]. Several analytic methods have been proposed to infer genetic interrelations from gene expression data. In the coarse-scale approach of clustering, the underlying conjecture is that co-expression is indicative of the co-regulation, thus clustering may identify genes that have similar functions or are involved in the related biological processes. The most widely used method is the unsupervised hierarchical clustering [ 18 ]. This approach has an increasing number of nested classes by similarity measurement and resembles a phylogenetic classification. If we know the number of clusters in advance, the k-means clustering [ 19 ] could assign gene elements into a fixed number k of clusters in a way to minimize the overall Pearson or Euclidean distances of each member internally in the same cluster. Other algorithms such as the neural-network-based self-organizing maps (SOM) [ 20 ], singular value decomposition (SVD) or principal component analysis (PCA) [ 21 ], and fuzzy clustering methods [ 19 ] also have their own advantages and limitations. Alternative supervised clustering algorithm of support vector machine [ 22 ], which uses prior biological information of cluster for training, would enhance the accuracy of clustering. However, the nature of clustering algorithms apparently cannot uncover the causal interactions between genes just by grouping. Regarding the causality of pathways, the clustering analysis needs to cooperate with sequence motif detection [ 23 ]. It is also important to note that models using clustering analysis are static and thus can not describe the dynamic evolution of gene expression, even in the type of time-course microarray data. A statistical model of Bayesian network [ 24 ] was proposed to model genetic regulatory networks. Basically, the technique uses a probabilistic score to evaluate the networks with respect to the expression data and searches for the network with the optimal score. The dynamic Bayesian network [ 25 ] was proposed to learn the network structure and parameters by maximizing the posterior probability via Bayes rule of prior probability and marginal likelihood. Another algorithm of Boolean networks [ 26 ] can also be employed to model the dynamic evolution of gene expression, where the state of a gene can be simplified to being either active (on, 1) or inactive (off, 0). The probabilistic nature of Bayesian networks is capable of handling noise inherent in both the biological processes and the microarray experiments. This makes Bayesian networks superior to Boolean networks, which are deterministic in nature. The validity of dynamic Bayesian networks is evaluated by the sensitivity-specificity score ratios [ 25 ], which depend on the training size, the degree of accuracy of prior assumption. A genetic regulatory network based on the first order differential equation with given decay rates was discussed in [ 27 ]. In this study, the dynamic system approach could be employed to model how a target gene's expression profile is regulated by its upstream regulatory genes from the system causality point of view. Then, with the causal dynamic model, the upstream regulatory function can be extracted from the expression profile of the target gene by the optimal estimation method, i.e. maximum likelihood estimation. Since merely the second-order differential equation is employed to model the dynamic evolution of the target gene, only a few parameters need to be estimated. Furthermore, the derived regulatory function is closely related to the causal upstream information of the pathway and will create a basis for inferring the regulatory pathway from the system biology point of view. In either eukaryote or prokaryote, signaling regulatory pathways are considered as responses to the physiological activities or the deviation from homeostasis, which would affect the normal states of an organism. Among these signaling regulatory pathways, cell cycle [ 17 ] is one of the most conspicuous features of life which plays an important role in growth and cellular differentiation in all organisms. In plants, the stress-induced pathways [ 28 ] are very important to survivability under the abiotic environmental treatment such as drought, salinity and cold[ 29 ]. If these critical pathways can be identified from quantitative analysis in silico , the defect of biological processes would be predicted and corrected before hand. Our aim is to construct signaling regulatory pathways quantitatively by the system inference approach with a dynamic model and microarray data. In this study, a second-order differential equation, which has been widely used to model many physical dynamic systems with good characteristics, is proposed to model the time-profile evolutional behavior of a target gene. The regulatory function is taken as the driving input of the dynamic equation of the target gene. Using the dynamic equation and microarray data, we first extract the regulatory function for each target gene. According to the extracted regulatory function, we deduce their upstream regulators to trace back upstream signaling pathways. Then, upstream regulatory genes are taken as target genes to trace back their upstream regulatory genes. Iteratively, we can construct the whole regulatory pathway to the genome wide using the dynamic regulatory model and microarray data from the system biology point of view. Finally, we give some independent validation of our approach by repeating the analysis with randomly reshuffling the time order of microarray data and see if the proposed pathways are destroyed. We have applied our dynamic system approach to two genetic regulatory pathways with microarray data sets publicly available on net [ 15 , 30 ]. One is the circadian regulatory pathway in Arabidopsis thaliana [ 31 , 32 ], and the other is the metabolic shift pathway from fermentation to respiration in yeast Saccharomyces cerevisiae [ 33 ]. The circadian system is an essential signaling pathway that allows organisms to adjust cellular and physiological processes in anticipation of periodic changes of light in the environment [ 34 - 38 ]. According to the synchronously dynamic evolution of microarray data, we have successively identified the core signaling transduction from light receptors to the endogenous biological clock [ 39 , 40 ], which is coupled to control the correlatively physiological activity with paces on a daily basis. On the other hand, the diauxic shift [ 41 ] from the exhausted fermentable sugar of anaerobic metabolism to aerobic growth is correlated with widespread changes in the expression of genes involved in fundamental cellular processes such as carbon metabolism, protein synthesis, and carbohydrate storage. [ 28 , 31 , 32 , 42 - 47 ] The architecture of the signaling pathway correlative to glycolysis or gluconeogenesis during the diauxic shift is properly built up. With the dynamic system approach, not only the regulatory abilities between causal genes could be derived, but also the delays of regulatory activity are specified. These quantitative characteristics will help determine the intrinsic frameworks of connectivity in the above interesting pathways from the system biology point of view. Results The proposed methods in this study would be divided into four steps. In the first step, a dynamic model using the second-order differential equation is developed to describe the expression profile data as output and the regulatory function as input to denote the implicit characteristics of each gene with some parameters. With the help of the second-order dynamic model, we would then extract the upstream regulatory function from the expression profile of the target gene using the optimal estimation method. In the third step, the regulatory function estimated will help seek the correlative regulatory signals from the upstream paths. Iteratively, we can reconstruct the whole signaling regulatory pathway by linking up the upstream regulatory paths. Finally, some biological filters using available biological knowledge are employed to prune the constructed signaling regulatory pathway to improve the accuracy of the proposed method. I. Dynamic system description of signaling regulatory model The second-order differential equation is well used in the description of dynamic system evolved from the causality of gene regulatory function. Let X i ( t ) denote the expression profile of the i -th gene at time point t . The following second-order differential equation is proposed to model the expression level of the i -th gene, where G i ( t ) is the upstream regulatory function to influence the expression profile X i ( t ) of the i -th gene while a i , and b i are the parameters that characterize the dynamic inherent property of the gene like degradation and oscillation, and ε i ( t ) is the noise of current microarray data or the residue of the model. In general, the second-order differential equation has been widely used to model dynamic systems to characterize efficiently the dynamic properties of damping and resonance of systems in physics and engineering. Obviously, the clue of upstream regulatory pathways is in G i ( t ). Thus, the first step is to detect the upstream regulatory function G i ( t ) from both dynamic equation in (1.1) and microarray data. However, to detect the input regulatory function G i ( t ) from both equation (1.1) and microarray data directly is not easy. In this situation, a Fourier decomposition technique is employed to decompose G i ( t ) as a synthesis of some harmonic sinusoid functions so that the signal detection problem of G i ( t ) is reduced to a simple parameter estimation problem. Accordingly, we can decompose G i ( t ) by the following Fourier series, Then the detection of G i ( t ) becomes how to estimate the Fourier coefficients of α n and β n , which are the magnitudes of different harmonics of cos( nωt ) and sin( nωt ), for n = 0,..., N in equation (1.2), respectively. In science and engineering, the Fourier series has been widely employed to synthesize any continuous functions with finite energy. The estimation of α n , β n and the detection of G i ( t ) in equation (1.2) are given in Methods in the sequel. As a result of parameter estimation in Methods, the detection of regulatory function G i ( t ) could be derived as follows, Since the input regulatory function G i ( t ) of a target gene is usually due to the transcriptional binding or some physical interactions from the upstream regulatory genes, in the following, we would trace back to the corresponding regulatory genes from input regulatory function of the target gene. II. Inference of the regulatory pathway via Apparently the input regulatory function G i ( t ) in equation (1.1) contains the driving information for the target gene's expression from the upstream regulatory genes. The identified regulatory function from equation (1.3) could be interpreted as the regulatory connectivity through transcriptional binding or protein-protein interaction imposed on the i -th target gene. Nevertheless, the expression data of protein type which should be considered directly in practice are by now unavailable and unreliable to trace back upstream regulatory genes. Instead, the expression data on mRNA level which is now widely available from microarray assays would make tracing back the upstream regulatory pathway possible under proper assumptions. All along the paper we assume that the expression levels of mRNA transcripts are proportional to the actual number of corresponding proteins in the cell. This assumption is indeed a strong approximation since post-transcription is known to play a very important role in down regulating the number of the transcription factor in the cell. Before the inference of upstream regulatory genes, it is reasonable to confine the effect of the regulatory genes on the regulated target gene. The saturated activity of expression level reveals that the regulatory ability cannot extend unlimitedly. The sigmoid function is often chosen to express the nonlinear saturation with proper parameters. Here, we apply the sigmoid transformation to represent the 'on' and 'off activities of the regulatory genes on binding or not to motifs of the target gene. So the regulatory signal shown below with the parameter set of θ j = { γ , M j , τ j } is the sigmoid transformation of X j ( t ), the expression profile of the j -th regulatory gene. where γ is the transition rate, M j is the mean expression of the j -th regulatory gene's profile, and τ j is the corresponding signal transduction delay. The delay activity should be considered in order to describe the signal transduction delay τ j from the j -th regulatory gene to the target gene. The delay τ j would be computed by statistical correlation between the regulatory signal transformed from the j -th regulatory gene and the identified regulatory function of the target gene. The delay τ j is determined by the following maximum correlation criterion, where r τ is the correlation between and under variable delay τ . If there are many τ to achieve the maximum correlation in (2.2), then only the smallest one is chosen. Using the correlation method, we trace back R i regulatory genes whose regulatory signals are most correlated with the regulatory function of the i th target gene, i.e. choose R i genes with maximum correlation but with smaller τ j in (2.2). The determination of number R i will be discussed later. Then, we construct the regulatory pathway by tracing back R i regulatory genes from the identified regulatory function of the target gene as the following kinetic relationship, where c ij is the pathway kinetic parameters from the regulatory gene j to the target gene i , R i are the searched upstream regulatory genes, the constant c i 0 represents the basal level to denote the regulatory function other than upstream regulatory genes, and e i ( t ) is the residue of the model. Furthermore, to estimate the pathway kinetic parameters c ij , equation (2.3) for m time points should be written in the following regression form, where , . We assume that each element in the error vector, e i ( t k ), k = {1,..., m }, is an independent random variable with a normal distribution with zero mean and variance σ 2 . By maximum likelihood parameters estimation method (see Methods), the estimates of σ 2 and Ω i are given as follows, which is solved as and It should be noted that with the combination of biological knowledge about the transcriptional factors, protein phosphorylation, post-transcriptional and specific enzyme regulation of target genes, lots of putative and verified genes correlated with the target genes are pruned by this biological filter for the more efficient and accurate searching of R i upstream regulatory genes in equation (2.3). For example, suppose the expression profile of gene j has a high correlation with regulation function G i ( t ) of target gene i. However, if gene j is not a transcription factor, protein phosphorylation, post-transcription or specific enzyme of target gene i, it will be deleted from the candidates of R i upstream regulatory genes because it may be only a co-expressed gene with the target gene corregulated by the other gene. On the contrary, a verified regulatory gene should be recruited into the candidates even with small correlation with G i ( t ). Finally, we take the well-known Akaike Information Criterion (AIC) into account for determining the number R i of regulatory signal [ 42 ], The first term in AIC is the residual variance and the second term R i is the number of regulatory genes. AIC includes both the estimated residual variance and model complexity in one statistic, which decreases as σ 2 decreases and increases as R i increases. AIC has been widely employed to determine the complexity of system modeling science and engineering [ 42 ]. The optimal number R i of the upstream regulatory genes will be determined by the minimization of the AIC value in equation(2.7). Now, for the selected target genes in the interesting pathway, we could search for the optimal R i upstream regulatory genes by AIC in equation (2.7) after the biological filtering and determine their pathway kinetic parameters c ij of regulatory signal by equation (2.6). After biological filter pruning, if the number of candidates of regulatory genes is still less than R i determined by AIC, then some genes, which are highly correlative to G i ( t ) but not of transcription factors or signaling proteins of target gene i, should be recruit into candidates to uncover regulatory relationships that were not suspected to be connected. After the combination of equations (2.3) and (1.1), the whole regulatory pathway is obtained as for i = {l, 2,..., L }, and L is the number of target genes in the pathway. The sub-paths related to the i -th target gene in the interesting pathway could be detected by the inference algorithm. Then, it is natural that the whole regulatory pathway would be constructed by the links of all the sub-paths. We also outline the whole flowchart of our dynamic inferring algorithm as shown in Figure 1 for an overview. Discussion Data set of analysis The two famous modeling organisms, Arabidopsis thaliana and yeast Saccharomyces cerevisiae , have been well studied biologically and their microarray assays are abundant. Thus, we chose different types of pathways, one is the plant behavior under environmental variation and the other is the cellular metabolism in response to exhaustion of external source, as examples in this study. In other words, two signaling pathways, i.e. circadian regulatory pathway in Arabidopsis thaliana and metabolic shift pathway from fermentation to respiration in yeast Saccharomyces cerevisiae , are constructed from microarray data to confirm the accuracy of our proposed method. For cells grown in the light/dark cycle according to circadian rhythm, Harmer and colleagues [ 15 ] used highly reproducible oligonucleotide-based arrays representing about 8200 different genes to determine steady-state mRNA levels in Arabidopsis thaliana that are measured in replicate hybridization of 12 samples harvested every 4 hours over 2 days. With their investigation on the circadian regulatory system, Harmer et al. have provided an abundance of correlated genes for the regulatory pathway inference. As for the metabolic pathway, an cDNA microarray assay from DeRisi et al. [ 30 ], containing approximately 6400 distinct expression sequence tags (ESTs) in yeast Saccharomyces cerevisiae , is harvested at seven successive 2-hour intervals after an initial nine hours of growth under the diauxic shift. Adoption of the diauxic shift data set would make possible the inference of metabolic shift pathways. Process of raw microarray data With the second-order equation and the optimal estimation method, the dynamic model should be developed first for the regulatory scheme of target genes in the signaling regulatory pathway. Because the raw microarray data sample of the biological assays that will be analyzed is small with less than 15 data points for an individual gene, the cubic spline method is used to interpolate the observed data to increase the data points of each gene's time-course microarray data. As shown in Figure 2 , the expression profiles of Cry1 (CRYTOCHROME 1) and PhyA (PHYTOCHROME A) genes in the circadian regulatory pathway of Arabidopsis thaliana are interpolated by the cubic spline method among raw data points on the left-hand side. Similarly, Pgi1 (PHOSPHOGLUCOSE ISOMERASE 1) and Pgm2 (PHOSPHOGLUCOMUTASE 2) genes in the metabolic shift pathway of yeast Saccharomyces cerevisiae are on the right-hand side. After the expression profiles are smoothed by the cubic spline technique, we can obtain the data of the first derivative and the second derivative more accurately and abundantly. Extraction of regulatory information After data expansion by the cubic spline method, we would have enough data to estimate the parameters of the regulatory dynamic model of the target gene from equation (2.4). Following the dynamic model in equation (3.1), the parameters which characterize the dynamic regulatory mechanism are estimated successfully for each target gene in the pathway. By dynamic model fitting, the expression profiles of the mentioned genes in Figure 2 can be reconstructed in Figure 3 with time progression again. Hence, we not only could predict the dynamic evolution of the target gene's expression profile accurately, but also deduce the regulatory function simultaneously as the scheme of Figure 4 . The regulatory information between target genes and their upstream genes can be extracted properly with this method. Inference of the regulatory pathway For illustrations, the inferring strategy is applied to the selected core genes (X 1 ~X 13 and Y 1 ~Y 11 ) in two pathways of the circadian regulatory system in Arabidopsis thaliana and the metabolic shift pathway in yeast Saccharomyces cerevisiae to recognize their upstream regulatory genes, respectively. Their regulatory abilities with signal transduction delays are shown in the form of dynamic equation in Table 1 and Table 2 , respectively. These regulatory abilities implying different degrees of influence are converted into a red-colored line as positive regulation (activation) and a blue-colored line as negative regulation (inhibition) for each target gene. Then, according to the dynamic regulatory equations in Table 1 and Table 2 , the pathways of the circadian regulatory system and the metabolic shift pathway are described in Figure 5 and Figure 6 , respectively. a. Pathway of circadian regulatory system The circadian rhythm controls processes ranging from cyanobacteria cell division to human wake-sleep cycles. In plant, especially for Arabidopsis thaliana , the growth and development have adapted to the diurnal cycling of light and dark [ 28 , 31 , 32 , 42 , 44 , 46 - 49 ]. The ability of plants to respond to light is achieved through some photoreceptors. Two classes of photoreceptors are well known to form the photo-transduction pathway under the circadian regulatory system in Arabidopsis thaliana [ 50 ]. One is the crytochrome of blue-light photoreceptors, containing Cry1 and Cry2 . The other is the phytochrome of mainly red-light photoreceptors, including PhyA , PhyB , PhyD and PhyE . In the photo-transduction related genes (Table 1 and Figure 5 ), containing both crytochrome ( Cry1 and Cry2 ) and phytochrome (PhyA, PhyB, PhyD and PhyE ) , Cry1 [X 6 ] and Cry2 [X 10 ] are commonly regulated by Lhy [X 3 ] (LATE ELONGATED HYPOCOTYL) in reciprocal ways with significant values (0.7569 in Eq.(6) and -1.8773, Eq.(10) of Table 1 , respectively), implying the essentially regulatory role of Lhy on crytochrome genes. In addition, from Eq.(10) in Table 1 , we further observe that Ccal [X 4 ] (CIRCADIAN CLOCK ASSOCIATED 1) has the greatest positive regulation (2.3465) on Cry2 , meaning that Cry2 is jointly regulated by Lhy and Ccal . Because the binding sites of Lhy and Ccal found in the promoter regions of Cry2 [ 51 ] are consistent with our inference, the transcriptional binding might be the mechanism of Cry2 affected by both Lhy and Ccal . In addition, the mutual activations of phosphorylation between Cry1 [X 6 ] and PhyA [X 7 ] in Eq.(6) and Eq.(7) of Table 1 are specifically identified consistent with the previous work [ 52 ]. At present, little is known about the nature of interactions between these two classes of photoreceptors. From Eq.(10) in Table 1 , Cry2 [X 10 ] is also positively regulated by PhyA [X 7 ] with 0.5-hr activation delay similar to that in Cry1 (Eq.(6) in Table 1 ). Therefore, PhyA is considered as a post-transcriptional regulator of phosphorylation to crytochrome within 1.0-hr after transcription. On the other hand, PhyB [X 11 ] down-regulates Cry2 with a significant effect (-0.7141) while Cry2 [X 10 ] up-regulates PhyB (0.0511) weakly by feedback (see Eqs.(10), (11) in Table 1 .). The mutual interactions between Cry2 and PhyB in nuclear speckles that are formed in a light-dependent fashion are also confirmed by Mas et al . [ 48 ]. Because Cry1 and Cry2 are both negatively co-regulated by PhyD [X 8 ] and PhyE [X 12 ] significantly (see Eqs.(6), (10) in Table 1 ), PhyA has apparently different behavior from PhyB , PhyD , and PhyE in activating crytochrome. This might suggest the mechanism that PhyA mediates the blue light by up-regulating Cry1 and Cry2 , whilst PhyB , PhyD , and PhyE would mediate the red light by inhibiting blue photoreceptors [ 53 , 54 ]. In the mainly red-light photoreceptors of phytochrome (PhyA, PhyB, PhyD and PhyE) in Figure 5 , undoubtedly Lhy [X 3 ] and Ccal [X 4 ], well-known biological clock genes in the circadian system [ 40 , 46 ], are core regulators involved in the transcriptions of both phytochrome (see Eqs.(7), (8), (11), and (12) in Table 1 ) and crytochrome (see Eqs.(6), (10) in Table 1 ) via feedback transcriptional binding. Similarly, Gi [X 15 ] (GIGANTEA) in Figure 5 has been identified as a manifested regulator to all the phytochromes (also see Eqs.(7), (8), (11), and (12) in Table 1 ), although Gi sequence lacks any motifs suggesting that it is a transcription factor of phytochromes [ 55 ]. Hence, Gi might be a post-transcriptional regulatory factor. However, there is another gene Elf3 [X 16 ] (EARLY FLOWERING 3) opposite to Gi on phytochrome, especially for PhyA, PhyB and PhyE (Eqs.(7), (11) and (12) in Table 1 ). Because of lower regulatory ability than transcription factor Lhy or Ccal , Elf3 might play the same role as Lhy and Ccal . Just as expected, Elf3 contains glutamine-rich motif suggesting that it is a transcription factor [ 56 ]. Before entrance of the biological oscillator of the circadian system formed by Toc1 , Lhy , and Ccal , a crucial gene of Pif3 [X 9 ] (Figure 5 ) is mediated significantly by PhyA [X 7 ] (-0.7631) and PhyB [X 11 ] (0.1223) (see Eq.(9) in Table 1 ). This is consistent with the post-transcriptional interactions of Pif3-PhyA and Pif3-PhyB complexes. As a core gene in the biological oscillator, Toc1 [X 13 ] is transcriptionally regulated by Lhy [X 3 ] (0.7009) and Ccal [X 4 ] (-1.4704) whilst Pif3 [X 16 ] (-0.1698) is presumably considered as the bridge between Toc1 and phytochrome (Eq.(13), Table 1 ). From Lhy and Ccal point of view, they are both positively affected simultaneously by Pif3 implying the regulation on the transcriptional level [ 57 ]. In addition, Toc1 inhibits both Lhy and Ccal to form the structure of mutual transcriptional regulation (please compare Eqs.(3), (4) with Eq.(13) in Table 1 ). So we conclude that Lhy and Ccal function as principal transcription factors. We also infer some downstream pathways of Chs [X 5 ] (CHALCONE SYNTHASE), Pap1 [X 1 ], and Co [X 2 ] (CONSTANS) in Figure 5 . Chs is known as correlated with UV-B protection. It seems that Ccal and Lhy have greater effect (2.7078, -0.7631, respectively) on Chs than Pap1 (-0.0455) as a transcription factor (Eq.(5) in Table 1 ). This might mean that Chs is regulated by Pap1 in a small scale with amplifying effect on the cis -regulatory level. Co is recognized as a pivotal gene of photoperiodic regulation of flowering. Indeed, strong regulations from Ccal and Lhy are identified to show that Co is regulated with a large-scale attenuation effect on the cis -regulatory level (Eq.(2) in Table 1 ). In the overview of the circadian system in Figure 5 , most red lines of activating regulation are found in the photo-transduction pathway between phytochrome (light blue ovals) and crytochrome (light yellow ovals) implying the chain interactions after the external light input. By the feedback regulations of Lhy and Ccal (orange ovals), represented by black lines with more linking to upstream genes, the photo-transduction pathways are stabilized to provide oscillation. On the other hand, more blue lines of inhibitive interactions are revealed in the biological-clock regulatory pathways relevant to Co , Pap1 , and Chs (light green ovals) underlying the anti-phase functional regulation between these output pathways and the oscillator. In addition, the essential signal transduction factors of Fkf1 , Gi , Elf3 , and Pif3 (gray ovals) make some critical links between the functional blocks mentioned above in the circadian system[ 58 ]. Finally, in order to validate the proposed approach, an independent validation is also given by randomly reshuffling the time order of microarray experiment [see Additional file 2 ] but with the same choices of target gene and regulatory genes, as shown in Figure 7 . It is seen that the proposed circadian regulatory pathway in Figure 5 is destroyed by reshuffling the experimental data. b. Metabolic shift pathway Sugars, such as glucose and sucrose, are excellent carbon sources for yeasts and almost all of the energy requirements of the cell can be satisfied by glycolysis [ 6 , 45 , 59 - 61 , 63 - 66 ]. Saccharomyces cerevisiae can switch from fermentatioon at high levels of glucose to respiration at low levels of glucose with major changes in metabolic activity (diauxic shift). In their experiment on the diauxic shift [ 30 ], DeRisi et al. inoculated cells from an exponentially growing culture into fresh medium and grew them at 30 for 21 hrs. This offers a resource to infer the possible allosteric regulation of enzymatic activities, protein modification and transcriptional regulation as shown in Table 2 . In addition, the scheme of the corresponding inferred pathway is shown in Figure 6 . In the overview of the inferring relationships in Table 2 , the gluconeogenesis from Pyk1 to pgm2 and the partial fermentation from Pyk1 (PYRUVATE KINASE 1) to Adh1 and Adh2 (ETHANOL DEHYDROGENASE ISOZYME 1, 2) are unraveled as a result of the diauxic shift, so two sub-pathways in opposite directions are concluded. In the fermentation direction, Pykl [Y 4 ] encoding an enzyme, which catalyzes PEP (Phosphoenolpyruvate) to pyruvate, is negatively regulated (-5.8763) by Pck1 [Y 26 ] (Eq.(4) in Table 2 ). Pck1 could be intepreted here as an indirect upstream transcription factor or regulatory gene for Pyk1 due to its function of decarboxylation and phosphorylation of oxalacetat in the presence of a nucleoside triphosphate and a divalent metal ion to yield PEP. Another Gcrl [Y 15 ] gene is also identified as the strongest positive regulation (5.9829) to Pyk1 (also see Eq.(4) in Table 2 ), which is putatively considered as a transcription factor. This candidate transcription factor Gcrl of Pdc1 [Y 6 ] (PYRUVATE DECARBOXYLASE ISOZYME 1) plays a more essential role (-2.5615, Eq.(6) in Table 2 ) than Rap1 [Y 17 ] (0.1164), and Pyk1 [Y 4 ] is an upstream regulatory factor coding an enzyme with the most positive effect (3.1295) on Pdc1 according to the production of acetaldehyde from pyruvate. In the last kernel of the fermentation, Adh1 [Y 7 ] and Adh2 [Y 3 ] are involved in the ethanol metabolism of carbohydrate storage. Adh2 is implicated to up-regulate Adh1 (0.5145, Eq.(7) in Table 2 ) under the catabolism from ethanol to acetaldehyde and is significantly up-regulated by Adh1 (1.0746, Eq.(3) in Table 2 ) to produce ethanol reversely. The mutual regulations of these two isozymes are within a tiny activation delay of 0.5-hr implying their close relationship. In addition, Gcr2 [Y 16 ] and Sfp1 [Y 30 ] with consistently dominant negative influences on Adh2 and Adh1 respectively would be at the transcriptional level presumably (see Eqs.(3), (7) in Table 2 ). In the sub-pathway of glyconeogenesis, Eno2 [Y 2 ] (ENOLASE ISOZYME 2) is regulated by Pck1 [Y 26 ] (-0.7195) in the same way as Pyk1 while the main transcription factor is Stp2 [Y 31 ] with significantly positive regulation (0.2147, see Eq.(2) in Table 2 ). As seen in Table 2 , a causal cascade of Eno2, Gpm1 [Y 8 ] (PHOSPHOGLYCERATE MUTASE l), Tpi1 [Y 10 ] (TRIOSE-P ISOMERASE 1), Fba1 [Y 11 ] (ALDOLASE 1), and Pgi1 [Y 9 ] (PHOSPHOGLUCOSE ISOMERASE 1) indicates the construction of a trunk of the glyconeogenesis (see Eqs.(8), (9), (10), and (11) in Table 2 ). Among them, Rap1 [Y 17 ] and Gcrl [Y 15 ] are the common regulators of Gpm1 , Pgi1 , and Tpi1 . This means that Rap1 and Gcrl might be the most important regulators in the glyconeogenesis pathway by the transcriptional binding. Finally, Pgm2 [Y 5 ] (PHOSPHOGLUCOMUTASE 2) co-regulated by Glk1 [Y 27 ] (GLUCOKINASE 1), Hxk1 [Y 28 ], and Hxk2 [Y 29 ] (HEXOKINASE 1, 2) significantly confers another pathway leading to the synthesis of UDP-GLU from Glucose-6-P (see Eq.(5) in Table 2 ). In the overview of the metabolic shift pathway in Figure 6 , extremely significant regulations (vivid red lines or blue lines) from most transcription factors (gray ovals) means that transcriptional regulations are feasibly identified. However, in the fermentation sub-pathway (light blue ovals), the mutual regulations between Adh1 and Adh2 are apparent when compared with the obscure relationships in glyconeogenesis (light yellow ovals). Interestingly, three transcription factors Gcr1, Gcr2 and Rap1 (black-line signals) appear to have very significant effects on the metabolic shift pathway. Finally, in order to validate the proposed method, an independent validation is also given by randomly reshuffling the time order of microarray experiment but with same choices of target gene and regulatory gene, as shown in Figure 8 . Obviously, the proposed metabolic shift in Figure 6 is destroyed by reshuffling the experimental data. Conclusion Microarray expression analysis by the dynamic system approach offers an opportunity to generate functional regulation interpretation on the genome-wide scale. The crucial ontology behind using dynamic system techniques is that the causality between gene expression profiles could be identified according to the differential equation underlying a dynamic system. Therefore, because the microarray data were harvested with time progression, the simultaneously varied gene expressions implicated in a genetic regulatory system would be detected to infer the regulatory pathways in spite of the versatile interactions such as transcriptional control, protein phosphorylation, or specific enzyme regulation. The clustering method answers the problem of what is the functional catalogue of a specific gene by the identification of resembling patterns of gene expressions. Similarly, the co-regulations of upstream genes in our method also imply their concurrent functions. In contrast to the clustering algorithm, the causality of time-course data has been smoothly drawn by our dynamic method. The Bayesian networks were used merely for forward probabilistic estimation with time transition lacking in the feedback linkages. This unidirectional problem would not happen in our algorithm. Owing to the quantitative regulatory abilities of our model, we have a greater diversity of regulatory influence than the Boolean networks, which are deterministic with merely two states. In our dynamic system approach, we not only can link qualitatively the upstream genes to the downstream ones iteratively, but also indicate quantitatively their regulatory relationships, including the regulatory abilities and the activation delays. In terms of the regulatory abilities, the comparison between the upstream regulatory genes of a target gene can inspire us to ask which one is significant biologically and whether it is a positive or negative influence on the investigated gene. Moreover, the speculation of activation delays benefits the empirical reference by providing us when the upstream regulatory genes might interact with their target genes. Since any gene can be considered as a target gene to trace back its upstream regulators, these regulators are then considered as target genes to trace back their upstream regulators. Iteratively, the genetic regulatory pathway (or network) can be constructed to the genome-wide. According to the qualitative and quantitative features imbedded, two regulatory pathway examples are characterized as in Figure 5 and Figure 6 for the identification of the proposed method. In addition, using the Akaike Information Criterion (AIC), a proper number of regulatory genes would be affirmed. As a result, many links overlap with well-known regulatory and signaling pathways in the previous literature and several putative ones are also found. Furthermore, the activation or repression relationships inferred via the microarray data would distinctly uncover the overall effect of regulatory interactions among casual genes in pathways on the transcriptional level. In the two pathways under investigation, we have a more detailed understanding about the regulatory interactions among phytochrome, crytochrome and biological clock in the circadian regulatory system. On the other hand, the sophisticated knowledge of the metabolic pathway after the diauxic shift can be unfolded properly in our analysis. Furthermore, the independent validation of our approach is also given by randomly reshuffling the time order of microarray experiment. We found that the proposed pathways in Figures 5 and 6 are all destroyed as shown in Figures 7 and 8 , respectively. The successful analysis of these two pathways implies the development of a valid and high-throughput method. All of the programs have been released [see Additional file 1 ] There are some shortcomings in our study. First, although the time-course microarray data are available, its lower samplings will distort the real changes of gene expressions, especially for quick dynamic evolution. A more sampling experiment with respect to the intrinsic turnover rate is expected to have more precise analysis. Secondly, a regulatory gene with larger activation delay would not be recognized because the less activation delay criterion is used, but this might be overcome by properly relaxing the criterion. Thirdly, activation profiles under the proteome should be highly correlated with the transcriptional profiles to elevate the interpretation of our system model. In general, the synchronous time-course microarray assay is more suitable to underlie the transcriptional binding among causal genes, but an inference of physical interactions in the post-transcriptional level also has sufficient feasibility in our study. In the near future, the most pressing task is to investigate our presumed paths in the laboratory. As the pathway construction algorithms are further developed, we expect this system approach to have immense impact in elucidating the underlying molecular mechanisms of pathways in a variety of organisms, especially after the maturation of the protein chips. Ultimately, we envision that biologists will perform routine pathway inference to seek some novel regulations and to identify the evolutionarily conserved links. Methods I. Detection of regulatory function Gi(t) in equation (1.2) After the decomposition of G i ( t ) in equation (1.2), we substitute equation (1.2) into equation (1.1) to obtain the following dynamic equation for the expression profile of the i -th gene, In the above dynamic equation, parameters a i , b i , α n , and β n should be estimated by the time profile of microarray data of the i -th gene, i.e. these parameters should be specified so that the simulating output X i ( t ) of the dynamic model in equation (3.1) should meet the empirical expression profile of the i -th gene. The least-squares estimation method is employed to solve this parameter estimation problem. To make the dynamic model effective, the dynamic equation in (3.1) should meet the expression profile at all time points t = t 1 ,…, t m and is then arranged in a vector differential form. Consequently, the vector differential form underlined in this equation is applied to m time points in order. where , , and m denotes the number of time points. In the next step, formula (3.2) can be translated into a differential matrix equation as follows, Y i = A i Φ i + E i (3.3) where , Φ i = [ a i b i α 0 β 0 … α N β N ] T , and are in vector forms, while is a matrix. To estimate the relevant unknown parameters in Φ i , the least-squares method below is used to derive the optimal parameters estimation of , Actually, the modeling error could be concluded into E i as the noise of the gene-expression profile or of the microarray chips. So the consideration of modeling error makes equation (3.3) approach more the reality. By the way, in order to get accurate data of and from the expression profile of the target gene, the cubic spline should be employed to interpolate the time profile of the target gene. Furthermore, the choice of N is based on the tradeoff between the accuracy of approximation in (1.2) and the complexity of parameter estimation in (3.4). In this study N = 6 is chosen because these harmonics are enough to approximate regulation functions. II. Maximum likelihood Estimate of kinetic parameters Ω i in equation (2.4) Maximum likelihood method for Ω i in equation (2.4) is given as follows: The log-likelihood function for given m data points is then described by The necessary condition for the maximum likelihood estimation of variance σ 2 is , by which equation (2.5) is obtained. Substituting equation (2.5) into equation (4.2) yields, meaning that we can find the maximum likelihood estimation of Ω i by minimizing the value of σ 2 in equation(2.5). Then, the maximum likelihood estimate in equation (2.6) is obtained by in equation(2.5). Authors' contributions W.C. Chang and Chang-Wei Li carried out the computational studies and analysis. B.S. Chen gave the topic and suggestions. All authors read and approved the final manuscript. Supplementary Material Additional File 1 The raw data has been transform to .mat file, which is one of the Matlab files. This file has been divided into three parts: Name, Time, and Profile. [see Additional file 1 ] Name denotes the names of microarray data. Time denotes the time points of microarray data. Profile denotes the gene profiles of microarray data. Click here for file Additional File 2 The 'Shuffled_data' is the Shuffled raw data. [see Additional file 2 ] Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555938.xml
520814	cDNA array-CGH profiling identifies genomic alterations specific to stage and MYCN-amplification in neuroblastoma	Background Recurrent non-random genomic alterations are the hallmarks of cancer and the characterization of these imbalances is critical to our understanding of tumorigenesis and cancer progression. Results We performed array-comparative genomic hybridization (A-CGH) on cDNA microarrays containing 42,000 elements in neuroblastoma (NB). We found that only two chromosomes (2p and 12q) had gene amplifications and all were in the MYCN amplified samples. There were 6 independent non-contiguous amplicons (10.4–69.4 Mb) on chromosome 2, and the largest contiguous region was 1.7 Mb bounded by NAG and an EST (clone: 757451); the smallest region was 27 Kb including an EST (clone: 241343), NCYM , and MYCN . Using a probabilistic approach to identify single copy number changes, we systemically investigated the genomic alterations occurring in Stage 1 and Stage 4 NBs with and without MYCN amplification (stage 1-, 4-, and 4+). We have not found genomic alterations universally present in all (100%) three subgroups of NBs. However we identified both common and unique patterns of genomic imbalance in NB including gain of 7q32, 17q21, 17q23-24 and loss of 3p21 were common to all three categories. Finally we confirm that the most frequent specific changes in Stage 4+ tumors were the loss of 1p36 with gain of 2p24-25 and they had fewer genomic alterations compared to either stage 1 or 4-, indicating that for this subgroup of poor risk NB requires a smaller number of genomic changes are required to develop the malignant phenotype. Conclusions cDNA A-CGH analysis is an efficient method for the detection and characterization of amplicons. Furthermore we were able to detect single copy number changes using our probabilistic approach and identified genomic alterations specific to stage and MYCN amplification.	Background Neuroblastoma (NB) is one of the most common pediatric solid tumors, and accounts for 7–10% of all childhood cancers. The prognosis of patients with NB varies according to the stage, age and MYCN amplification status. Stage 1 disease is essentially curable, whereas patients with stage 4 disease, in particular those with MYCN amplification, remain largely incurable despite advances in cancer therapeutics [ 1 ]. Genomic alterations in NB have been investigated by cytogenetic, and molecular methods including spectral karyotyping and metaphase comparative genomic hybridization (M-CGH) [ 2 - 6 ]. Based on these studies several genomic alterations have been reported to correlate with prognosis including amplification of the MYCN oncogene (found in 30% of NB) [ 1 , 7 ], gains of 17q (>50%) and loss of 1p36 (30–35%) [ 1 , 8 , 9 ]. Other recurrent changes including losses of 3p, 4p, 9p, 11q, and 14q, as well as frequent gain of chromosome 7 have also been suggested to have relevance to the development and progression of these tumors [ 9 ]. Recently array-based CGH (A-CGH) on BAC and cDNA microarrays has been used to investigate the genomic alterations with high resolution [ 10 - 14 ]. cDNA A-CGH has been successfully utilized to detect amplification and to investigate the direct effects of genomic changes over gene expression level by using the same microarray for both A-CGH and gene expression analysis [ 14 - 16 ]. In this study, we applied A-CGH, on cDNA microarrays containing 42,000 elements, to systematically identify common aberrant genomic alterations in NB of various stages. We have applied a probabilistic approach to detect single-copy losses and gains of chromosomal regions. Our study has three principal aims: 1) Detection and high resolution mapping of amplicons in NB. 2) Detection of low copy number genomic alterations using a probabilistic approach. 3) Establishing a map of genomic imbalances in NB profiling samples with good (stage 1) and poor (stage 4 with or without MYCN amplification) prognosis. Results Amplicon Mapping by A-CGH Totally around 24,000 qualified array cDNA clones were applied for data analysis in 12 NB cell lines and 32 NB primary tumor samples (see Table 1 for sample information). Fig. 1 shows the number of clones as well as the average spacing for each chromosome. We first determined the sensitivity of A-CGH to detect the copy number of highly amplified genes. We here chose MYCN since it is the most commonly amplified gene in NB and correlates with the biological behavior of these tumors. Fig. 2A shows the linear regression plot of the MYCN amplification results from A-CGH and Quantitative-PCR (Q-PCR). We found that the slope of the fitting line was 0.35, and therefore an observed ratio of 2 by A-CGH corresponds to Q-PCR ratio of ~6. In order to identify the amplified regions, we initially selected genes with A-CGH ratio ≥2 for at least two contiguous clones in genome sequence order. Only two chromosomes (2p and 12q) showed amplifications by this criterion exclusively in the MYCN amplified samples. Focusing on 2p (Fig. 2B ), we found 6 independent non-contiguous amplicons (10.4–69.4 Mb). For the MYCN amplicon, the largest contiguous region was 1.7 Mb and bounded by NAG and an EST (clone: 757451) in three tumor samples, whilst the smallest region was 27 Kb including an EST (clone: 241343), NCYM , and MYCN . We identified 9 previously reported co-amplified genes ( HPCAL1 , ODC1 , NSE1 , NAG , DDX1 , NCYM , POMC , DNMT3A , ALK, MEIS1, TEM8 ) [ 16 , 20 - 27 ], and detected the novel amplification of several known genes ( NCOA1 , ADCY3 , PPP1CB , CGI-127 , LBH , CAPN13 , GalNac-T10 , EHD3 , XDH , SRD5A2 , CGI-27 , AMP18 ) and ESTs. Three of the cell lines (CHP134, IMR-5 and IMR-32) contained two amplicons in 2p13-15. The first (66.6–67.6 Mb) included previously reported amplified gene MEIS1 , and the size of the second amplicon was 0.3 Mb (69.1–69.4 Mb), which was bounded by LOC200504 and TEM8 . In addition to chromosome 2p, we identified another amplicon on 12q14-q15 in a single tumor (NB21); bounded by PRO2268 (68.9 Mb) and RAB3IP (69.9 Mb) containing one previously reported amplified gene ( MDM2 ) [ 28 ] as well as several novel amplifications ( CPM , CPSF6 , LYZ , GAS41 , SNT-1 , CCT2 , VMD2L3 , and RAB3IP ) (Fig. 2C ). We verified the amplification of NSE1 , NAG , DDX1 , MYCN and TEM8 by Q-PCR (data not shown). Simultaneous gene expression profiling by using the same cDNA arrays for all samples showed that 47% of the amplified genes correlate with gene expression (using a correlation coefficient cutoff 0.5; data not shown). Table 1 Summary of Neuroblastoma Information Sample No Sample Label Age (yr.mo) Sex Diagnosis MYCN Source C1 CHP-134B 1.1 M 4 + NCI C2 GI-LI-N 1.11 M 4 + ICLC C3 IMR-32 1.1 M 4 + ATCC C4 IMR-5 ND ND ND + ICLC C5 LAN-1 2 M 4 + ICLC C6 LAN-5 0.4 M ND + NCI C7 SK-N-BE(2) 2.2 M 4 + ATCC C8 SK-N-DZ 2 F ND + ATCC C9 SMS-KCNR 1.2 M 4 + NCI C10 SH-SY5Y 4 F 4 - ATCC C11 SK-N-AS 8 F 4 - ATCC C12 SK-N-FI 11 M ND - ATCC C13 SK-N-SH 4 F 4 - ATCC T1 NB19 12.9 F 1 - DZNSG T2 NB20 1.3 M 1 - DZNSG T3 NB229 0.3 M 1 - CHTN T4 NB248 0.6 M 1 - CHW T5 NB29 0.3 M 1 - DZNSG T6 NB33 1.5 F 1 - DZNSG T7 NB34 1.2 M 1 - DZNSG T8 NB43 1.1 F 1 - DZNSG T9 NB44 1.6 M 1 - DZNSG T10 NB5 0.3 M 1 - DZNSG T11 NB7 1.3 F 1 - DZNSG T12 NB9 1.1 M 1 - DZNSG T13 NB16 3.11 F 4 - DZNSG T14 NB205 3.11 F 4 - CHTN T15 NB217 2 M 4 - CHTN T16 NB24 0.7 M 4 - DZNSG T17 NB246 3.7 M 4 - CHW T18 NB247 1.5 M 4 - CHW T19 NB26 1 M 4 - DZNSG T20 NB30 0.11 F 4 - DZNSG T21 NB31 1.4 F 4 - DZNSG T22 NB32 1.2 M 4 - DZNSG T23 NB35 2.7 F 4 - DZNSG T24 NB8 4.7 M 4 - DZNSG T25 NB14 0.11 M 4 + DZNSG T26 NB21 5.3 M 4 + DZNSG T27 NB249 0.8 M 4 + CHW T28 NB251 0.9 F 4 + CHW T29 NB252 0.10 F 4 + CHW T30 NB266 2 F 4 + CHW T31 NB27 10.6 M 4 + DZNSG T32 NB28 1.8 F 4 + DZNSG Abbreviations used are ND: not determined ATCC: American Type Culture Collection, CHTN: Cooperative Human Tissue Network CHW: The Children's Hospital at Westmead, Australia, DZNSG: German Cancer Research Center, Heidelberg, ICLC: Interlab Cell Line Collection, NCI: National Cancer Institute, NIH. 1 Naming convention "C" denotes cell lines, "T" tumors, such that C1 is cell line number 1. Figure 1 Distribution of cDNA clones in our microarray. Total 23975 unique UniGene clusters remained from the initial 42591 clones after quality filtering. Number of clones in each chromosome was represented in gray bar on the left side. Average spacing (chromosome size/number of clones in the chromosome) was represented in black dot on the right side. Figure 2 Amplifications in MYCN amplified samples . A. Regression analysis of MYCN ratio (sample vs . normal) obtained from A-CGH and real-time Q-PCR in all of neuroblastoma samples including 13 cell lines and 32 primary tumors. The slope of the regression line is 0.35 indicating that an observed A-CGH ratio of 2 is equivalent to a Q-PCR ratio of 5.7. B. Independent amplicons in chromosome 2p. All amplified genes are listed under each amplicon in genome order. Map position, genome sequence position (Mb) and samples containing the specific amplicon are listed for each amplicon. The percentage of the MYCN amplified samples harboring these amplicons are shown in brackets following the gene name for all clones present in our microarray (gray), the remainder of the clones are predicted genes found in the NCBI database that are mapped between the boundaries of the amplicon. Amplicons were selected based on the criteria of A-CGH ratio ≥2 for at least two contiguous clones in genome sequence order. In cases where a single clone has a ratio <2 but the ratio of its adjacent clones is greater than 2, that single clone was still considered as a part of amplicon. : previously reported amplification. C. Amplicon in chromosome 12q in tumor NB21. Detection of low-level DNA copy number alterations To test the sensitivity of A-CGH to detect single copy number changes, we performed A-CGH with DNA from cell lines containing different numbers of X chromosomes (1–5 copies) [ 12 ] and compared them to a sample with 2 copies of X chromosomes. The observed mean fluorescence ratio of all clones across X chromosome was calculated (Fig. 3A ). For single copy deletions we observed an A-CGH ratio 0.9 (expected 0.5). The regression slope was 0.3, similar to that for the MYCN above (Fig 2A ). The underestimation of the expected ratio by A-CGH demonstrated that it is difficult to detect single-copy changes using pre-set threshold-based approaches. Figure 3 Sensitivity of A-CGH to detect the low-level copy number alteration . A. Measurement of X-chromosomal copy number. A-CGH was performed to analyze the copy number of genes in the X-chromosome. Female DNA (XX) was used as the reference DNA. Male DNA (XY), female DNA (XX), and DNA samples containing different number of X-chromosome (XXX, XXXX, XXXXX) were used as test DNA, with an expected ratio of test/reference of 0.5, 1.0, 1.5, 2.0, and 2.5 respectively for X-chromosome. Mean fluorescence ratios (±SEM) of autosomal DNAs (blue diamonds) and X-chromosomal DNAs (red circles) from each experiment are shown. The slope of the regression line is 0.3. B. Visualization of p-values derived from the topological statistics as described in the Methods along the X-chromosome from samples containing different X-chromosomal copy numbers. Each column represents a different experiment; and each row represents the p value for the alteration at a given SW-locus (a sliding window of 40 adjacent clones, details in Methods), ordered by genome map position from Xpter to Xqter. Red represents gain and green loss. The intensity of the color shows the level of significance according to the p-value shown in the color scale. In order to increase the sensitivity for detecting low copy number changes, we applied a probabilistic approach utilizing t-statistics and the local genomic sequence mapping information of each of the cDNA clones on our arrays. To validate our method, we re-analyzed the A-CGH data generated from the cell lines containing 1–5 copies of the X chromosome as described above, and we were able to detect a single copy loss and gain of X chromosome where the expected ratio was 0.5 and 1.5 respectively (Fig. 3B ). In addition, we used the reported results from the literature as an independent validation. The cell line SK-N-AS is deleted within 1p36.2-p36.3, which has been investigated by FISH and southern blot analysis [ 29 , 30 ]. The proximal SK-N-AS deletion breakpoint was mapped to between NPPA and PLOD , while the distal breakpoint is proximal of CDC2L1 . The deletion detected by our method is bordered by KIAA0495 and CTNNBIP1 , which is within the region reported. They are in a very good accordance. In addition, we also compared the 17q gain results for four NB cell lines (CHP134, IMR-5, SMS-KCNR, and SK-N-AS) with the results in literature by FISH and Q-PCR [ 31 ]. Our results confirmed the gains in 17q for all 4 cell lines and the loss in 17p in SK-N-AS detected by FISH. We next analyzed the A-CGH data using this method to detect genome-wide alterations of DNA copy number in our NB samples. Using this t-statistics, we identified DNA copy number alterations that involved the majority of the chromosomes in both primary tumors and cell lines (Fig. 4 ). We confirmed previously reported genomic changes, including gains of whole chromosome 1, 2, 6, 7, 8, 12, 13, 17, 18 and 22, and losses of 3, 4, 9, 11, and 14; partial gains of 1q, 2p, 11p, 12q and 17q; partial losses of 1p, 3p, 4p, 9p, 11q and 14q [ 9 , 32 ]. The most common changes were losses on chromosome 1p, 4 and 11q; gains on 2p, 7, and 17q. Figure 4 Genome-wide analysis of DNA copy number alteration by A-CGH. Samples were grouped based on sample type, MYCN amplification status and tumor stage. Each column represents a different sample; and each row represents a p-value of a given SW-locus using a sliding window of 40 adjacent clones, ordered in genome order across the whole genome. Black triangles on the right side of image represent centromere positions. Cell-: cell line without MYCN amplification; Cell+: cell line with MYCN amplification; Stage 4-: tumor in stage 4 without MYCN amplification; Stage 4+: tumor in stage 4 with MYCN amplification. On the right is shown an enlarged view of the region around the MYCN gene (2p24) for the amplified NB samples. Stage specific genomic alterations In order to identify the recurrent regions of genomic alterations that are specific to stage and MYCN amplification status, we partitioned the tumors into three subgroups (stage 1, stage 4 MYCN not amplified (4-) and amplified (4+)), and analyzed the frequency of genomic changes at each SW-locus (as defined in Methods) for each subgroup. Since cell lines may contain tissue culture related genomic alterations, we only used primary NB tumor samples for this analysis. The frequency of alterations for a given SW-locus was estimated using the average probability ( P ) value as described in the Methods. Fig. 5 shows the graphic depiction of the P associated with each SW-locus for all possible pair-wise comparisons among the three subgroups for gains or losses: Stage 1 vs . 4-, 4+ vs . 4- and 1 vs . 4+. As expected, since the majority of the loci show no change, they were plotted to values of co-ordinates around (0.5, 0.5) (shown in black). Regions altered in both classes with similar frequency are plotted close to the y = x diagonal, whilst off-diagonal points represented regions primarily altered in one or the other of the classes (termed differential imbalance). Loci that plot around (0,0) reflect loci altered in both groups (termed common regions). The colored points were selected by using our criterion for "common" and "specific" SW-locus (see Methods). In summary, we found alterations that were common to all three subgroups, which included gain of 7q32, 17q21, 17q23-24 and loss of 3p21. We also found genomic imbalances that were specific for each of the subgroups and common regions of gain for stage 1 and 4- tumors. Of note there were no shared alterations of 4+ with 1 or 4- besides the regions common to all. A detailed description and map positions for all these recurrent regions are provided in the Table 2 , and a graphic representation of these imbalances is shown in Fig. 5 , where we will discuss in detail in the discussion section. Figure 5 Genomic alterations specific to stage and MYCN status. Shown is the graphic depiction of the average p-values ( P ) of genomic alteration of each SW-locus (using a sliding window of 20 adjacent clones) within each tumor subgroup. All possible pair-wise comparisons among the three subgroups for gains or losses (stage 1 vs 4-, 4+ vs 4- and 1 vs. 4+) are shown. The frequency of alteration is estimated by P such that P = 0.15 is equivalent to a frequency of 70% and the lower the P the higher the frequency (details in Methods). Different colors were used to represent different clusters. Magenta: loci common to all groups; cyan: common to 1 and 4-; blue: 1 specific; green: 4- specific; red: 4+ specific, and black: all remaining loci. Colored dots were enlarged for easier visualization. Table 2 Recurrent regions related to MYCN status and stage Imbalance Chr. Cytoband Start (Mb) End (Mb) Clone No. Common to 1, 4-, & 4+ gain 7 7q32 133.19 137.05 26 gain 17 17q21 42.13 53.53 157 gain 17 17q23 55.41 61.8 99 gain 17 17q24 65.59 73.44 70 loss 3 3p21 45.72 46.77 13 Shared by 1 & 4- gain 7 7p14 32.41 32.83 11 gain 7 7q11 69.89 72.34 28 gain 7 7q36 148.74 152.97 39 gain 17 17q12 35.45 36.41 15 gain 17 17q21 39.56 40.03 16 Specific to 1 gain 2 2p22 32.23 33.33 12 gain 17 17p13 2.12 2.59 11 gain 17 17p13 5.67 7.17 22 gain 17 17p13 8.23 9.42 17 loss 8 8p12 38.32 39.84 12 loss 8 8q22-23 109.54 116.64 13 loss 11 11q12 60.87 63.09 29 loss 14 14q12 28.94 31.32 11 loss 14 14q23 58.04 58.76 12 loss 19 19q13 43.4 44.05 13 Specific to 4- gain 7 7p15 24.26 25.89 12 gain 7 7q34 139.53 141.16 22 loss 11 11q21 96.25 96.83 11 loss 11 11q22 108.12 110.64 11 loss 11 11q23 114.31 120.39 51 loss 11 11q24 125.28 126.76 20 loss 11 11q25 131.98 134.08 19 Specific to 4+ gain 2 2p24-25 10.97 20.08 38 loss 1 1p36 6.16 13.42 78 loss 1 1p36 15.22 16.13 20 Discussion Amplicons in Neuroblastoma In our study we found that unlike breast cancers [ 14 ] NBs do not have a wide variety of different amplicons or amplified genes. We identified 6 independent amplicons on 2p and one on 12q and precisely defined boundaries for all amplicons. Several genes related to angiogenesis and oncogenesis were in these novel amplified regions including TEM8 (tumor-specific endothelial marker), mapped to 2p13.1, which has been shown to have elevated expression during tumor angiogenesis [ 33 ]. Indeed this gene was recently reported to be amplified in the cell line IMR-32 in accord with our data [ 27 ]. The gene is expressed in human endothelium and has been implicated in colorectal cancer. Our present study showed TEM8 was amplified and over-expressed (data not shown) in several neuroblastoma cell lines. The significance of amplification of TEM8 in neuroblastoma cell lines but not endothelial cells raises an intriguing possibility that these tumor cells themselves contribute to the angiogenic process and requires further investigation. We also identified amplification of GAS41 (glioma amplified sequence) mapped to 12q14-q15 in one tumor sample. GAS41 , a transcription factor ubiquitously expressed with the highest expression in human brain, was previously shown frequently amplified in human gliomas [ 34 ]. ALK (anaplastic lymphoma kinase) receptor, an oncogene and reported highly expressed in neuroblastoma [ 26 ], was identified to be amplified in two of our tumor samples. In addition to these genes, most of those newly identified amplified genes have not been implicated previously in neuroblastoma tumorigenesis and progression; therefore a further characterization of these genes might provide the biological insights to neuroblastoma biology. Interestingly, all amplicons occurred in MYCN amplified samples, and we have not found a single amplicon in MYCN single copy samples. Additionally, using our search criterion (A-CGH ratio >2 corresponding to copy number >6 see above) we found no evidence of amplifications in other chromosomal regions. This was in conflict with a study by Satito-Ohara et al. who found evidence of high level gains 9 NB cell lines and amplification as evidenced by a homogeneous staining region (HSR) in one line [ 35 ]. This difference could be explained by potential artifacts that arise in cell lines in tissue culture or be as a result of under detection by our study because of the relatively small number of tumor samples in our study, and would require confirmation in a larger sample set. Detection of low level of genomic changes In this study, we have applied a t-statistics-based method to explore genomic alterations in cancer from data generated by A-CGH on a cDNA microarray platform. Our method efficiently dealt with the low sensitivity of cDNA microarrays to detect low copy changes. The microarrays we utilized contain 42,000 clones, containing around 24,000 unique UniGene clusters with an average coverage of one cluster every 125 Kb. Underestimation of DNA copy number ratios by cDNA A-CGH data made it difficult to detect low level of gains and losses using ratio threshold based approaches, which was addressed in our study and previous reports [ 14 - 16 ]. The algorithm we have implemented included an efficient noise reduction strategy by combining ratios within a sliding window of clones as has been previously described [ 13 , 14 ]. However the incorporation of t-statistics demonstrated several advantages over the sole reliance of moving average for detecting genomic changes. It provides confidence levels for detecting genomic changes at each DNA location (SW-locus) in terms of p-values. This has theoretical advantages over the original raw ratio, because it incorporates an estimate of possible statistical errors in the analysis by giving a p-value attached to each genomic change within a SW-locus. By this method we were able to detect 1.5 fold changes of gene copy number as shown in our X-chromosome validation experiment. However, all these advantages are traded for a loss of resolution: genomic imbalances much smaller than the window-size cannot be detected and the boundaries of instable regions are blurred. Therefore we should choose the smallest window that has the desired level of statistical significance. The effective resolution can be obtained by analyzing the correlation of overlapping sliding windows. The integrated autocorrelation time is an estimator of the minimal distance for windows to be effectively uncorrelated [ 36 ] even when they overlap. For the sliding window t-test in our algorithm this distance can be calculated to be half the window size w, thus the number (N) of total unique Unigene clusters is reduced to 2N/w for the effectively independent measurements of the DNA copy number. Our results indicate that a window size 20 is needed in order to detect the lowest possible DNA change (one copy change) with reasonable statistical significance. According to the discussion above, this window size reduces the approximately 24,000 quality filtered unique Unigene clusters to 2 24,000/20 = 2400 independent estimates. This resolution is comparable to typical BAC arrays. For the stronger signals, less noise reduction is required. To detect 2-copy number DNA changes, only a small window size 5 is needed, therefore the resolution will be 4 fold higher. Although cDNA A-CGH is known not as sensitive as BAC A-CGH for the detection of low level of DNA copy number changes, currently we are able to obtain the comparable detection by using the probabilistic approach. In addition, with cDNA array it is possible to identify genomic amplification at the gene level and investigate the direct effect of gene copy number change over gene expression level in parallel, which will be addressed in future studies. Conclusions In this study we explored the genomic alterations in NB from the data generated by A-CGH on a cDNA microarray platform. We have not found genomic alterations universally present in all (100%) three subgroups of NBs, although such a region would be interesting since it may harbor specific genes that are uniquely responsible for NB tumorigenesis. We therefore focused on commonly altered regions where >70% of tumors showed changes in a given region, for our three different subgroups (Fig. 5 and Table 2 ). We found only a few of imbalances occurring in all three subgroups, of which gain of 17q21-24 and loss of 3p21 have been previously described in NB biology [ 8 , 37 ]. Apart from these regions stage 4+ tumors did not have any other regions that commonly change with the other two stages, whereas stages 1 and 4- had several common alterations. Stage 4- tumors demonstrated several unique changes of which losses in 11q has been previously described in MYCN single copy NB [ 38 ] and acts as a possible marker of unfavorable phenotype independent of MYCN amplification [ 39 ]. Remarkably stage 4+ disease appears to have very few genomic alterations when compared with Stage 1 and 4- implying that MYCN amplification is sufficient to drive these tumors to an aggressive phenotype, and although other genomic changes occur, including loss of 1p36 as shown by us and others [ 40 ], it does not require extensive changes. This is in agreement with the murine MYCN transgenic model of NB where the MYCN transgene itself is enough for tumor development, but these tumors develop additional genomic changes characteristic of NB [ 41 ]. Based on these results we found that cDNA A-CGH analysis is an efficient method for the detection and characterization of amplicons. We confirmed the previously reported amplified genes and also identified novel amplifications in neuroblastoma. Furthermore our probabilistic approach allows the detection of single copy number changes from cDNA A-CGH and can be applied to other CGH platforms including BAC or oligonucleotides based arrays. Methods Tumors, cell lines, and genomic DNA Thirty-two snap frozen neuroblastoma specimens were obtained from 12 patients with stage 1, and 20 patients with stage 4 of which 12 were MYCN -amplified and 8 were MYCN single-copy tumors. The original histological diagnoses were made at tertiary hospitals with extensive experience in diagnosis and management of neuroblastoma. Additionally, 12 neuroblastoma cell lines including 8 MYCN -amplified and 4 MYCN single copy samples were used in the study. Details of individual sample are summarized in Table 1 . The conditions for cell cultures were done as described previously [ 17 ]. High molecular weight genomic DNA was extracted from interphase of a Trizol preparation for RNA extraction according to the manufacturer's instructions (Invitrogen, Gaithersburg, MD). Genomic DNA was treated with RNase A and protease (Qiagen, Valencia, CA), and purified by phenol/chloroform extraction followed by ethanol precipitation. We obtained normal genomic DNA samples (male, female or 1:1 mixture of male and female) from Promega, and genomic DNA samples containing the different numbers of X chromosomes (XXX, XXXX, and XXXXX) from the NIGMS . Microarray experiments Preparation of glass cDNA microarrays was performed according to a previously published protocol [ 18 ]. Image analysis was performed using DeArray software [ 19 ]. The cDNA library containing 42,000 clones was obtained from Research Genetics (Huntsville, AL) and clones were printed on two microscope glass slides as a set. Approximately 50% of the cDNAs on the microarrays were either known genes or similar to known genes in other organisms, whereas the remainders were anonymous ESTs. For A-CGH experiments on cDNA microarrays, 20 μg of genomic DNA from neuroblastoma tumor or cell line samples were sonicated and purified with QIAquick PCR purification column (Qiagen, Valencia, CA). Three micrograms of sonicated DNA were labeled with aminoallyl-dUTP (Sigma) in a 25-μl reaction, including random hexamer (0.24 μg/μl, Roche), dATP, dCTP and dGTP (125 μM each), dTTP (25 μM), aminoallyl-dUTP (100 μM) and high concentration of Klenow fragment (2.5 U/μl, NEB). The labeling reaction was purified with QIAquick PCR purification column. Cy3 and Cy5 dyes were coupled to the reference DNA (1:1 mixture of normal male and female DNA) and sample DNA respectively. Cy3- and Cy5-labeled probes were then combined along with human Cot-1 DNA (50 μg, Invitrogen) and yeast tRNA (100 μg, Invitrogen). The mixture was concentrated and re-suspended in 32 μl of hybridization buffer (50% formamide, 10% dextran sulfate, 4 × SSC, and 2% SDS). The hybridization mix was first heated at 75°C for 10 min, then at 37°C for an hour, and finally loaded to the pre-hybridized array. The hybridization was performed at 37°C overnight. The washing procedure was performed as described previously [ 17 ]. Real time quantitative PCR Real time PCR was carried out using SYBR Green PCR core reagents according to the manufacturer's instructions (Applied Biosystems, Foster City, CA). Each DNA sample was analyzed in triplicate using the ABI PRISM 7000 Sequence Detector. For quantitative PCR, 10 ng of genomic DNA was used for SYBR green PCR assay. Serial dilutions of neuroblastoma cell line CHP134 DNA were used as templates for a standard curve, and the normal genomic DNA was used as a calibrator. The normalization was performed as described using BCMA and SDC4 as reference genes [ 20 ]. Data analysis Fluorescence ratios were normalized for each microarray by setting the average log ratio for each subarray elements equal to zero (commonly referred to as "pin-normalization"). The data was quality-filtered by removing those clones with quality lower than 0.5 in more than 20% of all the samples [ 19 ]. For the clones that passed this filter, if the quality for a specific sample is lower than 0.5, then its fluorescence ratio is replaced by the average ratio value of all other samples with the good quality. The clones were finally assigned to UniGene Cluster (Build 154 September 2002). For the UniGene clusters represented by multiple clones, mean fluorescence ratios of those clones are used. After these processes we had 23975 unique UniGene clusters remaining from the initial 42591 clones. Map positions for the cluster were assigned by Blat searches against the "Golden Path" genome assembly ( ; June, 2002 Freeze). Throughout this publication, all genomic coordinates are given with the respect to this assembly. Finally the clusters were sorted according to their starting position of sequence on each individual chromosome. Detection of single copy changes To identify the alterations of copy number along the genome, we compared the distribution of the ratios in a sliding window of 20 clones in genomic order with a "null distribution" using a t-test. The p-value for genomic change (P gc ) obtained in this way is assigned to the center of the sliding window (referred to as the "SW-locus" throughout this manuscript). The t-test is valid in this instance because the observed distribution of P gc for the loci in random order matches the expected theoretical distribution. The null-distribution used in t-test represents the unaltered part of genome. To identify the cDNA clones for the null-distribution, we start with the whole genome. The SW-loci corresponding to portions of the DNA that are amplified or deleted with a p-value smaller than 0.05 are removed recursively from the null dataset, until the null dataset is stable and there is no more amplified or deleted SW-locus in the null data set. Finally, the confidence of identified genomic alterations is visualized in a pseudo-color map in which color intensity represents the log of p-values (red for gain and green for loss). Estimation of frequency of genomic changes among the samples The probabilistic approach above provides P-values for the presence of genomic alteration in a given sample. In order to estimate the frequency of a genomic alteration we can set a threshold ratio and identify how many samples have a ratio value outside this threshold in the same genomic region. The disadvantage of this method is that different ratio thresholds will give different frequencies. We therefore applied another approach to avoid the use of ratio thresholds. To determine the frequency of loss or gain that correlate with the stage or MYCN amplification status, we first calculated the mean of the P gc or P , for each group. This value is proportional to the frequency f with change = N with change /N total of their occurrence, where N with change is the number of samples in a subgroup with a given genomic imbalance and N total is the overall number of samples in that subgroup. This is valid as follows. For all loci in which there are no genomic imbalances the observed P gc will follow the flat theoretical distribution with a mean (expectation value) < P > = 0.5. Therefore, for those cases where we are sure of a genomic imbalance P gc is close to 0 (for example p < 0.01), whereas for the samples in which there are no changes P gc = 0.5. According to the formula: P = ((N no change × 0.5) + (N with change × 0))/N total = (1 - f with change ) × 0.5, the lower the P the higher the frequency of a genomic change in that SW-locus. Thus the P can be used to determine the frequency of a given change e.g. P = 0.15 corresponds to a frequency of ~70% of the samples with that given change. Determination of recurrent regions We first define a SW-locus with P < 0.15 in a specific subgroup as altered. This threshold corresponds to roughly a fraction of >70% of all tumor samples harboring that alteration in each subgroup. We define an altered SW-locus as common in all tumors, if the SW-locus passes the threshold for each of the three subgroups. A SW-locus is called differential in one subgroup with respect to another subgroup, if the frequency of genomic change is at least 3 times higher in one subgroup as compared to another. A SW-locus is defined as specific if the locus in one subgroup is differential with respect to each of the other subgroups; a SW-locus is defined as shared in two groups, if in both groups it is differential with respect to the third subgroup. Abbreviations NB: neuroblastoma A-CGH: array-comparative genomic hybridization M-CGH: metaphase-comparative genomic hybridization Stage 1-: stage 1 without MYCN amplification Stage 4-: stage 4 without MYCN amplification Stage 4+: stage 4 with MYCN amplification BAC: Bacterial artificial chromosome Author's contributions QC carried out all experiments and participated in data analysis. SB performed array CGH data analysis and statistical analysis. QC and SB drafted the manuscript. JSW, CCW, NC and CS were involved in the microarray production and the manuscript edition. ALK and BG were involved in the data analysis and the manuscript edition. FW, FB, MS and DC provided the tumor samples and patient information and were also involved in the manuscript edition. JK principal investigator of the project, participated in its design and is the final editor of the manuscript. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC520814.xml
423138	Functional Fluorescent Ca2+ Indicator Proteins in Transgenic Mice under TET Control	Genetically encoded fluorescent calcium indicator proteins (FCIPs) are promising tools to study calcium dynamics in many activity-dependent molecular and cellular processes. Great hopes—for the measurement of population activity, in particular—have therefore been placed on calcium indicators derived from the green fluorescent protein and their expression in (selected) neuronal populations. Calcium transients can rise within milliseconds, making them suitable as reporters of fast neuronal activity. We here report the production of stable transgenic mouse lines with two different functional calcium indicators, inverse pericam and camgaroo-2, under the control of the tetracycline-inducible promoter. Using a variety of in vitro and in vivo assays, we find that stimuli known to increase intracellular calcium concentration (somatically triggered action potentials (APs) and synaptic and sensory stimulation) can cause substantial and rapid changes in FCIP fluorescence of inverse pericam and camgaroo-2.	Introduction Central to the study of neuronal networks is the simultaneous measurement of activity at many locations. While important results have been obtained using multiple patch recordings ( Stuart et al. 1993 ; Markram 1997 ; Markram et al. 1997 ) and microelectrode arrays ( Meister et al. 1994 ), patch recordings are limited to a few points and electrode arrays can only record spiking activity or compound field potentials. Furthermore, electrical recordings cannot resolve activity in fine branches of individual neurons and are blind to biochemical signals. Optophysiological approaches have, therefore, become strong competitors and complementors of valuable electrophysiological methods for studying neural activity. First attempts used intrinsic optical signals ( Cohen et al. 1968 ) followed by specific chromophores for sensing membrane voltage by absorption ( George et al. 1988 ) or fluorescence changes (for a review see Cohen et al. 1978 ). Other dyes were found ( Gorman and Thomas 1978 ) and later specifically designed that respond to changes in intracellular calcium (Ca 2+ ) concentration (for a review see Tsien 1992 ). Although changes in membrane potential are the most direct measurement of neuronal activity, the large fractional changes achievable with Ca 2+ -dependent fluorophores led to a rapid adoption of Ca 2+ measurements ( Tank et al. 1988 ; Ross et al. 1990 ; Sugimori and Llinas 1990 ), which acquired additional importance with the discovery that the induction of synaptic plasticity in many cases requires a substantial rise in local [Ca 2+ ] ( Malenka et al. 1988 ; Yang et al. 1999 ). Ca 2+ furthermore plays a role in morphological changes of neurites ( Yuste and Bonhoeffer 2001 ) and in gene regulation ( Morgan and Curran 1986 ). Loading a population of cells with Ca 2+ indicators has proven difficult in adult neural tissues. While there has been a recent advance ( Stosiek et al. 2003 ), it is unclear how cell-type specificity could ever be achieved by techniques of bulk loading synthetic indicators. Great excitement, therefore, greeted the molecular engineering, several years ago ( Miyawaki et al. 1997 ; Persechini et al. 1997 ), of GFP variants that are Ca 2+ -sensitive (fluorescent calcium indicator proteins [FCIPs]). Two classes of genetic Ca 2+ indicators have been designed that use different mechanisms of action. The first class, called “cameleons” ( Miyawaki et al. 1997 ; Miyawaki et al. 1999 ; Nagai et al. 2002 ), depends on changes in the efficiency of fluorescence resonance energy transfer between two spectral variants of green fluorescent protein (GFP) that are connected by a Ca 2+ -sensitive linker. The second class uses a single GFP fluorophore that contains a Ca 2+ -dependent protein as a sequence insert ( Baird et al. 1999 ; Griesbeck et al. 2001 ; Nagai et al. 2001 ). In addition to solving the loading problem, major advantages of genetic indicators are the prospect of targeting specific cell types by using appropriate promoters and the possibility of combining long-term studies of neuronal activity and morphology ( Grutzendler et al. 2002 ; Trachtenberg et al. 2002 ). The ability to express FCIPs in intact animals has in recent years allowed the measurement of [Ca 2+ ] transients in worm ( Kerr et al. 2000 ; Suzuki et al. 2003 ), fruitfly ( Fiala et al. 2002 ; Reiff et al. 2002 ; Liu et al. 2003 ; Wang et al. 2003 ; Yu et al. 2003 ), zebrafish ( Higashijima et al. 2003 ), and, more recently, mouse ( Ji et al. 2004 ). Thus far, there are still no reports of transgenic mice that express functional FCIPs in the brain. Clearly, expression of a functional indicator in the mammalian brain would enable the measurement of neuronal population activity with much higher spatial and temporal resolution than are offered by currently used noninvasive methods, such as functional magnetic resonance imaging, positron emission tomography, and intrinsic signal reflectance imaging. Moreover, in combination with two-photon imaging ( Denk et al. 1990 ; Denk and Svoboda 1997 ), transgenic indicators would allow the simultaneous recording of Ca 2+ signals in neurons and neuronal compartments from multiple sites in vitro and in vivo. In this paper we demonstrate that FCIPs can be transgenetically introduced into mice under the control of the tetracycline (TET) regulation system (for a review see Gossen and Bujard 2002 ) and are expressed and function widely throughout the nervous system. Results Construction of Transgenic Mice To select indicators for the generation of transgenic mice, we first screened a number of FCIPs in HeLa cells (see Materials and Methods ) for brightness, large [Ca 2+ ]-depen-dent fluorescence changes, and inducibility. The FCIPs were flash pericam, inverse pericam (IP), G-CaMP, camgaroo-2 (Cg2), and the cameleons YC 2.12 and YC 3.12 ( Griesbeck et al. 2001 ; Nagai et al. 2001 , 2002 ; Nakai et al. 2001 ). Figure 1 A depicts the genetic design of FCIPs. We found relative fluorescence changes (ΔF/F) of approximately +170% and −40% for Cg2 ( n = 5 cells) and IP ( n = 2 cells), respectively ( Figure 2 A), and therefore selected Cg2 and IP for the generation of transgenic animals. In addition, we chose YC3.12, which showed inconclusive results in the screening but is optimized for expression at 37 °C (see Figure 1 A). In our rough screen we did not find detectable responses for any of the other indicators. Figure 1 Genetic Designs of the FCIPs and the TET System (A) Genetic design of fluorescence Ca 2+ indicator proteins: (i) yellow fluorescent protein, (ii) Cg2, (iii) IP, and (iv) came-leon YC3.12. (B) Operating principles of the TET regulatory system (for details see Gossen and Bujard 2002 ). “L” indicates short linker sequence. Figure 2 Expression and Functional Tests in Cell Culture (A) HeLa cells expressing tTA and Ptetbi-luciferase/Cg2 or Ptetbi-luiferase/IP and imaged by confocal microscopy. Top row: low [Ca 2+ ] (0.1 μM), bottom row: high [Ca 2+ ] (25 mM) and ionomycin. Relative fluorescence changes are indicated as %ΔF/F. (B) Ratios (rlu-FL/rlu-RL) of FL to RL activity measured in mouse ear fibroblast cell cultures from all DNA-positive founders in the absence (red) and presence (green) of Dox (see Materials and Methods ). Circles (solid and open) indicate the lines that were selected for crossing to the transactivator lines. Solid circles indicate lines that showed smooth fluorescence. The TET system ( Figure 1 B; for a review see Gossen and Bujard 2002 ) was chosen because it allows combinatorial targeting of different neural cell populations using genetic crosses ( Mayford et al. 1996 ). In addition, we wanted to have temporal control over expression in order to test whether the indicator protein is inactivated by constitutive expression throughout development. The three selected FCIPs (Cg2, IP, and YC3.12), were placed under the control of the bidirectional TET promoter (Ptetbi) ( Baron et al. 1995 ). The opposite side of the Ptetbi contained the firefly luciferase (FL) gene ( Baron et al. 1995 ; Hasan et al. 2001 ). This allows, using the ear fibroblast method, the screening of founders for the presence of the functional gene without the need for a second generation of crosses into activator lines ( Schoenig and Bujard 2003 ). Of a total of 46 candidate founder animals ( Figure 2 B), the best four to six founders for each construct (judged by FL/renilla luciferase [RL] luminescence in the fibroblast assay, see Materials and Methods ; Figure 2 B) were selected for mating with mice expressing the TET-dependent transactivator (tTA) under control of the alpha-calmodulin/calcium-dependent kinase II (αCaMKII) promoter ( Mayford et al. 1995 , 1996 ). All selected founders showed, in addition to strong expression, efficient regulation of luciferase activity by doxycycline (Dox) (200- to 20,000-fold increase in luciferase activity with Dox; Figure 2 B). All of the following experiments were conducted in the absence of TET derivatives, leaving the controlled genes active. Dox Inducibility and Expression Patterns Brain slices from double-positive animals (i.e., harboring both Ptetbi-FCIP/luc and αCamKII-tTA genes) were used for analysis of expression patterns by immunohistochemistry ( Figure 3 A– 3 F) and two-photon microscopy (acute brain slices: Figure 3 H– 3 J; whole-mount retina: Figure 3 K). Expression of FCIPs was apparent only in double-positive animals and in the absence of Dox (strong to moderate levels in several lines; Figure 3 A– 3 E), and Dox strongly suppressed the expression of FCIPs ( Figure 3 A). Expression levels varied by line: high in MTH-YC3.12-7, MTH-YC3.12-8, MTH-Cg2-7, MTH-IP-12, and MTH-Cg2-19; moderate in MTH-Cg2-14 and MTH-IP-1; and low in the remaining lines (e.g., Figure 3 E; Figure 2 B, open circles). FCIP-positive cells included hippocampal and neocortical pyramidal cells and vomeronasal and main olfactory receptor neurons (see Figure 3 G for axon fiber projections in the accessory and main olfactory bulb), as well as granule cells and a few mitral cells in the olfactory bulb ( Figure 3 F and data not shown). Expression of FCIPs was robust in hippocampal areas CA1 and CA3 and in the mossy fiber area of the dentate gyrus (data not shown); cortical and retinal ganglion cell dendrites were clearly identifiable ( Figure 3 F, 3 J, and 3 K). The pattern of expression appeared to be a mosaic subset of that of αCamKII. There was little obvious variation between lines in gene-expression patterns in most areas except in hippocampal areas CA1 and CA3 and in the dentate gyrus (data not shown). In high- and moderate-expression lines (MTH-Cg2-7, MTH-Cg2-14, MTH-Cg2-19, MTH-IP-1, MTH-IP-12, MTH-YC3.12-7, and MTH-YC3.12-8; closed circles in Figure 2 B), consistent with their luciferase activities, cytosolic fluorescence ( Figure 3 H, 3 J, and 3 K) was smooth with occasional bright spots, with nuclei usually less fluorescent. Sometimes unusually bright neurons were seen, usually located near the slice surface and presumably damaged. In these cells the nuclei were as bright as or brighter than the cytosol ( Figure 3 H, arrow). In low-expressing lines (the majority: MTH-Cg2-[3, 6, 15, 17], MTH-IP-[5, 6, 15], and MTH-YC3.12-[3, 4, 5, 6]), fluorescence was punctate ( Figure 3 I). Figure 3 Doxycycline and tTA-Dependent FCIP Expression Immunohistochemical assay (A–F) using rabbit polyclonal GFP antibodies/peroxidase-DAB system: (A) YC3.12, single-positive (MTH-YC3.12-7), double-positive (MTH-YC3.12-7, αCamKII-tTA), and Dox-treated double-positive (MTH-YC3.12-7, αCamKII-tTA). (B) Cg2, single-positive (MTH-Cg2-7) and doubles-positive (MTH-Cg2-7, αCamKII-tTA). (C) IP, single-positive (MTH-IP-12) and double-positive (MTH-IP-12, αCamKII-tTA). (D) Moderate-expression line of Cg2 (MTH-Cg2-14, αCamKII-tTA). (E) Low-expression line (MTH-Cg2-15, αCamKII-tTA). (F) FCIP distribution in various brain areas. (G) Fluorescence in fixed brain slices from the accessory and the main olfactory bulb. (H–K) Two-photon images of acute, living brain slices. (H) Neurons in both CA1 and striatum usually show nuclear exclusion. (I) punctate expression in low-expressing lines (also see Figure 2 B, open circles); example from CA1 and cortex. Maximum intensity projection of two-photon 3D stacks taken from a brain slice (J) and a whole-mount retina (K). We compared the spectral properties of smooth and punctate fluorescence in one high-expression line (MTH-Cg2-7) and one moderate-expression line (MTH-Cg2-14), using confocal imaging spectrometry with excitation at 488 nm. The emission spectra of smooth and punctate fluorescence were similar to each other and to Cg2-expressing HEK cells ( Figure 4 A and 4 B; a two-photon image of punctate fluorescence is shown in Figure 4 B, right, arrowheads). Punctate fluorescence was occasionally observed in wild-type and double-negative (−/− Pbitet-FCIPs and −/− αCamKII-tTA ) mice, but there it had very different, much broader emission spectra ( Figure 4 C). In three of the seven double-positive mice examined, an additional distinct peak was seen at 600 nm, which was never seen in either single-positive or C57/BL6 wild-type mice ( Figure 4 D and data not shown). In the further analysis we concentrated mostly on the more promising smooth-fluorescence lines. Figure 4 Fluorescence Spectra and FCIP Mobility (A) Fluorescence distribution and emission spectra of Cg2 in cultured HEK cells and in neurons (dendrites and soma) in an acute brain slice (MTH-Cg2-7). (B) Punctate fluorescence and corresponding emission spectra (MTH-Cg2-7). “*” denotes emission spectrum of a punctate fluorescence in a different brain slice (not shown). Note smooth and punctate fluorescence also in the two-photon image on the right (MTH-Cg2-14). (C) Punctate fluorescence in a double-negative littermate of MTH-Cg2-7. (D) Image (left) and emission spectrum (right) of two-photon-excited fluorescence in an acute brain slice (MTH-Cg2-7). (E) Indicator mobility by two-photon fluorescence photobleaching recovery (IP, MTH-IP-12). To determine what proportion of fluorescent protein is bound or sequestered, and hence immobile, we performed two-photon fluorescence-recovery-after-photobleaching experiments ( Svoboda et al. 1996 ) on both somata and neurites of a high-expressing line (IP, MTH-IP-12) and found that roughly half of the indicator protein is mobile ( Figure 4 E and data not shown). Punctate fluorescence and the immobile fraction found in the bleach−recovery experiments suggest that a significant fraction of transgenetically expressed FCIPs interact with other cellular components, possibly via binding of the M13 or calmodulin sequences in FCIPs to their normal cellular targets. In Vivo Two-Photon Imaging To evaluate the achievable signal levels in intact animals, we performed in vivo two-photon imaging through the thinned skull in adult anesthetized mice (MTH-YC3.12-8, MTH-YC3.12-7, MTH-IP-1, and MTH-Cg2-7). Imaging up to and occasionally beyond a depth of 500 μm was possible ( Figure 5 ). Densely packed neurites were clearly visible, consistent with the staining patterns seen in acute slices (see Figure 3 J) and in histochemical preparations (see Figure 3 F). These results show that FCIP fluorescence is sufficiently strong for high resolution in vivo morphological imaging. Figure 5 In Vivo Two-Photon Imaging Through the Thinned Skull Yellow cameleon 3.12 at different depths (MTH-YC3.12-8) (A) and with high resolution (MTH-YC3.12-7) (B). (C) IP at different depths (MTH-IP-1). Functional Responses Next we tested the Ca 2+ response properties of FCIPs using a variety of different preparations and stimulation methods. Somatic recordings in slices Mainly to test temporal response characteristics, we performed a series of somatic electrical recording and synaptic stimulation experiments on pyramidal cells in brain slices. Targeted whole-cell tight-seal recordings of FCIP-expressing layer-2/3 cortical cells (the cell identity was confirmed by the overlap of fluorescence from the FCIP and that from Alexa 568, which was contained in the pipette) showed that FCIP-expressing cells have normal electrophysiological properties ( Figure 6 A). Somatic FCIP fluorescence (recorded with a CCD camera; Figure 6 A, left) showed small changes (ΔF/F ∼4%, 10 trials, MTH-Cg2-14; Figure 6 A, lower right) in response to trains of current-injection-triggered APs ( Figure 6 A, upper right). In hippocampal pyramidal cells in area CA1, recorded electrically with sharp high-resistance microelectrodes, two-photon scans of the somatic fluorescence showed larger changes (ΔF/F ∼10%, MTH-Cg2-19; Figure 6 B) with a smaller number of APs. The difference in response size may have been due to washout of FCIP into the patch pipette and to the lack of optical sectioning in the CCD measurements, which contributes to an unknown extent to the resting fluorescence from outside the recorded cell. For sharp-electrode recordings, transient fluorescence increases of up to 100% were usually seen during the break in (Cg2, MTH-Cg2-14, and MTH-Cg2-19; data not shown). Figure 6 FCIP Responses to Direct and to Synaptic Stimulation in Acute Brain Slices (A) Whole-field imaged responses of Cg2-positive cells in cortex to bursts of APs evoked by somatic current injection (whole-cell recording electrode indicated schematically); responses in the recorded (red) and in a nonrecorded (green) soma and in a region with no cell body (blue). (B) Two-photon line scan (lower trace) through the soma of a hippocampal CA1 pyramidal neuron during a burst of APs evoked by somatic current injection through a high-resistance microelectrode. (C) Whole-field-imaged responses to synaptic stimulation in cortex (five pulses at 100 Hz, 10 μA); ΔF/F image is shown below. Fluorescence and voltage responses with and without pharmacological block of glutamate channels (note suppression of APs and unmasking of inhibitory synaptic potentials). Synaptic stimulation First, cortical slices from a Cg2 animal (MTH-Cg2-14) were imaged using a CCD camera. Stimulation effectiveness was monitored in a distantly located soma (≈200 μm) by whole-cell tight-seal recording ( Figure 6 C). Short trains of stimuli (five pulses, 0.1 ms long, at 100 Hz, 10 μA; Figure 6 C, lower right) elicited fluorescence increases localized to an area near the stimulating electrode ( Figure 6 C, upper right). Peak ΔF/F ranged from 3%–8% (15 trials, in three slices). The fluorescence increase began in the frame following the stimulus onset ( Figure 6 C, right) and was as fast as responses seen in similar experiments with synthetic indicators ( Larkum et al. 2003 ). Smaller fluorescence changes were observed in the soma of the recorded neuron ( Figure 6 C, upper right, solid green trace). Small changes could also be seen in the neuropil as far as 150 μm from the stimulation site (data not shown). Fluorescence changes were largely abolished by glutamate-receptor blockers 6-cyano-7-nitroquinoxaline-2,3-dione (40 μM) and 2-amino-5-phosphovaleric acid (100 μM) ( Figure 6 C, upper right, dotted traces), indicating that they were mediated by synaptic activation. Similar experiments were performed using two-photon imaging, which allows optical sectioning and hence better spatial localization and signal-to-noise ratio. Synaptic stimulation again led to reproducible and rapid fluorescence changes. Changes were now much larger for both Cg2 and IP (ΔF/F ∼20%–100%, MTH-Cg2-19, 42 trials from three slices [ Figure 6 D]; ΔF/F approximately 15%–40%, MTH-Cg2-7, 35 trials from two slices [ Figure 6 E]; ΔF/F approximately −30%, MTH-IP-12, five trials from one slice [ Figure 6 F]). Changes were spatially inhomogeneous, showing “hotspot” structures possibly due to the activation of individual synaptic sites (see Figure 6 D, lower panel, trace 7; MTH-Cg2-19). Similar results were obtained in dentate gyrus mossy fibers ( Figure 6 E, note black and blue regions of interest in trace 3). In some experiments response amplitudes started to decrease after a few trials, presumably because of bleaching or photo damage (data not shown). Figure 6 Continued (D–F) Two-photon-imaged responses to synaptic stimulation in the hippocampus. (D) CA1 region with Schaffer collateral stimulation (eight individual response traces and the averaged trace are shown, region of interest indicated in the “response” image). Averaged images (five frames) during rest and response, and their difference, respectively. In localized hot spots, responses reach 100% (panels and traces shown below). (E) Similar response amplitudes and kinetics are seen in the dentate gyrus with mossy fiber stimulation (note that the number of stimuli was only 20). (F) IP responses recorded by a two-photon line scan through a CA1 soma; stimulation (20 pulses at 200 Hz) of neurites (approximately 50 μm away from the somata). Light-evoked responses in retinal whole mount In several mouse lines with YC3.12 (for example MTH-YC3.12-8), a subset of ganglion cells was strongly labeled (see Figure 3 K) but no light-induced Ca 2+ responses were seen (eight cells in two retinas tested), consistent with YC3.12 results in other tissues. In lines expressing Cg2 (MTH-Cg2-14), fluorescence became too weak and bleached too quickly for optophysiological measurements (eight cells in two retinas tested; data not shown). In one of two IP-expressing mice tested ( Figure 7 A and 7 B; MTH-IP-12), the fluorescence levels were high enough to follow axons and primary dendrites. After the onset of laser excitation, the fluorescence in the cells decreased and then stabilized at a slightly lower level ( Figure 7 C, asterisk). This effect was more pronounced at higher laser power (data not shown) and probably reflects the development of a steady state between photobleaching and diffusional replenishment from outside the excitation volume. Stimulation with spots of visible light evoked transient decreases in fluorescence, i.e., increases in intracellular [Ca 2+ ], in both soma and primary dendrites ( Figure 7 C and 7 D). Seven of 12 cells tested in the two IP mice displayed obvious light-evoked somatic Ca 2+ responses. The variation in response amplitude between cells may in part be due to heterogeneity of the labeled cell population. Figure 7 Light-Evoked Ca 2+ Responses in Retinal Ganglion Cells (A) Intact, light-sensitive retinal whole mount with Sulforhodamine 101 (red) in the extracellular space. Blood vessels are red; IP-positive (MTH-IP-12) retinal ganglion cells are green; and unstained ganglion cells are dark. (Scale bar: 50 μm). (B) Projection of an image stack reveals the IP-labeled primary dendrites of the retinal ganglion cells. (C) Time course of Ca 2+ response measured by high repetition rate image scan (62.5 Hz) of a soma: The cell responds with a decrease in fluorescence to the onset of the laser (asterisk) and to the repeated light stimulation (arrows). (D) Averaged (four repetitions) light-stimulus-evoked Ca 2+ response (black trace; gray traces are single trials) measured in the soma (above) and in the primary dendrite (below) of a retinal ganglion cell. In vivo imaging of odor responses in the olfactory bulb FCIPs are expressed in the olfactory bulb in afferent sensory axons and granule cells, with relative expression levels varying somewhat with FCIP type and across lines (data not shown). Substantial changes in fluorescence were observed for both IP and Cg2 (MTH-IP-12 and MTH-Cg2-19) in response to odor stimulation ( Figure 8 ). We do not know, however, the exact fraction of non-FCIP autofluorescence contained in the resting signal. Figure 8 In Vivo Imaging of Odor-Evoked Ca 2+ Signals with Transgenic Indicators in the Olfactory Bulb (A–C) IP (MTH-IP-12). (A) Raw fluorescence image. (B) Time course of fluorescence signal in the corresponding regions outlined in (C) (matching line colors). The black trace shows respiratory activity. (C) Color-coded map showing the relative change in fluorescence evoked by different odors in each pixel during the first second of the odor response. (D–F) Cg2 (MTH-Cg2-19). (D) Raw fluorescence image. (E) Time course of fluorescence signal in the corresponding regions outlined in (F) (matching line colors). (F) Color-coded maps showing the relative change in fluorescence evoked by different odors in each pixel during the first second of the odor response. Odor-evoked overall fluorescence changes were, as expected, negative in IP animals (MTH-IP-12; Figure 8 B, responses seen in 41 of 41 trials) and positive in Cg2 animals (MTH-Cg2-19; Figure 8 E, responses seen in 54 of 54 trials). The signals consisted of a sustained component and a periodic component, which was phase-locked to the animal's respiration ( Figure 8 B and 8 E). During some of the late sustained component, the well-known negative intrinsic response ( Spors and Grinvald 2002 ) is likely to be superimposed. Signals were significant even for low odor concentrations (e.g., 0.1% 2-Hexanone; Figure 8 C), and they increased with concentration. The largest fluorescence changes seen were −8% for IP and +3% for Cg2. The time course of the signals was consistent with the [Ca 2+ ] dynamics in sensory afferents ( Wachowiak and Cohen 2001 ). Maps of odor-evoked fluorescence changes were constructed during early response times, thereby minimizing the contribution of the slow intrinsic signal. Odor-evoked spatial patterns of Ca 2+ signals were widespread in MTH-IP-12 ( Figure 8 C) and more localized in MTH-Cg2-19 ( Figure 8 F) but in both cases were more diffuse than maps of afferent glomerular activity measured with a synthetic indicator in sensory axon terminals ( Wachowiak and Cohen 2001 ). This is presumably because FCIP is also expressed in granule cells. These receive input from secondary dendrites of mitral cells, which project for several hundreds of micrometers around each glomerulus. Nevertheless, each odor evoked a unique activity map, and odors known to evoke similar maps of glomerular afferent activity (e.g., methyl benzoate and benzaldehyde) evoked similar activity patterns in transgenic animals ( Figure 8 C and 8 F). The more localized signals seen with Cg2 may be due to its lower affinity for Ca 2+ , reporting only high [Ca 2+ ] in the vicinity of activated glomeruli. Alternatively, Cg2 might be expressed more strongly in olfactory receptor axons. Discussion We have demonstrated that genetically encoded FCIPs can be stably expressed in mice and are functional. Transgenically expressed FCIPs showed changes of up to 100% in response to synaptic stimulation (see Figure 6 D– 6 F). These changes are smaller than those seen in cell culture or protein extracts ( Griesbeck et al. 2001 ; Nagai et al. 2001 ), suggesting that a fraction of protein, possibly immobile and sequestered, is nonresponsive. The size of the immobile fraction seen in bleach-recovery experiments (see Figure 4 E) fluctuates strongly around a mean value of roughly 50% of the total FCIP fluorescence. Fluorescence changes evoked by electrical stimulation in the neuropil show that FCIPs respond quickly to Ca 2+ influx. The relative fluorescence changes recorded in slices in response to synaptic stimulation are large when measured by two-photon microscopy (see Figure 6 D– 6 F) but are substantially smaller with wide-field microscopy (see Figure 6 C), presumably because signals from activated and nonactivated cells inevitably mix because of a lack of optical sectioning in the wide-field case. In whole-cell tight-seal recordings somatic signals may fade additionally due to washout of responsive protein. Since camgaroos and pericams are intrinsically pH-sensitive ( Baird et al. 1999 ; Griesbeck et al. 2001 ; Nagai et al. 2001 ) it is possible that the fluorescence changes contain a component due to changes in [H + ] (pH) rather than [Ca 2+ ]. It is, however, unlikely that the changes we saw are dominated by pH effects for the following reasons. In the case of Cg2, stimulation-induced pH changes ( Yu et al. 2003 ) should lead to a decrease in fluorescence while we see an increase (see Figure 6 D– 6 F). For IP the change due to pH would be in the same direction, but, in particular, the size of the changes seen during two-photon measurements (see Figure 6 F; ∼30%) are almost an order of magnitude larger than what one might expect from pH changes that occur with high [K + ] stimulation ( Yu et al. 2003 ), but see Wang et al. (1994) , who found much larger pH changes, albeit with massive glutamate application. Furthermore, changes of pH are much slower than those seen in our synaptic stimulation experiments. The robust and fast FCIP signals detected in response to sensory stimulation in vivo confirm that FCIPs are suitable for their main intended use, the imaging of activity from populations of neurons in living animals . Crucial for addressing functional questions in neuronal networks will also be the cell-type specificity of expression, which we have demonstrated here for the population of αCamKII-positive neurons. Our success rate in generating functional transgenic mouse lines for Cg2 and IP was moderate (five of 36 animals that were transgenic, according to DNA typing; see Figure 2 B). YC3.12 was a disappointment and did not yield functional lines. This is consistent with a previous attempt to generate transgenic mice expressing YC3.0 under the control of the β-actin promoter ( Tsai et al. 2003 ). There, animals were produced that also showed mosaic expression patterns and had only very small functional signals (ΔR/R ∼1%–2%) when tested by wide-field imaging of cerebellar slices undergoing synaptic stimulation (A. Miyawaki, V. Lev-Ram, and R.Y. Tsien, unpublished data). An early suggestion that indicator proteins become nonfunctional after expression for an extended period of time (O. Griesbeck, personal communication) certainly does not apply to IP and Cg2 since we found large responses even in mice aged 8–12 wks, which had been expressing indicators since the onset of αCamKII expression before birth. However, even in strongly expressing smooth-fluorescence lines, a substantial fraction of the indicator protein was found to be immobile and potentially nonfunctional at various ages. It is surprising that punctate fluorescence occurs predominantly in weak lines since precipitation typically occurs at high concentrations. It could, however, be that a limited number of binding and sequestration sites for FCIPs exist in the cell and that only after these sites are saturated does the accumulation of mobile (see Figure 3 J), cytosolic, and responsive FCIP begin. The accumulation of any functional indicator protein might, therefore, require expression levels above a (rather high) threshold. Such a threshold might explain why in the majority of lines we find weak, nonresponsive, and punctate fluorescence even when there is significant luciferase activity (open circles in Figure 2 B). Binding and/or sequestration of FCIP does, of course, raise the specter of interference of the indicator with biochemical processes inside labeled cells. While subtle effects cannot be ruled out at this point, we did not see any obvious abnormalities either at the whole-animal level or at the level of cellular morphology. The labeled neurons, furthermore, are connected synaptically (see Figures 6 C, 7 , and 8 ). It will, however, be important to understand and if possible remedy the reasons for the formation of precipitates. Perhaps the use of Ca 2+ -sensing motifs ( Heim and Griesbeck 2004 ) that lack affinity for proteins normally expressed in neurons will eliminate puncta and the immobile fraction. It is unclear why high expression levels appear to be achievable in brain using the TET promoter system but not other, cellular promoters. One explanation for weakness or outright lack of expression might be that sequences within the Ca 2+ -indicator genes silence cellular promoters ( Robertson et al. 1995 ; Clark et al. 1997 ) but not the tTA-responsive promoters (including Ptetbi). The reason for this, in turn, might be that cellular promoters (unlike the TET promoter) contain a substantial number of transcriptional control elements (upstream activator sequences), any of which might be sensitive to chromatin-induced silencing by the FCIP gene ( Lemon and Tjian 2000 ). This is supported by the observation that attempts to create transgenic mice expressing IP under the control of a 3.5-kb gonadotrophic-releasing-hormone promoter fragment resulted in many lines that had the gene inserted but showed at best weak and punctate expression (D.J. Spergel and P.H. Seeburg, personal communuication), similar to our low-expression lines, while the same promoter fragment drove hGFP2 (an EGFP variant) to high levels ( Spergel et al. 1999 ). Similarly, only weak expression was observed when YC3.1 was placed under control of the neuron-specific enolase promoter ( Futatsugi et al. 2002 ; A. Miyawaki, personal communication). One of the remaining problems hampering imaging of population activity with FCIPs is that expression, while cell-type specific, is not complete, i.e., not in every cell of one type, even in lines that express FCIPs at high levels. One possible explanation is position effect variegation (PEV), which occurs when a transgene integrates adjacent to a heterochromatin domain in the genome. In such a situation, expression of the locus variegates, being active in some cells and silent in others ( Saveliev et al. 2003 ; Schotta et al. 2003 ). If this is the case, it might be possible to avoid mosaicism by generating lines free from PEV by cloning FCIP genes into a bacterial artificial chromosome ( Shizuya et al. 1992 ) derived from a TET responder line that is not prone to PEV ( Hasan et al. 2001 ). In any case, it appears that the use of the TET system allows the expression of genetically encoded Ca 2+ indicators in mice, albeit for an unexpected reason. Unlike other promoter systems, such as the CMV promoter, which also appears to resist gene silencing, the TET system allows cell specificity via the expression of the transactivator, which appears to be controllable by cellular promoters without gene silencing. The creation of transgenic mouse lines expressing functional Ca 2+ indicators opens the way for the measurement of neural activity patterns in mammals in vivo and in vitro. While the sensitivity of genetic indicators does not (yet) quite reach those of synthetic compounds, it is sufficient for single-trial measurements at least in some applications (see Figure 8 ). Perhaps the greatest advantage is that the genetically encoded indicators alleviate the labeling problem in general and allow the observation of activity in targeted cell populations without the need to load cell or tissue preparations with synthetic indicators using potentially harmful procedures. In vivo imaging of FCIPs will permit analysis of population activity using minimally invasive procedures such as imaging through the intact skull. Applications for FCIPs are similar to those for intrinsic signal imaging, but FCIPs provide substantially higher spatiotemporal resolution and signal-to-noise ratio. Compared to voltage-sensitive dyes, transgenic Ca 2+ indicators yield substantially larger signals and obviate surgical procedures for dye loading. Another important advantage of transgenic Ca 2+ indicators is that the optical signal can be interpreted more specifically because of its defined cellular origin. In combination with two-photon microscopy, neuronal activity can be mapped at high resolution, down to the level of individual dendritic branchlets and maybe spines, possibly even in awake, behaving animals ( Helmchen et al. 2001 ). In addition, FCIP lines may be crossed with mouse lines in which the expression of genes of interest has been manipulated. For example, the combination of FCIP mice with lines harboring modifications of plasticity-driven genes ( Nakazawa et al. 2002 ) or genes that cause neurodegenerative diseases ( Wong et al. 2002 ) might help us to understand how specific genes are involved in the construction and experience-dependent modification of brain circuitry. Materials and Methods Screening of indicators and generation of transgenic mice Genes encoding six different Ca 2+ indicators (flash pericam, IP, CaMP, Cg2, and cameleons YC2.12 and YC3.12) and FL were cloned into a Ptetbi vector (Clontech, Palo Alto, California, United States). The resulting plasmids (Ptetbi-FL/FP, Ptetbi-FL/IP, Ptetbi-FL/CaMP, Pbi-FL/Cg2, Ptetbi-FL/YC2.12, and Ptetbi-FL/YC3.12) were sequenced and transfected into HeLa cells that stably express tTA ( Gossen and Bujard 1992 ). Cells were then tested for [Ca 2+ ]-dependent fluorescence changes to establish functioning of the indicators (see Figure 1 and Figure 2 A). The transgene insert, devoid of vector sequences, was purified by sucrose gradient ( Mann and McMahon 1993 ) and used for the generation of transgenic animals, using the DNA-microinjection method ( Gordon and Ruddle 1982 ) in the facility of the Zentrum für Molekulare Biologie at the University of Heidelberg. All procedures were performed in accordance with German federal guidelines for animal experiments. Screening of founders using cultured ear fibroblasts Ear fibroblast cultures were prepared for every DNA-positive founder animals using the procedure described by Schoenig and Bujard (2003) . Cells were trypsinized after reaching confluency and plated into 6-well plates divided into sets with and without Dox (4-[Dimethylamino]-1,4,4a,5,5a,6,11,12a-octahydro-3,5,10,12,12a-pentahydroxy-6-methyl-1,11-dioxo-2-naphthacenecarboxamide; Sigma-Aldrich, St. Louis, Missouri, United States). When cells reached 50% confluency, both the Dox-plus and the Dox-minus cultures were transfected with 0.5 μg of synthetic reverse tTA (rtTA-M2s) ( Gossen et al. 1995 ; Urlinger et al. 2000 ) and 0.5 μg of RL plasmids (Promega, Mannheim, Germany) using lipofectamine-2000 DNA-transfection reagents, as recommended by the vendor (Invitrogen Life Technologies, Carlsbad, California, United States). After 48 h, cells were washed once with PBS and incubated in 0.5 ml of lysis buffer on ice (Promega). 50-μl aliquots from each lysate were tested for FL activity and RL activity (Lumat LB9501; Berthold Technologies, Wildbad, Germany). The ratio of FL to RL activity was used to correct for DNA transfection efficiency. Individual transfections and measurements were done in duplicate, usually resulting in normalized activity values that agreed within 5%. Visualizing GFP in fixed brain slices Brains from double-positive animals (identified by PCR of tail DNA) were fixed in 4% paraformaldehyde in PBS for 4 h and washed twice with PBS. Brain slices were cut to a thickness of approximately 70 μm using a vibratome (VT 1000S; Leica Instruments, Wetzlar, Germany). Distribution of Ca 2+ indicator protein was determined by staining with GFP-specific polyclonal rabbit antibodies (Clontech) ( Krestel et al. 2001 ) and the DAB peroxidase system (Vectastain ABC Kit; Vector Laboratories, Burlingame, California, United States) or by direct observation of fluorescence with an upright microscope (Zeiss, Oberkochen, Germany) equipped with GFP filters. Two-photon imaging All two-photon measurements described in the following sections were done using custom-built two-photon microscopes. Fluorescence was two-photon-excited by a mode-locked Ti-sapphire laser (Coherent Mira F900, 930 nm, 100 fs, 78 MHz) coupled into a custom-built imaging system. The objective used was a 40X/0.8 NA water immersion lens (Nikon, Tokyo, Japan). A photomultiplier-based whole-field detector ( Denk et al. 1995 ) detected emitted light in the range around 535 nm, optimized for yellow fluorescent protein signals. Scanning and image acquisition were controlled using custom software (developed by R. Stepnoski, Lucent Technologies, Murray Hill, New Jersey, United States, and M. Muller, Max-Planck Institute for Medical Research, Heidelberg, Germany). Data analysis was performed with ImageJ ( http://rsb.info.nih.gov/ij/ ) and IgorPro (Wavemetrics, Lake Oswego, Oregon, United States). Fluorescence recovery after photobleaching and spectral analyses Regions rich in neurites were repeatedly scanned in the two-photon microscope. After a first bleaching run (see Figure 4 E, red trace, first and last of ten images shown), scanning was interrupted for 15 s. A second bleach run was performed, then scanning was interrupted by 93 s before the final bleach run (see Figure 4 E, blue and green traces, pictures 3/4 and 5/6, respectively). Spectral recordings were performed with a confocal microscope (TCS SP2 AOBS; Leica) using an excitation wavelength of 488 nm. Fluorescence emission was measured by recording image sequences with overlapping shifted spectral windows (10 or 20 nm wide) covering the range of 500–650 nm. Spectra were then calculated for different regions of interest and analyzed using the Leica LCS software, Microsoft Excel, and ImageJ. Preparation of living brain slices Parasagittal and transverse brain slices (300 μm in thickness) from mice (between postnatal days 18 and 60) were prepared according to published procedures for hippocampal experiments ( Hoffman et al. 2002 ) and for cortical experiments ( Waters et al. 2003 ) using a custom-built vibratome (Max Planck Institute, Heidelberg). Mice were deeply anesthetized with halothane. After decapitation the brain was quickly removed and placed into ice-cold, oxygenated artifical cerebrospinal fluid (ACSF; Biometra Biomedizinische Analytik, Gottingen, Germany) containing 125 mM NaCl, 25 mM NaHCO3, 2.5 mM KCl, 1.25 mM NaH2PO4, 1 mM MgCl2, 25 mM glucose, and 2 mM CaCl2 (pH 7.4). For the two-photon imaging experiments, hippocampus slices were incubated at 37 °C for 30 min and then allowed to reach room temperature gradually before being used for experiments over a period of several hours. For the cortical imaging experiments, slices were kept at 34 °C for the duration of the experiment. In all cases the slice chamber was continuously perfused with ACSF. Whole-cell recording and synaptic stimulation in brain slices Acute brain slices (see above) were maintained in ACSF. Whole-cell tight-seal recordings were made using pipettes made from borosilicate glass (5–10 MΩ) containing 135 mM K gluconate, 4 mM KCl, 10 mM HEPES, 10 mM Na2-phosphocreatine, 4 mM Mg-ATP, and 0.3 mM Na-GTP. Recording pipettes also contained 0.2% biocytin and 1–5 μM Alexa 568, which is spectrally distinct from the FCIPs and can be used to unambiguously identify the recorded cell. For synaptic stimulation we used saline-filled glass electrodes or tungsten microelectrodes (1 MΩ) (World Precision Instruments, Berlin, Germany). Synaptic transmission was blocked in some experiments by the addition of 40 μM 6-cyano-7-nitroquinoxaline-2,3-dione and 100 μM 2-amino-5-phosphovaleric acid. Imaging and analysis in cortical and hippocampal slices Wide-field images were taken with a MicroMax, 512 × 512 back-illuminated CCD camera (Roper Scientific, Tucson, Arizona, United States) binned 5 by 5. For experiments involving extracellular stimuli we calculated the change in fluorescence relative to the prestimulus period (ΔF/F) for each binned pixel frame by frame. The prestimulus period of regions of interest was fitted with an exponential curve, which was then subtracted from the entire fluorescence time course to correct for bleaching. No corrections were made for autofluorescence, so that the relative FCIP fluorescence changes are likely to be larger. When analyzing action-potential-induced signals from individual neurons, we calculated the fluorescence changes by subtracting the average of two nearby areas from the total fluorescence to account for background fluorescence (the intracellular FCIP concentration may, however, have been reduced by dialysis into the recording pipette—see Discussion ). Two-photon image sequences were collected approximately 50 μm away from the stimulating electrode. For synaptic stimulation five prestimulus frames were collected (64 mm × 64 mm, 128 ms/frame) to record baseline fluorescence (100 pulses at 100 Hz or 20 pulses at 100 Hz). Image sequences were 6 or 12 s long. The background level (average intensity with the laser off) was subtracted from every frame in the image sequence. The average of the five prestimulus (rest) images was subtracted from the average of five “response” images (response minus rest). The built-in smoothing function of ImageJ was used to reduce the noise further. The brightness of the difference image was enhanced for better display. Localized small fluorescent structures or hot spots in neurites were identified visually. For IP, traces with high time resolution were acquired using 64-pixel line scans at 500 Hz. Fluorescence was averaged over the width of the soma. The fluorescence from a neighboring region was subtracted to account for background. Fluorescence changes (percent ΔF/F) were calculated relative to the resting fluorescence. Tissue preparation for light-evoked responses in an intact retinal whole mount Mice were dark-adapted for several hours before the experiments and all subsequent procedures were carried out under dim red illumination to minimize photobleaching. Animals were anesthetized with halothane and subsequently killed by cerebral dislocation or by intraperitoneal injection of pentobarbital. Immediately afterward, both eyes were removed and dissected free in Ames medium (Sigma-Aldrich). A piece was cut from a retina and placed photoreceptor side down into the recording chamber, and maintained at 35 °C in Ames medium continuously perfused with oxygen. The remaining retina was kept for further use. Ca 2+ imaging and visual stimulation The stimulation and imaging procedures were as described elsewhere ( Euler et al. 2002 ). In brief, simple light stimuli (bright spots, 300 μm in diameter, on dark background) were projected repetitively onto the receptive field of a labeled retinal ganglion cell (light spot centered on the cell body) while monitoring Ca 2+ -mediated fluorescence (emission 520 BP 30 nm) changes in retinal ganglion cells using a custom-built two-photon microscope. The laser (Coherent Mira 900F) was tuned to 925 nm (for YC3.12, the laser was tuned to 870 nm, see Results ) to keep direct photoreceptor excitation at a minimum and prevent bleaching. To visualize the retinal tissue, Sulforhodamine 101 (2 mg/l; Sigma-Aldrich) was added to the extracellular medium. In vivo imaging in the olfactory bulb Mice were anesthetized and dissected as previously described ( Wachowiak and Cohen 2001 ). Odors were delivered through a custom-built flow dilution olfactometer. Dilutions are given relative to the stable vapor in the olfactometer's reservoir. Series of images were collected at rates of 5–15 Hz with a cooled CCD camera (CoolSnapHQ; Photometrics, Huntington Beach, California, United States) mounted on a custom-built upright fluorescence microscope that was equipped with a 20×, 0.95 NA water immersion objective (Olympus, Tokyo, Japan) and the following filter sets (Chroma Technology, Rockingham, Vermont, United States): HQ495/30, Q520LP, and HQ545/50 for IP and Cg2, and D436/20, 455DCLP, and D535/30 for YC3.12. For each pixel and frame, the change in fluorescence relative to the pre-odor period (ΔF/F) was calculated. Trials without odor stimulation were subtracted to correct for bleaching. For the display of activity maps, ΔF/F images taken during the first second of odor stimulation were averaged and low-pass spatially filtered. Later times were not included in activity maps to avoid the contribution of intrinsic signals ( Spors and Grinvald 2002 ). Respiratory activity was measured with a piezoelectric strap wrapped around the animal's thorax. In vivo two-photon imaging Mice were anesthesized with urethane (1.5 mg/g) and body temperature was maintained at 37 °C. For two-photon imaging, a custom-built headplate with an imaging window (4 mm × 3 mm) was glued to the top of the skull using cyano-acrylate (UHU, Buhl-Baden, Germany) and attached to a fixed metal bar before thinning of the skull. The combination of rigid headplate and thinned-skull reduces respiration and cardiac-pulsation-induced brain motion. The microscope objective was positioned at an angle so that the optical axis was perpendicular to the surface of the cortex. Table 1 Summary of Functionality Tests Recorded by Either Wide-Field or Two-Photon Imaging WF, wide-field; 2P, two-photon; −, decrease; IR, inconclusive results; NT, not tested	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC423138.xml
556014	Phosphorylation of HIV Tat by PKR increases interaction with TAR RNA and enhances transcription	Background The interferon (IFN)-induced, dsRNA-dependent serine/threonine protein kinase, PKR, plays a key regulatory role in the IFN-mediated anti-viral response by blocking translation in the infected cell by phosphorylating the alpha subunit of elongation factor 2 (eIF2). The human immunodeficiency virus type 1 (HIV-1) evades the anti-viral IFN response through the binding of one of its major transcriptional regulatory proteins, Tat, to PKR. HIV-1 Tat acts as a substrate homologue for the enzyme, competing with eIF2α, and inhibiting the translational block. It has been shown that during the interaction with PKR, Tat becomes phosphorylated at three residues: serine 62, threonine 64 and serine 68. We have investigated the effect of this phosphorylation on the function of Tat in viral transcription. HIV-1 Tat activates transcription elongation by first binding to TAR RNA, a stem-loop structure found at the 5' end of all viral transcripts. Our results showed faster, greater and stronger binding of Tat to TAR RNA after phosphorylation by PKR. Results We have investigated the effect of phosphorylation on Tat-mediated transactivation. Our results showed faster, greater and stronger binding of Tat to TAR RNA after phosphorylation by PKR. In vitro phosphorylation experiments with a series of bacterial expression constructs carrying the wild-type tat gene or mutants of the gene with alanine substitutions at one, two, or all three of the serine/threonine PKR phosphorylation sites, showed that these were subject to different levels of phosphorylation by PKR and displayed distinct kinetic behaviour. These results also suggested a cooperative role for the phosphorylation of S68 in conjunction with S62 and T64. We examined the effect of phosphorylation on Tat-mediated transactivation of the HIV-1 LTR in vivo with a series of analogous mammalian expression constructs. Co-transfection experiments showed a gradual reduction in transactivation as the number of mutated phosphorylation sites increased, and a 4-fold decrease in LTR transactivation with the Tat triple mutant that could not be phosphorylated by PKR. Furthermore, the transfection data also suggested that the presence of S68 is necessary for optimal Tat-mediated transactivation. Conclusion These results support the hypothesis that phosphorylation of Tat may be important for its function in HIV-1 LTR transactivation.	Background Since its isolation in 1983 [ 1 , 2 ], human immunodeficiency virus type 1 (HIV-1) continues to cause 5 million new infections each year, and since the beginning of the epidemic, 31 million people have died as a result of HIV/AIDS [ 3 ]. One of the major mechanisms employed by the immune system to counteract the effects of viral infections is through an antiviral cytokine – type 1 interferon (IFN). However, while IFN is able to inhibit HIV-1 infection in vitro [ 4 ], it has not been effective in the treatment of HIV-1 infections in vivo . Furthermore, the presence of increasing levels of IFN in the serum of AIDS patients while viral replication continues and the disease progresses [ 5 - 7 ] indicates that HIV-1 must employ a mechanism to evade the antiviral effects of IFN. In response to viral infection, IFN induces a number of genes including the dsRNA-dependent protein kinase R (PKR). PKR exerts its anti-viral activity by phosphorylating the alpha subunit of translation initiation factor 2 (eIF2α), which results in the shut-down of protein synthesis in the cell [ 8 ]. The importance of PKR in the host antiviral response is suggested by the fact that most viruses including vaccinia [ 9 ], adenovirus [ 10 ], reovirus [ 11 ], Epstein-Barr virus [ 12 ], poliovirus [ 13 ], influenza [ 14 ], hepatitis C [ 15 , 16 ], human herpes virus [ 17 - 19 ], and SV40 [ 20 ], employ various mechanisms to inhibit its activity. HIV-1 is no exception and we and others have shown that PKR activity is inhibited by HIV via the major regulatory protein, Tat [ 21 - 23 ]. Productive infection by HIV-1 results in a significant decrease in the amounts of PKR [ 23 ] and HIV-1 Tat protein has been shown to act as a substrate homologue of eIF2α, preventing the phosphorylation of this factor and allowing protein synthesis and viral replication to proceed in the cell [ 21 , 22 ]. During the interaction between Tat and PKR the activity of the enzyme is blocked by Tat and Tat itself is phosphorylated by PKR [ 21 ] at serine 62, threonine 64 and serine 68 [ 22 ]. HIV-1 Tat is a 14 kDa viral protein involved in the regulation of HIV-1 transcriptional elongation [ 24 - 26 ] and in its presence, viral replication increases by greater than 100-fold [ 27 , 28 ]. It functions to trigger efficient RNA chain elongation by binding to TAR RNA, which forms the initial portion of the HIV-1 transcript [ 29 ]. The interaction between Tat and TAR is critical for virus replication and mutations in Tat that alter the RNA-binding site result in defective viruses. Furthermore, virus replication can be strongly inhibited by the overexpression of TAR RNA sequences that act as competitive inhibitors of regulatory protein binding [ 30 ]. While a number of reports have shown that PKR and Tat protein interact, and furthermore, that Tat is phosphorylated by PKR, none have yet addressed the issue of the functional consequences for the phosphorylation of the Tat protein. Here we examine the phosphorylation of Tat by PKR and its effect on TAR RNA binding and HIV-1 transcription, and show that the phosphorylation of Tat results in Tat protein binding more strongly to TAR RNA. Removal of the residues reported to be phosphorylated by PKR resulted in decreased Tat phosphorylation and a significant loss of Tat-mediated transcriptional activity. Results The phosphorylation of HIV-1 Tat by PKR increases its interaction with TAR RNA We first confirmed the capability of our PKR preparation immunoprecipitated from HeLa cells to phosphorylate synthetic Tat protein (aa 1–86) (Figure 1a ), and we determined the optimal phosphorylation time of Tat by PKR as 60 minutes (Figure 1b ). We also confirmed that Tat was not phosphorylated by PKR in the absence of ATP, or by ATP alone (data not shown). Figure 1 Phosphorylation of HIV-1 Tat86 by PKR. (a) PKR was immunoprecipitated from HeLa cell extracts and activated with synthetic dsRNA in the presence of γ- 32 P-ATP. This activated 32 P-PKR was used to phosphorylate 0.5, 1 and 5 μg of synthetic Tat86 in the presence of γ- 32 P-ATP, at 30°C for 15 minutes. Proteins were separated by 15% SDS-PAGE. (b) PKR immunoprecipitated from HeLa cell extracts, and activated with dsRNA and ATP, was used to phosphorylate 2 μg of synthetic Tat86 at 30°C for the times indicated. To address the issue of the consequences of PKR phosphorylation on Tat function we investigated the ability of phosphorylated Tat (herein called Tat-P) and normal Tat (Tat-N) to bind to HIV-1 TAR RNA. Synthetic Tat protein (aa 1–86) was phosphorylated in vitro using PKR previously immunoprecipitated from HeLa cells. An electrophoretic mobility shift assay (EMSA) was performed to observe any difference in the binding of Tat-N and Tat-P to TAR RNA (Figure 2a ). It can be seen that Tat-N was able to form a specific Tat-TAR complex that could be effectively competed off using a 7.5-fold excess of cold TAR RNA. Tat-P was also able to form a specific Tat-TAR complex that clearly contained more TAR RNA than non-phosphorylated Tat. This complex could also be competed off using cold TAR but some residual complex was left suggesting that the Tat-P-TAR complex was more resistant to competition with cold TAR than the Tat-N-TAR complex. Figure 2 EMSA of Tat-N, Tat-P and TAR RNA showing dissociation of the Tat-TAR complex with increasing salt concentration. (a) PKR immunoprecipitated from HeLa cell extracts, and activated with dsRNA and ATP, was used to phosphorylate 2 μg of synthetic Tat86 at 30°C for 1 h, in the presence (Tat-P) or absence (Tat-N) of γ- 32 P-ATP. TAR RNA was synthesized in vitro from pTZ18TAR80 using a commercial kit, and either γ- 32 P-dCTP or unlabelled dCTP. The Tat-TAR RNA binding reaction was allowed to proceed in binding buffer at 30°C for 10 minutes. Each reaction contained 200 ng of either Tat-N or Tat-P, and approximately 70 000 cpm of 32 P-TAR RNA (lanes 1 and 2), or approximately 70 000 cpm of 32 P-TAR RNA and 7.5 × the volume of unlabelled TAR RNA (lanes 3 and 4). The Tat-TAR complexes formed were resolved on a 5% acrylamide/0.25X TBE gel. (b) The Tat-TAR binding reactions were performed at 30°C for 10 minutes in binding buffer containing various concentrations of NaCl: 25 mM (lanes 2 and 8), 50 mM (lanes 3 and 9), 100 mM (lanes 4 and 10), 200 mM (lanes 5 and 11), 500 mM (lanes 6 and 12), and 1000 mM (lanes 7 and 13). Lanes 2–7 show the dissociation of the Tat-N-TAR complex, and lanes 8–13 show the dissociation of the Tat-P-TAR complex. Lane 1 is TAR RNA only. As Tat-P appeared to bind more readily to TAR, we next investigated the differences in the binding efficiency of Tat-N and Tat-P with TAR RNA. EMSA were performed in the presence of increasing concentrations of NaCl (from 25–1000 mM). The progressive dissociation of the Tat-N-TAR RNA complex with increasing concentrations of salt in the buffer was observed (Figure 2b , lanes 2–7) while Tat-P-TAR complexes under the same conditions were clearly more stable (lanes 8–13). For example, at 500 mM NaCl the Tat-N-TAR complex was almost completely dissociated (lane 6) while the Tat-P-TAR complex was still clearly observed (lane 12). Even at the maximum salt concentration (1000 mM), the Tat-P-TAR complex can still be seen (lane 13), while the Tat-N-TAR complex was completely dissociated. These results suggest that Tat86 phosphorylated by PKR binds TAR RNA more efficiently and more strongly than normal Tat. Efficient phosphorylation of Tat requires particular residues Brand et al . [ 22 ] reported that PKR was able to phosphorylate Tat at amino acids serine-62, threonine-64 and serine-68. We therefore wished to know if any of these residues were critically important in the ability of Tat to bind TAR RNA. To this end, we created a series of Tat proteins containing mutations of all possible combinations of S62, T64 and T68 and investigated the phosphorylation of the resulting mutant Tat protein. A series of seven Tat mutants were made using alanine scanning (Figure 3a ) and cloned into the bacterial expression vector pET-DEST42, which contains a C-terminal 6 × His tag to allow purification using metal affinity chromatography. The resulting constructs were validated by sequencing before the mutant Tat proteins were expressed and purified (Figure 3b ). Protein yields varied between 40–170 g/mL and all mutants were full length, as confirmed by western blotting using an anti-His antibody (data not shown). Figure 3 Construction of HIV-1 Tat phosphorylation mutants. (a) Amino acid sequence of HIV-1 Tat wild-type and mutants. Changes to alanine at serine 62, threonine 64 and serine 68 are indicated for each mutant, and compared to the wild-type protein. Mutations were introduced by site-directed mutagenesis into pET-DEST42-HIS-Tat86. (b) Competent BL21(DE3)pLysS cells, transformed with pET-DEST42-HIS-Tat86 wild-type or mutants, were grown and lysed with 6 M guanidine-HCl, pH 8.0. The suspension was cleaned of cell debris and loaded onto a packed metal affinity resin. The resin was washed and the HIS-tagged Tat proteins were eluted with 6 M guanidine-HCl, pH 4.0. The fractions collected were dialysed in 0.1 mM DTT and then analysed by 15% SDS-PAGE and stained with Coomassie blue. Tat lanes show fractions containing HIS-tagged Tat proteins; M lanes, 14 kDa marker; C lanes, BL21(DE3)pLysS cell extract. Activated PKR was used to phosphorylate each of the Tat mutants as above and the reaction was allowed to proceed for 2, 5, 10, 15, 30, 45 and 60 minutes. The phosphorylated proteins were analyzed by SDS-PAGE and visualized by autoradiography (Figure 4 ). As can be seen from the figure, the phosphorylation of each protein by PKR varied and was the most efficient for wild-type Tat and the least efficient for the triple mutant, Tat S62A.T64A.T68A, where no sites for PKR phosphorylation were available. Scanning densitometry and non-linear regression analysis was performed and the extent of phosphorylation after 15 minutes was measured for each protein and expressed as a percentage of the wild-type protein (which is set to 100%) (Figure 5a ). This time was chosen from non-linear regression analysis of the wild-type protein that indicated enzymatic phosphorylation of the wild-type protein was active at this time point. Non-linear regression analysis was performed to calculate the maximal phosphorylation for each protein (P max ), and the time required to reach half-maximal phosphorylation (K 0.5 ) (Figure 5b ). Figure 4 PKR phosphorylation of HIV-1 Tat wild-type and mutants. HIV-1 Tat wild-type and mutant proteins were expressed in BL21(DE3)pLysS cells from pET-DEST-42 expression clones, and purified by passage through a TALON™ cobalt affinity resin. PKR was immunoprecipitated from HeLa cell extracts, and activated with dsRNA in the presence of ATP. The phosphorylation reactions contained 2 μg of Tat protein, 6 μL of activated PKR suspension, and DBGA to a final volume of 12 μL. Phosphorylation was preformed at 30°C for the times indicated, in the presence of 2 μCi of γ- 32 P-ATP. Protein samples were analyzed by 15% SDS-PAGE. This figure only shows one representative gel out of three separate phosphorylation experiments performed for each protein. Figure 5 PKR phosphorylation of HIV-1 Tat wild-type and mutants after 15 minutes and phosphorylation kinetics. (a) Proteins were phosphorylated by activated PKR at 30°C for 15 minutes in the presence of γ- 32 P-ATP. The reaction was stopped by the addition of protein loading buffer and incubation at 4°C. Samples were analyzed by 15% SDS-PAGE. Graph shows the results for three separate experiments. (b) Non-linear regression analysis of PKR phosphorylation curves of wild-type and mutant proteins was performed using a one-site binding hyperbola, which describes the binding of a ligand to a receptor and follows the law of mass action. K 0.5 is the time required to reach half-maximal phosphorylation. Phosphorylation of the single mutants was rapid and specific with maximal phosphorylation values (P max ) for S62, T64 and T68 of 98.6%, 87.5% and 81.6% respectively compared to the wild type (P max = 82.8%) and K 0.5 values of 10.9 min, 5.2 min and 0.8 min (wild-type = 5.5 min). This observation was also applicable to the Tat S62A.T64A mutant, which exhibited 87% phosphorylation (Figure 5a ) (P max = 82.1%, K 0.5 = 5.5 min). However, the percentage of phosphorylation at 15 minutes for the other double mutants and for the triple mutant decreased to 68% for Tat T64A.S68A, 48% for Tat S62A.S68A, and 56% for Tat S62A.T64A.S68A. These values also correlated well with the higher P max values (172.8%, 256.8% and 189.7% respectively) and K 0.5 values (54.9 min, 109.7 min and 62.2 min respectively) for each mutant, indicating slower, less efficient and non-specific phosphorylation. The phosphorylation of HIV-1 Tat by PKR enhances viral transcription To examine the effect of Tat phosphorylation on its transactivation ability mammalian expression constructs containing the Tat mutants were prepared and transfected into HeLa cells. To measure Tat-specific transcription, we co-transfected with pHIV-LTR-CAT as well as with β-actin-luciferase to normalize for transfection efficiency. The transfection reaction was optimized for DNA concentration, transfection reagent concentration, and time. The results for three separate transfections are shown in Figure 6 and expressed as percentage of wild-type Tat. As expected, no transactivation of the HIV-1 LTR was observed in the untransfected control or in the absence of pHIV-LTR-CAT, and basal transcription was present at low levels (0.08-fold) in the absence of Tat. We observed significant decreases in transactivation with mutant Tat, even when a single phosphorylation site was mutated. There was a general trend to low activity as more mutations were introduced. Thus, the average transactivation by the single mutants, Tat 62A, T64A and S68A, was 58%, transactivation by the double mutants, Tat S62A.T64A, T64A.S68A and S62A.S68A, was 41%, while the triple mutant, Tat S62A.T64A.S68A, exhibited only 24% transactivation. Figure 6 Transactivation of the HIV-1 LTR by HIV-1 Tat wild-type and mutants. Duplicate wells of confluent HeLa cells were transfected for 6 h with pcDNA3.2-DEST-Tat, pHIV-LTR-CAT and β-actin luciferase. Cells were harvested 24 h post transfection and assayed for CAT activity, luciferase activity and protein concentration. The graph shows the results of three separate experiments. The differences in LTR activation observed for the individual single mutants were not large, indicating that the absence of any one of these phosphorylation residues reduced the ability of Tat to activate the HIV-1 LTR but that no single residue was more important than the other. As in the phosphorylation data, Tat S62A.T64A behaved similarly to the single mutants. The mutations that had the greatest effect were the T64A.S68A, S62A.S68A, and the triple mutant. Of the three residue combinations, the absence of T64 and S68 together had the greatest negative effect on transactivation, inducing a 3-fold decrease, which was comparable to that observed for the triple mutant (4-fold). The absence of S62 in combination with S68 also had a marked effect on transactivation, reducing it 2.5-fold. On the other hand, the absence of S62 in combination with T64 reduced transactivation 1.8-fold. This suggests that the absence of S62 and T64 either singly or in combination is not as important for Tat-mediated transactivation as when these residues are absent in combination with S68, and may indicate a more important role for S68 in Tat transactivation. These data correlate with observations previously obtained in PKR phosphorylation experiments with these Tat mutants. Discussion HIV-1 inhibits the antiviral effects of IFN by the direct binding of its Tat protein to PKR [ 21 ]. In the infected cell, Tat blocks the inhibition of protein synthesis by PKR, thus allowing viral replication to proceed. As a consequence of this interaction, Tat becomes phosphorylated at S62, T64 and S68 [ 22 ]. Here we have examined the consequences of this phosphorylation on Tat function and have shown that it results in increased and stronger binding of Tat to TAR RNA. Tat protein is an essential regulatory protein during viral transcription and binds to the positive elongation factor B (P-TEFb), through its cyclin T1 subunit, and to TAR RNA to ensure elongation of viral transcripts [ 31 ]. Since protein phosphorylation is a well-known regulatory mechanism for the control of transcription by a number of eukaryotic and viral proteins, and since phosphorylation of Rev, the other major regulatory protein of HIV-1, increases its ability to bind to RNA [ 32 ], it was important to determine if phosphorylation of Tat also resulted in the modification of its function. The binding of Tat and TAR RNA is a necessary step for Tat to mediate viral transcription elongation [ 33 - 35 ]. In electrophoretic mobility shift assays, we show that Tat-P bound more TAR RNA than Tat-N, and the Tat-P-TAR complex was more resistant to competition by excess unlabelled TAR RNA. Moreover, when the NaCl concentration in the binding buffer reached 1000 mM, the dissociation of the Tat-N-TAR complex was approximately 5 times greater than that of the Tat-P-TAR complex. Together, these observations appear to indicate faster, greater and stronger binding of Tat to TAR RNA after phosphorylation by PKR. Interestingly, phosphorylated HIV-1 Rev protein has been shown to bind RNA seven times more strongly than non-phosphorylated protein, and the non-phosphorylated Rev-RNA complex dissociates 1.6 times more rapidly than the phosphorylated complex [ 32 ]. However, the precise mechanism by which phosphorylated Tat accomplishes this remains to be elucidated. It may be that the phosphorylation of Tat changes its secondary structure. This may result in an increased net positive charge by either exposing basic amino acids or masking negative amino acids, and this increases the attraction to negatively charged RNA, as in the case of cAMP response element binding protein (CREB) phosphorylation by protein kinase A and glycogen synthase kinase-3 [ 36 ]. On the other hand, phosphorylation of Tat may change the conformation of the adjacent RNA-binding domain of Tat, as observed with the phosphorylation of proteins such as HIV-1 Rev [ 32 ] and serum response factor (SRF) [ 37 ]. We examined the effect of phosphorylation on Tat-mediated transactivation of the HIV-1 LTR in vivo with a series of mammalian expression constructs carrying the wild-type tat gene or mutants of the gene with alanine substitutions at one, two, or all three of the serine/threonine PKR phosphorylation sites. Firstly, we investigated the in vitro phosphorylation of Tat by PKR using Tat proteins expressed and purified from analogous bacterial expression constructs. These were subject to different levels of phosphorylation by PKR and displayed distinct kinetic behaviour. Nonlinear regression analysis of the proteins indicated that PKR could not phosphorylate S62 or T64 alone in the absence of S68. These results suggest a cooperative role for the phosphorylation of S68 in conjunction with S62 and T64, although the mechanism involved and the reason for cooperation require further investigation. Overall, a gradual reduction in phosphorylation was observed as the number of mutated phosphorylation sites increased, and any phosphorylation observed with the triple mutant was shown to be non-specific, thus confirming previous published results identifying S62, T64 and S68 as the only PKR phosphorylation sites [ 22 ]. However, these findings do not exclude the possibility that there could be other sites within Tat that could be subject to phosphorylation by other kinases. Co-transfection experiments with the mammalian expression constructs showed a 4-fold decrease in LTR transactivation with the Tat triple mutant which could not be phosphorylated by PKR. A gradual reduction in transactivation was observed as the number of mutated phosphorylation sites increased – a 2-fold reduction with the removal of one site, and 2.5-fold with the removal of two sites. Furthermore, the transfection data also suggested that the presence of S68 is necessary for optimal Tat-mediated transactivation, since its absence in conjunction with one or both of the other residues yielded the lowest levels of transcription. These results were in agreement with the in vitro phosphorylation data and support the hypothesis that phosphorylation of Tat may be important for its function in HIV-1 LTR transactivation. It is relevant to note that even in the absence of all three PKR phosphorylation sites the level of transcription was still 3-fold above baseline. This may imply that Tat can still transactivate in the absence of PKR phosphorylation, although at much reduced efficiency, and/or that the protein may be phosphorylated by other kinases at other sites, for example, PKC which phosphorylates Tat at S46 [ 38 ]. Alternatively, it may be that phosphorylation could be progressive between PKR and one or more other kinases as in the case of CREB protein [ 36 ]. Furthermore, the identification of a phosphatase in enhanced Tat-mediated transactivation [ 39 ] could point to a possible, finely tuned interplay and balance between kinases and phosphatases in Tat-mediated HIV-1 transcription. The mechanism by which the absence or presence of phosphorylation affects transactivation still requires further investigation. It could be that the introduction of an increasing number of mutations in the region 62–68 which lies next to the nuclear localization signal (aa 49–58) leads to conformational changes that prevent the protein from entering the nucleus. However, HIV-1 subtype C viruses which are rapidly expanding, carry mutations in Tat R57S and G63Q within and close to the basic domain, and yet exhibit increased transcriptional activity [ 40 ]. On the other hand, the phosphorylation of serines and threonines may facilitate the rapid folding and conformation of the protein necessary for full function as in the case of HIV-1 Rev [ 32 ]. Rev from the less pathogenic HIV-2 contains alanines in place of the serines required for phosphorylation [ 41 , 42 ]. It is possible to envisage a similar situation for Tat, where phosphorylation of the protein by PKR and possibly by other kinase(s) may also lead to rapid folding and changes in conformation. These changes may allow it to bind to more TAR RNA, more strongly, which in turn may lead to the formation of a stronger and more stable Tat-TAR-P-TEFb complex ensuring hyperphosphorylation of the RNAPII CTD and subsequent, successful viral transcript elongation. Conclusion Overall, these results suggest that the phosphorylation of Tat by PKR plays a key role in the ability of Tat to transactivate the HIV-1 LTR, allowing the virus to use the natural antiviral responses mediated by interferon to further its own replication. This may, in part, explain the observation of increasing IFN levels in patients with advanced AIDS. The gradual reduction in transactivation observed with the decreasing absence of phosphorylation residues suggest that the presence of all PKR phosphorylation sites within the protein may be required for the optimal function of Tat in transactivation, and that the absence of S68, especially when in combination with T64, has a greater negative impact on transactivation. Methods Plasmids and proteins The plasmid, pTZ18-TAR80 was a kind gift from Dr. E. Blair, and was used for in vitro transcription of TAR RNA after digestion with Hin D III. A β-actin luciferase reporter gene plasmid was used as a transfection control to normalize transfection efficiency and was provided by Assoc. Prof. Nick Saunders, CICR, University of Queensland, Brisbane. The pHIV-LTR-CAT construct used in transfection experiments, the destination vector, pET-DEST42 (Invitrogen, CA, USA), and the pET-DEST42-Tat86 construct were a gift from Dr. David Harrich, QIMR, Brisbane. The mammalian expression vector, pcDNA3.2-DEST was purchased from Invitrogen (CA, USA) and was used as the destination vector for the construction of the Tat86 wild-type and mutant constructs. Synthetic HIV-1 Tat(1–86) protein was a gift from Dr. E. Blair. The protein is a chemically synthesized, full-length HIV-1(Bru) Tat (amino acids 1–86). Histidine-tagged HIV-1 Tat86 was expressed in BL21(DE3)pLysS cells (Invitrogen, CA, USA) and purified in the laboratory of Dr. David Harrich, QIMR, Brisbane. Histidine-tagged HIV-1 Tat86 phosphorylation mutants were prepared as described elsewhere in this method. PKR was prepared as described elsewhere in this method. Preparation of histidine-tagged HIV-1 Tat86 phosphorylation mutants Bacterial expression constructs were prepared using the prokaryotic expression vector, pET-DEST42-Tat86. Mutations were introduced in the tat gene at the three PKR phosphorylation sites: serine 62, threonine 64 and serine 68, by site-directed mutagenesis using complementary synthetic oligonucleotide primers (Proligo, Genset Pacific, Lismore, Australia) encoding the mutation of the residue, or residues, to alanine. The reaction for site-directed mutagenesis contained 32 μL distilled water, 5 μL Pfu I 10X reaction buffer (Promega, USA), 100 ng pET-DEST42-Tat86, 5 μL 5' oligonucleotide primer at a concentration of 25 ng/μL, 1 μL 10 mM dNTP mix, and 3 Units Pfu I DNA polymerase (Promega, USA). The reaction was subjected to PCR with the following cycling conditions: 95°C for 30 seconds, 18 cycles at 95°C for 30 seconds/55°C for 1 minute/68°C for 15 minutes, hold at 4°C. Electrocompetent JM109 cells were prepared in the laboratory and transformed with 2 μL of PCR reaction. Minipreps were prepared from selected ampicillin-resistant colonies and sequenced to confirm the mutation in the construct. Mammalian expression constructs were prepared using Gateway Cloning Technology (Invitrogen, USA) to transfer the mutated tat genes from pET-DEST42-Tat86 wild type and mutants to the mammalian expression vector, pcDNA3.2-DEST, according to the protocol supplied by the manufacturer. Expression and purification if HIS-tagged Tat mutant proteins Competent BL21(DE3)pLysS cells (Dr. David Harrich, QIMR, Brisbane, Australia) were transformed with 1 μL of pET-DEST42-His-Tat86 wild-type or mutants, and plated. A single ampicillin resistant colony was resuspended in 10 mL of LB broth/amp and incubated overnight at 37°C. This culture was added to 500 mL of LB broth/amp and incubated in an orbital shaker, at 37°C until the OD 600 was 0.6. The culture was inoculated with IPTG (Roche, Germany) to a final concentration of 200 μg/mL and incubation was continued for a further 2 hours. Cells were pelleted; the pellet was resuspended in 2 volumes of 6 M guanidine-HCl, pH 8.0 and incubated at room temperature overnight. The suspension was centrifuged at 14500 × g for 20 minutes, and the supernatant was centrifuged at 100 000 × g for 30 minutes. The supernatant was loaded onto a 1 mL equilibrated, packed resin (TALON™ Metal Affinity Resin, BD Biosciences Clontech, USA). To equilibrate, the resin was washed twice with 10 mL of Milli-Q water and charged by incubating with 5 mL of 0.3 M CoCl 2 at room temperature for 5 minutes. The resin was then washed extensively with water, and equilibrated in 6 M guanidine-HCl, pH 8.0. The HIS-tagged protein was allowed to bind to the resin by incubation on a rocking platform, at room temperature, for 1 hour. The resin was then sedimented at 700 × g for 2 minutes, and washed with 6 M guanidine-HCl, pH 8.0 for 5 minutes. The resin was sedimented as above and washed with 6 M guanidine-HCl, pH 6.0 for 5 minutes. The resin was loaded onto an empty column (Poly-Prep ion exchange column, Bio-Rad, USA), and the wash allowed to flow through. The HIS-tagged protein was eluted with 4 mL of 6 M guanidine-HCl, pH 4.0, and collected in 500 μL fractions. Fractions were dialysed in 0.1 mM DTT in PBS, at room temperature, overnight, and then centrifuged at 14500 × g for 2 minutes. To identify fractions containing the HIS-tagged protein, 5–20 μL aliquots were analysed by 15% SDS-PAGE and stained with Coomassie blue. Fractions containing protein were assayed for protein concentration (Bio-Rad Protein Assay Dye Reagent Concentrate, Bio-Rad, USA), and by Western blot against a 1:1000 dilution of monoclonal anti-poly HISTIDINE Clone HIS-1 antibody (Sigma Aldrich, USA). Aliquots of fractions were stored at -80°C in 10 mM DTT in PBS. In vitro phosphorylation assays PKR was purified from HeLa cell extracts as described previously [ 43 ]. Briefly, confluent HeLa cells in 75 cm 2 flasks were lysed in 1 mL of Buffer 1 (20 mM Tris, pH 7.6, 50 mM KCl, 400 mM NaCl, 1 mM EDTA, 1% Triton X-100, 20% glycerol, 200 μM PMSF, 5 mM mercaptoethanol), and centrifuged at 13500 × g for 30 minutes at 4°C. The supernatant was incubated in ice, for 30 minutes, with 2 μL of a 1:10 dilution of specific monoclonal antibody 71/10 (Dr. A. Hovanessian, Pasteur Institute, France), and then at 4°C overnight with 65 μL of protein G-sepharose (Amersham Biosciences, Sweden), with continuous rotation. Protein G-sepharose-PKR was sedimented, washed three times with Buffer 1, and three times with DBGA (10 mM Tris, pH 7.6, 50 mM KCl, 2 mM magnesium acetate, 20% glycerol, 7 mM β-mercaptoethanol). PKR was activated by incubating 120 μL of this suspension with 80 μL of DBGB (DBGA + 2.5 mM MnCl 2 ), synthetic dsRNA (Sigma Aldrich, USA) to a final concentration of 0.5 μg/mL, and 20 μL of 2 mg/mL ATP (Sigma Aldrich, USA), at 30°C for 15 minutes. Phosphorylation reactions for Tat proteins contained 2 μg of HIV-1 Tat, unless otherwise indicated in the figure legend, 6 μL of activated PKR suspension, and DBGA to a final volume of 12 μL. Phosphorylation was performed at 30°C for 1 hour, unless otherwise stated, in the presence of 2 μCi of γ- 32 P-ATP (Perkin-Elmer, USA). For measuring the extent of phosphorylation of the mutant Tat proteins, phosphorylation was stopped after 2, 5, 10, 15, 30, 45, and 60 minutes by the addition of protein loading buffer. Samples were analysed by 15% SDS-PAGE, and proteins were visualized by autoradiography, and scanning densitometry in a STORM 860 phosphorimager with ImageQuant ® software (Molecular Dynamics, USA). Electrophoretic mobility shift assay (EMSA) TAR RNA was synthesized from 0.8 μg of pTZ18TAR80 using a commercial in vitro transcription system (MAXIscript™ T7 kit, Ambion, USA) according to the protocol supplied with the kit. HIV-1 Tat was phosphorylated (Tat-P) with activated PKR for 1 hour, as described above, or in the absence of γ- 32 P-ATP (Tat-N). Tat-P and Tat-N were allowed to equilibrate at 30°C for 10 minutes in Binding Buffer (10 mM Tris, pH 7.6, 1 mM DTT, 1 mM EDTA, 50 mM NaCl, 0.05% glycerol, 0.09 μg/μL BSA), before incubating at 30°C for 10 minutes with 2.5 × 10 5 cpm of 32 P-TAR RNA. The Tat-TAR RNA complexes were separated on a 5% acrylamide/0.25X TBE gel (0.45 M Tris, 0.45 M boric acid, 0.1 M EDTA, pH 8.0), for 3–4 hours, at 10 mA, and visualized by autoradiography. Transfection assays Transfections were performed in duplicate in 6-well plates. HeLa cells were diluted in Modified Eagle's Medium (Invitrogen, USA) supplemented with 10% foetal bovine serum (Trace Scientific, Melbourne, Australia), antibiotics and glutamine (Invitrogen, USA), to yield 5 × 10 5 cells/mL. Each well was seeded with 2 mL of this cell suspension, and incubated at 37°C/5% CO 2 for 24 hours or until the cell monolayer was 80–90% confluent. A solution of 625 μL of serum-free medium and 10 μg of total DNA (3.3 μg β-actin-luciferase, 3.3 μg pcDNA3.2-DEST-Tat, 3.3 μg pHIV-LTR-CAT) was mixed with 600 μL of serum-free medium containing 25 μL of Lipofectamine 2000 (Invitrogen, USA), and incubated at room temperature for 20 minutes. The cells were washed twice with serum-free medium, inoculated with the DNA-Lipofectamine mixture, and incubated at 37°C for 6 hours. The DNA solution was replaced with complete medium and the cells wee incubated as above for 24 hours. The cells were harvested and assayed for CAT activity using the CAT ELISA kit (Roche, Switzerland) according to the protocol supplied with the kit, for luciferase activity using the Luciferase Assay System (Promega, USA) according to the supplied protocol, and for protein concentration (Bio-Rad Protein Assay Dye Reagent Concentrate, Bio-Rad, USA). Competing interests The author(s) declare that they have no competing interests. Authors' contributions LEM was responsible for the experiments described and contributed to the drafting of the manuscript. TW performed the optimization experiments for the phosphorylation of Tat by PKR. DH participated in the design of the study, provided reagents and critically read the manuscript. NAJM conceived and coordinated the study, and contributed to the drafting of the manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC556014.xml
546207	Severe falciparum malaria in Gabonese children: clinical and laboratory features	Background Malaria continues to claim one to two million lives a year, mainly those of children in sub-Saharan Africa. Reduction in mortality depends, in part, on improving the quality of hospital care, the training of healthcare workers and improvements in public health. This study examined the prognostic indicators of severe falciparum malaria in Gabonese children. Methods An observational study examining the clinical presentations and laboratory features of severe malaria was conducted at the Centre Hospitalier de Libreville, Gabon over two years. Febrile children aged from 0 to 10 years with Plasmodium falciparum infection and one or more features of severe malaria were enrolled. Results Most children presenting with severe falciparum malaria were less than 5 years (92.3% of 583 cases). Anaemia was the most frequent feature of severe malaria (67.8% of cases), followed by respiratory distress (31%), cerebral malaria (24%) hyperlactataemia (16%) and then hypoglycaemia (10%). Anaemia was more common in children under 18 months old, while cerebral malaria usually occurred in those over 18 months. The overall case fatality rate was 9%. The prognostic indicators with the highest case fatality rates were coma/seizures, hyperlactataemia and hypoglycaemia, and the highest case fatality rate was in children with all three of these features. Conclusions Prompt and appropriate, classification and treatment of malaria helps identify the most severely ill children and aids early and appropriate management of the severely ill child.	Introduction Each year 500 million infections and up to 2.7 millions deaths are attributable to malaria [ 1 ], about 90% of these deaths occur in children in sub-Saharan Africa [ 2 ]. Eighty percent of the deaths occur during the first 24 hours following admission [ 3 - 5 ]. Despite a better understanding of pathophysiology and management of malaria, childhood mortality remains unacceptably high [ 6 ]. The acquisition of malaria immunity is closely linked to the level of transmission and severe Plasmodium falciparum infection is very rare after the age of 5 years in highly endemic areas. Presentations of severe malaria are different at different ages and in areas with different levels of transmission. In Gabon, a country of about 1.2 million people, malaria is the main cause of neurological, haematological and infectious emergencies at the Centre Hospitalier de Libreville (CHL), the country's tertiary referral centre [ 7 ]. Malaria transmission is hyperendemic and perennial with an entomological inoculation rate of 50 per person-year [ 8 ]. Previous studies in Africa have shown three frequent presentations of severe falciparum malaria: cerebral malaria, metabolic malaria (hyperlactaemia, acidosis or respiratory distress) and severe anaemia [ 3 , 5 , 9 ]. The case fatality rate of severe anaemia, however, is low and in some studies it is not an independent predictor of death [ 9 , 10 ]. In Africa, malaria mortality remains high for a number of reasons including limited access to healthcare and increased drug resistance [ 6 ]. Better classification of severe malaria could aid clinicians caring for children with severe malaria to avoid diagnostic delays, identify the children most likely to die and thus improve management by targeting resources to the sickest children. This prospective observational study was designed to determine the clinical and laboratory features that identify those children most severely ill with malaria. Methods This study was carried out in the CHL between 1st August 2000 and 31st July 2002. The CHL is the largest public hospital in the country, situated in Libreville, the capital city (pop. 500,000). Services at the CHL include an Emergency Department (20 beds), a Paediatrics Unit (80 beds) and an Intensive Care Unit (12 beds). Ethical permission for the study was granted by the Gabonese Ministry of Health. Febrile children were referred the study team and seen on admission by a Malaria Clinical Research Unit (MCRU) clinician, summary data were recorded on a pro-forma sheet. A blood sample (2 ml) was drawn (anti-coagulated with EDTA) for quantitative examination of a blood film for malarial parasites using the Lambaréné method [ 11 ], measurement of haemoglobin concentration, white cell count, platelet count (STKS, Coulter Corporation) and blood lactate and glucose concentrations. Within 15 minutes of blood sampling, blood lactate and glucose concentrations were measured using Accusport™ hand held analyser (Bohringer, Manheim, Inc., Germany) and One-touch Analyzer (LifeScan, Inc., USA) [ 11 ]. Blood films were defined as negative if there were no asexual forms of P. falciparum in 100 high power fields of a thick film. The schematic process of the inclusion is shown in Figure 1 . Figure 1 Schematic process of the screening of febrile children in the Centre Hospitalier de Libreville, Gabon. Febrile children (or those with history of fever in the last 48 hours) were considered for inclusion in the study if they were: aged 0 to 10 years of age (inclusive), had malaria (> 2 asexual forms of P. falciparum seen on blood film) and had one or more of the following features of severity: [ 9 , 12 , 13 ]: Blantyre coma score (BCS) ≤ 2 defining cerebral malaria, repeated observed seizures (3 or more observed in 24 hours), lactate concentration in whole blood or capillary blood ≥ 5 mmol/L, glucose concentration in whole blood or capillary blood ≤ 2.2 mmol/L, severe anaemia (haemoglobin concentration of < 5 g/dL) and/or haematocrit concentration < 15%). Children were excluded from the study if informed consent was not obtained from a relative or if an alternative diagnosis was made clinically or by investigation (such as cerebrospinal fluid examination, chest radiography or blood culture). Respiratory distress was defined as the presence of one or more of these features [ 14 ]: abnormalities in respiratory rate (according to the age), rhythm (Kussmaul's or Cheyne-Stokes's breathing) and signs of distress such as nasal flaring, intercostal or subcostal recession. Management All children enrolled were hospitalised and treated with parenteral quinine (12.5 mg salt/kg/day, intravenous; Quinimax*, Sanofi-Synthelabo, France) without a loading dose, followed by oral quinine when tolerated. Pyrexial children received paracetamol suppositories (60 mg/kg/day, rectal; Efferalgan™, Bristol- UPSA, France). Seizures were controlled with diazepam (0.3 mg/kg, iv or 0.5 mg/kg rectal; Valium™, Roche, France). Severe anaemia was corrected by transfusion of packed red cells (15 ml/kg over 4 hours) screened for blood borne infections. Hypoglycaemia was treated with a slow intravenous injection of hypertonic glucose 40 %(Braun, Germany) at a dose of 1 ml/kg. Nasal oxygen at 6 1/minute was given to children with respiratory distress. Follow up An MCRU clinician performed a full physical examination daily until discharge for each child. Laboratory assessments including parasitaemia, blood glucose and lactate were performed during hospitalization as necessary. The outcome (survived, death) was recorded. Statistical analysis Statistical analysis was carried out with Epi info 6.04 (ENSP-Epiconcept- InVS, Corp.) and Stata Statistical Software (version 7.0, Stata Corporation, College Station, Texas, USA). Normality of data distribution was checked using either Shapiro-Wilks or Kolmogrov -Smirnov test. Normally distributed data were analysed by two-tailed Student's T test and non-normally distributed data with the Mann-Whitney U statistic. Proportions were compared with χ 2 tests with Yates' correction or Fisher's exact test. ANOVA test were used for multiple comparisons of variances, with Tukey's post hoc test. Assessments of prognostic factors were conducted with logistic regression model. A p value < 0.05 was considered as significant. Specific prevalence for each subgroup has been defined as the ratio of the number of cases observed in this sub-group over the population of this same sub-group. Hyperlactataemia is usually defined as a blood lactate concentration of ≥ 5 mmol/L. As the Accusport has been found to have poor agreement with the gold standard YSI 2300 [ 11 ] or YSI 1500 sport [ 15 ], a definition of hyperlactataemia as blood lactate concentration ≥ 10 mmol/L was used to increase the specificity in the analysis. Results Demographic and clinical data During the study period (1st August 2000-31st July 2002), 8,036 febrile children were screened for malaria at the hospital. The data for 7,980 of these children were analysed: 4,368 male (54.5%) and 3,612 female (45.5%). Seventy-five percent these were less than 5 years of age. 3,156 (39.3%) of the 8,036 febrile children screened in the MCRU had a positive blood film for P. falciparum . A lower prevalence of malaria was seen in children aged< 6 months (3.7%, n = 118, p < 0.001). Specific prevalence of malaria rises after the 6 first months of life until it reaches a maximum at 47 months (47.5%), after which it declines again. Severe anaemia was most frequent in children less than 24 months old with 68 % of the cases of severe malarial anaemia occurring before this age. In contrast, the highest specific prevalence of cerebral malaria was found in children aged > 12 months. Sixty five percent of cases of cerebral malaria occurred in children aged between 12–48 months. Characteristics of severe malaria The admission clinical, laboratory and parasitological characteristics of the 583 children with severe malaria are shown in the Figure 2 . Two hundred and ninety nine of the severe malaria cases were male (51.3%) and 536 (92%) of the children with severe malaria were less than 5 years old. A history of vomiting, seizures and anti-malarial treatment before admission were reported in 322 (55.2%), 267 (45.8%) and 315 (54%) of the children with severe malaria respectively. Despite significant fluctuations in rainfall, the number of malaria cases per month was sustained during the study period, confirming the perennial transmission of malaria in this region. Figure 2 Admission characteristics of the study population (All results are mean ± SD except those specified. Severe anaemia was the commonest feature of severe malaria present in 395 (67.8%,) of the children. Neurological presentations (either coma or repeated convulsions) were present in 228 (39.2%) of the children. Respiratory distress occurred in 181 (31%) of the children ; hyperlactataemia in 73 (15.7%) and hypoglycaemia in 33 (6.2%) of the children with severe malaria. Hyperparasitaemia (> = 20% of circulating infected red blood cells) was relatively rare, occurring in only 24 (4.1%) of the children with severe malaria. Renal failure, acute pulmonary oedema and spontaneous bleeding are uncommon complications of childhood malaria [ 5 , 9 , 10 ] and were not seen in this study. Circulatory collapse was not found alone in this study and was not considered in the statistical analysis. One hundred and forty six (25%) children with severe malaria were afebrile on admission to hospital. Case rate fatality Figure 3 shows a Venn diagram summarizing clinical and laboratory features of severe malaria and case fatality rates. Figure 3 Venn diagram of features of 463 cases with a complete dataset of clinical measures. Case fatality rates shown in brackets. Of 583 patients, 52 died (40% male), giving an overall case fatality rate of 8.9%. Seven children were lost to follow-up. Forty seven deaths (90%) occurred within the first 24 hours after admission. Neurological sequelae were present in 27 (5 %) of the 531 survivors. The case fatality rate was significantly higher in females than males (10.9% vs.6.9%, p = 0.006). Children who died were older than those who survived (mean (SD) = 35.4 (41.9) vs. 25.0 (18.2) months, p = 0.0009). Of the 52 deaths in the course of the study, 32 (61.5%) presented with cerebral malaria, 32 (61.5%) presented with respiratory distress, and 20 (50% of the 40 with measurements) had hyperlactataemia. Thirty (60%) of the children who died presented with convulsions, and 13 (25%) with hypoglycaemia. Twenty four (46.2%) of those who died had severe anaemia. Mortalities in each sub-group demonstrated that hyperparasitaemia (1 death in 24, 4%) and severe anaemia (24 deaths in 395, i.e. 6%) had a better prognosis than cerebral malaria (32 deaths in 142, i.e. 22.5%) and hypoglycaemia (15 deaths in 60, i.e. 25%). Thirty-two (17.7%) of the 181 patients with RD died, as did 30 (11%) of the 264 patients with convulsions and 20 (27.4 %) of the 73 patients with hyperlactataemia. A multiple logistic regression model identified coma (OR = 3.6, 95% CI = 1.8–7.1, p < 0.001), hyperlactataemia (OR = 6.98, 95% CI = 3.5–13.8, p = 0.0001), respiratory distress (OR = 2.0, 95% CI = 1.0–3.9, p = 0.033) and hypoglycaemia (OR = 4.0, 95% CI = 1.7–9.4, p = 0.001) as independent predictors of a fatal outcome. Severe anaemia, hyperparasitaemia and thrombocytopaenia were not shown to be predictors of death (Figure 4 ). Figure 4 Prognostic indicators at the time of admission. OR – odds ratio, SD – standard deviation. Discussion Malaria remains a serious health problem in sub-Sahara Africa. It was the most common reason for neurological emergencies during 2001 at the CHL [ 7 ]. This study was designed to describe the epidemiology, clinical and laboratory presentations of severe falciparum malaria in childhood presenting to CHL, in order to improve the diagnosis, classification and appropriate management of malaria. It is not possible to exclude absolutely all children with alternative diagnoses on clinical examination and simple investigation alone, which is a problem shared by all other similar studies on severe malaria [ 3 - 5 , 9 , 10 , 14 , 16 - 18 ]. The small numbers with alternative diagnoses should not affect the conclusions of this or other studies. Most cases (92%) of severe malaria were in children less than 5 years old. Similar observations have been made in another group of children hospitalized for malaria in Gabon [ 19 ]. Elsewhere, severe malaria tends to occur in older children [ 20 ]. Differences in the age of presentation of severe malaria may be the result of lower background immunity or other undefined variables [ 21 ]. The study confirms the stable and perennial transmission of malaria in Gabon, which contrasts with reports from other countries in West Africa where malaria transmission is predominantly at the end of the long rainy season [ 22 , 23 ]. Fever is a characteristic feature of P. falciparum infection, but a sizeable proportion of these children (25%) with severe malaria were afebrile on admission as observed elsewhere [ 23 ]. Self-medication with antipyretic or antimalarial agents was common (about 50% of the children) and may contribute to this finding. There are obvious implications for the diagnosis of malaria, which may be underestimated using clinical criteria alone. Severe anaemia was the most frequent feature of severity in this study, but was associated with decreased mortality. A similar observation in a recent Ghanaian study showed a better outcome in children with severe anaemia [ 17 ]. These findings confirm that severe malaria anaemia has a lower case fatality rate than other complications of severe malaria, consistent with several other studies where severe anaemia was not an independent predictor of in-hospital mortality [ 9 , 10 , 18 ]. The case fatality rate of severe anaemia without other markers of severe malaria is 1 to 2%, where blood transfusion is available [ 3 , 9 , 10 , 14 , 18 ] raising questions about the value of severe anaemia as a defining feature of the syndrome of severe malaria. Despite the increasing toll of HIV infection, and the continuing burden of diarrhoeal disease, malnutrition and respiratory tract infections, malaria remains a major cause of childhood death in endemic regions [ 1 ]. The overall case fatality rate of severe malaria in the study was 8.9% (52 deaths/583 cases), which is in keeping with studies from other geographic areas, where case fatality rates range between 8 and 40% [ 4 , 5 , 9 , 14 , 16 , 20 , 22 , 24 , 25 ]. Most of these deaths (90% in this study) occurred in the first twenty-four hours of hospital admission, a finding also in keeping with other studies [ 5 ]. The independent prognostic indicators in this study were cerebral malaria, respiratory distress, hypoglycaemia and hyperlactataemia. These observations are entirely consistent with a large number of studies where the independent predictors of a fatal outcome in malaria are impaired consciousness and metabolic dysfunction (as measured by hyperlactataemia, hypoglycaemia, acidosis or respiratory distress) [ 3 , 5 , 9 , 10 , 14 , 16 - 18 ]. The metabolic complications of malaria are complex and a number of interrelated measures have been used in different studies. Severe malaria is associated with a metabolic acidosis [ 16 ] and hyperlactataemia [ 5 ]. Respiratory distress has been associated with acidosis and hyperlactataemia in some studies [ 26 ]. These features of metabolic malaria probably all result from increased anaerobic metabolism due to tissue hypoxia [ 27 ]. Estimates of the prevalence of hypoglycaemia have been reported in Africa, ranging from 8% to 34% [ 28 , 29 ]. In severe childhood malaria hypoglycaemia results from impaired gluconeogenesis and increased tissue demand for glucose [ 27 , 28 ] and quinine induced hyperinsulinaemia. Blood glucose concentrations should be monitored in all children hospitalised for malaria especially those who receive quinine. The definition of hyperlactataemia used in this study was a blood lactate concentration higher than the conventional cut-off (≥ 5 mmol/L). This was necessary because of the limitations of the analyser used, but probably means that the frequency of true hyperlactataemia was underestimated. The Accusport™ analyser used has been shown to have poor agreement with "gold standard" machines [ 11 , 15 , 30 ]. Hyperlactataemia is a frequent and serious complication of severe malaria in childhood [ 5 , 9 , 10 ], which may be due microcirculatory sequestration of parasitized erythrocytes resulting in increased production of lactate by anaerobic glycolysis [ 31 ]. A recent study showed a correlation between hyperlactataemia and high plasma glutamine levels in severe malaria. This correlation may reflect impaired gluconeogenesis [ 31 ]. Lactate disposal is proportional to blood lactate concentration and can be increased by dichloroacetate [ 27 , 32 ]. Lactic acidosis, as confirmed in this study, is an established strong predictor of a fatal outcome in falciparum malaria in African children [ 5 , 9 , 10 ] and may prove a target for further interventional studies to improve survival. Respiratory distress was present in 31% of these children. This is higher than the frequency reported in other studies of severe malaria: 4.9% in Burkina [ 20 ], 6.4% in Togo [ 22 ] and 13.7% in Kenya [ 14 ]. These differences may partly be explained by low inter-observer agreement for this variable, geographical variations in disease pattern as well as differing definitions of severe malaria. Results from many studies consistently show that respiratory distress is a life-threatening syndrome in childhood malaria [ 14 , 33 ]. Respiratory distress was significantly associated with both hyperlactataemia and cerebral malaria in this study. The Blantyre coma score has long been established in children as a good indicator of cerebral dysfunction in malaria [ 3 ] and has enabled better standardization of studies on cerebral malaria in African children. The case-fatality rate associated with cerebral malaria (22%) is similar to that in Gambian children (27%) [ 4 ] but is higher than that observed for Kenyan (17%) [ 14 ] and Malawian children (15%) [ 3 ]. It has been postulated that with the higher the level of malaria transmission, immunity is acquired earlier, perhaps altering the presentation of severe malaria from predominantly a cerebral syndrome to that of severe anaemia [ 34 , 35 ]. The clinical and laboratory presentations of severe malaria are described in a hospitalized population of children in Gabon. The severe cases are likely to be only the "tip of the iceberg", many children living far from health care units may die whilst travelling to the nearest hospital. Most deaths from malaria occurred in the first 24 hours of admission, which highlights the need for early recognition of the most severely ill children. Early diagnosis and classification of severe malaria would allow appropriate management, including basic adjunctive therapy such as to prevent hypoglycaemia, and better use of scarce healthcare resources. Together these improvements could contribute to a reduction in the intractably high mortality due to the disease. Authors' contributions AD is a MCRU clinician. He participated to the study and wrote the article. PN conducted the study for his MD thesis. RT and MB are paediatricians who participated in the study. TP is a UK collaborator who participated in the study and helped write the article. MM was involved with the intensive care of the children. UMR participated in the study as a MCRU clinician. JJ helped to write the article. EK did all the statistical analysis. ENM, PK, SK and MK coordinated the realization of the study and edited the final version approved by all authors.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC546207.xml
514541	Retrieving sequences of enzymes experimentally characterized but erroneously annotated : the case of the putrescine carbamoyltransferase	Background Annotating genomes remains an hazardous task. Mistakes or gaps in such a complex process may occur when relevant knowledge is ignored, whether lost, forgotten or overlooked. This paper exemplifies an approach which could help to ressucitate such meaningful data. Results We show that a set of closely related sequences which have been annotated as ornithine carbamoyltransferases are actually putrescine carbamoyltransferases. This demonstration is based on the following points : (i) use of enzymatic data which had been overlooked, (ii) rediscovery of a short NH 2 -terminal sequence allowing to reannotate a wrongly annotated ornithine carbamoyltransferase as a putrescine carbamoyltransferase, (iii) identification of conserved motifs allowing to distinguish unambiguously between the two kinds of carbamoyltransferases, and (iv) comparative study of the gene context of these different sequences. Conclusions We explain why this specific case of misannotation had not yet been described and draw attention to the fact that analogous instances must be rather frequent. We urge to be especially cautious when high sequence similarity is coupled with an apparent lack of biochemical information. Moreover, from the point of view of genome annotation, proteins which have been studied experimentally but are not correlated with sequence data in current databases qualify as "orphans", just as unassigned genomic open reading frames do. The strategy we used in this paper to bridge such gaps in knowledge could work whenever it is possible to collect a body of facts about experimental data, homology, unnoticed sequence data, and accurate informations about gene context.	Background As a consequence of the deluge of completely sequenced genomes belonging to a large array of species, one can expect to identify many homologues of enzymes which have been previously well studied at the experimental level. This seems to be the general rule and the public sequence databanks (DDBJ/EMBL/GenBank) are now inundated by putative amino acid sequences which have been annotated uniquely by the widely used two-step process : (1) detection of a homologous relationship by a pairwise sequence similarity search at the level of primary structure and (2) inference of functional similarity from this detected homology. However, the opposite might be true. For various reasons one can either miss or misinterpret the actual function of a putative protein when annotating by homology, resulting in a wrong function transfer. Several studies have already emphasized this point (see, for example, [ 1 - 3 ]). On the other hand, beside these now well identified errors which are often due to automatic processes, more subtle mistakes may occur when some of the numerous effects of divergent evolution are overlooked. In particular, one of the insufficiently appreciated problems of functional assignment is that homologous proteins might catalyse different biochemical reactions. Here, we discuss an instance of erroneous annotation (misannotation) in genes of nitrogen metabolism which to our knowledge has not yet been brought up. We explain why this is so and draw attention to the fact that similar cases must actually be rather frequent. Results and Discussion Annotating distant carbamoyltransferases Our group ([ 4 , 5 ]) is presently involved in deciphering the evolutionary relationships between two ubiquitous and essential proteins, aspartate carbamoyltransferase (ATCase, EC 2.1.3.2) which catalyses the first committed step of de novo pyrimidine biosynthesis and ornithine carbamoyltransferase (OTCase, EC 2.1.3.3) which plays a crucial role in both anabolism and catabolism of arginine. In a recent study of the phylogeny of the 245 available OTCases (paper in preparation), we confirmed the existence of two families, OTC alpha and OTC beta, previously proposed on the basis of phylogenetic studies [ 4 ]. However, the advent of many new sequences further led to a more complex topology of the distance tree schematized on Fig. 1 . It appears now that a significant number of sequences which have been annotated as OTCases are distantly related to either family as outlined on Fig. 1 . These sequences, which are forming several clusters emerging in different locations between the root and the two families OTC alpha and OTC beta, have been provisionally annotated as UTC (unknown carbamoyltransferases). Indeed, among these UTC we found two sequences which do not correspond to classical OTCases. Figure 1 A schematic evolutionary tree of carbamoyltransferases. A distance tree of the 245 available ornithine carbamoyltransferases has been obtained using the PhyloTree programme [33]. This tree was rooted by a few paralogous aspartate carbamoyltransferases belonging to either family ATC I or ATC II ([4], [5]). The root is indicated by a closed circle. The two families OTC alpha and OTC beta are schematized as triangles. The homologous carbamoyltransferases which are branching far from these ATC or OTC families are labelled as unknown (UTC) even if some of them have been annotated as ornithine carbamoyltransferases (see text). The few UTC sequences which have been annotated as carbamoyltransferase-like are enclosed in ovals. The clade of reannotated putrescine carbamoyltransferases is framed in red. The YgeW protein encoded by Escherichia coli and its close homologue from Clostridium botulinum are both located on a long branch emerging at the basis of this OTCases tree. YgeW is annotated as belonging to the ATCase/OTCase family (see, for example, the SwissProt knowledgebase [ 6 ]). On the branch which is next to the root we find the sequence of a protein which has been reported to be essential for arginine biosynthesis in the anaerobic bacterium Bacteroides fragilis [ 7 ]. This protein has been crystallized and characterized as a carbamoyltransferase-like protein since it does not display OTCase activity in vitro [ 7 ]. Indeed, several of its residues have been substituted in sites which are viewed as crucial for OTCase activity. Moreover, Dashuang et al . [ 7 ] indicated that a similar protein has been found in Xylella fastidiosa . Our phylogenetic data are in agreement with this observation since the protein annotated as OTCases in two strains of X. fastidiosa and its close relative present in two species of the Xanthomonas genus are found to branch close to that of B. fragilis . Therefore, the functional identification of these different UTC is certainly not straightforward and requires further investigations. Furthermore, it occurred to us that, more than thirty years ago, another carbamoyltransferase was discovered by Roon and Barker [ 8 ]. A putrescine carbamoyltransferase (PTCase, EC 2.1.3.6) was found to be synthesized by the Gram-positive bacterium Streptococcus (now Enterococcus ) faecalis when it was grown on agmatine but not arginine as primary energy source. This PTCase was easily separated from the OTCase synthesized by the same organism grown on arginine [ 8 ]. This putrescine carbamoyltransferase had further been studied by V. Stalon's group ([ 9 - 12 ]). Two features of this study – which had apparently been overlooked in recent genome annotations – appear now to be crucial for the interpretation of the data shown on Fig. 1 . (1) Wargnies et al . [ 9 ] showed that the putrescine carbamoyltransferase of E. faecalis had a weak but unambiguous OTCase activity (7.4% in terms of Vmax, with K M for putrescine and L-ornithine of 0.029 mM and 13.0 mM respectively); (2) A short NH 2 -terminal sequence (29 residues) was published ten years later [ 13 ]. Since the complete genome of E. faecalis has now been sequenced [ 14 ], we could identify that one of the three putative ornithine carbamoyltransferases encoded by this genome, the open reading frame EF0732 annotated as ArgF-2 [ 14 ], is actually the putrescine carbamoyltransferase previously studied by the group of Stalon. A family of putrescine carbamoyltransferases In a second step, we extended this reannotation of a wrong OTCase as a PTCase to six other sequences encoded by Lactococcus lactis , Streptococcus mutans, Pediococcus pentosaceus, Lactobacillus brevis (and a very close partial sequence in Lactobacillus sakei ) , Listeria monocytogenes and Mycoplasma mycoides , respectively. Indeed, these eight sequences, which have been annotated as either ArgF or ArcB (Table 1 ), (i) share high identity at the level of their amino acid sequence; (ii) they form a monophyletic group (Fig. 1 ) and (iii) they match the previously published E. faecalis NH 2 -terminal 29 residues sequence [ 13 ]. Moreover, as shown on Fig. 2 , these sequences share several specific motifs which are not found in the homologous OTCases. These motifs, especially the five longer ones, are well conserved, even in M. mycoides which is however more distant. When these motifs are used together to query either the UniProt knowledgebase [ 15 ] or the nr (non-redundant) database using the PHI-Blast programme, we obtain only these putative PTCase sequences (including that of M. mycoides ) to the exclusion of any other carbamoyltransferase. Table 1 The seven ornithine carbamoyltransferases sequences which have been reannotated as putrescine carbamoyltransferases. Species name gene name CDS Swissprot TREMBL Web link Enterococcus faecalis argF-2 EF0732 Q837U7 Lactococcus lactis argF2 LL1700 Q9CEY4 Streptococcus mutans arcB SMU.262 Q8DW19 Pediococcus pentosaceus a - Scaffold 18 Gene 459 - Lactobacillus brevis a - Scaffold 15 Gene 476 - Lactobacillus sakei b argF - Q8RPX3 Listeria monocytogenes arcB LMO0036 Q8YAS7 Mycoplasma mycoides arcB MSC_0703 CAE77322 a : draft genome b : fragment Figure 2 Long motifs specific of putrescine carbamoyltransferases. The amino acid sequence of the open reading frame EF0732 of E. faecalis annotated as ArgF-2 [14] but now reannotated as putrescine carbamoyltransferase is shown. The NH 2 -terminal 29 residues sequence [13] which helped to this reannotation is indicated by a red bar above the respective residues. The amino acids which are specifically conserved in putrescine carbamoyltransferases and not in ornithine carbamoyltransferases are printed in bold. Inside the long motifs (> 15 residues) found along the whole sequence and numbered as "Motif 1" to Motif 5"are indicated the few substitutions present in the other sequences (lines above the E. faecalis sequence) including those of M. mycoides which is more distant (line in italic below the E. faecalis sequence). The residues forming the respective binding sites of carbamoylphosphate and of either putrescine or ornithine are also in bold and framed. Gene context, another tool for gene reannotation In a third step, the reannotation of this clade of OTCase sequences as PTCases was confirmed by a comparative study of the neighbourhood of their encoding genes present in the four genomes completely sequenced and published ( E. faecalis , Lc. lactis , S. mutans and L. monocytogenes ). As shown on Fig. 3 , the same set of neighbouring genes were present in these four species. We have successively a transcriptional regulator, the reannotated PTCase, an amino acid permease (probably an antiporter), a conserved hypothetical protein and finally the carbamate kinase ArcC-3 (EC 2.7.2.2). In a next step, we found that the so-called conserved hypothetical protein is homologous to the agmatine deiminase (or agmatine iminohydrolase, EC 3.5.3.12) of Bacillus cereus [ 15 ]. Fig. 3 further shows that the gene order found in E. faecalis , is completely conserved in Lc. lactis and S. mutans and slightly modified in L. monocytogenes . Figure 3 Conservation of the gene context of putrescine carbamoyltransferases. The open reading frame (ORF) EF0732 of E. faecalis is shown with neighbouring ORFs. Under each numbered ORF (EF0731 to EF0735) is indicated the putative function proposed by the annotators of this complete genome [14]. Below is the reannotation we are proposing. This schematic representation is repeated for the similar genomic regions present in three other completely sequenced genomes, those of Lc. lactis , S. mutans and L. monocytogenes . The homologous sequences are indicated by using the same color : white for transcriptional regulator, yellow for PTCase, pink for permease, green for agmatine deiminase, blue for carbamate kinase, respectively. Thus, these four clusters of genes appear to encode the full set of enzymes which are expected to form the catabolic agmatine deiminase pathway [ 10 ]. Agmatine deiminase, PTCase and carbamate kinase were already known to become coinduced by agmatine in E. faecalis when it is used as sole energy source [ 10 ], strongly suggesting that these gene clusters are functional operons. In Pseudomonas aeruginosa PAO1, the homologous agmatine deiminase is encoded by the gene aguA belonging to an operon aguBA induced by agmatine and N-carbamoylputrescine ([ 16 , 17 ]) but in this species N-carbamoylputrescine is converted by a N-carbamoylputrescine amidohydrolase (EC 3.5.1.53, the aguB product) into putrescine and CO 2 + ammonium rather than into putrescine and carbamoylphosphate. More recently, a similar pathway for polyamine biosynthesis has been identified by homology in higher plants [ 18 ]. In the alternative pathway corresponding to the analogous sets of genes shown on Fig. 3 , it is thus a PTCase which catalyzes the second step and catabolically converts N-carbamoylputrescine to putrescine and carbamoylphosphate which can further be used to synthesize ATP via carbamate kinase [ 19 ]. Furthermore, when we compare the clusters of genes shown in Fig. 3 to those surrounding gene argF or arcB , encoding the catabolic OTCase functioning in the arginine deiminase (ADI) pathway present in many microbial genomes (see [ 20 - 22 ]), we observe a very similar distribution, namely a transcriptional regulator, the arginine deiminase (EC 3.5.3.6), an arginine/ornithine antiporter and a (sometimes two) carbamate kinase. There is thus a very close analogy between the set of genes encoding the enzymes catalyzing the different steps of the agmatine deiminase pathway found in these different Firmicutes and that encoding the enzymes catalyzing the different steps of the arginine deiminase pathway. Conclusions Genome annotation requires both reliable tools for identifying gene function and manual expertise. The frustration due to the high percentage of orphan genes found in all genomes is often compounded with another – more vicious – problem which may occur when a very strong sequence similarity is obscuring the actual functional identity of another kind of orphan. The analysis described in this paper illustrates the difficulty in identifying such a potential source of misannotation and delineates at least two fundamental parameters which must be considered especially when the results appear to be straightforward. First, one must keep in mind that proteins sharing a high level of identical residues may have different functions. A routine step for challenging the functional annotation of any putative coding sequence should be a phylogenetic analysis. Any CDS found to branch far from its homologues in an evolutionary tree, as observed in the case of the carbamoyltransferases (Fig. 1 ), should be examined with caution before assigning it a putative function. Another example of the usefulness of the phylogenetic approach to correct misannotations can be found in a comparative analysis of ureohydrolases [ 20 ]. The second parameter which must be considered is the striking lack of information in the various public databases. For example, in the case studied here (the putrescine carbamoyltransferase EC 2.1.3.6) it is reported that there is no sequence available in various first-rate databases specialized in enzymatic and/or metabolic data such as ENZYME [ 23 ], BRENDA [ 24 ], KEGG [ 25 ], BIOCYC [ 26 ], etc...as well as in the Gene Ontology (GO:0050231) Consortium [ 27 ]. A significant part of this deficit of information appears to be due to not correlating biochemical data [ 8 - 11 ] previously published and well recorded in BRENDA [ 24 ], for example, with the incomplete amino acid sequence which was not taken into account although it had been published by the same group [ 13 ] who studied this enzyme. The specific point we would like to stress in this paper reflects a more general gap – which is widely ignored – between the enormous quantity of information buried in the sequence data and the refined knowledge built up over several decades of studies on gene regulation and protein biochemistry (recorded in [ 23 ] to [ 27 ]). In this respect, experimentally studied proteins not correlated with sequence data also qualify as "orphans". In the present case, such a resulting gap in knowledge could be bridged only because we used the experimental approach detailed in this paper. After being alerted by the unusual topology of the phylogenetic tree (Fig. 1 ) and the rediscovery of the partial sequence [ 13 ], a confirmation of the reannotation as PTCases was obtained when considering their signature (Fig. 2 ) and the gene context (Fig. 3 ) which differentiate them clearly from their OTCase homologues. The strategy we used to identify such orphan sequences could work in any other case where it is possible to collect a body of facts about experimental data, homology, unnoticed sequence data, and accurate informations about gene context. Note that we incidentally used such a strategy to annotate the genes encoding a putative agmatine deiminase in the genomes listed in Fig. 3 . It is highly probable that this approach can be applied to many similar cases. Therefore, our community should feel encouraged to dig in old lab books, unpublished data buried in doctoral thesis and similar documents, in order to retrieve information crucial for correct genome annotation. Moreover, it becomes urgent to design new approaches in order to efficiently explore what has been called the "bibliome" [ 28 ]. This could help to bridge important gaps in knowledge – such as exemplified in this paper- which lead to numerous errors in genome annotation. Accordingly, it would become possible to (re)annotate conserved hypothetical proteins for which there is an apparent lack of information in the various public databases. Methods Collecting sequences Near 450 carbamoyltransferases (ATCases and OTCases) sequences were collected from the public databases SwissProt, TREMBL and TREMBLNEW [ 15 ]. To facilitate the management of these data which are continuously growing up with the onset of new completely sequenced genomes, we assemble them in a relational database (available on request). Moreover, in the case of unpublished but completely sequenced genomes, it was often possible to recover bona fide sequences from specific sequencing groups sites (Joint Genome Institute [ 29 ] and Sanger Institute [ 30 ]) using either BlastP or tBlastN queries. We retained only unpublished sequences aligning along their whole length with bona fide carbamoyltransferases and sharing no less than 30% identity with it, using at least two distantly related seeds. Reconstructing phylogenetic trees Rooted phylogenetic trees were derived from multiple alignements of ATCases and OTCases using two different approaches. (1) New sequences were manually added and aligned to the previously published [ 4 ] multiple alignement using the BioEdit sequence alignment editor [ 31 ]. These additions were made effortless by introducing each new sequence near its closest partner (the first hit in a routine BlastP check). This processive approach minimized the risk of introducing any bias when adding numerous new sequences. However, the soundness of this manual alignement was routinely checked using automatic programmes (both ClustalX and DARWIN, see below) to verify that we did not miss any conserved motifs. We further ascertained this multiple alignement (especially the introduction of gaps) by using the informations available from the known 3D structures of ATCases and OTCases. Maximum parsimony and distance trees were derived from this alignment using the PROTPARS and NEIGHBOR programmes of the PHYLIP package [ 32 ], respectively. This PHYLIP package was further used to derive confidence limits for each node of either parsimony or distance trees using a bootstrap approach (programmes SEQBOOT and CONSENSE). (2) The PhyloTree programme of the DARWIN package [ 33 ] allows to build a multiple alignement and to derive a distance tree which is an approximation to maximum likelihood tree since the deduced evolutionary distances are weighted by computing their variance when reconstructing the tree. Authors' contributions NG dug up the "ancient" data on putrescine carbamoyltransferase, contributed his knowledge about carbamoyltransferases and made important additions to the manuscript. YX brought essential informations about the genetics and biochemistry of the enzymes involved in arginine metabolism and their evolution. DGN participated in the collection of new carbamoyltransferase sequences and their manual alignment and identified which sequence of E. faecalis is the putrescine carbamoyltransferase. BL carried out the phylogenetic analyses, the gene context study and drafted the manuscript which was further improved (and approved) by all authors.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC514541.xml
524496	Description and evaluation of an EBM curriculum using a block rotation	Background While previous authors have emphasized the importance of integrating and reinforcing evidence-based medicine (EBM) skills in residency, there are few published examples of such curricula. We designed an EBM curriculum to train family practice interns in essential EBM skills for information mastery using clinical questions generated by the family practice inpatient service. We sought to evaluate the impact of this curriculum on interns, residents, and faculty. Methods Interns (n = 13) were asked to self-assess their level of confidence in basic EBM skills before and after their 2-week EBM rotation. Residents (n = 21) and faculty (n = 12) were asked to assess how often the answers provided by the EBM intern to the inpatient service changed medical care. In addition, residents were asked to report how often they used their EBM skills and how often EBM concepts and tools were used in teaching by senior residents and faculty. Faculty were asked if the EBM curriculum had increased their use of EBM in practice and in teaching. Results Interns significantly increased their confidence over the course of the rotation. Residents and faculty felt that the answers provided by the EBM intern provided useful information and led to changes in patient care. Faculty reported incorporating EBM into their teaching (92%) and practice (75%). Residents reported applying the EBM skills they learned to patient care (86%) and that these skills were reinforced in the teaching they received outside of the rotation (81%). All residents and 11 of 12 faculty felt that the EBM curriculum had improved patient care. Conclusions To our knowledge, this is the first published EBM curriculum using an individual block rotation format. As such, it may provide an alternative model for teaching and incorporating EBM into a residency program.	Background Evidence-based medicine (EBM) strives to provide a systematic approach to integrating the best research evidence with clinician expertise and patient preferences to provide better patient care [ 1 ]. While the potential for answering clinical questions using online resources is high [ 2 ], a recent study found resident physicians rarely used web-based or other evidence-based sources to answer clinical questions, preferring instead another person or a pocket reference [ 3 ]. Many authors have argued for the importance of teaching EBM skills during residency training, and several have cited evidence to support the desirability of integrating EBM training with other aspects of the residency program [ 4 - 8 ]. Such a curriculum presents several challenges, including finding sufficient time to teach EBM skills to interns and developing ways to integrate and reinforce EBM among residents and faculty. While there have been several studies of residency EBM curricula [ 4 , 9 - 11 ], none, to our knowledge, has operated in the framework of an intern block rotation. In this paper we describe an EBM curriculum based on an individual block rotation and designed to integrate and reinforce EBM skills throughout the residency program; we also report on its evaluation by interns, residents and faculty. Methods Setting The UCSF family practice residency program is based at San Francisco General Hospital (SFGH), a large county hospital serving the urban poor. The program runs a busy family practice inpatient service (at SFGH) as well as the Family Health Center, an outpatient clinic that includes both continuity practices and acute care services. Within this setting, we formulated 3 primary goals for our EBM curriculum: (1) to teach interns basic EBM concepts and skills; (2) to disseminate and reinforce EBM skills to second and third year residents and faculty; and (3) to apply EBM to the care of patients. This study was approved by the UCSF Committee on Human Research. Curriculum The EBM curriculum for UCSF family practice residents began, in its current form, summer of 2001. The core of the curriculum is a 2-week, individual EBM rotation for interns. In contrast to the usual format of multiple lectures or learning modules scattered throughout residency, the two-week individual block format provides for a concentrated time in which to learn EBM. It also allows tailoring of the rotation to fit the residents' backgrounds and interests. During each of the two weeks, residents have 3 half-day clinics, with the remainder of the time available for EBM. The EBM portion of the rotation fulfills the ACGME requirements for resident research and scholarly activity. The rotation begins with a meeting with the EBM faculty Director (DT) during which the structure and goals of the rotation are communicated, the intern's knowledge and experience related to EBM are assessed by review of his or her experience and by a pre-rotation test of EBM skills and knowledge (described below). Interns receive reference materials [ 1 , 12 ] and a binder including detailed instructions for the rotation and key articles. The intern, together with the rotation faculty director (DT) and the medical librarian co-director (JH), attends family practice inpatient service rounds to obtain one or two clinical questions directly bearing on the care of one or two patients. With faculty guidance, questions are then formatted into the standard EBM structure identified by the acronym PICO (for p atient/ p roblem, i ntervention, c omparison, and o utcome). The entire process generally takes 5 to 10 minutes and includes modeling of how to formulate an appropriate 'answerable' question by senior residents and the inpatient attending faculty. The question(s) generated are then used by the librarian Co-director as material for a tutorial on developing search strategies and using high-quality web-based EBM resources. The emphasis of the tutorial is to introduce the intern to the concept of information mastery [ 5 ]. An initial assessment of the intern's searching experiences and use of EBM resources helps to tailor the tutorial session. In the tutorial, the intern is introduced to essential EBM concepts, including clinical question development, levels of evidence search strategies, and appraisal techniques [ 13 ], and learns to translate the 'answerable' clinical question into a 'searchable' one. Based on the type of clinical question, the intern reviews the possible levels of evidence and study designs and formulates a valid search strategy. As the intern searches for the evidence, the medical librarian provides input into alternative search strategies and information resources. The results of this and any subsequent searches are discussed with the faculty EBM Director and a 1-page response is developed in the form of a critically appraised topic or CAT [ 1 ]. Each CAT is structured to include the 'clinical bottom line' answer to the question; the clinical scenario that generated the questions; a summary of the evidence; a critical appraisal of the evidence; and citations. Results are presented to and discussed with the inpatient team. CATs are stored at the EBM website for future reference by residents and faculty [ 14 ]. During the first week, the intern completes a web-based EBM tutorial [ 15 ] which covers critical evaluation of articles about diagnosis, therapy, prognosis, meta-analysis and decision making, and how to ask clinical questions. In the second week the intern receives another 1 or 2 clinical questions from the inpatient service which are searched and answered as above. The intern also prepares and presents a journal club, which is attended by faculty, residents and medical students. A research article is selected by the intern (with guidance from the EBM Director), which could potentially change a primary care practice around a clinical question. The article is critically reviewed using an EBM approach [ 1 , 12 ], presented and discussed by the group. During the initial 3 months of the curriculum, interns completed a written test of their EBM skills and knowledge before and immediately following the rotation. The tests were adapted, with permission, from a similar instrument developed by Sean Schafer, MD and Katie Ramos, PhD at the University of California Fresno Family Practice Residency Program [ 16 ]. Scores improved post-rotation in all 3 areas tested: EBM terms and concepts 81% to 97%; quantitative skills 51% to 80%; question formulation and searching 71% to 92%, with the total score increasing from 63% to 87%. As others have noted, progress in the use of EBM depends on the availability of information support services to resident and faculty at the point of patient care [ 17 ]. In conjunction with our EBM rotation we have also improved support for residents and faculty searching for evidence-based answers to clinical questions, through the development of our EBM-filtered information support website [ 15 ], access to a librarian information specialist (JH), and provision of PDAs to our interns who did not have their own. We estimate that training 13 interns per year requires approximately 70 hours of librarian time and 200 hours of the faculty Director's time. In the event that the medical librarian Co-Director (JH) is not available, the faculty Director (DT) provides coverage for this portion of the rotation. Absence of the faculty Director for one or two days can usually be covered by schedule modifications and communication by telephone and e-mail. An extended absence of the faculty Director requires another faculty member assuming supervisory responsibility. Evaluation We evaluated our EBM curriculum with respect to our 3 primary goals using pre- and post-rotation questionnaires (completed by each EBM intern) and by a survey of family practice residents and faculty. The EBM intern questionnaires consisted of pre- and post-rotation self-assessments of confidence in EBM knowledge and skills (Goal 1). Self-confidence in EBM knowledge and skills was assessed by asking the intern to rate his or her level of comfort from 1 = very uncomfortable to 5 = very comfortable for (1) MEDLINE searching to answer a clinical question; (2) use of Web-based EBM resources to answer a clinical question; and (3) use of EBM principles to critically evaluate articles. At the end of the rotation interns were asked to identify the most useful and least useful aspects of the rotation and what could make the rotation better. In addition, residents completed and returned to the Residency Coordinator a standard evaluation of the rotation. We evaluated the dissemination and reinforcement of EBM skills and knowledge to residents and faculty (Goal 2) and the application of EBM to patient care (Goal 3) by surveying residents (all of whom had previously completed the EBM rotation as interns) and faculty. These surveys were distributed and collected by a program assistant. Surveys were labeled with a code number for each resident and faculty and results were reported in aggregate to provide anonymity. The resident survey asked "How frequently have you continued to apply EBM concepts and tools from the rotation to answer clinical questions?" and "How frequently have EBM concepts and tools been reinforced via teaching by faculty or senior residents?" For both questions, response options were 1 = never, 2 = seldom (less than once per month on average); 3 = occasionally (1 to 3 times per month on average); 4 = often (1 or 2 times a week on average); and 5 = frequently (3 or more times per week on average). For faculty and residents who had had a clinical question answered by the EBM intern on the inpatient service were asked. "How often did the EBM answer provide useful information?" and "How often did the answer change your management of a patient?" For both questions, response options were 1 = less than 25% of the time; 2 = 25% to 75% of the time; and 3 = more than 75% of the time. Finally, all faculty and residents were asked if they believed that "the presence of the EBM rotation improves the quality of patient care within the family practice residency program." Results During the period from July 1, 2001 to September 30, 2003, 30 EBM interns presented 30 journal clubs and generated 74 CATs in response to questions from the inpatient service. Journal club presentations and CATS are electronically archived and make accessible via the residency website [ 14 ]. Interns rated their experience during the rotation a median of 7 (superior) on a scale from 1 to 9. Written comments were also elicited and showed that one-on-one meetings with the EBM faculty (DH and JH) and exposure to web-based EBM resources were consistently valued. The only areas suggested for improvement were (1) to drop written material that repeated content of website tutorial (which we did) and (2) to have back-up EMB questions in the event that an appropriate question cannot be generated by the inpatient service (we have used this option only rarely). As shown in Table 1 , residents' confidence in their level of EBM knowledge and skills significantly increased from prior to the rotation in all 3 areas assessed (Goal 1). Table 1 Comparison of interns' confidence* in EBM skills pre- and post-rotation (n = 10) Mean Pre-Rotation Score Mean Post-Rotation Score Difference P-value Use of PubMed/Medline 4.11 4.78 +0.67 <.01 Use of other Web-based EBM resources 2.74 4.67 +1.93 <.001 Use of EBM tools and principles (critical appraisal) 2.58 4.44 +1.86 <.001 Mean confidence score 3.14 4.63 +1.39 <.001 * Assessed from 1 = very uncomfortable to 5 = very comfortable Surveys to evaluate goals 2 and 3 were completed by 21 of 25 residents (response rate of 84%) and 12 of 13 faculty (response rate or 92%). As seen in Table 2 , 86% of residents reported that they have continued to apply EBM concepts and tools learned in the rotation to clinical questions at least occasionally and 81% reported having had EBM concepts and tools reinforced occasionally or often by faculty or senior residents. Eleven of the 12 faculty (92%) agreed that the EBM curriculum had increased their use of EBM concepts and tools in teaching and 75% felt that the curriculum had increased their use of EBM concepts and tools in their own clinical practice. Table 2 Residents' report of how often they applied EBM concepts and tools to patient care, and how often EBM concepts and tools are reinforced by senior residents and faculty (n = 21) Item Never Seldom (< 1/month) Occasionally (1 to 3 times/month) Often (1 to 2 times/week) Frequently (≥3 times/week) N (%) N (%) N (%) N (%) N (%) Applied to patient care 0 (0) 3 (14) 7 (33) 8 (38) 3 (14) Reinforced by senior residents and faculty 0 (0) 4 (19) 12 (57) 5 (24) 0 (0) As shown in Table 3 , most residents and inpatient attending faculty felt that the EBM answer to the inpatient clinical question provided useful information 25% to 75% of the time. Only 1 resident (and none of the faculty) reported the EBM answer provided useful information less than 25% of the time. The majority of the residents (65%) and faculty (83%) reported that the information led to a change in patient management 25% to 75% of the time, with the remainder reporting a change in management less than 25% of the time. All residents and 11 of the 12 faculty (92%) agreed that the EBM rotation had improved the quality of patient care within the residency program. Table 3 Percent of time that the answer to EBM question provided useful information or led to a change in patient management on the Family Practice Inpatient Service Item Residents (n = 20) Faculty (n = 6) <25% 25% to 75% >75% <25% 25% to 75% >75% N (%) N (%) N (%) N (%) N (%) N (%) Provided useful information 1 (5) 12 (60) 7 (35) 0 (0) 3 (50) 3 (50) Led to a change management of the patient 7 (35) 13 (65) 0 (0) 1 (17) 5 (83) 0 (0) One Resident, who joined in the second year, had not been on the inpatient service during a time when an EBM intern was present. ** Answered by the 6 faculty members who attend on the inpatient service. Discussion While the above findings support the acceptance and perceived utility of our EBM program, they do not provide a formal measure of its effectiveness. Changes in resident knowledge, skills and confident were measured over a short period of time. The frequency of reinforcement of EBM and the impact of EBM on clinical care was by resident and faculty report and may have been biased. No attempt was made to observe changes in physician behaviors or patient outcomes. There were no measures of EBM use prior to the introduction of the EBM rotation and no comparison group was available. It is therefore not possible to objectively determine to what degree current levels of awareness and utilization of EBM are the result of the rotation. It is also not possible to separate out the effects of the 2-week EBM rotation from the adjunct changes of establishing a medical information website or promoting the use of PDAs, except to the extent that the questions asked specifically about the 2-week EBM rotation. Our block EBM rotation differs substantially from the approaches reported in previous studies [ 9 , 11 , 16 ] in that it utilizes a concentrated, individual experience that combines formal learning (via tutorials and a web-based course) with immediate application of EBM to answer an important clinical questions for individual patients – the target goal for EBM training. An advantage of our approach is the ability to tailor the curriculum to the background and needs of each intern and to provide interns with dedicated time during which they can rapidly acquire EBM knowledge and skills and apply them to "real time" clinical questions under the supervision of a faculty member and a librarian information specialist. We believe that our curriculum could be implemented by any faculty member with a working knowledge of EBM. The website tutorial [ 15 ] provides the core didactic portion for the rotation. Moreover, directing the rotation naturally increases the experience and expertise of the faculty member involved. Interns reported greater confidence in their search skills after the two week rotation. Two randomized controlled trials and a controlled before-after study have demonstrated benefits of training in electronic search skills [ 17 - 19 ]. Interns also reported significantly greater confidence in their ability to apply EBM knowledge and principals in critical appraisal. This is consistent with previous studies that have found training improves critical appraisal skills for residents and practicing physicians [ 20 ]. One study that evaluated the impact of a 1-month pilot program to use EBM methods on an inpatient service, reported results roughly similar to ours [ 21 ]. Our curriculum also includes the archiving of EBM answers in a standardized format for future reference by residents and faculty. While our study did not include an evaluation of the usefulness of the archived EBM answers generated by the rotation, archiving EBM answers for web-based access has been shown to provide a useful resource for resident physicians in an internal medicine residency program [ 22 ]. Previous studies have found only small increases in residents' knowledge and skills from journal clubs alone, leading to the suggestion that journal clubs should be used as a component of EBM training rather then being 'stand alone' activities [ 6 - 8 ]. While we did not attempt to evaluate the effects of the journal club per se, including it as a component of our EBM curriculum is concordant with this suggestion. Other investigators have reported educational interventions aimed at improving faculty knowledge and skills in medical informatics and EBM. One study reported increases in faculty self-rating of EBM skills following an intervention consisting of 2 half-day workshops and substantial amount of individual mentoring [ 23 ]. We found that faculty reported our EBM rotation has increased their use of EBM in their clinical practice, as well as their teaching of EBM; this was echoed by residents. While it is difficult to compare the two studies, our findings suggest that integrating EBM into the residency via resident training may improve faculty application of EBM to clinical care and in their teaching, and may be a cost-effective way to reinforce faculty EBM skills. Our EBM curriculum, based on an individual 2-week EBM rotation in the first year, appears to be successful in increasing resident's EBM skills and confidence. In addition, resident and faculty both perceive EBM as being incorporated and reinforced beyond the rotation, and that the presence of EBM in the residency improves the quality of patient care. We hope that our experience provides a useful model for teaching and integrating EBM into a busy, resource-limited, family practice residency. Competing interests The authors declare that they have no competing interests. Authors' contributions All authors participated in the development of the curriculum. DT conceived of and conducted the evaluation. JH and PSS reviewed the survey instrument. JH and PSS reviewed and suggested changes to the paper. JH drafted the subsection describing the teaching the use of electronic. All authors have read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524496.xml
514555	The National Women's Health Study: assembly and description of a population-based reproductive cohort	Background Miscarriage is a common event but is remarkably difficult to measure in epidemiological studies. Few large-scale population-based studies have been conducted in the UK. Methods This was a population-based two-stage postal survey of reproductive histories of adult women living in the United Kingdom in 2001, sampled from the electronic electoral roll. In Stage 1 a short "screening" questionnaire was sent to over 60,000 randomly selected women in order to identify those aged 55 and under who had ever been pregnant or ever attempted to achieve a pregnancy, from whom a brief reproductive history was requested. Stage 2 involved a more lengthy questionnaire requesting detailed information on every pregnancy (and fertility problems), and questions relating to socio-demographic, behavioural and other factors for the most recent pregnancy in order to examine risk factors for miscarriage. Data on stillbirth, multiple birth and maternal age are compared to national data in order to assess response bias. Results The response rate was 49% for Stage 1 and 73% for the more targeted Stage 2. A total of 26,050 questionnaires were returned in Stage 1. Of the 17,748 women who were eligible on the grounds of age, 27% reported that they had never been pregnant and had never attempted to conceive a child. The remaining 13,035 women reported a total of 30,661 pregnancies. Comparison of key reproductive indicators (stillbirth and multiple birth rates and maternal age at first birth) with national statistics showed that the data look remarkably similar to the general population. Conclusions This study has enabled the assembly of a large population-based dataset of women's reproductive histories which appears unbiased compared to the general UK population and which will enable investigation of hard-to-measure outcomes such as miscarriage and infertility.	Background Despite improvements in obstetric care in the UK over the past fifty years, it is estimated that around one in five pregnancies will end in miscarriage (fetal death before 24 weeks) [ 1 , 2 ]. The personal and public health impact of pregnancy loss is a neglected area in medical research and strategies of prevention remain outside mainstream medical services. Although many large-scale population-based studies of miscarriage risk have been conducted elsewhere [ 3 - 10 ], relatively few such studies have been conducted in the UK, and most of these have been occupational [ 11 - 14 ]. There are no registers of miscarriage or routine data collection systems which would allow linkage of miscarriages to individual women in the UK . There are thus no national prevalence estimates which can be used as reference for UK-based clinical or epidemiological studies. In addition, although there is now greater knowledge of how the risk of miscarriage changes with maternal age and previous history of miscarriage [ 6 ], the influence and interaction of biological, behavioural and social risk factors are less well-understood. The lack of reliable information on risk factors, and the confusion surrounding ad hoc reports of spurious associations, makes research in this area of great importance. Studies of miscarriage have tended to be clinical-based, and are thus subject to selection bias. For example, gestations are later among miscarriages reaching hospital-based clinics. Many miscarriages are managed at home, and some are not reported to a clinician. Not only is miscarriage hard to measure, and different clinical sources rarely see the full range of cases, but reported risks of miscarriage tend to be pregnancy-rather than woman-based: estimates of risk tend to relate to the proportion of pregnancies ending in miscarriage, and there are very few studies examining the risk of experiencing one, two or more miscarriages, or the chances of conceiving following a miscarriage [ 15 ]. Large prospective cohort studies are theoretically the ideal design, but take time and are prohibitively expensive [ 2 ]. An alternative and practical approach is a survey asking the women themselves for their full reproductive history, including all fetal losses at all gestations. An increasing number of couples are also seeking help for problems achieving a pregnancy. Although it is estimated that up to 15% couples experience such problems [ 16 ], few population-based prevalence studies have been conducted in the UK, particularly where fertility problems have been treated solely by the general practitioner using ovarian stimulation. We now report on a large UK population-based survey of reproductive health, the National Women's Health Study. The study design was developed from several other large epidemiological surveys of reproductive outcome which showed that a postal method could be used to obtain full reproductive histories from large study populations [ 13 , 14 , 17 , 18 ]. The aim of the study was to obtain population-based prevalence estimates relating to miscarriage and infertility, and to obtain good quality data on potential risk factors for miscarriage to be used when advising and counselling women who have suffered miscarriage and those who wish to reduce their risk of future pregnancy loss. The design of the study, together with response rates and description of the study population, is presented in this report. Further reports on risk factors for miscarriage, plus population-based estimates of miscarriage and of pregnancies conceived using assisted reproduction techniques will follow. Methods Sample selection This was a population-based cross-sectional postal survey of reproductive histories of adult women living in the United Kingdom in 2001, designed to enable the construction of a retrospective population-based reproductive cohort and a case-control study of risk factors for miscarriage. A sample of women was randomly selected from electronic electoral registers for England, Wales, Scotland and Northern Ireland held by the company Eurodirect [ 19 ]. All UK citizens aged 18 and over are eligible to vote; registration is voluntary, but in 2001 around 98% of the entire resident population were on the electoral register [ 20 ], the remainder being largely non-UK citizens and iterant population. At the time of survey there was no opt-out clause for those who did not wish to be on an electronic version of the electoral register, so the sampling frame contained all UK residents eligible (and registered) to vote. In order to reduce possible biases associated with memory, we aimed for a sample aged 55 years and under at survey. Date of birth is not, however, routinely recorded on the electoral register. To avoid unnecessary mailing and expense, we therefore made use of a probabilistic process offered by Eurodirect based on forename, whereby the sampling frame was restricted to women thought likely to be aged 55 and under on the basis of their name. This process was based on empirical data relating to birth certificates going back to the beginning of the 20th century, from which it could be calculated that, for example, those named "Elsie" are likely to be aged over 55, and those named "Kylie" under 55 years. Predictions are further refined by examination of combinations of names within a household (a "Jane" married to or living with an Alfred likely to be older than a "Jane" married to or living with a "Darren") and length of residency (e.g. someone registered to vote at the same address for 12 years has to be over 30). We requested a random sample of 61,000 women likely to be aged 55 and under (sample size calculations based on achieving at least 80% power for key risk factors in the case-control analysis, and cost). After removing those known to be under age 18 at study (those turning 18 in the year of registration are allowed to register early, giving date of birth), the final sample consisted of 60,814 women. The study received approval from the Trent Multi-Centre Research Ethics Committee and the Ethics Committee of the London School of Hygiene & Tropical Medicine. Postal survey The postal survey had two stages. Stage one consisted of a single-page "screening" questionnaire which asked for details of all pregnancies experienced by study participants, as well as periods of infertility and infertility treatment. This form was sent to the whole sample and included "opt-out" boxes to be ticked if the recipient had never been pregnant and had never attempted to have children, and/or was over age 55, and/or did not wish to take part. The second stage of the study consisted of a longer postal questionnaire which was sent to all those responding to Stage 1 who had ever been pregnant or who reported ever attempting to conceive and who agreed to be re-contacted. Excluded from this second stage were women who had had one or more termination for non-medical reasons (i.e. for reasons other than that a defect had been identified in the fetus or that continuing the pregnancy would put the mother at risk) and no other pregnancies. The Stage 2 questionnaire requested more general detail about the women (including height, age at menarche, educational level, marital status and details of infertility problems, treatment and diagnosis, if appropriate); detailed information on all pregnancies (including whether the pregnancy was the planned, the result of infertility treatment, father's date of birth and whether father had remained the same); plus socio-demographic and behavioural details relating to the most recent pregnancy. These details included questions relating to weight at start of pregnancy, nausea, smoking, coffee and alcohol consumption, diet, vitamin intake, ill health, air travel, sexual intercourse, occupation and stress levels. The most recent pregnancy was selected to minimise biases related to recall, and since it could be at the start, middle or end of the reproductive careers of these women whose ages at survey ranged from 18 to 55 years potential biases relating to ending reproductive careers on a "success" were not expected to be large. For those whose most recent pregnancy had ended in miscarriage (defined as fetal death at <24 weeks gestation), brief information relating to clinical management of miscarriage and the advice given was also requested. Permission to access clinical notes relating to outcomes reported in the questionnaire, and to contact the women for further study if needed, was also requested. In order to increase the number of cases for the case-control analysis of risk factors for miscarriage, women who had had a miscarriage recently (since 1995) but whose last pregnancy was not a miscarriage were sent a third questionnaire. This was a shortened version of the Stage 2 questionnaire, containing only those questions relating to biological, socio-demographic and behavioural details of the most recent pregnancy, but now requesting these details in relation to the most recent miscarriage. Such women then had two pregnancies in case-control analyses and standard errors were computed using a robust method based on the "sandwich estimate" to account for this statistically. A free telephone helpline was run throughout the study, to answer queries and refer on to other organizations for professional help, if appropriate, and this was well used. Statistical methods All analyses in this paper were performed using Stata statistical software [ 21 ]. To investigate possible selection bias we compared stillbirth and multiple delivery rates with rates in the general population. For this we obtained annual registered stillbirth risks and registered multiple delivery rates by maternal age for England and Wales, 1980–2001 [ 22 ] (data for 2002 was estimated from that for 2001). Standardised registered stillbirth ratios (SRSR) and standardised multiple delivery rates (SMDR) were then calculated using logistic regression analysis (offsetting the log odds of the population risk) [ 23 ]. The unit of analysis for stillbirths was a registered birth. A registered livebirth is defined as a baby born alive at any gestation, registered stillbirth being defined as a fetal death at 28 weeks or more gestation until the end of 1992, and at 24 weeks or more gestation from 1993 onwards. Where gestational age was not available from Stage 2 data, a pregnancy was considered to be a stillbirth if it was so described. Forty-one (40%) of the total 102 stillbirths in the analysis fell into this category. For multiple delivery, the unit of analysis was a pregnancy containing at least one livebirth or registered stillbirth (as described above). For the purposes of the analyses presented in this paper (comparisons with the general population), a pregnancy was only considered multiple if it contained two or more babies who were liveborn or (registered) stillborn in order to be consistent with the definitions used in the national data. Thus, for example, a twin pregnancy occurring before 1993 and resulting in a livebirth and a fetal death at less than 28 weeks was considered to be a singleton pregnancy in this analysis. Average maternal age at first birth, if live, was also compared with that in the general population. Annual average maternal age at first (registered) birth, if live, was obtained with denominators for England and Wales, 1980–2001 [ 22 ] and re-calculated for 5-year periods. This national data was available for births within marriage only. Marital status of mother at time of birth was known only for the most recent pregnancy (or most recent miscarriage since 1995) in this dataset. For the NWHS average maternal age was therefore calculated for all first registered births, if live. No formal statistical comparisons of maternal age were made, partly because the numbers were so large that slight, non-meaningful, nuances in the data would give a statististically significant result, and render the comparison meaningless, and partly because the average ages in the general population, though comparable, were expected to be similar but slightly older in the general population data owing to the fact that the data related to births within marriage only. Births where the date of birth or maternal age were not known were excluded from all comparisons with population data. Results Stage 1 The response to the first stage of the study is summarised in Table 1 . 29,721 (49%) of all the questionnaires were returned to us, though for 3,591 (6%) this was to say that the addressee had moved, and for 70 (0.1%) that the woman had died. A total of 26,050 questionnaires were returned by the addressee, a response rate of 46% assuming that all questionnaires not returned undelivered had reached the correct recipient. Of these, 11% (5% overall) did not wish to participate in the study, and a further 21% were aged over 55 (n = 5,499) or were otherwise ineligible (n = 65). 27% of the 17,748 women who were eligible on the grounds of age, reported that they had never been pregnant and had never attempted to conceive a child, the remaining 13,035 women reporting their full reproductive history. Table 1 The National Women's Health Survey – response rates STAGE 1 No. Crude % Adjusted 1 % TOTAL QUESTIONNAIRES POSTED 60,814 100% - Returned undelivered 2 3,661 6% - Responded 26,050 43% 46% Did not wish to participate 2,738 5% 5% Aged >55 years or otherwise ineligible 3 5,564 9% 10% Aged < = 55 years but never attempted to have children 4,713 8% 8% Aged < = 55, ever attempted to have children 13,035 21% 23% Among whom, - Never pregnant 340 3% - - Ever pregnant 12,695 97% - STAGE 2 TOTAL QUESTIONNAIRES POSTED 10,828 100% - Returned undelivered 16 0.2% - Responded 7,882 73% 73% No longer wished to participate 180 2% 2% Completed questionnaire 7,702 71% 71% Among whom, - Attempted pregnancy, never pregnant 194 3% - Ever pregnant 4 7,508 97% 1 Adjusted for undelivered mail 2 Includes 70 women who died before the study start 3 Under 18 at study start (6 th November 2001); male; foreign national; or too ill to participate 4 344 women who had had a miscarriage since 1995, but whose last pregnancy was not a miscarriage, were sent a second stage 2 questionnaire and were asked to supply details in relation to their most recent miscarriage. 285 (83%) of the women responded to this third questionnaire. 12,695 women aged under 55 at survey had been pregnant at least once. These 12,695 women, whose average age at survey was 40.5 years, had started their reproductive careers from 1963 to 2002, 75% having their first pregnancy in 1980 or later (Table 2 ). 486 women had conceived their first pregnancy less than 40 weeks before the study commenced, 126 of whom were pregnant when they filled in the questionnaire. Overall these 12,695 women reported a total of 30,661 pregnancies, 80% of which occurred in 1980 or later. Outcome of these pregnancies is described in Table 2 . Table 2 NWHS Stages 1 and 2 – description of women reporting one or more pregnancy, and of the pregnancies they reported STAGE 1 STAGE 2 n (%) n (%) TOTAL NO. WOMEN IN ANALYSIS 12,695 (100) 7508 (100) Age at survey (years) <30 1247 (9.8) 685 (10.6) 30–34 2007 (15.8) 1284 (20.6) 35–39 2618 (20.6) 1629 (28.6) > = 40 6678 (52.6) 3910 (39.3) Not known 145 (1.1) - Mean age (SD) 1 40.5 (8.45) 40.4 (8.24) Year of first pregnancy <1980 3201 (25.2) 1798 (24.0) 1980–84 1902 (15.0) 1131 (15.1) 1985–89 2091 (16.5) 1259 (16.8) 1990–94 2158 (17.0) 1356 (18.1) 1995–99 2079 (16.4) 1406 (18.7) 2000–02 788 2 (6.2) 558 3 (7.4) Not known 476 (3.8) - Total number of pregnancies reported per woman 1 2607 (20.5) 1403 (18.7) 2 5077 (40.0) 3162 (42.1) 3 2962 (23.3) 1749 (23.3) 4 1573 (12.4) 818 (10.9) 5 285 (2.2) 229 (3.1) > = 6 191 (1.5) 147 (1.9) Median (range) 2 (1 – 18) 2 (1 – 18) Pregnancy history No dates given for any pregnancies 436 (3.4) - All pregnancies occurred before 1980 1495 (11.8) 853 (11.4) Pregnancies before and after 1980 1707 (13.5) 945 (12.6) Pregnancy history commenced 1980 onwards 9057 (71.3) 5710 (76.1) All pregnancies conceived after 31/03/2000 486 (3.8) 329 (4.4) TOTAL REPORTED PREGNANCIES 30661 (100) 18391 (100) Outcome of pregnancy Livebirth, surviving >7 days 24081 (78.9) 14782 (80.4) Livebirth, early neonatal death 95 (0.3) 56 (0.3) Stillbirth 188 (0.6) 110 (0.6) Miscarriage 4 3512 (11.5) 2326 (12.7) Ectopic 226 (0.7) 102 (0.6) Termination for medical reasons 5 312 (1.0) 89 (0.5) Termination for non-medical reasons 6 1424 (4.6) 562 (3.1) Molar pregnancy 47 (0.2) 26 (0.1) Ongoing (current) pregnancy 482 (1.6) 338 (1.8) Not known 294 (1.0) - Year of pregnancy end <1980 6093 (19.9) 3486 (18.0) 1980–84 4503 (14.7) 2623 (14.3) 1985–89 5028 (16.4) 3000 (16.3) 1990–94 5549 (18.1) 3434 (18.7) 1995–99 5808 (18.9) 3865 (21.0) 2000–02 2721 7 (8.9) 1983 8 (10.8) Not known 959 (3.1) - 1 Where date of birth given 2 Includes 486 women whose first pregnancy was conceived after 31 st March 2000, 126 of whom were currently pregnant for the first time at time of survey 3 Includes 329 women whose first pregnancy was conceived after 31 st March 2000, 73 of whom were currently pregnant for the first time at time of survey 4 Fetal death at <24 weeks gestation. Includes missed miscarriages (fetal death at <24 weeks without spontaneous expulsion of fetus) and blighted ova (anembryonic pregnancy) 5 Termination of pregnancy because of a defect identified in the baby, or because continuing the pregnancy would put the mother's health at risk 6 Termination of pregnancy for reasons other than a defect identified in the baby or risk to mother's health 7 1,718 of these pregnancies were conceived after 31 st March 2000 8 1,232 of these pregnancies were conceived after 31 st March 2000 Stage 2 11,424 (88%) women ever attempting to have children (successfully or unsuccessfully) agreed to participate in the second stage of the study. Of these 596 (5%) were not sent a Stage 2 questionnaire, 212 because they had only ever had one or more termination of pregnancy for non-medical reasons, and 384 because their Stage 1 form arrived back after mailing had ended. A total of 10,828 women were thus sent a second stage questionnaire. The response to this second stage was high (73%), though 2% of women had decided that they no longer wished to participate (Table 1 ). The 7,702 women completing a Stage 2 questionnaire, and the 18,391 pregnancies they reported, are described in Table 2 . Their characteristics are almost identical to those of Stage 1, indicating that Stage 2 responders were an apparently unbiased subset of those responding to Stage 1. 5,777 (75%) women responding to Stage 2 gave signed consent for us to access their medical notes, with 6,963 (90%) agreeing to be contacted again in the future, if required. Comparison with national data Comparisons of Stage 1 data, and the subset Stage 2 data, with national rates are presented in Table 3 . There was no evidence to suggest that stillbirth differed from expectation in either Stage 1 (SRSR 115 (95% CI 94 – 139), P = 0.17), or Stage 2 data (SRSR 102 (95% 79 – 132), P = 0.86). Multiple delivery was also in line with expectation from national rates for both stages (Stage 1 SMDR 111 (95% CI 99 – 126), P = 0.08), Stage 2 SMDR 108(95% CI 93–126, P = 32)). Although the inference from this is unambiguous for both stages of the study, the point estimates were noted to be closer to unity for Stage 2 data where almost all pregnancies had known gestational age. This reflects the fact that there might be some slight misclassification of registered stillbirth prior to 1993 in the Stage 1 data where gestational age was only known for 61% of reported stillbirths, some of which might legally be classified as miscarriages. Table 3 Comparison with population birth data of reported births in Stages 1 and 2 1 of the National Women's Health Study occurring since 1980 2 REGISTERED STILLBIRTH 3 No. stillbirths 3 Total livebirths & stillbirths 3 SRSR 4 (95% CI) Stage 1 1980–2002 102 18,740 115 (94 – 139) Stage 2 1980–2002 59 12,061 102 (79 – 132) MULTIPLE (REGISTERED) DELIVERY 5 No. multiple deliveries 5 Total deliveries 5 SMDR 4 (95% CI) Stage 1 1980–2002 264 18,391 111 (99 – 126) Stage 2 1980–2002 169 11,887 108 (93 – 126) AVERAGE MATERNAL AGE AT FIRST 6 (LIVE)BIRTH (years) No. first 6 livebirths Mean (SD) age 7 England & Wales 8 Mean age Stage 1 Year of delivery 1980–84 1,724 25.2 (4.12) 25.5 1985–89 1,916 25.9 (4.56) 26.4 1990–94 2,058 27.1 (4.85) 27.8 1995–99 2,026 28.6 (5.01) 29.0 2000–02 699 29.4 (5.06) 29.6 Stage 2 1980–84 1,032 25.5 (4.02) 25.5 1985–89 1,182 26.0 (4.45) 26.4 1990–94 1,325 27.3 (4.78) 27.8 1995–99 1,432 28.8 (4.81) 29.0 2000–02 540 29.7 (4.89) 29.6 1 Stage 2 data are a subset of Stage 1 data (see methods). 2 Pregnancies with missing maternal age have been excluded from this analysis. 3 Registered stillbirths 1980–2002, defined as fetal death at ≥ 28 weeks prior to 1992, or at ≥24 weeks thereafter. 41 (40%) of stillbirths had no gestational age, but were described as stillbirths by the mother. Unit of analysis is a baby; multiple births counted as many times as there are babies. Denominator contains all reported livebirths and registered stillbirths 1980–2002. 4 Standardised Registered Stillbirth Ratio (SRSR) and Standardised registered Multiple Delivery Ratio (SMDR). Standardised for maternal age (5-year intervals) and single year of birth using data for England and Wales 1980–2002. 5 Unit of analysis is a delivery (pregnancy) containing one or more registered live or stillbirth; multiple pregnancies counted once only. Multiple pregnancies containing only one registered birth (with another non-registrable outcome, such as miscarriage) considered as singleton in this analysis. 6 First registered birth, if live. 7 NWHS data relates to livebirths both within and outside marriage 8 Livebirths within marriage only Age at first (live) birth was remarkably similar to national data for both Stage 1 and Stage 2 data (Table 3 ). Exactly as expected, though showing no evidence to suggest any biases with respect to maternal age, average age at first birth was very slightly higher for the national data, since it related to births within marriage only, whereas the NWHS data related to all births (marital status at delivery was unknown). Discussion Using a novel method, the National Women's Health Study has enabled a large UK population-based dataset to be assembled, comprising full reproductive histories, including any history of infertility, for 13,035 women, 12,695 of whom had conceived 30,661 pregnancies. We have obtained further detailed information for 7,702 of these women (18,391 pregnancies), including fertility diagnoses for both male and female partner (if appropriate), and lifestyle and behavioural risk factors for the most recent pregnancy. Seventy-five percent of these women consented to their medical notes being accessed in relation to information reported in the questionnaire, and 90% agreed to be contacted again, thus providing the means to carry out a population-based cohort study of these women at some time in the future. UK population-based data, collected at government level by England & Wales, Scotland and Northern Ireland, relate to registered births (live and still) and terminations of pregnancy, with Scotland also routinely collecting maternity data on hospital deliveries at any gestation. The National Women's Health Study goes one step further than this, providing the whole reproductive picture. Rather than being a pregnancy-based, cross-sectional survey, the data collected for each woman covers the complete spectrum of reproductive outcomes from infertility problems through miscarriage, ectopic pregnancies and terminations (for both medical and non-medical reasons), to live and stillbirths, and does not rely on legal definitions for inclusion in the dataset. Furthermore, unlike most epidemiological studies of adverse reproductive outcome such as miscarriage, the data source is not clinical (which, for miscarriage, leads to inevitable biases relating to gestational age), but relates to women selected randomly from the UK electoral register. And for outcomes such as infertility no other data currently exist to enable estimation of how many pregnancies in the population as a whole result from fertility treatment. The study does rely on maternal recall and this could be a source of bias. Studies of self-reported reproductive history and exposures relating to reproductive events have, however, found maternal recall to have acceptably high reliability, and to be little affected by time from event [ 24 - 26 ]. In terms of the key reproductive indicators of stillbirth, multiple delivery rates and maternal age at first birth, the data look remarkably similar to the general population. We therefore feel confident that response was unlikely to be related to adverse reproductive outcome. Indeed, the average age at survey of around 40 years, coupled with average ages at first birth which are exactly as would be expected from general population data, could be seen to indicate that non-responders to the survey tended to concentrate among younger women who had not yet tested their fertility. In addition, we feel confident that those responding to the more detailed Stage 2 questionnaire are an unbiased sample of those responding to Stage 1. Both Stage 1 and Stage 2 data can thus can be considered unbiased with respect to reproduction, and representative of patterns among all women in the UK population who have ever tried to have children, hence prevalence estimates might be taken as unbiased estimates of hard-to-measure outcomes such as miscarriage and pregnancies conceived through assisted reproduction techniques. Such data will be invaluable as population-based reference data for epidemiological studies of reproduction. In addition to both pregnancy-and woman-based population prevalence estimates, further papers to follow include reports of case-control analyses of behavioural and lifestyle risk factors for miscarriage. Conclusions In summary, we have assembled a large population-based dataset of women's reproductive histories which appears representative of the general UK population and which will enable investigation of hard-to-measure outcomes such as miscarriage and infertility. Competing interests None declared Authors' contributions NM and PD initiated the research and participated in protocol design, data collection, analysis and writing the paper. SP participated in data collection and analysis. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC514555.xml
545068	Design and validation of the Health Professionals' Attitudes Toward the Homeless Inventory (HPATHI)	Background Recent literature has called for humanistic care of patients and for medical schools to begin incorporating humanism into medical education. To assess the attitudes of health-care professionals toward homeless patients and to demonstrate how those attitudes might impact optimal care, we developed and validated a new survey instrument, the Health Professional Attitudes Toward the Homeless Inventory (HPATHI). An instrument that measures providers' attitudes toward the homeless could offer meaningful information for the design and implementation of educational activities that foster more compassionate homeless health care. Our intention was to describe the process of designing and validating the new instrument and to discuss the usefulness of the instrument for assessing the impact of educational experiences that involve working directly with the homeless on the attitudes, interest, and confidence of medical students and other health-care professionals. Methods The study consisted of three phases: identifying items for the instrument; pilot testing the initial instrument with a group of 72 third-year medical students; and modifying and administering the instrument in its revised form to 160 health-care professionals and third-year medical students. The instrument was analyzed for reliability and validity throughout the process. Results A 19-item version of the HPATHI had good internal consistency with a Cronbach's alpha of 0.88 and a test-retest reliability coefficient of 0.69. The HPATHI showed good concurrent validity, and respondents with more than one year of experience with homeless patients scored significantly higher than did those with less experience. Factor analysis yielded three subscales: Personal Advocacy, Social Advocacy, and Cynicism. Conclusions The HPATHI demonstrated strong reliability for the total scale and satisfactory test-retest reliability. Extreme group comparisons suggested that experience with the homeless rather than medical training itself could affect health-care professionals' attitudes toward the homeless. This could have implications for the evaluation of medical school curricula.	Background In 2003, the Department of Health and Human Services reported that there are between two to three million people in the United States who experience homelessness each year [ 1 ]. According to Gelberg and Arangua [ 2 ], such estimates provide only a partial picture of the problem: among the U.S. population, 14% (26 million people) have been homeless at some time in their lives and 5% (8.5 million people) have been homeless within the past five years. Yet even as this number grows, the homeless continue to be subjected to broad stereotyping and stigmatization, both of which make it easier to ignore them. It is not surprising then that homeless persons are reluctant to obtain needed, continuous medical care within traditional outpatient settings. This is particularly problematic because they often have competing or immediate needs [ 3 , 4 ], multiple health problems [ 5 , 6 ], and increased morbidity and mortality [ 7 , 8 ]. The disinclination of the homeless to seek care may be due in part to the way in which many health-care workers respond to them. A less investigated but possibly equally important circumstance is the attitudes that health-care professionals have toward the homeless. As members of the larger society, these professionals often harbor the same preconceived ideas and biases toward the homeless that the rest of society does. Such judgmental attitudes can, and often do, emerge during the provider-patient encounter, thus limiting the effectiveness of medical treatment of the homeless [ 9 ]. As a countermeasure to prevailing negative attitudes and widespread stigmatization toward marginalized persons, recent medical literature has drawn on a well-known practice in the social sciences [ 10 , 11 ] and has begun calling for greater emphasis on the dimensions of compassion and humanism in medical education [ 12 - 14 ]. Humanism in psychology became popular in the 1950s when Rogers [ 15 ] began practicing client-centered therapy, which allows the relationship between therapist and client to develop so that the client can be guided within the framework of the therapeutic encounter. According to Branch [[ 16 ], p. 1067], humanism in medicine may be defined as "the physicians' attitudes and actions that demonstrate interest in and respect for the patient, and address the patient's concerns and values." As physicians interact with homeless patients, a heightened sense of these patients' vulnerability may evoke greater empathy and humanism, two attitudes that can only have a positive impact on the quality of care that the physicians provide. Although it would be helpful to have a better understanding of physicians' current opinions about homeless patients, only two instruments have been designed and validated to measure attitudes toward homeless persons: the Attitudes Toward the Homeless Inventory (ATHI) [ 17 ], and the Attitudes Towards the Homeless Questionnaire (ATHQ) [ 18 ]. The ATHI surveys American college students' attitudes toward homelessness but does not specifically address the attitudes of health-care providers. In addition, it was not designed to measure a health-care provider's desire or confidence in his or her ability to deliver health care to homeless persons. Nevertheless, Buchanan et al. [ 19 ] recently used the ATHI to measure the attitudes of 12 primary care residents before and after a 2-week rotation in homeless health care and found that they felt more comfortable affiliating with homeless people after the course. The ATHQ, which was designed and validated in the United Kingdom, does assess the attitudes of health-care providers toward the homeless, but differences in health systems terminology (e.g., NHS for National Health Service, and terms related to homelessness, such as sleep rough) would make its use in its original form problematic in the United States. For this reason, we decided to use the ATHI for comparison purposes because its language is more consistent with American English. The purpose of our study was to develop and validate a new measure, the Health Professionals' Attitude Toward the Homeless Inventory (HPATHI), an instrument that could be used to assess United States medical students' and physicians' attitudes toward homeless persons and to measure their level of interest and confidence in their ability to deliver health-care services to the homeless population. The objectives of this study were to: 1) describe the process of designing and validating the new instrument, and 2) discuss the usefulness of the instrument for assessing the impact of educational experiences working with the homeless on the attitudes, interest, and confidence of medical students and other health-care professionals. Methods The development and validation of the HPATHI occurred in three phases. The first phase consisted of the identification of items for the inventory; the second phase involved a pilot test of the initial instrument with a group of medical students; and the final phase entailed modifying the instrument and administering it in its revised form to our target population – health-care professionals, including primary care physicians, primary care residents, and third-year medical students. Throughout this three-phase process, we sought to ascertain the reliability and validity of the HPATHI. Three types of validity were tested. Content validity was evaluated using the Delphi technique [ 20 ], which seeks consensus on instrument items among a panel of experts. Concurrent validity was computed by correlating the target population's responses to the HPATHI with their responses to the ATHI. Construct validity was established by using the extreme groups method [ 21 ], by conducting an exploratory factor analysis, and by conducting item analyses to assess the relationship of each individual item to the instrument's overall scale and hypothetical subscales. Stages of development and validation of the HPATHI In Phase 1, a group of physicians and nurse practitioners, all experts in homeless health care, was recruited through snowball sampling [ 22 ] to serve on a Delphi panel [ 20 ]. These individuals were identified as experts because they work with the homeless on a regular basis and are members of the National Health Care for the Homeless Clinicians' Network. They received a list of statements about the homeless, homelessness, and homeless health care that was compiled by selecting and adapting appropriate items from the AHTI [ 17 ] and the ATHQ [ 18 ]. We, as the authors of this study, added additional items that were based on our experiences and our review of the literature. The experts were asked to classify the statements into one of three categories ("essential," "interesting but not essential," and "irrelevant") and to generate other items that they considered to be necessary to explore health-care professionals' attitudes toward the homeless. The panel reached a consensus on the items to be used in the instrument through successive rankings of the items. The statements were then combined into an instrument with a five-point Likert scale (1 = strongly disagree; 2 = disagree; 3 = neither agree nor disagree; 4 = agree; and 5 = strongly agree) organized around three thematic areas: a) attitudes toward the homeless; b) interest in working with the homeless; and c) confidence in one's ability to work with the homeless. During Phase 2, we administered the first draft of the HPATHI to a convenience sample of third-year medical students enrolled at Baylor College of Medicine (BCM) in Houston, Texas. These students were selected because they showed an interest in working with the homeless by registering for an elective course that allows students to provide health care to the homeless in a clinic setting [ 23 ]. Of the 100 students who attended class on the day that the instrument was administered, 72 completed it and were asked to provide contact information for the retest. Student responses to the instrument were analyzed using Cronbach's alpha coefficient to establish its internal consistency. Two weeks later, 34 of the 63 students who provided e-mail addresses completed the instrument a second time, thereby providing the data to determine the HPATHI's test-retest reliability. We then conducted an item analysis of redundant items or those with poor item-to-scale correlations. In Phase 3, we prepared Web-based versions of the HPATHI and of the ATHI and sent the Internet link to our target population with a standard invitation to complete both instruments. Primary care physicians who serve as faculty in the BCM Department of Family and Community Medicine, family practice residents in the BCM Department of Family and Community Medicine, general internal medicine residents in the BCM Department of Family and Community Medicine, and students from the BCM Medical School served as a convenience sample of individuals completing the instrument over a six-month period. Subsequent data analysis consisted of: 1. Conducting an exploratory factor analysis using a Promax rotation to examine the structure of the HPATHI. 2. Estimating the internal consistency reliability of the online version of the HPATHI using Cronbach's alpha coefficient; 3. Determining the concurrent validity of the HPATHI against the ATHI; 4. Comparing extreme groups on their responses to the HPATHI. The extreme groups were determined according to their level of training (preclinical medical students vs. physicians) and according to their experience in working with the homeless (<1 month vs. >1 year). This study was approved by the BCM Institutional Review Board for educational research with human subjects. Data analysis was conducted using SPSS Version 12.0 statistical software for Phases 1 and 2 and SAS ® (V8.2 and V9.1) for Phase 3. Results Phase 1: instrument development Of the 23 panel members who received the first list of 24 statements, 16 (70%) reviewed them, rank ordered them, and generated an additional 26 statements. We rearranged the combined list of 50 statements according to the proposed ranking ("essential," "interesting but not essential," and "irrelevant") and returned them to the 16 panel members for a second ranking. Based on their responses, six statements were eliminated and the remaining 44 were returned to the 16 panel members for a third ranking. Nine panel members returned responses after this iteration, and we eliminated nine additional items according to their recommendations. The remaining 35 items then constituted the first draft of the instrument we named the Healthcare Professionals' Attitude Toward the Homeless Inventory (HPATHI). Although seven panel members did not participate in the third iteration, their rankings in the second iteration corresponded to the final list of items. Phase 2: pilot administration of the 35-item HPATHI The sample population of third-year medical students completed a pilot administration of the 35-item HPATHI; the subset of students who responded two weeks later underwent a second administration of the same instrument. Table 1 displays the means and standard deviations of student responses to both administrations of the HPATHI. Items 2, 5, 6, 11, 15, 16, 20, and 23 were reverse-coded for the analysis so that a higher total mean on the instrument would indicate a positive attitude toward the homeless. Table 1 Means and standard deviations for HPATHI* test and retest Statement Test n = 76 Retest n = 34 M SD M SD 1. Homeless people are victims of circumstance. 3.26 .839 3.24 .819 2. Most homeless people are mentally ill. 2.90 .875 3.24 .955 3. Homeless people have the right to basic health care. 4.60 .620 4.68 .475 4. Homelessness is a major problem in our society. 4.36 .698 4.50 .707 5. Homeless people choose to be homeless. 3.71 .777 3.91 .712 6. Homeless people are lazy. 3.82 .738 3.94 .694 7. Health care dollars should be directed toward serving the poor and homeless. 3.83 .888 4.03 .717 8. Doctors should address the physical and social problems of the homeless. 4.17 .839 4.47 .563 9. Doctors have a duty to care for the homeless. 3.94 .977 4.18 .834 10. Caring for the homeless is pointless since they do not follow-up. 4.14 .612 4.24 .654 11. Providing medical care for the homeless is futile. 4.04 .740 4.21 .592 12. I am comfortable being a primary care provider for a homeless person with a major mental illness. 3.03 1.14 3.44 1.05 13. I feel comfortable being part of a team when providing care to the homeless. 4.32 .535 4.15 .643 14. I feel comfortable providing care to different minority and cultural groups. 4.42 .710 4.18 .673 15. I feel overwhelmed by the complexity of the problems that homeless people have. 2.97 .839 3.09 .996 16. I understand that my patients' priorities may be more important than following my medical recommendations. 3.97 .769 4.24 .741 17. I entered medicine because I want to help those in need. 4.42 .687 4.62 .551 18. I am interested in working with the underserved. 3.96 .941 3.88 .946 19. I enjoy addressing psychosocial issues with patients. 3.69 .973 3.74 1.08 20. I resent the amount of time it takes to see homeless patients. 3.79 .604 3.94 .547 21. I enjoy learning about the lives of my homeless patients. 3.58 .921 3.82 .834 22. I believe social justice is an important part of health care. 3.75 1.06 3.74 1.14 23. I believe caring for the homeless is not financially viable for my career. 3.24 .831 3.26 .864 24. I am too pressed for time to investigate psychosocial issues routinely. 3.37 .941 3.56 .894 25. I feel overwhelmed by the number of problems that homeless people have. 2.79 .844 2.88 .844 26. My knowledge regarding the problem of homelessness is adequate. 2.58 .931 2.65 .884 27. I can provide care for the homeless effectively. 2.97 .878 3.00 .921 28. Homeless people come from all walks of life. 4.42 .622 4.65 .485 29. Most homeless people tend to be drug addicts or alcoholics. 3.10 .735 3.38 .697 30. I think mentally ill homeless people refuse to get treatment. 3.67 .692 3.82 .797 31. Homeless people are dangerous, aggressive, and physically threatening. 3.97 .556 3.97 .627 32. There are only a few children among the homeless population. 4.31 .493 4.32 .684 33. All people have a right to basic health care. 4.50 .805 4.59 .657 34. I feel it is important to provide care to all socio-economic groups. 4.54 .670 4.62 .604 35. Most poor people have adequate access to health care through the public system. 2.15 1.02 2.03 .937 Totals 114 6.93 116 7.75 * Statements 24–35 were excluded from HPATHI after analysis. Cronbach's alpha coefficient for the first administration was 0.87; the test-retest reliability coefficient (Pearson r) was 0.69. Through an item analysis we discarded 12 items that were either highly correlated with other items, and were thus considered repetitious, or that had item-scale correlations less than 0.20. Phase 3: administration of the HPATHI to the target population One hundred and sixty health-care professionals (24 primary care physicians, 15 primary care residents, 47 clinical medical students, 71 preclinical medical students, and 3 medical students who did not specify their education level) from one academic institution completed the HPATHI; 147 of them also completed the ATHI. Table 2 displays the means and standard deviations for both instruments by gender, by level of training (primary care physician; primary care resident; clinical medical student; preclinical medical student), and by experience with the homeless (no experience; <1 month, >1 month but <1 year, and >1 year). Table 2 Comparison of the HPATHI and ATHI by gender, level of training, and experience HPATHI ATHI N Mean SD N Mean SD Total 160 3.90 0.34 147 3.36 0.28 Gender Female 103 3.95 0.30 96 3.38 0.26 Male 57 3.81 0.38 51 3.32 0.30 Level of Training† MS1 55 3.89 0.32 54 3.34 0.26 MS2 16 4.10 0.23 16 3.47 0.21 MS3 28 3.91 0.41 26 3.38 0.36 MS4 19 3.84 0.30 18 3.33 0.25 Resident 15 3.73 0.22 13 3.30 0.25 Faculty 24 3.93 0.36 18 3.38 0.30 Experience None 31 3.72 0.33 28 3.21 0.23 <1 month 64 3.91 0.30 62 3.37 0.27 >1 month & <1 year 30 3.88 0.32 25 3.37 0.22 1 – 3 years 17 4.10 0.28 16 3.49 0.31 >3 years 18 4.05 0.38 16 3.46 0.33 † On the HPATHI, 3 respondents failed to indicate a level of training and on the ATHI, 2 respondents failed to indicate a level of training. The exploratory factor analysis (principal components), using a Promax rotation to account for the relationship among the factors, yielded a three-factor structure that explained 39% of the variance of the data. Factor 1 consisted of items 12, 13, 14, 17, 18, 19, 20, 21, 22, and 23 and was labeled Personal Advocacy; factor 2 consisted of items 1, 3, 4, 7, 8, 9, and 15 and was labeled Social Advocacy; and factor 3 consisted of items 5, 6, 10, and 11 and was labeled Cynicism. Table 3 presents the items with their loadings in each factor. Table 3 Factor loadings for the 23-item HPATHI Factor 1 Factor 2 Factor 3 1. Homeless people are victims of circumstance. -0.095 0.518 0.064 3. Homeless people have the right to basic health care. -0.096 0.644 0.154 4. Homelessness is a major problem in our society. -0.124 0.677 0.140 5. Homeless people choose to be homeless. 0.008 0.233 0.469 6. Homeless people are lazy. -0.026 0.330 0.559 7. Health-care dollars should be directed toward serving the poor and homeless. 0.180 0.575 0.079 8. I am comfortable being a primary care provider for a homeless person with a major mental illness. 0.329 0.466 -0.071 9. I feel comfortable being part of a team when providing care to the homeless. -0.015 0.514 0.188 10. I feel comfortable providing care to different minority and cultural groups. -0.007 0.152 0.725 11. I feel overwhelmed by the complexity of the problems that homeless people have. 0.028 0.056 0.748 12. I understand that my patients' priorities may be more important than following my medical recommendations. 0.469 -0.206 0.202 13. Doctors should address the physical and social problems of the homeless. 0.438 -0.046 0.395 17. I entered medicine because I want to help those in need. 0.485 0.003 0.088 18. I am interested in working with the underserved. 0.516 0.168 0.085 19. I enjoy addressing psychosocial issues with patients. 0.697 0.106 -0.224 20. I resent the amount of time it takes to see homeless patients. 0.613 -0.214 0.097 21. I enjoy learning about the lives of my homeless patients. 0.788 -0.039 -0.188 22. I believe social justice is an important part of health care. 0.509 0.404 -0.154 23. I believe caring for the homeless is not financially viable for my career. 0.504 -0.096 0.042 The HPATHI was further shortened to 19 items by the deletion of four more items, which either were not represented in the three-factor structure (items 2, 14, and 16) or had an adverse effect on the subscale's reliability (item 15). The three subscales were also significantly related to each other: factor 1 had Pearson's r correlations of 0.47 (n = 160; p < 0.001) with factor 2 and 0.43 (n = 160; p < 0.001) with factor 3; and factor 2 had a Pearson's r correlation of 0.48 (n = 160; p < 0.001) with factor 3. Table 4 displays the descriptive statistics and measurement properties for the 19-item HPATHI total and subscales. These three factors, if taken as subscales for the HPATHI, showed satisfactory Cronbach's alpha coefficients: 0.75, 0.72, 0.72, and 0.84 respectively for factor 1 (mean = 3.86; sd = 0.47), factor 2 (mean = 4.06; sd = 0.46), factor 3 (mean = 4.06; sd = 0.50), and total scale (mean = 3.96; sd = 0.38). Table 4 Measurement properties for reduced HPATHI scale and subscales Descriptive Statistics Subscale Statistics Full-scale Statistics Item Mean SD Scale ¶ Item-Total Correlation Cronbach's Alpha if Deleted Item-Total Correlation Cronbach's Alpha if Deleted 1 2.55 0.76 SA 0.28 0.73 0.27 0.84 3 1.45 0.61 SA 0.52 0.66 0.40 0.83 4 1.68 0.71 SA 0.48 0.67 0.39 0.83 5 2.20 0.80 C 0.42 0.73 0.40 0.83 6 2.14 0.65 C 0.53 0.65 0.50 0.83 7 2.15 0.79 SA 0.48 0.67 0.55 0.82 8 1.81 0.67 SA 0.49 0.67 0.50 0.82 9 2.01 0.72 SA 0.47 0.67 0.39 0.83 10 1.76 0.60 C 0.59 0.62 0.47 0.83 11 1.68 0.64 C 0.54 0.65 0.44 0.83 12 2.82 0.96 PA 0.31 0.75 0.31 0.84 13 1.79 0.58 PA 0.43 0.73 0.47 0.83 17 1.63 0.58 PA 0.38 0.73 0.37 0.83 18 1.84 0.78 PA 0.51 0.71 0.52 0.82 19 2.23 0.97 PA 0.51 0.71 0.47 0.83 20 2.09 0.74 PA 0.40 0.73 0.36 0.83 21 2.19 0.78 PA 0.55 0.71 0.46 0.83 22 2.00 0.85 PA 0.49 0.71 0.54 0.82 23 2.66 0.94 PA 0.34 0.74 0.31 0.84 ¶ PA = Personal Advocacy; SA = Social Advocacy; C = Cynicism The Pearson's correlation coefficient between the HPATHI and the ATHI was 0.68 for the HPATHI's total scale (concurrent validity) (Table 4 ). For the extreme group comparisons, no significant difference was found between preclinical medical students and primary care physicians in their responses to the HPATHI (F = 1.05; df = 3, 156; p = 0.371). On the other hand, respondents who had more than one year of experience with the homeless scored significantly higher than those who had less than one month of experience (F = 6.19; df = 2, 157; p = 0.003) (Table 5 ). When the individual hypothetical subscales were considered, all items were either moderately or strongly correlated with their respective subscales (range of Pearson's correlation coefficients, 0.38 to 0.68). However, when the entire instrument was considered, the item analysis showed that items 1, 2, 15, and 16 had low item-scale correlations (Pearson's r < 0.24). Table 5 Extreme group comparisons by level of training and experience with the homeless N Mean SD Level of Training MS1 & MS2 71 3.94 0.31 Residents & Faculty 39 3.85 0.33 Experience with the Homeless‡ Less than 1 month 95 3.85 0.32 More than 1 year 35 4.01 0.33 ‡ Differences between the means of the two groups by experience were statistically significant (p < 0.01) Discussion Using the Delphi method to select the survey items helped ensure the HPATHI's content validity. Not only were the items selected and validated by homeless health-care experts, but the items chosen for inclusion in the HPATHI were based on findings from current literature on the issue. In addition, by correlating the HPATHI with the ATHI, we were able to demonstrate satisfactory concurrent validity. Construct validity for the HPATHI (i.e., attitudes toward the homeless) was determined by the extreme group comparisons, the item analyses, and the factor analysis. Results of the extreme group comparisons showed that experience with the homeless, rather than medical training, is a significant factor that correlates with health-care professionals' attitudes towards the homeless and their interest in working with the homeless population. Moreover, individuals who had more extensive experience with the homeless showed more positive attitudes toward and interest in homeless patients. Therefore, increasing the opportunities to provide direct patient care to the homeless might improve the attitudes of health-care professionals toward this group. The factor analysis suggested that we do have potential subscales in the instrument that may offer meaningful information regarding the general attitudes of health-care professionals who work with the homeless. The three subscales appear to represent: personal advocacy , which contains the items that reflect a personal commitment to work with the homeless; social advocacy , which consists of the items that reflect society's responsibility to care for the homeless population; and cynicism , which encompasses the items that reflect a negative attitude and a sense of futility in working with the homeless. A limitation of this study is that only 160 health-care professionals participated in the online administration of the HPATHI. This small sample size limits the effectiveness of the factor analysis. Moreover, our sample focused only on the medical profession as represented by medical students and primary care residents and physicians. The participation of a larger number of health-care professionals from other medical specialties would strengthen the study. Conclusions The development of the HPATHI has many implications for and applications to future research on homeless health care, although its most valuable use may be to assess the attitudes toward homeless persons of medical students, residents, and practicing physicians. Throughout its various iterations, the instrument demonstrated strong internal consistency reliability for the total scale and satisfactory test-retest reliability. The scales identified by the factor analysis also showed satisfactory internal consistency reliability. The information collected from the expert panel and the literature search on attitudes toward the homeless and their health-care status proved to be an excellent framework for determining which statements to retain for the final instrument. The inter-item correlations and the correlations among the subscales indicate that the items are measuring similar underlying constructs within an overall theme – the attitudes of health-care professionals toward the homeless. Despite the limited sampling, the research process has demonstrated that the HPATHI is a reliable and valid instrument that has the ability to assess the attitudes of health-care professionals toward the homeless population. We believe that the instrument may also be used in the future within the academic framework of medical schools to determine if attitudinal changes are affected by training experiences occurring with the homeless. To determine whether this is the case, we intend to administer the HPATHI as a pre/post-test survey to students who enroll in the homeless health-care track. Moreover, over the next year, we plan to keep on assessing the attitudes of health-care professionals toward the homeless by having new groups respond simultaneously to the ATHI and the HPATHI, as was done in Phase 3 of the original study. Additionally, we intend to include participants from other medical schools in the United States and to expand our sample to other health-care professionals who traditionally work with the homeless, to further test the instrument's overall validity. Competing interests The author(s) declare that they have no competing interests. Authors' contributions DB contributed to the conception, design, and acquisition of data, and drafted and revised the article. FMM contributed to the conception, design, and acquisition of data, helped draft the article, and carried out the data analysis. SK contributed to the conception, design, and acquisition of data, helped draft the article, and carried out the data analysis. DR contributed to the draft of the manuscript and critically revised it for important content. DLC contributed to the conception, design, and acquisition of data. AM contributed to the conception, design, and acquisition of data. RJV contributed to the interpretation of the data and critically revised it for important content. All authors read and approved the final manuscript Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545068.xml
526281	Effectiveness of behavioural graded activity compared with physiotherapy treatment in chronic neck pain: design of a randomised clinical trial [ISRCTN88733332]	Background Chronic neck pain is a common complaint in the Netherlands with a point prevalence of 14.3%. Patients with chronic neck pain are often referred to a physiotherapist and, although many treatments are available, it remains unclear which type of treatment is to be preferred. The objective of this article is to present the design of a randomised clinical trial, Ephysion, which examines the clinical and cost effectiveness of behavioural graded activity compared with a physiotherapy treatment for patients with chronic non-specific neck pain. Methods Eligible patients with non-specific neck pain persisting longer than 3 months will be randomly allocated to either the behavioural graded activity programme or to the physiotherapy treatment. The graded activity programme is based on an operant approach, which uses a time-contingent method to increase the patient's activity level. This treatment is compared with physiotherapy treatment using a pain-contingent method. Primary treatment outcome is the patient's global perceived effect concerning recovery from the complaint. Global perceived effect on daily functioning is also explored as primary outcome to establish the impact of treatment on daily activity. Direct and indirect costs will also be assessed. Secondary outcomes include the patient's main complaints, pain intensity, medical consumption, functional status, quality of life, and psychological variables. Recruitment of patients will take place up to the end of the year 2004 and follow-up measurement will continue until end 2005.	Background Prevalence and incidence Neck pain is a common complaint that causes substantial morbidity in western countries with a reported prevalence ranging from 9.5 to 22% [ 1 , 2 ]. Of all musculoskeletal pains in the Netherlands, neck pain is one of the three most reported with a point prevalence of 21%; it is more often reported by women than men [ 3 ]. In 1996 total related costs were estimated to be US $686.2 million, which is about 1% of the total Dutch health care expenditures [ 4 ]. Most neck complaints are continuous or recurrent [ 3 ]. When the neck pain persists for more than 3 months it is defined as chronic, and the related prevalence is 14.3% [ 3 , 5 ]. Although the prevalence of neck pain is stable over different age groups, the incidence of chronic neck pain increases with age [ 3 , 6 ]. There are many potential causes of neck pain, but mostly no specific underlying pathology is found so that it is designated as non-specific [ 7 ]. Although not a life- threatening disease, neck pain can negatively affect patients' quality of life, cause pain and stiffness, and may result in substantial medical consumption, absenteeism and disability [ 4 , 8 ]. In the Netherlands, patients with neck pain are often referred for physiotherapy. Moreover, physiotherapy accounted for 84% of the total direct medical neck pain costs in 1996 [ 4 ]. Although physiotherapists can apply various treatments, no formal guidelines are yet available. Treatment models Two treatment models have been described in the literature, both of which are applicable within the field of physiotherapy. The first, a biomedical model, considers pain to be a sign of physiological damages and treatment according to this model aims to remove the pathologic condition so that the pain will no longer occur [ 9 , 10 ]. Moreover, treatment is guided by the amount of pain a patient experiences, leading to a pain-contingent approach [ 11 ]. According to the second, a biopsychosocial model, pain is not necessarily caused by underlying pathology or impairment but can persist long after the initial pathology has healed; psychological and social factors may be important in the development and maintenance of complaints [ 12 , 13 ]. According to the principles of this biopsychosocial model, behavioural therapies assume that maladaptive behaviours are learned and, therefore, can be modified through new learning experiences [ 10 , 14 ]. Three different approaches are known: respondent, operant and, cognitive behavioural therapy [ 9 , 15 , 16 ]. The present study mainly employs an operant behavioural approach, as described by Fordyce and applied by Lindström et al [ 11 , 17 ]. According to this approach, the treatment focuses on decreasing pain behaviour (operants) and increasing healthy behaviour, and consists of behavioural graded activity on a time-contingent basis [ 11 , 18 ]. Available evidence Many conservative physiotherapeutic treatments are available for treating neck pain, but there is insufficient evidence to allow to conclude that one type of treatment is more effective then others [ 19 , 20 ]. In a review on chronic pain, operant behavioural therapy was found to be beneficial to waiting list control groups on outcomes such as pain experience, mood effect other than depression, social role, and for the expression of pain behaviour [ 21 ]. Compared to other treatments, operant behavioural therapy is only beneficial for the expression of pain behaviour and role functioning [ 21 ]. Another review showed little evidence that biopsychosocial multidisciplinary rehabilitation is more effective than other rehabilitation methods for neck and shoulder pain, but the authors found only two relevant studies that satisfied the criteria for their review [ 22 ]. When examining the effectiveness of behavioural treatment for chronic pain another difficulty is that no standard protocol exists for the application of these treatments. As a result, a wide range of techniques described in the literature has been labelled as behavioural [ 23 ]. In summary, it remains unclear which type of conservative, including behavioural, treatment is to be preferred in the management of chronic neck pain. Therefore, this study, Ephysion (Effectiveness physiotherapy in neck pain), aims to evaluate the clinical and cost effectiveness of an operant behavioural programme (i.e. behavioural graded activity) compared with a physiotherapy treatment in patients with chronic non-specific neck pain. In addition, we aim to identify subgroups of patients who benefit most from one of the two treatments, and to identify the most important determinants for recovery from chronic non-specific neck pain. Why a design article Because a biased study design can produce incorrect conclusions, the design of a trial should be carefully examined before adopting its conclusions [ 24 ]. A design article allows to examine the design objectively without being influenced by the study results, to check any resulting articles for protocol deviations, and may also reduce the temptation to search for associations during data analysis rather then presenting hypotheses in advance [ 25 ]. Further, a published protocol informs others about which studies are in process thus reducing duplication of research effort [ 25 ]. Finally, a design article prevents publication bias in the case that future articles are not published, because study results can be retrieved from the author and the study can therefore still be included in future reviews [ 25 , 26 ]. Methods Study design A randomised clinical trial (RCT) has been designed to assess the effectiveness of behavioural graded activity compared with physiotherapy treatment in patients with chronic non-specific neck pain. The study design has been approved by the Medical Ethics Technical Commission of the Erasmus MC, University Medical Centre in Rotterdam and is in compliance with the Helsinki Declaration. Selection of patients and informed consent Forty general practitioners (GP) in region West Brabant in the Netherlands will select the patients. Patients are eligible if they are aged between 18 and 70 years old, have suffered from neck pain for over three months, and have an adequate knowledge of the Dutch language. Excluded are patients diagnosed with a specific disorder (e.g. a slipped disc, a tumour or a lesion in the cervical spine), those who have had physical/manual therapy during the previous six months, those with a chronic disease (e.g. rheumatoid arthritis or coronary artery disease), or those who have to undergo surgery in the near future. Eligible patients will receive an information leaflet from their GP and the GP then informs the research department. Thereafter, the research assistant contacts the patient, provides additional information about the implications of participation, re-checks the eligibility of the patient, and completes the informed consent procedure. Sample size The sample size for this study is calculated according to the global perceived effect (GPE). Based on previous studies, a 20% difference in GPE is expected after completion of either treatment (9 weeks) and is considered to be clinically relevant; 160 patients are needed to detect this difference. In this calculation a power (1 - β) of 80% is taken into account. Thus, the inclusion of 80 patients per treatment group is planned. Randomisation An independent examiner using a computer-generated randomisation schema performs randomisation. To prevent unequal distribution, patients are pre-stratified based on three important prognostic factors: gender, age and the severity of the complaint, which are recorded at baseline [ 27 ]. Further, unequal group sizes are prevented by using a 6-block randomisation that equalizes allocation to the two treatment groups per stratum after every sixth patient [ 28 ]. After randomisation, patients choose a physiotherapist within the allocated treatment group. Then, to ensure that the treatment starts as soon as possible, the research assistant makes the first appointment for treatment. Blinding Patients are told to receive physiotherapy but are blinded to allocation of the two treatments; the content of the treatments is not described in the information leaflet. This enhances the quality of the study, because the patients themselves measure the effect of treatment. GPs are also blinded for allocation to prevent accidentally informing the patients of the allocated treatment. The physiotherapists are not blinded for allocation, but the physiotherapists from each treatment group are kept strictly separate and are not involved in the outcome measurement. Finally, the primary investigator is blinded for patients' allocation but the research assistant is not; neither is involved in the outcome measurement. Physiotherapists and Interventions After receiving written information, 34 physiotherapists in region West Brabant will participate in either the physiotherapy treatment (PT) or the graded activity programme (GAP). To optimise the contrast between the two treatments, both groups are strictly separated throughout the study. The PT group consists of 16 physiotherapists and the GAP group of 18 physiotherapists. The PT physiotherapists participate in a meeting to standardize the physiotherapy treatment. The GAP physiotherapists are instructed on the behavioural graded activity approach during a two-day theoretical and practical training course. Both interventions are performed in an outpatient setting. A maximum of 18 treatments per patient is set and each treatment takes about 30 minutes, which is in accordance with medical insurance policy in the Netherlands. Before treatment starts, physiotherapists receive a completed questionnaire about the patient's main complaints [ 29 ]; this questionnaire reveals the three daily activities which are considered the most important complaints to the patient. Physiotherapists can use these three activities in the process of formulating the patient's primary therapy aim. In both treatments, the physiotherapist starts with a physical examination of the patient and an anamnesis. Then an individually tailored program will be applied and the process recorded after each treatment session using a specially designed form. The physiotherapy treatment The content of the physiotherapy treatment is decided by consensus among the participating PT physiotherapists. Treatment is according to a biomedical model, which implies guidance based on the amount and severity of pain that the patient experiences. By consensus, the physiotherapy treatment is divided into the patient's primary therapy aim, three general treatment goals, and several techniques to attain those goals. The primary therapy aim is defined as the result the patient wants to achieve by the end of therapy. A general treatment goal is a goal for each single treatment and could, therefore, differ per treatment session. Table 1 shows the three general treatment goals, together with the techniques physiotherapists can choose to attain them. In daily practice a broad spectrum of treatment techniques are available, but in this study the techniques to be used consist of physiotherapy techniques with a strong focus on exercises. Moreover, manipulative techniques, acupuncture and other (alternative) techniques are excluded, as are physiotherapeutic applications such as ultrasound or diathermy. Behavioural graded activity An operant approach was the basis of the behavioural graded activity programme as used in this study. The treatment is according to a biopsychosocial model, which implies that it is guided by the patients' functional abilities and that time-contingent methods are used to increase the activity level of the patient [ 11 ]. The behavioural graded activity programme has three phases; a baseline phase, a treatment phase, and a generalization phase. These phases are not bound to strict time limits but can gradually merge into each other. Before starting the baseline phase, the treatment vision and the patient's ideas about pain and its causes are discussed. The development and maintenance of pain will be explained and patients are reassured that it is safe to move and to increase their level of activity [ 11 , 13 , 30 ]. Both are explained by means of a pain model, which has been derived from the fear-avoiding-model of Vlaeyen et al. [ 13 ]. Thereafter primary therapy aims are formulated based on the patient's main complaints, which are described as three daily activities and were revealed in the baseline questionnaire. For each of these activities, a baseline level of intensity is determined based on a pain-contingent measure. This means that patients perform each activity at least three times, each time until they have to stop because of their pain. Afterwards, patient and physiotherapist together set a start quota and time-contingent treatment quotas for each activity. The quotas will be based on the patient's mean baseline scores, primary therapy aims [ 17 ], and on the behaviour that can be derived from the baseline measure. If necessary, facilitating disorder-oriented exercises can be added to the treatment as preparation for the activities that were pointed out as main complaints. The same approach as used for the main complaint is used for these exercises. During the treatment phase, patients systematically increase the time-contingent quotas to enable them to reach their personal aims within a pre-set therapy time period. To ensure a successful experience during the first exercise, the start quota is below the mean baseline score. The pre-set exercise quotas have to be strictly followed; neither over-performance nor under-performance is allowed. During this phase the patient has to practice at home and document every activity or exercise on a performance chart. These charts will be discussed in the following treatment session and achievements will be reinforced while disregarding pain behaviours. Positive reinforcements of healthy behaviour and the patient's experiences of success are considered to be important to enhance the patient's motivations. The generalization phase takes place at the end of the treatment phase. In this phase generalization of learned behaviour and management of relapses will be discussed. Outcome measurement Baseline questionnaires are sent after inclusion, which is as soon as possible after patients have consulted their GP. Outcome of intervention will be assessed at 4 and 9 weeks after randomisation; however, if the treatment is not finished at 9 weeks, the patients will receive an additional questionnaire (Ts) after finishing the treatment. Follow-up assessments are planned at 26 and 52 weeks after randomisation. All outcome measures are reported by means of mailed questionnaires. Table 2 presents the outcome variables, the instruments used and the moments at which they are measured. Primary treatment outcome of this study is the global perceived effect, which is used to assess recovery from the complaint [ 31 ]. In addition, the global perceived effect in daily functioning was explored in order to also establish impact of treatment on daily activity. Both treatment outcomes (recovery of complaint and functioning in daily activity), are assessed on a 7-point Likert-scale, ranging from completely recovered (1) to worse than ever (7). Costs are measured using a combination of questionnaires to collect data on direct medical costs (e.g. the amount of received treatment and additional therapy received), and indirect costs due to sick leave and disability. Secondary outcome measures include main complaints, pain intensity, medical consumption, coping, functional status, quality of life, and psychological variables. Prognostic factors are measured including demographic variables, the baseline variables and the psychological variables (table 2 ). Analyses Descriptive statistics will be used to examine comparability of baseline data between PT and GAP, and to check if randomisation was successful. Before this analysis, decisions about differences considered to be clinically relevant are made and, if necessary, adjustment will be made for these differences in multivariate analysis. Further, all outcome data will be screened for normality and, if necessary, logarithmic transformations or non-parametric methods of analysis will be applied. The first aim is to evaluate the clinical and cost effectiveness of GAP compared to PT. Clinical effectiveness will be examined with a Student's t-test (continuous), a Chi-square test (dichotomised) or a Wilcoxon test (not normally distributed) according to the intention-to-treat principle. This means that patients will be analysed in the treatment group to which they are randomly allocated. For missing data, imputation techniques will be used. When the dropout rate is 10% or more, or loss to follow-up is 20% or more, per-protocol analysis will be performed. The results on primary outcome will be dichotomised into improved versus not improved. Improved implies completely recovered and much improved, whereas not recovered implies slightly improved, not changed, slightly worsened, much worsened, and worse than ever [ 31 ]. Cost effectiveness will be calculated from a societal perspective. Costs (direct as well as indirect) will be related to the treatment effects, based on the primary outcome measure, by calculating cost-effectiveness ratios. The second aim is to identify subgroups of patients that benefit most from one of the two treatments. The following subgroups will be investigated: duration and severity of the complaint, depression, and fear of movement. The third aim is to identify important variables for recovery. For this purpose multivariate analysis will be performed to investigate the influence of prognostic variables and patient characteristics on the outcome. Separate analyses will be conducted to investigate prognostic factors for short-term (3 months) and long-term (12 months) recovery. Discussion This study is designed to evaluate the clinical and cost effectiveness of a behavioural graded activity programme compared with a physiotherapy treatment in patients with chronic non-specific neck pain. Since physiotherapists perform both treatments in this study, contrast between the two treatments is a very important issue. There are contrasts both in the composition of the treatment and the way the physiotherapists approach the patient. With regard to the composition, the graded activity programme (GAP) starts with a systematically performed baseline measurement; this is in contrast to the physiotherapy treatment (PT), where treatment is based on history taking and physical examination. In GAP quotas are set based on the patient's behaviour, whereas in PT they are set based on pain levels and training principles. After quotas are set GAP uses a time-contingent treatment approach, which involves a pre-set systematic increase in activities. In contrast, PT uses a pain-contingent approach, which means that treatment is adapted to the patient's reaction to previous treatment sessions. Furthermore, GAP uses a hands-off approach, whereas PT may contain hands-on techniques, such as massage, traction etc (Table 1 ). This study addresses an important question because chronic neck pain is a common complaint and it remains unclear which type of physiotherapeutic treatment is most effective. Recruitment of patients will take place until up to the end of 2004; follow-up measurement will continue up to end 2005. Competing interests The authors declare that they have no competing interests. Authors' contributions APV and BWK conceived the study, developed the design of the randomised clinical trial and participated in writing the article. MG is an expert in the field of graded activity and contributed to the content of the article. CJV advised on the content of the article. FV conducts the research, participated in the completion of the study design and wrote the article. All authors have read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC526281.xml
535570	Francis Crick's Legacy for Neuroscience: Between the α and the Ω	The legacy of Francis Crick is explored by two scientists who were influenced by his work	‘You’, your joys and your sorrows, your memories and your ambitions, your sense of personal identity and free will, are in fact no more than the behavior of a vast assembly of nerve cells…” — Crick (1994, p. 3) Francis Crick was an evangelical atheist. He believed that scientific understanding removed the need for religious explanations of natural phenomena. From James Watson's and his early work, the structure of DNA explained the α, the origins of life. This was a starting point; from the elucidation of the structure of DNA, there was an explosion, a massive diversity of science that in part removed the need to postulate a creator or a creation myth. Francis still felt that life was no less astonishing just because it was biological and natural in origin. He had a consistent and completely rational world view without a need to invoke vitalism, or any non-material force ( M. Crick 2004 ). And in the last decades of his life, he applied this philosophy to the Ω, consciousness. Once the structure of DNA was known, the physicist George Gamow formed the RNA Tie Club, with Francis and eighteen others including his close friends Leslie Orgel and Sydney Brenner (2001) ; it was an ingathering that sowed seeds for future molecular biologists ( Judson 1996 ). DNA had become the “α,” the beginning ( Bronowski 1978 ), not just of Francis's career, but of a whole new culture of scientific life and understanding ( Crick 1966 ). Ten years later, the secrets of DNA transcription and translation unmasked, Francis turned to consciousness. He admitted he knew little at first, only that the structure of consciousness was as tough a problem as DNA's structure. DNA was certainly not played out, but the Ferrier Lectures in the Proceedings of the Royal Society of London by David Hubel and Torsten Wiesel were just available, tempting Francis with an almost physicist's view of neurons in action. Hubel and Wiesel wrote of functional architectures, embedded in beautiful, almost crystalline structure. The comprehension of mind invoked by a biological mechanism appeared ripe for the sort of thoughtful, theoretical science he had applied to DNA. Francis was now sixty years old and moved from Cambridge to the Salk Institute in La Jolla, California. Francis began with the brightest young minds he could find. David Marr was a young mathematician and physiologist whose doctoral thesis on a theory of mammalian brain function at Cambridge had brought him into some contact with Brenner and Francis. A professor at the Massachusetts Institute of Technology, he began working with Tomasio Poggio of the Max Plank Institute in Tübingen on a computational theory of neuroscience. Following an invitation from Francis, Poggio and Marr spent the month of April, 1979 extending their intense examination of the core problems of visual perception. They spent hours sitting at the most western end of the Salk Institute, at the cafeteria or in Francis's office, gazing into the Pacific Ocean with all its daily changes, discussing not only architecture of visual cortex and visual perception, but the ramifications of a good theory of brain function. We know of these conversations, as the probing of Marr by Francis is captured in the final chapter of Marr's now classic book “Vision” ( Marr 1982 ). (Although Marr speaks of a three-way conversation, judging from our own experiences as Francis's younger colleagues, the interlocutor simply seems to be Francis.) Marr had been diagnosed with acute leukemia in the winter of 1978 ( Marr and Vaina 1991 ). The one-month visit to the Salk Institute was an intellectual gift, for eighteen months later, Marr died. Francis had simultaneously lost a young friend and colleague who had brought an “incisive mind and creative energy” ( Crick 1994 , p. 77) and his best new ideas of a theoretical neurology to the brain ( Marr 1969 , 1970 ). And he saw the tragedy of Marr being cut off from solving the big problems for which he was so clearly destined. During those early years, Francis must have thought that consciousness was tractable—if only the right way of thinking was brought to bear on it. Francis's brain was capable of collecting and filing away many disparate data, which he could then combine uniquely and imaginatively, leading to that “dramatic moment of sudden enlightenment that floods the minds when the right idea clicks into place” ( Crick 1990 , p. 141). Whatever his initial thoughts about the nature of the problem, Francis soon came to realize that the problem of consciousness was even tougher than he imagined, that the “click” was not happening with consciousness. In 1988, he wrote, “I have yet to produce any theory that is both novel and also explains many disconnected facts in a convincing way” ( Crick 1990 , p. 162). Over the quarter century he was at the Salk Institute, Francis did propose solutions to some smaller problems in neuroscience ( Sejnowski 2004 ) and brought consciousness into the scientific fold ( Rich and Stevens 2004 ). But something else was going on quietly and behind the scenes. Francis was building an army to help him take on consciousness. This was not empire building with Francis as the head of a group of directed scientists in the Cambridge or German model. Francis continually encouraged and assisted young scientists to approach the hardest problems of the brain. Marr and Poggio were just the first recruits he helped embolden. He started his long-time collaboration with Christof Koch, once a post-doctoral trainee with Poggio, on “The Problem of Consciousness” ( Crick and Koch 1990 , 1992 ). His door was always open to graduate students, postdoctoral trainees, faculty who wanted to discuss those problems as many others and we can attest. Francis could be found daily at tea time, an ingathering of the Salk Institute computational and vision laboratories of Simon LeVay, Terry Sejnowski and Thomas Albright, surrounded by graduate students and post-doctoral trainees, with conversation ranging across science—Francis listening to their stories of their explorations and encouraging them to reach beyond their horizons. Francis had a “love of the truth and helped others to move to the truth” ( Watson 2004 ). Francis Crick in his office. Behind him is a model of the human brain that he inherited from Jacob Bronowski. (Photo: Marc Lieberman) When Francis worked on the structure of DNA, he had some simple facts, such as Chargaff's Laws, and means to make point mutations from which it could be determined how function followed structure. But not a single neuroanatomist knew how many neurons actually converged in their input to a particular single cell. No one knew how to eliminate a specific cell type from a circuit— to make a point mutation, so to speak, in the structure of consciousness. His 1979 article in Scientific American, “Thinking about the Brain,” did not have much impact at the time, even when it explicitly described three needed methods: first, a method by which all the connections to a single neuron could be stained; second, a method by which “all neurons of just one type could be inactivated, leaving the others more or less unaltered”; and third, a means to differentially stain each cortical area, “…so that we could see exactly how many there are, how big each one is and exactly how it is connected to other areas.” By the mid-1980s, Francis had realized that these massive holes in our understanding of the most simple brain facts were not being filled. Something needed to be done. Over the twenty years since the RNA Tie Club, molecular biology had matured. Francis actively began encouraging the inclusion of the critical tools of molecular biology in the study of neural circuits and perception; in his thinking, molecular biology was critical to understand how the brain worked because it provided tools. He would encourage junior scientists, postdoctoral trainees, and faculty—all those who had visited him over the years—to think about using these tools. He would give short homilies about the plethora of sub-types of neurons in the retina; would not the cortex be at least as rich in possibilities? Molecular tools could unravel this knot. As we reminisced after Francis's death, we discovered that Francis had spoken with each of us on these molecular methods, across a twenty-year interval. In the mid-1980s, Francis spoke with Ralph, pressing him to consider how he might do highly specific lesions of single neuron types in motion cortex using molecular identifiers. At the time, the only tools imaginable were some sort of killer antibody approach. Twenty years later, Ed recalls Francis continuing to encourage this cross-disciplinary molecular and systems approach. It was absolutely imperative to Francis's vision of the maturation of neuroscience that there would be a conjoining of molecular biology and systems neuroscience. We are sure we were not unique in hearing this call; with how many others had he shared his vision? The science of the mind is a thinker's game. It is chess against the grandest masters, biological evolution and natural selection—and we are just learning to move the pieces. Our viewpoint is often myopic, with our noses pressed against the back row of the chessboard. It is hard to see the pieces, let alone their arrangement or the strategies they are forming. Francis may not have had the overview needed to reveal evolution's gambit, but he knew the moves needed to clear the “tangle of difficulties” ( Crick 1994 , p. 77) that prevented an unfogged view of his opponent's pieces. Francis hoped for simplicity. He wrote, “Curiously enough, in biology it is sometimes those basic problems that look impossibly difficult to solve which yield most easily…. The biological problems that are really difficult to unscramble are those where there is almost infinity of plausible answers and one has to painstakingly attempt to distinguish between them.” ( Crick 1990 , p. 157–158). Watson and Crick had picked the right pieces of information to construct their model. Francis early on had had the same hopes to open the doors of consciousness (to paraphrase Huxley 1963 ). Watson and Crick used their intuition to fill in the gaps. But Francis found that there were just too many possibilities, and the gaps in knowledge were still just too big for consciousness. In 1999, Francis felt that gentle and informal direction was not enough. Thus, he convened a meeting of molecular biologists and neuroscientists at the Salk Institute to encourage them to work together. He brought scientists including Tom Albright, Ursula Bellugi, Ed Callaway, Rusty Gage, Steve Heinemann, Terry Sejnowski, Chuck Stevens, and Inder Verma into one room and said it was time to get serious. He reminded them of the advantages of genetic methods for targeting specific cell types within complex neural circuits, and he reiterated the need for methods that could be used to identify, manipulate, and observe neural circuits in action. Not only were methods to be used in transgenic mice, but also methods based on viral vectors were needed to study the visual system of monkeys. From this, a number of initiatives moved forward, with studies ranging from the molecular biology of Williams syndrome to basic molecular tool building ( Naldini et al. 1996 ; Blomer et al. 1997 ; Bellugi et al. 1999 ; Pfeifer et al. 2001 ; Zhao et al. 2001 ; Kaspar et al. 2002a , 2002b , 2003 ; Lechner et al. 2002 ; Lein et al. 2004 ). Today the tools are emerging at an ever faster pace, at least in part due to Francis's maneuvers behind the scenes and his encouragement of junior scientists. Time is curing Francis's bout of scientific prematurity ( Stent 1972 ). Individual cell types will soon be reversibly inactivated ( Johns et al. 1999 ; Lechner et al. 2002 ; Slimko et al. 2002 ; Ibanez-Tallon et al. 2004 ); viral methods of tracing connections will start to fill in the gaps; new sensor methods for simultaneously recording from hundreds and thousands of identified neurons are coming ( Guerrero and Isacoff 2001 ; Zemelman and Miesenbock 2001 ; Tsien 2003 ). There is a new field Francis termed “molecular psychology” or “molecular biology of systems neuroscience”; Albright simply calls it Neuroscience. In 2001, Francis was diagnosed with colon cancer. He realized that the problem of the neural correlates of consciousness might outlast him. Francis was walking with a cane, still not waiting for anyone, nor allowing anyone to wait for him. He continued to find time for new faces in the field and continued to work on consciousness. While he had made many strides forward, he saw the race for him was winding down. He had had his hope for understanding the structure of consciousness. He had laid the groundwork. He decided to encapsulate his ideas in a “Framework” paper with Koch ( Crick and Koch 2003 ). For many of us it was clear that he was laying out where he would go, had he enough time. Each of the points of the Framework could form a major research initiative. Perhaps they should. But the central point is the approach to understanding consciousness; it is both structural and functional, peering forward into the future into what the shape might be. It was clear to his friends and colleagues that Francis was leaving a last testament. As the cancer finally caught up with Francis, he focused on the role of the rarely studied claustrum ( Sherk 1986 ). He wrote internal memos, brought friends and colleagues to working lunches at home with Odile, his wife of fifty-five years. Why do this? Why all this focus on another part of the brain, when only months remained? (Indeed it turned out to be weeks.) Was it his way of saying goodbye, of bringing his extended family close again? We think not. Francis wanted to make sure his plan went forward. He stressed to his visitors queries about the origins of the claustrum, its molecular biology, its role in consciousness. He was using his framework, pointing out the route to understanding the Ω of his career. Francis was doing what he truly loved to his last moments. He needed to be doing science, perhaps more than ever, to take him away from the physical pain that he surely felt. He had built his army. Perhaps none of us even knew we had enlisted, but we had. And he was setting us off on the long march forward into a time that soon would not be for him. Francis died on the cusp of a new age of molecular systems neuroscience. Soon, we will have the tools and the data, but we will not have Francis. Francis had existed between the α of DNA and the Ω of consciousness. And for a man who never believed in the afterlife, he had indeed achieved immortality.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535570.xml
517706	MUSCLE: a multiple sequence alignment method with reduced time and space complexity	Background In a previous paper, we introduced MUSCLE, a new program for creating multiple alignments of protein sequences, giving a brief summary of the algorithm and showing MUSCLE to achieve the highest scores reported to date on four alignment accuracy benchmarks. Here we present a more complete discussion of the algorithm, describing several previously unpublished techniques that improve biological accuracy and / or computational complexity. We introduce a new option, MUSCLE-fast, designed for high-throughput applications. We also describe a new protocol for evaluating objective functions that align two profiles. Results We compare the speed and accuracy of MUSCLE with CLUSTALW, Progressive POA and the MAFFT script FFTNS1, the fastest previously published program known to the author. Accuracy is measured using four benchmarks: BAliBASE, PREFAB, SABmark and SMART. We test three variants that offer highest accuracy (MUSCLE with default settings), highest speed (MUSCLE-fast), and a carefully chosen compromise between the two (MUSCLE-prog). We find MUSCLE-fast to be the fastest algorithm on all test sets, achieving average alignment accuracy similar to CLUSTALW in times that are typically two to three orders of magnitude less. MUSCLE-fast is able to align 1,000 sequences of average length 282 in 21 seconds on a current desktop computer. Conclusions MUSCLE offers a range of options that provide improved speed and / or alignment accuracy compared with currently available programs. MUSCLE is freely available at .	Background Multiple alignments of protein sequences are important in many applications, including phylogenetic tree estimation, secondary structure prediction and critical residue identification. Many multiple sequence alignment (MSA) algorithms have been proposed; for a recent review, see [ 1 ]. Two attributes of MSA programs are of primary importance to the user: biological accuracy and computational complexity (i.e., time and memory requirements). Complexity is of increasing relevance due to the rapid growth of sequence databases, which now contain enough representatives of larger protein families to exceed the capacity of most current programs. Obtaining biologically accurate alignments is also a challenge, as the best methods sometimes fail to align readily apparent conserved motifs [ 2 ]. We recently introduced MUSCLE, a new MSA program that provides significant improvements in both accuracy and speed, giving only a summary of the algorithm [ 2 ]. Here, we describe the MUSCLE algorithm more fully and analyze its complexity. We introduce a new option designed for high-throughput applications, MUSCLE-fast. We also describe a new method for evaluating objective functions for profile-profile alignment, the iterated step in the MUSCLE algorithm. Current methods While multiple alignment and phylogenetic tree reconstruction have traditionally been considered separately, the most natural formulation of the computational problem is to define a model of sequence evolution that assigns probabilities to all possible elementary sequence edits and then to seek an optimal directed graph in which edges represents edits and terminal nodes are the observed sequences. This graph makes the history explicit (it can be interpreted as a phylogenetic tree) and implies an alignment. No tractable method for finding an optimal graph is known for biologically realistic models, and simplification is therefore required. A common heuristic is to seek a multiple alignment that maximizes the SP score (the summed alignment score of each sequence pair), which is NP complete [ 3 ]. It can be achieved by dynamic programming with time and space complexity O( L N ) in the sequence length L and number of sequences N [ 4 ], and is practical only for very small N . Stochastic methods such as Gibbs sampling can be used to search for a maximum objective score [ 5 ], but have not been widely adopted. A more popular strategy is the progressive method [ 6 , 7 ] (Figure 1 ), which first estimates a phylogenetic tree. A profile (a multiple alignment treated as a sequence by regarding each column as a symbol) is then constructed for each node in the binary tree. If the node is a leaf, the profile is the corresponding sequence; otherwise its profile is produced by a pair-wise alignment of the profiles of its child nodes (Figure 2 ). Current progressive algorithms are typically practical for up to a few hundred sequences on desktop computers, the best-known of which is CLUSTALW [ 8 ]. A variant of the progressive approach is used by T-Coffee [ 9 ], which builds a library of both local and global alignments of every pair of sequences and uses a library-based score for aligning two profiles. On the BAliBASE benchmark [ 10 , 11 ], T-Coffee achieves the best results reported prior to MUSCLE, but has a high time and space complexity that limits the number of sequences it can align to typically around one hundred. In our experience, errors in progressive alignments can often be attributed to one of the following issues: sub-optimal branching order in the tree, scoring parameters that are not optimal for a particular set of sequences (especially gap penalties), and inappropriate boundary conditions (e.g., seeking a global alignment of proteins having different domain organizations). Misalignments are sometimes readily apparent, motivating further processing ( refinement ). One approach is to use a progressive alignment as the initial state of a stochastic search for a maximum objective score ( stochastic refinement ). Alternatively, pairs of profiles can be extracted from the progressive alignment and re-aligned, keeping the results only when an objective score is improved ( horizontal refinement ) [ 12 ]. Implementation The basic strategy used by MUSCLE is similar to that used by PRRP [ 13 ] and MAFFT [ 14 ]. A progressive alignment is built, to which horizontal refinement is then applied. Algorithm overview MUSCLE has three stages. At the completion of each stage, a multiple alignment is available and the algorithm can be terminated. Stage 1: draft progressive The first stage builds a progressive alignment. Similarity measure The similarity of each pair of sequences is computed, either using k -mer counting or by constructing a global alignment of the pair and determining the fractional identity. Distance estimate A triangular distance matrix is computed from the pair-wise similarities. Tree construction A tree is constructed from the distance matrix using UPGMA or neighbor-joining, and a root is identified. Progressive alignment A progressive alignment is built by following the branching order of the tree, yielding a multiple alignment of all input sequences at the root. Stage 2: improved progressive The second stage attempts to improve the tree and builds a new progressive alignment according to this tree. This stage may be iterated. Similarity measure The similarity of each pair of sequences is computed using fractional identity computed from their mutual alignment in the current multiple alignment. Tree construction A tree is constructed by computing a Kimura distance matrix and applying a clustering method to this matrix. Tree comparison The previous and new trees are compared, identifying the set of internal nodes for which the branching order has changed. If Stage 2 has executed more than once, and the number of changed nodes has not decreased, the process of improving the tree is considered to have converged and iteration terminates. Progressive alignment A new progressive alignment is built. The existing alignment is retained of each subtree for which the branching order is unchanged; new alignments are created for the (possibly empty) set of changed nodes. When the alignment at the root is completed, the algorithm may terminate, return to step 2.1 or go to Stage 3. Stage 3: refinement The third stage performs iterative refinement using a variant of tree-dependent restricted partitioning [ 12 ]. Choice of bipartition An edge is deleted from the tree, dividing the sequences into two disjoint subsets (a bipartition). Edges are visiting in order of decreasing distance from the root. Profile extraction The profile (multiple alignment) of each subset is extracted from the current multiple alignment. Columns containing no residues (i.e., indels only) are discarded. Re-alignment The two profiles obtained in step 3.2 are re-aligned to each other using profile-profile alignment. Accept/reject The SP score of the multiple alignment implied by the new profile-profile alignment is computed. If the score increases, the new alignment is retained, otherwise it is discarded. If all edges have been visited without a change being retained, or if a user-defined maximum number of iterations has been reached, the algorithm is terminated, otherwise it returns to step 3.1. Visiting edges in order of decreasing distance from the root has the effect of first re-aligning individual sequences, then closely related groups Algorithm elements In the following, we describe the elements of the MUSCLE algorithm. In several cases, alternative versions of these elements were implemented in order to investigate their relative performance and to offer different trade-offs between accuracy, speed and memory use. Most of these alternatives are made available to the user via command-line options. Four benchmark datasets have been used to evaluate options and parameters in MUSCLE: BAliBASE [ 10 , 11 ], SABmark [ 15 ], SMART [ 16 - 18 ] and our own benchmark, PREFAB [ 2 ]. Objective score In its refinement stage, MUSCLE seeks to maximize an objective score, i.e. a function that maps a multiple sequence alignment to a real number which is designed to give larger values to better alignments. MUSCLE uses the sum-of-pairs (SP) score, defined to be the sum over pairs of sequences of their alignment scores. The alignment score of a pair of sequences is computed as the sum of substitution matrix scores for each aligned pair of residues, plus gap penalties. Gaps require special consideration (Figure 3 ). We use the term indel for the symbol that indicates a gap in a column (typically a dash '-'), reserving the term gap for a maximal contiguous series of indels. The gap penalty contribution to SP for a pair of sequences is computed by discarding all columns in which both sequences have an indel, then applying an affine penalty g + λe for each remaining gap where g is the per-gap penalty, λ is the gap length (number of indels in the gap), and e is the gap-length penalty (sometimes called the extension penalty). Progressive alignment Progressive alignment requires a rooted binary tree in which each sequence is assigned to a leaf. The tree is created by clustering a triangular matrix containing a distance measure for each pair of sequences. The branching order of the tree is followed in postfix order (i.e., children are visited before their parent). At each internal node, profile-profile alignment is used to align the existing alignments of the two child subtrees, and the new alignment is assigned to that node. A multiple alignment of all input sequences is produced at the root node (Figure 1 ). Similarity measures We use the term similarity for a measure on a pair of sequences that indicates their degree of evolutionary divergence (the sequences are assumed to be related). MUSCLE uses two types of similarity measure: the fractional identity D computed from a global alignment of the two sequences, and measures obtained by k -mer counting. A k -mer is a contiguous subsequence of length k , also known as a word or k -tuple. Related sequences tend to have more k -mers in common than expected by chance, provided that k is not too large and the divergence is not too great. Many sequence comparison methods based on k -mer counting have been proposed in the literature; for a review, see [ 19 ]. The primary motivation for these measures is improved speed as no alignment is required. MAFFT uses k -mer counting in a compressed alphabet (i.e., an alphabet in which symbols denote classes that may contain two or more residue types) to compute its initial distance measure. The alphabet used in MAFFT is taken from [ 20 ], and is one of the options implemented in MUSCLE. Trivially, identity is higher or equal in a compressed alphabet; it cannot be reduced. If the alphabet is chosen such that there are high probabilities of intra-class substitution and low probabilities of inter-class substitution, then we might expect that detectable identity (and hence the number of conserved k -mers) could be usefully extended to greater evolutionary distances while limiting the increase in matches due to chance. We have previously shown [ 21 ] that k -mer similarities correlate well with fractional identity, although we failed to find evidence that compressed alphabets have superior performance to the standard alphabet at lower identities. We define the following similarity measure between sequences X and Y: F = Σ τ min [ n X ( τ ), n Y ( τ ) ] / [min ( L X , L Y ) - k + 1 ]. (1) Here τ is a k -mer, L X , L Y are the sequence lengths, and n X ( τ ) and n Y ( τ ) are the number of times τ occurs in X and Y respectively. This definition can be motivated by considering an alignment of X to Y and defining the similarity to be the fraction of k -mers that are conserved between the two sequences. The denominator of F is the maximum number of k -mers that could be aligned. Note that if a given k -mer occurs more often in one sequence than the other, the excess cannot be conserved, hence the minimum in the numerator. The definition of F is an approximation in which it is assumed that (after correcting for excesses) common k -mers are always alignable to each other. MUSCLE also implements a binary approximation F Binary , so-called because it reduces the k -mer count to a present / absent bit: F Binary = Σ τ δ XY ( τ ) / [min ( L X , L Y ) - k + 1 ]. (2) Here, δ XY ( τ ) is 1 if τ is present in both sequences, 0 otherwise. As multiple instances of a given k -mer in one sequence are relatively rare, this is often a good approximation to F . The binary approximation enables a significant speed improvement as the size of the count vector for a given sequence can be reduced by an order of magnitude. This allows the count vector for every sequence to be retained in memory, and pairs of vectors to be compared efficiently using bit-wise instructions. When using an integer count, there may be insufficient memory to store all count vectors, making it necessary to re-compute counts several times for a given sequence. Distance measures Given a similarity value, we wish to estimate an additive distance measure. An additive measure distance measure d(A, B) between two sequences A and B satisfies d(A, B) = d(A, C) + d(C, B) for any third sequence C, assuming that A, B and C are all related. Ideal but generally unknowable is the mutation distance , i.e. the number of mutations that occurred on the historical path between the sequences. The historical path through the phylogenetic tree extends from one sequence to the other via their most recent common ancestor. The mutation distance is trivially additive. The fractional identity D is often used as a similarity measure; for closely related sequences 1 - D is a good approximation to a mutation distance (it is exact assuming substitution at a single site to be the only allowed type of mutation and that no position mutates more than once). As sequences diverge, there is an increasing probability of multiple mutations at a single site. To correct for this, we use the following distance estimate [ 22 ]: d Kimura = -log e (1 - D - D 2 /5) (3) For D ≤ 0.25 we use a lookup table taken from the CLUSTALW source code. For k -mer measures, we use: d kmer = 1 - F . (4) Tree construction Given a distance matrix, a binary tree is constructed by clustering. Two methods are implemented: neighbor-joining [ 23 ], and UPGMA [ 24 ]. MUSCLE implements three variants of UPGMA that differ in their assignment of distances to a new cluster. Consider two clusters (subtrees) L and R to be merged into a new cluster P , which becomes the parent of L and R in the binary tree. Average linkage assigns this distance to a third cluster C : d Avg PC = ( d LC + d RC )/2. (5) We can take the minimum rather than the average: d Min PC = min [ d LC , d RC ]. (6) Following MAFFT, we also implemented a weighted mixture of minimum and average linkage: d Mix PC = (1 - s ) d Min PC + s d Avg PC , (7) where s is a parameter set to 0.1 by default. Clustering produces a pseudo-root (the last node created). We implemented two other methods for determining a root: minimizing the average branch weight [ 25 ], as used by CLUSTALW, and locating the root at the center of the longest span. Sequence weighting Conventional wisdom holds that sequences should be weighted to correct for the effects of biased sampling from a family of related proteins; however, there is no consensus on how such weights should be computed. MUSCLE implements the following sequence weighting schemes: none (all sequences have equal weight), Henikoff [ 26 ], PSI-BLAST [ 27 ] (a variant of Henikoff), CLUSTALW's, GSC [ 28 ], and the three-way method [ 29 ]. We found the use of weighting to give a small improvement in benchmark accuracy results, e.g. approximately 1% on BAliBASE, but saw little difference between the alternative schemes. The CLUSTALW method enables a significant reduction in complexity (described later), and is therefore the default choice. Profile functions In order to apply pair-wise alignment methods to profiles, a scoring function must be defined for a pair of profile positions, i.e. a pair of multiple alignment columns. This function is the profile analog of a substitution matrix; see for example [ 30 ]. We use the following notation. Let i and j be amino acid types, p i the background probability of i , p ij the joint probability of i and j being aligned to each other, S ij the substitution matrix score, f x i the observed frequency of i in column x of the first profile, f x G the observed frequency of gaps in that column, and α x i the estimated probability of observing i in position x in the family. (Similarly for position y in the second profile). Estimated probabilities α are derived from the observed frequencies f , typically by adding heuristic pseudo-counts or by using Bayesian methods such as Dirichlet mixture priors [ 31 ]. A commonly used profile function is the sequence-weighted sum of substitution matrix scores for each pair of letters, selecting one from each column (PSP, for profile SP): PSP xy = Σ i Σ j f x i f y j S ij . (8) Note that S ij = log ( p ij / p i p j ) [ 32 ], so PSP xy = Σ i Σ j f x i f y j log ( p ij / p i p j ). (9) PSP is the function used by CLUSTALW and MAFFT. It is a natural choice when attempting to maximize the SP objective score: if gap penalties are neglected, maximizing PSP maximizes SP under the constraint that columns in each profile are preserved. (This follows from the observation that the contribution to SP from a pair of sequences in the same profile is the same for all alignments allowed under the constraint). MUSCLE implements PSP functions based on the 200 PAM matrix of [ 33 ] and the 240 PAM VTML matrix [ 34 ]. In addition to PSP, MUSCLE implements a function we call the log-expectation (LE) score. LE is a modified version of the log-average (LA) profile function that was proposed on theoretical grounds [ 35 ]: LA xy = log Σ i Σ j α x i α y j p ij / p i p j . (10) LE is defined as follows: LE xy = (1 - f x G ) (1 - f y G ) log Σ i Σ j f x i f y j p ij / p i p j . (11) The MUSCLE LE function uses probabilities computed from VTML 240. Note that estimated probabilities α in LA are replaced by observed frequencies f in LE. The factor (1 - f G ) is the occupancy of a column. Frequencies f i must be normalized to sum to one if indels are present (otherwise the logarithm becomes increasingly negative with increasing numbers of gaps even when aligning conserved or similar residues). The occupancy factors are introduced to encourage more highly occupied columns (i.e., those with fewer gaps) to align, and are found to significantly improve accuracy. We avoid these complications in the PSP score by computing frequencies in a 21-letter alphabet (amino acids + indel), and by defining the substitution score of an amino acid to an indel to be zero. This has the desired effect of down-weighting column pairs with low occupancies, and can also be motivated by consideration of the SP function. If gap penalties are ignored, then this definition of PSP preserves the optimization of SP under the fixed-column constraint by correctly accounting for the reduced number of residue pairs in columns containing gaps. Gap penalties We call the first indel in a gap its gap-open ; the last its gap-close . Consider an alignment of two profiles X and Y, and a gap of length λ in X in which the gap-open is aligned to position y o in Y and the gap-close to y c . The penalty for this gap is b ( y o ) + t ( y c ) + λe , where b and t are costs for opening and closing a gap that vary according to the position in Y, and e is a length cost (sometimes called a gap extension penalty) that does not vary by position. A fixed length cost allows a minor optimization of the scoring scheme [ 14 ]. Consider a global alignment of sequences X and Y having lengths L X and L Y . If a constant δ (the center ) is added to each substitution matrix score and δ /2 is added for each gapped position, this adds the constant value δ ( L X + L Y )/2 to the score of any possible alignment, and the set of optimal alignments is therefore unchanged. Given a scoring scheme with substitution matrix S ij and extension penalty e , we can thus choose δ /2 = e and instead use S' ij = S ij + 2 e and e' = 0 to obtain the same alignment. The constant 2 e can be added to the substitution matrix at compile time, and no explicit extension penalty is then needed in the recursion relations. MUSCLE uses this optimization for the PSP function, but not for LE (where the center must be added at execution time after taking the logarithm). Let f y o be the number of gap-opens in column y in Y and f y c be the number of gap-closes in column y . MUSCLE computes b and t as follows (Figure 4 ): b ( y ) = g /2 (1 - f y o ) (1 + h w ( y ) H ), (12) t ( y ) = g /2 (1 - f y c ) (1 + h w ( y ) H ). (13) Here, g is a parameter that can be considered a default per-gap penalty, h w ( y ) is 1 if y falls within a window of w consecutive hydrophobic residues or zero otherwise, and H is a tunable parameter. By default, w = 5, H = 1.2. The factor g /2 (1 - f y o ) is motivated by considering the SP score of the alignment. The gap penalty contribution to SP for a pair of sequences ( A ∈ Y, B ∈ X) is computed by discarding all columns in which both sequences have an indel, then applying an affine penalty g + λe for each remaining gap. It is convenient here to consider that half of the per-gap penalty g is applied to the open position and half to the close position. Suppose a gap G is inserted into X such that the gap-open is aligned to position y in Y. If a sequence s ∈ Y has a gap-open at y , then the SP score includes no open penalty for G induced by any pair ( s , t ) : t ∈ X. The multiplier (1 - f y o ) therefore corrects the gap-open contribution to the SP score due to pre-existing gaps in Y. (It should be noted that even with this correction, there are other issues related to gaps and PSP still does not exactly optimize SP under the fixed-column constraint). The increased penalty in hydrophobic windows is designed to discourage gaps in buried core regions where insertions and deletions are less frequent. Note that MUSCLE treats open and close positions symmetrically, in contrast to CLUSTALW, which treats the open position specially and may therefore tend to produce, in word processing terms, left-aligned gaps with a ragged right margin. Terminal gaps A terminal gap is one that opens at the N-terminal position of the sequence to which it is aligned or closes at the C-terminal; as opposed to an internal gap . It has been suggested [ 9 , 36 ] that global methods have intrinsic difficulties with long deletions or insertions. We believe that these difficulties are often due to the choice of penalties for terminal gaps. CLUSTALW, which charges no penalty for terminal gaps, tends to fail to open a needed internal gap and thus fail to align terminal motifs; MAFFT, which charges the same penalty for terminal and internal gaps, sometimes aligns small numbers of residues to a terminal by inserting an unnatural internal gap. By default, MUSCLE penalizes terminal gaps with half the penalty of an internal gap. This is done by setting b (1), the open penalty at the C-terminal, and t ( L ), the close penalty at the N-terminal, to zero (Figure 4 ). The option of always applying full penalties, as in MAFFT, is also provided. We found that the compromise of a half penalty for terminal gaps gave good results for a wide range of input data, but that further improvements could sometimes by achieved by the following technique. If the length ratio of the two profiles to be aligned exceeds a threshold (by default, 20%), then MUSCLE constructs four different alignments in which gaps at both, one or neither terminals are fully penalized. A conservation score is defined by subtracting all gap penalties (both internal and terminal) from the alignment score, leaving a sum over profile functions only. The alignment with the highest conservation score is used. Tree comparison In progressive alignment, two subtrees will produce identical alignments if they have the same set of sequences at their leaves and the same branching orders (topologies). We exploit this observation to optimize the progressive alignment in Stage 2 of MUSCLE, which begins by constructing a new tree. Unchanged subtrees are identified, and their alignments are retained (Figure 5 ). A progressive alignment of the changed subtrees is constructed, producing the same alignment at the root that would be obtained starting from the leaves. Tree comparison is performed by the following algorithm. Consider two trees A and B with identical sets of N leaves. Leaves are identified by consecutive integers ( ids ) 1 through N . Call a pair of nodes, one from each tree, equivalent if they are the same leaf or they are internal nodes and their children are equivalent. The left/right position of a child is not considered; in other words, subtree rotations are allowed (because they do not change the results of a progressive alignment). Traverse A in prefix order (children before their parent), assigning internal nodes ids N + 1 through 2 N in the order visited. When visiting an internal node P A , take the ids of its two child nodes L A and R A and use them as indexes into a lookup table pointing to nodes in B . If (a) L A is equivalent to a node L B in B and R A is equivalent to a node R B , and (b) L B and R B have the same parent P B , then assign P B the same id as P A , to which it is equivalent. When the traversal is complete, a node b in B is equivalent to some node in A if and only if b has an id. This procedure is O( N ) time and space. Defaults, optimizations and complexity analysis We now discuss the default choices of algorithm elements in the MUSCLE program and analyze their complexity. Complexity of CLUSTALW It is instructive to consider the complexity of CLUSTALW. This is of intrinsic interest as CLUSTALW is currently the most widely used MSA program and, to the best of our knowledge, its complexity has not previously been stated correctly in the literature. It is also useful as a baseline for motivating some of the optimizations used in MUSCLE. The CLUSTALW algorithm can be described by the same steps as Stage 1 above. The similarity measure is the fractional identity computed from a global alignment, clustering is done by neighbor-joining. Global alignment of a pair of sequences or profiles is computed using the Myers-Miller linear space algorithm [ 37 ] which is O( L ) space and O( L 2 ) time in the typical sequence length L . Given N sequences and thus N ( N - 1)/2 = O( N 2 ) pairs, it is therefore O( N 2 L 2 ) time and O( N 2 + L ) space to construct the distance matrix. The neighbor-joining implementation is O( N 2 ) space and O( N 4 ) time, at least up to CLUSTALW 1.82, although O( N 3 ) time is possible; see e.g. [ 38 ]. A single iteration of progressive alignment computes a profile of each subtree from its multiple alignment, which is O( N P L P ) time and space in the number of sequences in the profile N P and the profile length L P , then uses Myers-Miller to align the profiles in O( L P ) space and O( L P 2 ) time. There are N - 1 internal nodes in a rooted binary tree and hence O( N ) iterations. It is often assumed that L P is O( L ), i.e. that O(0) gaps are introduced in each iteration. However, we often observe the alignment length to grow approximately linearly, i.e. that O(1) gaps are added per iteration. For example, taking the average over all iterations in all alignments in version 3 of the PREFAB benchmark, Stage 1 of MUSCLE adds 2.8 gaps per iteration to the longer profile. It is therefore more realistic to assume that L P is O( L + N ), making one iteration of progressive alignment O( NL + L 2 ) in both space and time. This analysis is summarized in Table 1 . Initial distance measure One might expect (a) that a more accurate distance measure would lead to a more accurate final alignment due to an improved tree, and (b) that errors due to a less accurate distance measure might be eliminated by allowing Stage 2 to iterate more times. Neither of these expectations is supported by our test results (unpublished). Allowing Stage 2 to iterate more than once with the goal of further improving the tree gave no significant improvement with any distance measure. Possibly, the tree is biased towards the MSA that was used to estimate it, and the MSA is biased by the tree used to create it, making it hard to achieve improvements. The most accurate measure on a pair of sequences is presumably the fractional identity D computed from a global alignment, but use of D in step 1.1 does not improve average accuracy on benchmark tests. The 6-class Dayhoff alphabet used by MAFFT proved to give slightly higher benchmark accuracy scores, despite the fact that other alphabets were found to correlate better with D [ 21 ]. We also found that the use of the binary approximation F Binary gave slightly reduced accuracy scores even when Stage 2 was allowed to iterate. The default choice in MUSCLE is therefore to use the Dayhoff alphabet in step 1.1 and to execute Stage 2 once only. While the impact on the average accuracy of the final alignment due to the different options is not understood, we observe that a better alignment of a pair of sequences is often obtained from a multiple alignment than from a pair-wise alignment, due to the presence of intermediate sequences having higher identities. It is therefore plausible that D obtained from the multiple alignment in step 2.1 may be more accurate than D obtained from a pair-wise alignment in step 1.1, and this may be relatively insensitive to the method used to create the tree for Stage 1. But this leaves unexplained why k -mer counting appears to be as good as or better than D in Stage 1. Computing F from a pair of sequences is O( L ) time and O(1) space, so for all pairs the similarity calculation is O( N 2 L ), compared with O( N 2 L 2 ) in CLUSTALW. For a typical L around 250, combined with an order of magnitude improvement due to the simplicity of k -mer counting compared with dynamic programming, this typically gives a three orders of magnitude speed improvement for computing the distance matrix in MUSCLE compared with CLUSTALW. The default strategy is therefore well justified as a speed optimization, and has the added bonus of providing a small improvement in accuracy. Clustering MUSCLE implements both UPGMA and neighbor-joining. We found UPGMA to give slightly better benchmark scores than neighbor-joining; UPGMA is therefore the default option. We expect neighbor-joining to give a better estimate of the correct evolutionary tree (see e.g. [ 38 ]). However, it is well-known that alignment accuracy decreases with lower sequence identity (see e.g. [ 39 ]). It follows that given a set of profiles, the two that can be aligned most accurately will tend to be the pair with the highest identity, i.e. at the shortest evolutionary distance. This is exactly the pair selected by the nearest-neighbor criterion in UPGMA. By contrast, neighbor-joining selects a pair of evolutionary neighbors, i.e. a pair having a common ancestor. When mutation rates are variable, the evolutionary neighbor may not be the nearest neighbor (Figure 6 ). This explains why a nearest-neighbor tree may be superior to the true evolutionary tree for guiding a progressive alignment. Neighbor-joining is naively O( N 4 ) time, although this can be reduced to O( N 3 ). UPGMA is naively O( N 3 ) time as the minimum of an N 2 matrix must be found in each of N - 1 iterations. However, this can be reduced to O( N 2 ) time by maintaining a vector of pointers to the minimum value in each row of the matrix. We are again fortunate to find that the most accurate method is also the fastest. Dynamic programming The textbook algorithm for pair-wise alignment with affine penalties employs three dynamic programming matrices; see e.g. [ 40 , 41 ]. A more time-and space-efficient implementation can be achieved using linear space for the recursion relations and a single matrix for trace-back (Kazutaka Katoh, personal communication). Consider sequences X and Y length L X , L Y . We use the following notation: X x is the x th letter in X, X x the first x letters in X, S xy the substitution score (or profile function) for aligning X x to Y y , b X x the score for a gap-open in Y that is aligned to X x , t X x the score for a gap-close aligned to X x , U xy the set of all alignments of X x to Y y , M xy the score of the best alignment in U xy ending in a match (i.e., X x and Y y are aligned), D xy the score of the best alignment ending in a delete relative to X (X x is aligned to an indel) and I xy the score of the best alignment ending in an insert (Y y is aligned to an indel). A match is preceded by either a match, delete or insert, so: M xy = S xy + max { M x -1 y -1 , D x -1 y -1 + t X x -1 , I x -1 y -1 + t Y y -1 } (14) We assume that a center parameter has been added to S xy such that the gap extension penalty is zero. By considering all possible lengths for the final gap, D xy = max( k < x ) [M ky + b X k +1 ]. (15) Here, k is the last position in X that is aligned to a letter in Y. Extract the special case of a gap of length 1: D xy = max { max( k < x -1) [M ky + b X k +1 ], M x -1 y + b X x }. (16) Hence, D xy = max { D x -1 y , M x -1 y + b X x }. (17) Similarly, I xy = max { I xy -1 , M xy -1 + b Y y }. (18) Let the outer loop iterate over increasing x and the inner loop over increasing y . For fixed x , define vectors M curr y = M xy , M prev y = M x -1 y , D curr x = D xy , D prev x = D x -1 y ; for fixed x , y define scalars I curr = I xy , I prev = I xy -1 . Now we can re-write (14), (17) and (18) to obtain the following recursion relations: M curr y = S xy + max { M prev y -1 , D prev y -1 + t X x -1 , I prev y -1 + t Y y -1 } (19) D curr y = max { D prev y , M prev y + b X x } (20) I curr = max { I prev , M prev y + b Y y }. (21) An L X × L Y matrix is needed for the trace-back that produces the final alignment. Inner loop The inner-most dynamic programming loop, which computes the profile function, deserves careful optimization. We will consider the case of PSP; similar optimizations are possible for LE. PSP = Σ i Σ j f x i f y j S ij = Σ i f x i W y i , where W y i = Σ j f y j S ij . The vector W y i is used L X times, and it therefore pays to compute it once and cache it. Observe that a typical profile column contains << 20 different amino acids. We sort the frequencies in decreasing order; the summation Σ i f x i W y i is terminated if a frequency f x i = 0 is encountered. This typically reduces the time spent in the summation, especially when sequences are closely related. As with W y i , the sort order is computed once and cached. Observe that the roles of the two profiles are not symmetrical. It is most efficient to choose X, for which frequency sort orders are computed, to be the profile with the lowest amino acid diversity when averaged over columns. With this choice, the summation terminates earlier on average then if the other profile is identified as X. Note that out of N - 1 iterations of progressive alignment, a minimum of and maximum of N - 1 profile-profile alignments will include at least one profile containing one sequence only, and in the refinement phase exactly N of the 2 N - 1 edges in the tree terminate on a leaf. At least half of all profile-profile alignments created in the MUSCLE algorithm therefore include a profile of one sequence only. Special cases where one or both profiles is a single sequence can be handled in separate subroutines, saving overhead due to unneeded loops that are guaranteed to execute once only. This optimization is especially useful for the LE function as it enables the logarithm to be incorporated into the W vector. Diagonal finding Many alignment algorithms are optimized for speed, typically at some expense in average accuracy, by using fast methods to identify regions of high similarity between two sequences, which appear as diagonals in the similarity matrix. The alignment path is then constrained to include these diagonals, reducing the area of the dynamic programming matrix that must be computed. MAFFT uses the fast Fourier transform to find diagonals. MUSCLE uses a different technique which we have previously shown [ 21 ] have comparable sensitivity and to be significantly faster. We use a compressed alphabet to find k -mers in common between two sequences, then attempt to extend the match. In the case of diagonal identification we found compressed alphabets to significantly out-perform the standard amino acid alphabet [ 21 ]. Currently, MUSCLE uses 6-mers in the Dayhoff alphabet for diagonal finding, as for the initial distance measure, though other alphabets are known to give slightly better performance [ 21 ]. A candidate diagonal is rejected if there is any overlap (i.e., if a single position in one of the sequences appears in two or more diagonals) or if it is less than a minimum length (default 24). The ends of the diagonal are deleted (by default, the first and last five positions) as they are less reliable. Despite these heuristics, we find the use of diagonal-finding to reduce average accuracy and to give only modest improvements in speed for typical input data; this option is therefore disabled by default. Similar results are seen in MAFFT; the most accurate MAFFT script is NWNSI [ 14 ], in which diagonal-finding is also disabled. Additive profiles Both the PSP and LE profile functions are defined in terms of amino acid frequencies and position-specific gap penalties. The data structure representing a profile is a vector of length L P in which each element contains frequencies for each amino acid type and a few additional values related to gaps. We call this data structure a profile vector , as distinct from a profile matrix , an explicit N × L P multiple alignment containing letters and indels. For N > 20, using profile vectors reduces the cost of computing the profile function compared with profile matrices, and is therefore preferred for use in dynamic programming. In CLUSTALW and MAFFT, the implementation of progressive alignment builds a profile matrix at each internal node of the tree, which is used to compute a profile vector. This procedure is O( NL P ) = O( N 2 + NL ) in time and space, becoming expensive for large N . Observe that the count of a given amino acid in a column in the parent matrix is the sum of the counts in the two child columns that are aligned at that position (Figure 7 ). With a suitable sequence weighting scheme, it is therefore possible to compute the amino acid frequencies of the parent profile vector from the frequencies in the two child profile vectors and the alignment path. This is an O( L P ) procedure in both time and space, giving a significant advantage for N >> 20. Three issues must be addressed to fully implement this idea: the sequence weighting scheme, inclusion of occupancy factors and position-specific gap penalties, and construction of a profile matrix (i.e., the final multiple alignment) at the root node. Sequence weighting For the frequencies in the parent profile vector to be a linear combination of the child frequencies, the weight assigned to a sequence must be the same in the child and parent profiles. This requirement is not satisfied, for example, by the Henikoff or PSI-BLAST schemes, which compute weights based on a multiple alignment. We therefore choose the CLUSTALW scheme, which computes a fixed weight for each sequence from edge lengths in the tree. Gap representation To compute gap penalties, we need the frequencies f o of gap opens and f c of gap closes in each position. In the case of the LE profile function, we additionally require the gap frequency f G . These can be accommodated by storing f o , f c and f e in the profile vector, where f e is the frequency of gap-extensions in the column (meaning that indels are found in a given sequence in the column, the preceding column and in the following column; i.e., a gap-close is not counted as an extension). These three occupancy frequencies are sufficient for computing the profile function and the position-specific gap penalties b and t . Note that we can compute the frequency f G of indels, as needed for the occupancy factor in the profile function, as follows: f G = f o + f c + f e . (22) Now consider the problem of computing the occupancy frequencies in the parent profile vector, given only the child occupancy frequencies and the trace-back path for the alignment. Consider first a diagonal edge in the path, i.e. an edge that does not open or extend a gap, following another diagonal edge. In this case, the occupancy frequencies are computed similarly to amino acid frequencies (as a sum in which a child frequency is weighted according to the total weight of the sequences in its profile). For horizontal or vertical edges, i.e. edges that open or extend gaps, the parent occupancy frequencies can be computed by considering the effect of the new column of indels (Figure 8 ). It is straightforward to work through all cases and show that the three frequencies f o , f c and f e are sufficient for their values in the parent profile vector to be computed in O( L P ) time from the child profile vectors and alignment path. Construction of the root alignment By avoiding the use of profile matrices, the complexity of a single progressive alignment iteration is reduced from O( L P 2 + NL p ) space and O( L P 2 + NL P ) time to O( L P 2 ) = O( L 2 + NL ) space and time. The NL term in the time complexity is now due only to the increase in profile length, and is therefore typically much smaller than before. The root alignment is constructed by storing the alignment path produced at each internal node. For each input sequence, the path to the root is followed, inserting the gaps induced by each alignment path at each internal node. This procedure is O( NL P log N ) = O( N 2 log N + NL log N ) time, and requires O( NL P ) = O( NL + N 2 ) space for storage of the paths. This is expensive for large N , and we therefore optimize this step by using a device we call an e-string , a type of edit string. E-strings An alignment path can be considered as an operator on a pair of sequences that inserts indels into those sequences such that their lengths become equal. Conventionally, an alignment path is represented as a vector of three symbols representing edges in the graph, say M, D and I (for match, delete and insert, i.e. a diagonal, horizontal or vertical edge). Note that indels in one sequence are inserted only by Ds, indels in the other are inserted only by Is. Define an e-string e to be a vector of \| e \| integers interpreted as an operator that inserts indels into a string by scanning it from left to right (Figure 9 ). A positive integer n means skip n letters of the string; a negative integer - n means insert n indels at the current position. We require the vector to be in its shortest form, so signs always alternate. We represent an alignment path as a pair of e-strings, one for each sequence, assigned to the appropriate edges in the tree. We will typically find that \| e \|, the length of the e-string, is much less than L P , the length of the alignment path. Now consider the effect of applying two consecutive e-strings ("multiplying" them). This can be expressed as a third e-string, which can be efficiently computed in O(\| e \|) time from the multiplicands. For each leaf (input sequence), the product is computed of e-strings on the path to the root (Figure 10 ). The final e-string obtained at the root is then applied to the sequence. This method does not reduce the big-O time or space complexity, but is much faster than a naive implementation. Brenner's method Steven Brenner (personal communication) observed that a multiple alignment can be alternatively be obtained by aligning each sequence to the root profile. This requires O( NL P 2 ) time, giving a lower asymptotic complexity in N at the expense of an additional factor of L P . This method gives opportunities for errors relative to the "exact" e-string solution (when a sequence misaligns to its copy in the profile), but can also lead to improvements by allowing the sequence to correctly align to conserved motifs that were not apparent when the sequence was added. (Note the resemblance to the refinement stage, which begins by re-aligning individual sequences to the rest). The chances for error are reduced by constraining the alignment to forbid gaps in the root profile. Our tests show that this method gives comparable average accuracy to the e-string solution but to be slower for up to at least a few thousand sequences, and e-strings are therefore used by default. Refinement complexity In the following, we assume that an explicit multiple alignment (profile matrix) of all sequences is maintained, and determine the complexity of each step in Stage 3. Step 3.1 determines the bipartition induced by deleting an edge from the tree. This is O( N ) time, and sufficiently fast that there is little motivation for further optimization. Step 3.2 extracts profiles for the two partitions from the current multiple alignment and computes their profile vectors, which is O( NL P ) time and space. Step 3.3 performs profile-profile alignment, which is O( L P 2 ) time and space. Step 3.4 computes the SP score, which is O( N 2 L P ) time and O( NL P ) space (discussed in more detail shortly). A single iteration of Stage 3 is thus O( N 2 L P + L P 2 ) time and O( NL P + L P 2 ) space. There are O( N ) edges in the tree, so executing this process for all edges is O( N 3 L P + NL P 2 ) time and O( NL P + L P 2 ) space, which is O( N 4 + N 3 L + NL 2 ) time and O( N 2 + NL + L 2 ) space. Assuming that a fixed maximum number of iterations of Stage 3 is imposed, this is also the total complexity of refinement. We now consider optimizations of the refinement stage. Anchor columns A multiple alignment can be divided vertically at high-confidence ( anchor ) columns. Each vertical block is then refined separately, improving speed and reducing space due to the O( L 2 ) factor in dynamic programming. This strategy has been used by several previous algorithms, including PRRP [ 13 ], RASCAL [ 42 ] and MAFFT. In MUSCLE, the following criteria are used to identify anchor columns. The profile function (LE or PSP) must exceed a threshold, the averaged profile function over a window around the position must exceed a (lower) threshold, and the column may not contain a gap. In addition, the contribution to the averaged score from a single column has a ceiling, reducing skew in the averaged score due to exceptionally high-scoring columns. These heuristics are designed to avoid anchor columns that have high scores but are either artifacts (similar residues found by chance in unrelated regions) or are too close to variable regions. When performing a profile-profile alignment, each anchor column and its two immediate neighbors (which form the boundaries of vertical blocks) are required to remain aligned; i.e., terminal gaps are forbidden except at the true terminals. Introducing this constraint overcomes a small degradation in average alignment quality that is otherwise observed. This implies that the degradation is sometimes due to cases where a well-conserved region is divided into two parts by an anchor column, one of which becomes short enough that it misaligns to a similar short motif elsewhere. SP score Notice that computation of the SP score dominates the time complexity of refinement and of MUSCLE overall, introducing O( N 4 ) and O( N 3 L ) terms. We are therefore motivated to consider optimizations of this step. We first consider the contribution SP a to the SP score from amino acids; gap penalties require special treatment. Let s and t be sequences, x be a column, s [ x ] be the amino acid of sequence s in column x , and S ( i , j ) be the substitution score of amino acids i and j . It is convenient to impose an (arbitrary) ordering on the sequences and amino acid types. Then, SP a = Σ x Σ s Σ t > s S ( s [ x ], t [ x ]). (23) Define δ ( s , i , x ) = 1 if s [ x ] = i , 0 otherwise, and n i [ x ] = Σ s δ ( s , i , x ). We say n i [ x ] is the count of amino acid type i in column x . We can now transform the sum over pairs of sequences into a sum over pairs of amino acid types: SP a = Σ x Σ i n i Σ j n j > i S ( i , j ) + 1/2 Σ x Σ i ( n i 2 - n i ) S ( i , i ). (24) Frequencies are computed as: f x i = n i [ x ]/ N . (25) Using frequencies, For simplicity, we have neglected sequence weighting; it is straightforward to show that (26) applies unchanged if weighting is used. Note that (23) is O( N 2 L P ) but (25) and (26) are O( NL P ). For N >> 20, this is a substantial improvement. Let SP g be the contribution of gap penalties to SP, so SP = SP a + SP g . It is natural to seek an O( NL P ) expression for SP g analogous to (26), but to the best of our knowledge no solution is known. Note that in MUSCLE refinement, the absolute value of the SP score is not needed; rather, it suffices to determine the difference in the SP scores before and after re-aligning a pair of profiles. Let SP( s , t ) be the contribution to the SP score from a pair of sequences s and t , so SP = Σ s Σ t > s SP( s , t ), and denote the two profiles by X and Y. Then we can decompose SP into intra-and inter-profile terms as follows: SP = Σ s ∈X Σ t ∈X: t > s SP( s , t ) + Σ s ∈Y Σ t ∈Y: t > s SP( s , t ) + Σ s ∈X Σ t ∈Y SP( s , t ) (27) Note that the intra-profile terms are unchanged in any alignment that preserves the columns of the profile intact, which is true by definition in profile-profile alignment. This follows by noting that any indels added to align the profiles are guaranteed to be external gaps with respect to any pair of sequences in the same profile. It therefore suffices to compute the change in the inter-profile term: SP XY = Σ s ∈X Σ t ∈Y SP( s , t ). (28) This reduces the average time by a factor of about two. We can further improve on this by noting that in the typical case, there are few or no changes to the alignment. This suggests computing the change in SP score by looking only at differences between the two alignments. Let π - be the alignment path before re-alignment and π + the path after re-alignment. The change in alignment can be specified as the set of edges in π - or π + , but not both; i.e., by considering a path to be a set of edges and taking the set symmetric difference Δ π = ( π - ∪ π + ) - ( π - ∩ π + ). The path π + after re-alignment is available from the dynamic programming traceback. The path π - before re-alignment can be efficiently computed in O( L P ) time. Note that in order to construct the profile of a subset of sequences extracted from a multiple alignment, those columns that contain only indels in that subset must be deleted. The set of such columns in both profiles is therefore available as a side effect of profile construction, and this set immediately implies π - . It is a simple O( L P ) procedure to compute Δ π from π - and π + . Note that SP a is a sum over columns, and there is a one-to-one correspondence between columns and edges in π . The change in SP a can therefore be computed as a sum over columns in Δ π , with a negative sign for edges from π - , reducing the time complexity from O( NL P ) to O( N \|Δ π \|). We now turn our attention to SP g . We say that a gap G intersects Δ π if and only if any indel in G is in a column in Δ π , and denote by Γ the set of gaps that intersect Δ π . If a gap does not intersect Δ π , i.e. does not have an indel in a changed column, its contribution to SP g is unchanged. It therefore suffices to consider penalties for gaps in Γ, again with negative signs for edges from π - . The construction of Γ is straightforward in O( NL P ) time. Finally, a sum over pairs in Γ is needed, reducing the O( N 2 ) component to the smallest possible set of terms. Dimer approximation We next describe an approximation to SP that can be computed in O( NL P ) time. Define a two-symbol alphabet {X, -} in which X represents any amino acid and - is the indel symbol. There are four dimers in this alphabet: XX, X-, -X and --, which denote by no-gap, gap-open, gap-close and gap-extend respectively. Re-write a multiple alignment in terms of these dimers, adopting the convention that dimer ab composed of symbol a in column x -1 and symbol b in column x is written in column x . Now consider the contribution to SP g of an aligned pair of dimers, written as ab ↔ cd . Clearly XX↔X- adds a gap-open penalty; XX↔-X adds a gap-close (Figure 11 ). To avoid double-counting, we will include only the penalty contribution of indels in the second column. Then XX↔X- adds a per-gap penalty, but XX↔-X adds zero because the second column does not contain a gap. External indels must be discarded; so, for example, --↔-- adds zero. In fact, aligning two identical dimers always contributes zero because any indel in the second column is found in both sequences and is therefore external. The contribution of all possible pairs of dimers is unambiguous, with the exception of -X↔--, which can add a per-gap or extend penalty (Figure 12 ). We approximate this case by assigning it a penalty of tg , where g is the default per-gap penalty and t is a tunable parameter, set to 0.2 by default. With this approximation, dimers can be treated like amino acids: the scores for each aligned pair of dimers forms a substitution matrix (Figure 13 ), and SP g can be computed by summing substitution scores over all pairs of sequences. We can now apply Equation 26, re-interpreting the frequency vectors f as having 24 components (20 amino acids and four dimers), and compute the change in SP by considering only those columns in Δ π . We find use of the dimer approximation to marginally reduce benchmark scores. By default, MUSCLE therefore uses the exact SP score for N ≤ 100 and the dimer approximation for N > 100, where the higher time complexity of the exact score becomes more noticeable. Evaluation of profile functions We have previously attempted a systematic comparison of profile functions [ 30 ]. The methodology used in that work demanded careful optimization of affine gap parameters for each function. This proved to be time-consuming and tedious, and we therefore tried the following alternative approach, inspired by the notion that a good profile function should be good at discriminating correctly aligned pairs of profile positions from incorrectly aligned pairs. The protocol begins with a set of pair-wise structural alignments. With the sequence of each structure as a query, we used PSI-BLAST to search the NCBI non-redundant protein sequence database [ 43 ], giving a multiple sequence alignment (profile) for each structure. Note that we use the term profile in this context to mean the sequence alignment produced by PSI-BLAST, not the scoring matrix. Using the structural alignments as a guide, we then created a database in which columns from the PSI-BLAST profiles were aligned to each other, giving a large set of pairs of alignment columns that we consider to be correctly aligned (the "true" database, although there are undoubtedly misaligned sequences and hence some incorrect pairs). By selecting the same number of pairs of columns at random from structures in different FSSP families, we created a similar ("false") database of unrelated pairs. A profile function was evaluated by computing the score of each pair of columns in both the true and false databases, and then sorting the results in order of increasing score. The results can be displayed by reviewing the sorted list and, for each score S in the list, plotting the number of true pairs with score ≤ S ( x axis) against the number of false pairs with score ≤ S ( y axis); we call the resulting graph a discrimination plot . Ideally, all true pairs would score higher than all false pairs, in which case the profile function would be a perfect discriminator and would always produce perfect alignments. A function that perfectly discriminates will appear as a Γ-shaped plot; a function that has no ability to discriminate will appear as a diagonal plot along the line x = y . If a function F has a discrimination plot that is always above another function G (i.e., D F ( x ) > D G ( x ) ∀ x , where D F is the discriminator plot for F as a function of x ), then F has a superior ability to discriminate true from false pairs compared with G. If the plots intersect, the situation is ambiguous and neither function is clearly superior. We used sets of structural alignments from [ 30 ] (PP) and [ 44 ] (PP2). PP contains 588 structure pairs with sequence identity ≤ 30%, z-score ≥ 15, RMSD ≤ 2.5Å and an alignment length of ≥ 50 positions. These criteria were designed to select pairs of structures with low sequence identity and high structural similarity. PP2 contains 500 pairs selected from the FSSP database [ 45 ] with ≤ 30% sequence identity, z-score ≥ 8 and ≤ 12, RMSD ≤ 3.5Å and alignment length ≥ 50. The criteria for PP2 were designed to select challenging alignments with low sequence identity and relatively high structural divergence, leading to a high frequency of gaps and therefore, presumably, a stronger dependence on accurate identification of sequence similarity. Results on PP2 show the LE function to have higher discrimination than all other tested functions (historically, the LE function was designed by systematic trial and error using a wide range of different profile functions with feedback from discrimination plots). This is illustrated in Figure 14 , in which the discrimination plot for LE on PP2 is compared with several other functions: PSP, LA, Yona and Levitt's [ 46 ], LAMA [ 47 ]. Using PP, we again find that LE is superior to LA (not shown), but the comparison with PSP is ambiguous as the discrimination plots intersect (Figure 15 ). A major advantage of this approach is that no gap penalties are required, with the result that once the databases have been constructed, a new function can be tested in seconds rather than the days or weeks that were needed with the earlier methodology. However, some caveats are in order. We are using PSI-BLAST as a gold standard for creating profiles, but PSI-BLAST may introduce biases both due to its selection of sequences for inclusion in the profile and due to errors in alignments of those sequences to the query. If the profile function will be used to align PSI-BLAST profiles, then this is an appropriate experimental design. But in the case of multiple sequence alignment, where profiles are produced iteratively by the profile function itself, the results may not be directly applicable. We also note that any monotonic transformation of the profile function leaves the discriminator plot unchanged as it does not change the sort order of the scores. (A monotonic transformation is F' = m (F) where m ( x ) is a monotonically increasing function). However, a monotonic transformation may change the alignments produced by a profile function, so we can regard high discrimination as a necessary but not sufficient condition for a good profile function. One can turn this into a virtue by noting that the discrimination plot allows the relative probability of true versus false to be determined from a score. It is therefore possible to numerically determine a log-odds function from the discrimination plot, which can be evaluated by table look-up. Using discrimination plots for PP2, we found the optimal transformation for LE to be close to linear, in contrast to other functions we tried, including PSP (results not shown). This observation further encouraged us to explore the performance of LE in an MSA algorithm. Testing on multiple alignment benchmarks we find LE to give superior results on BAliBASE, but statistically indistinguishable results on other databases (results not shown). MUSCLE therefore uses LE as the default choice as it sometimes gives better results but has not been observed to give lower average accuracy on any of our tests. It is also useful to introduce a method with a distinctively different scoring scheme as an alternative that may give better results on some input data and may provide unique features for incorporation into jury or consensus systems. One drawback of LE is its relatively slow performance due to the need to compute a logarithm for each cell of the dynamic programming matrix. Complexity of MUSCLE The complexity of MUSCLE is summarized in Table 2 . We assume L P = O( L + N ), the e-string construction for the root alignment, and a fixed number of refinement iterations. Results MUSCLE offers a variety of options that offer different trade-offs between speed and accuracy. In the following, we report speed and accuracy results for three sets of options: (1) the full MUSCLE algorithm including Stages 1, 2 and 3 with default options; (2) Stages 1 and 2 only, using default options (MUSCLE-prog); and (3) Stage 1 only using the fastest possible options (MUSCLE-fast), which are as follows: F Binary is used as a distance measure (Equation 2), the PSP profile function is used, and diagonal finding is enabled. Alignment accuracy In Tables 3 and 4 we report the speed and accuracy of MUSCLE v3.3, CLUSTALW v1.82, Progressive POA, a recently published method that is claimed to be 10 to 30 times faster than CLUSTALW for large alignments [ 48 ], and the MAFFT script FFTNS1 v3.82, the fastest previously published method known to us. On the advice of one of the authors of Progressive POA, we used command-line options selecting global alignment with truncated gap scoring (Catherine Grasso, personal communication). We report results both using distance matrices computed by BLAST (POA-blast) and using the distance method built into the program (POA). We use four sets of reference alignments: BAliBASE v2, PREFAB v3, SABmark v1.61, and a version of SMART from July 2000. The accuracy score is Q , the number of residue pairs correctly aligned divided by the length of the reference alignment. For more discussion of the reference data, assessment methodology and a comparison of MUSCLE with T-Coffee and NWNSI, the most accurate MAFFT script, see [ 2 ]. Execution speed To compare speeds for a larger number of sequences, we created a test set by using PSI-BLAST to search the NCBI non-redundant protein sequence database for hits to dienoyl-coa isomerase (1dci in the Protein Data Bank [ 49 ]), selecting the highest-scoring 1,000 sequences. This set of sequences had average length 282, maximum length 454 and average pair-wise identity 20%. We aligned randomly chosen subsets of from 200 to 1,000 sequences with each program and noted the total execution time. In the case of 1,000 sequences, the resulting alignments had from 1,100 from 1,400 columns, confirming that it is unrealistic to assume that L P is O( L ). Results are shown in Figure 16 . We have previously shown that MUSCLE-prog is faster than FFTNS1 on a set of 5,000 sequences, for which we estimated that CLUSTALW would require approximately one year [ 2 ]. In this test, MUSCLE-fast is approximately 3× faster than FFTNS1 for 200 sequences, and 5× faster for 1,000 sequences. This trend continues for larger numbers of sequences (complete results not shown), showing that MUSCLE-fast has lower asymptotic complexity, due largely to the use of additive profiles for progressive alignment compared with the profile matrices constructed by FFTNS1. Conclusions MUSCLE demonstrates improvements in accuracy and reductions in computational complexity by exploiting a range of existing and new algorithmic techniques. While the design–typically for practical multiple sequence alignment tools–arguably lacks elegance and theoretical coherence, useful improvements were achieved through a number of factors. Most important of these were selection of heuristics, close attention to details of the implementation, and careful evaluation of the impact of different elements of the algorithm on speed and accuracy. MUSCLE enables high-throughput applications to achieve average accuracy comparable to the most accurate tools previously available, which we expect to be increasingly important in view of the continuing rapid growth in sequence data. Availability and requirements MUSCLE is a command-line program written in a conservative subset of C++. At the time of writing, MUSCLE has been successfully ported to 32-bit Windows, 32-bit Intel architecture Linux, Solaris, Macintosh OSX and the 64-bit HP Alpha Tru64 platform. MUSCLE is donated to the public domain. Source code and executable files are freely available at .	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517706.xml
534108	A systematic review of the effectiveness of antimicrobial rinse-free hand sanitizers for prevention of illness-related absenteeism in elementary school children	Background Absenteeism due to communicable illness is a major problem encountered by North American elementary school children. Although handwashing is a proven infection control measure, barriers exist in the school environment, which hinder compliance to this routine. Currently, alternative hand hygiene techniques are being considered, and one such technique is the use of antimicrobial rinse-free hand sanitizers. Methods A systematic review was conducted to examine the effectiveness of antimicrobial rinse-free hand sanitizer interventions in the elementary school setting. MEDLINE, EMBASE, Biological Abstract, CINAHL, HealthSTAR and Cochrane Controlled Trials Register were searched for both randomized and non-randomized controlled trials. Absenteeism due to communicable illness was the primary outcome variable. Results Six eligible studies, two of which were randomized, were identified (5 published studies, 1 published abstract). The quality of reporting was low. Due to a large amount of heterogeneity and low quality of reporting, no pooled estimates were calculated. There was a significant difference reported in favor of the intervention in all 5 published studies. Conclusions The available evidence for the effectiveness of antimicrobial rinse-free hand sanitizer in the school environment is of low quality. The results suggest that the strength of the benefit should be interpreted with caution. Given the potential to reduce student absenteeism, teacher absenteeism, school operating costs, healthcare costs and parental absenteeism, a well-designed and analyzed trial is needed to optimize this hand hygiene technique.	Background With the recent emergence of severe acute respiratory syndrome (SARS), a newly discovered infectious disease, the importance of primary infection control measures have been highlighted [ 1 , 2 ]. Routine handwashing with soap and water has been cited by the World Health Organization (WHO) as being " the most important hygiene measure in preventing the spread of infection " [ 3 ]. This statement has been reiterated by both the United States Centers for Disease Control (CDC) and Health Canada, in reference to reducing the transmission of SARS, the influenza virus, and other infectious pathogens [ 3 - 5 ]. The epidemiological evidence supporting the effectiveness of this basic measure in healthcare settings dates back to the mid-nineteenth century [ 6 , 7 ]. Ignaz Semmelweis, a Hungarian obstetrician, implemented routine handwashing with chlorinated lime by maternity ward staff, as a mechanism to reduce the incidence of puerperal fever [ 6 , 7 ]. This simple routine elicited dramatic results, reducing the mortality rate from 13–18 percent to 2 percent [ 6 , 7 ]. These findings have been replicated numerous times in hospital environments- underlining the magnitude of routine handwashing [ 8 , 9 ]. The elementary school environment is also negatively impacted by outbreaks of disease causing microorganisms [ 10 , 11 ]. These occasional outbreaks result in increased student and teacher absenteeism, increased healthcare expenditures, and an overall decline in the children's learning environment [ 11 ]. The United States CDC has estimated that the average school-aged child missed approximately one week annually due to illness-related absenteeism in 1995 [ 12 ]. Despite the scientifically proven evidence of the effectiveness of handwashing, and the increasing promotion of proper hand hygiene techniques, observational studies in school settings have indicated that handwashing practices are often lacking [ 13 , 14 ]. Guinan et al . (1997) reported that proper handwashing compliance, with soap and water, in school-aged children ranged from 8 to 28 percent. Reported reasons for the observed inadequacy in compliance included insufficient time during the day, and the use of substandard washing facilities in hard to access locations of the school environment [ 13 , 14 ]. In attempts to overcome the obstacles of routine handwashing in school environments, antimicrobial rinse-free hand sanitizers are being used as an alternative hand hygiene technique. The concern is that such programs may be carried out in the absence of evidence of effectiveness in the school environment. Thus, it is timely to review the evidence currently available for the effectiveness of antimicrobial rinse-free hand sanitizer programs in reducing absenteeism due to communicable illness. The aim of this systematic review was to determine whether antimicrobial rinse-free hand sanitizer interventions are effective in preventing illness-related absenteeism in elementary school children. Methods A detailed written protocol was prepared and reviewed in advance (complete protocol can be obtained from the corresponding author). Search strategy A comprehensive search was conducted to identify all relevant studies regardless of publication status. Six electronic databases were searched for studies published in any language. The databases included: Biological Abstracts (1990-May 2003), CINAHL (1982–2003), the Cochrane Controlled Trials Register (1981–2003), EMBASE (1980-May 2003), HealthSTAR (1975-May 2003), and MEDLINE (1966-May 2003). A detailed search strategy was developed for use in MEDLINE and an iterative process was completed to refine the MEDLINE search for each database. Descriptions of the database search strategies are presented in Appendix 1 (see Additional File 1 ). OVID served as the primary search interface, and the SDI feature was used to monitor newly posted citations, the most recent date September 30, 2004. Due to the low occurrence of studies in this subject area, no filters were used to identify specific study types or reviews. The reference lists of all relevant articles were reviewed for additional studies. A letter was sent to all corresponding authors of the articles identified by hand-search, excluding two newly eligible citations posted between May 2003 and September 2004 [ 15 , 16 ], or by searching bibliographic databases. Additionally, contact experts and the industrial companies that manufactured the antimicrobial hand gels used in the included trials (GOJO Industries and Woodward Laboratories, Inc.) were contacted in attempts to identify other eligible trials. A detailed list of contacts is provided in Appendix 2 (see Additional File 2 ). Finally, conference proceedings for the American Journal of Infection Control (2000–2004) and recently published issues of the American Journal of Infection Control (February 2003 to August 2004) were searched by hand. Eligibility criteria Studies were evaluated for inclusion on the basis of four criteria: target population, intervention, outcome, and study design. The target population of interest consisted of elementary school children between 4 and 12 years of age (including senior kindergarten and grades 1 through 8). The interventions of interest were those that administered antimicrobial rinse-free hand hygiene programs compared with no intervention or placebo treatment arm in a school setting. The outcome of interest was the comparison of the number of absences due to communicable illnesses in children who received the antimicrobial rinse-free hand hygiene intervention with the number of such absences in those who received a placebo or no intervention. We included cluster randomized controlled trials (RCTs) and cluster non-randomized controlled trials regardless of publication status. Study selection All relevant citations, titles and abstracts, were imported into a reference database where duplicates were manually removed. Priority in downloading was given to MEDLINE. Reviews were excluded, but the bibliographies from such articles were examined for relevant studies. The screening was completed in an unblinded manner; there is inconclusive evidence that blinding introduced bias into the process [ 17 ]. One individual (EM) independently screened the titles and abstracts of each citation and identified all citations for full review. One reviewer was deemed appropriate as it was thought that the reviewer would error on the side of caution. Hard copies of all potentially relevant citations were retrieved, and two reviewers (EM, NLS) independently assessed each article using the aforementioned eligibility criteria, excluding the newly published citations in which eligibility was assessed by EM [ 15 , 16 ]. Disagreements were discussed and a final decision was made by means of open consensus. A pilot test assessing the eligibility criteria on a sample of articles was not performed. Recent studies by both Juni et al . 2002 and Moher et al . 2003 indicate that the exclusion of trials in languages other than English (LOE) does not bias measures of effectiveness- however, both are cautionary, advocating language inclusive search strategies [ 18 , 19 ]. Due to limited fiscal resources, an English language restriction was applied at this level, however the number of citations in LOE that met eligibility criteria will be noted. Data abstraction Two reviewers (EM, NLS) independently abstracted data from all studies meeting the eligibility criteria, excluding one abstract where EM independently abstracted pertinent information [ 15 ], using pre-printed data collection forms presented in Appendix 3 (see Additional File 3 ). Information pertaining to the descriptive details of the study (e.g., year published, language of publication, publication status), design (e.g., randomized controlled trial), population (e.g., age, grade level), intervention (e.g., type of antimicrobial rinse-free hand sanitizer, inclusion of an educational component), and primary outcome(s) (e.g., absences due to illness) were collected. Adverse advents were not considered due to the relatively benign nature of the intervention. Reviewers resolved differences by means of open consensus. For the case of crossover study designs, data from the both arms of the study was abstracted. A pilot test assessing the data collection form on a sample of articles was not performed. Quality assessment Two reviewers (EM, NLS) independently assessed the quality of each of the included studies, excluding the abstract by Thompson (2004) previously mentioned [ 15 ], using the previously validated 3-item Jadad scale, which assesses the quality of the report in terms of randomization generation, double blinding, and withdrawals and drop-outs by intervention group [ 20 ]. Studies were not given a quantitative score; rather this was used as a qualitative tool. (Due to the nature of the intervention, not all items apply). Additionally, if trials were randomized, allocation concealment was assessed and qualitatively evaluated as adequate, inadequate or unclear [ 21 ]. Disagreements were resolved through open consensus. A pilot study applying the quality assessment criteria on a subset of studies was not completed. Data analysis Data synthesis and analysis was performed in accordance with the Cochrane Reviewers' Handbook [ 22 ]. Firstly, data was qualitatively synthesized to examine the overall pattern of studies with respect to study design, population, intervention, and outcome characteristics. Sources of clinical and statistical heterogeneity were identified and results were examined. All data was abstracted as reported. The primary outcome, frequency of absences due to communicable illness was analyzed. Percent relative differences were presented along with 95 percent confidence intervals as the estimate of intervention effectiveness in the four studies which calculated rate and risk ratios as the measure of association [ 10 , 27 , 31 , 41 ]. Percent relative differences and 95 percent confidence intervals were calculated enabling the results to be compared between studies, without altering the measure of association reported in the studies (rate and risk ratios). In the studies where data could not be abstracted, measures of association were reported [ 15 , 16 ]. The validity of performing a quantitative synthesis was considered, however based on a qualitative inspection of heterogeneity and estimates of intervention effectiveness this was not deemed appropriate. Thus, sensitivity and subgroup analyses were not performed, and publication bias was not assessed quantitatively. Results Description of studies A flow diagram of the search results is illustrated in Figure 1 . From the searches of the electronic databases, a total of 211 citations were identified, of which 70 were duplicates, resulting in the identification of 141 unique citations. In all, 18 potentially relevant trials were retrieved from the searches of the relevant databases. Hand-searching of the reference lists of relevant articles and conference proceedings resulted in a further 7 trials, which were also reviewed for consideration. Thus, 25 studies were determined potentially relevant [ 10 , 23 - 46 ]. Using both titles and abstracts, no trials meeting the inclusion criteria in LOE were found during the study selection phase. Figure 1 Flow diagram outlining the results of literature search and review of studies retrieved After review of the full text of these studies, 21 were excluded for the following reasons: no outcomes of interest (n = 1), inappropriate population (n = 1), inappropriate interventions (n = 9), inappropriate study design (n = 3), irrelevant subject matter (n = 5), and review (n = 2). Thus, a total of 4 trials fulfilled the inclusion criteria [ 10 , 27 , 31 , 41 ]. However, during the time between manuscript submission and revision, 2 additional citations were deemed to be eligible [ 15 , 16 ], bringing the total to n = 6 eligible studies, one of which was a published manuscript [ 16 ], and the other a published abstract [ 15 ]. Of the 6 remaining studies, 2 were crossover studies [ 16 , 27 ], 1 was a placebo-controlled cluster randomized controlled trial (RCT) [ 41 ], 2 were cluster non-randomized controlled trials (NRCT) [ 10 , 31 ], and the published abstract was a cluster trial, however randomization was unclear [ 15 ]. McNemars's test was used to assess observer agreement, chi-square = 2.00; df = 1, p = 0.1573; and there was 92 percent agreement between the two reviewers with respect to study relevance for the initial four trials included [ 10 , 27 , 31 , 41 ]. All of the relevant trials are described in Table 1 , and an assessment of their quality of reporting is presented in Table 2 (the study by Thompson 2004 was excluded as only the abstract was available). Our overall agreement was 89 percent with respect to data collection for the five included studies outlined in Table 1 . For quality abstraction, percent agreement was 80 percent. When examining items relating to blinding and assessment of withdrawals and dropouts, observer agreement was 100 and 80 percent, respectively. Kappa statistics were not calculated as sample size was insufficient. Table 1 Characteristics of studies included in the systematic review, demographics and descriptive statistics Author(s), Year, and Country Source of Funding Study Population Definition Illness-related Absenteeism Unit of analysis Study Duration Intervention Arm Control Arm Cluster Randomized Controlled Trials (RCTs) White 41 2001 United States Industry (Woodward Laboratories, Inc.) 1 private and 2 public elementary schools grades K-6; 72 initial classes (n = 1626 students) * Target group : • 16 classes, n = 388 students Control group : • 16 classes, n = 381 students • GI or respiratory-related • Parents reported events Grouped by classroom 5 weeks • Education: presentation and video describing germs and proper handwashing techniques (1 hr session) • Large involvement of both parents and school staff • Each child received 1-oz bottle of SAB (CleanHands ® ) (alcohol-free) instant hand sanitizer • Instructed to use spray under teacher supervision to supplement handwashing • Education: presentation and video describing germs and proper handwashing techniques (1 hr session) • Large involvement of both parents and school staff • Each child received 1-oz bottle of placebo formulation • Instructed to use spray under teacher supervision to supplement handwashing Crossover Dyer 27 2000 United States Industry (Woodward Laboratories, Inc.) Private elementary school (K-6); 2 classrooms per grade level, n = 30 students per classroom Target group : • Children grades K-6 • 1 classroom per grade level • n = 210 students Control group : • Children grades K-6 • 1 classroom per grade level • n = 210 students • GI (symptoms including vomiting, abdominal pain, and diarrhea) • Respiratory-related (symptoms included cough, sneezing, sinus trouble, bronchitis, fever alone, pink eye, headache, mononucleosis, and acute exacerbation of asthma) • Parents reported events Grouped by classroom 10 weeks (4 weeks first arm, 2 week washout period, 4 weeks second arm) • Education: presentation and video describing germs and proper handwashing techniques (1 hr session) • Each child received 1-oz bottle of SAB (CleanHands ® ) (alcohol-free) instant hand sanitizer • Instructed to use spray under teacher supervision to supplement handwashing • Education: presentation and video describing germs and proper handwashing techniques (1 hr session) • Instructed to wash hands as usual Morton 16 2004 United States Maine Administrative School District #35 in Eliot, and South Berwick, Maine; Erie Scientific donated AlcoSCRUB ® 1 elementary school in northern New England, grades K-3; 17 classrooms and n = 285 students eligible PHASE 1: Target group : • 9 classrooms Control group : • 8 classrooms • PHASE 2: • reversed • GI (symptoms including influenza, diarrhea, nausea, or vomiting (with or without fever)) • Respiratory-related (symptoms included nasal congestion, cough, or sore throat (with or without fever)) • Parents reported events Grouped 100 days (46 day first arm, 1 week washout period, 47 day second arm) • Education: guardians provided with study information and a contact number for the school nurse; additionally, monthly updates were provided • Education: students received a carefully planned education program; 45 minute "Germ Unit", Glo Germ™ presentation • Reinforcement: 1 st week: daily reminders given to students, after 1 st week reinforcement given weekly and after holidays; each classroom visited twice by school nurse during two arms • AlcoSCRUB ® dispensers were furnished in each classroom, located near the classroom entrance at a height that was accessible to all students • Gel use was monitored, and reinforcement was given to classes with low use • Education: information was presented by school nurse about proper handwashing • Instructed to wash hands as usual Cluster Non-randomized Controlled Trials (NRCTs) Hammond 31 2000 United States GOJO Industries, Inc. 18 public elementary schools grades K-6 in 6 school districts * Target group : • 5 school districts: District 1 (Ohio; K-5; n = 1440 students), District 2 (Ohio; 2,3; n = 266 students), District 3 (Delaware; 3,4; n = 110 students), District 4 (Tennessee; K-6; n = 680 students), District 5 (California; K-5; n = 579 students) Control group : • 5 school districts: District 1 (K-5; n = 1136 students), District 2 (2,3; n = 552 students), District 3 (3,4; n = 113 students), District 4 (K-6; n = 592 students), District 5 (K-5; n = 612 students) • Infectious process such as cold, flu, and gastroenteritis (common infectious illnesses such as pink eye, abscesses, and skin infections were not included) • No information regarding reporting of events Grouped by school and grouped by classroom 10 months • Each test classroom was equipped with a dispenser of PURELL instant alcohol-based hand sanitizer; also placed in other locations around the school • Reinforcement by study co-ordinators every 3 months • No intervention Guinan 10 2002 United States GOJO Industries, Inc. 5 elementary schools; 4 schools had 4 classrooms, 1 school had 2 classrooms (coed and single sex schools) Target group : • Children grades K-3 • 4 schools with 2 classrooms, 1 school with 1 classroom • n = 145 students Control group : • Children grades K-3 • 4 schools with 2 classrooms, 1 school with 1 classroom • n = 145 students • Infectious process such as cold, flu, and gastroenteritis • Children and parents reported events • Had to be 5 days between episodes to count Grouped by classroom 3 months • Education: presentation and video describing germs and proper handwashing techniques (1 hr session) • Each test classroom was equipped with a dispenser of an alcohol-based hand sanitizer • No intervention Cluster Controlled Trials-randomization unclear Thompson 15 2004 United States (abstract only) Not available 5 grade two classrooms, and 1 one/two combination classroom, n = 138 children Target group : • 3 classrooms Control group : • 3 classrooms • Illness = cold, flu, conjunctivitis, and gastrointestinal symptoms • Teachers recorded absences n/a n/a • Age appropriate interactive learning session • Alcohol-based hand sanitizer installed in intervention classrooms n/a * Not including drop-outs or withdrawals † n/a = not available Table 2 Quality assessment of trials meeting inclusion criteria Author(s), Year, and Country Quality Assessment Allocation Concealment Additional Comments Cluster RCT White 41 2001 United States Study was described as randomized however did not explain method of randomization; participants and study-coordinators were blinded; description of withdrawals and dropouts provided: of the 72 initial classes (1626 students), 32 classes (16 target and 16 control; 769 students) participated (remainder dropped from analysis) Unclear Sample size calculation not defined; statistical methods unclear; parents required to sign detailed informed consent form; study reviewed and approved by the two school boards; soap and handwashing not monitored; clustering not accounted for Crossover Dyer 27 2000 United States Study was not formally randomized; neither participants or study-coordinators were blinded; description of withdrawals and dropouts provided: no exclusions from the population were necessary Not Relevant Sample size calculation not defined; statistical methods unclear; no parental consent form; study not approved by a formal university institutional review board (approved by school board of education); limited SES diversity; soap and handwashing not monitored; clustering not accounted for Morton 16 2004 United States Study was described as randomized however did not explain method of randomization; neither participants or study-coordinators were blinded; description of withdrawals and dropouts provided: of the 17 initial classes (285 students), 17 classes ((253 students, 120 girls and 133 boys), non-consent = 22 children, adverse-events = 10 children) Unclear Sample size calculation not defined; data not in a format which could be easily extracted; study approved by Board of Education, and the Institutional Review Board at the state's largest hospital; a consent form was sent to all parents and guardians; clustering not accounted for Cluster NRCTs Hammond 31 2000 United States Study was not formally randomized; neither participants or study-coordinators were blinded; description withdrawals and dropouts provided: 1 school district did not comply with protocol; 25/3080 students did not participate or complete the protocol (in each case, data was not used for results) Not Relevant Sample size calculation not defined; no parental consent form; formal review not mentioned; clustering not accounted for Guinan 10 2002 United States Study was not formally randomized; participants and study-coordinators were not blinded; description withdrawals and dropouts not provided Not Relevant Sample size calculation not defined; statistical methods unclear; no parental consent form; formal review not mentioned (approved by each school); limited SES diversity (high SES); performed in peak absenteeism season; clustering not accounted for All six trials were conducted in the United States, 4 with reported industrial sponsorship, and were published between 2000 to 2004. The trials varied in size (range = 138 to 6080 students; range = 1 to 18 schools), and geographic locations (Pennsylvania [ 10 ], California [ 27 , 37 ], Ohio/Tennessee/Delaware/California [ 31 ], New England [ 16 ]). One of the studies assessed only private schools [ 28 ]; another study assessed both private and public schools [ 41 ], three assessed only public schools [ 10 , 16 , 31 ], while one's type of school assessed was not available [ 15 ]. Additionally, there was considerable variation in the type of school included both within and between trials: Christian private school, public elementary schools, same-sex schools and co-ed schools. The duration of the studies ranged from 5 weeks to 10 months, the longest trial being that of the RCT [ 41 ]. The trials also varied with respect to the intervention administered. White et al . (2001) and Dyer et al . (2000), provided each student with alcohol-free instant hand sanitizer [ 27 , 41 ], whilst Hammond et al . (2000), Guinan et al . (2002), Morton et al . (2004), and Thompson et al . (2004) provided each class with alcohol-based instant hand-sanitizer dispensers [ 10 , 15 , 16 , 31 ]. Education was concurrently provided for both the control and intervention arms in two studies [ 16 , 27 , 41 ], education on germs and hygiene was provided only to the intervention arm in one study [ 10 ], one study did not provide any education however study reinforcement was provided for teaching staff [ 31 ], and one study provided education to the intervention arm but as only the abstract was available for this study it was unclear if the control arm received any education [ 15 ]. Methodological quality The quality of reporting of the 5 trials that were examined in detail was low. Only one study was described as being randomized and double-blinded, however, it failed to describe a detailed and appropriate method of randomization and allocation concealment was unclear [ 41 ]. Four of the five studies, as previously mentioned, discussed withdrawals and dropouts however the description was quite basic and detailed flow-diagrams outlining the passage of participants through the trial were not supplied [ 16 , 27 , 31 , 41 ]. White et al . (2001) reported a significant number of dropouts (857 of 1626 students did not complete the study) however no explanations were offered [ 41 ]. Four studies received industrial sponsorship either by GOJO industries or Woodward Laboratories. In addition, two studies received financial support from another external source [ 16 , 41 ]. Other characteristics of poor quality reporting included: no sample size calculation defined for all five studies, and the statistical methods were vague. No studies took into consideration clustering when analyzing their results. Our overall agreement for all items of quality was greater than 80 percent; again observer agreement statistics were not calculated as the sample size was insufficient. Primary outcome All six studies varied in their definition of communicable illness-related absenteeism, refer to Table 1 . Out of the five studies with published manuscripts, four of the studies reported the estimate of intervention effectiveness in terms of risk/rate ratios, subsequently calculating percent relative effect, and one reported and odds ratio for a pair-matched study; results are reported in Table 3 , Table 4 , and Table 5 . The percent relative effect measures the decreased rate of the occurrence of absenteeism when the rate ratio is the measure of association or the decreased risk of absenteeism when the relative risk is the measure of association. Tests of significance were completed in all five of the studies using chi-squared tests or t tests however no confidence intervals were calculated. Two of the studies used rate ratios as the measure of association [ 10 , 31 ], two studies used relative risks as the measure of association [ 27 , 41 ], and the other used the odds ratio [ 16 ]. Table 3 , 4 and 5 distinguish between the measures of association used. All studies found a statistically significant effect of the antimicrobial rinse-free hand gel interventions. In the study by Hammond et al . (2000), the experimental group had a 20% (95% CI = 19–21%) reduction in absences due to communicable illness, the experimental group in the trial completed by Guinan et al . (2002) had a 49% (95% CI = 42–56%) reduction, White et al . (2001) demonstrated a 33% (95% CI = 17–45%) reduction in the experimental group, Dyer et al . (2000) had a 34 % (95% CI = 10–50%) reduction in absences due to communicable illness in the experimental group in the first phase and a 56 % (95% CI = 31–72%) in the second phase, and Morton et al . (2004) reported a significant odds ratio for McNemar's test (chi-square = 7.787; p = 0.0053). Table 3 Absences due to communicable illness, person-time incidence rates and percent relative effects of a non-alcoholic rinse-free hand sanitizer Intervention Control Trials No. of students No. of absences (no. of student-days) Absenteeism rate per 100 student-days No. of absences (no. of student-days) Absenteeism rate per 100 student-days Percent Relative Effect (95% CI)* White et al. 41 770 153 (9615) 1.59 222 (9459) 2.35 33 (17, 45) Dyer et al. 27 Phase 1 420 70 (4136) 1.69 105 (4120) 2.55 34 (10, 50) Phase 2 420 28 (4156) 0.674 63 (4140) 1.52 56 (31, 72) * Percent relative effect = (1- intervention rate/control rate)100 95 percent confidence interval = (1–95% UCL of Rate Ratio)100 to (1–95% LCL of Rate Ratio)100 Table 4 Absences due to communicable illness, cumulative incidence rates and percent relative effects of an alcoholic rinse-free hand sanitizer Intervention Control Trials No. of students No. of absences (No. of students) Absenteeism risk No. of absences (No. of students) Absenteeism risk Percent Relative Effect (95% CI) Hammond et al. 31 6080 7441.5 (3075) 2.42 9066 (3005) 3.02 20 (19, 21) Guinan et al. 10 290 140 (145) 0.97 277 (145) 1.91 49 (42, 56) * Percent relative effect = (1- intervention risk/control risk) 100 95 percent confidence interval = (1–95% UCL of Risk Ratio)100 to (1–95% LCL of Risk Ratio)*100 Table 5 Measures of association reported for studies in which no data could be extracted Trials No. of students Raw data reported Measures of association, and statistical tests reported Thompson 15 138 days absent per student in intervention group = 2.30 days absent per student in control group = 3.20 • Overall reduction in absenteeism due to illness was 28 percent for children using alcohol hand rub Morton et al. 16 253 N = 211 absent overall n = 42 never absent due to illness n = 103 ill regardless of participation in the control or the AlcoSCRUB ® group n = 69 ill in control group n = 39 ill in AlcoSCRUB ® group • McNemar's test for dichotomous variables with paired subjects, was used to assess strength of intervention: chi-square = 7.787; p = .0053 • Odds of being ill decreased by 43 percent with use of AlcoSCRUB ® Discussion Many studies have examined the importance of preventing the transmission of infectious diseases in the school environment, one such studied completed by Cramer et al . 1999, indicated this item to be of great concern for the parent's of school-aged children [ 47 ]. The most common infections transmitted in school environments are respiratory (influenza, pharyngitis etc) and diarrheal illnesses (i.e., Norwalk virus). Most of the infections occur at a constant low level but occasionally outbreaks do occur resulting in increased absenteeism and involvement of public health authorities. Since hands are the primary mechanism of transmission of these illnesses, proper hand hygiene techniques have been endorsed as the first defence at reducing the risk of transmission [ 1 - 5 , 8 , 10 , 28 ]. In health care settings, the routine use of antimicrobial alcohol based hand gels has been endorsed as an alternative to handwashing when hands are not visibly soiled [ 48 - 50 ]. Effectiveness in the hospital setting has not been easy to document given the relative low incidence of documented infections that can be specifically related to nosocomial transmission relative to the high number of handwashing opportunities in specific environments such as intensive care unit settings. Can the evidence from these six trials reported here be used to promote this type program in elementary schools at the present time? This systematic review of antimicrobial rinse-free hand sanitizers for prevention of illness-related absenteeism in elementary school children is the first review, of the author's awareness, to assess this issue. Although randomized controlled trials are the study design least likely to provide biased estimates of effect, due to the nature of school-based interventions, the inclusion of both randomized and non-randomized cluster controlled trials was required [ 51 ]. Of the six studies that met our inclusion criteria, three were non-randomized cluster controlled trials. Recent evidence indicates that non-randomized designs overestimate the effect of an intervention, thus the magnitude of the results should be interpreted with caution [ 52 ]. Four of the six studies used an alcohol-based product, the other two using a benzalkonium chloride based disinfectant. The FDA in the United States has indicated that insufficient data exits to classify the latter compounds as safe and effective to use as antiseptic handwashes. They are also adversely affected in the presence of organic material such as food residues, which may be an issue in schools [ 53 ]. Four studies were industry sponsored, and five were flawed due to the lack of sample size calculations. The five studies included were of low quality and methodologically weak. The only blinded randomized study using a placebo incorporated in this review was reported to be randomized and double-blinded, however, a description of the randomization technique was not discussed in the report and allocation concealment was unclear [ 41 ]. Additionally, this study suffered from a large proportion of withdrawals and drop-outs, thus the results had to be cautiously interpreted. Current studies have indicated that poor quality studies are associated with exaggerated treatment effects [ 52 ]. Although all studies reported statistically significant effects of the antimicrobial rinse-free hand gel in the experimental group, the aforementioned evidence suggests the reader should interpret these results cautiously. Thus, a clear delineation of the effectiveness of the intervention cannot be resolved from this review. Several limitations were encountered when completing this review, the major one being the scarcity of high quality studies. Additionally, although content experts, primary authors and industrial companies were contacted, no grey literature was found. The possible existence of unpublished non-significant trials should not be discounted. The validity of performing a quantitative synthesis was considered, however based on a qualitative inspection of heterogeneity and estimates of intervention effectiveness this was not deemed appropriate. Sources of heterogeneity included study designs, population characteristics, intervention characteristics, case definition and primary outcome measure. Thus, sensitivity and subgroup analyses were not performed, and publication bias was not assessed quantitatively. Another limitation was the fact that one reviewer was used to do the broad screen of articles and review the two citations identified between September 2003 and the present time. This may have biased the results; however, it is believed that this reviewer would overestimate the citations to be included. Conclusions In wake of the recent worldwide emergence of Severe Acute Respiratory Syndrome (SARS), the importance of proper hand hygiene has been brought to the spotlight. Comprehensive hand hygiene programs with occasional reinforcement are an inexpensive intervention, which potentially can work for a broad population, with minimal adverse effects. Future research should concentrate on developing study protocols that are scientifically sound with regards to randomization generation, blinding, allocation concealment and other factors that will minimize or avoid bias. Hand hygiene programs are the most important infection control measure in the school environment and have potentially large public health and economic implications therefore their design, implementation, and analysis should be carried out with the rigour. Competing interests The author(s) declare that they have no competing interests. Authors' contributions EM conceived and designed the study as part of a graduate course in systematic reviews, reviewed trials for inclusions, abstracted data, participated in data analysis, and drafted and revised the manuscript. NLS participated in initial study design, reviewed trials for inclusion, abstracted data, participated in data analysis, and revised the manuscript. Both EM and NLS agreed upon the final revision. Appendices Appendix 1 – Syntax for searches Appendix 2 – List of corresponding authors, content experts and industrial companies contacted Appendix 3 – Data collection form Pre-publication history The pre-publication history for this paper can be accessed here: Supplementary Material Additional file 1 Additional file 1 - Syntax for searches Click here for file Additional file 2 Additional file 2 - List of corresponding authors, content experts and industrial companies contacted Click here for file Additional file 3 Additional file 3 - Data collection form Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC534108.xml
517935	Effects of age and leg length upon central loop of the Gastrocnemius-soleus H-reflex latency	Background central loop of the gastrocnemius-soleus H-reflex latency (T c ) that looks promising in the diagnosis of S1 radiculopathy; has been investigated in a few studies and only two of them have focused on the constitutional factors affecting it. Although leg length has been shown to contribute to the T c , the role of age is controversial. More confusing, none of the previously performed studies have used strict criteria to rule out subclinical neuropathy, so the results could be misleading. This study has been performed to determine the influence of leg length and age on T c among a carefully selected group of healthy volunteers. Methods after screening forty six volunteers by taking history, physical examination and a brief electrophysiologic study; forty of them were selected to enroll into the study. T c was obtained in all the study subjects and leg length and age were recorded for correlational analyses. Results this group was consisted of 26 males (65%) and 14 females (35%) with the age range of 19–65 years (Mean ± SD: 37 ± 10.7) and leg length range of 29.5–43 centimeters (36.4 ± 3.4). Mean ± SD for T c was 6.78 ± 0.3. We found a significant correlation between T c and leg length (p value= 0.003, r = 0.49 and confidence interval 95% = 0.59–0.88), no significant correlation was found between age and T c (p value= 0.48, r = 0.11), also we obtained the regression equation as: T c = 0.04L + 5.28 Conclusions in contrast to leg length, age was not correlated with T c . Future studies are required to delineate other contributing factors to T c .	Background The H-Reflex evaluates S 1 radiculopathy [ 1 ]. The measured latency, however, is neither specific nor sensitive for S 1 spinal nerve disease, as it traverses a long pathway. Pease et al [ 2 ] were the first who described the central loop of the gastrocnemius-soleus H-Reflex latency (Central S 1 loop latency or T c ) and suggested it might be promising in the diagnosis of S1 radiculopathy [ 1 , 2 ]. Unfortunately, T c has been the subject of few studies, and as far as we know, only 5 articles [ 2 - 6 ] have been published on this issue so far. Among them, two have specifically evaluated the constitutional factors contributing to T c . Leg length has been shown to have a significant effect on T c . It is controversial whether age entails a similar effect. Wang et al [ 5 ] found a direct correlation between age and Tc. This observation was not confirmed by Ghavanini et al in an independent study [ 6 ]. The current study has been performed to determine the influence of leg length and age on T c . Methods We enrolled 46 volunteers to this study after obtaining informed consent. Following a standard history taking, all of them underwent physical examination and a brief electrophysiologic evaluation [ 7 ] to rule out asymptomatic polyneuropathy, including determination of: right peroneal nerve conduction velocity (PNCV), distal motor latency of right deep peroneal nerve (PDML) and standard gastrocnemius-soleus H reflex latency (T p ). We defined our exclusion criteria as: history of sacral radiculopathy or diabetes mellitus or any other disease with potential to cause neuropathy, any abnormality in neurological or musculoskeletal physical examination, or any of the following findings: PNCV less than 40 m/s, PDML more than 5 ms or prolonged T p (according to Braddom and Johnson's study [ 8 ]. Since we were supposed to rule out subclinical peripheral neuropathy and one component of the related electrodiagnostic study was measuring the distal motor latency for the deep peroneal nerve, the temperature at the dorsum of the foot was kept almost at 32°Celsius. The leg length of each person was measured as the distance from middle of the midpopliteal crease to the point at the most proximal part of the medial malleolus, in centimeters. Subject's age to the nearest year was also recorded. For obtaining T c , we used DANTEC 2000 c equipment, the sensitivity, sweep, and filter were set at: 0.2–1 mv/div , 5 ms/div , and 2–10,000 Hz respectively. The technique was the same as described in the Literature [ 1 , 2 ]. Briefly: the volunteers lied prone on the examining table with the feet off the edge of the plinth. The E 1 was placed at the middle of the line connecting midpoint of popliteal crease to the point at the most proximal part of the medial malleolus, and the E 2 over the Achilles tendon (both were surface electrodes). The ground electrode was posed proximal to E 1 and a disc electrode (anode) was placed on the anterior superior iliac spine. Then we inserted a monopolar 70 mm needle (cathode) at a point 1 cm medial to the posterior superior iliac spine, perpendicular to the frontal plane, and retracted it just a little after reaching the sacrum. Stimulus duration of 1 ms at 0.5 HZ was then applied while increasing current intensity to obtain both H and M waves simultaneously. M wave is the earlier wave and H is the later one. The interpeak latency was measured in milliseconds (ms) and recorded as T c . This measurement was only performed on the right lower extremity. Descriptive statistics were applied to depict Mean ± SD of age, leg length and Tc. The independent effect of leg length and age on T c was assessed by multiple regression model. The analyses were performed using SPSS 10.0 software. Kolmogrov-Smirnov test was used for evaluating the normal distribution of the variables. Results From 46 subjects who volunteered to participate in this study; five cases were excluded after history taking and physical examination (two because of history of sacral radiculopathy, two because of diabetes mellitus and one because of asymmetry in ankle reflexes) and one case after electrophysiologic evaluation; thus we completed the study with 40 subjects. Subjects' characteristics are shown in table 1 . Table 1 subjects' characteristics Number of subjects 40 (26 males, 14 females) Age range (years) 19–65 Leg length (centimeters) 29.5–43 T c ± SD 6.78 ± 0.3 The group consisted of 26 males (65%) and 14 females (35%). Kolmogrov-Smirnov test showed normal distribution of the variables. You are provided with the information below: (Mean ± SD) Age: 37.0 ± 10.7 years (range: 19–65); leg length: 36.4 ± 3.4 cm (range: 29.5–43); Tc= 6.78 ± 0.3 There was a significant correlation between T c and leg length (P value = 0.003, r= 0.49, CI 95% = 0.59–0.88). There was no correlation between T c and age (p value = 0.48, r = 0.11) We also found this regression equation: T c = 0.04L + 5.28 (L is leg length in centimeters, T c is represented in milliseconds.) Discussion In this study we found a significant correlation between leg length and Tc, but we were unable to show such a relation between age and T c . Pease et al were the first, studied T c [ 2 , 4 ], and reported Mean ± SD of 7 ± 0.3 ms which is very close to our results (Tc= 6.78 ± 0.3). They didn't specifically consider the leg length, age or any other potential confounding variables to T c . Zhu et al [ 3 ] evaluated 60 persons and reported Mean T c : 6.8 ms and its SD: 0.33 ms, again close to our results. They also reported that T c and person's height were correlated but didn't study any correlation between age and T c . Wang et al [ 5 ] evaluated 40 persons and found this regression equation: T c = 0.02A + 0.003H + 0.92 (H: Height and A: Age), and stated that age is a contributing factor on T c . Another research was performed by Ghavanini et al [ 6 ], in which 39 subjects were evaluated. The reported T c ± SD was 6.9 ± 0.4; two regression equations were also suggested: T c = 0.097Tp + 4.045 and Tc = 0.051L + 4.92 (L=leg length in centimeters); results are close to ours, and age was not found to affect T c . A summary of the above data plus detailed demographic data are provided in the table 2 . Table 2 comparing related studies Subjects' characteristics Tc Group size (persons) Mean age (yr) Mean L (cm) Mean H(cm) Mean SD Correlation (r) with A-H-L Present study 40 37.0 36.4 M:38.1 F:33.2 ? 6.78 0.3 No-?-0.49 Pease et al [2] 20 ? ? ? 7.0 0.3 ?-?-? Zhu et al [3] 60 43 ? 169 6.8 0.33 No-0.54-? Wang et al [5]* 40 ? ? ? ? ? ?-?-? Ghavanini et al [6]** 39 41 M:39.8 F:37.0 M:172.2 F:159.5 6.9 0.4 No-0.56–0.62 A: age (yr); L: leg length (cm); H: height (cm); Tc: central loop of the H-reflex latency (ms); M: male; F: female; No: no correlation was found; ?: not reported : suggested a regression equation: Tc = 0.02A+0.03H+0.92 *: Suggested two regression equations: Tc = 0.051L+4.928; Tc = 0.097Tp+4.04 Limitations In this study we focused on age and leg length as potential contributing factors on the T c . we didn't control, randomize or observe other possible confounding (contributing) factors with potential to affect this parameter. We observed a significant correlation between leg length and T c (P value = 0.003, r= 0.49, CI 95% = 0.59–0.88), that is compatible to a previous published work [ 3 ] (r = 0.54, p value less than 0.01). Had we found any association between T c and age, the question might have been raised that subclinical neuropathy of old age could have been contributive; obviously, this is not the case in our study. Although F-wave has been used to evaluate the possibility of proximal neuropathy; it was not measured in this study. Alternatively, we measured H-reflex latency to exclude proximal neuropathy [ 11 ]. It should be emphasized that noninvasive methodologies for the diagnosis of subclinical S1 radiculopathy are now available [ 12 ]. It is also acceptable to stimulate the S1 spinal nerve at the S1 foramen by magnet, instead of deep tissue needling; nevertheless, we used more popular techniques for this study. Conclusions We found that between age and leg length, only the latter can affect T c . It may be reasonable to consider leg length for calculating T c and to "narrow" the normal limits. Further studies with larger sample sizes are required for detecting other contributing factors and standardizing T c according to leg length. Competing interests None declared. Abbreviations T c : central loop of the gastrocnemius-soleus H-Reflex latency T p : gastrocnemius-soleus H-Reflex latency PNCV: right peroneal nerve conduction velocity PDML: right peroneal nerve distal motor latency Authors' contribution SS: examining the cases, calculation of TC, writing the paper. MRAG: suggesting the research, supervision and helping with calculation of TC. AA: examining the cases, calculation of TC PJ: statistical consultant (data analysis) Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517935.xml
520828	Aspergillus antigen induces robust Th2 cytokine production, inflammation, airway hyperreactivity and fibrosis in the absence of MCP-1 or CCR2	Background Asthma is characterized by type 2 T-helper cell (Th2) inflammation, goblet cell hyperplasia, airway hyperreactivity, and airway fibrosis. Monocyte chemoattractant protein-1 (MCP-1 or CCL2) and its receptor, CCR2, have been shown to play important roles in the development of Th2 inflammation. CCR2-deficient mice have been found to have altered inflammatory and physiologic responses in some models of experimental allergic asthma, but the role of CCR2 in contributing to inflammation and airway hyperreactivity appears to vary considerably between models. Furthermore, MCP-1-deficient mice have not previously been studied in models of experimental allergic asthma. Methods To test whether MCP-1 and CCR2 are each required for the development of experimental allergic asthma, we applied an Aspergillus antigen-induced model of Th2 cytokine-driven allergic asthma associated with airway fibrosis to mice deficient in either MCP-1 or CCR2. Previous studies with live Aspergillus conidia instilled into the lung revealed that MCP-1 and CCR2 play a role in anti-fungal responses; in contrast, we used a non-viable Aspergillus antigen preparation known to induce a robust eosinophilic inflammatory response. Results We found that wild-type C57BL/6 mice developed eosinophilic airway inflammation, goblet cell hyperplasia, airway hyperreactivity, elevations in serum IgE, and airway fibrosis in response to airway challenge with Aspergillus antigen. Surprisingly, mice deficient in either MCP-1 or CCR2 had responses to Aspergillus antigen similar to those seen in wild-type mice, including production of Th2 cytokines. Conclusion We conclude that robust Th2-mediated lung pathology can occur even in the complete absence of MCP-1 or CCR2.	Background Monocyte chemoattractant protein-1 (MCP-1, also known as CCL2) and its receptor, CCR2, have been the focus of intense interest due to increasing awareness of their association with debilitating human diseases, including asthma [ 1 - 3 ] and pulmonary fibrosis [ 4 - 7 ]. Since MCP-1 attracts and activates a variety of cells, including monocytes, immature dendritic cells, basophils, natural killer cells, and a subset of T lymphocytes [ 8 - 17 ], MCP-1 may have multiple roles in the immune response. Models of Th1 or Th2 inflammation applied to mice deficient in either MCP-1 or CCR2 have clearly shown important roles for this chemokine and its receptor in the development of inflammation [ 18 - 24 ]. However, results obtained using allergen-induced models of asthma (ovalbumin and cockroach antigen) in CCR2-deficient mice are varied, showing either increased, decreased or unchanged Th2 inflammation and airway hyperreactivity (AHR) [ 25 - 27 ], possibly due to differences in the allergen models or strains of mice used. These experiments with CCR2-deficient mice do not directly address the role of MCP-1, which is just one of several MCP chemokines that can bind to CCR2. Although MCP-1-deficient mice have been reported to have defects in Th2 responses [ 18 , 19 ], the effects of MCP-1 deletion in allergen-induced allergic experimental asthma have not been previously reported. In addition to Th2 inflammation, airway fibrosis is another important feature of human asthma. Blease and colleagues [ 28 , 29 ] examined the contributions of MCP-1 and CCR2 to the development of fibrosis following intratracheal administration of Aspergillus fumigatus conidia to A. fumigatus sensitized mice. Airway fibrosis was significantly increased in mice treated with MCP-1 neutralizing antibody and in CCR2-deficient mice. However, these increases in fibrosis were seen in the setting of impaired clearance of conidia and a markedly increased neutrophilic inflammatory response, suggesting that the increased fibrosis might be attributable simply to an impaired antifungal response. Previous studies involving other models of allergic asthma applied to CCR2-deficient mice did not examine whether airway fibrosis occurred in these models or whether development of fibrosis was dependent on CCR2 expression [ 25 - 27 ]. Consequently, the role of MCP-1 and CCR2 in the development of allergen-induced lung fibrosis is not well established. In this study, we hypothesized that the effects of MCP-1 are mediated through CCR2 and that MCP-1 and CCR2 are independently required for the development of experimental allergic asthma. To test this hypothesis, we subjected mice deficient in either MCP-1 or CCR2 to an Aspergillus antigen model of Th2-cytokine-driven allergic asthma associated with significant airway fibrosis and measured pulmonary inflammation, cytokine production, AHR and fibrosis. Methods Mice Breeding pairs of Mcp-1 +/+ and Mcp-1 -/- mice [ 19 ] and Ccr2 +/+ and Ccr2 -/- mice [ 21 ] were generated as previously described. Mice were bred and maintained under specific pathogen-free conditions in the Laboratory Animal Resource Center at San Francisco General Hospital. All mice were backcrossed nine times with C57BL/6 mice (Jackson Laboratory, Bar Harbor, ME). Deletion of Mcp-1 or Ccr2 genes was confirmed by PCR. Similar numbers of male and female six-week-old mice were used for the study. The UCSF Institutional Animal Care and Use Committee approved all experimental protocols. Aspergillus Antigen Sensitization Protocol The Aspergillus fumigatus antigen preparation consisted of a mixture of culture filtrate (300 μg protein/mouse) and mycelial extract (80 μg protein/mouse) in PBS (Cellgro by Mediatech, Inc, Herndon, VA). Culture filtrates and mycelial extract were prepared as described previously [ 30 ]. For sensitization, anesthetized six-week old mice were given 50 μl of Aspergillus antigen intranasally five times at four-day intervals. Control mice were given 50 μl of PBS according to the same schedule as Aspergillus antigen-treated mice. All measurements and samples were obtained from mice four days after the final Aspergillus antigen administration, which was 20 days after the first challenge. Our group has previously found that airway reactivity measured four days after the final Aspergillus antigen challenge was similar to reactivity measured at earlier time points (on the same day as the final challenge or one day after the final challenge) [ 30 ]. Determination of Airway Reactivity Mice were anesthetized and paralyzed by intraperitoneal injection of etomidate (28 mg/kg) (Bedford Laboratories, Bedford, OH) and pancuronium bromide (0.1 mg/kg) (Baxter Healthcare Corporation, Irvine, CA). A tracheal cannula was inserted via a midcervical incision and the mice were ventilated using a Harvard model 683 rodent ventilator (9 μl/g tidal volume, 150 breaths per minute) (Harvard Apparatus, Holliston, MA). Using a whole body plethysmograph, airflow resistance was calculated during baseline breathing and in response to serially increasing doses of intravenous acetylcholine chloride (0.032, 0.100, 0.316, 1.00, and 3.16 μg/gm body weight) (Sigma, St. Louis, MO). The log of the concentration of acetylcholine (μg/gm) required for a 200% increase in total lung resistance, designated log PC 200 , was reported. Bronchoalveolar Lavage (BAL) After completion of the airway physiology measurements, the lungs were lavaged five times with 0.8-ml aliquots of sterile PBS. The lavage fluid was pooled and centrifuged, and the cell pellet was treated with red-blood-cell lysing buffer (Sigma, Saint Louis, MO). After being washed, the samples were resuspended in PBS. Total leukocytes were counted using a hemacytometer. Differential cell counts were determined by cytocentrifugation and Diff-Quik staining (Dade Behring Inc., Newark, DE) followed by microscopic examination of at least 300 cells. Thoracic Lymph Node Isolation and Lung Histology Thoracic lymph nodes were harvested from mice exposed to Aspergillus antigen. Lungs were then removed en bloc and the left mainstem bronchus was firmly sutured closed. The left lung was removed by cutting the left mainstem distal to the suture. It was then frozen in liquid nitrogen and stored at -70°C until processed for hydroxyproline content. The right lung was inflated to 20 cm water pressure with 10% neutral buffered formalin (VWR Scientific Products, West Chester, PA) and fixed in 10% formalin for more than 48 h. Fixed lungs were embedded in paraffin, sectioned at 5 μm thickness, and stained with either hematoxylin and eosin (H&E), periodic acid Schiff (PAS), or trichrome by the Pathology Department of San Francisco General Hospital using standard protocols. The proportion of peribronchial inflammatory cells that were eosinophils was determined by counting inflammatory cells surrounding airways with lumens of 100–200 μm (measured on the short axis) on H&E stained sections. We analyzed 500 total cells (100 cells from each of five airways) for each animal studied. Analysis of Cytokine Production by Cells To prepare single-cell suspensions for cytokine analyses, isolated lymph nodes were gently minced using a syringe plunger and cells were passed through 70-μm cell strainers. Red blood cells were removed by hypotonic lysis at room temperature. Lymph node cells were counted, centrifuged, and resuspended in RPMI medium 1640 (Cellgro by Mediatech, Inc, Herndon, VA) supplemented with FCS (10% vol/vol) (Hyclone, Logan, UT), penicillin (100 U/ml) (Cellgro by Mediatech, Inc, Herndon, VA), streptomycin (100 μg/ml) (Cellgro by Mediatech, Inc, Herndon, VA), phorbol 12 myristate 13-acetate (PMA) (25 ng/m) (Sigma, Saint Louis, MO), and ionomycin (1 μg/ml) (Sigma, Saint Louis, MO) to a final concentration of 5 million cells per ml. Cells were then aliquoted into 96-well plates and incubated at 37°C. After 40 h, cell supernatants were harvested and stored at -70°C until they were analyzed. ELISA for IL-4, -5, -13 and IFN-γ was performed on stimulated lymph node cell supernatant per the manufacturer's protocols (R&D Systems, Minneapolis, MN). Determination of MCP-1 For quantitation of MCP-1 levels in BAL fluid, C57BL/6 wild-type mice were treated with the previously described Aspergillus antigen protocol. Four days after the final Aspergillus antigen administration, lungs from Aspergillus antigen- and PBS-treated mice were lavaged two times with 0.6-ml aliquots of sterile PBS. The samples were centrifuged and the supernatants were collected and stored at -70°C until analysis. ELISA for MCP-1 was performed on cell-free BAL fluid per the manufacturer's protocol (R&D Systems, Minneapolis, MN). Measurement of Serum Total IgE Concentration Sera were obtained from blood collected by cardiac puncture from Aspergillus antigen- or PBS-treated mice after airway responsiveness measurement. Serum total IgE concentration was determined by a sandwich ELISA using complementary antibody pairs for mouse IgE (clone R35-72 and R35-118) obtained from Pharmingen (Pharmingen, San Diego, CA) according to the manufacturer's instructions. Color development was achieved using streptavidin-conjugated horseradish peroxidase (Pharmingen, San Diego, CA) followed by addition of HRP substrate (ABTS, Sigma, Saint Louis, MO). Determination of Lung Hydroxyproline Content Lungs were analyzed for hydroxyproline content as previously described [ 31 ] with slight modification. Lungs were homogenized in distilled water and incubated with 50% trichloroacetic acid on ice for 20 min. Samples were centrifuged and the pellet was mixed with 12 N hydrochloric acid and baked at 110°C for 14–18 h until samples were charred and dry. The samples were resuspended in 2 ml deionized water by incubating for 72 h at room temperature applying intermittent vortexing. Serial dilutions of trans-4-hydroxy-L-proline standard (Sigma, Saint Louis, MO) were prepared. 200 μl of vortexed sample (or standard) was added to 500 μl 1.4% chloramine T/0.5 M sodium acetate/10% isopropanol (Fisher Scientific, Pittsburgh, PA) and incubated for 20 min at room temperature. Next, 500 μl of Ehrlich's solution (1.0 M p-dimethylaminobenzaldehyde, 70% isopropanol/30% perchloric acid) (Fisher Scientific, Pittsburgh, PA) was added, mixed, and incubated at 65°C for 15 min. After samples returned to room temperature, the optical density of each sample and standard was measured at 550 nm and the concentration of lung hydroxyproline was calculated from the hydroxyproline standard curve. Statistical Analysis Statistical significance for treatment effect was determined by analysis of variance with post-ANOVA t tests corrected for multiple comparisons using Bonferroni adjustment. These statistical analyses were performed using statistical software STATA 5.0 (Stata Corporation, College Station, TX) and R [ 32 ] (The R Foundation for Statistical Computing, Vienna, Austria). All tests were two-tailed with a p-value of 0.05 for statistical significance. Results Aspergillus antigen airway challenge induces MCP-1 production We used a model system that involved repeated intranasal challenges with Aspergillus antigen over a 20-day period. To determine whether antigen challenge induces MCP-1 production in the airway, we measured MCP-1 protein levels in BAL fluid from wild-type C57BL/6 mice on day 20. MCP-1 levels were markedly higher in Aspergillus antigen-treated mice (46.3 ± 12.7 pg/ml, mean ± SE) than in PBS-treated mice (5.8 ± 1.3 pg/ml), (P = 0.01). MCP-1- and CCR2-deficient mice develop airway inflammation in response to Aspergillus antigen The Aspergillus antigen induction of MCP-1 was accompanied by a significant degree of lung inflammation as assessed by BAL fluid cell counts and lung histology. In wild-type mice, Aspergillus antigen induced a >20-fold increase in BAL fluid cell numbers (Fig. 1 ) and the development of prominent infiltrates in peribronchovascular spaces and scattered infiltrates in the lung parenchyma (Fig. 2A and 2B ). The inflammatory infiltrates consisted of numerous eosinophils as well as other cell types. Figure 1 Aspergillus antigen induced similar increases in BAL fluid cell counts in wild-type, Mcp-1 -/- and Ccr2 -/- mice. Total cells, macrophages, eosinophils, and lymphocytes are expressed as mean BAL fluid total cell counts ± SE from wild-type, Mcp-1 -/- and Ccr2 -/- mice (PBS-treated, N = 5 mice/group; Aspergillus antigen-treated, N = 8 mice/group; Aspergillus antigen exposure and sample collection are described in methods). Neutrophils represented <0.5% of total cells for all groups. The data shown are from one experiment and representative of three separate experiments. Asterisks () indicate values that are statistically significantly different (p < 0.001) compared to PBS controls. To determine the airway inflammatory response to Aspergillus antigen in the absence of MCP-1 or its receptor, CCR2, we used mice with targeted disruptions of the genes that encode MCP-1 and CCR2. Since mouse strain differences are associated with major differences in antigen reactivity in many model systems, the mice used here were produced by extensive backcrossing into a C57BL/6 genetic background. Both MCP-1- and CCR2-deficient mice developed marked airway inflammation in response to Aspergillus antigen (Figs. 2C and 2D ). The BAL fluid cell counts from Aspergillus antigen-treated MCP-1- and CCR2-deficient mice revealed significantly greater numbers of all cell types than in PBS-treated controls (p < 0.001). The numbers of macrophages, lymphocytes and neutrophils were not significantly different from those in Aspergillus antigen-treated wild-type mice (Fig. 1 ). The BAL fluid eosinophil response in MCP-1- and CCR2-deficient mice was slightly (~30–40%) smaller than in wild-type mice, but this difference did not reach statistical significance (Fig. 1 ). The fraction of peribronchial inflammatory cells that were eosinophils was not significantly different among wild-type mice (51 ± 13%, mean ± standard deviation), CCR2-deficient mice (52 ± 6%), and MCP-1-deficient mice (37 ± 13%) (N = 5 mice/group). These findings indicate that there was a robust inflammatory response to Aspergillus antigen even in the absence of MCP-1 or CCR2. Figure 2 Aspergillus antigen-induced lung inflammation appears similar in wild-type, Mcp-1 -/- and Ccr2 -/- mice. H&E stained lung sections from PBS- or Aspergillus antigen-treated wild-type, Mcp-1 -/- and Ccr2 -/- mice. Representative normal airway from wild-type control mice (A) (similar findings from Mcp-1 -/- and Ccr2 -/- control mice are not shown). Representative lung sections from Aspergillus antigen-treated wild-type (B), Mcp-1 -/- (C) and Ccr2 -/- mice (D) demonstrate intense peribronchiolar and perivascular inflammation. Aspergillus antigen exposure and sample collection are described in methods. Magnification: 20× objective. MCP-1- and CCR2-deficient mice develop AHR and produce mucus in response to Aspergillus antigen To determine airway reactivity to acetylcholine in mice exposed to Aspergillus antigen or to vehicle (PBS) alone, we compared airway reactivity of PBS- and Aspergillus -antigen-treated mice 4 days after the final challenge as described in the methods section. Measurements from this time point were previously found to be comparable to those from earlier time points [ 30 ]. In the experiment shown in Fig. 3 , the PBS-treated group included a mixture of wild-type, Mcp-1 -/- , and Ccr2 -/- mice since preliminary experiments showed similar airway reactivity between PBS-treated wild-type, Mcp-1 -/- , and Ccr2 -/- mice (not shown). Aspergillus -antigen-treated wild-type, Mcp-1 -/- , and Ccr2 -/- mice each had significantly lower PC 200 values than did PBS-treated controls (P < 0.001), indicating the development of AHR (Fig. 3 ). Although there appeared to be a trend toward less airway reactivity in Aspergillus -antigen-treated Mcp-1 -/- and Ccr2 -/- mice than in Aspergillus -antigen-treated wild-type mice, this trend was not statistically significant and was not observed in two additional Aspergillus -antigen-challenge experiments comparing wild-type mice to either Mcp-1 -/- or Ccr2 -/- mice separately (data not shown). Figure 3 Aspergillus antigen induced AHR in wild-type, Mcp-1 -/- and Ccr2 -/- mice. Airway reactivity in response to intravenous acetylcholine was measured invasively. Data are expressed as log PC 200 and lower values indicate higher airway response. Aspergillus antigen exposure and the airway measurement protocol are described in methods (PBS-treated, N = 12 mice; Aspergillus antigen-treated, N = 8–10 mice/group;). The data shown are from one experiment and representative of three separate experiments. Asterisks () indicate values that are statistically significantly different (p < 0.001) compared to PBS controls. To determine if Aspergillus -antigen challenge results in increased mucus production, we analyzed lung histology by PAS-staining. As shown in Fig. 4A , there was minimal PAS staining in the airway epithelium of control mice. In contrast, Aspergillus -antigen-treated mice from all three groups showed accumulation of PAS-stained material in epithelial cells (Fig. 4B , 4C , 4D ), indicating that Aspergillus antigen airway challenge resulted in mucus production by goblet cells. These findings indicate that Aspergillus antigen induces AHR and mucus production even in the absence of MCP-1 or CCR2. Figure 4 Aspergillus antigen induced goblet cell hyperplasia in wild-type, Mcp-1 -/- and Ccr2 -/- mice. Representative PAS-stained lung sections from PBS-treated wild-type mice (A) showed minimal PAS-positive staining (similar findings from Mcp-1 -/- and Ccr2 -/- control mice are not shown). Aspergillus antigen-treated wild-type (B), Mcp-1 -/- (C) and Ccr2 -/- mice (D) showed magenta staining in epithelial cells, which represents mucus. Aspergillus antigen exposure and sample collection are described in methods. Magnification, 40× objective. Th2 cytokine and IgE production is similar in Aspergillus antigen-treated wild-type, Mcp-1 -/- and Ccr2 -/- mice To determine if deletion of MCP-1 or CCR2 alters the cytokine response to Aspergillus antigen, we assayed Th1 and Th2 cytokines in stimulated cell supernatants prepared from thoracic lymph nodes isolated from Aspergillus antigen-treated mice. (PBS-treated mice had much smaller thoracic lymph nodes and it was not possible to reliably obtain sufficient numbers of cells from these mice for comparison.) MCP-1- and CCR2-deficient mice had concentrations of the cytokines IL-4, IL-5, IL-13 and IFN-γ generally similar to those in wild-type mice (Fig. 5A , 5B , 5C , 5D ). There was a trend toward lower IL-4 production in cells from Ccr2 -/- mice, but this difference was not statistically significant. In addition, sera from Aspergillus -antigen-treated mice and control mice were assayed for serum total IgE levels. As shown in Fig. 5E , Aspergillus antigen induced increases in serum IgE in wild-type, Mcp-1 -/- , and Ccr2 -/- mice similar to those in control mice. Figure 5 Aspergillus antigen-treated wild-type, Mcp-1 -/- and Ccr2 -/- mice demonstrated intact Th2 cytokine production and induction of IgE. For cytokine determination, draining lymph node cells from Aspergillus antigen-treated wild-type, Mcp-1 -/- and Ccr2 -/- mice were isolated and stimulated with PMA/ionomycin for 40 hr and cytokine levels for IL-4 (A), IL-5 (B), IL-13 (C), and IFN-γ (D) were quantitated by ELISA. Serum IgE (E) from Aspergillus antigen-treated wild-type, Mcp-1 -/- and Ccr2 -/- mice and control mice were measured by ELISA. In (A-D), bars represent mean ± SE; in (E), results are expressed as the common log of IgE concentration where each circle represents a single PBS- or Aspergillus antigen-treated mouse and horizontal lines represent the mean of each group (PBS-treated, N = 5 mice/group; Aspergillus antigen-treated, N = 8–9 mice/group). Aspergillus antigen exposure and sample collection are described in methods. Asterisks () indicate values that are statistically significantly different (p < 0.001) compared to PBS controls. Aspergillus antigen-induced lung fibrosis develops in the absence of MCP-1 or CCR2 To determine whether Aspergillus antigen-induced airway fibrosis develops in the absence of MCP-1 or CCR2, we measured lung hydroxyproline content in PBS- and Aspergillus -antigen-challenged mice (Fig. 6 ). Aspergillus antigen treatment resulted in a two-fold increase in lung hydroxyproline, a measure of collagen content. This effect was very similar in wild-type, Mcp-1 -/- , and Ccr2 -/- mice. Histopathologically, lung sections from PBS-treated mice had normal lung architecture and minimal evidence of trichrome staining (Fig. 7A ). Lung sections from mice treated with Aspergillus antigen had clear increases in trichrome staining in a peribronchiolar distribution (Fig. 7B , 7C , 7D ). There were no apparent differences in trichrome staining in wild-type mice as compared to either MCP-1- or CCR2-deficient mice after allergen challenge. Figure 6 Aspergillus antigen induced similar lung fibrosis in wild-type, Mcp-1 -/- and Ccr2 -/- mice. Left lungs from Aspergillus antigen- or PBS-treated wild-type, Mcp-1 -/- and Ccr2 -/- mice were analyzed for total hydroxyproline content as described in methods. Results are expressed as mean ± SE (N = 10 mice/group). Aspergillus antigen exposure and sample collection are described in methods; data are representative of two separate experiments. Asterisks () indicate values that are statistically significantly different (p < 0.001) compared to PBS controls. Figure 7 Increased airway subepithelial collagen deposition after treatment with Aspergillus antigen. Representative lung sections from PBS-treated mice show minimal trichrome staining around small airways (A) (similar findings from Mcp-1 -/- and Ccr2 -/- control mice are not shown). Increased trichrome staining is noted around small airways in Aspergillus antigen-treated wild-type (B), Mcp-1 -/- (C) and Ccr2 -/- (D) mice. Blue staining around airways represents collagen. Aspergillus antigen exposure and sample collection are described in methods. Magnification, 20× objective. Discussion We hypothesized that MCP-1 and its receptor, CCR2, are independently required for the development of Aspergillus -antigen-induced allergic asthma. We found that wild-type C57BL/6 mice challenged with Aspergillus antigen developed robust Th2 responses associated with pulmonary inflammation, AHR, mucus production and fibrosis. Surprisingly, neither MCP-1 nor CCR2 was critical for the development of these lung pathologies, since robust responses were also seen in mice with deletions of genes encoding either protein. These results demonstrate that neither MCP-1 nor CCR2 are required for the development of experimental allergic asthma induced by exposure to Aspergillus antigen. Our results stand in contrast to some previous reports showing important roles for MCP-1 or CCR2 in other models of allergic asthma [ 25 , 27 , 33 ]. Although the precise explanation of these differences is not clear, there are several experimental factors that may contribute. For example, the choice of antigen and the route of sensitization differ between models. We used antigens prepared from Aspergillus , an important allergen in some people with asthma, and administered it exclusively to the respiratory tract, presumably a relevant route for sensitization in asthma. Previous studies have used ovalbumin [ 25 , 26 , 33 ] or cockroach antigen [ 27 ] and have used intraperitoneal antigen injections to sensitize prior to antigen challenge. CCR2-deficient mice have been shown to have defects in recruitment of antigen-presenting cells to the peritoneum [ 21 , 34 , 35 ], suggesting that CCR2 could be important for sensitization when antigen is administered to the peritoneum. Another factor that differs between studies is timing. We studied mice at 4 days after the final allergen challenge, when all aspects of the Aspergillus antigen-induced experimental asthma phenotype are present. Campbell et al. found that the administration of MCP-1 antibody could inhibit AHR in cockroach antigen sensitized and challenged mice at very early time points (1 and 8 h post challenge) but not later (24 h after challenge) [ 27 ]. The effect on AHR at 1 and 8 h was ascribed to MCP-1's ability to activate mast cells, which are important in some asthma models but not in others [ 36 ]. Genetic background may also be an important factor, since mouse strains vary widely in their response to airway antigen challenge [ 37 ]. Previous experimental asthma studies involving CCR2-deficient mice have used mice of mixed genetic backgrounds [ 25 - 27 ], whereas we used mice that had been backcrossed nine times to C57BL/6 and therefore have a more homogenous genetic background. Some of the specifics of our experimental system may therefore account for the lack of a requirement for MCP-1 and CCR2. However, MacLean et al. [ 26 ] used an allergic asthma model involving ovalbumin, intraperitoneal sensitization, and mice of mixed genetic backgrounds and found that CCR2-deficient mice had intact responses to allergen challenge. This indicates that the lack of a requirement for CCR2 is not unique to a single asthma model. It also highlights the difficulty in pinpointing the experimental factors that account for the diverse results reported by various investigators. Of note, neither MCP-1 nor CCR2 was critical for inflammatory cell migration to the lungs after Aspergillus antigen challenge. We found that Aspergillus antigen-induced monocyte recruitment (as measured by counting BAL fluid macrophages) was intact in both MCP-1- and CCR2-deficient mice. While intact alveolar macrophage recruitment in response to airway instillation of Saccharopolyspora rectivirgula has been reported in CCR2-deficient mice [ 38 ], other in vivo models have demonstrated requirements for MCP-1 and CCR2 in monocyte/macrophage recruitment [ 19 , 39 - 42 ]. Our finding indicates that other chemoattractants are sufficient for maximal monocyte/macrophage recruitment in this Aspergillus antigen model. In support of this observation, a recent microarray-based analysis of gene expression changes in a similar asthma model found that 14 different chemokines (including MCP-1/JE) were induced by Aspergillus antigen challenge [ 43 ]. However, we did find that MCP-1 and CCR2 may have indirect effects on eosinophil recruitment in response to Aspergillus antigen. While there was marked eosinophil recruitment to the lungs in MCP-1- and CCR2-deficient mice, there was a trend toward fewer eosinophils than in wild-type mice. Since MCP-1 is not a chemoattractant for eosinophils (which lack CCR2), this trend suggests that MCP-1 may have indirect effects on eosinophil recruitment in this model. A more dramatic decrease of eosinophil recruitment has been seen following neutralization of MCP-1 in another model, but that effect was associated with other signs of impaired Th2 immunity [ 33 ]. Although there may be some role for MCP-1 and CCR2 in eosinophil recruitment, robust inflammatory responses to Aspergillus antigen occurred even in the complete absence of either of these molecules. In contrast to our results indicating a robust Th2 response in MCP-1- and CCR2-deficient mice after Aspergillus antigen challenge, diminished Th2 cytokine production has been reported in studies of MCP-1 neutralization or deletion in different models [ 19 , 20 , 33 , 44 , 45 ]. In studies involving CCR2-deficient mice, the results have been more heterogenous, suggesting that CCR2 deletion may increase [ 25 , 28 ], decrease [ 24 ], or have no effect on Th2 responses [ 26 ]. As mentioned previously, the explanation for these different Th2 responses in CCR2-deficient mice is not clear, and may suggest that complex pathways involving other CCR2 ligands or MCP-1 receptors [ 46 ] are operational in different models of inflammation. However, if these pathways exist and were important in the model we used, we would have expected to find that deletion of MCP-1 and CCR2 had different effects. Instead, we observed that MCP-1- and CCR2-deficient mice were similar in all respects, including cytokine production, IgE production, and AHR. Our results support the idea that the role of MCP-1 and CCR2 in the development of allergic responses may be dependent upon the experimental model used. The role of MCP-1 and CCR2 in the development of allergen-induced airway fibrosis has not been extensively explored. Previous findings of increased pulmonary fibrosis in CCR2-deficient mice compared to wild-type mice after treatment with Aspergillus conidia were accompanied by neutrophilic inflammation and the inability of CCR2-deficient mice to clear the organism normally [ 28 , 29 ]. Consequently, the persistence of Aspergillus organisms in the airway may have altered the fibrotic response. Other studies involving different experimental systems have suggested that MCP-1 and CCR2 may directly or indirectly contribute to the development of fibrosis. Gharaee-Kermani et al. [ 47 ] found that MCP-1 directly induced increased production of collagen by cultured fibroblasts, although the role of CCR2 was not explored in that report. MCP-1 and CCR2 may also indirectly influence fibrosis via their effects on inflammatory cells. Previous studies showed that CCR2-deficient mice developed less pulmonary fibrosis in response to three different stimuli, including intratracheal bleomycin instillation, than did wild-type mice [ 48 , 49 ]; however, those studies did not test the requirement for MCP-1 in the development of fibrosis. In C57BL/6 mice, bleomycin induces a robust inflammatory response that consists of neutrophils and lymphocytes, with a smaller component of eosinophils [ 50 ], in contrast to our allergen model. Thus, it is possible that the relative abundance or types of recruited cells in response to a particular airway challenge greatly influence the character or extent of lung fibrosis mediated by MCP-1 or CCR2.. Therefore, based on these previously published results we might have expected MCP-1 and CCR2 to be critical to the development of allergen-mediated fibrosis. However, we found that MCP-1-deficient and CCR2-deficient mice each developed marked fibrosis following Aspergillus antigen challenge, similar to wild-type mice. Our result, in contrast to the reported requirement for CCR2 in the development of bleomycin-induced pulmonary fibrosis, suggests that different cell types and mediators may be operational in allergen-induced airway fibrosis than those observed in bleomycin-induced lung fibrosis. Conclusions In conclusion, this study demonstrates that pulmonary inflammation, Th2 immune responses, Th2-mediated airway pathology, and lung fibrosis are remarkably intact despite the complete absence of MCP-1 or CCR2 in an Aspergillus antigen-driven model of allergic airway disease. Previous studies have demonstrated roles for MCP-1 and CCR2 in other models of inflammation and fibrosis, including different allergic airway disease models [ 25 , 27 , 33 ]. Those findings indicate that the role of MCP-1 and CCR2 in allergic responses and in fibrosis depends on the models used, although it is difficult to identify which experimental factors determine whether MCP-1 and CCR2 are required. Both MCP-1 and CCR2 may be good therapeutic targets for some diseases. However, the variable involvement of these potential targets in animal models indicates that it may be extremely challenging to predict which human diseases are most likely to benefit from this approach. Abbreviations AHR, airways hyperreactivity; BALF, bronchoalveolar lavage fluid. Authors' contributions LLK conceived of the experiment, carried out all experiments and prepared the manuscript. MWR assisted in collection and analysis of mouse samples. XLB performed all mouse airway measurements. SC and XH performed antigen challenge and assisted in collection and analysis of mouse samples. IFC and BJR provided the targeted knock-out mice, provided expert advice and interpretation of the study's results. DJE participated in the study's design, coordination and final revisions of the manuscript. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC520828.xml
545939	The 1999 international emergency humanitarian evacuation of the Kosovars to Canada: A qualitative study of service providers' perspectives at the international, national and local levels	Background In response to the Kosovo crisis, Canada received 5,500 Albanian Kosovar refugees in 1999 as part of the emergency humanitarian evacuation and settlement effort. This study attempts to describe the experiences of service providers at the international, national, and local levels, involved in the organization and delivery of health and settlement services in Canada for the Kosovar refugees. Methods A qualitative case study design using key informant interviews was used. Nominated sampling was used to identify 17 individuals involved in the organization and delivery of health and settlement. Key themes were identified and recommendations made to provide a framework for the development of policy to guide response to future humanitarian emergencies. Results Six themes emerged: (1) A sense of being overwhelmed, (2) A multitude of health issues, (3) critical challenges in providing health care, (4) access to health and settlement services, (5) overall successes and (6) need for a coordinated approach to migration health. Conclusions For those involved, the experience was overwhelming but rewarding. Interviewees' major concerns were the need for a more comprehensive and coordinated approach to the flow of medical information and handling of specific health problems.	Background The Kosovo crisis resulted in the largest population displacement in Europe since World War II. During this time, over 300,000 Albanian Kosovars were expelled from Kosovo, [ 1 ] resulting in a Complex Humanitarian Emergency. (CHE). The United Nations High Commission for Refugees (UNHCR) managed a Humanitarian Evacuation Program (HEP) to move the Kosovars to new international destinations, including Canada. Canada received approximately 5,500 refugees in 1999, with approximately 500 arriving to the city of Hamilton, in southern Ontario. Some Kosovars stayed temporarily in military bases and arrived after sponsorship housing had been arranged, while others arrived directly to temporary reception "houses" established at two Hamilton hotels. The reception in Hamilton included the organized provision of health services by physicians, nurses, dentists and optometrists, and settlement services from the Settlement and Integration Services Organization (SISO). This HEP was unique for many countries, including Canada, and involved a complex coordinated effort on the part of international, national, and local organizations traditionally involved with immigrants and refugees, settlement, and health. This event was an important opportunity for health and settlement organizations operating at international, national, and local levels to critically examine and reflect on their efforts to organize and deliver health and settlement services to new immigrants and refugees. The goal of this study was to explore the main challenges and successes of the Kosovar arrival, from international, national, and local perspectives and to develop recommendations to advise and guide planning of the future complex humanitarian emergencies, using interviews with key informants. Methods A qualitative case study methodology was used with semi-structured interviews of key informants who were involved in the humanitarian evacuation at the international, national and local levels in Hamilton, Canada. This method was deemed most appropriate to provide a framework for understanding the challenges and successes associated with Complex Humanitarian Emergencies and generate recommendations for future efforts. The background of research team members was diverse. Medicine, nursing, anthropology, sociology and settlement were represented. Some members had been directly involved with the Kosovar settlement process at the local level (NF, SR, MJ). Sampling A purposive sampling strategy was used, in which appropriate key informants known to the researchers were approached for participation first. Snowball sampling, whereby additional participants were identified by the initial respondents, was used to increase the diversity of respondents. These sampling approaches are used when certain known individuals are likely to have in-depth information on a topic [ 2 ]. Individuals from agencies involved with the 1999 Kosovar HEP were included. There was a very specific attempt to sample individuals from the international to the local levels even if they had no direct association with the Hamilton group. The sampling included three Health Canada officials who were known to be involved at the international and national levels and two Citizenship and Immigration Canada (CIC) officials involved at the provincial and local levels. Also sampled were two representatives from the Department of Social and Public Health Services in Hamilton, five local healthcare professionals (one family physician, two nurse practitioners, one dentist, one optometrist), and five local settlement providers. All of the CIC and the local key informants were referred and identified by the local settlement agency SISO as groups that had been integral to the Kosovo efforts in Hamilton. The international and national key informants were identified through Health Canada and internationally as people who were involved in the effort. An initial group of nine key informants identified eight additional participants. All 17 key informants approached, agreed to participate. Interviews Interviews were conducted in English between April and July 2001. The interviewer (EM) took extensive field notes to supplement the taped and transcribed interviews. Of the 17 key informant interviews, the three international and national ones were conducted by telephone, and 14 were in person at the workplace of the interviewee. Six participants were male and 11 were female. Interviews were approximately 30 minutes in length. The interviews were based on an interview guide and the main part of the interview focused on the organization and delivery of health services to the Kosovars. Questions were further refined during the study. Specific probes were used to follow up on open-ended questions, where appropriate. Consultation with the local settlement agency (SISO) took place throughout the research process. Key informants were told that the purpose of the study was to gather information that would help in planning for similar events in the future. All participants provided verbal consent and were assured of confidentiality of responses. All interviews were tape recorded with the participants' permission and all but one (due to technical difficulty) was transcribed. Ethics approval for the study was received from McMaster University Research Ethics Board. Questions pertained to the participants' involvement in the process, challenges, flow of information and communication, and involvement in health care services and settlement. Data Analysis Two research team members (EM and NF) initially independently reviewed the transcripts, coded categories and themes, and then compared results. A third researcher with expertise in qualitative research, who was not involved in the project, also reviewed the transcripts using NVivo software version 2.0 (QSR International Pty Ltd, Melbourne, Australia) and compared results. The interviewer's field notes were also used for comparison with the data. Ambiguities were resolved and themes were developed from categories through discussion among the research group members and re-reading of transcripts. This process continued until no new themes emerged from the interviews. It was felt that saturation was achieved after 17 interviews, and data collection was ended. The process was iterative, thus the later interviews probed new emerging themes identified in previous interviews. Results Six core themes emerged from the data analysis. Theme 1: A Sense of Being Overwhelmed There was agreement among the participants that the Kosovo crisis in 1999 was very unusual in nature. Participants stressed that there was a lot of "scrambling" to prepare the infrastructure to receive a large number of people in a very short period of time. Respondents/informants were surprised by the magnitude, immediacy, scope and scale of the response required. Most described feeling overwhelmed. Although the local settlement agency (SISO) had begun preparations by meeting with various local agencies and organizations weeks in advance, they were given only three days notice by national immigration authorities that the Kosovars were arriving locally. One Hamilton Social and Public Health Services representative noted "We had many community people come together to talk about how we could plan for this big influx of refugees... we knew were in dire straits". One settlement worker called it "organized chaos". One local health care professional explained the triage process and it was evident that a major organizational effort was required to process the arrivals. "We had to do our own triage at the hotel to find out whether there was anybody needing medications, anyone with heart conditions, anyone with diabetes ... we had to scan the place to find out, and even then we didn't have medical records that came with them... We just went around asking: Do you know about anybody who is pregnant? Do you know about anybody who has heart conditions? Do you know about anybody who is diabetic? ... And we tried to identify those people or they presented in the little clinic". Theme 2: A Multitude of Health Issues Participants described facing a multitude of challenges in responding to the health needs of Kosovar refugees. Women's health services, including pregnancy care and abortion services were required. Dental care, mental health services and general curative services were also required. Participants were struck by the poor oral health of the Kosovars. Many Kosovars had been deprived of any curative or preventative care in Kosovo for a number of years prior to the 1999 crisis. One local health care professional observed, "I noticed with the Kosovars...that a lot of them had not had any health care for a long time so they had many of the things that we take for granted. Immunizations being up to date and those sorts of things had not been done.... They only saw doctors if they were absolutely on their deathbed. So sometimes, you sort of had a lot of catching up to do in terms of getting their health up to date" Chronic and poorly managed conditions such as diabetes, hypertension and renal failure were common. An interviewee with an international perspective noted, "I think the world, from a health point of view, went into Kosovo thinking that all refugees were like the Great Lakes and Rwanda...The challenges [in Kosovo] were more of chronic diseases, diseases of socialization, hypertension, diabetes, renal failure...I think that people have learned to be a bit more comprehensive in their approach to conflicts and emergencies." Theme 3: Critical Challenges in Providing Health Care – Lack of information at the local level The participants expressed three main challenges which made their work more difficult: (i) tuberculosis screening, (ii) the lack of medical records and tuberculosis test results and (iii) mental health issues. Several participants noted that the tuberculosis screening process was not optimally organized. One local healthcare professional lamented, "I was reassured, but without any documentation to back it, that everyone had been screened for TB...When I contacted Public Health, they had not received any notification ...Certainly I didn't know where or how to get that information, and it seemed to me that Public Health didn't either". Communication to local health care providers about test results (tuberculosis and radiographs) was often described as sub-optimal. Immigration officials in Europe attempted to keep medical information flowing. One interviewee from an international perspective explained that, "We provided by fax and e-mail a summary of the medical conditions on the aircraft, so that they can be dealt with appropriately on arrival in Canada...We also used colour-coded cards and things so that (when) people who got off the plane (who needed care) they did receive expedient care". However, many local respondents expressed concern over the location of medical records and the inability to access this information in a timely manner. A local healthcare professional expressed, "At no point did I receive any medical documents about people with serious or chronic medical conditions that required care...if that information was available, I never got it". It was clear among the participants that mental health concerns were prevalent and service providers struggled with the delivery of appropriate mental health services. One local healthcare professional explained, "...things like headaches presenting when really the underlying condition was one of stress and anxiety, distress. Most of them came under the umbrella of what we might call, presenting with trivial complaints, but really what was beneath it all was stress and anxiety..". Theme 4: Access to Health and Settlement Services The rapid settlement of the Kosovars resulted in the local collaboration and coordination of health and settlement services at a single geographic site. This multi-agency collaboration was thought to enhance access and provision of services. One local healthcare professional described, "[The Kosovars] had, I thought, very well organized access...The SISO organization provided chauffeurs and translators, and administrators to pull all those areas together. So...if a refugee needed to go to a lab, they were driven, they were translated for, and they were brought back...". Since translators organized through SISO were available at the local health care sites, language was not perceived as a significant challenge by local health care providers. There was an awareness that the care provided was transitional in nature. This created some hesitancy in initiating treatment, as frequently there was uncertainty about the medical follow-up arrangements. Mental health issues and the lack of opportunity for identification and communication with future care providers were identified as concerns. While the Kosovars received medical coverage through the Interim Federal Health (IFH) program, local health care providers felt that the amount and nature of coverage was inadequate for services such as home care, dental care and optometry. One local healthcare professional described the issues surrounding home care, "One access issue that came up was the fact that under their IFH coverage, IFH does not cover home care, and without special arrangements to be in place...they don't qualify for it...There were a lot of people...who had walked miles and miles...and I saw [some] who had really bad foot conditions, infections and ulcers requiring daily treatment, soakings, dressings, bandage changes... Normally, those are things in the community that we would involve home care in...". In several instances, both the optometrist and dentist interviewed provided services that were not covered, free of charge, but in general, there was much confusion and uncertainty regarding the payment scheme for professionals. A dentist explained, "..we were not informed as a health professional that if we were approached by people from Kosovo that plans were available for treatment...we knew absolutely nothing until we started asking the questions..". Theme 5: Overall successes Specific examples of successes included provision of comprehensive onsite health care integrating both health and settlement services, the use of nurse practitioners to allow physicians to focus on more complex cases, the policy to keep families intact, and the positive media coverage that contributed to an atmosphere of acceptance. A local settlement provider observed, "The Kosovars was something that the whole nation took on...We saw what kind of role the media can play in making the host community aware what other people are going through, explaining that refugees are not to be seen as invaders but as people who are in need of welcoming, and the Kosovars received one of the best, I think at least in my experience, one of the warmest welcomes...". Services available to the Kosovar refugees were deemed better compared to that provided to other refugee groups. This was a concern to many participants. Some stated that all refugees should be offered the same standard of high quality settlement services as those made available to the Kosovars. Another local settlement provider noted, "Refugees, regardless from where, should be treated the same way because it creates resentment not just in other refugee communities that have come to Canada, but [it] creates resentment among the workers that are providing services and sometimes struggling to get resources for a group of people...The Kosovars should be an example of how refugees should be treated in general". Theme 6: Need for a Coordinated Approach in Migration Health A number of participants suggested that we should learn from this experience with the Kosovars and prepare to put a contingency plan in place for similar events in the future. Better communication and organization were repeatedly stressed. One participant used the term 'Migration Health' to describe such an overall coordinated approach. Several participants commented on the importance of the coordinated approach as a necessary societal investment. One interviewee from an international perspective noted, ...a lot of people who are working on the receiving end are simply following the process for receiving refugees.... there may not be the resources to look at the longer term issues: primary health care, health education, explaining how the health systems work, looking at some of the parameters that may influence longer term mortality and morbidity: dietary counselling, smoking cessation counselling, primary preventive health care procedures that we do in Canada that may not take place in the developing world." Most participants suggested that health care and settlement providers need to enhance their cultural sensitivity and cultural competence and better understand the health conditions of the displaced individuals in their country of origin. Discussion Despite organized attempts to coordinate efforts at different stages of the migration process, communication gaps and the sheer size of the influx resulted in challenges at the local level. Officials and service organizers at the international/national levels were unaware of these local gaps at the time of the evacuation. The flow of medical information and health records is an example. Primary care health workers needed to have easy access to targeted health information about the Kosovars, however this information was not available. If this information had been more easily available, health care services may have been more streamlined, and unnecessary duplication of lab tests and radiographs could have been avoided. Health care providers did not know what to 'expect'. Mental health, dental care and communicable diseases (specifically tuberculosis) were identified as requiring further specialized planning. This finding is similar to other refugee experiences in Australia [ 3 ] and Canada [ 4 ]. Our study was also consistent with other refugee literature suggesting that there is often gap in addressing refugee women's health services [ 5 - 7 ]. There was a consensus among participants that this international evacuation represented an improved approach and a good foundation on which to organize refugee health and settlement in the future. Informants' concerns about local preparedness and the need for future advanced planning was consistent with recent United Nations High Commission for Refugees (UNHCR) findings [ 8 ]. Participants wanted to see a contingency plan developed for the future with enhanced communication and better organization. These wishes have been incorporated into the detailed recommendations. A potential limitation of this study is the time elapsed between the refugees' arrival and the interviews. Asking people to recall events that took place approximately 18 months previously may have influenced the nature and detail of the information obtained. However, it may also have enabled informants to recover from the initial emotional reaction and to see these events from a different perspective. It must be stated that the 'success' as defined by key informants may be very different than 'success' from the refugees' perspective. This study examined the perceptions of those involved with the Kosovars. The majority (but not all) of the local health professionals and settlement workers interviewed had worked in the local 'settlement houses' site where there was a shorter notice of arrival of the refugees. This may have contributed to the sense of 'chaos' echoed in many of the comments. Many of the participants in this study made observations about the diverse nature of complex emergencies. The importance of logistics and planning has also been described in other studies [ 8 - 10 ]. Some countries and jurisdictions are starting to develop 'rapid response' protocols for similar situations. After the Kosovar influx, Australia developed a surveillance, triage, clinical and database system ( 'Operation Safe Haven' ). Triage questionnaires for primary health care based on International guidelines were developed [ 11 ]. Operation Safe Haven produced a template for refugee "acute health response" system. Using the Australian template combined with findings from this study, we propose the following attributes for an International Rapid Response System (IRRS): 1. Establish lead organizations at the different levels (international, national, provincial, local) 2. Clearly define roles for different organizations involved 3. Establish communication linkages between lead organizations at different levels 4. Identify strategies for flow of information ("situation reports") from authorities to local organizations 5. Establish a protocol for triage/rapid assessment of health, settlement needs and cultural preferences 6. Establish a system of medical charts that follow individual refugees though the process 7. Establish links to primary and specialized health care especially for urgent communicable disease and mental health issues 8. Identify a plan for the provision of urgent dental and eye care 9. Establish a surveillance/data-tracking system to collect essential health information, tack service use, and provide the ability to conduct quality assurance assessments 10. Use information technology – key internet links; background data; briefing; high quality background and cultural information about refugee groups 11. Introduce better training of professionals who will be dealing with refugees. Includes increasing cultural competence of health care and settlement providers, women's health, mental health, chronic illness, dental care and current health coverage for refugees. Global health issues should be introduced into health school curricula. 12. Local health care professionals need access to better information regarding the background, circumstances and organizational arrangements relating to refugees. Regular, ongoing information sharing sessions with health professionals involved with refugees, public health and government would facilitate communication when the next refugee crisis occurs. There must be political commitment at all levels. Conclusions For those involved in the Kosovar Humanitarian Evacuation Program, the experience was both overwhelming and rewarding. Many perceived that a superior effort was made for the Kosovars compared to other groups of refugees and that positive media coverage contributed to a warm and effectively organized reception. Interviewees' major concerns were the need for a more comprehensive and coordinated approach to the flow of information and handling of specific health problems. List of Abbreviations Used Complex Humanitarian Emergency: CHE United Nations High Commission for Refugees: UNHCR Humanitarian Evacuation Program: HEP Settlement and Integration Services Organization: SISO Tuberculosis: TB Interim Federal Health: IFH International Rapid Response System: IRRS Competing Interests The author(s) declare that they have no competing interests. Authors' Contributions NF and LRC designed and implemented the study, analysed the data and critically revised the manuscript. SR, MJ, MH, JK, SR critically revised the manuscript, EM conducted the interviews, assisted with study design and implementation, data analysis, and drafted the manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545939.xml
545083	A novel Mixture Model Method for identification of differentially expressed genes from DNA microarray data	Background The main goal in analyzing microarray data is to determine the genes that are differentially expressed across two types of tissue samples or samples obtained under two experimental conditions. Mixture model method (MMM hereafter) is a nonparametric statistical method often used for microarray processing applications, but is known to over-fit the data if the number of replicates is small. In addition, the results of the MMM may not be repeatable when dealing with a small number of replicates. In this paper, we propose a new version of MMM to ensure the repeatability of the results in different runs, and reduce the sensitivity of the results on the parameters. Results The proposed technique is applied to the two different data sets: Leukaemia data set and a data set that examines the effects of low phosphate diet on regular and Hyp mice. In each study, the proposed algorithm successfully selects genes closely related to the disease state that are verified by biological information. Conclusion The results indicate 100% repeatability in all runs, and exhibit very little sensitivity on the choice of parameters. In addition, the evaluation of the applied method on the Leukaemia data set shows 12% improvement compared to the MMM in detecting the biologically-identified 50 expressed genes by Thomas et al. The results witness to the successful performance of the proposed algorithm in quantitative pathogenesis of diseases and comparative evaluation of treatment methods.	Background Recently, microarray technology has provided the means for simultaneous screening and analysis of thousands of genes. Although an enormous volume of data is being produced by microarray technologies, the full potential of such technologies cannot be accessed without the ability to sift through the noisy signals to obtain useful information. The large data sets produced by microarray technology have resulted in the need for reliable, accurate, and robust methods for microarray data analysis. A major challenge is to detect genes with differentially expression profile across two experimental conditions. In many studies, the two sample sets are drawn from two types of tissues, tumours or cell lines, or at two time points during the course of a biological processes. The computationally simple methods used for such analysis, including the methods of identifying genes with fold changes (such as the popular Log-ratio graphs) [ 1 ], are known to be unreliable due to the fact that in such methods the statistical variability of the data is not properly addressed. While various parametric methods and tests such as two-sample t-test [ 2 ] have been applied for microarray data analysis, strong parametric assumptions made in these methods as well as their strong dependency on large sample sets restrict the reliability of such techniques in microarray problems. The nonparametric statistical methods, including the Empirical Bayes (EB) method [ 3 ], the significance analysis specialized for microarray data (such as SAM [ 4 ]) and the mixture model method (MMM) [ 5 ] have been specialized and applied for microarray data analysis. It is claimed and argued that the new extensions of the MMM are among the best methods producing biologically-meaningful results [ 5 , 6 ]. In this paper, without ignoring the potential applicability of non-parametric methods in microarray processing applications, due to the claims made in [ 6 ], we have restricted the comparison of our methods only to the MMM based methods. The major disadvantages of the above-mentioned methods, especially the MMM, include the lack of repeatability of the results under different runs of the algorithm, and the sensitivity of the algorithm on parameter initialization. A reliable microarray analysis method must be reproducible and applicable to different data sets under different experimental conditions. More specifically, an accurate microarray processing method is expected to pinpoint, with a relatively high degree of accuracy and robustness, genes with elevated expression levels that are related to the experimental condition in all runs. The main objective of this paper is to design and test an extension of the MMM whose results are reproducible, more biologically meaningful, and significantly less sensitive to the models' parameters. The paper is organized as follows. In Algorithms section, a review of the MMM and its recent extensions, Mod2MMM, together with the detailed description of the proposed method are given. In Results and Discussion section, the K5M algorithm is first applied to the well-studied Leukaemia data set [ 7 ] that is often treated as a benchmark problem to compare different algorithms with each other. Once the desirable performance of the proposed algorithm is verified against the Leukaemia data set, the algorithm is applied to a new data set [[ 8 - 14 ] and [ 15 ]] that deals with the pathogenesis of Hypophosphatemia, which is a common X-linked metabolic bone disorder in human and mouse. Finally, the Conclusion section is in the end. Algorithms MMM & its recent extensions We start this section with a brief review of the existing MMM based techniques. Consider Y ij as the expression level of gene in array i ( i = 1, ..., n ; j = 1, ..., j 1 , j 1 + 1, ..., j 1 + j 2 ), where the first j 1 and last j 2 arrays are obtained under two conditions. A general statistical model for the resulting data is: Y ij = a i + b i x j + ε ij (1) Where x j = 1 for 1 ≤ j ≤ j 1 and x j = 0 for j 1 + 1 ≤ j ≤ j 1 + j 2 . In addition, ε ij is a random error with mean 0. From the above formulation, it can be seen that a i + b i is the mean expression level of the first condition, and a i is the mean expression level of gene i in the second condition. The method requires that both j 1 and j 2 , the number of data sets for each experiment condition, be even. The t-test statistic type scores (2) and (3) are calculated on the pre-processed data. Here, a i is a random permutation of a column vector that contains j 1 /2 1's and j 1 /2 -1's and b i contains j 2 /2 1's and j 2 /2 -1's. Since the data are not assumed to be normally distributed, the distribution functions f 0 and f are estimated as in (4) and (5), respectively. The null distributions, f 0 and f , are estimated directly in a nonparametric model for gene expression data. Where φ (z; μ i , V i ) symbolizes the normal density function with mean μ i , variance V i , and the mixing proportions π i define the linear combination of the normal basis function. We use Φ g 0 to represent all unknown parameters {( π i , μ i , V i ): i = 1, ..., g 0 } in a g 0 -component mixture model. The number of normal basis functions, i.e. g 0 can be estimated by the EM algorithm, which maximizes the log-likelihood function of (6) to obtain the maximum likelihood estimation of . Within K iterations, the EM algorithm is expected to find the local maxima for all unknown parameters. It is recommended to run the EM algorithm several times with various random starting parameters and choose the final estimate as the one resulting the largest log-likelihood [ 6 ]. As mentioned above, using random starting points causes the result of the MMM instable and avoids reproducibility of the results. More specifically, in each run the MMM algorithm may give different number of expressed genes, which is not desirable in biological studies. This issue will be addressed in our proposed method. After finding the optimized for different g 0 's, the algorithm selects the sub-optimal g 0 corresponding to the first local minimum of BIC or AIC [ 16 ]. where v g 0 is the number of independent parameters in Φ g 0 . Then, the algorithm uses the resulting g 0 as the number of normal functions to fit f 0 . The same procedure is performed to estimate the sub-optimal number of normal functions to estimate f . As mentioned above, with the fixed number of normal functions, the parameters of functions f and f 0 are iteratively updated for a number of iterations. When the iterations are terminated, the likelihood ratio is estimated based on the final estimations of f 0 and f : LR ( Z ) = f 0 ( Z ) / f ( Z ) (9) A bisection method [ 17 ] with a Bonferroni adjustment is used to determine the cut-off points [ 18 ] for decision-making. This means that for a threshold value s , if LR ( Z ) < s , then the gene is identified to have significantly altered expression in two experiments. It is possible to determine the rejection region numerically, i.e. for any false positive rate α , the threshold value s = s ( α ) can be calculated from the following integral: In literature of microarray processing, α = 0.01 is often used as the genome wide significant level, so the gene-specific significance level is: α * = α /(2 n ) Recently a new modification of the MMM algorithm, Mod2MMM hereafter, was introduced [ 6 ]. This method points out a problem in constructing the test and null statistics and indicates that the true distribution of z may be different from the null distribution of Z , which can lead to invalid inference. The modified algorithm starts with the assumption that j 1 ≥ 2 j 2 [ 6 ], and constructs the new z and Z as you can follow in appendix1. For the cases where j 1 ≥ j 2 but j 1 < 2 j 2 , j 1 observations drawn under condition one are split into two equally-sized parts to calculate , v i (1 a ) and , v i (1 b ) respectively. To calculate and v i (2) about j 1 /2 observations are drawn under condition two. While this modification can address the differences in the distributions of f and f 0 , the stability of the parameter estimation step still remains a major problem. The main difference between the conventional MMM and its recent extensions are that the conventional MMM disregards the fact that the true distribution of z (the statistical variable under study) may be different from the null distribution of the statistics Z (as defined below). This assumption can potentially lead to invalid inference. A modified version of the MMM (Mod2MMM hereafter), introduced in [ 6 ], assumes that the denominator and the numerator of one of t-statistic-type score zi may not be independent. This method addresses the issue by constructing new z i and Z i variables as will be discussed later. A concern over all existing MMM based methods (including Mod2MMM) that greatly affects the results is associated with the way mixed distributions are estimated. In the MMM, Expectation Maximization (EM) algorithm [ 19 ] is often used to optimize the parameters of fitted mixture distribution functions of two t-statistic-type scores related with genes expression level. Starting the EM algorithm with random values as the parameters of the normal basis functions to estimate distributions makes the results depend highly on the exact initialization, and always makes variations in the results. On the other hand, if all parameters of the normal functions in the mixture model distributions are set without iterative optimization, the set values may never result to an accurate model of the data set in hand. We propose a modified version of MMM to address this problem. Our modified MMM (K5M hereafter) combines K-mean clustering and the EM estimation to not only optimize most of the parameters with the EM iteratively but also apply K-means to optimize other sensitive parameters to ensure complete reproducibility of the algorithm. The experimental results indicate superior robustness of the proposed algorithm compared to the conventional MMM and other recently introduced extensions of the MMM [ 6 ]. Proposed method (K5M) In order to address the stability and reproducibility of the MMM, we propose a new modified approach for the MMM that estimates the distribution function of z by using mixture of normal distributions in a stable and reliable way. The following observations made in the experimental study of the MMM for gene expression analysis were the main motivations for the proposed changes to the MMM: 1 The observed variations in the parameter estimation process in some versions of the MMM can be attributed to the algorithm's attempt to iteratively update the means and variances of the normal distributions using often noisy data. In experimental studies, often the direct observation of the data reveals specific points where centers (means) can be positioned and the scattering patterns that can give reliable estimates on the variance of each cluster. However, the iterative updating of model parameters with noisy data and based on some random starting points often misses the true optimal points and even creates variations and fluctuations in parameter estimation in many runs. 2 Even when variations do not occur, two runs of the algorithm can result to significantly different estimations of f and f 0 . This in turns results to lists of differentially expressed genes in different runs. More specifically, a set of two typical runs of the algorithm on the same data set can result to two lists that are very different both in number of the genes as well as the exact genes picked up by the algorithm. In our study of the conventional MMM and Mod2MMM, two runs with the same algorithm (on the same data) resulted to lists whose size vary between 50 and 200. 3 The literature of other areas of research utilizing normal basis function for estimation including neural networks indicates that in order to have more robustness in different runs and have reproducible results, the means and variances of the basis functions must be estimated and fixed during the iteration on the coefficients [ 20 ]. This is due to the fact that updating means and variances makes the estimation process a nonlinear one that is highly sensitive and very likely to become unstable. However, when updating the values of coefficients only, the problem is reduced to a reliable linear estimation that is much more robust and stable. 4 Based on the observations mentioned above, in our proposed method, finding the distribution of z is regarded partially as a clustering problem, i.e. the means and variances of the normal distributions are estimated as the prototypes of a clustering step. Specifically, if z is distributed in a one-dimensional space, wherever there is a mass of z , there is a cluster with mean μ i and variance V i , which are identified by the members of that cluster. Hence applying a clustering method is capable of estimating the means and variances of each normal distribution. The key is to use a simple clustering technique such as K-mean to estimate the mixture distributions f 0 and f based on K normal distributions. While the algorithm can use K-means to find the optimal values of means and variances, the coefficients π i 's need to be optimized using an optimization process such as the EM. Based on the above discussion, the proposed algorithm can be described in the following two steps: Step 1 : Using BIC, find the sub-optimal number of normal distributions for both f and f 0 (as described above). These optimal numbers are then used as the number of clusters in K-means technique. Step 2 : Using K-means clustering technique, for both f and f 0 find the best mean μ i and variance V i for all clusters. Step 3 : With the obtained values of μ i , V i and using the EM algorithm, iteratively update the values of the optimized π i for all clusters (both f and f 0 ), i.e. A reasonable number of clusters is expected to be obtained from the first step of the algorithm, and the estimation results of the two bellow data sets in Tables 1 and 4 show that the used K (calculated based on AIC) is satisfactory. Table 3 shows the results of the MMM and K5M methods for the run with an unequal variance and four normal distributions for both f and f 0 . The MMM creates the likelihood ratio (LR) statistics plotted in Figure 1 , the K5M with K = 4 forms the LR statistics plotted in Figure 2 , and the K5M with K = 2 results to the LR plot of Figure 3 . Table 1 Comparison of the result of the K5M with the MMM and the Mod2MMM based on the Leukaemia data. Method Total detected genes ALL AML Total accepted genes out of 50 genes [22] MMM 187 21 18 39 Mod2MMM 58 14 16 30 K5M, K = 3 185 25 20 45 K5M, K = 4 58 19 8 27 Table 3 Estimation of fitted and by MMM (in the optimum run) and K5M. f 0 f MMM = (0.1859, -0.2231, 0.0322, 0.0638) = (-0.0387, 0.4381, 0.1600, -0.1933) = (0.3215, 0.3522, 0.7692, 0.337) = (3.2288, 3.397, 2.6393, 4.6982) = (0.1672, 0.2353, 0.4048, 0.1925) = (0.0687, 0.0509, 0.0263, 0.0725) K5M = (1.1111, 1.1264, 0.3115, -0.3329) = (1.7867, -0.6817, -2.354, 0.3324) = (0.4589, 0.4640, 0.1879, 0.1807) = (2.9432, 0.5583, 4.24, 0.5027) = (0.1914, 0.1963, 0.3120, 0.3001) = (0.0583, 0.1018, 0.0294, 0.0442) Table 4 The top ten most significant genes provided by K5M and MMM. GenBank Accession IDs Gene/ Protein Description Rank based on MMM Rank based on K5M D00073 Kidney/ carrier activity 1 1 AA815845 Unknown 2 2 AF085696 ion transportation/ K+ channel, inward rectifier/renal salt flow 3 3 AW047688 Brain/Hypothalamus 4 4 M12660 Kidney/ Complement protein H gene 5 5 AI847513 Brain/ Hypothalamus 7 6 AA919924 Phosphate metabolism/inositol-1(or4)-monophospha te Activity 6 7 X69966 Dilation of the proximal renal tubules and extensive vacuolization of tubule epithelium 8 8 AF103809 Elevated kidney levels of lysosomal enzymes 9 9 AA711516 Barstead mouse myotubes MPLRB5 10 10 Figure 1 Likelihood ratio statistics as a function of Z value based on the MMM method. Figure 2 Likelihood ratio statistics as a function of z based on the K5M with K = 4. Figure 3 Likelihood ratio statistics as a function of z based on the K5M with K = 2. It is worth mentioning that due to the random initialization in K-means and the random initialization of the coefficients π i 's, in each run, it is expected that the number of identified differentially expressed genes fluctuate slightly. However, as indicated above, since the K- means clustering algorithm is known to a robust method, and considering the fact that in the EM estimation process, only a linear estimation is performed, it is expected that the robustness of the proposed algorithm be much more than the other version of the MMM based algorithms. This observation, as have been shown before, is supported by our experimental results. In addition, our experimental indicate that the most expressed genes are identified in all runs or the algorithm and in each run one or two new genes with less expression ratio are added to this set. Results and discussion In this section, first the two applications and their corresponding data sets are described and then the results produced by the proposed method (i.e. K5M) is compared with the other MMM based methods on two data sets. The detailed description of the methods is given in MMM & its recent extensions Section. Leukaemia dataset In this section, we apply the nonparametric MMM method with and without the proposed modifications to the Leukaemia data presented in [ 7 ]. The objective of this application is to identify the most important genes involved in development of different types of Leukaemia. The dataset used for this analysis includes 27 acute lymphoblastic leukaemia (ALL) samples and 11 acute myeloid leukaemia (AML) samples for 7129 genes. The main goal is to find genes with differential expression between ALL and AML cases. A second goal is to compare the result of MMM and Mod2MMM (as introduced in MMM & its recent extensions Section) with K5M and test the robustness of K5M. The genome-wide significance level is chosen α = 0.01 (according to Benferroni adjustment used in the MMM based methods). Each sample in the dataset is pre-processed as in [ 21 ], by subtracting its median and dividing the resulting variable by its quartile range (i.e. the difference between the first and the third quartile). Results of Leukaemia study Thomas et al [ 22 ] used known biological information to identify the most important genes in Leukaemia and provided biological justifications for these identified genes. They introduced 50 genes out of the identified genes as the most expressed and related genes to the disease, including 25 most expressed genes for AML and 25 for ALL. We treat Thomas et al's list as the biology knowledge base and compare the capabilities of the computational techniques to correctly identify the genes discussed in [ 22 ] by processing the dataset. The comparison of the result obtained by the K5M with those of the MMM and the Mod2MMM is summarized in Table 1 . As can be seen in Table 1 , The MMM has identified 187 differentially expressed genes [ 21 ], among which the total of 39 genes are in the list of genes obtained by Thomas et al [ 22 ]. The Mod2MMM method found 30 genes of the Thomas's list. The K5M algorithm, determines 45 genes that are identified in the Thomas's list, i.e. the proposed algorithm successfully identifies 90% of biological result. This means that K5M improved the detection of expressed genes 12% compare to the MMM and 30% compare to the Mod2MMM for the Leukaemia data, i.e. our method identified more genes from the list of the 50 truly expressed genes identified by Thomas et al [ 22 ]. As the BIC suggested the optimum number of clusters K = 4 for the MMM, the K5M is applied with K = 4 also. Running K5M with different number of clusters leads to the different but reasonably similar results. As the number of the clusters increase, the number of expressed genes decreases. Table 1 shows that the K5M with K = 3 identifies the total of 185 differentially expressed genes, while with K = 4 the total of 58 genes are identified, however; the 58 genes found with K = 3 are the most expressed genes among 185 genes found by K = 4. This result shows the consistency of the K5M method. In order to further compare the performance of the MMM and K5M on the leukaemia data, The ROC curve is plotted based on False Positive rate and True Positive rate of the data set calculated as in [ 5 ]. The area under each curve is the measure of test accuracy. As can be seen in Figure 5 , the area under the K5M curve is more than the area under the MMM curve, therefore the K5M is providing a more accurate classification than the MMM. Figure 5 ROC curves for the MMM and K5M based on the leukaemia data set. The area under the K5M curve is more than the area under the MMM which shows the K5M method is more accurate than the MMM. Hypophosphatemia dataset The following study is the main application for which the proposed method was specialized and therefore is described in more details. Hypophosphatemia is a common X-linked metabolic bone disorder in human. Hypophosphatemia results from phosphate wasting in the renal tubules. Phosphate that is normally reabsorbed from the urine is excreted. It appears that elevated levels of FGF-23 activate the excretion of phosphorous by the kidneys. Previous studies have demonstrated an impairment of the high- affinity, low capacity Na+ dependent phosphate co-transport system [ 23 , 24 ]. The main animal model used to study this disease is the Hyp mouse. Hyp mice have a mutation of the Phex gene [ 25 , 9 ]. The disease is characterized by low reabsorption of phosphate, bone disease, and bone abnormalities in the lower extremities. The genes active in the regulation of phosphate re-absorption in the kidney are not well understood. It is also not clear whether mutations of the Phex gene block renal adaptation to low phosphate diet. Hyp mice have a primary osteoblast defect and defects in vitamin D metabolism. Parabiosis experiments on normal and Hyp mice have revealed that there is an intrinsic osteoblast defect in Hyp mice rather than an intrinsic renal abnormality. Hyp kidneys transplanted into normal mice reabsorbed phosphorus at normal levels. Kidneys transplanted from normal mice into Hyp mice began phosphate wasting in the Hyp mice. The mechanism that leads to the excessive excretion of phosphorous is unknown. On a low phosphate diet a normal mouse will activate systems to conserve phosphate by increasing re-absorption. The genes activated in the normal mouse on the low phosphate diet, and the genes with differential expression between normal and Hyp mice should indicate the systems involved in the phosphorus homeostasis. In an attempt to identify these genes, nutritional experiments were performed on normal and Hyp mice [[ 9 , 8 - 14 ] and [ 15 ]]. Normal and Hyp mice were placed on low phosphate diets for 3 – 5 days. Tissue samples from the kidneys of test and control mice were collected. 16 samples were analyzed using Affymetrix GeneChip mouse U74A arrays- 4 samples for each experiment state. The mRNA of 12,488 genes was analyzed. Two GeneChip microarrays were done for each diet for normal mice and three microarrays for each diet for the Hyp mice for a total of 10 arrays. To investigate this, 5-week-old normal and Hyp were fed a control (1.0% P) or low phosphate (0.03% P) diet for five days. The four group experiments are shown in Table 2 . Table 2 Four experimental groups in the Hyp mice data sets. In this paper, The comparisons are done between group 1 and group2, and between group 3 and group 4. Diet Control Low Phosphate Genotype Normal Group1 Group2 Hyp Group3 Group4 In this study, we consider the gene expression signal less than 100 as noise caused by the microarray machine, and in the pre-processing step we ignored the genes whose expression signals in both conditions are less than 100. The following two specific goals are considered in this study: 1. To identify the genes in whose mRNA expressions are altered by low phosphate diet in normal mice. 2. To determine the effect of Hyp mutation on this response, i.e. identifying the genes in Hyp condition that are differentially expressed across the normal and low phosphate diet experiments. Results of Hypophosphatemia study The Hyp dataset includes five samples for each group. In order to make the number of data samples even, we used four samples of each group. For this data set, since j1 = j2, the Mod2MMM cannot be applied. In MMM method, five mixture models are used to estimate f 0 and f (distributions under two experimental different conditions) with number of normal basis functions ranging from 1 to 5, i.e. The MMM algorithm was run several times and the run with maximum log-likelihood was chosen as the final model. Bayesian Information Criterion (BIC) [ 26 ] was used to determine the number of components. To find the rejection region for a given model, the bisection method is used. In this paper we assume α = 0.01, and therefore the gene-specific significance level used here is calculated as: α * = 0.01/(95.44 * 2) = 5 * 10 -7 Using bisection method [ 17 ], as discussed in Section 4, the value of s is obtained as s = 3 × 10 -6 .. Both the MMM and K5M were run 100 times. Figure 4 presents the number of genes expressed in each run of the MMM. The difference between the number of identified differentially expressed genes in two runs with the minimum and the maximum number of genes amounts to 150 genes. This clearly indicates the high degree of inconsistency and irreproducibility of the results obtained by the MMM. The number of genes expressed in each run of the K5M indicates that all genes are the same in all runs and therefore indicates 100% repeatability and robustness of the proposed method. Figure 4 Histogram of the number of genes expressed in each run by the MMM method which shows the strong variability (x-axis shows number of runs). The ten most significant genes expressed by the low phosphate diet in the normal mouse identified by the MMM, and the ten most significant genes provided by K5M are represented in Table 4 . As can be seen in Table 5 , the most differentially expressed genes are same for the MMM and K5M. Out of these 10 genes, six are directly related to the kidney's functions. For this data set, the main advantage of the K5M is its consistency and robustness as discussed above. A similar procedure is conducted to accomplish the second goal of this study, i.e. identifying the role of Hyp condition on the most definitely expressed gene in normal and low phosphate diet microarrays. The ten most significant genes that are differentially expressed across the two experimental conditions, i.e. Normal Low Phosphate and Hyp Low Phosphate, are listed in table 6. As shown in the table 6, again eight genes are related directly to the kidney's function. These further witnesses to the capability of the proposed technique to discover the genes that are truly involved in the biological study. Table 5 The top ten significant genes, by comparing group 3 and group 4 in table 2, provided by K5M and MMM. Accession IDs Gene/ Protein Description Rank based on MMM Rank based on K5M AF028071 Kidney/ apical plasma membrane, Basolateral plasma membrane 3 1 D26352 Kidney/calcium ion binding 1 2 AA815845 Unknown 2 3 D00073 Kidney/ carrier activity 5 4 AB00603 Monooxygenase activity, oxidoreductase activity 9 5 U97079 GTP binding, protein binding, phosphate binding 7 6 AI315650 Detected in Kidney 6 7 X71922 Kidney/ growth factor activity, hormone activity 11 8 D43797 Kidney/carrier activity, sodium, excitatory glutamate symporter activity Identified as a non expressed gene 9 X81059 Protein phosphate 2 Identified as a non expressed gene 10 Conclusions In this paper, we proposed a technique to improve the repeatability, and robustness of the mixture model method by using the K-mean clustering method in estimating the distributions. Our proposed method finds the distribution of the variables partially based on a clustering procedure and an EM optimization process. The method is applied to analyze two microarray data sets, Leukaemia data set and a data set reflecting the effect of the low phosphate diet on regular and Hyp mice [ 8 ] data. The experimental results indicate 100% robustness and repeatability of the results in different runs and provide 12% improvement (compared to the mixture model method) in detecting the relevant genes in both studies. Authors' contributions Maryam Zaheri, and Ali A. Rad were in charge of writing the codes and programming aspects of the paper. Siamak Najarian and Javad Dargahi's primary role was to perform a literature review on mixture model techniques, identify the aspects of the method that need to be improved, and provide suggestions to address these shortcomings. Kayvan Najarian's primary roles were to design improvments to the algoritm (based on the literature review and overal modifications suggested by Siamak Najarian and Javad Dargahi), prepare and pre-process the data (for both datasets), partcipate in preperation of the Hyp dataset, define the Hyp problem interpret the results and finally write and edit the manuscript. Appendix 1 The Mod2MMM makes a new z and Z based on the following formula: Where: And:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545083.xml
546213	Genome-wide and Ordered-Subset linkage analyses provide support for autism loci on 17q and 19p with evidence of phenotypic and interlocus genetic correlates	Background Autism is a neurobehavioral spectrum of phenotypes characterized by deficits in the development of language and social relationships and patterns of repetitive, rigid and compulsive behaviors. Twin and family studies point to a significant genetic etiology, and several groups have performed genomic linkage screens to identify susceptibility loci. Methods We performed a genome-wide linkage screen in 158 combined Tufts, Vanderbilt and AGRE (Autism Genetics Research Exchange) multiplex autism families using parametric and nonparametric methods with a categorical autism diagnosis to identify loci of main effect. Hypothesizing interdependence of genetic risk factors prompted us to perform exploratory studies applying the Ordered-Subset Analysis (OSA) approach using LOD scores as the trait covariate for ranking families. We employed OSA to test for interlocus correlations between loci with LOD scores ≥1.5, and empirically determined significance of linkage in optimal OSA subsets using permutation testing. Exploring phenotypic correlates as the basis for linkage increases involved comparison of mean scores for quantitative trait-based subsets of autism between optimal subsets and the remaining families. Results A genome-wide screen for autism loci identified the best evidence for linkage to 17q11.2 and 19p13, with maximum multipoint heterogeneity LOD scores of 2.9 and 2.6, respectively. Suggestive linkage (LOD scores ≥1.5) at other loci included 3p, 6q, 7q, 12p, and 16p. OSA revealed positive correlations of linkage between the 19p locus and 17q, between 19p and 6q, and between 7q and 5p. While potential phenotypic correlates for these findings were not identified for the chromosome 7/5 combination, differences indicating more rapid achievement of "developmental milestones" was apparent in the chromosome 19 OSA-defined subsets for 17q and 6q. OSA was used to test the hypothesis that 19p linkage involved more rapid achievement of these milestones and it revealed significantly increased LOD* scores at 19p13. Conclusions Our results further support 19p13 as harboring an autism susceptibility locus, confirm other linkage findings at 17q11.2, and demonstrate the need to analyze more discreet trait-based subsets of complex phenotypes to improve ability to detect genetic effects.	Background Autism (OMIM # 209850) is a neurobehavioral disorder involving deficits in language and social abilities and patterns of repetitive behaviors, restricted interests and resistance to change. The most recent estimate of population prevalence for the broader autism spectrum indicates a rate of 34/10,000 (~1/300) [ 1 ], with a male: female ratio of 4:1 [ 2 , 3 ]. Evidence from various studies indicates idiopathic autism has a complex genetic etiology. Twin studies show a concordance of 60% among monozygotic (MZ) twins and 0% among dizygotic (DZ) pairs for classic autism, but this increases to 92% for MZ pairs and 10% for DZ pairs when a broader phenotype of related social and language abnormalities is included [ 4 , 5 ]. The sibling recurrence risk is suggested to be ~3–10% but may be underestimated as a result of "stoppage rules" [ 6 - 8 ], and the relative risk is thus 30–100 times that in the general population [ 5 , 7 ]. Heritability is estimated at 90%, which is among the highest for psychiatric disorders. While the data do not strongly endorse any one model for inheritance, twin and family studies support a multilocus etiology with as many as 10–20 loci (reviewed in [ 9 - 11 ]). Genome-wide screens of multiplex autism families for susceptibility loci [ 12 - 22 ] have identified a few genomic regions in common across multiple studies; 7q and 2q have received the greatest attention [ 17 , 19 , 20 , 23 - 28 ], with support from chromosomal abnormalities affecting these regions in idiopathic autism (reviewed in [ 29 ]). Genetic studies of autism are substantially complicated by clinical and locus heterogeneity, and it is possible that epistatic or epigenetic mechanisms may play important roles in genetic etiology [ 9 , 30 ]. Analytical strategies that address the latter concerns are limited, and most studies to date have focused on analysis of main effects using a global autism diagnosis to define affection status. Moving forward, more sophisticated approaches are being proposed in which trait-based subsets of the broader autism phenotype are used in genetic analyses. Similarly, given the interdependence of genes and their protein products within biological systems, analytical approaches that address potential interaction between susceptibility loci will also be critical to characterizing gene-phenotype relationships in autism. We report a second generation 10-cM microsatellite-based genomic screen of multiplex autism families. The dataset for this screen includes 71 families recruited by the Tufts/New England Medical Center, a well-characterized set of 85 families from the Autism Genetics Resource Exchange (AGRE), and 2 families from Vanderbilt University. Several sites of suggestive linkage are identified, although none meet criteria for genome-wide significance. The loci with greatest support for linkage were 17q11.2 and 19p13; the latter site demonstrated significantly increased allele-sharing when the Ordered-Subset Analysis (OSA) algorithm was employed using a quantitative trait-based autism phenotypic subset related to specific "developmental milestones" as a covariate to rank families. Methods Sample and Demographics The demographics for the 158 family dataset comprising the studies in this report are shown in Table 1 . Families were recruited through three sites: (a) 71 families from the Tufts/NEMC site, (b) 2 families from the Vanderbilt University site, and (c) the remainder of families (85) were chosen from the AGRE repository based on our own recruitment criteria. Multiplex families (mostly affected sibling-pairs) had one affected individual who met full criteria for autistic disorder based on Autism Diagnostic Interview-Revised (ADI-R; [ 31 - 33 ])) algorithm scores, while the second individual either met criteria or in some cases was under the cut-off by only one or two points. Exclusion criteria included dysmorphic features, abnormal karyotype, diagnosis of fragile X syndrome, and other genetic disorders of known etiology. Individuals were assessed by the respective groups using the ADI-R at a developmental age >18 months; Tufts/NEMC and Vanderbilt groups included individuals between the ages of 4 and 22; in cases in which ADI-R interviews were performed initially at <4 years, they were repeated when the probands reached 4 years of age. All individuals were additionally assessed using the Autism Diagnostic Observation Schedule [ 32 , 34 ] and the Vineland Adaptive Behavior Scales – Interview Edition [ 35 , 36 ]. Genotype data and statistical analysis DNA from Tufts and Vanderbilt samples was obtained from peripheral blood or immortalized lymphoblastoid cell lines using the PureGene Kit (Gentra Systems). While a minority of families from the Tufts/NEMC cohort had been genotyped previously [ 13 ], both new and previously genotyped families were genotyped by deCODE (Reykjavik, Iceland) using their 500 marker (~8 cM intermarker spacing) panel and corresponding genetic map [ 37 ]. Genotype data were obtained from the AGRE website [ 38 ] for families whose samples were purchased from the AGRE repository and included in this study. Clinical procedures and genotyping for the AGRE sample has been described previously [ 18 , 39 ]. AGRE samples and corresponding genotype data had a distinct but overlapping panel of markers compared to the Tufts and Vanderbilt families. AGRE genetic markers were placed on the deCODE map, with order and spacing properly insured through exhaustive comparisons between genotyped markers, available genetic maps, and physical DNA sequence assemblies in both public and Celera databases. Genotype data for each chromosome underwent thorough error detection and genotype confirmation. Initially, data were tested for Mendelian inconsistencies using PEDCHECK [ 40 ] and RELPAIR [ 41 ], followed by SIMWALK2 [ 42 ] for haplotype construction to detect genotyping errors reflected by unlikely double recombinants. In the event of a highly improbable genotype, the data for that marker were excluded for the family. Allele frequencies were estimated using genotype data from all unrelated individuals in the combined dataset, consisting of more than 300 chromosomes. Allele frequencies were compared with available data from other Caucasian populations, and no significant differences were observed (data not shown). The LAPIS program of the PEDIGENE system [ 43 ] was used to output appropriate analysis files for the different programs. Linkage was analyzed using both model-dependent and model-independent methods. For autosomes, two-point and multipoint heterogeneity LOD (HLOD) scores were calculated under both dominant and recessive models using Allegro [ 44 ]. Disease allele frequency was estimated to be 0.01 and 0.1 for dominant and recessive models, respectively. Phenotypic status was only considered for affected individuals, and other family members were designated as having an unknown phenotypic status. Nonparametric allele-sharing LOD* values were calculated using affected relative pair data based on an exponential model using the S pairs scoring function as recommended by McPeek [ 45 ]. NPL scores and corresponding P-values were also calculated by Allegro. Data from the X chromosome were analyzed using ASPEX [ 46 ] and FASTLINK [ 47 ] to calculate two-point and multipoint MLOD scores. Peak parametric (HLOD) or nonparametric LOD* scores ≥1.5 were considered as "suggestive" evidence for linkage and listed in Table 2 , along with corresponding peak marker, deCODE cM location, and chromosomal band position. The nonparametric genome-wide significance threshold [ 48 , 49 ] for linkage at the P = 0.05 level was determined by conducting simulations using Merlin [ 50 ] with the current dataset. The Simulate option in Merlin was used to produce 1000 random datasets that preserve the properties of the original data for marker informativeness, spacing and missing data patterns. An empirical significance threshold was determined by using the 95 th percentile of the resulting distribution. OSA [ 51 ] identifies genetically more homogeneous subsets of the overall data by ordering families according to covariate trait values in ascending or descending order. OSA takes the first family and calculates an allele-sharing LOD* score. In an iterative process, OSA successively adds families, re-calculating LOD* scores with each addition, and it identifies the division in the dataset at which maximum linkage is obtained on the chromosome being analyzed. Permutation testing is used to determine the empirical significance of the observed results. OSA has been applied with success to identify or increase evidence in support of linkage to complex disease susceptibility loci [ 52 - 54 ]. To explore potential genetic interaction or other genetic correlations between sites of main effect (i.e. suggestive linkage), OSA was applied using family-specific LOD scores as the covariate trait. Families were ranked according to the family-specific LOD score at peak sites demonstrating LOD scores ≥1.5. Allele-sharing analysis was performed for the other six chromosomes using the OSA algorithm. For instances of empirically significant increases in evidence for linkage, we explored the nature of the genetic correlation to ask whether it reflected clinical correlations in the respective subsets. We employed ADI-based factor subsets, identified by principal components analyses of ADI/ADI-R items, to represent putative phenotypic subsets in autism [ 55 , 56 ]. The ADI-based variable clusters correspond to "(1) language, (2) social intent, (3) developmental milestones, (4) rigid-compulsive behaviors, (5) savant skills, and (6) sensory aversion", as determined by Folstein and colleagues [ 55 ]; and (7) "insistence on sameness" as described by Cuccaro and colleagues [ 56 ]. We thus compared the seven ADI-based factor score means (both the mean of family means and the mean of affected individuals) using a t-test for the families above and below the OSA-determined split in the dataset resulting in maximal linkage. Subsequent analysis involved specific examination of the "developmental milestones" cluster. The milestones factor indexes on the following ADI items: " (1) To walk unaided; (2) to sit unaided on flat surface; (3) age of first single words; (4) age of first phrase; (5–6) acquisition of bladder control: daytime, night; (7) acquisition of bowel control. " Analysis of the "developmental milestones" factor as a potential phenotypic subset related to the autism linkage correlations was performed by applying the OSA algorithm. We used "developmental milestones" family means, normalized via SAS and Box-Cox transformation procedures, as an ascending ranking covariate. LOD* scores were calculated according to the OSA algorithm, and the resulting increase in linkage achieved with the OSA-determined family subset was analyzed through permutation testing. Approval for these studies was granted by the respective Institutional Review Boards at Tufts University School of Medicine/New England Medical Center and Vanderbilt University Medical Center. Additionally, all studies were performed with informed consent provided by the families participating in the research. Results Seven chromosomes revealed one or more regions of linkage with a model-dependent or model-independent LOD score ≥1.5 (Figure 1 ). No locus reached the empirically derived genome-wide significance level of 2.92. These suggestive loci include 3p25, 6q23, 12p12, 16p12-p13, 17q11, 17q21 and 19p13 (Table 2 ). Data provide the most compelling support for 17q11.2 and 19p13 as harboring autism susceptibility loci. For 17q11.2, peak linkage was observed at 53 cM, corresponding to marker D17S1294 (Table 2 ), at which we see a multipoint HLOD of 2.85. Nonparametric multipoint analysis revealed an allele-sharing LOD* score of 2.13 and an NPL score of 2.84 (P = 0.0024). A second telomeric peak can be distinguished on 17 at ~69 cM, corresponding to 17q21.2. Marker D17S1299 at this site yielded a HLOD of 1.9, a LOD* of 1.66 and an NPL score of 2.26 (P = 0.012). The more proximal peak at ~53 cM lies in close proximity (~150 kb) to the serotonin transporter ( SLC6A4 ) locus, long considered to be an attractive functional candidate gene for autism and other neuropsychiatric conditions. Figure 2 shows multipoint LOD score plots for both dominant and recessive parametric (HLOD) and nonparametric allele-sharing LOD* values for chromosomes 17 and 19. The second most significant result was observed on 19p13, where peak linkage was detected at marker D19S930, revealing a multipoint HLOD of 2.55 at ~40 cM (Table 2 and Figure 2 ). Nonparametric analyses detected a LOD* of 1.92 and a corresponding NPL of 2.77 (P = 0.003). As with chromosome 17, the multipoint analyses show a second more telomeric peak, corresponding to marker D19S113. The recessive HLOD at this site was 2.20, with model-independent LOD* and NPL values of 1.39 and 2.10 (P = 0.018), respectively. To address the possibility of gene-gene interaction, we applied the OSA approach with family-specific LOD scores as the ranking trait. Families, almost all of which are affected sib-pairs, were ranked in both ascending and descending order using family-specific LOD scores. The three most significant correlations are presented in Figure 3 . Using chromosome 19 LOD scores as the covariate, the results on chromosome 17q, while non-significant (P = 0.1), showed an increase in linkage at the more distal peak on 17q21.1 from a LOD* of 1.7 to 3.6 and identified an optimal subset of 52 families. Applying the same covariate, a significant increase was seen on chromosome 6q, with a smaller, completely overlapping, 30-family optimal subset. This subset resulted in an increase in LOD* values from 1.0 to 3.6 at ~164 cM (P = 0.004). Another significant finding involves the 7q region, possibly representing the most replicated site of linkage in autism (reviewed in [ 9 , 29 ]). Given a substantial focus on this region over several years, we lessened our criteria to examine any other chromosome demonstrating a LOD score >1. Application of OSA using chromosome 7q linkage data, again ranking families based on LOD scores in a descending manner, lead to a significant increase in linkage on 5p at ~41 cM from a LOD* of 1.1 to 3.3 in a 41-family subset. Thus, in these three cases, notwithstanding the nonsignificance of the 19p13/17q21 result, there is a positive correlation of linkage in varying but overlapping subsets of the data between these respective pair-wise locus combinations. To further explore the basis of the observed results, we tested the hypothesis that underlying phenotypic correlates might explain genetic correlations. We compared the mean values for the seven factor traits in the optimal subsets compared to the means of the remaining families using a t-test, both under assumption of equal and unequal variances. This comparison for all seven available factors revealed a nominally significant differences in the chromosome 19 optimal subsets identified from OSA analysis of chromosomes 17 (52 families) and 6 (30 families) for the "developmental milestones" cluster. The families in the optimal OSA subset have lower scores and therefore are more rapidly achieving developmental milestones. A similar procedure for the chromosome 7-based subset revealed no obvious differences in any of the factors (data not shown). To directly test the hypothesis that chromosome 19 linkage was related to reduced affection for the "developmental milestones" factor, we performed an OSA analysis in which families were ranked in ascending order based on mean values for the milestones factor score. Figure 2 shows the results from this analysis, which generated increased evidence for linkage to 19p13 with peak LOD* scores increasing from 1.9 to 3.4. Permutation testing revealed this increase to be empirically significant (P = 0.04), thus further supporting 19p13 as harboring a genetic risk factor for autism. Discussion We have presented evidence in support of autism susceptibility loci on chromosomes 17q and 19p. Our results suggest that the 19p locus is related to a phenotypic profile involving a more rapid achievement of particular "developmental milestones". Features indexed in this ADI-based factor are: (1) ability to walk unaided; (2) ability to sit unaided on a flat surface; (3) age of first single words; (4) age of first phrase; (5–6) acquisition of bladder control: daytime and night; and (7) acquisition of bowel control. Analyses leading to this conclusion also showed positive genetic correlations between optimal OSA-defined subsets contributing to linkage at 19p13 and increases in linkage at loci on 17q21 and 6q23. A similar positive genetic correlation was shown for chromosomes 7q and 5p, however this observation lacks evidence of an underlying phenotypic relationship based on available ADI variable clusters. While the increase in linkage at 17q21 was not empirically significant, the differences in "milestone" score means between the optimal chromosome 19 subsets seen for both chromosomes 17 (52 families) and 6q (30 families) were significant. These exploratory data led to the significant finding of increased linkage in the single direct test of our hypothesis concerning the phenotypic correlation related to chromosome 19 linkage. Despite the significance of the final results on 19, we remain cautious in the interpretation of the overall results. As with a number of other genomic screens in autism, no single main effect locus achieved genome-wide significance. Support for a number of these loci, particularly at 17q11.2 and 19p13 comes from similar suggestive linkage in other genomic screens for autism. Although not all screens detect these loci (not an uncommon finding in linkage studies for complex genetic disorders), the evidence is strong regarding an effect at 19p, within 10 cM of our peak: (1) Shao et al reported an MMLS = 1.21 and an MLOD = 1.38 [ 14 ]; (2) the Paris Autism Research International Sibpair Study (PARIS) an MMLS = 1.37 [ 12 ]; the International Molecular Genetic Study of Autism Consortium (IMGSAC) reported an MLS of 1.16 [ 15 ]; the Mt. Sinai group reported an NPL of 1.56 which increased to 2.31 when only families with obsessive-compulsive behaviors were considered for this region [ 22 ]. Similarly, several groups have reported evidence for linkage at 17q11. The recently published AGRE follow-up genomic screen identified an MLS of 2.83 near SLC6A4 [ 21 ]. A genome scan for attention deficit/hyperactivity disorder (ADHD) identified an MLS of 2.98 near this locus [ 57 ]. An IMGSAC follow-up screen for autism [ 27 ] reported a maximum multipoint LOD score of 2.34 at HTTINT2 in the SLC6A4 gene on chromosome 17q11.2. Our own more preliminary analysis of linkage in this region with a highly overlapping dataset to that in the current study, revealed very similar results [ 58 ]. Given our inclusion of some AGRE families, it is not completely unexpected that 17q11.2 linkage is similar to that seen the larger AGRE 2 nd -stage screen [ 21 ], however AGRE families only represented about half of the overall dataset. Families recruited from the Tufts/NEMC site are clearly contributing to this linkage based on the LOD score-based optimal family subset compositions. The 17q21 locus is worth further consideration. Our data support the premise that the adjacent linkage peaks represent distinct loci and are not an artifact of primary linkage at 17q11.2. The evidence for linkage at 17q21, while weaker than that at 17q11.2 only 16 cM centromeric, specifically showed an, albeit nonsignificant, interlocus correlation with 19p13 linkage. Linkage at 17q11.2 in this subset of families actually decreases slightly. Of particular interest is the fact that the distal region harbors the integrin β3 ( ITGB3 ) locus, which was identified recently from a genome-wide quantitative trait locus (QTL) association screen for platelet serotonin levels [ 59 ]. We see nominal evidence of linkage to autism at this site, and ~20–25% of individuals with autism have elevated levels of circulating serotonin. The other "suggestive" (LOD ≥ 1.5) loci reported here have also been detected in other genome-wide scans for autism loci. A broad region of 7q has been detected in most screens [ 17 , 20 , 23 , 25 , 27 , 28 ]. The 16p region has been identified by IMGSAC, and others [ 15 , 18 , 22 , 27 ]. Chromosomal abnormalities have also been reported for several of these regions in cases of autism (reviewed in [ 11 ]). Linkage at 3p was reported by at least two groups [ 14 , 17 ]. Linkage has also been reported at our 6q locus by at least one other group [ 12 ]. Thus, while not significant, the replication of these linkage observations provides support for the likelihood that many of these loci represent true sites of main effect in autism. The application of OSA to detect putative interlocus correlations between the 19p13 and 17q21, 19p13 and 6q23, and between 7q35 and 5p are limited to some degree in significance by their highly exploratory and hypothesis-generating nature. Given the number of comparisons between loci, and the number of comparisons between optimal subset pairs (on 19p or 7q) for the traits means, the potential for type I error is increased. Therefore our interpretation must be cast alongside appropriate caveats. Nevertheless, the multiple exploratory comparisons generated a hypothesis: that linkage to 19p13 was related to a more rapid achievement for specific milestones . We tested this hypothesis with a single analysis revealing an empirically significant increase in linkage at 19p13. Our results of autism linkage and its increase using an ascending milestone score covariate in OSA, taken in the context of replicated observations of suggestive linkage by other groups, strengthens support for the presence of an autism gene at this site. In the end, ultimate interpretation will rely upon replication of these phenomena with independent samples to confirm these observations. Finally, our results highlight the utility of using trait-based subsets of autism to identify putative susceptibility loci for this complex disorder. We and others have hypothesized a likely increased specificity of individual risk genes and corresponding alleles for traits or subphenotypes comprising the broader autism spectrum. Therefore methods such as OSA with power to identify more homogeneous samples and QTL (quantitative trait locus) linkage and association analyses should provide greater sensitivity in the discovery of disease genes in the context of locus and clinical heterogeneity. Additionally, OSA or other forms of conditional linkage analyses, have the ability to uncover potential interactions between loci, an important concept since the inherent interdependence of proteins in common pathways or networks acting during development and normal neuronal function could be easily imagined to act genetically in concert with one another. Conclusions We report evidence to support linkage of autism to 17p11.2 and 19p13. Exploratory analyses to test for correlations between suggestively-linked loci, using the OSA method revealed positive correlations of linkage (i.e. in overlapping families) between 7q and 5p, 19p and 6q, and possibly 19p and 17q22, distal to peak linkage at 17q11.2. Comparing mean scores for ADI-derived factor traits from families above and below the OSA-defined split maximizing linkage, suggested a positive correlation between 19p13 linkage and a more rapid achievement of "developmental milestones" as measured by items in this cluster of ADI variables. We tested this hypothesis by applying OSA with descending "developmental milestone" scores as the ranking covariate, and detected an empirically-significant increase in linkage to 19p13. These findings further support evidence for an autism susceptibility locus in 19p13 and underscore the utility in applying trait subsets in complex disorders to identify genetic risk factors. Competing interests The authors declare that they have no competing interests. Authors' contributions JLM and LMO were responsible for conducting the genome-wide linkage analyses; JLM conducted all OSA analyses, was largely responsible for coordinating and executing the bulk of the reported studies, was key to drafting and editing the manuscript text, and in preparing the figures. CL and JLH provided input into the design and interpretation of statistical results, and provided guidance for conducting the genome-wide simulations required to determine the significance threshold. JLH established the Core-based infrastructure in the Center for Human Genetics Research that facilitated this work and provided very helpful input into the development of the manuscript. SEF developed, with her group, the ADI-based clusters so crucial to examining phenotypic subsets in this paper, which she helped to edit. SEF was responsible for overseeing recruitment and the phenotypic assessment of families from her research group, then at Tufts/NEMC. GC is the clinical coordinator at the Vanderbilt site and oversees ascertainment, recruitment and detailed phenotypic assessment of affected individuals. KG is the Vanderbilt data coordinator and is responsible for management and oversight of family information, pedigree data, and status of DNA, blood or cell line samples for family members. She is directly responsible for preparation of Table 1 of this paper. JSS and JLH are Principal Investigator (PI) and co-PI, respectively, of the current study, and together conceived of and implemented the experimental strategy in close consultation with all other co-authors. JSS initially drafted the manuscript and with JLM incorporated changes suggested by co-authors. JSS was responsible for preparation of all final versions of figures from earlier versions provided by co-authors. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC546213.xml
535564	The association of patient trust and self-care among patients with diabetes mellitus	Background Diabetes requires significant alterations to lifestyle and completion of self management tasks to obtain good control of disease. The objective of this study was to determine if patient trust is associated with reduced difficulty and hassles in altering lifestyle and completing self care tasks. Methods A cross-sectional telephone survey and medical record review was performed to measure patient trust and difficulty in completing diabetes tasks among 320 medically underserved patients attending diabetes programs in rural North Carolina, USA. Diabetes tasks were measured three ways: perceived hassles of diabetic care activities, difficulty in completing diabetes-related care activities, and a global assessment of overall ability to complete diabetes care activities. The association of patient trust with self-management was examined after controlling for patient demographics, physical functioning, mental health and co-morbidities. Results Level of patient trust was high (median 22, possible max 25). Higher trust levels were associated with lower levels of hassles (p = 0.006) and lower difficulty in completing care activities (p = 0.001). Patients with higher trust had better global assessments of overall ability to complete diabetes care activities (p < 0.0001). Conclusion Higher patient trust in physicians is associated with reduced difficulty in completing disease specific tasks by patients. Further studies are needed to determine the causal relationship of this association, the effect of trust on other outcomes, and the potential modifiability of trust	Background Diabetes mellitus is a disease associated with significant morbidity and mortality [ 1 , 2 ]. Patients with diabetes have higher rates of coronary artery disease, retinopathy, neuropathy and nephropathy [ 1 ]. Many of these complications can be prevented with appropriate medical care[ 3 , 4 ] This care, however, often requires significant alterations in lifestyle and strict adherence to self-care tasks, such as checking blood sugars, and taking medications by the patient [ 5 ]. Previous research has shown that patients with diabetes and other diseases often have difficulty in adopting lifestyle changes and completing self-care tasks[ 6 ]. The cause of this poor adherence is multi-factorial and includes patient characteristics such as educational level, [ 7 - 9 ] treatment regimen characteristics, [ 10 ] and characteristics of the clinical setting [ 11 , 12 ]. Patient's perception on the intrusiveness of treatment regimens and their perceived self-efficacy in completing the task have also been demonstrated to affect adherence to diabetes treatment [ 13 , 14 ]. Aspects of the patient-physician relationship such as communication and empathy have been shown to be important to patient's adherence [ 15 - 18 ] and ability to complete self-care tasks [ 7 , 19 , 20 ]. Patient trust is another component essential to the doctor-patient relationship [ 21 ]. Defined variously as a set of beliefs or expectations that the doctor will act in the patient's best interest, or as a reassuring feeling of confidence in the doctor, [ 22 , 23 ] trust provides the foundation for many of the other aspects of the relationship such as communication and empathy [ 23 ]. Despite the significance of trust, there has been relatively little research on this aspect of the doctor-patient relationship. Most studies examining trust have focused on characteristics that predict trust levels [ 24 - 28 ] or the influence insurance characteristics have on trust [ 29 , 30 ]. Limited studies have been done examining the relationship between patient trust and medical outcomes. Safran et al found that adherence to physician recommended lifestyle changes was significantly associated with patient trust levels [ 31 ]. Higher levels of patient trust have also been associated with lower hemoglobin A1c levels in patients with type 2 diabetes [ 32 ]. None have measured trust and its effect on patient's ability in completing self-care tasks. Given the underlying importance of trust in the doctor-patient relationship and the effect of the doctor-patient relationship on adherence and self-care, we hypothesized that higher levels of patient trust in physicians may be associated with reduced difficulty of self-care tasks. To explore this hypothesis, we conducted a survey of patients with diabetes to determine if level of trust is associated with the outcomes of difficulty and hassles in completing self-care tasks. Methods Data Source Project IDEAL (Improving Diabetes Education, Access to Care and Living) was an initiative to increase the quality of care and quality of life for underserved patients with diabetes in North Carolina. Full details of the study have been published elsewhere [ 33 ]. Briefly, fourteen clinical sites across North Carolina were funded by the Kate B. Reynolds Charitable Trust to establish programs to improve the quality of care provided to underserved patients with diabetes. Each developed its own unique diabetes intervention. Evaluation of the initiative was conducted by the Wake Forest University School of Medicine Department of Public Health Sciences using both quality of care and quality of life indicators. To assess the effectiveness of the initiative, a random sample of patients from each program was selected for inclusion at each point in time (1998 pre-initiative and 2001 post-initiative). Quality of care was assessed through a chart review of randomly selected patient. Diabetes Quality of Care (DQIP) [ 34 ] and Health Plan Employer Data and Information Set (HEDIS) [ 35 ] measures were used as standards for quality of care. Quality of life was assessed through a telephone survey of the same patients, allowing both quality of care and quality of life indicators to be linked. The telephone survey was composed of portions of the Medical Outcomes Study Short Form 36 (SF-36) [ 36 ], the DQIP Patient Measures [ 34 ] and Diabetes-39 Health Related Quality of Life instruments [ 37 ]. To determine the relationship between self-management and level of patient trust, an established trust instrument was added to the post intervention survey. Trust Instrument To measure patient trust, we used the Wake Forest University Trust Scale. This scale has been tested in multiple populations and found to be reliable with good construct validity [ 38 , 39 ]. We used the 5 question institutional trust scale as many of the patients were cared for by multiple physicians. This trust measure is a summation of five items that are adapted from a longer trust instrument to ask the patient's view of doctors in general [ 39 ]. The items are: "The doctors at 'this clinic' will do whatever it takes to get patients all the care they need;" "The doctors at 'this clinic' are extremely thorough and careful;" "The medical skills of the doctors here are not as good as they should be;" "You have no worries about putting your life in the hands of the doctors at 'this clinic;" "All in all, you trust the doctors at 'this clinic' completely." A five point scale was used with answers ranging from strongly disagree to strongly agree (1–5), with a higher score indicating a higher level of trust. Self-Management Measures Patient self-care measures were taken from the telephone survey. The methods for this survey have been described elsewhere [ 40 ]. We examined 3 measures: patients' perceived hassle associated with self-care tasks (hassles), patients' difficulty in completing recommended care activities (difficulty), and a global assessment of their ability to care for their disease (care). Briefly, hassle was measured with a 7-item scale that asked patients to rate how much of a problem or hassle it has been to complete diabetes-related tasks over the past 4 weeks. Items in the scale included remembering to take medication and test blood sugar, making meal plans, avoiding particular foods, having to keep a care schedule in mind, organizing the daily routine around a medical care activity, and total time spent in managing their disease. This scale is part of the DQIP instrument [ 34 ]. All questions were measured using a 5-point scale from "no problem" (1) to "a major hassle" (5) with a higher value equalling more hassle. Respondents also had the choice of indicating the question "does not apply". This resulted in a variable number of questions being answered for each patient. To account for this variation, we calculated an average response for each patient by summing the responses and dividing by the number of questions answered by that patient. Our second outcome, difficulty in completing care activities, was measured in a 5-item scale also taken from the DQIP measures [ 34 ]. This scale asked patients about the difficulty they had in completing 5 specific care activities exactly as their doctor suggested. Activities included 1) taking their medication as prescribed, 2) exercising regularly, 3) following their diet, 4) checking their blood sugar, and 5) checking their feet for wounds or sores. Answers were scored on a 5-point scale from "not at all difficult" (1) to "extremely difficult" (5) with an option for "does not apply". Again, to account for variation in the number of questions answered, we calculated an average score, excluding measures that did not apply. The final adherence measure we examined was a global measure of care. This single item asked patients to rate how they "...are at taking care of their diabetes". Possible responses on a 5-point scale ranged from "I stay right on top of it at all times" (1) to "I let it slip way too much" (5). Independent Variables We considered several modifying factors that might influence the relationship between trust and self-care measures. These factors were selected either because they have been shown to influence patient self-care or because they have been used in previous studies on patient trust and outcomes. Modifying factors were obtained from either the survey or through the chart review. Patient's age, gender, race (white, non-white) and insurance (any, none) were obtained from chart review. A co-morbidity score (none, 1–2, greater than 2) was generated based on the number of additional diabetes-related diseases (hypertension, coronary artery disease, non-traumatic amputation, nephropathy, neuropathy, peripheral vascular disease and smoking) identified during chart review. The Medical Outcomes Study Short Form 36 (MOS SF-36) [ 36 ] subscales for physical and mental health were included as part of the patient interview and were used to assess health-related quality of life. The length of the relationship with the health care professional and the number of visits to a provider for diabetes care were obtained from the telephone survey. The use of insulin by the patient was collected from the chart review and was included in the analyses. Imputed values were created for missing measures and analyses were run with and without imputed variables to detect the effect of these variables. Statistical Analysis To determine the association of patient trust and our outcomes of patient hassles, difficulty in completing care activities and global assessment, we used Generalized Estimating Equation[ 41 ] methods to account for clustering of patients within IDEAL program sites. A regression analysis using the xtgee command in STATA/SE 7.0 (STATA Corporation, College Station, TX) was performed for each outcome of interest. Models adjusted for age, gender, race, insurance status, number of co-morbidities, insulin use, number of visits for diabetes in the last year, length of relationship with the provider, physical functioning, and mental health. Results Three hundred and twenty-six individuals with diabetes had information from both the telephone survey and chart review portions of the study. The response rate for the telephone survey was 67%. The mean age of the respondents was 60, over two-thirds were female and forty percent were non-white (see table 1 ). Most respondents (75%) had some sort of insurance, primarily Medicare or a Medicare HMO. Most (79%) had least one co-morbid condition and the average number of co-morbid diseases was one. Average score on the MOS-SF 36 physical functioning and mental health scale was 57.0 (sd 29.8) and 73.5 (sd 21.2), respectively. Diabetes control was generally good with the average hemoglobin A1c 7.3%, although the range was wide (4.4%–14.6%). Twenty four percent of the sample had insulin listed as a treatment in their medical records. When asked about the length of time they had been seeing a provider or clinic, 15% responded less than 1 year, 21% 1–2 years, and 65% more than 2 years. The average number of visits to a provider for diabetes care in the last 12 months was 3. The level of patient trust in their physician was high, with an average and median level of trust of 22 (sd 3, range 11–25). Table 1 Description of the Sample (n = 326) Age, mean 60 (sd 12.0, range 19–89) Gender, female 67% (n = 217) Race, non-white 40% (n = 120) Insurance, none 25% (n = 82) Co-morbidities, any (%) 79% (n = 228) Co-morbidities, mean (%) 1.1 (sd 1.0, range 0–4) Hemoglobin A1c, mean 7.3% (sd 1.7, range 4.4–14.6) Physical Health (max 100) 57.0 (sd 29.8) Mental Health (max 100) 73.5 (sd 21.0) Number of years with provider/clinic* Less than 1 year 13% (n = 44) 1 to 2 years 19% (n = 62) More than 2 years 58% (n = 190) Number of visit in last 12 months 3 (sd 1.0) Patient Trust (max 25) 22 (sd 3, range 11–25) * Number does not equal 100% due to participant who responded that they did not have a regular source of diabetes care When asked about hassles associated with their diabetes, respondents reported the highest degree of hassle with avoiding foods they enjoyed (higher score equals more hassle: mean 2.4, SD 1.4, range 1–5) (table 2 ). Taking medication (mean 1.2, SD 0.7, range 1–5) and testing blood sugar (mean 1.4, SD 0.9, range 1–5) were the care activities associated with the lowest level of hassle. Similarly, patients reported low levels of difficulty when asked about taking their medication for their diabetes (higher score equals more difficulty: mean 1.2, SD 0.5, range 1–4), while they reported higher levels of difficulty in following a diet for their diabetes (mean 2.1, SD 1.0, range 1–5) and exercising regularly (mean 2.5, SD1.3, range 1–5). Overall, the patients surveyed reported high global assessments of their self care for their diabetes (higher score equals worse self care: mean 2.0, SD 0.9, range 1–5). Table 2 Average score for global outcome measures and individual components of each measure Outcome Measure Mean (SD, Range) Hassles associated with diabetes care* Average for entire scale 1.6 (0.7, 1–4.5) Remembering to take diabetes medicine 1.2 (0.7, 1–5) Remembering to test blood for sugar 1.4 (0.9, 1–5) Making meal plans 1.6 (1.2, 1–5) Avoiding foods you enjoy 2.4 (1.4, 1–5) Keeping schedule in mind at times 1.7 (1.2, 1–5) Organizing daily routine around diabetes care 1.4 (0.9, 1–5) Total time spent managing diabetes 1.5 (1.0, 1–5) Difficulty in performing self care activities Average of entire scale 1.7 (0.5, 1–3.8) Taking medications as prescribed 1.2 (0.5, 1–4) Exercising regularly 2.5 (1.3, 1–5) Following diet 2.1 (1.0, 1–5) Checking blood for sugar 1.5 (0.9, 1–5) Checking feet for wounds or sores 1.2 (0.5, 1–5) Global assessment of ability to care for diabetes* 2.0 (0.9, 1–5) * Higher equals more hassles associated with care activity Higher equals more difficulty in completing care activities * Higher equals lower assessment of ability to care In bivariate analyses, a higher levels of hassles was associated with younger age (p < 0.001), non-white race (p = 0.03), and lower score for mental health. There was no association with gender, insurance status, co-morbidities, use of insulin, physical health, length of the relationship with the provider, or the number of visits for diabetes in the past year. A higher difficulty in completing care activities was associated with younger age (p = 0.002), female gender (0.003), non-white race (p = 0.03), worse physical health (p < 0.001) and worse mental health (p = 0.001). There was no association with insulin use, co-morbidities, length of relationship, or number of visits in the past year. Better self rated ability to care for diabetes was also associated with older age (p = 0.004), better physical health (p = 0.02), and better mental health (p < 0.0001). There was no association with gender, race, co-morbidities, use of insulin, length of relationship, or number of visits in the past year. In multivariate regression analysis, level of patient trust was significantly associated with each of the outcomes we examined. Patients with higher levels of trust in their physician reported lower levels of hassles with disease (GEE parameter -0.03, 95% CI -0.05 to -0.008, p = 0.006). Similar finding were seen when self-reported difficulty was examined. Higher level of patient trust was again associated with less reported difficulty in caring for diabetes (GEE parameter -0.02, 95% CI -0.03 to -0.01, p = 0.001). Our final patient outcome, global assessment of ability to care for diabetes, also showed a significant association with higher levels of patient trust associated with higher self-reported ability to care for diabetes (GEE parameter -0.05, 95% CI -0.08 to -0.02, p < 0.0001). Thus, if a patient responded "strongly agree" to all the trust questions rather than "agree" (representing a 5-point increase in level of trust), self-reported hassles would decrease by approximately 0.2 points (about 30% of a standard deviation in hassles). Similarly, for a 5-point increase in trust, self-reported difficulty would decrease by 0.1 points (about 20% of a standard deviation in difficulty) and global ability to care for diabetes would decrease by 0.3 (about 33% of a standard deviation in global ability to care for diabetes). (table 3 ) Table 3 General estimating equations regression parameters for difference in outcomes associated with a 5-point difference in trust Outcome GEE parameter estimates (95% CI) Hassles associated with diabetes care -0.2 (-0.3 to -0.05) Difficulty in performing self care activities -0.1 (-0.20 to -0.05) Global assessment of ability to care for diabetes -0.3 (-0.40 to -0.10) Discussion Diabetes is a disease requiring many types of interventions to prevent the associated morbidity and mortality. These interventions include medications to control elevated glucose levels and finger sticks to check blood sugars. Additionally, significant alterations in lifestyle such as increasing exercise and changing the type of food one eats are an essential portion of the treatment regimen. In our sample of low-income patients with diabetes, we found low reported hassle of completion of diabetes tasks, low reported difficulty in completing diabetes related tasks and good self reported ability to stay on top of their disease. The patients in this study did report more difficulty in making lifestyle changes such as exercising regularly and following recommended diets than taking medications. Other studies examining adherence in diabetic patients have found similar findings. Ruggiero et al, in a nationwide survey of individuals with diabetes, found that over 90% reported always or usually taking their medication but only 64% always or usually followed dietary recommendations and less than half always or usually exercised [ 6 ]. Higher levels of trust were associated with lower reported levels of hassles, lower self-reported difficulty in completing care activities, and improved self-reported global ability to take care of diabetes. These results support our hypothesis of increased completion of self-management tasks with higher levels of patient trust in physicians. There are several possible causes for this association. Patients who are actively engaged in their medical care and jointly make health care decisions with their provider may have less difficulty and hassles in performing self-care activities that they had input on. These patients may also have higher trust levels because they have been engaged as an active participant in the health care decision by their provider. Additionally, medical regimens used to treat chronic disease are complicated. Patients may not fully understand the medical rationale behind particular recommendations such as exercise and diet. Additionally, exercise and diet may not result in immediate improvement in symptoms and often cause initial discomfort or feelings of deprivation, thereby providing little positive feedback and reinforcement. Patients who have high levels of trust may be willing to overcome the initial discomfort and maintain faith that these difficult changes will ultimately be beneficial to them. Our results indicate that an intervention to improve trust that resulted in an increase of 5-points (moving from agree to strongly agree on all questions) would result in an improvement of 0.1–0.3 in our outcome scores. This improvement is equivalent to the difference in self care scores seen between a 60 year old patient and a 30 year old patient. Additionally, trust was significantly associated with all self care measures while insulin use was not. Few studies have examined the association of completion of self-care tasks with patient trust. Safran et al surveyed adults employed by the state of Massachusetts using the Primary Care Assessment Survey which includes a section on patient trust [ 31 ]. They found that patient trust was significantly associated with patient satisfaction and self-reported adherence to lifestyle changes. Thom et al conducted a longitudinal survey of patients cared for in primary care clinics and found that higher levels of patient trust were associated with greater self-reported adherence to prescribed medications [ 42 ]. Our study differs from these by its focus on low-income patients and on measurements of difficulty of self-care in a common, chronic disease. These results add to growing evidence that patient trust is important to patient outcomes, but are limited by the cross-sectional design of the study. Although the results support the contention that higher levels of trust result in decreased hassles and difficulty in completing self-management task, it is possible that the reverse is true. Longitudinal studies of individuals with newly diagnosed chronic diseases such as diabetes are needed to define the temporal nature of this association. Additionally, our outcome measures did not assess whether patients were actually performing the activities, rather, the instruments were designed to determine the level of difficulty and hassles. It is possible that individuals viewed particular activities as very difficult and having a high degree of hassle and still completed the task. Future studies with measurements of both patient perception and actual completion rates are needed. The shortened trust instrument we used lacked the important dimension of patient-centered care. Patients who feel their concerns are listened to and work in partnership with their health care provider may be more likely to attempt new self care activities. It is likely that the inclusion of this dimension through questions such as the providers' ability to listen to or advocate for the patient would strengthen the association of trust with reduced difficulty in completing self care tasks. We also did not collect information on several important mediators of patient trust and self-care such as the empathy level of the health care provider, educational level of the patient, and patients' overall perceived self-efficacy. These additional factors may provide further elucidation into the relationship of patient trust and ability to complete self care task. Despite these limitations, this study adds to our understanding of the predictors of patient self-management of disease. Further studies comparing strategies for measuring trust, measurement of trust at multiple points in time and the linkage of patient trust with clinical outcomes are needed. Conclusions Trust is an integral component of the doctor-patient relationship. Increasingly, external forces such as changes in reimbursement for medical services and increases in the cost of malpractice insurance place pressure on this relationship [ 43 ]. These pressures may erode trust [ 43 ]. If trust is associated with self-care tasks such as adherence to medication or lifestyle changes, the turbulent changes in the health care industry may result in worse health outcomes. Efforts to define the impact of trust on patient ability and comfort in performing self-care activities and to identify the modifiable determinants of trust may translate into improved health outcomes, especially for vulnerable populations. Competing interests The author(s) declare that they have no competing interests. Authors' contributions DEB wrote the text and conducted with the statistical analysis. FB assisted with the statistical analysis and the writing of the paper. RAB was responsible for selection of the quality of care markers, oversaw data abstraction and assisted with writing the article. VTD was responsible for data collection and clinic interventions and assisted with editing the article. RTA was responsible for selection of the quality of life markers and assisted with editing the article. DCG was responsible for the scientific direction of the study, design, and assisted with editing the article All authors read and approved the final manuscript Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535564.xml
509289	Cellular Immune Response to Parasitization in Drosophila Requires the EBF Orthologue Collier	Drosophila immune response involves three types of hemocytes (‘blood cells’). One cell type, the lamellocyte, is induced to differentiate only under particular conditions, such as parasitization by wasps. Here, we have investigated the mechanisms underlying the specification of lamellocytes. We first show that collier (col), the Drosophila orthologue of the vertebrate gene encoding early B-cell factor (EBF), is expressed very early during ontogeny of the lymph gland, the larval hematopoietic organ. In this organ, Col expression prefigures a specific posterior region recently proposed to act as a signalling centre, the posterior signalling centre (PSC). The complete lack of lamellocytes in parasitized col mutant larvae revealed the critical requirement for Col activity in specification of this cell type. In wild-type larvae, Col expression remains restricted to the PSC following parasitization, despite the massive production of lamellocytes. We therefore propose that Col endows PSC cells with the capacity to relay an instructive signal that orients hematopoietic precursors towards the lamellocyte fate in response to parasitization. Considered together with the role of EBF in lymphopoiesis, these findings suggest new parallels in cellular immunity between Drosophila and vertebrates. Further investigations on Col/EBF expression and function in other phyla should provide fresh insight into the evolutionary origin of lymphoid cells.	Introduction Hematopoiesis in Drosophila shares several features with the analogous process in vertebrates. A first population of embryonic hemocyte precursors (prohemocytes) is specified from the head mesoderm very early during embryogenesis. At the end of larval stages and the onset of metamorphosis, a second population of hemocytes is released from a specialised hematopoietic organ, the larval lymph gland ( Rizki and Rizki 1984 ; Tepass et al. 1994 ; Campos-Ortega and Hartenstein 1997 ; Evans et al. 2003 ; Holz et al. 2003 ). Both populations give rise to plasmatocytes, which are dedicated phagocytes, and crystal cells, which are responsible for melanisation of pathogens. Lymph glands contain precursors of a third type of hemocyte that is not generated in embryos, the lamellocyte. Lamellocytes are large, adhesive cells devoted to the encapsulation of foreign bodies too large to be phagocytosed; these cells differentiate only in response to specific conditions, such as parasitization of larvae by Hymenoptera ( Lanot et al. 2001 ; Sorrentino et al. 2002 ). Striking similarities with vertebrate hematopoiesis were revealed when it was shown that Serpent (Srp), a GATA factor, and Lozenge (Lz), a transcription factor related to Runx1/AML1, are required for the development of hemocytes and of crystal cells, respectively ( Rehorn et al. 1996 ; Lebestky et al. 2000 ; Orkin 2000 ). However, except for the observation that gain-of-function mutations in the Janus kinase Hopscotch and in the Toll receptor lead to constitutive production of lamellocytes ( Harrison et al. 1995 ; Luo et al. 1995 ; Qiu et al. 1998 ), the mechanisms and factors underlying the specification of this cell type remain unknown ( Evans et al. 2003 ; Meister 2004 ). During our search for genes involved in specification of lamellocytes, we observed that collier (col) is expressed in the lymph glands at the end of embryogenesis ( Kambris et al. 2002 ). The gene col encodes the Drosophila orthologue of mammalian early B-cell factor (EBF) ( Hagman et al. 1993 ; Crozatier et al. 1996 ), a key factor controlling B-cell lymphopoiesis in mice ( Lin and Grosschedl 1995 ; Maier and Hagman 2002 ). We show here that Col activity is required for specification of the lamellocyte lineage in Drosophila. On the basis of Col expression and col mutant phenotypes, we propose that this factor confers an instructive function on a discrete subpopulation of cells in the Drosophila definitive hematopoietic organ. Results/Discussion Col Expression Identifies Lymph Gland Precursors in Early Embryos We first observed that Col is expressed in Drosophila lymph glands at the end of embryogenesis ( Figure 1 ). In the absence of a specific molecular marker, the embryonic anlage of lymph glands has been mapped to the thoracic lateral mesoderm by lineage analysis of transplanted cells ( Holz et al. 2003 ). By histochemical staining, we observed that Col is expressed in two discrete clusters of cells in the dorsal mesoderm of thoracic segments T2 and T3, starting at the germ-band extension, when lymph gland hemocyte precursors become specified (stage 11; Figure 1 A) ( Holz et al. 2003 ). These clusters of Col-expressing cells grow closer during germ-band retraction before coalescing to form the paired lobes of the lymph glands (early stage 13; Figure 1 B and 1 C). Double staining for Col and Odd-skipped, a lymph gland marker expressed from that stage onward ( Ward and Skeath 2000 ), confirmed that Col-expressing cells are lymph gland precursors ( Figure 1 E). Thereafter, only three to five cells located at the posterior tip of each lobe maintain high levels of Col expression, although low levels are still detected in the other cells of the lymph glands and in some pericardial cells ( Figure 1 D, 1 F, and 1 G). Col expression thus identifies a few cells of the thoracic dorsal mesoderm as the lymph gland primordium and distinguishes a specific posterior region of this hematopoietic organ ( Figure 1 H). The embryonic hematopoietic primordium has been defined as the cephalic domain of Srp expression at the blastoderm stage ( Rehorn et al. 1996 ; Lebestky et al. 2000 ). Srp is not detected, however, in lymph gland precursors prior to stage 12 (Berkeley Drosophila Genome Project gene expression report [ http://www.fruitfly.org/cgi-bin/ex/insitu.pl ] ; Lebestky et al. 2003 ). Consistent with this result, larval hematopoietic progenitors expressing Col are observed in srp 6G (an amorphic allele; Rehorn et al. 1996 ) mutant embryos ( Figure 1 I), indicating that the specification of the embryonic and larval lymph gland progenitors may involve different processes. Figure 1 Col Expression during Lymph Gland Ontogeny (A) Col expression in lymph gland precursors is first observed in two separate clusters of cells (black arrows) in the dorsal-most mesoderm of thoracic segments T2 and T3 at stage 11 (stages according to Campos-Ortega and Hartenstein [1997] ). Col expression in the head region is ectodermal (parasegment 0) and related to its function in head segmentation ( Crozatier et al. 1999 ). (B and C) The clusters of Col-expressing cells get closer between stage 12 and early stage 13 (B) before coalescing (C). (D and E) Col expression becomes progressively restricted to the posterior-most cells of the forming lymph glands (arrowhead) during stage 14, as shown by the partial overlap between Odd-skipped (Odd) and Col expression. (F and G) Enlarged view of lymph glands after completion of embryogenesis, stage 16. Col expression marks the prospective PSC ( Lebestky et al. 2003 ) in a dorsal-posterior position (arrowheads). (H) Schematic representation of Col expression in the lymph glands and pericardial cells in stage 16 embryos. (I) A srp 6G mutant embryo arrested at stage 13. Col is expressed in the presumptive lymph gland primordium (black arrow), although it is not possible to distinguish between high and low levels of expression. All embryos are oriented anterior to the left. (A–C), (G), and (I) are lateral views; (D–F) are dorsal views. (B), (C), and (E–G) are higher magnifications of the dorsal thoracic region. White arrows in (A) and (I) indicate Col expression in a developing dorsal muscle ( Crozatier and Vincent 1999 ). Lamellocyte Differentiation in Response to Parasitization Requires Col Activity Expression of Col in the embryonic lymph gland prompted us to investigate its possible function during larval hematopoiesis. Loss-of-function mutations of col (e.g., col 1 ) are lethal at the late embryonic stage ( Crozatier et al. 1999 ), but lymph glands form normally, indicating that Col activity is not required for formation of the organ per se. Rescue of the embryonic lethality by expressing the col cDNA under the control of a truncated col promoter that is active in the head ectoderm but not in the lymph glands ( Crozatier and Vincent 1999 ) thus allowed us to analyse hematopoiesis in col 1 larvae. The presence of plasmatocytes and crystal cells in the circulation of these mutants indicated that col is not required for specification of either of these lineages ( Table 1 ). We then tested the competence of col 1 larvae to respond to wasp (Leptopilina boulardi) parasitization by producing lamellocytes. This dedicated cellular response is maximal in wild-type (wt) larvae 48 h after wasp egg-laying ( Figure 2 A and 2 B) ( Lanot et al. 2001 ). No circulating lamellocytes were detected in the hemolymph of parasitized col 1 larvae; as a consequence, the wasp eggs were not encapsulated and they developed into parasitic larvae ( Figure 2 C). That this phenotype completely lacked lamellocytes was confirmed by using a lamellocyte marker, misshapen-lacZ, provided by the enhancer trap line l(3)06949 ( Braun et al. 1997 ). Whereas in wt larvae, numerous lacZ -positive cells could be seen adhering to and surrounding wasp eggs, no such cells were detected in col 1 larvae ( Figure 2 D and 2 E). To ascertain that the absence of lamellocytes was the consequence solely of the col mutation, we tested col 1 in transheterozygous combinations with two other col loss-of-function alleles and over the deficiency Df(2R)AN293 ( Crozatier and Vincent 1999 ). In no case did we observe lamellocyte differentiation (we tested 10–20 larvae for each genotype) in response to parasitization by L. boulardi, thereby confirming the critical requirement for Col activity in rendering hematopoietic precursors competent to differentiate into lamellocytes. Although gain-of-function mutations that lead to constitutive activation of either the Janus kinase or the Toll signalling pathways result in hematopoietic defects, including differentiation of lamellocytes in the absence of infestation ( Harrison et al. 1995 ; Luo et al. 1995 ; Qiu et al. 1998 ), col 1 is, to our knowledge, the first identified loss-of-function mutation that abolishes lamellocyte production upon parasitization. Figure 2 col Requirement for Lamellocyte Differentiation (A–C) 4′,6-diamidino-2-phenylindole (DAPI) staining of hemocytes from wt (A and B) and from col 1 (C) third instar larvae. (A) Uninfected larva; (B) and (C) infected larvae. Plasmatocytes (inset in [A]) are always present, whereas lamellocytes (inset in [B]) are detected in the hemolymph of wt (B) but not col 1 (C) larvae 48 h after infestation by L. boulardi. In col 1 mutants, the wasp eggs are not encapsulated (white arrows) and develop into larvae (bottom right organism in [C]). (D–F) Lamellocytes expressing the P- lacZ marker l(3)06949 ( Braun et al. 1997 ) surround the wasp eggs in wt larvae (D), are completely absent in infected col 1 mutant larvae (E), and differentiate in the absence of wasp infection following enforced Col expression in hematopoietic cells ( srpD-Gal4/UAS-col larvae) (F). (G) srpD-Gal4/UAS-col pupa showing the presence of melanotic tumors. Bars: 50 μm. Table 1 Circulating Hemocytes in Third Instar Larvae Values are expressed as mean (SD). Hemocyte types were counted as described in Duvic et al. (2002) a Crystal cells were counted in the three posterior-most segments b The strong adhesive properties of lamellocytes preclude an accurate counting of individual cells c Observed in a fraction of the larvae Bal, balancer chromosome; ND, not determined Enforced Col Expression Triggers Lamellocyte Differentiation in the Absence of Immune Challenge We then asked whether forced expression of Col in hematopoietic cells could induce lamellocyte differentiation in the absence of infestation. Because the e33C-Gal4 line, which drives expression in lymph glands ( Harrison et al. 1995 ) but also epidermis and some other tissues, was lethal in combination with UAS-col, we designed a new Gal4 driver. The driver srpD-Gal4 contains distal elements of the srp gene promoter and drives expression of a UAS reporter gene in prohemocytes and hemocytes (see below) ( Waltzer et al. 2003 ), with a low level of expression in pericardial cells and the fat body (data not shown). Although embryonic-lethal at 25 °C, the srpD-Gal4/UAS-col combination was viable when embryos were allowed to develop to the second larval instar at 18 °C before shifting to 25 °C. Examination of hemolymph samples from late third instar larvae expressing Col under the control of the srpD-Gal4 driver revealed the presence, in a fraction of the larvae, of numerous lamellocytes identified on the basis of both cell morphology and expression of misshapen-lacZ ( Figure 2 F; Table 1 ). Around 5% of all larvae developed melanotic tumors ( Figure 2 G), which have been previously observed in other genetic contexts that lead to overproduction of lamellocytes ( Hou and Perrimon 1997 ). This phenomenon is considered to be a consequence of an autoimmune reaction in which hemocytes encapsulate self-tissue ( Sparrow 1978 ). Thus, we conclude that enforced col expression in hematopoietic cells can induce differentiation of lamellocytes in the absence of immune challenge. We also observed a concomitant drop in the number of circulating crystal cells ( Table 1 ), consistent with the hypothesis that lamellocytes and larval crystal cells could differentiate from a common precursor ( Evans et al. 2003 ). No production of lamellocytes was observed, however, when col expression was targeted to already specified crystal cells or plasmatocytes by using the lz-Gal4 ( Lebestky et al. 2000 ) and hml-Gal4 drivers ( Goto et al. 2003 ), respectively. This indicates that lamellocytes differentiate only when col expression is forced in yet-uncommitted progenitors. Col-Expressing Cells Play an Instructive Role At the end of larval stages, the lymph gland is composed of four to six paired lobes. The two anterior (primary) lobes that formed in the embryo ( Figure 1 ) contain prohemocytes, plasmatocytes, and crystal cells, whereas the posterior (secondary) lobes, which form during the third larval instar, contain predominantly prohemocytes, suggesting that they correspond to a more immature stage of development ( Shrestha and Gateff 1982 ; Lanot et al. 2001 ). Col expression in the anterior primary lobes was found to be restricted to a posterior cluster of about 30–40 posterior cells ( Figure 3 A– 3 C). Consistent with, on average, three to four cell divisions between embryo hatching and the third larval instar—as observed both in circulating hemocytes and imaginal tissues ( Schubiger and Palka 1987 ; Qiu et al. 1998 )—these cells are likely to represent the entire progeny of the three to five cells that strongly express Col in the late embryo (see Figure 1 E and 1 F). They remain clustered at the posterior end of the primary lobes throughout larval development. Col is expressed in a variable number of cells in secondary lobes ( Figure 3 A and 3 C) but is never observed in circulating hemocytes. Despite the dramatic burst of lamellocyte production that occurs in lymph glands when larvae are parasitized (see Figure 2 ) ( Lanot et al. 2001 ; Sorrentino et al. 2002 ), the number and posterior clustering of Col-expressing cells were unchanged ( Figure 3 D and 3 E). This indicates that the small group of Col-expressing cells are not likely to be the direct precursors of lamellocytes, but rather that they play an instructive role in orienting hematopoietic precursors present in the lymph glands towards the lamellocyte lineage. Figure 3 Col Expression in Lymph Glands of Third Instar Larvae (A and B) Col is expressed in the primary lobes, in a posterior cluster of cells (arrow), and in a variable number of secondary lobes. Low expression is also detected in some pericardial cells (asterisks), the significance of which remains unknown. PI, propidium iodide. (C) Schematic representation of the lymph glands and Col expression in late third instar larvae. (D and E) Col expression 24 h (D) and 48 h (E) after wasp infection; despite strong cell proliferation, including in secondary lobes, Col expression remains unchanged (black arrow). (F–H) Overlap between Ser-lacZ ( Bachmann and Knust 1998 ) and Col expression in PSC cells; note a few scattered Ser-expressing cells that do not stain for Col. Bars: 50 μm (A, B, D, and E) ; 10 μm (F–H). Col expression in a posterior cluster of cells of the primary lobes is reminiscent of that of Serrate (Ser), a Notch ligand ( Lebestky et al. 2003 ). The Ser/Notch pathway has recently been shown to be essential for crystal cell development ( Duvic et al. 2002 ; Lebestky et al. 2003 ). Analysis of clones of Ser mutant cells in the larval lymph glands further indicated that Ser-expressing cells are responsible for activation of Lz expression in surrounding cells and their commitment to a crystal cell fate ( Lebestky et al. 2003 ). Together with the Ser expression pattern, this observation led the authors to propose that the posterior cluster of Ser-expressing cells could act as a signalling centre, which they termed the posterior signalling centre (PSC). Through double-labelling experiments, we confirmed the overlap between Col and Ser expression (as visualised by Ser-LacZ [ Bachmann and Knust 1998 ]) in the posterior cells of the primary lobe ( Figure 3 F– 3 H). However, Ser, but not Col, is expressed in scattered cells throughout the primary lymph gland lobes in addition to the PSC ( Figure 4 ) ( Lebestky et al. 2003 ). Figure 4 PSC-Specific Gene Expression Is Dependent upon Col Activity PSC-specific expression of col , Ser-lacZ, and Ser (arrowhead in [A], [C], and [E]) is lost in col 1 mutant larvae (B, D, and F); only Ser expression in scattered cells is maintained (arrow in [E] and [F]). Bar: 50 μm. PSC-Specific Gene Expression Is Dependent upon Col Activity Because Col expression and function suggested that the PSC was playing an instructive role in orienting other lymph gland cells towards the lamellocyte fate, we asked whether Col was necessary for the PSC to form properly. We looked at col and Ser expression in col 1 mutant lymph glands, using in situ hybridisation for col because Col antibodies do not recognise the Col 1 protein ( Crozatier and Vincent 1999 ). In wt larvae, consistent with the results of immunostaining, col transcripts were restricted to the PSC ( Figure 4 A). In contrast, we could not detect col expression in col 1 mutant lymph glands ( Figure 4 B). Furthermore, expression both of Ser-lacZ and Ser in the PSC ( Figure 4 C and 4 E) was also abolished ( Figure 4 D and 4 F), indicating that proper specification of PSC identity is dependent upon Col activity. Although Ser expression was lost from the PSC region, it was still observed in scattered cells in the primary lobe ( Figure 4 E and 4 F, arrows), suggesting that Ser-lacZ expression reflected the presence of a PSC-specific transcriptional enhancer without reproducing the entire Ser expression pattern. Evidence for a Bipotential Crystal Cell/Lamellocyte Precursor Ser signalling through the Notch signalling pathway is critical for the specification of crystal cell precursors ( Duvic et al. 2002 ; Lebestky et al. 2003 ). However, numerous crystal cells differentiate in col mutant lymph glands, including in secondary lobes, despite the loss of Ser expression in the PSC (see Figures 4 E, 4 F, 5 A, and 5 B). These data, together with the clonal analysis of Lebestky et al. (2003) , lead us to conclude that crystal cell development is triggered by signalling from the scattered Ser-expressing lymph gland cells, rather than from the PSC itself. In contrast, no differentiating lamellocytes could be detected in col mutant lymph glands, even under conditions of wasp infestation that induced massive lamellocyte differentiation in wt glands ( Figure 5 C– 5 F), confirming the key role of the PSC in this process. Figure 5 Col-Expressing Cells Play an Instructive Role in Lamellocyte Production Expression of the crystal cell marker doxA3 ( Waltzer et al. 2003 ) (A, B, and G); of the lamellocyte markers α-ps4 (M. Meister, unpublished data) (C–F and H) and L1 ( Asha et al. 2003 ) (J); and of Col (I and J); in wt (A, C, and E), col loss-of-function mutant (B, D, and F), and srp-Gal4/UAS-col (G–J) larvae. In (E) and (F), larvae were taken 48 h after infestation. An increased number of doxA3 -positive cells (B) parallels the absence of lamellocyte differentiation (F) in col 1 mutant lymph glands. Conversely, lamellocyte differentiation and a reduced number of doxA3 -positive cells are observed upon enforced Col expression (G and H). Double staining for Col and L1 shows that Col-expressing cells and differentiating lamellocytes do not overlap in the lymph gland. (I) shows ectopic Col expression compared to expression in the PSC (arrowhead; not visible in [J]). Antibody and in situ probes are indicated on each panel. In all panels, larvae are oriented with the head to the left: a single primary lobe is shown, with sometimes a few secondary lobes. Bar: 50 μm. We then looked at the production of crystal cells and lamellocytes in lymph glands with enforced Col expression ( srpD-Gal4/UAS-col ; Figure 5 G– 5 J). Very few crystal cells and numerous lamellocytes were observed, consistent with the circulating hemocyte picture ( Figure 5 G and 5 H). The srpD-Gal4 -driven Col expression in the lymph gland is not uniform. Some cells express high levels when compared to the PSC, whereas many others show no detectable expression. A similar pattern was also observed in combination with UAS-lacZ ( Figure 5 I; data not shown). Double-labelling experiments showed that the lymph gland cells induced to differentiate into lamellocytes surround but do not overlap with the Col-expressing cells ( Figure 5 J), confirming the instructive role of Col-expressing cells. In all genotypes that we tested, we found equally large numbers of plasmatocytes in the lymph glands (data not shown), which indicates that this cell type is not affected by col loss-of-function and gain-of-function mutations. Altogether, the absence of lamellocytes after parasitization that is associated with the increase in the number of crystal cells in col mutant lymph glands, and the opposite situation in srpD-Gal4/UAS-col lymph glands ( Figure 5 ; Table 1 ), support the existence of bipotential crystal cell/lamellocyte precursors. A Model for Induction of Lamellocytes in Response to Parasitization In summary, our data show that (i) Col expression defines a specific group of cells within the lymph glands; (ii) lamellocyte differentiation, which is an exclusive feature of lymph gland hematopoiesis, depends upon Col activity; and (iii) the massive production of lamellocytes that follows parasitization does not involve changes in Col expression. We thus propose a two-step signalling model for induction of lamellocytes in response to wasp egg-laying ( Figure 5 A). According to this scheme, Col endows PSC cells with the competence to respond to a primary signal emitted by plasmatocytes as these permanent immune supervisors form a first layer around the parasite egg ( Russo et al. 1996 ). Subsequently, PSC cells send a secondary signal that orients prohemocytes towards the lamellocyte fate. The production of lamellocytes upon enforced col expression suggests that the need for the primary signal to activate the secondary signal can be bypassed in overexpression experiments. Although several aspects of this model remain to be translated into molecular terms, it certainly sheds a new light on the genetic control of hemocyte lineages in Drosophila. Concluding Remarks B- and T-lymphocytes mediate adaptive immunity, a phylogenetically recent component of the immune system as it is found only in gnathostomes ( Kimbrell and Beutler 2001 ; Mayer et al. 2002 ). How adaptive immunity emerged during evolution, and was built on top of the innate immune system by which it is controlled and assisted, remains a fascinating question. The requirement for Col function in the Drosophila cellular immune response, and EBF function in B-cell development in vertebrates, suggests that Col/EBF function was co-opted early during the evolution of cellular immunity. A puzzling question remains, however, of how the cell-autonomous function of EBF in B-cell development, and the non–cell-autonomous function of Col in lamellocyte development, could relate to an ancestral Col/EBF function. We would like to propose that the ancestral expression of Col/EBF in a subset of hematopoietic cells conferred on these cells the ability to respond to signals from circulating immune supervisors (generically designated as macrophages in Figure 6 ) and provide a secondary line of defence against specific intruders. This cell-specific property in turn laid the ground for the emergence of the vertebrate lymphoid cells on one side and the Drosophila PSC on the other. Although admittedly highly speculative, this proposal takes into account the following considerations. B-cell development represents the default fate of lymphoid progenitors ( Schebesta et al. 2002 ; Warren and Rothenberg 2003 ). Although specification of B-cells critically depends on EBF (and the basic helix-loop-helix protein E2A), commitment depends on another gene, Pax5. The Pax5 −/− pro–B-cells retain the ability to generate a whole range of both ‘innate’ myeloid and lymphoid cells ( Nutt et al. 1999 ; Rolink et al. 1999 ; Mikkola et al. 2002 ). Thus, the ontogeny of the B-cell lineage from preexisting myeloid cell types has occurred through several steps, one key event being the co-opting of Pax5, acting downstream of EBF, for which there is no known counterpart in Drosophila hematopoiesis. Second, the co-opting of Col activity for lamellocyte differentiation in larval hematopoiesis most likely came on top of a preexisting hematopoietic system, such as that operating in Drosophila embryos ( Evans et al. 2003 ; Meister 2004 ). Further investigation of Col/EBF functions in intermediate phyla should provide more insight into the diversity of myeloid lineages and ontogeny of the lymphoid lineages during evolution. Figure 6 A Model for Lamellocyte Specification (A) A model for the induction of lamellocyte differentiation in the Drosophila lymph glands in response to wasp parasitization. Col enables PSC cells to respond to a primary signal (S1) that is likely emitted by plasmatocytes upon their encounter with a parasite ( Russo et al. 1996 ; Meister 2004 ). As a result, the PSC cells send a secondary signal (S2) that causes prohemocytes to develop into lamellocytes. Notch (N) signalling instructs a fraction of prohemocytes to become crystal cells ( Duvic et al. 2002 ; Lebestky et al. 2003 ). The circular arrow indicates that increased proliferation leading to increased numbers of crystal cells and lamellocytes follows parasitization ( Sorrentino et al. 2002 ). (B) Schematic view of hematopoiesis in Drosophila and mouse . Left: Lymph gland cells contain two types of hematopoietic cells, PSC cells and uncommitted precursors. These precursors can give rise to either plasmatocytes or crystal cells. Crystal cell precursors can also give rise to lamellocytes upon receiving a signal from the PSC cells expressing Col (dotted arrows); this signalling is itself dependent upon a communication between circulating plasmatocytes and the PSC (A). Right: In mice, hematopoietic stem cells (HSC) give rise to common myeloid precursors (CMP) and common lymphoid precursors (CLP) (adapted from Orkin [2000] and Schebesta et al. [2002] ). Signalling between CMP- and CLP-derived cells is an essential component of adaptive immunity. Col and EBF functions, in Drosophila and vertebrate hematopoiesis, respectively, suggest an ancestral role in their conferring on a subset of hematopoietic cells the ability to respond to signals from circulating immune supervisors (generically designated here as macrophages) and to provide a secondary line of defence against specific intruders. Materials and Methods Fly stocks and hemocyte counting Unless otherwise stated, all fly stocks were maintained at 25 °C on standard medium, and genotypes were verified with marked balancer chromosomes. For wasp infection, second instar larvae were submitted to egg-laying by L. boulardi for 2–4 h, then allowed to develop at the appropriate temperature and analysed 24 or 48 h later. Hemocyte observation and counting, and lacZ staining of lamellocytes, were as previously described ( Braun et al. 1997 ; Duvic et al. 2002 ). Transgenic constructs and flies The srpD-Gal4 transgene: A distal promoter fragment, between 8.8 and 6 kb upstream of the srp transcription start site and a 340-bp fragment overlapping this site were amplified by PCR using 5′-GCTAGCGACGCGTGATGCAACTTAATCAA-3′ and 5′-CTGCAGTTTATGAATGGAAGACGCGGACG-3′ primers, and 5′-CTGCAGACGGCCAAGTCCAACAACAACAA-3′ and 5′-GGATCCCTGTTGCTGCTGTAACTGTTGAT-3′ primers, respectively, then fused before subcloning upstream of the Gal4 coding sequence in a pCaSpeR vector. Transgenic lines were obtained by standard procedures. Because they are embryonic-lethal at 25 °C, the srpD-Gal4/UAS-col animals were kept at 18 °C before shifting to 25 °C at the second larval instar. Immunostaining and in situ hybridisation. Immunostaining and in situ hybridisation of larval lymph glands and embryos were performed as in Crozatier and Vincent (1999) using rabbit anti-Col (1:250), rat anti-Ser (gift from K. Irvine; 1:500), mouse anti-β-galactosidase (Promega, Madison, Wisconsin, United States; 1:1000), mouse lamellocyte-specific L1 (gift from I. Ando; 1:10), and guinea-pig anti-Odd-skipped (gift from J. B. Skeath; 1:100). Peroxidase and Alexa Fluor 546 or 488 labelled secondary antibodies (Molecular Probes, Eugene, Oregon, United States) were used at a 1:500 dilution. In some cases, lymph glands were incubated for 30 min at 37 °C in a propidium iodide solution in the presence of RNase. Mounting in Vectashield medium (Vector Laboratories, Burlingame, California, United States) preceded observation by confocal microscopy (Zeiss LSM 510 [Zeiss, Oberkochen, Germany] and Leica SP2 [Leica, Wetzlar, Germany]). Single-stranded digoxigenin-labelled RNA probes were synthesised from corresponding cDNAs cloned in pGEM (Promega).	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC509289.xml
387823	Emergence of a Peaceful Culture in Wild Baboons	null	For most animal species, behavioral attributes are largely the product of interactions between genes and environment, with behavioral patterns preserved by natural selection. Birds, for example, know instinctively what type of nest to build for their offspring; salamanders don't need lessons to swim. But when it comes to primates—including humans—a good deal of behavior is learned. Primates exhibit a wide range of behaviors, not just among species but also among populations and even individuals. Yet the nature versus nurture debate still rages, particularly when it comes to understanding the roots of aggression. While bonobos are famous for using sex to resolve disputes, aggression is far more common in most primate species—again humans included. Our closest relative, the chimpanzee, has a reputation for being among the most belligerent, with rhesus monkeys and baboons not far behind. For many of these species, bouts of violence are often followed by gestures of reconciliation, such as grooming or, in the case of chimps, kissing. Since most primates live in social groups, it may be that such conciliatory measures serve to maintain some semblance of social structure, offsetting the disruptive effects of aggression. (To learn more about primate behavior and aggression, see the primer by Frans de Waal in this issue [DOI: 10.1371/journal.pbio.0020101 ].) In baboons, “grooming” is a socially rewarding behavior. (Photograph, with permission, by Robert Sapolsky) Primatologists characterize these behavioral differences as “cultural” traits, since they arise independent of genetic or environmental factors and are not only shared by a population (though not necessarily a species) but are also passed on to succeeding generations. Such cultural traditions have been documented in African chimp populations, which display over 39 behaviors related to “technology” (such as using stones to crack nuts), grooming, and courtship. While most of these cases involve either tools, foraging, or communication, Robert Sapolsky and Lisa Share report evidence of a higher order cultural tradition in wild baboons in Kenya. Rooted in field observations of a group of olive baboons (called the Forest Troop) since 1978, Sapolsky and Share document the emergence of a unique culture affecting the “overall structure and social atmosphere” of the troop. In his book A Primate's Memoir , Sapolsky studied the activities and lifestyle of the Forest Troop to explore the relationship between stress and disease. In typical baboon fashion, the males behaved badly, angling either to assume or maintain dominance with higher ranking males or engaging in bloody battles with lower ranking males, which often tried to overthrow the top baboon by striking tentative alliances with fellow underlings. Females were often harassed and attacked. Internecine feuds were routine. Through a heartbreaking twist of fate, the most aggressive males in the Forest Troop were wiped out. The males, which had taken to foraging in an open garbage pit adjacent to a tourist lodge, had contracted bovine tuberculosis, and most died between 1983 and 1986. Their deaths drastically changed the gender composition of the troop, more than doubling the ratio of females to males, and by 1986 troop behavior had changed considerably as well; males were significantly less aggressive. After the deaths, Sapolsky stopped observing the Forest Troop until 1993. Surprisingly, even though no adult males from the 1983–1986 period remained in the Forest Troop in 1993 (males migrate after puberty), the new males exhibited the less aggressive behavior of their predecessors. Around this time, Sapolsky and Share also began observing another troop, called the Talek Troop. The Talek Troop, along with the pre-TB Forest Troop, served as controls for comparing the behavior of the post-1993 Forest Troop. The authors found that while in some respects male to male dominance behaviors and patterns of aggression were similar in both the Forest and control troops, there were differences that significantly reduced stress for low ranking males, which were far better tolerated by dominant males than were their counterparts in the control troops. The males in the Forest Troop also displayed more grooming behavior, an activity that's decidedly less stressful than fighting. Analyzing blood samples from the different troops, Sapolsky and Share found that the Forest Troop males lacked the distinctive physiological markers of stress, such as elevated levels of stress-induced hormones, seen in the control troops. In light of these observations, the authors investigated various models that might explain how the Forest Troop preserved this (relatively) peaceful lifestyle, complete with underlying physiological changes. One model suggests that nonhuman primates acquire cultural traits through observation. Young chimps may learn how to crack nuts with stones by watching their elders, for example. In this case, the young baboon transplants might learn that it pays to be nice by watching the interactions of older males in their new troop. Or it could be that proximity to such behavior increases the likelihood that the new males will adopt the behavior. Yet another explanation could be that males in troops with such a high proportion of females become less aggressive because they don't need to fight as much for female attention and are perhaps rewarded for good behavior. But it could be that the females had a more direct impact: new male transfers in the Forest Troop were far better received by resident females than new males in the other troops. Sapolsky and Share conclude that the method of transmission is likely either one or a combination of these models, though teasing out the mechanisms for such complex behaviors will require future study. But if aggressive behavior in baboons does have a cultural rather than a biological foundation, perhaps there's hope for us as well.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC387823.xml
535558	Use of the bootstrap in analysing cost data from cluster randomised trials: some simulation results	Background This work has investigated under what conditions confidence intervals around the differences in mean costs from a cluster RCT are suitable for estimation using a commonly used cluster-adjusted bootstrap in preference to methods that utilise the Huber-White robust estimator of variance. The bootstrap's main advantage is in dealing with skewed data, which often characterise patient costs. However, it is insufficiently well recognised that one method of adjusting the bootstrap to deal with clustered data is only valid in large samples. In particular, the requirement that the number of clusters randomised should be large would not be satisfied in many cluster RCTs performed to date. Methods The performances of confidence intervals for simple differences in mean costs utilising a robust (cluster-adjusted) standard error and from two cluster-adjusted bootstrap procedures were compared in terms of confidence interval coverage in a large number of simulations. Parameters varied included the intracluster correlation coefficient, the sample size and the distributions used to generate the data. Results The bootstrap's advantage in dealing with skewed data was found to be outweighed by its poor confidence interval coverage when the number of clusters was at the level frequently found in cluster RCTs in practice. Simulations showed that confidence intervals based on robust methods of standard error estimation achieved coverage rates between 93.5% and 94.8% for a 95% nominal level whereas those for the bootstrap ranged between 86.4% and 93.8%. Conclusion In general, 24 clusters per treatment arm is probably the minimum number for which one would even begin to consider the bootstrap in preference to traditional robust methods, for the parameter combinations investigated here. At least this number of clusters and extremely skewed data would be necessary for the bootstrap to be considered in favour of the robust method. There is a need for further investigation of more complex bootstrap procedures if economic data from cluster RCTs are to be analysed appropriately.	Background The greater complexity of cluster randomised controlled trials (RCTs) compared with their individually randomised counterparts has led to much methodological work concerning their design and analysis[ 1 ]. However, the analysis of cost data from these trials has received little attention to date. The conceptual issues arising in this context have been explored [ 2 ] but, briefly, there are two problems. The first is that many trials randomise only a small number of clusters. This can sometimes produce inadequate randomisations where, for example, all clusters with a characteristic related to outcome are allocated to one treatment arm [ 3 ]. It may also be difficult to make inferences on cluster level covariates and between-cluster variability. Although in theory this problem is just as relevant to clinical data, in practice methods of analysis based on the t -test are fairly robust to moderate violations of the assumptions of normality and homogeneity of variances [ 4 ]. The situation with highly skewed continuous economic data, however, may be more serious. Indeed, at either the individual or cluster level, skewed data are potentially very problematic, particularly with small numbers of clusters. Appeals to normality of data may not be reasonable for the distribution of cluster means, given variation in medical practice, social and geographical factors. In individually randomised trials, problems of skewness and small sample sizes have sometimes resulted in confidence intervals with poor coverage properties (such as negative lower limits for mean costs). In such circumstances economic data have been analysed using methods such as the nonparametric bootstrap [ 5 ], first proposed by Efron [ 6 ]. This relies on computer-intensive resampling methods rather than a formula and commensurate appeals to the central limit theorem. In essence, by treating the sample at hand as the population, repeated resampling with replacement from this 'population' and calculation of a parameter of interest builds up a picture, the 'empirical distribution' of this parameter, based on so-called ' B bootstrap estimates' of the parameter of interest. This can be used to construct directly the required confidence interval by, for instance, reading off the 2.5% and 97.5% percentiles of the distribution. Bootstrapped confidence intervals may, therefore, be asymmetric and be better able to deal with skewed data. The bootstrap approach is flexible but does assume that the data are independently and identically distributed [ 7 , 8 ]. When stratification, cluster sampling or probability weights are introduced into sampling this assumption is violated and the bootstrap as described above will give incorrect inferences. Work has been carried out in the 1980s and 1990s to generalise the bootstrap to survey sampling and regression analysis [ 7 , 9 ]. The bootstrap is included in some standard statistical packages, but it is often overlooked that confidence intervals from this may have poor coverage properties when there is a small number of clusters – a phenomenon that is common in cluster RCTs. This paper details the results of simulation studies to evaluate how clustered cost data might be analysed. Small numbers of clusters together with skewed data were utilised to ascertain how the bootstrap performed against a method of analysis commonly used in clustered clinical data. Thus it details the generation and analysis of a single outcome model. This single outcome model was primarily conceptualised as a model for cost data and the term cost is therefore used as short-hand. However, some of the scenarios presented may be equally applicable to clinical data. Methods The performances of confidence intervals for simple differences in mean costs utilising a robust (cluster-adjusted) standard error and from two cluster-adjusted non-parametric bootstrap procedures were compared in terms of confidence interval coverage. Specifically, the comparison was of the percentage of 20,000 simulations for which the estimated 95% confidence interval contained the true value of the treatment effect. If the observed coverage were to be 95% on average across the simulations, then 20,000 simulations would from simple binomial theory give a margin of error of approximately , which was considered acceptable. Over 20,000 simulations per parameter set, then, the following were noted: 1. The mean value (over all simulations) of the estimated treatment effect and the confidence limits under each of the three procedures, 2. The percentages of simulations for which the estimated confidence interval did not contain the true value of the treatment effect (zero) and whose lower/upper limit was greater/less than this true value, 3. The percentage of simulations for which the estimated confidence interval contained the true value for the treatment effect. This figure was simply 100 minus the sum of the two percentages in 2 above. Thus ideally the two figures in 2 should each be 2.5% whilst that for 3 should be 95% (the nominal level). It was decided to split observed non-coverage rates according to whether there was spurious positive treatment effect or a spurious negative one because skewed distributions were expected to have different implications for each of these. Data generation process The data were constructed by assuming there were n individuals in each of 2k clusters. Half of these were randomised to a hypothesised intervention group, whilst the other half were randomised to a control group. A random effects model incorporating a treatment dummy variable was used: E hij = α hi + β T h + ε hij h = 0,1; i = 1,... k ; j = 1,... n ; E ( α hi ) = 0; E ( ε hij ) = 0 Thus the j th individual in the i th cluster was randomised to receive treatment h . Each individual's outcome, E hij , comprised three elements: the effect of treatment, β T h , the cluster-specific effect, α hi , and the individual-specific effect, ε hij . In order to inform the values of these parameters, it was necessary to undertake some exposition of how cost data might be distributed in a cluster RCT. Conceptualising costs in a cluster RCT In attempting to construct realistic scenarios for cost data from a cluster RCT, three main factors were considered: 1. What is the nature of the distribution of individual patient costs expected to be in the population of patients normally eligible for treatment? 2. How representative of this population distribution are the cost distributions within clusters likely to be? This has implications for the ICC and the assumptions regarding the distributions. 3. How might the introduction of an intervention affect 2 above? As detailed previously, the existence of one unrepresentative cluster (such as a London teaching hospital) in one arm may affect the ICC, independently of treatment, or the treatment could directly change the ICC and the distribution [ 2 ]. Parameters varied in the simulation model This section formally sets out the parameters which, when varied in the simulation model, attempted to capture two potential extreme scenarios as well as situations in between. Under the first scenario, all clusters are equally representative of the population, leading to a high degree of skewness within most, if not all, clusters. Under such a scenario, the ICC is likely to be small and results would be expected to be robust to any incorrect assumptions made regarding the between-cluster distribution. Under the second scenario, within-cluster costs are expected to be more homogenous and much of the skewness in the cost data at the population level is attributable to differences in the cluster mean costs. As a result, the ICC would be expected to be much larger, and the between-cluster distribution might be expected to exhibit considerable skewness. Six factors were varied: 1. The between-cluster distributions, 2. The within-cluster distributions, 3. The ICC in the control group, 4. The ICC in the intervention group, 5. The number of clusters in each intervention arm, 6. The number of individuals in each cluster. The between-cluster distributions There were two distributional assumptions used for the cluster means. The first was the normal distribution and the second was the lognormal distribution, which, for a given mean and variance, exhibits higher kurtosis and skewness than the gamma distribution, the main alternative skewed distribution. For each of the ICC combinations given below, there were three possible combinations of the between-cluster distributions: 1. The cluster means in both the control and intervention groups were normally distributed 2. The cluster means in both the control and intervention groups were lognormally distributed 3. The cluster means in the (nominal) control group were normally distributed, whilst in the (nominal) intervention group they were lognormally distributed. The within-cluster distributions The same distribution was used for individual level data in both treatment arms in order not to further increase the number of parameter combinations possible. Lognormal data were used in order to provide a scenario applicable to individual level cost data. The ICC in the control group Given that the likely range of cost ICCs is largely unknown, values between zero and 0.5 were used. While high, the upper value is consistent with the one published source of cost ICCs [ 10 ]. The ICCs used were 0.01, 0.1 and 0.25 for one of the two treatment groups. The value 0.5 was not used for the control group but was achieved in the intervention group as a result of changes in the ICC (see below). The total variance, equal to the between-cluster variance plus the within-cluster variance, was arbitrarily fixed at 100. The ICC in the intervention group For a given value of the ICC in the control group, the intervention ICC can remain the same or it can change in a number of ways. In particular, the intervention ICC: 1. Remained the same as the control ICC, 2. Doubled, as a result of an appropriate increase in the between-cluster variance, 3. Doubled, as a result of an appropriate decrease in the within-cluster variance, 4. Halved, as a result of an appropriate decrease in the between-cluster variance, 5. Halved, as a result of an appropriate increase in the within-cluster variance. Although it is only the between-cluster variance or the within-cluster variance that changes at any one time (not both), the changes involved are large ones. Hence these extreme scenarios should cover a range of findings. The number of clusters in each group The number of clusters in each group was 6, 12 or 24. These figures reflect the small numbers of clusters recruited in many cluster RCTs and, coupled with the cluster sizes given below, they allowed alternative combinations of cluster size and number of clusters to be investigated for a given total trial size. The number of individuals in each cluster The cluster size was 25, 50 or 100. Mean cluster sizes between 50 and 100 are not unusual in health services research trials, but there is enormous variation in cluster size, depending upon treatment area and type of cluster[ 1 ]. Comparison of methods that allow for clustering The first method utilises a standard procedure, where 'standard' has been taken to mean a 95% confidence interval quoted for continuous data in packages such as Stata [ 11 ], utilising a point estimate and a Huber-White (robust) cluster-adjusted standard error [ 12 - 14 ]. Bootstrapping was performed for the two other methods, as described in Davison and Hinkley [ 7 ]. Under both methods the sampling structure was maintained in a bootstrap replication by selecting k clusters with replacement from the treatment group and selecting k clusters with replacement from the control group – in other words, resampling of clusters was stratified by intervention group. Under method 1 all individuals within a resampled cluster were then selected. Under method 2 a second level of bootstrap was performed on individuals within clusters selected at level one. The difference between the two randomisation group means was then calculated for each method. This was repeated to give 1000 bootstrap estimates (estimates performed on the resampled data) of the treatment effect. A bias-corrected and accelerated (BC a ) confidence interval was then estimated at the same nominal 95% level as for the robust method [ 8 ]. Given the nature of the BC a method, the resulting confidence interval need not be symmetric. The bootstrap methods are described in more detail below. Method 1 Under this procedure (BS1), clusters are bootstrapped and each resampled cluster is kept intact. This method is utilised by Stata [ 11 ] when the cluster() option is added to the bootstrap command. Suppose that within a randomisation group, for each of k clusters, n responses are obtained, y ij , such that y ij = α i + ε ij i = 1,... k ; j = 1,... n . The α i s are sampled randomly from the distribution F α and the ε ij s are independently sampled randomly from the distribution F ε . E( α i ) = 0; (1) It can be shown that for the bootstrap estimates (with superscript asterisks): When compared with (2) and (3) it should be noted that the expected variance and covariance of the resampled outcome data are slightly biased downwards. However, an estimator such as the sample mean is strongly consistent (in that its bias is zero and its variance tends to zero as the total sample size approaches infinity); the level of bias is small unless the number of clusters becomes very small. Method 2 An alternative method (BS2) involving resampling individuals as well as resampling whole clusters was also considered. This uses a first stage bootstrap applied to the estimated cluster means (sampling with replacement). The second stage, in which individuals are bootstrapped, involves resampling the deviations from the estimated cluster means. However, the estimated cluster means incorporate both within and between-cluster variability and any analysis that restricts itself to the cluster means will over-estimate the variance in these means [ 7 ]. By incorporating the deviations from the estimated cluster means we have, in effect, double-counted the within-cluster variance. Therefore, the cluster means were shrunk using Davison and Hinkley's shrinkage estimates, (see page 102 of their book): where c is given by ; if the right hand side is negative, it is reset to zero. The variance of the adjusted cluster means, , is then . The deviations from the estimated cluster means were also standardised to Finally, for all resampled clusters, the 'shrunken' mean is added to the standardised deviation for each resampled individual: This method, with the rescaling procedures will in future be referred to as the BS2 method, or double bootstrap. Results The primary focus of this work was on confidence interval coverage and rejection rates; the estimated confidence limits are, therefore, not presented but general inferences regarding the relative width of various confidence intervals can be made easily from the coverage and rejection rates presented below. Coverage rates for cost confidence intervals As a summary, Tables 1 to 3 show coverage rates for each of the three methods of analysis for each sample size combination when averaged over the 15 parameter/distribution combinations for a given control group ICC – that is, three distributional assumptions for each of five ICC combinations. Issues specific to variance or distribution combinations are presented below. In the three tables, within each box the first number represents the coverage of the robust confidence interval, the second represents the coverage of the BS1 method whilst the third figure represents that of the BS2 method. A number of results are immediately apparent: • All three methods produce coverage of less than 95%, the nominal level, • All three methods appear to be consistent with respect to the impact of the number of clusters per arm. In other words, as the number of clusters per arm increases, the observed coverage approaches 95% for each method, • The coverage of the robust method is within approximately 1.5% of 95%, • The BS1 method is always outperformed by the other two methods. In other words the other two methods always achieve coverage that is closer to 95%, • The BS2 method performs much better than the BS1 method but never as well as the robust method, • As the ICC in the control group increases (that is, across tables) the robust method performs slightly worse whilst the performances of the bootstrap methods are noticeably poorer, • When examining numbers along the diagonal in each figure (that is, for equivalent total sample sizes), the bootstrap methods perform much better for a large number of clusters and small cluster size compared with vice-versa. This is probably due to the slight downward bias in the second moments; the degree of bias is an inverse function of the number of clusters. From the results in these tables there does not appear to be much to commend the bootstrap, since both bootstrap methods are always outperformed by the robust method. Moreover, when examining the confidence interval coverage results split by distribution and variance combination (results not shown), there was only one parameter combination for which the robust method was outperformed and this was by an amount that was consistent with Monte Carlo sampling error. However, the rejection rates for confidence intervals were split to ascertain if any method was noticeably better at taking account of the skewness in the data. Rejection rates for confidence intervals When examining the rejection rates for six clusters of size 25 (results not shown), those of both bootstrap methods virtually always exceeded those under the robust method. The exceptions were for a control ICC of 0.01 when the within-cluster variance increased as a result of the intervention for two of the distributional combinations. For the third distributional combination the rejection rates were the same. Since the individual level data were always lognormally distributed (reflecting typical cost data at this level), an increase in the within-cluster variance for lognormally distributed data is likely to have a large effect on skewness, provided the ICC is not too large (which would dilute the effect of within-cluster factors). Thus, if the desired criterion is ability to match the nominal 2.5% rejection rate in any given direction, there are occasions when one or both bootstrap methods outperform the robust method in terms of the proportion of lower rejections. In particular, this appeared to happen when the ICC was moderate to large, together with an extremely skewed distribution of the treatment effect, typically achieved by a large change in the ICC and something other than a normal, normal (N, N) distribution combination. Under these circumstances the distribution of the treatment effect is most skewed, since the cluster means in the intervention group are exhibiting large skewness by way of a lognormal distribution whose (already moderate to large) variance has doubled. However, this must be balanced against the poor performance of the bootstrap methods in terms of the proportion of upper rejections, which tended to be particularly high compared with those from the robust method. Larger cluster size With a cluster size of 50 or 100 (results not shown), similar results were seen to those above, in that a large difference in the absolute value of the ICC together with extremely skewed distributions were required for the double bootstrap to achieve a rejection rate closer to the nominal level than the robust method. In particular, increases in the between-cluster variance accompanied by skewed distributions at the between-cluster level typically caused the double bootstrap to give better lower rejection rates. Larger number of clusters For 12 clusters of size 25 per arm (results not shown), all three methods performed better than for an equivalent sample size with fewer clusters per arm (6 clusters of size 50). There were very few instances in which the rejection rate exceeded that for 6 clusters of size 50. Even the BS1 method appeared to perform more consistently than when there were only six clusters per arm. Although it was still always outperformed by the BS2 and robust methods, its maximum rejection rate was 10.26%, compared with 14.24%. As before, when the ICC was very small (that is, most of the variability was within clusters), the double bootstrap method typically only outperformed the robust method when the within-cluster variance increased. In addition, for larger values of the ICC, changes in the between-cluster variance or the within-cluster variance could result in the double bootstrap outperforming the robust method, confirming the results obtained for six clusters of size 100. Lastly, normal distributions at the between-cluster level in both arms were sufficient to ensure that the robust method always performed better than the bootstrap. Larger number of clusters and larger cluster size All the trends identified in previous sections were replicated for 24 clusters of size 25 (see Tables 4 , 5 , 6 ). Interestingly, for ICC = 0.25, this sample size combination produced the largest number of occasions on which the BS2 method outperformed the robust method, namely eight. However, this should be balanced against the continued better coverage of the robust method overall. Discussion This work constitutes early stages in the further research that has been advocated to identify appropriate approaches to the analysis of cost outcomes from cluster RCTs [ 10 ]. The ICCs used in the present simulations were comparable to those estimated for costs in this previous study, but highly variable ICCs for costs at different levels and the magnitude of patient costs relative to total costs have both been emphasised as important issues [ 10 ]. Thus, decisions as to whether any of the scenarios investigated here are relevant to future trials will depend, in part, upon the issue of which cost component is most important. Limitations The main limitation of the work presented here is the lack of empirical data to inform the modelling. Empirical data on costs from cluster RCTs are required to investigate whether the bootstrap method should be evaluated in the presence of even more highly skewed data. The bootstrap methods of Davison and Hinkley also require testing in trials with non-constant cluster size. For the single bootstrap, in the presence of a non-constant cluster size, each bootstrap sample of clusters will have a different composition. The sample mean will exhibit a different degree of variability depending upon, for example, whether the bootstrap sample has happened to select many large or many small clusters. The result may be incorrect inferences about the variability in the sample mean. Moderate variability in cluster size or a large number of clusters might not be expected to have a large effect upon the estimated confidence limits, but again the researcher would have to be cautious. Whilst the double bootstrap can address this issue, this version resamples only the deviations from the cluster mean of an individual's 'own' cluster, potentially omitting valuable statistical information [ 7 ]. However, more complex bootstrap methods such as those of Rao and Wu and Carpenter et al do not suffer from this restriction and future work should allow the cluster size to vary [ 9 , 15 , 16 ]. Future research Despite these limitations, the results from the simulations present a coherent picture of the relative strengths of the methods of analysis that were compared. They also show that methods of analysis that can deal adequately with trial data incorporating a small number of clusters must be developed and investigated. For cost data the robust method gave confidence intervals with broadly correct coverage when the cluster size was constant. The Huber-White estimator can take account of non-constant cluster size, so future work should address its ability to give acceptable confidence intervals when data are skewed and the cluster size varies within the trial. Conclusions There are a number of general points that can be drawn from these results. First, when the between-cluster distribution is normal in both treatment arms, there is virtually no evidence in favour of using a bootstrap method. Second, when the ICC takes values of about 0.1 or greater, the double bootstrap can give a lower rejection rate which is closer to the nominal level than that achieved by the robust method, particularly when the between-cluster distributions are skewed. However, this is only common when the ICC changes as a result of the intervention. Third, the downward bias in the second moments of the bootstrap methods is particularly problematic. In general, 24 clusters per treatment arm is probably the minimum number for which one would even begin to consider the bootstrap in preference to traditional robust methods, for the parameter combinations investigated here. At least this number of cluster and extremely skewed data would be necessary for the bootstrap to approximate the results from the robust method with any consistency. The likelihood of such a scenario will clearly vary, but in any case these simulations have brought out a very important issue: a normal distribution for the cluster means is usually sufficient to eliminate the bootstrap from any consideration, regardless of the skewness of the individual data. This work has related to skewed cost (or potentially clinical) data. Whilst this is of interest, policymakers are more often interested in cost-effectiveness data. These have usually involved ratio statistics, which often cause major problems for traditional estimators. Future work should, therefore, investigate how well the cluster bootstrap deals with cost-effectiveness data and hence address the question of whether the disadvantages of the cluster bootstrap identified here are outweighed by its potential advantages in dealing with ratio statistics. Competing interests The author(s) declare that they have no competing interests. Authors' contributions TNF conceived of the study, wrote and conducted the simulation programs and drafted the manuscript. TJP participated in the design and coordination of the study, read, critically reviewed and contributed to drafts of the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535558.xml
533883	Rate of first recorded diagnosis of autism and other pervasive developmental disorders in United Kingdom general practice, 1988 to 2001	Background There has been concern that the incidence of autism and other pervasive developmental disorders (PDDs) is increasing. Previous studies have been smaller, restricted to autism (excluding other pervasive developmental disorders such as Asperger's syndrome), included boys only, or have not been based on a national sample. We investigated time trends in the rates of diagnosis of pervasive developmental disorders. Methods We analysed the rates of first diagnosis of pervasive developmental disorders among people registered with a practice contributing to the United Kingdom General Practice Research Database during the period 1988 to 2001. We included 1410 cases from over 14 million person-years of observation. The main outcome measures were rates of diagnosis of pervasive developmental disorders by year of diagnosis, year of birth, gender and geographical region. Results The rate increased progressively from 0.40/10,000 person-years (95% CI 0.30 to 0.54) in 1991 to 2.98/10,000 (95% CI 2.56 to 3.47) in 2001. A similar change occurred in the age standardised incidence ratios, from 35 (95% CI: 26–47) in 1991 to 365 (95% CI: 314–425) in 2001. The temporal increase was not limited to children born during specific years nor to children diagnosed in a specific time period. The rate of diagnosis of PDDs other than autism rose from zero for the period 1988–1992 to 1.06/10,000 person-years in 2001. The rate of diagnosis of autism also increased but to a lesser extent. There was marked geographical variation in rates, with standardised incidence ratios varying from 66 for Wales to 141 for the South East of England. Conclusions Better ascertainment of diagnosis is likely to have contributed to the observed temporal increase in rates of diagnosis of PDD, but we cannot exclude a real increase.	Background The term pervasive developmental disorder (PDD) refers to a range of disorders with onset in childhood characterised by abnormalities in the domains of language development, communication abilities and social interactions and by a rigid, repetitive pattern of behaviours and interests [ 1 ]. Within PDDs, autism refers to children meeting full diagnostic criteria for the above three domains of developmental impairments and onset before the age of three years. The other PDDs are Asperger's syndrome (a less severe form of PDD with the features described above but without language delay and with intelligence within the normal range), pervasive developmental disorder not otherwise specified (not fulfilling complete criteria for autism), and childhood disintegrative disorder (a severe form of autism developing in previously normal children who undergo massive regression between the ages of 2 and 10 years) [ 2 ]. Rett's disorder (occurring only in girls, characterised by acquired microcephaly, regression and neurological signs) is also classified as a PDD in ICD-10 but has a known genetic aetiology [ 3 ] and is not included in the incidence estimates presented in this paper. There have been a number of reports that the incidence of PDD is increasing [ 3 ] and the prevalence for all PDDs appears to be around the 60/10,000 in recent studies in the United Kingdom [ 4 ] and the United States [ 5 ]. There is some evidence that increased prevalence rates are due to a broadening of the concept of PDD and changes in diagnostic practice [ 6 , 7 ]. This has not gone unchallenged, however, and it is likely that increased awareness and changes in educational and social policies account for some of the recent upward trend in rates[ 6 , 8 ]. However, a real increase in the incidence cannot be ruled out [ 3 ]. As part of a case-control study to identify risk factors for autism and other PDDs [ 9 ], we have investigated time trends in the incidence of the diagnosis of PDDs using data obtained from the United Kingdom General Practice Research Database (GPRD). Methods The General Practice Research Database The GPRD was set up in 1987, then known as the VAMP (Value Added Medical Products) Research Bank [ 10 ]. It consists of the electronic clinical records of patients registered with contributing general practices and aims to include complete prescribing and diagnostic information. The practices included in the GPRD are broadly representative of all practices in England and Wales in terms of geographical distribution, practice sizes and the age and gender distributions of registered patients [ 11 ]. Contributing practices must meet a range of data quality criteria before they are included in the GPRD. The quality of the information in the database, including the completeness of recording of diagnoses made in medical facilities outside the practice, has been validated in a number of independent studies and has been found to be high [ 10 ]. There is excellent agreement between prescribing data from the GPRD and national data from the Prescription Pricing Authority [ 12 ]. General practices contributing to the GPRD originally used a software system called VAMP but during the late 1990s most practices switched to a new software system called VISION. At the start of the study period, general practitioners contributing to the GPRD used a modified version of the Oxford Medical Information Systems coding system (OXMIS) to record diagnoses [ 13 ], but in later years most practices used the READ coding system [ 14 ]. Each individual is assigned a unique identification number and all data that could identify individuals are removed from records before they are incorporated into the GPRD. Study population and denominators The information analysed in the present study comprised the electronic medical records of children born in 1973 or later and registered with general practices contributing data to the GPRD during the period 1 st January 1988 to the 31 st December 2001. Such children will have been eligible at some time in their life for MMR vaccination, an exposure of particular interest in the case-control study [ 9 ]. For a specific general practice, the start of the observation period for this study was taken as the later of either the 1 st January 1988 or the date at which the practice started contributing data to the GPRD. The end of the observation period was the date at which the practice stopped contributing data to the GPRD, or the 31 st December 2001 if this was earlier. Individual patients were included in the study during the times within the observation period that they were registered with a practice contributing to the GPRD. The number of practices included in the GPRD varied during the study period, rising from 34 in 1988 to 557 in 1996 then falling to 380 by 2001. Age- and gender-specific rates were calculated for people born in 1973 or later and registered with contributing general practices during each calendar year. Those registered for part of a particular year contributed time at risk for the period they were registered. People were classified by geographical location of their general practice by country for Wales, Northern Ireland and Scotland, and by the eight former administrative National Health Service regions for England. Identification of cases When a patient first registers with a practice contributing to the GPRD, past medical events and diagnoses judged to be clinically important by the general practitioner are recorded in the electronic record, with the date of each diagnosis where available. While a patient is registered with a contributing practice, all new diagnoses and all new drug therapies or changes in existing therapies are recorded contemporaneously. Cases were defined as people born in 1973 or later who had a first diagnosis of PDD entered into their general practice record while registered with a practice contributing to the GPRD during the study period. Cases were identified by searching the complete medical record of all people registered with participating practices during the study period for either OXMIS or READ clinical codes indicating a diagnosis of PDD: Appendix [see Additional file 1 ]. People with a diagnosis of a PDD made prior to their observation period, and people whose PDD diagnosis was the first entry recorded when they joined a practice, were classified as prevalent cases and were excluded from the incidence estimates. On the basis of the recorded diagnoses patients were classified as 'autism' or 'other PDD'. Those with autistic disorders and similar presentations were grouped in the autism category and other descriptions (such as Asperger's syndrome) were grouped in the other PDD category. Analysis Crude rates and age standardised incidence ratios for first diagnosis of all PDDs, autism, and other PDDs were calculated by year of diagnosis and gender and for different geographical regions. Because of low numbers of cases in some years, indirect standardisation was used with rates during the whole study period as the standard rates. The age-standardised incidence ratios presented take into account variations in the age structure of the population in different time periods and different geographical regions. Rates of first diagnosis were calculated by year of diagnosis and year of birth, both in two year intervals. Ethical approval Ethical approval for the study was obtained from the Scientific and Ethical Advisory Group of the GPRD and from the ethics committee of the London School of Hygiene and Tropical Medicine. Results We identified 1410 persons with a first recorded diagnosis of PDD during the study period: 1097 were categorised as autism and 313 as other PDD, for 294 (94%) of whom the diagnosis was Asperger's syndrome (appendix). Table 1 shows the crude rates of first diagnosis of PDD and age standardised incidence ratios by year of diagnosis and gender. The person-years at risk varied from year to year because of changes in the number of contributing practices, which increased until the mid 1990s and then fell. During the study period incidence rates increased progressively for both males and females. The patterns of crude rates and of standardised incidence ratios were very similar. The overall rate increased seven-fold from 0.40/10,000 person-years (95% CI 0.30 to 0.54) in 1991 to 2.98/10,000 person-years (95% CI 2.56 to 3.47) in 2001. A similar change occurred in the age standardised incidence ratios, from 35 (95% CI: 26–47) in 1991 to 365 (95% CI: 314–425) in 2001. The overall male to female ratio was 4.8 and during 1991 to 2001 ranged from 2.7 to 8.4 (chi-squared test for heterogeneity, p = 0.21), with no clear temporal trend. Table 2 shows the number of cases, person-years of follow-up and rate of first recorded diagnosis by date of diagnosis and by date of birth, both in two year intervals. The diagonally linked numbers identify rates among children of the same age in different years. The temporal increase observed in rates of diagnosis is not limited to children born during specific years or to children diagnosed in a specific time period. The rates of diagnosis for children at the same age, born in successive birth cohorts, increased in each successive cohort, including for relatively old ages at diagnosis. For example, for diagnoses at the ages of two to four years, the rates of new diagnosis per 10 000 person-years from 1992–93 to 2000–01 were 1.41, 2.70, 3.07, 5.56 and 7.74. Corresponding rates at ages eight to ten years were 0.36, 0.40, 0.67, 2.46 and 4.46. Table 3 shows the rates of first recorded diagnosis and age standardised incidence ratios of autism and other PDDs by year of diagnosis. There was a striking rise in the rate of diagnoses of other PDDs: from zero for the period 1988 to 1992 rising to 1.06/10,000 person-years by 2001. Over the study period the rate of diagnosis of autism also increased substantially. The patterns were broadly similar for males and females (data not shown). Few diagnoses of other PDDs were made before 1997, but by 2001 over one third of all diagnoses were PDDs other than autism. Table 4 shows the rates of first recorded autism and other PDDs by geographical area. There was marked geographical variation in rates of first diagnosis, with the lowest rate in Wales being less than half the rate in the South East of England (chi-squared test for heterogeneity in SIRs by region, p < 0.001). In general, the regional variations were less marked for autism than for other PDDs and the ratio of the diagnoses of autism compared with other PDDs varied greatly between areas (chi-squared test for heterogeneity, p = 0.0006). Discussion There was around a 10-fold increase in the rate of first recorded diagnoses of PDDs in United Kingdom general practice medical records from 1988–92 to 2000–01 (table 1 ). The increase was more marked for PDDs other than autism but the increase in autism was also striking (table 3 ). If these changes indicate a true increase in the incidence of the conditions it is of great public health importance. However, it is probable that the increase is due, at least in part, to changes in the ascertainment and diagnosis of the conditions. Factors that could have affected the results Some of the general practices contributing data to the GPRD will provide anonymised copies of hospital letters and specialist reports on individual patients. Of patients with a recorded diagnosis of PDD in the GPRD, including prevalent cases when first registered, 446 were registered with 203 general practices willing to provide this service. For 80 of these, medical records were not available as the patient was no longer registered with the general practitioner. We obtained complete case records including copies of hospital clinic letters and specialist reports for 318 (87%) of the remaining 366 people. These were reviewed by a psychologist (LH) and a random sample of 50 records were also reviewed by a child psychiatrist (EF), both of whom have long experience in the field of autism. They judged that a PDD was likely to be present in 294 children (92.5%)[ 15 ]. For the 211 patients (of 318) who were first diagnosed with a PDD after they entered the GPRD (and thus are included in this paper), a diagnosis of PDD was confirmed for 193 (91.5%). Thus the positive predictive value of a recorded diagnosis of PDD in the electronic record was high, but we were not able to assess the sensitivity of the ascertainment of cases, that is how often the diagnosis of a PDD may have been missed or not recorded. Of the 211 cases reviewed and included in this paper, 5 (2.4%) were classified as 'other-PDD' on the basis of their electronic record only, whereas in the validation, 32 (15.1%) met diagnostic criteria for a PDD other than autism. Thus it is likely that a proportion of people in the 'autism' diagnostic category in this paper have a form of PDD other than autism. The inaccuracy of diagnostic descriptions of different PDDs within the GPRD is likely to reflect changes in the definition of PDD over the past two decades, in particular a broadening of PDDs other than autism [ 1 , 16 , 17 ]. Many children with Asperger's syndrome or pervasive developmental disorder not otherwise specified would only have been assigned a diagnosis of PDD from the latter half of the 1990s. In the earlier period such children may either not have received a diagnosis of PDD at all or have been diagnosed as autism. Inflation in the number of cases in later years could have occurred as other PDD diagnoses came into widespread use and some previously undiagnosed children were diagnosed. For example the OXMIS coding dictionary, used by most general practitioners contributing to the GPRD until the mid-1990s, has only two possible codes for autism and no clinical code for Asperger's syndrome, and this diagnosis could only be assigned when practices started to use the READ coding system. These changes explain the low level of diagnoses of other PDDs until the mid 1990s (table 3 ). During the study period there is likely to have been an increase in the diagnosis of high-functioning autism, as professionals have become more aware that autism can occur in people of normal intelligence [ 3 ]. Greater ascertainment of high-functioning autism may partly explain the increased incidence of autism as well as in the other-PDD diagnoses. That there was also a marked increase in the rate of diagnosis of severe disorders, however, suggests that better detection of less severe cases alone can not explain all of the increase. Two previous studies have demonstrated falls in the rates of diagnosis of mental retardation [ 6 ] and of non-specific developmental disorders [ 18 ] during the 1990s as the rate of diagnosis of autism increased. These patterns could be partly due to improvements in the detection and diagnosis of autism. In the late 1980s and early 1990s autism was only likely to be diagnosed by specialist child psychiatrists who increased in number by 40% between 1988 and 2001 (personal communication: Royal College of Psychiatrists). Through the 1990s developmental and community paediatricians began to diagnose PDDs, and this, combined with increased awareness of autism and PDDs among the general public, may have contributed to increased ascertainment of the disorders [ 19 ]. The marked geographical variation in rates of diagnosis and in the ratio of diagnoses of autism to other PDDs (table 4 ) may reflect differences in service provision and parental awareness in different regions. It is likely that children with PDDs other than autism will generally be first diagnosed at later ages than children with autism. Greater ascertainment of other PDDs during the latter part of the study period (table 3 ) could have led to the observed increase in incidence being largely restricted to older age groups. However, the increased rates were observed for all age groups (table 2 ). Because of low numbers of cases in some years, indirect standardisation was used to calculate SIRs. When comparing SIRs, marked differences in the age distribution of the populations being compared can lead to a biased comparison [ 20 ]. However, the differences in age distributions were not great in our study and the patterns seen in the crude rates and the age standardised rates did not differ materially, suggesting the comparison of SIRs was valid. The accuracy of the denominator data may have changed during the study period. When patients move geographical area they may delay registering with a new general practice until they have a specific reason to visit them. Thus our estimates of person-years at risk may be too low for those moving into a practice and too high for those moving out. Person-years at risk may also be inflated as patients who emigrate may omit to inform their general practitioner and, furthermore, administrative delays and errors may result in individuals being registered with more than one general practitioner. For example, in 1997 in England 51 million people were registered with a general practitioner [ 21 ] whereas the total population size was only around 49 million people[ 22 ]. It has also been suggested that in recent years the period of time for which patients who have moved out of an area remain registered with a general practice in that area has shortened as health authorities and general practices have streamlined procedures [ 23 ]. This may mean that the inflation of denominators could have been greater in the early years compared to later years. However, these factors could explain only a very small part of the increased rates observed. An additional issue is that the person-years used as denominators in our analysis did not exclude the period following diagnosis. However, given the relative rarity of the disease, this will have produced only a very small inflation in the denominators. Comparison with other studies A previous study based on the GPRD assessed time trends in the diagnosis of autism from 1988 to 1999 [ 24 , 25 ]. The study did not examine other PDDs, was restricted to children aged less than 13 years and included a total of 305 cases. The previous paper was based on a sub-set of the GPRD data held by the Boston Collaborative Drug Surveillance Program. This is reflected in the person-years of observation contributing to the two studies – just under 3.1 million person years in the previous study compared with over 14 million person years in our paper. The two studies were undertaken in substantially different populations. However, the extent of the rise in incidence in the two studies was similar: from around 0.3/10,000 person-years in 1989/1990 to around 2.0/10,000 person-years by 1999 [ 24 ]. In addition, the increased risks of autism observed in successive birth cohorts were similar to those observed in our study [ 25 ]. A study in the West Midlands area of the United Kingdom based on diagnoses made at child development centres found cumulative incidence rates of autism for children between the ages of one to four years of 2.22/10,000 person-years in 1991–3 (based on 20 cases), rising to 4.75/10,000 person-years in 1994–6 (based on 42 cases) [ 26 ]. The corresponding cumulative rates from our study were 1.00/10,000 person-years in 1991–3 rising to 1.97/10,000 person-years for 1994–6. Restricting our data to the West Midlands region we observed rates of recorded diagnosis of autism of 1.57/10,000 person-years in 1991–3 and 1.56/10,000 person-years in 1994–6, based on 14 and 16 cases respectively. It is unclear why the rates observed in the West Midlands study were different from the rates we observed. However, these area and time period specific rates are based on small numbers of cases. The validation of cases in the GPRD can not necessarily be assumed to apply to all geographical areas. A recent study in North London estimated the prevalence of autism by year of birth and concluded there was a rise from 1979 to 1992 after which there was a plateau between 1992 and 1996 [ 27 ]. These authors excluded children with Asperger's. The review of clinical records of children in our study indicated that some children who would have been diagnosed as having autism in the early 1990s actually had Asperger's and would probably not have been classified as 'autism' in later years. This shift in diagnostic labelling from autism to Asperger's may explain, at least in part, the plateau in incidence described by Lingam et al. and would be compatible with the reduction in average age at diagnosis observed in that study. Conclusions This is one of the largest studies undertaken of trends in the incidence of autism and other PDDs. We describe striking increases in the rates of diagnosis of these conditions. However, much of the increase may be due to better ascertainment related to changes in diagnostic practice and improved recognition of the conditions. While review of clinical records confirmed that over 90% of diagnoses of PDDs recorded by general practitioners were likely to be correct, the nature of the study precluded us from assessing how often children with PDDs were not diagnosed. Thus the extent to which the increases in incidence we document are true increases in disease is uncertain. Abbreviations PDD pervasive developmental disorder GPRD General Practice Research Database Competing interests The authors declare that they have no competing interests. Authors' contributions AH, LS, EF, LR and PS designed the study. LH and EF undertook the validation of case reports. CC analysed the data. LS drafted the paper. All authors commented on drafts and approved the final version of the paper. Pre-publication history The pre-publication history for this paper can be accessed here: Supplementary Material Additional File 1 Appendix: Codes used to identify cases, numbers identified, and diagnostic classification used in this paper Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC533883.xml
535560	S-adenosylmethionine (SAM-e) for the treatment of depression in people living with HIV/AIDS	Background This study reports on clinical data from an 8-week open-label study of 20 HIV-seropositive individuals, diagnosed with Major Depressive Disorder (DSM-IV), who were treated with SAM-e (S-Adenosylmethionine). SAM-e may be a treatment alternative for the management of depression in a population reluctant to add another "pill" or another set of related side effects to an already complex highly active antiretroviral therapy (HAART) regimen. Methods The Hamilton Rating Scale for Depression (HAM-D) and the Beck Depression Inventory (BDI) were used to assess depressive symptomatology from 1,2,4,6 and 8 weeks after initiation of treatment with SAM-e. Results Data show a significant acute reduction in depressive symptomatology, as measured by both the HAM-D and the BDI instruments. Conclusions SAM-e has a rapid effect evident as soon as week 1 ( p < .001), with progressive decreases in depression symptom rating scores throughout the 8 week study.	Background The incidence of depression in people living with HIV (PLWH) has consistently been reported to be higher than the 12–15% rates reported for the general population [ 1 ]. Depression in this population is largely untreated. Estimates suggest that this comorbid condition of HIV affects 10% – 50% of PLWH [ 2 ]. The management of depression in people with HIV is important since it can reduce medication adherence. Depression can be life-threatening, given that the patient can deteriorate medically, and that untreated depression can result in suicide. Untreated depression may also result in substance abuse, a theory based on the 'self medication' theories of addictive behavior. Depression is a pressing problem at this point in the epidemic because the HIV population is aging and the incidence of depression increases with age [ 3 ]. In New York City, 27% of all people living with HIV are over 50 years of age and 66% are over 40 years age [ 4 ]. In two studies of PLWH age 50 and older, the rate of depressive symptomatology ranged from 58% – 68% [ 5 ]. Under-treated depression in the HIV population may be the result of multiple factors. People of color, women, substance abusers, and those living in poverty are less likely to have access to or to accept mental health outpatient services [ 5 - 7 ]. Many patients with HIV who are managing their disease are on complex regimens of medications (HAART) and dietary supplements to treat their illness [ 8 ]. Adding another pill with significant side effects is often viewed by both patient and primary treating physician as an unacceptable burden, either in terms of number of pills per day, or in terms of anticipated side-effect profiles, or risk/benefit analysis [ 9 ]. Primary care physicians are often unprepared to recognize depression, which can present as fatigue, weakness, insomnia or loss of appetite, and may seek medical explanations for depressive symptoms [ 7 ]. HIV-related dementia, which presents with apathy and amotivation, can also be mistaken for depression. Physicians may be unfamiliar with mental health interventions and place depression as a lower priority after management of the primary HIV infection [ 10 ]. The stigma of HIV remains large and the addition of the potential stigma of a mental illness (i.e. depression) further contributes to the patient's reluctance to pursue or engage treatments for depression [ 11 ]. Many patients fear that substance abuse, homosexuality, or HIV itself may be grounds for discrimination from a mental health provider [ 12 , 13 ]. SAM-e is a compound that has been used for many years in Europe as an antidepressant. It became available in the USA as a non FDA-regulated compound in 1999. SAM-e resembles a naturally occurring compound in the human body, formed by methionine and adenosine triphosphate. SAM-e donates an active methyl group to several acceptor molecules, including catecholamines, fatty acids, phospholipids, proteins and nucleic acids [ 14 ]. SAM-e appears to increase central turnover of dopamine and serotonin; for example it increases the main metabolite of central serotonin-5-HIAA (5-hydroxyindoleacetic acid) in the cerebrospinal fluid [ 15 ]. The efficacy of SAM-e as an antidepressant has been reported in a number of open-label trials [ 16 ] and in two small placebo-controlled clinical trials [ 17 , 18 ]. Kagan et al conducted a placebo-controlled trial that found SAM-e to be effective in the treatment of depressed inpatients [ 17 ]. A meta-analysis of SAM-e studies shows a greater response rate with SAM-e compared with placebo, with a global effect size ranging from 17–51% [ 19 ]. Side effects reported in these studies are fewer than with SSRIs specifically with respect to nausea, headache, dry mouth, and agitation or activation. In studies of parenteral SAM-e, activation of mania was noted in two cases [ 20 ]. SAM-e would seem to offer the potential for effective antidepressant treatment with fewer side effects for patients with medical illnesses. This was shown in a successful open label trial of SAM-e in Parkinson's patients [ 21 ] and in a double-blind study of fibromyalgia patients [ 22 ]. The aim of the study reported below was to assess efficacy, tolerability, and safety of SAM-e in a group of HIV-positive individuals diagnosed with Major Depressive Disorder (DSM-VI). Methods This study is an analysis of the complete dataset from an independently funded, IRB-approved project conducted at ACRIA, a 12-year-old, community-based HIV research and education agency in New York City. A preliminary report, based on a subset of these data, was presented at the 14 th International Conference on AIDS. A total of 20 patients were recruited between April 2000 and November 2000. Recruitment activities included distribution of information brochures to community physicians, announcements at public ACRIA treatment education workshops, and direct calls to persons listed in the ACRIA database. Individuals who responded to recruitment materials were screened via phone by the research coordinator using the PRIME-MD [ 23 ]. HIV-positive patients were then invited to be interviewed by a study psychiatrist at ACRIA for a clinical psychiatric examination. Patients who met DSM-IV criteria for Major Depression (assessed with the SCID-IV) and who did not meet any of the exclusion criteria were eligible for participation after giving informed consent [ 24 ]. Criteria for exclusion were (a) unstable medical illness, (b) pregnancy, lactation or refusal by participants to employ an acceptable contraceptive, (c) history of substance abuse in the prior month, (d) treatment with another psychotropic medication within two weeks prior to initiation of SAM-e treatment, (e) concurrent MAOI treatment, (f) active suicidal ideation and/or psychotic symptoms, (g) reversible medical pathology that is thought to be causing the depression, and (h) history of mania or diagnosed bipolar disorder. Patients received $15 for each completed study visit. Treatment was initiated at 200 mg of SAM-e twice a day with daily supplementation of 1,000 mcg Vitamin B12 and 800 mcg Folic Acid, which were provided by Pharmavite, LLC. Over the course of the study, the dose of SAM-e was individually adjusted up to 800 mg, twice daily. Patients reporting nausea or insomnia were not increased above 800 mg daily, while patients with no reported side effects could be increased to a total daily dose of 1600 mg. Dosing was adjusted according to the severity of symptoms and clinical treatment response. We defined clinical treatment response as an improvement in depressive symptomatology of greater than 50% reduction on patient scores on the BDI and HAM-D as our treatment endpoint [ 25 , 26 ]. A statistical data codebook was created and secondary statistical analyses were performed. Measures At each study visit (Baseline, Week 1, Week 2, Week 4, Week 6, and Week 8), patients completed a clinical interview and a series of structured questionnaires on medical/physical symptoms, substance abuse, dementia scales, as well as two instruments designed to measure depressive symptoms, self-administered Beck Depression Inventory (BDI and the clinician-administered HAM-D). BDI The BDI is a self-report measure for depressive symptoms [ 25 ]. Total scores range from 0 – 63 and are calculated by summing the scores to each of the 21 items. Higher scores on this measure indicate greater severity of depression. Scores above 30 indicate severe depression. Scores of 30 – 10 describe moderate depression, while scores less than 10 indicate the absence of depression. HAM-D The HAM-D is a 17-item, clinician-rated instrument. It was designed to assess the changes in severity of depressive symptomatology over time in patients who had been diagnosed with Major Depressive Disorder [ 26 ]. Scores above 24 reflect severe depression, while scores ranging from 17 – 7 indicate mild symptomatology. Scores below 7 indicate an absence of depression. Analyses Two separate analyses (e.g., Baseline -v- Week 1 and Baseline -v- Week 8) were identified as the most critical indicators of SAM-e efficacy. The first analysis represented evidence for acute onset of SAM-e, and the second indicated effectiveness of SAM-e over time. BDI and HAM-D scores were centered and the resulting Z-scores were used in a series of t-tests to assess the agreement between patient and investigator ratings of depression. Mean scores from both depression measures (BDI and HAM-D) were compared between time points, using a series of paired t-tests. Finally, an intent-to-treat analysis, using the last score carried forward, was conducted on patient BDI scores for all subjects who initiated treatment. Results Participants Thirty (30) subjects were screened by telephone and twenty (20) entered into the study. Of these, fifteen (15) were male, five (5) were female and the majority were people of color (50% Black, 30% Hispanic, 20% Caucasian). Table 1 presents a detailed demographic profile of the patients. Although the exclusion criteria included a prohibition against concurrent treatment with conventional antidepressants during the study period, we identified one patient who did not discontinue use of conventional antidepressant until Week 4 of the study. All subjects self-reported that they were not actively substance abusing at the time of the study; however, urine drug screens were not performed. Five (5) patients did not complete the study, but Week 1 data were recorded and were included in the analysis of early onset of therapeutic SAM-e effects. These five (5) subjects were excluded from subsequent analyses. Two (2) subjects did not return for study participation after Week 1: both had a history of IDU; one was co-infected with Hepatitis C and had a comorbid obsessive compulsive disorder (OCD). Three (3) patients did not return after Week 4, one with a history of IDU and one with a history of suicide attempts. Of those who dropped out of the study, two (2) met criteria for AIDS. Table 1 Demographic characteristics of patients Variable N % Median Range Age 20 45 24 – 57 Sex Male 15 75.0 Female 5 25.0 Race/Ethnicity Black 10 50.0 Hispanic 6 30.0 Caucasian 4 20.0 HIV diagnosis (year) 20 1992 1987 – 2000 CD-4 count (baseline) 20 320 5 – 1200 Log viral load (baseline) 20 3.40 1.70 – 4.48 Transmission risk MSM sex 4 20.0 Heterosexual sex 5 25.0 IDU 2 10.0 Multiple risks 9 45.0 Treatment response Table 2 contains the means, standard deviations and ranges for depression data from each study visit. Of the patients who provided Baseline and Week 1 follow-up data, there was clear evidence for a rapid therapeutic effect of SAM-e. The reduction in mean BDI scores from Baseline ( M = 33.5, SD = 11.1; Range: 15 – 55) to Week 1 ( M = 18.9, SD = 10.4; Range: 0 – 45) was significant, t (1, 18) = 5.15, p < .001. Mean HAM-D scores declined in a parallel manner, from Baseline ( M = 26.5, SD = 6.8; Range: 12 – 39) to Week 1 ( M = 16.8, SD = 7.3; Range: 2 – 29). This change was also statistically significant, t (1, 14) = 3.58, p < .01. Both depression assessment instruments reached the defined clinical treatment response threshold of a greater than 50 percent reduction in depression symptom scores. Table 2 BDI and HAM-D scores recorded at each study visit BDI HAM-D Study Visit N M(SD) Range N M(SD) Range Baseline 20 33.5(11.1) 15–55 20 26.5(6.8) 12–39 Week 1 19 18.9(10.4) 0–45 15 16.8(7.3) 2–29 Week 2 17 14.1(8.2) 0–25 13 10.7(5.5) 0–21 Week 4 17 8.8(7.8) 0–28 15 6.0(4.7) 0–15 Week 6 16 6.4(6.8) 0–20 14 5.2(5.7) 0–20 Week 8 16 5.0(4.7) 0–16 15 3.7(3.3) 1–13 The 15 patients who completed the study also demonstrated clinical treatment response over time. Mean BDI scores dropped significantly from Baseline ( M = 35.1, SD = 12.0; Range: 15 – 55) to Week 8 ( M = 5.1, SD = 4.8; Range: 0 – 16), t (1, 14) = 8.28, p < .001. Similarly, HAM-D scores were reduced from Baseline ( M = 26.7, SD = 6.3; Range: 12 – 36) to Week 8 ( M = 3.7, SD = 3.3; Range: 1 – 13). The reduction in HAM-D scores reached statistical significance, t (1, 13) = 9.92, p < .001. There were no significant differences in patient and physician ratings of depression at each study visit (Table 3 ). Figures 1 and 2 provide graphic representations of mean BDI and HAM-D scores across each time point, respectively. Remission of depression was defined as a HAM-D score ≦ 7, response to treatment was defined as ≧ 50% decrease in HAM-D scores. For the 15 patients who completed the study, the remission rate was 93%. Fourteen (14) of the 15 patients achieved a HAM-D of 7 or lower, while one patient received a HAM-D rating of 13. The response rate was also 93%, with 14 of the 15 patients achieving 50% or greater reduction in HAM-D scores. The patient who did not meet the response rate had a Baseline HAM-D of 12 and a Week 8 score of 7. Table 3 Comparison of BDI and HAM-D scores (study completers) Study Visit t df p Baseline -0.07 14 .95 Week 1 -1.39 14 .19 Week 2 -0.98 14 .35 Week 4 -0.64 14 .54 Week 6 0.35 13 .74 Week 8 0.71 14 .49 Figure 1 Mean BDI scores at each study visit (95% confidence intervals) Figure 2 Mean HAM-D scores at each study visit (95% confidence intervals) The intent-to-treat analysis yielded similar results. The last reported BDI and HAM-D scores for each of the 20 patients who initiated treatment (i.e., received at least one dose of SAM-e) were used in this analysis. Baseline BDI scores ( M = 33.5, SD = 11.1; Range: 15 – 55) were higher than Week 8 ( M = 6.6, SD = 6.1; Range: 0 – 21). HAM-D scores dropped from Baseline ( M = 26.5, SD = 6.8; Range: 12 – 39) to Week 8 ( M = 7.7, SD = 10.1; Range: 1 – 39). The reduction in self-reported (i.e., BDI) depressive symptomatology was significant, t (1, 19) = 8.85, p < .001. The results were equivalent for the psychiatrist-rated HAM-D, t (1, 18) = 7.23, p < .001. The intent-to-treat remission rate was 79%. Fifteen (15) of the 20 patients achieved a HAM-D of 7 or lower, while one patient did not receive a valid HAM-D rating and four (4) received HAM-D ratings greater than 7 (13, 22, 24, 39, respectively). The intent-to-treat response rate was 74%, with 14 of the 19 patients achieving 50% or greater reduction in HAM-D scores. Adverse effects No patients terminated their participation due to side effects. Two (2) patients reported nausea and one (1) reported diarrhea; all resolved spontaneously. One (1) patient had insomnia and a subjective feeling of 'high energy,' but did not meet criteria for hypomania. Discussion Recent data from the HIV Cost and Services Utilization Study (HCSUS) indicate that almost half the nation's adult HIV patients suffered from symptoms of mental disorders. The report indicates that the prevalence of mental disorders range from four to eight times higher than those found in the general populations. Over 60 percent of HIV-positive adults used mental health or substance abuse services during the period studied; 26% visited mental health specialists, 15% had group mental health treatment, 40% discussed emotional problems with medical practitioners and 30% used psychotherapeutic medications [ 27 ]. Depression remains under-reported and under-treated in the HIV population. Patient factors include stigma, fear of more medication, and perhaps the assumption that depression is a normal part of HIV disease, or due to HAART. Physicians tend to focus more on physical symptoms, and may be unaware of how to screen for depression in this population. Both physicians and patients are disadvantaged in accessing gay-affirmative, well-trained HIV mental health specialists who can disentangle HIV medical problems from substance abuse and mental illness. Risk factors for depression (e.g., history of substance use/abuse, lack of social support, stigma, etc.) are prevalent in those populations who are disproportionately affected by HIV. Many patients are reluctant to add more medication to their complex treatment regimens. The need for effective and safe treatments for depression is clear. Results of this study demonstrate that SAM-e significantly reduces depression in people living with HIV. This finding supports previous research demonstrating the efficacy of SAM-e for use in treating depression. More importantly, the therapeutic effect of SAM-e had an acute onset. There were significant reductions in the severity of depression among study patients within the first week of treatment. Unlike most antidepressants that require several weeks to reach maximum therapeutic efficacy, previous research has shown that serum levels of SAM-e peak within 24 hours of treatment. This may account for the rapid, effects of SAM-e seen on the HAM-D and BDI scores. Conclusions SAM-e is a safe, effective treatment for depression among people living with HIV. SAMe has few reported side effects, and in our population of PLWH, was extremely well tolerated. SAM-e in this group had rapid onset of clinical efficacy. Both of these results make it an appealing alternative therapy for the management of depression in the HIV population. The results of the current study must be interpreted in the context of the study's limitations. Limitations of this study include the lack of a placebo group, the lack of a double-blind design and small sample size. The criteria for participation and the small sample size may limit the generalizability of our findings to other patient populations. Without a placebo control group, we cannot assess two factors that might have affected our results. Although patients were treatment-experienced with other anti-depressants, we cannot rule out the potential placebo-response that may have contributed to the rapid therapeutic effect of SAM-e. Also, meeting with a study psychiatrist may have contributed to improved depressive symptom ratings. Nonetheless, the study results provide useful and clinically relevant information for treating depression in HIV-positive individuals. Future studies employing a double-blind, placebo-control design are warranted. The absence of depression does not imply wellness. Many people living with HIV/AIDS report feelings of sadness and depression but do not receive treatment for depression. Future studies should also assess health-related quality of life. Such protocols would determine if people with significant, but subclinical depressive symptomatology would benefit from SAM-e. Authors' contributions RASrecoded study data, conducted statisticalanalyses andco-wrote the final manuscript. DM provided details of the study design and procedures andwas responsible for extractingclinical data from patient records. IC and KJ served as clinical investigators and provided valuable comments on the final paper. SEKwrote the first draft of the paper and co-wrote the final manuscript. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535560.xml
535548	Merkel cell carcinoma in a malignant pleural effusion: case report	Background Merkel cell (neuroendocrine) carcinoma is a small round blue cell malignant neoplasm that primarily presents in the skin. The diagnosis of Merkel cell carcinoma in a pleural fluid is challenging because of the morphological similarity to many other malignant neoplasms. Immunohistochemical stains can be essential to establish the diagnosis of Merkel cell carcinoma. Case presentation A 77 year-old woman presented with a mass in her right buttock thought clinically to be a boil or sebaceous cyst. Upon histopathologic review including immunohistochemical analysis, a diagnosis of Merkel cell carcinoma was rendered. Wide-excision and sentinel lymph node biopsy revealed negative margins and no evidence of metastasis. Ten months later she complained of bone pain and a bone scan revealed multiple lesions. An abdominal CT scan revealed a T4 vertebral mass and local radiotherapy was administered. Two months later the patient presented with shortness of breath. A chest radiograph showed an effusion and thoracentesis was performed. The fluid was confirmed to contain metastatic Merkel cell carcinoma by cytology and immunohistochemical analysis. Conclusions Merkel cell carcinoma is an aggressive neoplasm that can, despite careful surgical management, occasionally present as a malignant pleural effusion in a relatively short time period. Immunohistochemical analysis can aid in confirming this rare outcome.	Background The Merkel cell is the namesake of an F. Merkel who in 1875 described distinct round cells in the basal layer of the epidermis [ 1 ]. The distinctive Merkel cell with its round shape and salt and pepper chromatin is known to be present in the epidermis and dermis [ 2 ]. Toker first described Merkel cell carcinoma in 1972 and along with Tang concluded the cells of origin to be derived from the neural crest [ 3 , 4 ]. More recently, Sibley discussed the idea that the cells may be derived from the pleuripotential basal cells of the epidermis that surround hair follicles and function as touch receptors in the skin [ 5 ]. Merkel cell carcinoma is a tumor of the dermis and can easily be mistaken for other malignant small round blue cell tumors such as lymphoma, amelanotic melanoma, and metastatic small cell carcinoma [ 6 ]. Historically, electron microscopy demonstrating cytoplasmic microfilaments and neurosecretory granules was the key to diagnosing Merkel cell carcinoma [ 7 ]. Immunohistochemical approaches are now used to confirm a diagnosis of Merkel cell carcinoma. Case Report A 77 year old female presented to her primary care physician with a complaint of a right buttock mass. Her past medical history was significant for colonic adenocarcinoma treated surgically, twelve years prior. The skin overlying the 2 × 3 cm firm, non-fluctuant, non-tender mass was erythematous. Initially the mass was thought to be a boil or carbuncle. Six weeks later during a visit with a surgical oncologist for routine follow-up of her colonic adenocarcinoma resected 13 years prior, the patient noted that the mass was becoming quite painful and enlarging rapidly. The patient was then taken to the OR for local excision. Pathology revealed a small round blue cell tumor forming large nests and infiltrating as single cells between thick strands of fibrous tissue in the dermis and extending to the subcutaneous fat. There were abundant mitoses, a lack of necrosis or obvious vascular invasion. The epidermis was uninvolved. The histologic and immunochemical findings supported the diagnosis of Merkel cell carcinoma. A wide local excision and sentinel lymph node biopsy were performed. One sentinel lymph node was negative for malignancy and there was no residual tumor identified in the excision. No adjuvant therapy was given. Ten months after the initial presentation the patient had increasing shoulder pain that radiated into the chest and flank. A CT of the thoracic spine revealed a 3.0 cm mass in the area of the fourth thoracic (T4) vertebra. A CT guided needle biopsy of the area revealed metastatic Merkel cell carcinoma. A bone scan revealed additional areas of concern in the right humerus and bilateral distal femurs. The patient received radiation therapy to the thoracic spine and right distal femur for palliation of bone pain. A year after first presenting for a right buttock mass, the patient was admitted to the hospital for increasing dyspnea. A chest radiograph revealed a large right pleural effusion, which was drained for both therapeutic and diagnostic purposes. The fluid was grossly bloody and cytology revealed abundant, round, basophilic single cells, many in mitosis, with characteristic granular, "salt and pepper" nuclear chromatin (Figure 1 ). A cellblock was made and immunohistochemical staining with CK20 revealed characteristic positivity in a dot like rim pattern (Figures 2 ). Figure 1 Pleural fluid at 630x demonstrating small round cells with salt and pepper chromatin. Note multiple mitotic figures Figure 2 Cell block stained with CK20 demonstrating sharp perinuclear dot like pattern. Conclusions The diagnosis of Merkel cell carcinoma in pleural fluid is challenging partially because of its rarity. To our knowledge this is the first case report of Merkel cell carcinoma in a pleural fluid characterized with immunohistochemical stains. Wason and Friedman described the only other reported case of pleural effusion due to metastatic Merkel cell carcinoma in 1985 [ 7 ]. In that report, the diagnosis was confirmed by electron microscopy that demonstrated cytoplasmic microfilaments and numerous dense-core, peripheral, neurosecretory granules. The light microscopic morphologic characteristics of Merkel cells in pleural fluid yields a large list of possible diagnosis including small cell lung carcinoma, carcinoid tumor, lymphoma, plasmacytoma, pancreatic carcinoma small cell type, and peripheral neuroectodermal tumor (PNET). Since Merkel cell carcinoma rarely appears in pleural fluid and small cell lung carcinoma frequently does the pathologist must have the salient cytomorphologic features of Merkel cell carcinoma in mind to make the diagnosis in the absence of clinical history. The most characteristic features of Merkel cell carcinoma in pleural fluid include convolution of nuclei and lack of prominent nuclear molding. Once Merkel cell carcinoma has entered the differential diagnosis the case can be pursued. With immunohistochemistry, it is now possible to definitively diagnose Merkel cell carcinoma in pleural fluid without electron microscopy. The immunohistochemical profile of Merkel cell carcinoma is distinct and has been well established on paraffin embedded tissue [ 8 ]. Tumor cells consistently express low molecular weight keratins in the form of a perinuclear dot like pattern typical of neuroendocrine carcinomas. Merkel cell carcinomas also label for neuron specific enolase (virtually all), chromogranin B (100%), chromogranin A (72%), secretoneurin (22%), and synaptophysin (39%) [ 5 , 9 ]. Among neuroendocrine carcinomas, CK20 positivity in a dot like rim pattern is seen only in Merkel cell and salivary gland small cell carcinomas [ 10 ] Other small cell tumors that are CK20 positive include bronchogenic small cell carcinoma (0.03%), small cell cervical carcinomas (9%), and small cell carcinomas of salivary glands (60%) but not in the characteristic dot-like pattern [ 11 ]. Neuron specific enolase is positive in almost all Merkel cell carcinomas [ 5 ]. Carcinoid tumors stain for synaptophysin and chromogranin, but not CK20. Reactivity with CD45 is diagnostic of a hematopoietic lineage; in addition, apart from vimentin, hematologic malignancies do not generally react with antibodies to other lineage specific markers. Plasmacytomas express CD138 and CK8 [ 10 ]. CD 99 positivity suggests PNET; however, one series found CD99 to be positive in 40% of Merkel cell tumors [ 12 ]. Metastatic Merkel cell carcinoma as the cause of a malignant pleural effusion is a rare occurrence. However, such an aggressive neoplasm as Merkel cell, which is often widely metastatic, may be a more common cause of malignant pleural effusion than recognized. Morphologically, the tumor cells in a body fluid could easily be mistaken for a malignancy of another origin. In the work-up of the pleural effusion caused by a small round blue cell process, inclusion of immunohistochemical markers specific to Merkel cell carcinoma may be prudent. Competing interests The authors declare that they have no competing interests. Authors' contributions	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535548.xml
517925	An imaging system for standardized quantitative analysis of C. elegans behavior	Background The nematode Caenorhabditis elegans is widely used for the genetic analysis of neuronal cell biology, development, and behavior. Because traditional methods for evaluating behavioral phenotypes are qualitative and imprecise, there is a need for tools that allow quantitation and standardization of C. elegans behavioral assays. Results Here we describe a tracking and imaging system for the automated analysis of C. elegans morphology and behavior. Using this system, it is possible to record the behavior of individual nematodes over long time periods and quantify 144 specific phenotypic parameters. Conclusions These tools for phenotypic analysis will provide reliable, comprehensive scoring of a wide range of behavioral abnormalities, and will make it possible to standardize assays such that behavioral data from different labs can readily be compared. In addition, this system will facilitate high-throughput collection of phenotypic data that can ultimately be used to generate a comprehensive database of C. elegans phenotypic information. Availability The hardware configuration and software for the system are available from wschafer@ucsd.edu .	Background The nematode Caenorhabditis elegans is among the most widely studied genetically-tractable experimental organisms. C. elegans is a soil-dwelling animal with a relatively simple and extremely well characterized anatomy; an adult hermaphrodite, for example, contains exactly 959 somatic cells, each with an identified position, morphology and cell lineage. Because of its short generation time, amenability to germline transformation, and completely sequence genome, it is ideally suited for classical and molecular genetic analysis. In particular, since the C. elegans nervous system is simple and well-characterized (the identity and connectivity of each neuron is known), it has become a facile model for studying the molecular basis for nervous system function. Robust behavioral phenotypes have been described for many C. elegans behaviors, including locomotion, egg-laying, mating and feeding, and these phenotypes have proven extremely useful for the genetic dissection of key aspects of neuronal function such as synaptic release, sensory transduction, and neuromuscular signalling [ 1 ]. Historically, a major limitation of such neurobiological studies in C. elegans has been the lack of quantitative methods for the evaluation of behavioral phenotypes. For example, the phenotypes of many behavioral mutants, even those defective in key aspects of neuronal signal transduction, appear subtle to a real-time human observer, and are difficult to assay without time and labor-intensive analysis of video recordings [ 2 - 4 ]. Even the phenotypes of mutants with grossly abnormal behavior are difficult to characterize precisely by manual observation. For example, mutants with striking defects in locomotion (uncoordinated, or Unc mutants) are typically classified using qualitative terms such as "coiler", "kinker", "sluggish", "slow" and "loopy" [ 5 , 6 ]. Since these descriptions are imprecise and subjective, it is extremely difficult if not impossible to assess the phenotypic similarity between two mutants based solely on such characterizations. Another challenge occurs in the analysis of behaviors such as locomotion and egg-laying, which can fluctuate over long time scales or involve infrequently-occurring events that are difficult to evaluate through real-time observation [ 7 ]. Furthermore, the quantitative assays that have been used in C. elegans behavioral studies (e.g. [ 8 ]) generally differ from lab to lab, and this lack of standardization has made effective comparison of data collected by different researchers difficult. To address these problems, we have developed an automated tracking and image analysis system for the quantification of C. elegans behavioral patterns. Using this system, it is possible to record the behavior of individual animals at high magnification over long time periods and to simultaneously quantify a large number of behaviorally-relevant features for subsequent analysis. This system has wide applications for the dissection of complex C. elegans behaviors, and will also make it possible to comprehensively classify the behavioral patterns of C. elegans mutants on a genome-wide scale. By making this system widely available to C. elegans neuroscientists, we intend to define a software architecture that can be continually optimized and upgraded to incorporate new parameters that are useful to worm researchers, as well as a hardware platform that can be expanded to provide additional mechanical capabilities for the research community. Methods To effectively capture the locomotion behavior of a freely-moving worm, it is necessary to acquire a sequence of images from which the animal's position, speed and body posture at any given point in time can be derived. C. elegans are small (1 mm) animals which, in the laboratory, are normally cultured on agar plates covered with a lawn of the common laboratory bacterium Escherichia coli. Nematodes move using an approximately sinusoidal wave motion that is propagated along the anterior/posterior axis in the dorsal/ventral plane. On an agar plate, the animal will normally lie on either its left or right lateral surface, making the waveform associated with movement visible from above. When crawling at maximum speed, an adult nematode travels at a rate of approximately 500 μm/s; thus, under the relatively high magnification (40–50 X) required to measure detailed features of body posture the worm can quickly crawl outside the field of view. It is therefore necessary to incorporate into the imaging system a motorized stage that can automatically follow the animal's movements and keep it in the microscope's visual field. The tracking system described here consists of (i) a Nikon SMZ-800 microscope with a stereoscopic zoom, for visualizing the animals; (ii) a Daedal motorized stage controlled by a National Instruments 4-axis controller, for maintaining the animal in the visual field; (iii) a Cohu monochrome analog CCD camera, for image acquisition; (iv) a Windows computer (PC) with a National Instruments video acquisition board, for tracking and image analysis (Figure 1 ). An optional VCR can also be included in the system for the purpose of cross verification of behavioral tracking. A complete parts list is in Additional file 1 ("hardware"). It should be noted that the software (see below) can, with very minor revisions, be adapted to other programmable motorized stages and frame grabbers that meet the industrial standards; questions about specific pieces of equipment can be addressed to the authors. Software Briefly, the software for the system consists of four basic modules. The first module, called the tracker, allows the system to follow the worm as it crawls around the plate by directing the movements of a motorized stage to maintain the animal in the center of the field of view. As the video acquisition board acquires digital images from the microscope field, the tracker program identifies the animal from each acquired image based on 1. size of the objective isolated from background and 2. the direction of the animal crawling in previous frame if more than two objectives are found Based on the coordinates of the animal's centroid within the field of view, the tracker directs the movements of the stage to recenter the animal in the visual field when animals approach the edge of the image frame. The program then saves an image (460 × 380) of the worm containing visual frame (i.e, the pixels composing the worm body plus the minimum enclosing rectangle), the position of the animal within the field of view, the position of the stage, the time the image was captured, and other information crucial for behavioral analysis. These data are saved into the widely used .avi multimedia format, with a MPEG-4 filter to significantly compress the size of data. Null images are not saved into the .avi format, and the user is notified of the null frame. The highest frame rate with which the tracker can perform these operations with our current hardware setting is 30 frame/sec. Next, the system contains a module (called the converter) that processes the raw images to simplify parameter estimation. First, the grayscale image is thresholded and converted to a binary image representing the worm outline. The image is further simplified by generating a morphological skeleton along the midline of each binary image and then distributing 30 skeleton points along this skeleton. A third module (called Lineup) then orders the backbone points from head to tail. To distinguish head and tail, minor user input is required to achieve 100% accuracy. In wild type, this user input (which involves identifying the head with a mouse click) is only required on 1% of the frames. Otherwise, all image processing is completely automated. Thus, for each raw acquired image, the system generates 4 representations of the animal (Figure 2 ) of increasing complexity: the centroid representing the animal's position, the set of ordered skeleton points representing the animal's body posture, the binary image, which provides information about the size and shape of the animal, and the grayscale image, which retains information about the translucency of the animal. Together the outputs of the first three modules then are used to extract quantitative image features that define the characteristic behavioral pattern of a particular mutant type. During image processing stage, aberrant frames (e.g. containing a 'worm' with suddenly abnormal length) are marked and removed, and the user is notified of the defective image. To obtain this information, the system has a fourth, parameter estimation module (called miner) that measures specific features based on grey/binary image, centroid, or skeleton point analysis that define important parameters related to locomotion or morphology. Broadly speaking, these include measurements of morphology, body posture, movement, and locomotion waveform. Morphological features include measurements of size, length, transparency, and elongation/eccentricity. Body posture features include measurements of body curvature as well as the occurrence of specific postures such as coils and omega turns. Movement features include centroid-based measurements of global speed and direction, skeleton point-based measures of local movement, and the occurrence of directional reversals and large turns. Waveform features include measurements of the frequency and amplitude of body bends, the flex of the animal's body during the locomotion wave, and the frequency and magnitude of foraging movements by the animal's nose. A total of 59 distinct features (Table 1 ) are measured by the system. For most of these features, three statistics (top 5% as maximum, mean and lower 5% as minimum) are calculated for each recording, giving a total of 144 measured parameters. A list of all the features and the algorithms used to generate them are found in supplemental data [see Additional file 2 "algorithms"]. Implementation The software is available in a PC version (compiled and benchmarked on a PC with 1 G Hz Pentium-III running Windows 2000 or XP). Software is written with C/C++, Labview 7.0 and Matlab (release 13), and complied with NI LabWindow 7.0. Installation disk and dataset samples are available upon request for non-profit academic usage with a license fee ($75, charged by National Instruments for the usage of their vision library; see Additional file 3 and 4 , "codes" and "filelist" for details). Worm behavioral image data are in AVI format with a standard MPEG-4 filter (Microsoft MPEG-4 v2). Quantitative morphological and behavioral data are outputted into two widely distributed formats: Microsoft Excel and Microsoft Access. Using this hardware configuration, it is possible to process a 2 Hz 1 min real time data set in less than 5 minute (from image data to final data). Thus, it is feasible to envision using the system to screen for specific behavioral phenotypes among mutagenized C. elegans. Applications We describe here a prototype for a standard, open-source system for automated phenotypic analysis of C. elegans behavior. We anticipate that such a system will be extremely useful to C. elegans neurobiologists, as machine vision offers a number of clear advantages over real-time observation for the characterization of behavioral phenotypes. First, it provides a precise definition of a particular mutant phenotype, facilitating quantitative comparisons between different mutant strains. For example, the waveform parameters have provided detailed information about the effects of neuronal G-protein signalling pathway genes on locomotion behavior. Even phenotypes that are extremely difficult to distinguish by eye (e.g. those of the calcium channel mutants unc-2 and unc-36 ) can be identified with relatively high reliability using the system [ 9 ]. In addition, it has been possible to use our system to reliably score behavioral events without labor-and time-intensive (and potentially biased) human scoring; for example, our system has been used to automatically detect directional reversals with high reliability in a touch avoidance assay [ 10 ]. Other specific postures such as coils can also be detected with high (>90%) reliability (Z. Feng, unpublished data). With appropriate controls, a standardized phenotyping system also makes it possible to compare behavioral data collected by different researchers in different labs with greater precision than is possible using qualitative observer-driven approaches. In particular, a computerized system makes it possible to comprehensively assay multiple aspects of behavior simultaneously, yielding a complex phenotypic signature that can be used for bioinformatic studies [ 11 ]. In the future, we hope to use the tools described here to generate a comprehensive C. elegans phenotypic database that could be used to explore the clustering and relative similarities of mutant phenotypes. Supplementary Material Additional File 1 Hardware list Click here for file Additional File 2 Algorithms for feature measurements Click here for file Additional File 3 Codes Click here for file Additional File 4 File list Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517925.xml
526256	The ProActive trial protocol – a randomised controlled trial of the efficacy of a family-based, domiciliary intervention programme to increase physical activity among individuals at high risk of diabetes [ISRCTN61323766]	Background Increasing prevalence of obesity and disorders associated with sedentary living constitute a major global public health problem. While previous evaluations of interventions to increase physical activity have involved communities or individuals with established disease, less attention has been given to interventions for individuals at risk of disease. Methods/design ProActive aims to evaluate the efficacy of a theoretical, evidence- and family-based intervention programme to increase physical activity in a sedentary population, defined as being at-risk through having a parental family history of diabetes. Primary care diabetes or family history registers were used to recruit 365 individuals aged 30–50 years, screened for activity level. Participants were assigned by central randomisation to three intervention programmes: brief written advice (comparison group), or a psychologically based behavioural change programme, delivered either by telephone (distance group) or face-to-face in the family home over one year. The protocol-driven intervention programme is delivered by trained facilitators, and aims to support increases in physical activity through the introduction and facilitation of a range of self-regulatory skills (e.g. goal setting). The primary outcome is daytime energy expenditure and its ratio to resting energy expenditure, measured at baseline and one year using individually calibrated heart rate monitoring. Secondary measures include self-report of individual and family activity, psychological mediators of behaviour change, physiological and biochemical correlates, acceptability, and costs, measured at baseline, six months and one year. The primary intention to treat analysis will compare groups at one-year post randomisation. Estimation of the impact on diabetes incidence will be modelled using data from a parallel ten-year cohort study using similar measures. Discussion ProActive is the first efficacy trial of an intervention programme to promote physical activity in a defined high-risk group accessible through primary care. The intervention programme is based on psychological theory and evidence; it introduces and facilitates the use of self-regulatory skills to support behaviour change and maintenance. The trial addresses a range of methodological weaknesses in the field by careful specification and quality assurance of the intervention programme, precise characterisation of participants, year-long follow-up and objective measurement of physical activity. Due to report in 2005, ProActive will provide estimates of the extent to which this approach could assist at-risk groups who could benefit from changes in behaviours affecting health, and inform future pragmatic trials.	Background This trial addresses the rise in the burden of disease associated with sedentary living: a major public health problem. Physical inactivity accounts for up to 11.7% of all deaths in developed countries [ 1 ] and it has been causally associated with coronary heart disease, diabetes, osteoporosis and some cancers. The rise in the prevalence of obesity in many countries may be associated with a decline in physical activity. Reversal of this trend will require not only public health programmes to increase activity at societal level, but also interventions to help high-risk individuals increase physical activity and maintain beneficial activity patterns [ 2 ]. This is a trial of such an intervention. It aims to overcome the limitations of previous studies through careful choice and characterisation of the target population, study design, measures, and the interventions under evaluation themselves. The intervention programme, based on theories and evidence from psychology about how best to support behavioural change and maintenance, is potentially generalisable to other settings, target groups and behaviours. Target population The study targets people with a parental family history of Type 2 diabetes and a sedentary lifestyle, who constitute a clearly identifiable high-risk population [ 3 ]. A consistent direct relationship exists between sedentary living and Type 2 diabetes [ 3 , 4 ]. People with a family history of diabetes have a three-fold increased risk of developing diabetes compared to those without; a risk that is magnified by physical inactivity and weight gain [ 3 , 5 ]. At least 40% of the excess risk associated with weight gain might be avoided in such people if their BMI did not exceed 30 kg/m 2 [ 2 , 3 ], and prospective studies support the idea that physical activity reduces weight gain [ 5 , 6 ]. Limitations of previous trials Most trials have evaluated increasing physical activity in the context of established disease. The few published trials of primary prevention in high-risk groups have methodological limitations. They have mainly evaluated brief interventions to increase exercise in the general population delivered through primary care practitioners [ 7 - 9 ], often with very short follow-up (a few weeks). Those such as the Activity Counselling Trial ("ACT"), offering follow-up for two years, are based on an unknown proportion of willing attendees at ambulatory care facilities [ 10 , 11 ]. Evaluation of exercise prescriptions delivered through leisure centres has not been encouraging [ 12 ]. Moreover, participants have been poorly characterised in terms of risk, and most studies have relied on self-reports of exercise. This may inflate differences between groups due to recall bias, and cannot capture changes in either total energy expenditure, or physical activity related energy expenditure [ 13 ]. Measuring total energy expenditure is relevant if an increase in one component of activity results in a decline in another, as activities are substituted, and measuring physical activity related energy expenditure (i.e. total energy expenditure adjusted for resting energy expenditure) is important if this is the aetiologically relevant factor. Three trials among individuals with impaired glucose tolerance in China [ 14 ], Finland [ 15 ] and USA [ 16 ], have established that intensive approaches to lifestyle changes including physical activity can delay progression to diabetes by 58% over three years [ 15 , 16 ] and possibly longer [ 14 ]. Few studies have modelled the long-term effects of physical activity, although available work suggests that interventions for primary prevention of Type 2 diabetes might be cost-effective [ 17 ]. A full review of primary prevention trials [ 5 ] identified no interventions aimed at increasing physical activity alone, without accompanying dietary change or intended weight loss. The current study addresses this gap, and previous methodological shortcomings, by careful characterisation of the participants, year-long follow-up, objective measurement of physical activity, and modelling of the relationship between current behaviour change and future disease risk. Limitations of previous interventions (i) Poorly specified interventions Many of the available trials evaluated interventions that were not explicitly based on psychological theory and evidence, and did not specify clearly which behaviour change techniques were applied by the providers [ 18 - 20 ]. In addition, interventions often used relatively ineffective behaviour change techniques, for instance giving people advice about behaviour change [ 21 , 22 ]. More effective interventions to promote physical activity have applied psychological theory and evidence about how best to support behaviour change [ 16 , 23 ]. The development of the ProActive intervention programme included a review of psychological theories and evidence, through systematic reviews [ 18 , 20 ] and expert meetings, and a one-year feasibility study among 15 willing participants and their families. Based on this work, the Theory of Planned Behaviour (TPB) [ 24 ] was selected as the theoretical framework to inform behavioural determinants targeted in the intervention. Determinants include beliefs and attitudes towards the behaviour (here physical activity), which are elicited at an individual level. Tailoring interventions to personal beliefs is an innovative, but theoretically appropriate application of the TPB. Systematic reviews and expert meetings then informed the selection of potentially effective techniques aimed at changing beliefs. They include reinforcement of positive beliefs, and problem solving in relation to negative beliefs, in order to strengthen motivation. (ii) Attention to adoption and maintenance of physical activity A range of behaviour change techniques with evidence for their effectiveness was used to bridge the gap between intention and action: goal setting and review, action planning, use of prompts, self-monitoring, and reinforcement [ 2 , 18 , 23 , 25 ]. Use of a causal model, linking measured beliefs and attitudes to behaviour will allow subsequent process analysis to better specify both determinants and intervention (Hardeman et al ., 2004. A causal modelling approach to the development of theory-based behaviour change programmes for trial evaluation. Submitted). Major challenges in promoting physical activity are maintenance of behaviour change, and the avoidance of drop out rates that can approach 50% [ 5 , 26 ]. Reviews suggest that theoretical advances in facilitating behaviour maintenance have not been applied in intervention programmes [ 27 , 28 ]. The highest levels of participation have been achieved by home-based interventions, involving frequent professional contact, and promoting enjoyable, informal exercise of moderate intensity, such as walking [ 2 , 8 ]. Behavioural maintenance may be best supported by building habits, using self-regulatory strategies such as repetition of behaviours over time in a constant environment, ongoing goal review, self-monitoring, reinforcement, and relapse prevention [ 28 , 29 ]. The use of mail and telephone contacts is a promising cost-effective approach [ 23 ], especially the use of frequent, brief, support calls [ 2 ]. All these approaches are incorporated into ProActive . ProActive objectives The primary objective of ProActive is to determine the effects of a theoretical- and evidence-based intervention programme on objectively measured physical activity after one year, in sedentary individuals at risk of diabetes and related metabolic abnormalities due to their family history. Three questions are posed: 1. Behaviour change: Can an innovative approach to increasing physical activity achieve clinically important change in this behaviour when offered to a group at increased risk of diabetes? 2. Disease impact: If so, what is the potential for the changes in behaviour achieved in mid-life to reduce the incidence of diabetes in later life? 3. Dose finding: How does delivery of the approach, at two levels of intensity, affect acceptability, efficacy and costs? The trial will estimate the extent to which physical activity and its key psychological mediators are altered by the intervention programme, and assess its acceptability to this high-risk group. It will document the extent to which behaviour change is associated with reduction in weight gain and improvement in physiological and biochemical correlates, and will model the potential impact of the intervention on future risk of diabetes. Intensive, face-to-face interventions may not be a feasible health service model, and there is some evidence that less intensive, continuous support may be as effective [ 2 , 23 ]. The intervention programme is therefore being evaluated at two levels of intensity: 'face-to-face' (delivered at the participants' homes and by telephone) and 'distance' (delivered by one home visit and telephone and correspondence) over one year, in order to inform the most cost-effective intervention programme for wider evaluation. If potential efficacy is demonstrated, we intend to proceed to a multi-centre pragmatic trial of the cost-effectiveness of the approach in practice. Methods/design ProActive is a four-year study with a complex randomised trial design [ 30 ], with central randomisation of willing participants to intervention programmes or comparison. The trial is managed from the Institute of Public Health, University of Cambridge, following MRC guidelines. Ethical approval has been obtained from the Eastern MREC, and West Suffolk, Cambridge, Huntingdon and West Essex LRECs. The study design and patient flows (achieved at recruitment closure, October 2003, and projected to end of study) are shown in Figure 1 . The focus of measurement is the adult offspring of a Type 2 diabetic parent, but the focus of the intervention programme is this individual within a family context. Participants were recruited via parents with diabetes on primary care registers (20 practices), or directly through records of their family history of diabetes (seven of the 20 practices). Sedentary individuals and their families were randomised to facilitation, either 'face-to-face' or 'distance', or to a comparison arm offering a leaflet providing brief advice on the benefits of activity. Psychological, physiological, anthropometric and biochemical data were collected at baseline and one year after randomisation, with psychological data also collected at six months after randomisation. Figure 1 Trial design and patient flows; Oct 2003 (recruitment closure) The study is explanatory in design, and the quality-assured intervention programmes are delivered by carefully trained and supervised family health facilitators with experience of working in primary care or the community, and backgrounds in health promotion, dietetics and nursing. Setting, recruitment and screening The study is set in urban, suburban and rural Cambridgeshire, Essex and West Suffolk, England, in the homes of participants and their families. The study population consists of offspring of people with Type 2 diabetes, aged 30–50 years, without a diagnosis of diabetes, and not considered very active based on self-report at the start of the study (see below). This age range defines a group at risk of weight gain [ 31 , 32 ]. Any individuals found at study entry to have fasting hyperglycaemia [ 33 ] were referred to their family doctor, but retained in the trial. Practice recruitment Once the relevant ethical and PCT approval had been obtained, 53 practice teams in the locality were approached by letter, inviting them to take part in the study, and highlighting the reimbursement of all costs involved. Personalised letters were sent to the practice manager (who we asked to collate responses and reply using a reply slip and Freepost envelope), all partners and nursing staff. Included with each letter was a brief summary of the study and a Research Information Sheet for Practices (RISP) form [ 34 ]. If no response was received, a follow-up phone call was made to the practice manager. A principal investigator and member of the trial team visited interested practice teams, to discuss the study in further detail. All relevant practice staff were encouraged to attend, particularly those who would be involved in the administration of proposed patient surveys. The 20 practices that agreed to take part then received a 'set-up' visit by the trial team. A 'Practice Survey Manual' was created for the practice staff, and the trial team supported the practice teams as needed throughout the survey period. Participant recruitment Initially participants were recruited through their parents; patients with Type 2 diabetes on the diabetes registers of 20 practices ('recruitment method 1'). Patients were written to by their general practitioner, with a description of the study, and asked to provide contact information for any offspring aged 30–50 years, living locally. Consent was also sought for the practice to pass the contact details of the offspring to the research team so that they could invite the offspring directly into the study. Piloting demonstrated feasibility and acceptability of the method, and one reminder was sent after three weeks if no reply was received. From 20 practices, 2631 patients were approached and 2025 (77%) replied, yielding 1238 potentially eligible offspring who were invited to take part in the study. The ratio of approximately one potentially eligible offspring to two patients with diabetes was half our pilot projections, so to increase recruitment we developed a second recruitment approach ('recruitment method 2'). This approach recruited potential participants with a recorded family history of diabetes directly from practices with family history registers, and was feasible in seven of the 20 practices. General practitioners wrote to all patients aged 30–50 years with a recorded family history of diabetes, enclosing a study information sheet, and asking those willing to complete and return to the practice a questionnaire to determine which family member(s) had diabetes, and of which type. Consent was sought for this information and contact details to be passed on to the research team. Using this method, with again one reminder letter, 1340 patients were written to, and 896 (67%) responses were received, with 283 patients interested and eligible. Both recruitment approaches provided 1521 potential trial participants. Practitioners used their discretion in applying both approaches to the exclusion of patients who were physically or mentally unwell. Study population: inclusion and exclusion criteria Activity levels Potential participants recruited by both methods were next written to by the research team with full information about the study and a screening activity questionnaire, describing occupational and leisure activity, based on published questionnaires [ 35 , 36 ], to exclude very active individuals. Two reminder letters with questionnaires were sent at two-week intervals if necessary, giving a response rate of 74%. Respondents were excluded if they reported their occupational activity as 'heavy manual work' [ 35 ]; or 'physical work' if their total score on the leisure questionnaire [ 36 ] was ≥ 20; or 'sedentary' or 'standing' work if their total leisure activity score was ≥ 30. This resulted in exclusion of approximately 30% of those screened, a figure that matches well with the proportion of the UK population designated as active in prevalence surveys [ 37 ]. Study requirements To fulfil measurement requirements participants had to be able to walk briskly, without help, on the flat for 15 minutes. Participants also had to live within reach of the measurement centre and the Family Health Facilitators; defined as a 30-minute average travel time from the study co-ordination centre. Other exclusion criteria included individuals with serious physical or psychiatric illness limiting programme involvement; people with life issues interfering with the study; those known to be pregnant or have diabetes before baseline measurement; and those planning to move away. As shown in Figure 1 , application of these criteria reduced the 837 'interested' responses to 465 potentially eligible individuals, who were telephoned by a trained interviewer to confirm eligibility. Eligible and interested individuals were then scheduled for baseline measurement at either the Ely Research Centre or the Addenbrooke's Hospital Wellcome Trust Clinical Research Facility, where written consent was obtained. Eligible offspring were registered with general practitioners in the Eastern Region of the UK. Prior to both baseline measures and randomisation, these doctors were individually informed about their registered patients' intention to participate in ProActive . Brief details of the trial were sent, together with a request for feedback if the practitioner had any concerns about the offspring's participation, or about the safety of the facilitators making home visits. Randomisation Randomisation was carried out centrally by the trial statistician, using a partial minimisation procedure that dynamically adjusted the randomisation probabilities in order to balance important covariates; body mass index, sex, age, physical activity (individually calibrated heart rate monitoring, see below), family size, and behavioural intentions. Randomisation thus used baseline measures. Thirty-two pairs of siblings and two sibling-triples were cluster randomised to the same study group to avoid contamination, and the remaining 295 participants (81%) were individually randomised. Overall, 365/465 (78%) of those eligible went forward to randomisation. Baseline measures and follow-up At baseline and the end of the study, all participants attend the study centre at either Ely or Cambridge for questionnaires, physiological and anthropometrical measures, and venesection. At six months, psychological and self-reported physical activity data are collected by postal questionnaires. Measures relating to the intervention programme evaluation are collected by the facilitators during the intervention, and we assess reported use of self-regulatory strategies by participants to increase their activity levels at six and twelve months. Compliance with follow-up In similar primary care based trials we have achieved attrition rates of 30% or less [ 38 , 39 ], and at current rates we will exceed the required 300 to complete the study (100 in each group, see Figure 1 ). Maximising retention is an important issue, particularly as the comparison group do not benefit from regular contact with a facilitator. At recruitment, the introductory leaflets for all three arms emphasised the importance of follow-up, irrespective of treatment group. Participants who drop out of the intervention programme are contacted by a principal investigator, and offered an opportunity to give feedback and to confirm drop out from the intervention programme only, or from trial measurement as well. Measurement The distribution of measures across baseline, six-month and one-year follow-up are shown in Table 1 . The principle outcome is an objective measurement of physical activity energy expenditure, the daytime physical activity ratio (dayPAR), which is the ratio of daytime energy expenditure to resting energy expenditure measured using heart rate monitoring with individual calibration for the heart rate-energy expenditure relationship [ 40 , 41 ]. This allows more precise quantification of the relationship between energy expenditure and relevant disease end points than self-report [ 13 ]. The method has been validated against the gold standard techniques of doubly-labelled water and whole-body calorimetry [ 42 ]. Physical activity is also measured by a validated questionnaire covering work, recreation and domestic activity over the previous month and year [ 43 ], and offspring report of usual physical activity patterns among family members and how they changed over the previous year. Table 1 Study measures Measures Baseline 6 months 12 months Questionnaire measures: 1. Godin / EPIC self-reported physical activity [36] 2. Short form State anxiety [45] 3. Risk / worry diabetes + 4. Theory of Planned Behaviour [24] + 5. General Questionnaire, comprising: A) Rose Angina questionnaire B) Smoking, Alcohol & Physical Activity + (smoking & physical activity only) C) Occupation & Social Class D) SF-36 & EQ-5D [44,46] 6. EPAQ (2) [43] 7. Clinical measurement questionnaire (physiological measures and family history) 8. Physical activity of family members + 9. Injury questionnaire + 10. Intervention programme satisfaction + * 11. Skills acquisition + * Physiological measures: [13,40-42,51] Cardiorespiratory fitness & dayPAR parameters Weight, height, % body fat, blood pressure, ECG Biochemical parameters (fasting plasma glucose, glycosylated haemoglobin, insulin, lipids) Blood stored for future genetic testing Costs: Cost to the NHS of facilitator training & salary + Costs of intervention programme delivery + + questionnaires developed for study * intervention programme participants only Oxygen uptake (ml O 2 /kg/body weight) is measured by indirect calorimetry during a submaximal graded treadmill exercise test, and maximal cardiorespiratory fitness (VO 2max ) is estimated using predicted maximal heart rate (i.e. 220 minus age) [ 40 , 42 ]. Self-report measures of well-being and quality of life include subjective health and energy (SF-36) [ 44 ], anxiety [ 45 ], worry about diabetes and perceived vulnerability, and EuroQol (EQ-5D) [ 46 ]. The frequency and severity of physical activity related injury is assessed by study questionnaire at one year. Psychological mediators of physical activity include intention to increase activity over the next year, and its predictors (attitude, subjective norm, perceived behavioural control). These key measures have been developed for the study following the recommendations of Ajzen [ 24 ]. Physiological correlates of behaviour include weight measured on standard scales calibrated at three monthly intervals, body fat percentage measured by bio-electrical impedance (Bodystat, Isle of Man, UK), and systolic/diastolic blood pressure, measured using an automatic sphygmomanometer (Accutorr, UK). Biochemical correlates include fasting plasma glucose, glycosylated haemoglobin, insulin and lipids. We are storing EDTA whole blood samples for future genetic testing. Sociodemographic factors and ECG are also documented at baseline. Cost of the intervention The economic analysis will explore the impact of a physical activity intervention programme on NHS costs. As the study is explanatory in design, we will not conduct a full cost-effectiveness analysis, but aim to provide a cost-description of the delivery of the intervention programmes. We are measuring the costs of delivering the 'face-to-face' and 'distance' intervention programmes via family health facilitators. These costs primarily comprise the training of facilitators, educational materials, travel, and the time that facilitators spend contacting and visiting families (including cancelled visits). Travel costs and contact time are recorded by the facilitators for every trial participant. The cost of facilitator time will be based on national average salaries, employment costs, qualifications, overheads and indirect costs [ 47 ]. Although we do not expect the ProActive intervention programme to have an impact on health service costs in the short term, we are monitoring health service utilisation (hospital, primary and community care) in the last 20% of participants recruited to the study. Participant safety The primary safety concerns for participants in ProActive are cardiovascular and musculoskeletal events associated with the laboratory procedures of treadmill exercise testing and injuries sustained as a consequence of increasing physical activity in everyday life. The cardiorespiratory fitness test used in this study is submaximal, and only undertaken following extensive screening procedures. If a participant exhibits a positive Rose angina questionnaire [ 48 ], a positive physical activity readiness questionnaire [ 49 ] or an abnormal ECG, they are referred to a clinical member of the measurement team for a more detailed medical review. If there are clinical concerns, participants are excluded from the study, and referred to their general practitioner. In over 3000 such tests undertaken by our group using this protocol, no significant adverse events have occurred. Supervising staff are trained and hold current cardio pulmonary resuscitation certificates. Ranges for acceptable results are set for all clinical measures. If these are exceeded, the information is sent to the general practitioner, and the participant informed and advised to consult. As the intervention programme is based on participants' own preferred activities, and emphasises small achievable goals set by the participants, the risk of excess injury is small. Group information about injury will be reported. Participants previously unaware of their familial risk of diabetes may experience anxiety related to awareness of their increased risk status. This is considered in facilitator training, and measures of anxiety, worry about diabetes and perceived vulnerability are included (see above). Data management, quality assurance and exclusion of bias Physiological and anthropometric measures are made in two centres by observers unaware of individuals' group allocation. Biochemical measures are made in one laboratory with established quality assurance systems. Randomisation was undertaken by the trial statistician, independently of the trial co-ordination team, and the data entry team are unaware of study group. The administrative database (participant information), dayPAR values and blood test results are managed in-house, with the latter being double entered. Numeric fields have limiters set so that values outside a defined range cannot be entered. Additionally, any blood results outside the 'normal range' are flagged for confirmation of value. Random checks on administrative data are performed regularly, checking the data on the database against paper records and correcting any errors found. Double data entry of all anthropometric and questionnaire measures is undertaken by an experienced, independent agency, blind to study group (Wyman Dillon Research and Data Management, Bristol, UK). In addition, random checks are applied as described above. Intervention (see Figure 1 ) Intervention programme contacts The family health facilitator contacts participants randomised to the 'face-to-face' and 'distance' interventions, and arranges a home interview including family members. At this introductory interview, personal reasons for increasing physical activity are elicited and reinforced, family participation is encouraged, and the relationships between physical activity, weight gain and prevention of Type 2 diabetes are explained and discussed. In the 'face-to-face' arm this is followed by four visits and two brief support telephone calls over five months. During these interactions the participant and willing family members learn strategies to increase physical activity, for instance selecting activities that they enjoy doing, setting achievable goals, defining action plans, self-monitoring, self-reinforcement and relapse prevention. Pedometers are available for self-monitoring among participants who have chosen walking as their goal. A key difference between this intervention and others currently under evaluation (e.g. ACT) is that there is no absolute target for physical activity defined at the outset. Family members are encouraged to make gradual and continuous increases in their activity, as much as they feel able to, on the understanding that all increases, if maintained, are beneficial. Follow-up continues by monthly telephone calls up to one year, to discuss any difficulties in applying the strategies, and to encourage family members to increase activity further. In the 'distance' arm, following the introductory meeting the intervention programme is delivered by six telephone calls over five months, and then monthly by post up to one year, with content similar to the 'face-to-face' arm. During the phone calls the facilitators encourage the participants to involve family members. Visits and telephone calls take approximately one hour and 45 minutes, respectively. Materials An arm-specific introductory leaflet is used, but otherwise materials are the same for the face-to-face and distance arms. All introductory leaflets include text to encourage retention in the trial. In the comparison arm the leaflets offer brief advice on the benefits of physical activity. Participants in the intervention programme arms are given an educational manual describing the strategies that participants are encouraged to use to increase their habitual activity in a step by step fashion. Promotion of fidelity of intervention delivery Various mechanisms are used to promote the fidelity of delivery of the intervention programme to the underlying psychological theories and intervention programme protocols. A detailed training manual and protocols for each contact were developed, and a Training Officer appointed. Facilitators attended a five day phased course in psychological theories, behaviour change techniques and experiential training in techniques, with six half-days initially, followed by refresher sessions at six months and continuing supervised practice by a clinical psychologist and through peer-appraisal. Facilitators complete a checklist for the introduction of and mastery of self-regulatory strategies by the participant after each contact, and monitor intervention programme attendance and drop-out for each participant. Assessment of fidelity and evaluation of the intervention programme An assessment of adherence by facilitators to the behaviour change techniques specified in the protocols was conducted among a random sample of 27 participants, using reliable coding frames and transcripts of the sessions. The intervention programme evaluation includes: an assessment of the frequency of meetings and telephone calls, proportion of progress reports and postcards sent and progress reports returned, satisfaction with the intervention programme, reported use of self-regulatory strategies by participants at six months and one year, and drop-outs at one year. Statistical procedures Sample size The sample size calculation was initially based on physical activity level (PAL), the ratio of total energy expenditure to estimated basal metabolic rate [ 40 , 41 ], and required 100 individuals completing one-year follow-up in each group. Prior to the measurement of any follow-up data, and endorsed by the Trial Steering Committee, a proposal was made to change the primary outcome measure to dayPAR, the ratio of daytime energy expenditure to resting energy expenditure, on grounds that this outcome consisted entirely of measured rather than estimated quantities. The calculations were based on the Ely cohort study data [ 41 , 50 ], in which the residual standard deviation of one-year change in dayPAR adjusting for baseline was 0.53. With 100 individuals in each group, there is 80% power to detect a difference in mean dayPAR of 0.18 between the combined intervention programme groups and the control group with a two-sided test at the 5% level of significance. This is equivalent to 2 MET hours/day, 30 minutes of brisk walking on the level, or 20 minutes of leisurely bicycling or swimming; a plausible and important increase. The observed difference in mean dayPAR between any pair of groups will be estimated with a 95% confidence interval having the width ± 0.15, equivalent to ± 1.75 MET hours/day, ± 25 minutes brisk walking or ± 15 minutes bicycling or swimming. Calculations were based on equal numbers in each group, and require 300 participants with outcome data at one-year follow-up. Recruitment of 400 participants allowed for 25% attrition after randomisation, and lower interim attrition rates will enable a lower recruitment target of 365 participants (Figure 1 ). Main analyses will be at one year, comparing combined 'face-to-face' and 'distance' versions of the intervention programme with 'brief advice', comparing 'face-to-face' with 'distance' modes of the intervention programme, and estimating the difference between each intervention programme group and 'brief advice' to inform a larger pragmatic trial. Analysis by intention-to-treat will retain individuals within their randomised group regardless of participation. Comparisons will involve an adjustment for baseline physical activity and other variables used in the randomisation. We will undertake sensitivity analyses, assuming a range of potential outcomes for non-completers, informed by available baseline and interim data on non-completers. Non-completers will have multiple data imputed with a 'missing at random' assumption and with sensitivity analyses to represent optimistic and pessimistic scenarios for drop out. Clustering effects by family will be estimated for the primary outcome. A secondary 'dose-response' analysis will use all three randomised groups, over baseline, six months and one year. A 'per protocol' analysis will also be undertaken among those completing the intervention programme. The incremental cost of delivering the 'face-to-face' intervention programme will be compared to the 'distance' and 'brief advice' groups. Modelling will comprise a series of stages Stage 1) The trial will provide evidence on the relationship between observed behaviour change, weight change, and biochemical and physiological correlates. Modelling is facilitated by reference to the Ely Cohort; a prospective population cohort study that began in 1990 and involved 1122 people without known diabetes [ 40 ]. Measurements identical to those used in the Ely Cohort Study are included in ProActive . Stage 2) Using models based on past cohort data, the influence of behaviour change on future diabetes incidence [ 40 , 41 , 51 , 52 ] will be projected, appropriately allowing for uncertainty in the parameter estimates. Simulation methods will be adopted. At this stage other risk factors (e.g. smoking, diet) will be assumed fixed. Stage 3) We will undertake sensitivity analyses on the projections at Stage 2, using a range of plausible assumptions about how behaviour change might affect other risk factors and hence indirectly influence future diabetes risk. Discussion ProActive is the first efficacy trial of physical activity promotion in a defined high-risk group accessible through primary care, evaluating an intervention programme based on theory and evidence. It supports increases in informal activity, through the introduction and facilitation of self-regulatory strategies with regular reinforcement by the facilitator. Due to report in 2005, ProActive has the potential to make substantial contributions to understanding the extent to which such approaches could assist the wide range of at risk groups who could benefit most from increasing their physical activity. The trial team brings together expertise in the epidemiology of diabetes [ 53 ] with intervention development and evaluation [ 18 - 21 , 30 ], measurement from beliefs to self-reported behaviour [ 26 ] and objectively measured energy expenditure [ 13 , 40 , 54 ] and trials [ 55 ], especially in primary care [ 38 , 39 ]. Their complementary contributions will allow both the answering of the main study questions in a robust manner, and the development of theory and method for future studies. Further exploratory work on interactions between genotype, social class and physical activity are planned, which may in the future lead to refinement in selection of the at-risk group. As an adjunct to the measurement of physical activity related energy expenditure by individually calibrated heart rate monitoring, we are also employing measurement of body movement using the MTI-Actigraph [ 54 ] on a proportion of the participants. The combination of the two measurement techniques has the potential to overcome the limitations with either method used alone, and improves the estimates of physical activity related energy expenditure [ 56 ], since the measurement errors associated with the methods are not positively correlated. In terms of the intervention itself, careful measurement along the hypothesised causal path from cognition, through self-reported behaviours to energy expenditure, will enable testing of the application of the Theory of Planned Behaviour in this setting, and of the relationship between the everyday activities that the programme has as its focus and the objectively measured physical activity (Hardeman et al ., 2004. A causal modelling approach to the development of theory-based behaviour change programmes for trial evaluation. Submitted). This will enable replication and further strengthening of effective intervention steps, as well as development of theory. Together, it is expected that the findings will inform the design of future larger scale and more pragmatic preventive programmes promoting physical activity in at-risk groups. List of abbreviations dayPAR = daytime physical activity ratio; the ratio of daytime energy expenditure to resting metabolic rate measured using heart rate monitoring with individual calibration ECG = electrocardiogram MET = metabolic equivalent PAL = physical activity level; the ratio of total energy expenditure to estimated basal metabolic rate measured using heart rate monitoring with individual calibration TPB = Theory of Planned Behaviour VO 2max = maximal oxygen uptake (ml O 2 /kg/min) Competing interests The authors declare that they have no competing interests. Authors' contributions ALK, NW, SG, SS, WH, DS, – Principal Investigators TP – Trial Statistician KW – Trial Co-ordinator Will H – Trial Economist UE – Physical activity measurement All authors read and approved the final manuscript. ALK is the paper guarantor. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC526256.xml
509299	Retroviral DNA Integration: ASLV, HIV, and MLV Show Distinct Target Site Preferences	The completion of the human genome sequence has made possible genome-wide studies of retroviral DNA integration. Here we report an analysis of 3,127 integration site sequences from human cells. We compared retroviral vectors derived from human immunodeficiency virus (HIV), avian sarcoma-leukosis virus (ASLV), and murine leukemia virus (MLV). Effects of gene activity on integration targeting were assessed by transcriptional profiling of infected cells. Integration by HIV vectors, analyzed in two primary cell types and several cell lines, strongly favored active genes. An analysis of the effects of tissue-specific transcription showed that it resulted in tissue-specific integration targeting by HIV, though the effect was quantitatively modest. Chromosomal regions rich in expressed genes were favored for HIV integration, but these regions were found to be interleaved with unfavorable regions at CpG islands. MLV vectors showed a strong bias in favor of integration near transcription start sites, as reported previously. ASLV vectors showed only a weak preference for active genes and no preference for transcription start regions. Thus, each of the three retroviruses studied showed unique integration site preferences, suggesting that virus-specific binding of integration complexes to chromatin features likely guides site selection.	Introduction Retroviral replication requires reverse transcription of the viral RNA genome and integration of the resulting DNA copy into a chromosome of the host cell. A topic of long standing interest has been the chromosomal and nuclear features dictating the location of integration target sites (reviewed in Coffin et al. 1997 ; Bushman 2001 ). Integration site selection has also gained increased interest because of its importance for human gene therapy. Retroviral vectors have been used extensively to deliver therapeutic genes carried in retroviral backbones. However, retroviral integration can take place at many locations in the genome, on occasion resulting in insertional activation of oncogenes (reviewed in Coffin et al. 1997 ; Bushman 2001 ). Recently, two patients undergoing gene therapy for X-linked severe combined immunodeficiency developed leukemia-like illnesses associated with integration of a therapeutic retroviral vector in or near the LMO2 proto-oncogene ( Check 2002 ; Hacein-Bey-Abina et al. 2003 ). Insertional activation of oncogenes by retroviral vectors has also been detected in animal models ( Li et al. 2002 ). With the availability of the complete human genome sequence, large-scale sequence-based surveys of integration sites have become possible ( Schroder et al. 2002 ; Laufs et al. 2003 ; Wu et al. 2003 ). Schroder et al. (2002) investigated targeting of human immunodeficiency virus (HIV) and HIV-based vectors in a human lymphoid cell line (SupT1) and found that genes were favored integration targets. Global analysis of transcription in SupT1 cells showed that active genes were favored for integration, particularly those that were active after infection with the HIV-based vector. Wu et al. (2003) examined targeting of murine leukemia virus (MLV) in human HeLa cells and found that MLV does not strongly favor integration in transcription units, but rather favors integration near sequences encoding mRNA 5′ ends. Here we contrast integration targeting by three retroviruses, avian sarcoma-leukosis virus (ASLV), HIV, and MLV, taking advantage of 1,462 new integration site sequences and matched transcriptional profiling data. We find that ASLV does not favor integration near transcription start sites, nor does it strongly favor active genes. For HIV, we find that active genes are favored for integration in two primary cell types, extending findings from previous studies of transformed cell lines. Cell-type-specific transcription was found to result in cell-type-specific biases in integration site placement. We also reanalyzed the MLV data from Burgess and coworkers ( Wu et al. 2003 ) in parallel, confirming that MLV integration is favored near transcription start sites. Thus it appears that each retrovirus studied to date has a unique pattern of integration site selection within the human genome, suggesting that there may be local recognition of chromosomal features unique to each virus. Results Integration Site Datasets Used in this Study The origins of the 3,127 integration sites studied are summarized in Table 1 . All were generated by acute infection of cells with retroviruses or with viruses generated from retroviral vectors. To isolate integration sites, DNA from infected cells was isolated, cleaved with restriction enzymes, then ligated to DNA linkers. Integration sites were amplified using one primer that bound to the viral DNA end and another that bound the linker, then amplification products were cloned and sequenced ( Schroder et al. 2002 ; Wu et al. 2003 ). Integration sites were mapped on the draft human genome sequence ( Figure 1 ) and local features at integration sites quantified. Figure 1 Relationship between Integration Sites and Transcriptional Intensity in the Human Genome The human chromosomes are shown numbered. HIV integration sites from all datasets in Table 1 are shown as blue “lollipops”; MLV integration sites are shown in lavender; and ASLV integration sites are shown in green. Transcriptional activity is shown by the red shading on each of the chromosomes (derived from quantification of nonnormalized EST libraries, see text). Centromeres, which are mostly unsequenced, are shown as grey rectangles. Table 1 Integration Site Datasets Used in This Study DOI: 10.1371/journal.pbio.0020234.t001 Three integration site datasets were newly determined in this study. Integration by an ASLV vector was analyzed in 293T-TVA cells, which are human 293T cells engineered to express the subgroup A avian retrovirus receptor. Integration by an HIV-based vector was characterized in two types of primary human cells, peripheral blood mononuclear cells (PBMCs) and IMR90 lung fibroblasts. Several previously described datasets were also subjected to further analysis in parallel—HIV integration sites in three transformed cell lines (SupT1 [ Schroder et al. 2002 ], H9, and HeLa [ Wu et al. 2003 ]) and MLV integration sites in HeLa cells ( Wu et al. 2003 ). The use of restriction enzymes to cleave cellular DNA during the cloning of integration sites could potentially introduce a bias in favor of isolating integration events closer to restriction sites. Previous work suggested that integration site surveys were not strongly biased. In one study, an experimental control based on integration in vitro indicated that the cloning and analytical methods used did not detectably bias the conclusions ( Schroder et al. 2002 ). In addition, HIV integration sites cloned by different methods showed a similar preference for active genes ( Carteau et al. 1998 ; Schroder et al. 2002 ; Wu et al. 2003 ), including sites isolated using several different restriction enzymes to cleave cellular DNA prior to linker ligation, and isolation using inverse PCR instead of ligation-mediated PCR. In this study we have added a computational method to address possible biased isolation. Each integration site was paired with ten sites in the human genome randomly selected in silico that were constrained to be the same distance from a restriction site of the type used for cloning as the experimentally determined integration site. Statistical tests were then carried out by comparing each experimentally determined integration site to the ten matched random control sites. In this way any bias due to the placement of restriction sites in the human genome was accounted for in the statistical analysis. All the collections of integration sites were analyzed in this manner, including data previously published in ( Schroder et al. 2002 ; Wu et al. 2003 ). However, direct analysis of the distribution of integration sites without this correction yielded generally similar conclusions, suggesting that restriction site placement did not introduce a strong bias. A detailed description of the statistical analysis is presented in Protocol S1 (p. 2). Integration in Transcription Units For HIV the frequency of integration in transcription units ranged from 75% to 80%, while the frequency for MLV was 61% and for ASLV was 57%. For comparison, about 45% of the human genome is composed of transcription units (using the Acembly gene definition). Analysis using the different catalogs of human genes suggests that somewhat different fractions of the human genome are transcribed, and new information indicates that an unexpectedly large fraction of the human genome may be transcribed into noncoding RNAs ( Cawley et al. 2004 ). However, for comparisons using any catalog of the human genes, the rate of integration in human transcription units determined experimentally was substantially higher than in the matched random control sites ( Protocol S1 , p. 3–11). We next assessed the placement of integration sites within genes and intergenic regions. A previous study revealed that integration by MLV is favored near transcription start sites, but no such bias was seen for HIV ( Wu et al. 2003 ). We investigated this issue for ASLV and reanalyzed all of the available data by comparison to the matched random control dataset ( Figure 2 ). For sites within genes, MLV showed a highly significant bias in favor of integration near transcription start sites ( p = 1.4 × 10 −14 ). No such bias was seen for ASLV ( p > 0.05). For the HIV datasets, two of the four sets showed modest bias (HIV/PBMC, p = 0.0007; HIV/H9 and HeLa combined, p = 0.027). For HIV, the observed statistical significance was sensitive to the details of analytical approach used and is of questionable biological importance. Thus, ASLV and HIV do not strongly favor integration at transcription start sites as was seen with MLV. Figure 2 Integration Intensity in Genes and Intergenic Regions Genes or intergenic regions were normalized to a common length and then divided into ten intervals to allow comparison. The number of integration sites in each interval was divided by the number of matched random control sites and the value plotted. A value of one indicates no difference between the experimental sites and the random controls. Viruses and cell types studied are as marked above each graph. The direction of transcription within each gene is from left to right. Note that our normalization method de-emphasizes favored MLV integration events just upstream of gene 5′ ends (outside transcription units), as reported by Wu et al. (2003) . We carried out an analysis specially designed to identify this effect and confirmed that the regions just upstream of gene 5′ ends are favored for MLV integration when reanalyzed with the matched random control data (unpublished data). Effects of Transcriptional Activity on Integration Transcriptional profiling analysis was carried out in some of the cell types studied, allowing the influence of transcriptional activity on integration site selection to be assessed. Transcriptional profiling was carried out on infected cells so that the data reflected the known influence of infection on cellular gene activity ( Corbeil et al. 2001 ; Schroder et al. 2002 ; Mitchell et al. 2003 ; van 't Wout et al. 2003 ). Prior to isolating RNA for microarray analysis, the SupT1, PBMC, and IMR90 cells were each infected with the HIV-based vector used for HIV integration site isolation. The 293T-TVA cells were infected with the RCAS ASLV vector prior to RNA isolation. RNA samples were harvested 24–48 h after infection. For analysis of MLV and HIV integration sites in HeLa cells, published transcriptional profiling data from uninfected cells were used ( Tian et al. 2002 ). The median expression level (average difference value) of genes hosting integration events was consistently higher than the median of all genes assayed on the microarrays. The ratios (targeted genes/all genes assayed) for HIV ranged from 1.6 to 3.0, indicating that integration targeting in human primary cells (PBMC and IMR90) favored active genes, as shown previously for transformed cell lines ( Schroder et al. 2002 ; Wu et al. 2003 ). For MLV and ASLV the ratios were each 1.3, lower than for HIV but still greater than the chip average. The analysis shown in Figure 3 reveals that the association of integration sites with gene activity was statistically significant for all the HIV datasets. For MLV and ASLV, there was a weak tendency for integration to favor active genes, but the trend did not achieve statistical significance. Figure 3 Influence of Gene Activity on Integration Frequency Expression levels were assayed using Affymetrix HU-95Av2 or HU-133A microarrays and scored by the average difference value as defined in the Affymetrix Microarray Suite 4.1 software package. All the genes assayed by the chip were divided into eight “bins” according to their relative level of expression (the leftmost bin in each panel is lowest expression levels and the rightmost the highest). Genes that hosted integration events were then distributed into the same bins, summed, and expressed as a percent of the total. The y-axis indicates the percent of all genes in the indicated bin. P values were determined using the Chi-square test for trends by comparison to a null hypothesis of no bias due to expression level. All average difference values were ranked prior to analysis, and the analysis was carried out on the ranked data. This was done to avoid possible complications due to differential normalization or other data processing differences arising during work up of the microarrays. All three HIV datasets showed reduced integration in the most highly expressed category of genes analyzed, suggesting that although transcription favors integration, very high level transcription may actually be less favorable ( Figure 3 ). A fourth dataset monitoring infection by an HIV-based vector in a T-cell line also showed this trend (M. Lewinski, P. S., J. R. E., and F. D. B., unpublished data). Possibly this finding is related to that of a previous study of integration by ALV in a model gene that also suggested that high level transcription may disfavor integration ( Weidhaas et al. 2000 ). Tissue-Specific Transcription Results in Tissue-Specific Biases in HIV Integration Site Selection We next investigated the effects of cell-type-specific transcription on integration site selection. For this analysis we used only the three HIV integration site datasets for which we had transcriptional profiling data from infected cells (i. e., SupT1, PBMC, and IMR90), to allow us to control for the effects of infection on transcription. Pairwise comparisons of the microarray datasets for the three cell types showed that the correlation coefficients ranged only from 0.64 to 0.79, indicating that transcriptional activity indeed differed among cell types. We reasoned that since active transcription favors integration, then the genes targeted by integration should on average be more highly expressed in the cell type that hosted the integration event than in either of the other two. Statistical analysis ( Figure 4 ) showed that transcription of targeted genes was higher on average in the cell type hosting the integration event than in either of the other two tested (all comparisons attain p < 0.05 using the Chi-square test for linear trend in proportions). We note, however, that the differences were quantitatively modest, perhaps because much of the cellular program of gene activity is shared among many cell types ( Caron et al. 2001 ; Mungall et al. 2003 ; Versteeg et al. 2003 ). Figure 4 Effects of Tissue-Specific Transcription on Integration Site Selection in Different Cell Types Genes hosting integration events by the HIV vector were analyzed for their expression levels in transcriptional profiling data from IMR90, PBMC, and SupT1 cells. For each gene hosting an integration event, the expression values from the three cell types were then ranked lowest (red), medium (orange), and highest (yellow). The values were summed and displayed separately for each set of integration sites: (A) IMR90 sites, (B) PBMC sites, and (C) SupT1 sites. In each case there was a significant trend for the cell type hosting the integration events to show the highest expression values relative to the other two ( p < 0.05 for all comparisons). Integration Site Selection and Transcriptional Domains We next analyzed factors influencing the placement of integration sites at the chromosomal level, taking into account both gene density and expression (see Figure 1 ). Transcriptional activity was quantified by counting the EST sequence copies for each gene present in a collection of nonnormalized EST libraries ( Mungall et al. 2003 ). EST sequences from many tissues were used to build up the map of transcriptional activity, thus focusing the analysis on transcriptional patterns common to many cell types ( Caron et al. 2001 ; Mungall et al. 2003 ; Versteeg et al. 2003 ). A detailed comparison of the relationship between EST frequency and integration frequency of HIV, MLV, and ASLV is shown for Chromosome 11 in Figure 5 . Integration frequency for HIV closely parallels the transcriptional intensity deduced from EST counts. Fewer sites are available for analyzing MLV and ASLV, but MLV may show a related trend, while it is unclear whether ASLV does so or not. Similar analysis of the other human chromosomes yields similar conclusions (unpublished data). Figure 5 Comparison of Transcriptional Intensity to Integration Intensity on Human Chromosome 11 All data were quantified in 2-Mb intervals. The top line shows summed EST data documenting the “transcriptional intensity” for each chromosomal interval (data from Mungall and al. [2003] ). The bottom three lines show the summed frequency of integration site sequences in each interval. The numbers of ESTs (top) or integration sites (bottom three) are shown on the y-axis. Statistical analysis was carried out comparing integration frequencies to (1) gene density or (2) transcriptional intensity, as measured by the EST counts. All analyses incorporated a comparison to the matched random control set of integration sites. Each type of vector showed a significant positive correlation with gene density (HIV, p = 1.8 × 10 −12 to 3.2 × 10 −38 , depending on the dataset; MLV, p = 2.4 × 10 −22 ; ASLV, p = 3.2 × 10 −9 ) and a stronger association with transcriptional intensity (HIV, p = 3.8 × 10 −19 to 9.7 × 10 −46 ; MLV, p = 5.1 × 10 −35 ; ASLV, p = 1.7 × 10 −11 ). Overall, ASLV showed the weakest association with gene density and transcriptional intensity. Thus, the analysis of transcriptional activity in the context of chromosomal location revealed significant effects of transcription on MLV and ASLV integration. This is in contrast to the study based on transcriptional profiling alone, in which the effect was not statistically significant—however, a similar trend was evident and the general conclusions similar (see Figure 3 ). It appears that adding information on chromosomal position to the gene expression data allowed quantitatively modest effects to reach statistical significance. Substructures within Chromosomal Regions Favorable for Integration Two lines of evidence indicated that the chromosomal regions favorable for integration can be subdivided into favorable and unfavorable segments. In the first study, a computational analysis was carried out to determine the length of the chromosomal segments yielding the best fit between transcriptional intensity and integration intensity. The sizes of the chromosomal regions analyzed were varied systematically from 25 kb to 32,000 kb, and the statistical significance determined for the correlations. This analysis revealed that the segment length yielding the best correlation was comparatively short, around 100–250 kb, the length of one or a few human genes. These conclusions held for HIV, ASLV, and MLV ( Protocol S1 , p. 16–65). An analysis of integration frequency near CpG islands also indicated substructure within regions favorable for integration. CpG islands are chromosomal regions enriched in the rare CpG dinucleotide. These regions commonly correspond to gene regulatory regions containing clustered transcription factor binding sites—consequently, CpG islands are more frequent in gene-rich regions. Previously Wu et al. (2003) reported that CpG islands were positively associated with MLV integration sites but that for HIV integration sites there was no influence. An analysis of the effects of proximity to CpG islands on integration frequency incorporating the matched random control dataset is shown in Figure 6 . The relative integration frequency near CpG islands was found to be much higher than expected by chance for MLV, as reported previously, and slightly higher than expected by chance for ASLV. For HIV, the region surrounding CpG islands was actually disfavored, and this was statistically significant in three out of four datasets. Thus for HIV, broadly favorable gene-dense chromosomal regions actually contain a mixture of favorable clusters of active genes and unfavorable CpG islands. For MLV, in contrast, CpG islands are quite favorable. Figure 6 The Effects of Proximity to CpG Islands Differs for HIV, MLV, and ASLV Integration The viral vectors and cell types studied are indicated by color. A value of one indicates no bias, less than one indicates disfavored integration, and more than one indicates favored integration. The x-axis (from plus or minus 1 kb to 50 kb) indicates distance from the edge of a CpG island in either direction along the genome. The statistical analysis specifically removed the favorable effects of being in a gene and being in a region containing expressed genes to highlight the effects of CpG islands alone. When effects of gene density and activity are left in, HIV integration goes from disfavored at short distances (less than 1 kb) to favored at longer distances (more than 10 kb). This is because at longer distances the association with genes is significant—many CpG islands are within 10 kb of a gene, and genes are favored targets for HIV integration. To carry out this analysis, the numbers of experimentally determined and matched control sites were fitted according to whether they were near a CpG island, whether they were in genes, and the level of the expression density variable. Each variable contributes a “multiplier” for the ratio of the number of experimental to control sites. The multiplier for “near CpG island” is shown ( see Protocol S2 , p. 9–12). High gene density in the human chromosomes is known to correlate with several other features, including high levels of gene expression, high densities of CpG islands, the occurrence of Giemsa-light chromosomal bands, and high G/C content ( Caron et al. 2001 ; Lander et al. 2001 ; Venter et al. 2001 ; Mungall et al. 2003 ; Versteeg et al. 2003 ). The effects of chromosomal banding pattern and G/C content were analyzed statistically and found to favor integration, as expected from the correlation with other favorable features ( Protocol S2 ). A Quantitative Model for Integration Intensity A statistical model was constructed to examine the relative contributions to integration intensity of (1) gene density, (2) gene activity, (3) proximity to CpG islands, (4) G/C content, and (5) location within genes ( Protocol S2 , p. 13–15). Inclusion of each of these parameters improved the fit of the model to the observed experimental datasets, but the quantitative contribution of each parameter varied among the different retrovirus types. The effects of being in a gene or a region with many expressed genes were most important for HIV and ASLV. For MLV, the distance from the transcription start site was the most important parameter. ASLV differed significantly from each of the other datasets ( p < 0.0001), and the model based on the above parameters predicted the placement of ASLV integration sites least well. Thus ASLV is least responsive to the effects of the parameters so far known to affect integration site selection in human cells. Integration Frequency in the Individual Human Chromosomes Figure 7 presents the frequency of integration in each of the different human chromosomes for HIV, MLV, and ASLV. For HIV, each of the datasets is shown separately. A statistical analysis was carried out comparing the observed frequency of integration in each chromosome to that expected from the matched random control ( Protocol S1 , p. 3). As can be seen from the figure, the frequencies were quite different among the different chromosomes. For example, the gene-rich Chromosome 19 showed more integration than expected by chance, while the gene-poor Chromosome 18 showed less integration. The differences in integration frequencies among chromosomes are in part a function of gene density. However, for unknown reasons, the datasets also differed significantly from each other ( p < 2.22 × 10 −16 ). Evidently there are factors—so far unknown—affecting integration targeting that operate at the level of whole chromosomes(a conclusion also reached by Laufs et al. [2003] ). Figure 7 Frequency of Integration in Human Chromosomes Human chromosome numbers are indicated at the bottom of the figure. The number of integration events detected in each chromosome was divided by the number expected from the matched random control. The line at one indicates the bar height expected if the observed number of integration events matched the expected number. Higher bars indicate favored integration, lower bars, disfavored integration. Most of the cell types studied were from human females; too little data were available for the Y chromosome for meaningful analysis. Discussion We report that ASLV, MLV, and HIV have quite different preferences for integration sites in the human chromosomes. HIV strongly favors active genes in primary cells as well as in transformed cell lines. MLV favors integration near transcription start regions and favors active genes only weakly. ASLV shows the weakest bias toward integration in active genes and no favoring of integration near transcription start sites. We expect that these same patterns will be seen for MLV and ASLV integration in different human cell types, because all four HIV datasets yielded similar results, though more data on additional cell and tissue types will be helpful to further evaluate the generality. One of the earliest models for chromosomal influences on integration targeting proposed that condensed chromatin in inactive regions disfavored integration, thereby concentrating integration in more open active chromatin ( Panet and Cedar 1977 ; Vijaya et al. 1986 ; Rohdewohld et al. 1987 ). Integration by HIV, ASLV, and MLV all showed at least a weak bias in favor of integration in active genes, consistent with the idea that open chromatin at active genes favors integration. Also consistent with this idea, heterochromatic regions at human centromeres and telomeres were found to disfavor integration. However, it seems unlikely that relative accessibility is the only feature directing integration site selection, because HIV, ASLV, and MLV each show such distinctive target sequence preferences. Studies of the Ty retrotransposons of yeast, close relatives of retroviruses, have revealed that interactions with bound chromosomal proteins can tether the Ty integration machinery to chromosomes and thereby direct integration to nearby sites ( Boeke and Devine 1998 ; Bushman 2003 ; Sandmeyer 2003 ; Zhu et al. 2003 ). Such a model may explain integration targeting by retroviruses as well ( Bushman 2003 ). HIV integration complexes might bind to factors enriched at active genes, while MLV complexes could bind to factors bound near transcription start sites. In support of this idea, in vitro studies have established that retroviral integrase enzymes fused to sequence-specific DNA-binding domains can direct integration preferentially to local regions when tethered at specific DNA sites ( Bushman 1994 ; Goulaouic and Chow 1996 ; Katz et al. 1996 ). The analysis of chromosomal regions favored for integration also suggested a role for locally bound proteins. Chromosomal regions enriched in active genes were generally favorable, but further analysis revealed interleaved favorable and unfavorable regions. Statistical tests indicated that favorable regions were typically short (100–250 kb), and for HIV these were interspersed with unfavorable regions near CpG islands. CpG islands are thought to be regulatory regions that bind distinctive sets of transcription factors. Thus, a simple model to explain targeting is that a distinctive set of sequence-specific DNA-binding proteins bound at or near CpG islands disfavor HIV integration, while proteins bound in active transcription units are favorable. For MLV, the proteins bound at CpG islands instead favor integration. For ASLV, it is possible that the viral integration machinery does not interact with factors bound in or near genes, explaining the more random distribution of integration sites in the genome. Such a pattern might have evolved to minimize disruption to the host cell chromosomes due to integration. Another possibility, however, is that ASLV does have stricter target site preferences during normal integration in chicken cells, but the targeting system does not function properly in the human cells studied here. According to this idea, putative chicken chromosomal proteins normally bind ASLV integration complexes and direct integration, but the homologous human proteins may be too different to interact properly. It should be possible to investigate this possibility by characterizing ASLV integration in chicken cells, now that the draft chicken genome sequence is completed ( Ren et al. 2003 ). One consequence of the above findings is that integration will differ from tissue to tissue as a consequence of cell-type-specific transcription. To assess effects of tissue-specific transcription, we analyzed HIV integration in three different cell types (SupT1, PBMC, and IMR90). Transcriptional profiling data showed that transcription was significantly different among the three. This allowed an analysis of integration targeting, which showed that highly expressed genes particular to each tissue were favored for integration in that tissue. However, the magnitude of the tissue-specific biases on integration were modest, probably because most of the cellular transcriptional program appears to be common among cell types ( Caron et al. 2001 ; Mungall et al. 2003 ; Versteeg et al. 2003 ). Additional mechanisms could also contribute to targeting. For example, we and others have detected statistically significant biases in integration frequency in whole chromosomes that do not appear to be fully explained by gene density or gene activity ( Schroder et al. 2002 ; Laufs et al. 2003 ; data reported here). Perhaps the intranuclear position of chromosomes may have an influence, since this has been proposed to be relatively fixed for cells of specific types but may differ among cell types ( Boyle et al. 2001 ; Chubb and Bickmore 2003 ). Our data indicate that ASLV has integration site preferences that may make it attractive as a vector for human gene therapy. MLV-based vectors have the unfavorable preference for integration near transcription start sites ( Wu et al. 2003 ). The adverse events arising during X-linked severe combined immunodeficiency gene therapy involved integration of an MLV vector near the transcription start region of the LMO2 proto-oncogene. HIV-based vectors strongly favor integration in active genes, which is also likely to be disruptive to the host cell genome ( Schroder et al. 2002 ; Wu et al. 2003 ). ASLV, in contrast, shows only weak favoring of integration in active genes, and no favoring of integration near transcription start sites. A quantitative model based on gene density, expression, and proximity to transcription start regions fit the ASLV data least well, indicating that ASLV has the weakest bias toward integration in these unfavorable locations. ASLV vectors are known to infect a variety of human cell types (e. g., Valsesia-Wittmann et al. 1994 ; Federspiel and Hughes 1997 ; Hatziioannou and Goff 2001 ; Katz et al. 2002 ) and can transduce nondividing cells ( Hatziioannou and Goff 2001 ; Katz et al. 2002 ), adding to their possible utility. More generally, this study, together with previous work ( Schroder et al. 2002 ; Wu et al. 2003 ), indicates that the selection of different retroviral integration systems can modulate the selection of integration target sites, and this may potentially be exploited for safer gene therapy. Materials and Methods Oligonucleotides used in this study Each oligonucleotide is described by its name, sequence (written 5′ to 3′), and use, in that order. Hinc II adaptor, GTAATACGACTCACTATAGGGCACGCGTGGTCGACGGCCCGGGCTGC, adapter for use with DNA cleaved by 6-cutter restriction enzymes, top strand; mNheIAvrIISpeII adaptor, P-CTAGGCAGCCCG-NH 2 , adapter for 6-cutter restriction enzymes, bottom strand; ASB-9, GACTCACTATAGGGCACGCGT, adapter primer for PCR for I/SupT1, PBMC, and IMR-90; SB-76, GAGGGATCTCTAGTTACCAGAGTCACA, HIV primer for PCR for I/SupT1; ASB-19, GAGATTTTCCACACTGACTAAAAGGGTC, HIV primer for PCR for I/PBMC and IMR-90; ASB-16, GTCGACGGCCCGGGCTGCCTA, adapter primer for nested PCR for I/SupT1, PBMC, and IMR-90; ASB-1, AGCCAGAGAGCTCCCAGGCTCAGATC, HIV primer for nested PCR for I/SupT1; ASB-20, CTGAGGGATCTCTAGTTACCAGAGTCA, HIV primer for nested PCR forPBMC and IMR-90; MseI linker +, GTAATACGACTCACTATAGGGCTCCGCTTAAGGGAC, MseI linker, top strand; MseI linker −, P-TAGTCCCTTAAGCGGAG-NH 2, MseI linker, bottom strand; MseI linker primer, GTAATACGACTCACTATAGGGC, MseI linker primer for first round of PCR; MseI linker nested primer, AGGGCTCCGCTTAAGGGAC, MseI linker nested primer for second round of PCR; BB389, GATGGCCGGACCGTTGATTC, inner ASLV LTR primer for second round of “nested” PCR; BB390, CGATACAATAAACGCCATTTGACCATTC, outer ASLV LTR primer for first round of PCR. Preparation of the ASLV- and HIV-based vectors To produce HIV vector particles, 293T cells were cotransfected with three plasmids: one encoding the HIV vector segment (p156RRLsinPPTCMVGFPWPRE) ( Follenzi et al. 2000 ), the second, the packaging construct (pCMVdeltaR9) ( Naldini et al. 1996 ), and the third, the gene for VSV-G (pMD.G) ( Naldini et al. 1996 ). Forty-eight hours after transfection, supernatants were collected, centrifuged to pellet cellular debris, then filtered through 0.45-μm filters. Viral particles were further purified by centrifugation at 23,000g and resuspended in 1/17 volume of fresh medium. ASLV particles were generated by transfecting the DF-1 chicken embryonic fibroblast cell line (ATCC CRL-12203) with the plasmid RCASBP(A)GFP (from Steve Hughes, National Cancer Institute, Frederick, Maryland). Supernatant was removed from the cells 4 d post transfection (when cells were nearly 100% GFP positive) and filtered though a 0.45-μm syringe filter. Infections. PBMCs were separated from human blood using a ficoll gradient (Amersham Biosciences, Little Chalfont, United Kingdom). 1 × 10 7 PHA, IL-2 prestimulated PBMCs, or IMR-90 cells (passage #36) at 30%–50% confluency (1–2 × 10 6 cells) were infected with the HIV-based vector at an moi of 10 (60 ng p24 per 5 × 10 5 cells). The vector was added to the cells with DEAE-dextran at a final concentration of 5 μg/ml. Forty-eight hours after infection, the cells were pelleted. For RNA isolation, cells were resuspended in 250 μl of PBS and 750 μl of TRIzol and frozen in liquid nitrogen. To determine the extent of infection, cells were analyzed by flow cytometry. For ASLV, supernatant containing RCASBP(A)GFP particles was added to 293T-TVA cells (293T 0.8 cells; a gift from John Young, Salk Institute) at 30%–50% confluency. Forty-eight hours post infection, green fluorescence was seen in approximately 30% of the cells, as determined by examination of the cultures with a fluorescence microscope. DNA was harvested at this point (DNeasy, Qiagen, Valencia, California, United States). RNA from infected cells was also isolated at 48 h post infection (TRIzol) and stored at −80 °C until used for transcriptional profiling analysis. RNA was isolated from infected cell cultures, and samples from each were used for hybridization on one Affymetrix (Santa Clara, California, United States) microarray. Integration site determination. HIV integration sites were cloned by ligation-mediated PCR essentially as described in Schroder et al. (2002) . ASLV integration site determination using ligation-mediated PCR was carried out essentially as described in Wu et al. (2003) . Oligonucleotides used are summarized above. All novel integration site sequences are deposited at the National Center for Biotechnology Information (NCBI) (accession numbers CL528318–CL529767). Integration sites from earlier studies were reanalyzed on the November 2002 freeze of the human genome sequence (using the University of California at Santa Cruz browser), and a few were excluded because they did not find matches of sufficiently high quality on the new draft sequence, accounting for slightly different numbers than in previous reports. This study used primarily the Acembly and Ensemble human gene catalogs; similar results were generally obtained when the Unigene, RefGene, or GeneScan catalogs of the human genes were used ( Protocol S1 , p. 3–10). Transcriptional profiling analysis Transcriptional profiling was carried out using Affymetrix microarrays as described in Schroder et al. (2002) . Gene expression levels (average difference values) were analyzed using Affymetrix Microarray Suite 4.1 software. All novel transcriptional profiling datasets reported here are deposited at NCBI (GEO dataset numbers GSE1407, GSE1408, GSE1409, and GSE1410). For the analysis in Figure 3 , a complication was introduced by the fact that the HU95A chips used have multiple probe sets for some genes but not others. In our analysis all probe sets were accepted and analyzed in these cases. Statistical analysis A detailed description of the statistical methods used is presented in Protocols S1 and S2 . Supporting Information Protocol S1 Association of Genomic Features with Integration—Part 1 (322 KB PDF). Click here for additional data file. Protocol S2 Association of Genomic Features with Integration—Part 2 (128 KB PDF). Click here for additional data file. Accession Numbers The GenBank ( http://www.ncbi.nlm.nih.gov/ ) accession numbers for the novel integration site sequences discussed in this paper are CL528318–CL529767. The GenBank GEO dataset numbers for the novel transcriptional profiling datasets reported here are GSE1407, GSE1408, GSE1409, and GSE1410.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC509299.xml
517931	Opioid modulation of GABA release in the rat inferior colliculus	Background The inferior colliculus, which receives almost all ascending and descending auditory signals, plays a crucial role in the processing of auditory information. While the majority of the recorded activities in the inferior colliculus are attributed to GABAergic and glutamatergic signalling, other neurotransmitter systems are expressed in this brain area including opiate peptides and their receptors which may play a modulatory role in neuronal communication. Results Using a perfusion protocol we demonstrate that morphine can inhibit KCl-induced release of [ 3 H]GABA from rat inferior colliculus slices. DAMGO ([D-Ala(2), N-Me-Phe(4), Gly(5)-ol]-enkephalin) but not DADLE ([D-Ala2, D-Leu5]-enkephalin or U69593 has the same effect as morphine indicating that μ rather than δ or κ opioid receptors mediate this action. [ 3 H]GABA release was diminished by 16%, and this was not altered by the protein kinase C inhibitor bisindolylmaleimide I. Immunostaining of inferior colliculus cryosections shows extensive staining for glutamic acid decarboxylase, more limited staining for μ opiate receptors and relatively few neurons co-stained for both proteins. Conclusion The results suggest that μ-opioid receptor ligands can modify neurotransmitter release in a sub population of GABAergic neurons of the inferior colliculus. This could have important physiological implications in the processing of hearing information and/or other functions attributed to the inferior colliculus such as audiogenic seizures and aversive behaviour.	Background Sounds are first converted into neuronal signals in the inner ear and then conveyed to the cerebral cortex via a number of discrete brain areas including the inferior colliculus. Each of these areas receives ascending pathways carrying signals from one or both ears and descending pathways from higher brain centres. The current knowledge of the neurochemical events occurring at each of these brain centres is limited [ 1 , 2 ]. In the inferior colliculus studies have been carried out to characterise the role of GABAergic neurons especially in sound localisation which is believed to be one of the main functions of this brain area [ 3 , 4 ]. Additionally the inferior colliculus has ben implicated in audiogenic seizures and aversive behaviour in which GABAergic neurons may also play an important role. [ 5 , 6 ] The neuronal communication occurring in the inferior colliculus is likely to be influenced by modulatory systems such as those of peptidergic neurotransmitters. Opiate receptor gene expression, immunoreactivity and activity in the inferior colliculus have been described [ 7 - 10 ] although detailed studies on the effect of opiate on GABA neurotransmitter release in this brain regions have not been carried out. Three classes of opiate peptides endorphins, dynorphins and enkephalins activate μ, κ and δ-opiate receptors subtypes respectively [ 12 ]. Recently a fourth related receptor ORL1 activated by the peptide nociceptin has been identified and its distinct pharmacology has been described [ 13 ]. All opiate receptors are associated with either Go or Gi subunits and they mediate inhibitory actions including pre-synaptic inhibition of neurotransmitter release. Different mechanisms of inhibition of neurotransmitter release have been reported in various tissues and neurons [ 14 ]. For example, in the periaqueductal gray stimulation of opiate receptors and their associated G-proteins results in the activation of potassium channels [ 15 ] while in the hippocampus, inhibition of the GABAergic activity by opioid is independent of potassium channel activation [ 16 ]. In previous work we have established the presence and distribution of opiate receptors in the adult and developing rat cochlea suggesting that the opiate system has a role in hearing function [ 17 , 18 ]. In order to extend our knowledge of the role of opiate system in hearing it is necessary to characterise its presence and role also in the auditory pathways. Our hypothesis was that opiate peptides can modulate synaptic function in the auditory pathways by pre-synaptically altering the release of other neurotransmitters. To test this hypothesis we have used opiate drugs to inhibit the release of [ 3 H]GABA from inferior colliculus slices Results and Discussion KCl-induced [ 3 H]GABA release Inferior colliculus slices pre-incubated with [ 3 H]GABA were perfused for 30 min and stimulated twice with 25 mM KCl to elicit neurotransmitter release. The eluate was collected in 1 ml fractions and the released radioactivity was assessed by scintillation counting. Figure 1 shows two examples of typical release profiles from slices perfused with either Krebs buffer throughout (control) or with Krebs buffer for fractions 1–7 and with Krebs containing 1 μM morphine for the remaining fractions, where both samples were stimulated with KCl at the time corresponding to fraction 4 and 12. The two peaks were referred to as S1 and S2 and occured approximately 2 fractions after the application of KCl due to the buffer volume contained in the tubes feeding into the incubation chamber. Values of the radioactivity eluted are expressed as fractional release which is the ratio of the radioactivity released in a particular fraction divided by the total amount of radioactivity contained in the tissue immediately prior to that fraction. The variation in the value of S1 of the two profiles shown in Fig 1 , both induced by KCl alone, reflects the variation in amount of tissue present in each of the elution chambers and illustrates the need for utilising the ratio of the two peaks (S2/S1) of each elution profile as a mean to detect the effect of the modulating drug. Morphine modulation of KCl induced [ 3 H]GABArelease The effect of different concentrations of morphine on KCl-induced [ 3 H]GABA release is shown in Figure 2 . Both 1 μM and 5 μM but not 100 nM morphine caused a significant decrease of [ 3 H]GABA release from the inferior colliculus slices. The effect of 1 μM morphine was antagonised by the antagonist naloxone (10 μM) which was perfused from one fraction before the addition of morphine. The perfusion of naloxone alone did not cause a significant effect on [ 3 H]GABA release. These data strongly indicate that morphine modulates the release of [ 3 H]GABA via activation of opiate receptors. The reduction in [ 3 H]GABA release calculated as the change in S2/S1 ratios in the presence and absence of morphine during S2 was 16% (p < 0.01). These data agree with previous reports on the presence of both GABA neurons and opiate receptors and peptides in the inferior colliculus [ 7 ]. In addition a functional inter-relationship is established between the two systems which could be of physiological significance. Specific role of μ opiate receptors In order to establish which of the opiate receptor subtypes are involved in the modulation of the [ 3 H]GABA release, morphine was substituted by either 1 μM DAMGO, DADLE or U69593 which specifically activate μ, δ and κ opiate receptors respectively (Fig. 3 ). Only DAMGO (1 μM) had a significant effect on [ 3 H]GABA release, an effect that was again antagonised by naloxone. DAMGO, as well as morphine, reduced the amount of [ 3 H]GABA release by 16% (p < 0.01) indicating that only μ opioid receptors participate in the regulation of GABA release. Higher concentrations of DAMGO (5 μM) did not have greater effects on [ 3 H]GABA release (not shown). Data from our lab (unpublished) and from others [ 9 , 10 ] indicate that mRNA transcripts or receptor binding for all three opiate receptor subtypes are present in the inferior colliculus. Further work is required to establish the roles of the δ - and κ-opioid receptors in the inferior colliculus. Receptor desensitisation A possible explanation for the relatively low effect of opiate agonist on [ 3 H]GABA release (16%) could be that during the exposure to opiate agonists, down-regulation of the opiate receptors may occur [ 19 ]. To address this possibility experiments were carried out in the presence of the protein kinase C inhibitor bisindolylmaleimide I (BIM). BIM has been shown to inhibit receptor desensitisation [ 20 , 21 ] and to reverse tolerance to opiate drugs which involves opiate receptor desensitisation. [ 22 , 23 ]. Because BIM is solubilised in DMSO additional control assays were carried out to check for the effect of the solvent. The results indicate (Fig. 4 ) that BIM had no effect on the extent of morphine inhibition of [ 3 H]GABA release. While there is no direct proof that BIM had its reported effect on the tissue, the data indicate that receptor desensitisation may not be the cause of the relatively low percentage effect of morphine. Co-localisation of μ-opiate receptors and GABAergic neurons Another possible explanation for the small (16%) reduction of [ 3 H]GABA release by opiate agonists may be the limited number of GABAergic neurons that express opiate receptors. To address this question inferior colliculi slices were double labelled with guinea pig antibodies against μ-opioid receptors and with rabbit antibodies against glutamic acid decarboxylase (the enzyme uniquely responsible for the synthesis of GABA) Species specific secondary antibodies conjugated to red and green fluorochromes allowed the detection of both antigens on the same slide (Fig. 5 ). Although these results were qualitative it was evident that staining of glutamic acid decarboxylase was more extensive than that of μ-opiate receptors and that only a few GABAergic neurons showed co-localisation of μ-opiate receptors. These data are consistent with the proposal that only a sub-population of GABAergic neurons are under the influence of opiate receptors. Establishing the nature of the GABAergic neurons that express opiate receptors will be an important task in understanding the role of opiate signalling in the inferior colliculus. Conclusions This study has demonstrated that in the rat inferior colliculus slices opiate agonists can inhibit KCl-induced [ 3 H]GABA release via activation of the μ-opiate receptor subtype. The amount of [ 3 H]GABA released in presence of opiate agonists was 16% lower than in control slices. This relatively low level of decrease is probably not due to receptor desensitization occurring during the assay but rather to a relatively small population of GABAergic neurons in the inferior colliculus expressing μ-opiate receptors. The small effect of the opiate compounds could also indicate that modulation of GABA release is not their major role, but it could still be of physiological significance. Together with its reported role in audiogenic seizures and aversive behaviour, the inferior colliculus is an important neuronal centre for auditory processing containing both ascending and descending fibres. The identification of the role of opiate peptides and possibly other modulatory system in the inferior colliculus and other areas of the auditory pathway may allow a better understanding of the mechanism of the hearing system and possibly offer a target for therapeutic intervention in hearing dysfunction. Alternatively, elucidation of the role opiate peptides in the inferior colliculus could provide information about regulation of audiogenic seizures and aversive behaviour. Methods Materials Opiate agonist and antagonists, (Morphine, DAMGO [d-Ala(2), N-Me-Phe(4), Gly(5)-ol]-enkephalin, DADLE [D-Ala2, D-Leu5]-enkephalin, U69593 and naloxone were purchased from SIGMA, UK. Antibodies against μ-opiate receptor AB1774 (guinea pig polyclonal) and glutamic acid decarboxylase AB1511 (rabbit polyclonal) and species specific pre-absorbed secondary antibodies (donkey anti rabbit IgG FITC and donkey anti guinea pig IgG rhodamine) were purchased from Chemicon UK. Both antibodies were raised against synthetic peptides, and have been used in several immunocytochemical investigations of rat tissue. [ 24 , 25 ] AB1511 recognises the two isoforms of the enzyme in a Western blot (65/68 KDa) while antibody AB1511 recognises μ-opiate receptors immunocytochemically in the same tissues as other simlilar antibodies and by insitu hybridisation (Chemicaon data sheets, ) Krebs carbonate buffer: NaCl 118 mM, KCl 4.84 mM, CaCl 2 2.4 mM, NaHCO 3 25 mM, MgSO 4 1.8 mM KH 2 PO 4 1.2 glucose 9.5 mM. Animals Sprague Dawley rats, approximately 200 g, were obtained from UCL Biological Services. All animal experiments were carried out in accordance to the Animal (Scientific Procedure) Act 1986, UK. Slices preparations Rats were stunned and killed by cervical dislocation. The skull was opened and the whole brain removed. The inferior colliculus was dissected out by two coronal transections, the first between the cerebellum and the inferior colliculus and the second between the inferior colliculus and the superior coliculus. The slice was placed horizontally and medullar tissue ventral to the inferior colliculus was removed. The inferior colliculus was then placed on a tissue chopper and sliced into 250 μm coronal sections. Individual slices were separated under a dissecting microscope in Krebs buffer. Neurotransmitter release As previously described, slices were incubated in 5 ml oxygenated (95% 0 2 / 5% CO 2 ) Krebs buffer containing GABA transaminase inhibitor aminooxyacetic acid (100 μM) at 32°C for 5 min [ 26 ]. [ 3 H]GABA was added to give a final concentration of 11 nM and incubated in a shaking water bath for 30 min. Slices were distributed into 6 superfusion chambers between filter papers (Brandel Superfusion System) and perfused at 0.5 ml/min with oxygenated Krebs buffer. Following a 30 min perfusion required to reach a steady state (non-stimulated) [ 3 H]GABA release, 2 min fractions (1 ml) were collected. In order to evoke sub-maximal GABA release the slices were perfused for 2 min with 25 mMKCl at 6–8 min and 22–24 min of the fractionation time (fraction 4 and 12). At the end of each experiment (17 fractions) the tissue and the filter papers were collected and incubated with 500 μl Soluene-350 for 20 min and neutralised with 200 μl of glacial acetic acid. Scintillant (Packard, Ultima Gold, 3 ml) was added to all tissue samples and eluted fractions and the radioactivity was measured by scintillation counting (Wallac 1409). Each drug treament was repeated in several experiments as indicated in the figures and the ratios of the S2/S1 peaks were averaged. Statistical significance of the effect of treatments was analysed by one way ANOVA using Excel (Microsoft, USA) Immunocytochemistry Inferior colliculi were dissected as decribed above and fixed in 4% paraformaldehyde in phosphate buffer saline (PBS: 137 mM NaCl, 2.7 mM KCl, 4.3 mM Na 2 HPO 4 .7H 2 O, 1.4 mM KH 2 PO 4 pH 7.3) for 1 hour, washed 3 times in PBS and incubated overnight at 4°C in 30%sucrose in PBS. Coronal sections (20 μm) were cut with a cryostat and collected on poly-lysine coated glass slides and allowed to dry. The sections were blocked in 10% normal donkey serum diluted in PBS containing 0.25% bovine serum albumin and 0.1% Triton X-100 (PBS-A) for 30 min at room temperature. Subsequently, they were incubated for 12 hours at 4°C with the combination of primary antibodies (rabbit anti GAD and guinea pig anti mu opioid receptor) diluted 1:500 in PBS containing 1% bovine serum albumin and 0.3% Triton X-100 (PBS-B).. Slides were washed 3 × 10 min with PBS-B and incubated with secondary antibodies diluted 1:200 in PBS-B, for 2 hours at room temperature. The secondary antibodies used were donkey anti rabbit conjugated with fluorescein (AP182F) and donkey anti guinea pig conjugated with rhodamine (AP182R). Finally, the sections were rinsed in PBS-B for 10 min and in PBS for 2 × 10 min and then mounted in Citifluor (Agar). The immunoreactivity was visualized under the confocal microscopy (LSM 510 META Carl Zeiss, Germany). Authors' contributions WT carried out the majority of the experiments, NJ carried out initial experiments and established assay conditions, JC provided expertise in neurotransmitter release studies and participated in the design of the study and analysis of the data, PP provided expertise in the double labelling studies, HD, AF and PG participated in the design of the study and analysis of the data, SOC conceived of the study and participated in its design and coordination. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517931.xml
544399	Audit of therapeutic interventions in inpatient children using two scores: are they evidence-based in developing countries?	Background The evidence base of clinical interventions in paediatric hospitals of developing countries has not been formally assessed. We performed this study to determine the proportion of evidence-based therapeutic interventions in a paediatric referral hospital of a developing country Methods The medical records of 167 patients admitted in one-month period were revised. Primary diagnosis and primary therapeutic interventions were determined for each patient. A systematic search was performed to assess the level of evidence for each intervention. Therapeutic interventions were classified using the Ellis score and the Oxford Centre for Evidence Based Medicine Levels of Evidence Results Any dehydration due to diarrhoea (59 cases) and pneumonia (42 cases) were the most frequent diagnoses. Based on Ellis score, level I evidence supported the primary therapeutic intervention in 21%, level II in 73% and level III in 6% cases. Using the Oxford classification 16%, 8%, 1% and 75% therapeutic interventions corresponded to grades A, B, C, and D recommendations, respectively. Overall, according to Ellis score, 94% interventions were evidence based. However, out of the total, 75% interventions were based on expert opinion or basic sciences. Most children with mild to moderate dehydration (52 cases) were inappropriately treated with slow intravenous fluids, and most children with non-complicated community acquired pneumonia (42 cases) received intravenous antibiotics Conclusions Most interventions were inappropriate, despite the availability of effective therapy for several of them. Diarrhoeal dehydration and community acquired pneumonia were the most common diagnoses and were inappropriately managed. Existing effective interventions for dehydration and pneumonia need to be put into practice at referral hospitals of developing countries. For the remaining problems, there is the need to conduct appropriate clinical studies. Caution must be taken when assigning the level of evidence supporting therapeutic interventions, as commonly used classifications may be misleading	Background Previous studies have shown that medical interventions based on scientific evidence range from 10 to 80% [ 1 - 4 ]. These studies were performed in a broad spectrum of patients and settings in developed countries. In paediatric practice, the proportion of evidence based interventions reported ranges from 75 to 91% [ 5 - 8 ]. These studies are clearly relevant to the quality of care in developed countries, with strong health systems, widely available high technology and qualified human resources. However, health services in developing countries are frequently weak, and they face too often severe lack of expensive technology. In addition, the level of qualification of health personnel may vary greatly within health facilities, even at referral level and in urban areas. Moreover, the prevalent childhood illnesses in developed countries are not necessarily those prevalent in developing countries. The assessment of quality of care at referral level for severely ill children is an important component of the efforts for reducing the child mortality rate in poor countries, for reducing the burden on health systems and for investing money in favour of high priority health interventions [ 9 , 10 ]. Thus we were prompted to assess the proportion of interventions based on sound scientific base in a paediatric referral setting of a developing country. Methods Referral care provided to children hospitalized in the paediatric department of the Instituto Especializado de Salud del Niño (IESN) was assessed for evidence base. IESN is a national paediatric hospital with more than 500 inpatient beds, serving mostly patients from deprived socioeconomic areas of Lima and inner cities of the country. The clinical records of 195 children aged 1 month through 16 years old and hospitalized during January 2003 were initially revised. One of the investigators (NYC) assessed the clinical records of children and assigned to each one a primary diagnosis and one or more primary therapeutic interventions, on the basis of the main clinical features and/or definitive diagnostic laboratory aids. Patients in whom a primary diagnosis was not possible to determine or those without a clear diagnosis were excluded. The primary intervention was defined as the treatment or other manoeuvre that represented the most important attempt to cure, alleviate, or care for the patient in respect of his or her primary diagnosis [ 3 ]. To determine the level of evidence for each primary intervention, Cochrane reviews were searched. If there was not such a systematic review, a search through PubMed (MEDLINE) was performed by one of the investigators (NYC). All Cochrane reviews were searched through their own search tools. For PubMed, the period of search was 1966 through 2002. The key words used included those related to the primary diagnosis (e.g., pneumonia). Limits: "Title", "All child: 0–18 years". Publication type was sequentially searched for "Practice Guideline", "Meta-Analysis", "Randomized Controlled Trial" and "Review". Articles in English or Spanish were included. The National Guideline Clearinghouse was additionally visited for additional references. Published recommendations for judging the quality of guidelines were used for deciding the selection of the guidelines [ 11 ]. In addition, Clinical Evidence was used whenever deemed pertinent. The level of evidence assigned to interventions was based on Ellis score (levels I, II and III) and the Oxford Centre for Evidence Based Medicine Levels of Evidence (grades A, B, C and D) [ 3 , 12 ]. Ellis score considers one or more interventions for a given diagnosis as one primary intervention. For comparison purposes we ranked each individual intervention through Oxford classification. As a next step to our study, we planned the dissemination of the results among the hospital policy makers and the suggestion of corrective courses of action for those interventions needing improvement. Results Results of the search strategies for PubMed are included as an appendix [See Additional File 1 ]. A guideline on management of pain in sickle cell disease was found in the National Guideline Clearinghouse website and the results of the search are also shown at the end of the appendix [See Additional File 1 ]. Overall, one hundred and ninety five clinical records were revised. Twenty eight clinical records were excluded because of undefined primary diagnosis (Table 1 ). One hundred and sixty seven remaining clinical records were further assessed. The most frequent primary diagnoses are shown in Table 2 , being diarrhoeal dehydration and pneumonia the main causes for hospitalization. The childhood prevalent diseases are quite constant throughout the year at our hospital and thus it is unlikely that the results would have been different if we had chosen another study period. Out of 167 primary interventions, 21% were supported by level I evidence, 73% were level II, and 6% were ranked as level III, according to Ellis classification [ 3 ]. Table 3 shows that most interventions classified as level I are referred to acute asthma exacerbations. Nebulized beta-agonists and systemic corticosteroids in bronchiolitis, and antibiotics for acute otitis media were considered level I according to Ellis. They were classified as D{5} according to Oxford Centre for Evidence Based Medicine Levels of Evidence (Table 3 ) [ 12 ]. Most assessed interventions were considered as level II (Table 4 ). Diarrhoeal dehydration and community-acquired pneumonia were the predominant diagnoses. Considering each prescription separately (fluid restriction, furosemide, spironolactone and captopril in heart failure, for example) we obtained 146 interventions. When we assessed them through the Oxford classification, 11% interventions were classified as grade B, 1% as grade C, and 88% as grade D. Appropriate interventions for the same diagnoses presented in Table 4 and ranked by Oxford classification are shown in Table 5 . Overtly unsubstantiated therapy according to Ellis classification is shown in Table 6 . Considering levels I and II evidence-based therapy, 94% of therapeutic interventions were evidence based through Ellis classification. Using the Oxford classification, we obtained 193 individualized therapeutic interventions, that were classified as Grades A (16%), B (8%), C (1%), and D (75%). Comparison of grade of recommendation of the prescribed intervention with the appropriate one is shown in Table 7 . It will be used as summary evidence documenting our current hospital health care quality standard. Discussion In this study 94% of therapeutic interventions were evidence-based by Ellis score. It may seem encouraging that more than 90% of therapeutic decisions in a referral paediatric hospital of a developing country are evidence based. However, the level II of evidence from Ellis includes interventions based in cohort studies, case-control studies, case series, expert's opinion, and even those based in basic sciences. We attempted therefore to classify the primary interventions according to more specific criteria. Using the Oxford classification, 75% of therapeutic interventions were based in expert opinions or in basic sciences (Grade of Recommendation D). Some limitations of our study include the possible author' bias when assigning the primary diagnosis and primary intervention. The assignment of diagnosis by the clinician may have been influenced by both the choice of treatment and the available evidence. Only one of us classified the primary intervention. In addition, we evaluated only a primary intervention for a single primary diagnosis. Actually, many patients had more than one diagnosis and obviously more than one therapeutic intervention. We used as evidence-base for rating the interventions assessed in our study, guidelines and evidence-based resources published in the developed world. This raises the issue of whether they are fully applicable to our setting. The most prevalent conditions found were diarrhoeal dehydration and community acquired pneumonia. For both of them we used British produced guidelines [ 20 , 21 ] because there were not Cochrane reviews on them and because the guidelines fulfilled recommended criteria for methodological quality of published guidelines [ 11 ]. The main recommendation of the guidelines on diarrhoea favours rapid oral rehydration over intravenous rehydration for children with mild to moderate dehydration [ 20 ]. This recommendation is based on several studies performed in both developed and developing countries and thus it can be applicable to both settings. The only concern on the applicability from setting to setting is that related to the osmolarity of the oral rehydration solution (ORS). The guidelines recommend a solution with 60 mmol/l of sodium, whereas a recent expert consensus found sufficient evidence to recommend the universal use of an ORS containing 75 mmol/l of sodium [ 49 ]. Regarding community acquired pneumonia, the British guidelines recommend antibiotic treatment for all children with pneumonia, due to the difficulties in identifying the aetiology, and they also specify criteria for hospitalization [ 21 ]. These recommendations are in agreement with the World Health Organization published guidelines [ 50 ]. The main difference is that the WHO guidelines rest on fast breathing and chest retraction for the diagnosis of pneumonia, whereas the British guidelines emphasize the role of chest x-rays. Chest x-rays are widely available in referral hospitals in developing countries and thus they should be used in addition to the clinical findings. We acknowledge that the evidence derived from studies performed in developed countries should be translated with caution to developing settings. However, when the native research is scarce or of low quality, we think that the transfer of knowledge from the developed countries is an acceptable approach, as far as the particular characteristics of patients in developing countries are considered on an individual basis. Dehydration due to diarrhoea and pneumonia were the most frequent diagnoses. Oral rehydration for diarrhoeal dehydration and antibiotics for pneumonia are considered as interventions with sufficient evidence for implementing them widely [ 9 , 10 ]. In our study, all children with mild to moderate dehydration were treated with slow intravenous infusion, and most children with uncomplicated community acquired pneumonia received intravenous antibiotics. In addition to their enormous potential for saving lives, outpatient antibiotic therapy for pneumonia and outpatient oral rehydration can drastically reduce the rate of hospitalizations, the hospital stay, the hospital mortality rate, and the costs incurred. At our hospital, the mean stay time for hospitalized children is 4.7 days, and the mean crude mortality rate is 3.6%. We estimated the cost of managing hospitalized children with pneumonia and diarrhoeal dehydration as US$ 10.6/day and US$ 8.6/day, respectively. These costs are referred only to hospital bed and laboratory tests. A substantial amount of money could be saved treating these conditions on an outpatient basis. We planned the dissemination of our results among the hospital policy makers. The tools that will be suggested for improving the standards include the development and systematic application of locally produced guidelines and/or the adaptation of published guidelines. A useful alternative that has been experienced for several years at our inpatient ward unit is to make available personal computers connected to Internet for attending physicians, residents and interns, and to encourage the use of online evidence-based resources. This last alternative may work better, particularly where there are motivated physicians who are able to lead the efforts for improving the health care standards. However, the ultimate decision to systematically introduce and monitor the suggested changes will rest on hospital managers. Such changes should also depend on taking into account the role of several other determinants of the clinical decision making by individual practitioners, such as continuous training, motivation, time, availability of drugs, equipments and supplies, supervision, and long-term health system strengthening strategies. Conclusions Caution must be taken when assigning the level of evidence that supports therapeutic interventions, as commonly used classifications may be misleading. Existing effective interventions for dehydration and pneumonia are not being implemented in developing countries, even at referral level, and thus there is the need to change the current medical interventional behaviours. For the remaining diagnoses, the majority of assessed interventions were based on weak or non-existent evidence, highlighting the need to conduct appropriate clinical studies. Competing interests The author(s) declare that they have no competing interests. Authors' contributions NYC and LH conceived and designed the study. All authors analyzed and interpreted the data and contributed to earlier drafts of the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: Supplementary Material Additional File 1 "Appendix: Literature search results" details of the Medline (PubMed) search strategy and of the results are provided in this additional file. The selected articles are in black, bold characters. In addition, the search results from the National Guideline Clearinghouse website are provided for painful crisis in sickle cell anaemia. Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544399.xml
535574	Mathematics Is Biology's Next Microscope, Only Better; Biology Is Mathematics' Next Physics, Only Better	Joel Cohen offers a historical and prospective analysis of the relationship between mathematics and biology	Although mathematics has long been intertwined with the biological sciences, an explosive synergy between biology and mathematics seems poised to enrich and extend both fields greatly in the coming decades ( Levin 1992 ; Murray 1993 ; Jungck 1997 ; Hastings et al. 2003 ; Palmer et al. 2003 ; Hastings and Palmer 2003 ). Biology will increasingly stimulate the creation of qualitatively new realms of mathematics. Why? In biology, ensemble properties emerge at each level of organization from the interactions of heterogeneous biological units at that level and at lower and higher levels of organization (larger and smaller physical scales, faster and slower temporal scales). New mathematics will be required to cope with these ensemble properties and with the heterogeneity of the biological units that compose ensembles at each level. The discovery of the microscope in the late 17th century caused a revolution in biology by revealing otherwise invisible and previously unsuspected worlds. Western cosmology from classical times through the end of the Renaissance envisioned a system with three types of spheres: the sphere of man, exemplified by his imperfectly round head; the sphere of the world, exemplified by the imperfectly spherical earth; and the eight perfect spheres of the universe, in which the seven (then known) planets moved and the outer stars were fixed ( Nicolson 1960 ). The discovery of a microbial world too small to be seen by the naked eye challenged the completeness of this cosmology and unequivocally demonstrated the existence of living creatures unknown to the Scriptures of Old World religions. Mathematics broadly interpreted is a more general microscope. It can reveal otherwise invisible worlds in all kinds of data, not only optical. For example, computed tomography can reveal a cross-section of a human head from the density of X-ray beams without ever opening the head, by using the Radon transform to infer the densities of materials at each location within the head ( Hsieh 2003 ). Charles Darwin was right when he wrote that people with an understanding “of the great leading principles of mathematics… seem to have an extra sense” ( F. Darwin 1905 ). Today's biologists increasingly recognize that appropriate mathematics can help interpret any kind of data. In this sense, mathematics is biology's next microscope, only better. Conversely, mathematics will benefit increasingly from its involvement with biology, just as mathematics has already benefited and will continue to benefit from its historic involvement with physical problems. In classical times, physics, as first an applied then a basic science, stimulated enormous advances in mathematics. For example, geometry reveals by its very etymology (geometry) its origin in the needs to survey the lands and waters of Earth. Geometry was used to lay out fields in Egypt after the flooding of the Nile, to aid navigation, to aid city planning. The inventions of the calculus by Isaac Newton and Gottfried Leibniz in the later 17th century were stimulated by physical problems such as planetary orbits and optical calculations. In the coming century, biology will stimulate the creation of entirely new realms of mathematics. In this sense, biology is mathematics' next physics, only better. Biology will stimulate fundamentally new mathematics because living nature is qualitatively more heterogeneous than non-living nature. For example, it is estimated that there are 2,000–5,000 species of rocks and minerals in the earth's crust, generated from the hundred or so naturally occurring elements ( Shipman et al. 2003 ; chapter 21 estimates 2,000 minerals in Earth's crust). By contrast, there are probably between 3 million and 100 million biological species on Earth, generated from a small fraction of the naturally occurring elements. If species of rocks and minerals may validly be compared with species of living organisms, the living world has at least a thousand times the diversity of the non-living. This comparison omits the enormous evolutionary importance of individual variability within species. Coping with the hyper-diversity of life at every scale of spatial and temporal organization will require fundamental conceptual advances in mathematics. The Past The interactions between mathematics and biology at present follow from their interactions over the last half millennium. The discovery of the New World by Europeans approximately 500 years ago—and of its many biological species not described in religious Scriptures—gave impetus to major conceptual progress in biology. The outstanding milestone in the early history of biological quantitation was the work of William Harvey, Exercitatio Anatomica De Motu Cordis et Sanguinis In Animalibus (An Anatomical Disquisition on the Motion of the Heart and Blood in Animals) (Harvey 1847), first published in 1628. Harvey's demonstration that the blood circulates was the pivotal founding event of the modern interaction between mathematics and biology. His elegant reasoning is worth understanding. From the time of the ancient Greek physician Galen (131–201 C.E.) until William Harvey studied medicine in Padua (1600–1602, while Galileo was active there), it was believed that there were two kinds of blood, arterial blood and venous blood. Both kinds of blood were believed to ebb and flow under the motive power of the liver, just as the tides of the earth ebbed and flowed under the motive power of the moon. Harvey became physician to the king of England. He used his position of privilege to dissect deer from the king's deer park as well as executed criminals. Harvey observed that the veins in the human arm have one-way valves that permit blood to flow from the periphery toward the heart but not in the reverse direction. Hence the theory that the blood ebbs and flows in both veins and arteries could not be correct. Harvey also observed that the heart was a contractile muscle with one-way valves between the chambers on each side. He measured the volume of the left ventricle of dead human hearts and found that it held about two ounces (about 60 ml), varying from 1.5 to three ounces in different individuals. He estimated that at least one-eighth and perhaps as much as one-quarter of the blood in the left ventricle was expelled with each stroke of the heart. He measured that the heart beat 60–100 times per minute. Therefore, the volume of blood expelled from the left ventricle per hour was about 60 ml × 1/8 × 60 beats/minute × 60 minutes/hour, or 27 liters/hour. However, the average human has only 5.5 liters of blood (a quantity that could be estimated by draining a cadaver). Therefore, the blood must be like a stage army that marches off one side of the stage, returns behind the scenes, and reenters from the other side of the stage, again and again. The large volume of blood pumped per hour could not possibly be accounted for by the then-prevalent theory that the blood originated from the consumption of food. Harvey inferred that there must be some small vessels that conveyed the blood from the outgoing arteries to the returning veins, but he was not able to see those small vessels. His theoretical prediction, based on his meticulous anatomical observations and his mathematical calculations, was spectacularly confirmed more than half a century later when Marcello Malpighi (1628–1694) saw the capillaries under a microscope. Harvey's discovery illustrates the enormous power of simple, off-the-shelf mathematics combined with careful observation and clear reasoning. It set a high standard for all later uses of mathematics in biology. Mathematics was crucial in the discovery of genes by Mendel ( Orel 1984 ) and in the theory of evolution. Mathematics was and continues to be the principal means of integrating evolution and genetics since the classic work of R. A. Fisher, J. B. S. Haldane, and S. Wright in the first half of the 20th century ( Provine 2001 ). Over the last 500 years, mathematics has made amazing progress in each of its three major fields: geometry and topology, algebra, and analysis. This progress has enriched all the biological sciences. In 1637, René Descartes linked the featureless plane of Greek geometry to the symbols and formulas of Arabic algebra by imposing a coordinate system (conventionally, a horizontal x-axis and a vertical y-axis) on the geometric plane and using numbers to measure distances between points. If every biologist who plotted data on x–y coordinates acknowledged the contribution of Descartes to biological understanding, the key role of mathematics in biology would be uncontested. Another highlight of the last five centuries of geometry was the invention of non-Euclidean geometries (1823–1830). Shocking at first, these geometries unshackled the possibilities of mathematical reasoning from the intuitive perception of space. These non-Euclidean geometries have made significant contributions to biology in facilitating, for example, mapping the brain onto a flat surface ( Hurdal et al. 1999 ; Bowers and Hurdal 2003 ). In algebra, efforts to find the roots of equations led to the discovery of the symmetries of roots of equations and thence to the invention of group theory, which finds routine application in the study of crystallographic groups by structural biologists today. Generalizations of single linear equations to families of simultaneous multi-variable linear equations stimulated the development of linear algebra and the European re-invention and naming of matrices in the mid-19th century. The use of a matrix of numbers to solve simultaneous systems of linear equations can be traced back in Chinese mathematics to the period from 300 B.C.E. to 200 C.E. (in a work by Chiu Chang Suan Shu called Nine Chapters of the Mathematical Art ; Smoller 2001 ). In the 19th century, matrices were considered the epitome of useless mathematical abstraction. Then, in the 20th century, it was discovered, for example, that the numerical processes required for the cohort-component method of population projection can be conveniently summarized and executed using matrices ( Keyfitz 1968 ). Today the use of matrices is routine in agencies responsible for making official population projections as well as in population-biological research on human and nonhuman populations ( Caswell 2001 ). Finally, analysis, including the calculus of Newton and Leibniz and probability theory, is the line between ancient thought and modern thought. Without an understanding of the concepts of analysis, especially the concept of a limit, it is not possible to grasp much of modern science, technology, or economic theory. Those who understand the calculus, ordinary and partial differential equations, and probability theory have a way of seeing and understanding the world, including the biological world, that is unavailable to those who do not. Conceptual and scientific challenges from biology have enriched mathematics by leading to innovative thought about new kinds of mathematics. Table 1 lists examples of new and useful mathematics arising from problems in the life sciences broadly construed, including biology and some social sciences. Many of these developments blend smoothly into their antecedents and later elaborations. For example, game theory has a history before the work of John von Neumann ( von Neumann 1959 ; von Neumann and Morgenstern 1953 ), and Karl Pearson's development of the correlation coefficient ( Pearson and Lee 1903 ) rested on earlier work by Francis Galton (1889) . Table 1 Mathematics Arising from Biological Problems The Present To see how the interactions of biology and mathematics may proceed in the future, it is helpful to map the present landscapes of biology and applied mathematics. The biological landscape may be mapped as a rectangular table with different rows for different questions and different columns for different biological domains. Biology asks six kinds of questions. How is it built? How does it work? What goes wrong? How is it fixed? How did it begin? What is it for? These are questions, respectively, about structures, mechanisms, pathologies, repairs, origins, and functions or purposes. The former teleological interpretation of purpose has been replaced by an evolutionary perspective. Biological domains, or levels of organization, include molecules, cells, tissues, organs, individuals, populations, communities, ecosystems or landscapes, and the biosphere. Many biological research problems can be classified as the combination of one or more questions directed to one or more domains. In addition, biological research questions have important dimensions of time and space. Timescales of importance to biology range from the extremely fast processes of photosynthesis to the billions of years of living evolution on Earth. Relevant spatial scales range from the molecular to the cosmic (cosmic rays may have played a role in evolution on Earth). The questions and the domains of biology behave differently on different temporal and spatial scales. The opportunities and the challenges that biology offers mathematics arise because the units at any given level of biological organization are heterogeneous, and the outcomes of their interactions (sometimes called “emergent phenomena” or “ensemble properties”) on any selected temporal and spatial scale may be substantially affected by the heterogeneity and interactions of biological components at lower and higher levels of biological organization and at smaller and larger temporal and spatial scales ( Anderson 1972 , 1995 ). The landscape of applied mathematics is better visualized as a tetrahedron (a pyramid with a triangular base) than as a matrix with temporal and spatial dimensions. (Mathematical imagery, such as a tetrahedron for applied mathematics and a matrix for biology, is useful even in trying to visualize the landscapes of biology and mathematics.) The four main points of the applied mathematical landscape are data structures, algorithms, theories and models (including all pure mathematics), and computers and software. Data structures are ways to organize data, such as the matrix used above to describe the biological landscape. Algorithms are procedures for manipulating symbols. Some algorithms are used to analyze data, others to analyze models. Theories and models, including the theories of pure mathematics, are used to analyze both data and ideas. Mathematics and mathematical theories provide a testing ground for ideas in which the strength of competing theories can be measured. Computers and software are an important, and frequently the most visible, vertex of the applied mathematical landscape. However, cheap, easy computing increases the importance of theoretical understanding of the results of computation. Theoretical understanding is required as a check on the great risk of error in software, and to bridge the enormous gap between computational results and insight or understanding. The landscape of research in mathematics and biology contains all combinations of one or more biological questions, domains, time scales, and spatial scales with one or more data structures, algorithms, theories or models, and means of computation (typically software and hardware). The following example from cancer biology illustrates such a combination: the question, “how does it work?” is approached in the domain of cells (specifically, human cancer cells) with algorithms for correlation and hierarchical clustering. Gene expression and drug activity in human cancer. Suppose a person has a cancer. Could information about the activities of the genes in the cells of the person's cancer guide the use of cancer-treatment drugs so that more effective drugs are used and less effective drugs are avoided? To suggest answers to this question, Scherf et al. (2000) ingeniously applied off-the-shelf mathematics, specifically, correlation—invented nearly a century earlier by Karl Pearson ( Pearson and Lee 1903 ) in a study of human inheritance—and clustering algorithms, which apparently had multiple sources of invention, including psychometrics ( Johnson 1967 ). They applied these simple tools to extract useful information from, and to combine for the first time, enormous databases on molecular pharmacology and gene expression ( http://discover.nci.nih.gov/arraytools/ ). They used two kinds of information from the drug discovery program of the National Cancer Institute. The first kind of information described gene expression in 1,375 genes of each of 60 human cancer cell lines. A target matrix T had, as the numerical entry in row g and column c, the relative abundance of the mRNA transcript of gene g in cell line c. The drug activity matrix A summarized the pharmacology of 1,400 drugs acting on each of the same 60 human cancer cell lines, including 118 drugs with “known mechanism of action.” The number in row d and column c of the drug activity matrix A was the activity of drug d in suppressing the growth of cell line c, or, equivalently, the sensitivity of cell line c to drug d. The target matrix T for gene expression contained 82,500 numbers, while the drug activity matrix A had 84,000 numbers. These two matrices have the same set of column headings but have different row labels. Given the two matrices, precisely five sets of possible correlations could be calculated, and Scherf et al. calculated all five. (1) The correlation between two different columns of the activity matrix A led to a clustering of cell lines according to their similarity of response to different drugs. (2) The correlation between two different columns of the target matrix T led to a clustering of the cell lines according to their similarity of gene expression. This clustering differed very substantially from the clustering of cell lines by drug sensitivity. (3) The correlation between different rows of the activity matrix A led to a clustering of drugs according to their activity patterns across all cell lines. (4) The correlation between different rows of the target matrix T led to a clustering of genes according to the pattern of mRNA expressed across the 60 cell lines. (5) Finally, the correlation between a row of the activity matrix A and a row of the target matrix T described the positive or negative covariation of drug activity with gene expression. A positive correlation meant that the higher the level of gene expression across the 60 cancer cell lines, the higher the effectiveness of the drug in suppressing the growth of those cell lines. The result of analyzing several hundred thousand experiments is summarized in a single picture called a clustered image map ( Figure 1 ). This clustered image map plots gene expression–drug activity correlations as a function of clustered genes (horizontal axis) and clustered drugs (showing only the 118 drugs with “known function”) on the vertical axis ( Weinstein et al. 1997 ). Figure 1 Clustered Image Map of Gene Expression–Drug Activity Correlations Plotted as a function of 1,376 clustered genes (x-axis) and 118 clustered drugs (y-axis). From http://discover.nci.nih.gov/external/CIM_example3/cgi_user_matrix.html . (updated 27 April 2000; accessed 7 October 2004). This image is more recent than the published image ( Scherf et al. 2000 ). Used by permission of John N. Weinstein. What use is this? If a person's cancer cells have high expression for a particular gene, and the correlation of that gene with drug activity is highly positive, then that gene may serve as a marker for tumor cells likely to be inhibited effectively by that drug. If the correlation with drug activity is negative, then the marker gene may indicate when use of that drug is contraindicated. While important scientific questions about this approach remain open, its usefulness in generating hypotheses to be tested by further experiments is obvious. It is a very insightful way of organizing and extracting meaning from many individual observations. Without the microscope of mathematical methods and computational power, the insight given by the clustered image map could not be achieved. The Future To realize the possibilities of effective synergy between biology and mathematics will require both avoiding potential problems and seizing potential opportunities. Potential problems. The productive interaction of biology and mathematics will face problems that concern education, intellectual property, and national security. Educating the next generation of scientists will require early emphasis on quantitative skills in primary and secondary schools and more opportunities for training in both biology and mathematics at undergraduate, graduate, and postdoctoral levels ( CUBE 2003 ). Intellectual property rights may both stimulate and obstruct the potential synergy of biology and mathematics. Science is a potlatch culture. The bigger one's gift to the common pool of knowledge and techniques, the higher one's status, just as in the potlatch culture of the Native Americans of the northwest coast of North America. In the case of research in mathematics and biology, intellectual property rights to algorithms and databases need to balance the concerns of inventors, developers, and future researchers ( Rai and Eisenberg 2003 ). A third area of potential problems as well as opportunities is national security. Scientists and national defenders can collaborate by supporting and doing open research on the optimal design of monitoring networks and mitigation strategies for all kinds of biological attacks ( Wein et al. 2003 ). But openness of scientific methods or biological reagents in microbiology may pose security risks in the hands of terrorists. Problems of conserving privacy may arise when disparate databases are connected, such as physician payment databases with disease diagnosis databases, or health databases with law enforcement databases. Opportunities. Mathematical models can circumvent ethical dilemmas. For example, in a study of the household transmission of Chagas disease in northwest Argentina, Cohen and Gürtler (2001) wanted to know—since dogs are a reservoir of infection—what would happen if dogs were removed from bedroom areas, without spraying households with insecticides against the insect that transmits infection. Because neither the householders nor the state public health apparatus can afford to spray the households in some areas, the realistic experiment would be to ask householders to remove the dogs without spraying. But a researcher who goes to a household and observes an insect infestation is morally obliged to spray and eliminate the infestation. In a detailed mathematical model, it was easy to set a variable representing the number of dogs in the bedroom areas to zero. All components of the model were based on measurements made in real villages. The calculation showed that banishing dogs from bedroom areas would substantially reduce the intensity of infection in the absence of spraying, though spraying would contribute to additional reductions in the intensity of infection. The model was used to do an experiment conceptually that could not be done ethically in a real village. The conceptual experiment suggested the value of educating villagers about the important health benefits of removing dogs from the bedroom areas. The future of a scientific field is probably less predictable than the future in general. Doubtless, though, there will be exciting opportunities for the collaboration of mathematics and biology. Mathematics can help biologists grasp problems that are otherwise too big (the biosphere) or too small (molecular structure); too slow (macroevolution) or too fast (photosynthesis); too remote in time (early extinctions) or too remote in space (life at extremes on the earth and in space); too complex (the human brain) or too dangerous or unethical (epidemiology of infectious agents). Box 1 summarizes five biological and five mathematical challenges where interactions between biology and mathematics may prove particularly fruitful. Box 1. Challenges Here are five biological challenges that could stimulate, and benefit from, major innovations in mathematics. Understand cells, their diversity within and between organisms, and their interactions with the biotic and abiotic environments. The complex networks of gene interactions, proteins, and signaling between the cell and other cells and the abiotic environment is probably incomprehensible without some mathematical structure perhaps yet to be invented. Understand the brain, behavior, and emotion. This, too, is a system problem. A practical test of the depth of our understanding is this simple question: Can we understand why people choose to have children or choose not to have children (assuming they are physiologically able to do so)? Replace the tree of life with a network or tapestry to represent lateral transfers of heritable features such as genes, genomes, and prions ( Delwiche and Palmer 1996 ; Delwiche 1999 , 2000a , 2000b ; Li and Lindquist 2000 ; Margulis and Sagan 2002 ; Liu et al. 2002 ; http://www.life.umd.edu/labs/Delwiche/pubs/endosymbiosis.gif ). Couple atmospheric, terrestrial, and aquatic biospheres with global physicochemical processes. Monitor living systems to detect large deviations such as natural or induced epidemics or physiological or ecological pathologies. Here are five mathematical challenges that would contribute to the progress of biology. Understand computation. Find more effective ways to gain insight and prove theorems from numerical or symbolic computations and agent-based models. We recall Hamming: “The purpose of computing is insight, not numbers” ( Hamming 1971 , p. 31). Find better ways to model multi-level systems, for example, cells within organs within people in human communities in physical, chemical, and biotic ecologies. Understand probability, risk, and uncertainty. Despite three centuries of great progress, we are still at the very beginning of a true understanding. Can we understand uncertainty and risk better by integrating frequentist, Bayesian, subjective, fuzzy, and other theories of probability, or is an entirely new approach required? Understand data mining, simultaneous inference, and statistical de-identification ( Miller 1981 ). Are practical users of simultaneous statistical inference doomed to numerical simulations in each case, or can general theory be improved? What are the complementary limits of data mining and statistical de-identification in large linked databases with personal information? Set standards for clarity, performance, publication and permanence of software and computational results.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535574.xml
545044	Dynamic changes of serum SARS-Coronavirus IgG, pulmonary function and radiography in patients recovering from SARS after hospital discharge	Objective The intent of this study was to examine the recovery of individuals who had been hospitalized for severe acute respiratory syndrome (SARS) in the year following their discharge from the hospital. Parameters studied included serum levels of SARS coronavirus (SARS-CoV) IgG antibody, tests of lung function, and imaging data to evaluate changes in lung fibrosis. In addition, we explored the incidence of femoral head necrosis in some of the individuals recovering from SARS. Methods The subjects of this study were 383 clinically diagnosed SARS patients in Beijing, China. They were tested regularly for serum levels of SARS-CoV IgG antibody and lung function and were given chest X-rays and/or high resolution computerized tomography (HRCT) examinations at the Chinese PLA General Hospital during the 12 months that followed their release from the hospital. Those individuals who were found to have lung diffusion abnormities (transfer coefficient for carbon monoxide [D L CO] < 80% of predicted value [pred]) received regular lung function tests and HRCT examinations in the follow-up phase in order to document the changes in their lung condition. Some patients who complained of joint pain were given magnetic resonance imaging (MRI) examinations of their femoral heads. Findings Of all the subjects, 81.2% (311 of 383 patients) tested positive for serum SARS-CoV IgG. Of those testing positive, 27.3% (85 of 311 patients) were suffering from lung diffusion abnormities (D L CO < 80% pred) and 21.5% (67 of 311 patients) exhibited lung fibrotic changes. In the 12 month duration of this study, all of the 40 patients with lung diffusion abnormities who were examined exhibited some improvement of lung function and fibrosis detected by radiography. Of the individuals receiving MRI examinations, 23.1% (18 of 78 patients) showed signs of femoral head necrosis. Interpretation The lack of sero-positive SARS-CoV in some individuals suggests that there may have been some misdiagnosed cases among the subjects included in this study. Of those testing positive, the serum levels of SARS-CoV IgG antibody decreased significantly during the 12 months after hospital discharge. Additionally, we found that the individuals who had lung fibrosis showed some spontaneous recovery. Finally, some of the subjects developed femoral head necrosis.	Introduction Severe acute respiratory syndrome (SARS) is a new infectious disease in humans. The first victim of SARS to be diagnosed was a businessman from the city of Foshan in Guangdong Province, China. SARS patients may present with a spectrum of symptoms and signs, ranging from relatively asymptomatic to fulminant pneumonitis and death [ 1 ]. Lung injury caused by the SARS coronavirus (SARS-CoV) is one of the main clinical manifestations in SARS patients, significantly affecting their prognosis. A regular follow-up survey of SARS patients in the convalescent phase would be helpful to evaluate any changes in acquired immune function, pulmonary function, bones and joints over the course of time. At present, there have been few reports about the relationship between the prognosis for recovery and the degree of lung injury caused by the SARS-CoV. In addition, a study of the serum levels of the specific IgG antibody against SARS-CoV is needed because it is the major immunologic protection to aid in recovery and is essential to avoid repeated infection with SARS-CoV. It has been 14 months since the World Health Organization officially declared the global outbreak of SARS to be under control [ 2 ]. The present study focused on the dynamic changes in the IgG antibody levels against SARS-CoV and in lung lesions in the discharged but recovering SARS patients as measured by lung function and imaging tests. The phenomenon of femoral head necrosis was also investigated in those SARS patients who complained of chronic bone and joint pain during the one year follow-up after discharge from the hospital. Methods All of the subjects of this study were discharged from Beijing Xiaotangshan Hospital, Beijing Armed Police Hospital, and Chinese 309 PLA Hospital, and all gave their informed consent. Study Protocol The subjects of our investigation were 383 clinically diagnosed SARS patients in the convalescent phase (160 male and 223 female, average age 38.2 ± 13.6 years) undergoing testing from May, 2003 to June, 2004. Each clinical diagnosis was based on the Clinical Diagnosis Standard for SARS Patients issued by the Ministry of Chinese Public Health [ 3 ]. All participants in the study had met the specified criteria for discharge from the hospital [ 4 ]. On the first visit, each patient was given a routine pulmonary function test (ventilation and diffusion function: SensorMedics 2200 pulmonary function test apparatus, U.S.A.), a chest X-ray examination and serum SARS-CoV specific antibody (SARS-CoV IgG) test at the Chinese PLA General Hospital, Beijing, P.R. China. Those individuals suspected of having pulmonary fibrotic changes received high resolution computerized tomography (HRCT) examination of their lungs. Individuals who complained of chronic pain in their bones and joints or who had difficulty walking received femoral head magnetic resonance imaging (MRI) examinations. Each patient returned a month after the first visit followed by one visit every 3 months. Serum SARS-CoV IgG was tested at each return visit. If negative results were obtained twice consecutively, the case was regarded as a misdiagnosis and the patient did not undergo a follow-up survey. Patients with positive SARS-CoV IgG and abnormal pulmonary diffusion received regular pulmonary function tests and those showing pulmonary fibrosis in imaging examinations received further regular HRCT examinations. Some individuals observed to have avascular necrosis of the femoral head received MRI examinations 3–6 months later. Clinical Diagnostic Criteria for the Patients with SARS Disease in Mainland China [ 3 ] (1) Epidemiological history (1.1) The individual has a history of close contact with SARS patients or is part of a cluster of cases of SARS or there is clinical evidence of having infected other patients. (1.2) The individual has a history of recent travel to an area where SARS cases have been reported within 2 weeks and secondary infected SARS cases have been found. (2) Symptoms and signs Acute onset of SARS generally begins with a prodrome of fever with a temperature >38°C, sometimes accompanied by chills, myalgia and anthralgia, headaches, and fatigue. Upper respiratory tract symptoms of catarrh are not prominent, although cough may be present. If present, it is mainly a dry cough, occasionally with blood streak sputum. Some individuals have chest discomfort, and severe cases may present with tachypnea, panting, and even respiratory distress. Generally, there are no obvious pulmonary signs among SARS patients. Wet rales and signs of lung consolidation, as well as decreased respiration and other signs of pleural effusion can occasionally be found. Note: Some patients do not show initial symptoms of fever, especially those who have had recent surgery or those having chronic diseases. (3) Routine laboratory examinations White blood cell counts are generally normal or below normal, with decreased absolute lymphocyte counts. (4) Chest radiological examinations The typical imaging profile of SARS is of multiple patchy opacities with bilateral distribution. The opacities are usually ground-glass in appearance, sometimes with air space consolidation, evolving progressively over the course of the disease. The evolution is very rapid in some cases, resulting in the confluence of lesions and large areas of opacification in a short time. If the chest radiological examination is negative, reexamination after 1 to 2 days should be done. (5) Antibiotic therapy is ineffective Suspected cases: In accordance with 1.1+2+3, 1.2+2+4 or 2+3+4. Clinically diagnosed cases: In accordance with 1.1+2+4, 1.2+2+4+5, or 1.2+2+3+4. SARS-CoV IgG Antibody Test The SARS-CoV IgG antibody in serum specimens from recovering SARS patients was assayed by the BGI-GBI Biotech Company with an enzyme-linked immunosorbent assay (ELISA) kit (No. S20030003, BGI-GBI Biotech Company, Beijing, P.R. China). The wells containing polystyrene microplate strips were coated with two recombinant SARS-CoV antigens that are well-characterized. Recovering SARS patients' serum samples in the diluent buffer (1:10) were incubated in the coated wells for 30 min at 37°C and then the wells were washed 5 times with the washing buffer. The dilutedenzyme-labeled anti-human IgG (100 μl) was added to the wells and incubated for 20 min at 37°C. The wells were washed 5 times with the washing buffer. A tetramethyl-benzidine substrate was then added to each well. The presence of specific antibodies was indicated by a yellow color developing after substrate addition. The reaction was terminated by addition of hydrochloric acid. The intensity of the color was measured spectrophotometrically at 450 nm to quantify the amount of antibody in the specimen. The optical density (OD) measured was compared with a standard calibration curve constructed for each lot, yielding concentration values for the samples. The OD values of both the positive and negative controls were determined. The threshold value for IgG was 0.18 OD units, calculated as the mean + 2 standard deviation (SD) levels of the readings given by 1000 control blood donor sera samples. If the OD was above the threshold value, the sample was considered to be positive for SARS-CoV IgG [ 5 ]. Pulmonary Function Test Each recovering SARS patient underwent a standard pulmonary function test (SensorMedics 2200, Yorba Linda, U.S.A) for forced expiratory volume in 1 second (FEV 1 ), vital capacity (VC), forced vital capacity (FVC), total lung capacity (TLC), transfer coefficient for carbon monoxide (D L CO), and carbon monoxide diffusion constant (D L CO/V A ) measured by means of the single-breath test. The hemoglobin level was also measured to adjust the D L CO value. The results were compared with those of age- and sex-matched controls and expressed as a percentage of predicted values. Pulmonary function was regarded as abnormal when the D L CO was less than 80% of predicted values (pred). This was considered a diffusion deficit [ 6 ]. Chest Radiography and Evaluation Frontal chest X-ray radiographs (CXR) were obtained at the first follow-up visit for each recovering SARS patient. If abnormities were found in the CXR or if the D L CO was <80% pred despite a normal CXR, the patient was sent for HRCT scanning (GE Light Speed, GE, U.S.A. 1-mm section in thickness with a 10-mm gap, supine position, scanning during inspiration, 1 second per scan, 140 kv, 200 mA). All CXR and HRCT images were assessed by three radiologists via a viewing console. The three radiologists were aware of the patients' clinical diagnosis at the time of their review of the radiographs. The final conclusions were established by consensus. Each segment of the lung was reviewed for ground-glass opacification, interstitial thickening, bronchiectasis, and architectural distortion. Abnormalities were magnified by means of a zoom function and were examined for intralobular interstitial, interlobular septal, or peribronchovascular interstitial thickening. Attention was also paid to the presence or absence of nodules or masses, cavitation or calcification, and emphysema. The presence of parenchymal bands, irregular interfaces (bronchovascular, pleural, or mediastinal), thickened interstitium, and traction bronchiectasis were considered as evidence of fibrotic changes [ 7 ]. Magnetic Resonance Imaging (MRI) Examination All MRI examinations were done using a 1.5 T Signa CVi imager (GE Medical Systems, Milwaukee, WI, U.S.A.). For the patients who complained of chronic bone and joint pain, coronal T 1 -weighted (spin echo; time to repetition [TR], 440–500; time to echo [TE) 11–14] scans of the hips were done. If there were any abnormalities noted in the T 1 -weighted images, further T 1 -weighted sagittal images and coronal short tau inversion recovery (inversion time 145, TR 3500–5000, TE 80–120) or turbo-spin-echo T 2 -weighted images with fat suppression (TR 2500–3000, TE 80–120) were obtained. Images 3 mm thick were obtained for the coronal studies, and 4 mm thick images were obtained for the sagittal studies. Osteonecrosis was diagnosed by the presence of a band of low signal intensity in T 1 -weighted images [ 8 ]. Statistical Analysis All data were expressed as the ± SD unless otherwise indicated. Statistical analyses were done by one-way analysis of variance (ANOVA), Student-Newman-Keuls, and Chi-square test for multiple comparisons. We used the STATA™ 7.0 statistical analysis software for Windows (STATA Statistical Software, Inc., U.S.A.) for evaluating the results of our study. With each statistical test, the criterion for significance was a p value of less than 0.05. Results The interval from hospital discharge to the first follow-up visit was 45.0 ± 20.7 days (Range: 11–104 days). Of the 383 individuals participating in our study, 311 patients (81.2%) tested positive for SARS-CoV IgG and 72 (18.8%) tested negative. (All patientswere tested twice for SARS-CoV IgG.) Of these, 33 patients (13 male and 20 female, average age 35.7 ± 12.1) with positive SARS-CoV IgG and abnormal pulmonary diffusion received regular follow-up examinations each month, from June to December in 2003, and every two months, from January to June in 2004. Tables 1 and 2 show the dynamic changes of SARS-CoV IgG in patients with positive tests for SARS-CoV IgG within the year after discharge, indicating that the serum SARS-CoV IgG remained at high levels, although it decreased significantly over the course of time. Table 1 Dynamic changes of serum SARS-CoV IgG antibody levels in patients recovering from SARS Samples (n) ± SD (OD units) May, 2003 35 1.240 ± 0.350 June, 2003 74 1.087 ± 0.284 July, 2003 172 1.203 ± 0.306 Aug., 2003 152 1.061 ± 0.376 Sept., 2003 123 1.105 ± 0.378 Oct., 2003 35 1.097 ± 0.282 Nov., 2003 77 0.835 ± 0.327†‡§¶* Dec., 2003 35 0.829 ± 0.232†§* Jan.–Feb., 2004 67 0.737 ± 0.169†‡§¶# Mar–Apr, 2004 34 0.678 ± 0.179†‡§¶# May–June, 2004 46 0.621 ± 0.181†‡§¶# F value 30.62 p value 0.0000 Note: Statistical analyses were done by one-way analysis of variance (ANOVA) and Student-Newman-Keuls for multiple comparisons, and values are given as mean ± SD; † p < 0.05 vs. SARS-CoV IgG antibody results in May, 2003. ‡ p < 0.05 vs. SARS-CoV IgG antibody results in June, 2003. § p < 0.05 vs. SARS-CoV IgG antibody results in July, 2003. ¶ p < 0.05 vs. SARS-CoV IgG antibody results in August, 2003. p < 0.05 vs. SARS-CoV IgG antibody results in Sepember, 2003. # p < 0.05 vs. SARS-CoV IgG antibody results in October, 2003. Table 2 Dynamic changes of serum SARS-CoV IgG antibody levels in 33 regular follow-up examinations of patients recovering from SARS Samples (n) ± SD (OD units) June, 2003 33 1.104 ± 0.267 July, 2003 33 1.325 ± 0.357 Aug., 2003 33 1.092 ± 0.249 Sept., 2003 33 1.121 ± 0.432 Oct., 2003 33 1.056 ± 0.309 Nov., 2003 33 0.895 ± 0.203‡¶ Dec., 2003 33 0.800 ± 0.170†‡§¶ Jan.–Feb., 2004 33 0.726 ± 0.163†‡§¶* Mar–Apr, 2004 33 0.675 ± 0.181†‡§¶* May–June, 2004 33 0.610 ± 0.167†‡§¶# F value 25.69 p value 0.0000 Note: Statistical analyses were done by one-way analysis of variance (ANOVA) and Student-Newman-Keuls for multiple comparisons, and values are given as mean ± SD; † p < 0.05 vs. SARS-CoV IgG antibody results in June, 2003. ‡ p < 0.05 vs. SARS-CoV IgG antibody results in July, 2003. § p < 0.05 vs. SARS-CoV IgG antibody results in August, 2003. ¶ p < 0.05 vs. SARS-CoV IgG antibody results in September, 2003. p < 0.05 vs. SARS-CoV IgG antibody results in October, 2003. # p < 0.05 vs. SARS-CoV IgG antibody results in November, 2003. There were 88 individuals (23.0%) with abnormal D L CO among the 383 patients participating in our study. Of the 311 individuals testing positive for SARS-CoV IgG, there were 85 with abnormal D L CO (27.3%, 85/311), in contrast to just 3 cases with abnormal D L CO among the 72 subjects testing negative for SARS-CoV IgG (4.2%, 3/72). There was a statistically significant difference between positive and negative SARS-CoV IgG groups in D L CO values (table 3 ). Among the 85 patients (29 male and 56 female, average age 42.2 ± 11.9 years) with abnormal D L CO and positive SARS-CoV IgG, 40 individuals received pulmonary function tests 4 times within the year at 42.0 ± 10.4, 70.0 ± 11.8 and 155.1 ± 42.9 day intervals. Among these 40 patients, there were 23 who exhibited abnormal D L CO at their second pulmonary function examination, 23 at the third examination, and 20 at the fourth examination (table 4 ). . Pulmonary fibrosis was detected by CXR and confirmed by HRCT examination in 72 SARS patients in the convalescent phase. Among these, there were 4 individuals with negative and 67 with positive SARS-CoV IgG. Of the 40 patients who received HRCT examinations at least 4 times, all showed improvement in the fibrotic condition (Figure 1 ). Figure 1 The results of chest HRCT examination in a SARS patient in the convalescent phase, showing marked reversal of pulmonary fibrosis. Of the 311 convalescent SARS patients with sero-positive SARS-CoV IgG, 78 received femoral head MRI examinations. The Imaging showed that 18 of these patients (23.1%, 18/78) had avascular necrosis of the femoral head. Of these 18 individuals, 8 had avascular necrosis of both femoral heads and 10 had avascular necrosis of one femoral head. Ten of the 18 patients showed first stage changes and 8 had secondary changes. During the 3–6 month follow-up visits for these individuals, there were no obvious changes in the avascular necrosis for these patient. Discussion Since the outbreak of SARS at the end of 2002, despite the great efforts that have been extended, the mechanisms, clinical characteristics, prognosis and effective therapeutics for this disease have not been adequately clarified. Both the SARS virus itself and the anti-viral therapy (such as high-dose glucocorticoids) used in treatment can cause various degrees of toxicity and side effects, including pulmonary fibrosis and avascular necrosis of the femoral head, even in the convalescent phase. Follow-up surveys of SARS patients in the convalescent phase are needed for recognizing the clinical characteristics of this disease and reevaluation of the therapeutic treatments [ 2 , 7 ]. In our study, 72 individuals (18.8%) showed negative results in the SARS-CoV IgG antibody test for at least two tests, suggesting that there may have been misdiagnosis of some clinically diagnosed SARS patients. Comparison of the Chinese clinical diagnosis standard (published April, 2003) [ 3 ] to the Center for Disease Control (CDC) SARS case definition (published April 30, 2003) [ 9 ], indicates that both of them emphasize the importance of epidemiological history, clinical manifestations and chest radiological changes for the clinical diagnosis of SARS disease. The CDC SARS case definition especially emphasizes the importance of laboratory criteria for the confirmation of a SARS diagnosis. This is accomplished by detecting the dynamic changes in the titration of specific antibodies against SARS CoV and positive detection of SARS-CoV RNA by PCR. In contrast, the Chinese clinical diagnosis standard did not mention the importance of a laboratory SARS-CoV test for the confirmation of a SARS diagnosis. This might have resulted in the misdiagnosis of SARS in some cases. During follow-up examinations, we found that those individuals with positive SARS-CoV IgG remained positive for a year, although the level of the antibody decreased gradually. Therefore, those inoculated with a SARS vaccine or infected by the SARS virus might not receive lifetime immunity, but only immunity for a limited duration. Certainly, our findings must be confirmed by further studies [ 7 , 8 ]. By regular examination of pulmonary function and CXR, we found that those with pulmonary fibrotic changes were able to heal on their own. The fibrotic tissue was absorbed and pulmonary diffusion and VC improved with time, suggesting that the mechanism of lung injury and lung fibrosis caused by the SARS-CoV may have a different pathophysiological process compared to other lung diseases, such as idiopathic pulmonary fibrosis or pulmonary fibrosis secondary to adult respiratory stress syndrome. The reason is not clear. However, in our follow-up study, we found some ground-glass-like changes in the HRCT images from SARS patients one year after discharge. This result shows that changes in the lung can still be observed in convalescents [ 7 , 9 ]. Recent concern has focused on a complication of SARS in the convalescent phase, when avascular necrosis develops on the femoral head. The morbidity of this condition is reported to be 15% to 30% in some SARS patients in Mainland China [ 8 , 10 ]. Among the 78 patients receiving an MRI examination, there were 18 cases of complicated necrosis of the femoral head to different degrees. The causes of this complication include SARS itself and the drugs (such as glucocorticoids) used in treatment, with the latter being more important than the former [ 11 - 13 ]. We didn't find any worsening or improvement of the avascular necrosis of the femoral head in these patients during our follow-up examinations. Although most patients received magnetotherapy, hyperbaric oxygen chamber therapy, local kerotherapy and Chinese traditional medicine to promote local blood circulation, there was no apparent short-term therapeutic effectiveness of these methods for recovery of the femoral head. In conclusion, SARS, as a new disease, remains unfamiliar to mankind. It has high rates of morbidity and mortality in the acute phase. A significant proportion of patients surviving the acute illness have impairment in their overall functional capacity and health status in the convalescent phase after discharge from the hospital. Follow-up surveys of SARS patients in the convalescent phase are needed to understand the clinical characteristics of this disease. Our findings suggest that follow up studies of these patients are required for a longer duration, including comprehensive assessments for detection and appropriate management of any persistent or emerging sequelae. These types of investigation may facilitate the search for effective therapeutics and aid in ultimately conquering this disease. Table 3 D L CO results for the convalescent SARS patients with sero-positive or sero-negative SARS-CoV IgG Positive Negative Total X 2 P value D L CO normal cases 226 69 295 D L CO abnormal cases 85 3 88 Total 311 72 383 17.7269 0.000 Note: Analyzed with Chi-square test. Table 4 Pulmonary function test results from the 4 follow-up examinations of 40 convalescent SARS patients ( ± SD) Follow-up* VC(% pred) FEV 1 (% pred) D L CO(% pred) D L CO/V A (% pred) Two months 87 ± 15 (51~114) 83 ± 13 (60~108) 69 ± 9 (47~79) 95 ± 14 (58~123) Four months 94 ± 14 (61~123) 90 ± 13 (65~121) 76 ± 11 (48~94) 99 ± 14 (67~126) Six months 100 ± 15† (66~136) 93 ± 12† (66~114) 76 ± 11† (52~98) 97 ± 14 (62~129) Eleven months 103 ± 15† (66~142) 96 ± 11† (67~115) 79 ± 12† (56~98) 97 ± 14 (59~128) F value 9.23 7.84 6.15 0.63 P value 0.0000 0.0001 0.0006 0.5936 Note: *: Indicating as time after discharge from acute illness. Statistical analyses were done by one-way analysis of variance (ANOVA) and Student-Newman-Keuls for multiple comparisons, and values are given as mean ± SD; †: Compared to those in the first follow-up exam, p < 0.05.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545044.xml
545050	Decreased gene expression from T7 promoters may be due to impaired production of active T7 RNA polymerase	Background Protein expression vectors that utilize the bacteriophage T7 polymerase/promoter system are capable of very high levels of protein production. Frequently, however, expression from these vectors does not reliably achieve optimal levels of protein production. Strategies have been proposed previously that successfully maintain high expression levels, however we sought to determine the cause of induction failure. Results We demonstrated that decreases in protein overproduction levels are not due to significant plasmid loss nor to mutations arising on the plasmid, but instead largely are attributable to chromosomal mutations that diminish the level of functional T7 RNA polymerase, resulting in decreased expression from the plasmid. Isolation of plasmid DNA from non-expressing strains and reintroduction of the plasmid into a T7 RNA polymerase-producing strain such as BL21(λDE3) reproducibly restored high level protein production. Conclusions Our results suggest that a major contributing factor to decreased expression levels in T7 based systems is chromosomal mutation resulting in loss of functional T7 RNA polymerase. Consistent with this hypothesis, we found that optimal protein overproduction was obtained reproducibly from T7 promoters using freshly transformed cells that had not been subjected to outgrowth during which mutations could accumulate.	Background Often the first step in protein purification is production of the protein of interest in Escherichia coli using a gene expression system that allows selective overexpression of a cloned gene [ 1 ]. In the most desirable situation, expression of the cloned gene will consume the majority of the cellular resources, such that at the time of harvest the target protein makes up the bulk of the cell content. Complete purification of the protein then requires fewer selective extraction procedures and results in a more concentrated product. It is clear that optimizing the overexpression step is an effective method for maximizing and facilitating protein purification. One commonly used system for achieving high levels of target protein production depends on the extremely selective nature of the bacteriophage T7 RNA polymerase for specific promoters [ 2 - 4 ]. In the pET system, the gene encoding T7 RNA polymerase is usually supplied by the host bacterial cell in the form of a λ lysogen which expresses the polymerase gene under control of the lacUV5 promoter [ 2 , 3 ]. The target protein is encoded on a plasmid in which the gene has been cloned so that transcription will be driven by the T7 RNA polymerase. Thus, expression of the T7 RNA polymerase gene is controlled by addition of IPTG to the growth media, and in turn, production of the polymerase controls expression of the plasmid-borne target gene. The T7 RNA polymerase system is extremely effective, resulting in very high levels of target protein production, and the controllable promoters allow even the cloning and expression of genes encoding highly toxic proteins in E. coli [ 2 , 3 ]. However, anecdotal evidence has suggested that many investigators observe less than optimal expression from this system and even the lack of detectable protein overproduction. In our hands, the level of synthesis of proteins varied, not only with the specific protein encoded on the plasmid but also from sample to sample of the same protein. Although poor protein synthesis, or complete lack of protein production, appears to be a common problem, strategies to cope with it are varied. Some protocols suggest "pre-induction" tests to determine specific colonies that will result in overproduction [ 5 ]. We found that this test, although simple, is tedious and can fail to identify expressor colonies. However, there are a number of effective strategies that have been proposed over the years [ 2 , 3 , 6 - 8 ]. While we support these recommendations, in this work we sought to determine the underlying cause of reduced expression from pET vectors in order to more effectively prevent problems with poor protein production levels. Results We initially examined the conditions required for optimal target protein production from pET plasmids because we observed inconsistent results upon induction of plasmid-borne genes. This inconsistency was most apparent with the plasmid that encoded SecB, pJW25. Following the protocol provided [ 5 ], we streaked a frozen glycerol stock of BL21(λDE3) containing pJW25 for single colonies on LB plates with ampicillin, picked 4 colonies and grew small cultures to analyze overproduction of SecB. Only 1 of the 4 cultures demonstrated detectable overproduction of the SecB protein, and the level of overproduction from that one was not remarkable. Random selection of a large number of additional colonies (greater than 100 total) resulted in only 25–30% that synthesized levels of SecB that were clearly greater than the uninduced control. Further, if a culture that was determined initially to overproduce SecB was inoculated into a larger volume (500 ml), overproduction of SecB in the larger culture was not always detected. As we intended to grow strains for overproduction in a 5 liter fermentor, it was critical that the culture used reproducibly synthesize high levels of target protein. While overproduction of SecB was unreliable from frozen stocks, we found that BL21(λDE3) that had been newly transformed with pJW25 overproduced SecB to a much higher level and that every colony examined synthesized SecB to a similar extent. Continued subculturing, however, resulted in a decrease or loss of overproduction (Figure 1A ). These results suggested that with time the bacteria were losing the ability to overproduce SecB. To verify that our observations were not unique to SecB, we examined a variety of other plasmids for overproduction of E. coli proteins in BL21(λDE3); the plasmid pJGGV encodes the secretory protein, proOmpA, pT7SecA produces SecA, the translocation ATPase [ 9 ], and pET11Tus encodes the Tus DNA binding protein [ 10 ]. Similar results were obtained with each of these plasmids. In all cases, newly transformed BL21(λDE3) gave rise to high levels of target protein production, while repeated subculturing decreased the level of expression, eventually to undetectable levels. The initial level of synthesis varied for each protein as did the severity of the effect of subculturing on overproduction. In all cases, however, the degree of overproduction was optimal immediately after transformation. Results are shown for data obtained with the pJW25 (SecB) and pJGGV (proOmpA) plasmids (Figure 1 ). To more accurately determine the effect of subculturing on target protein production, protein levels were monitored as the strains were allowed to grow. Newly transformed BL21(λDE3) were grown as overnight cultures, then repeatedly subcultured as described in Methods. After each subculturing, when cultures reached early log phase (A 600 = 0.4), a sample was removed and tested for the ability to overproduce plasmid-encoded protein. The amount of SecB synthesized by pJW25 decreased gradually with each subculturing, so that by the eighth subculture SecB protein production was minimal (Figure 1A ). At this point, four of eight cultures no longer made detectable levels of SecB; the other four were greatly reduced in secB expression (data not shown). The synthesis of proOmpA was even more dramatically affected by subculturing. Induction was abolished very quickly as there was no detectable overproduction in 15 out of 16 cultures after 3 subculturings (Figure 1B ). SecA retained reasonable, but low, protein production levels through 11 subculturings, while Tus synthesis was nearly abolished after only 3 subculturings (data not shown). In all cases, cultures that continued to overproduce protein did so at greatly reduced levels. These results indicated that some proteins are more deleterious to the cell and that continued growth, even without induction, results in alterations to the strain such that overproduction is no longer possible. We considered the following possible explanations for decreased synthesis of target protein with time: 1) the plasmid was not stably maintained in all bacteria, even with the constant presence of selective agent, 2) the plasmid was accumulating mutations that decreased expression levels, or 3) the bacterial strain accumulated mutations that inhibited production of functional T7 RNA polymerase, resulting in decreased or no expression from the plasmid. We examined each of these possibilities. Previous protocols highly recommend that strains be examined for loss of plasmid prior to induction, and suggest that this is the major contributor to decreased expression levels [ 2 , 3 ]. However, our strains were grown with constant selective pressure for plasmid maintainence and subculturings were always made at high dilutions (1:1000) to minimize transfer of any β-lactamase from lysed cells to the fresh culture. Therefore, we felt that, while plasmid loss may be occurring, it probably would not be sufficient to account for the dramatic decreases observed in protein levels. To address this point, we prepared plasmid DNA from strains that had been newly transformed and overproduced proteins at high levels and also from the same strains that had been subcultured repeatedly and no longer synthesized high levels of target protein. We found that the amount of DNA present as observed by agarose gel electrophoresis was quite similar in both strains, regardless of the plasmid examined (data not shown). While the results we obtained are not quantitative, they indicated that plasmid loss alone could not account for the reduced level of expression. To address possible plasmid loss more directly, we compared newly transformed strains and ones that had been subcultured repeatedly for their ability to form individual colonies on LB agar or LB agar containing ampicillin. As β-lactamase is a periplasmic enzyme, it is possible that sufficient leakage of the enzyme occurred to inactivate the ampicillin in liquid culture and allow growth of bacteria that no longer retained plasmid DNA despite the high dilution (1:1000) upon subculturing. If that were the case, we would expect the strain that had been subcultured many times to contain fewer plasmid-containing cells and therefore to form fewer colonies on LB+ampicillin. If no significant plasmid loss were occurring, we would expect the number of colonies to be equal whether or not ampicillin was present. We found that the pJW25-containing strain formed equivalent numbers of colonies on LB and on LB-ampicillin, even after subculturing eight times to result in a culture that did not overproduce SecB (data not shown). Thus, plasmid loss is not sufficient to account for the observed decrease in protein overproduction in these strains. The second hypothesis was that the plasmids had suffered mutations that rendered them unable to overexpress target genes, for example mutations to the T7 RNA polymerase binding site or to the Shine Dalgarno region. To observe such a great decrease in protein production, a mutated plasmid must confer a growth advantage that results in cells containing the mutated plasmid to overtake the culture. Therefore, strains that no longer overproduce significant amounts of target protein should contain mutated plasmid. To test this possibility, we isolated plasmid DNA from strains that no longer synthesized significant amounts of target protein and retransformed BL21(λDE3) with the plasmid DNAs. If plasmid mutations were occurring, we would expect a large number of the transformants to be poor expressors. On the other hand, if plasmid mutations were not responsible for decreased expression, then we would predict that the new transformants would produce high levels of protein. Indeed, this was the result we observed. Every colony examined, from every plasmid tested, synthesized target protein efficiently after retransformation. Results with pJW25 and pJGGV are shown in Figure 1 . The levels of protein production in newly transformed cells were similar to that of the original strain. This result indicated that plasmid mutation was not a significant cause of decreased target protein production from these plasmids. The third hypothesis was that mutation of the bacterial chromosome might lead to decreased production of active T7 RNA polymerase, in turn decreasing expression from the plasmid. We felt this explanation was probable as a single chromosomal mutation that diminished T7 RNA polymerase expression could be sufficient to abolish protein production from all resident plasmids and this mutant could quite likely overtake the culture in a fairly short period of time, even in the presence of antibiotic to maintain the plasmid. We examined this possibility by titering a mutant T7 phage, Δ4107, that requires host production of T7 RNA polymerase for growth [ 2 ]. T7 Δ4107 produced very large, clear plaques when grown on BL21(λDE3) (Figure 2A ). When the host strain was BL21(λDE3) that had been newly transformed with pJW25, the titer decreased to about 25% the original, and the plaques produced were slightly smaller (Figure 2B ). Furthermore, resistant colonies could be seen emerging within the plaques. After BL21(λDE3) containing pJW25 had been subcultured 10 times so that SecB was no longer detectably overproduced, the phage was again titered. The titer was about 25% of the original, but the plaques produced on this strain were very small and cloudy (Figure 2C ). These results are consistent with loss of functional T7 RNA polymerase. We would not predict complete loss of plaque forming ability as the strain culture probably consists of a mixed population at the time of plating; that is, while most cells in the population would be derived from a mutant in which there is little or no functional T7 RNA polymerase, there may be some cells that have not suffered a mutation and would therefore support growth of the phage. We conclude therefore, that a major cause of decreased production of proteins from pET plasmids may be a decreased level of functional T7 RNA polymerase due to mutations in the BL21(λDE3) chromosome. The loss of T7 RNA polymerase activity could be due either to mutation of the polymerase gene or excision of the λ prophage. To distinguish between these possibilities, we performed PCR using primers that amplify an 1100 base pair fragment of the T7 RNA polymerase gene. As template DNA, we used either BL21(λDE3), BL21(λDE3) freshly transformed with pJW25, or BL21(λDE3) containing pJW25 that had been subcultured eight times and no longer produced significant amounts of SecB. In all cases, a DNA product of the correct size was observed, indicating that prophage excision was not the major cause for decreased T7 RNA polymerase activity (data not shown). Therefore, our results suggest that mutation to the prophage, resulting in decreased levels of functional T7 RNA polymerase, was the predominant contributory factor for decreased production of target proteins. Discussion Optimal overproduction is an important first step towards high quantity purification of target proteins. It is critical that one be certain that the culture used will synthesize large quantities of the protein of interest, particularly if it will be used for large scale production procedures, such as growth in a fermentor. When used correctly, the T7 RNA polymerase/promoter system is an excellent choice for such overproduction. However, care must be taken to achieve optimal protein production levels. It had been suggested previously that plasmid loss is the primary cause for decreased expression from target genes in the pET system [ 2 , 3 ]. Our results differed. While plasmid loss may have occurred to a small extent, in the cases analyzed here the principal reason that lowered levels of protein production occurred was that mutations within the lysogen on the host bacterial chromosome resulted in decreased levels of functional T7 RNA polymerase. The fact that strains which no longer produced plasmid-encoded proteins also no longer supported vigorous growth of the mutant T7 phage, Δ4107, indicates that T7 RNA polymerase function was absent or greatly reduced in the majority of the population. Further, PCR amplification of the T7 RNA polymerase gene suggested that the reduction in polymerase function was due to mutation rather than to prophage excision. Surprisingly, there was no clear growth disadvantage to strains carrying pJW25 as assessed by monitoring growth over time (data not shown). We do not think that this finding negates our proposal that chromosomal mutation occurs, however. Rather, this may explain why we see loss of expression more rapidly with some plasmids than with others; for example, synthesis of proOmpA was drastically decreased after only three subculturings. Apparently, even in the uninduced state, sufficient expression of target protein occurred, so that a detrimental effect resulted. If that target protein is sufficiently detrimental, selection for polymerase mutations will occur. For this reason, the steps that are taken to ensure optimal expression from target genes must be directed at limiting opportunities for mutation to the T7 RNA polymerase and for those mutants to overtake the bacterial culture. Two methods have been suggested previously to ensure high levels of protein production; either screening individual colonies for protein synthesis [ 5 ] or testing the population of bacteria to assess the fraction of cells capable of overproduction [ 2 , 3 ]. The first is tedious and unreliable, while the second may be better used as an indicator of the degree of mutation that has occurred. However, as we found in this study, these precautions alone are not sufficient to ensure that the colony used for expression will be the highest protein producer possible. In particular, if any further growth of the strain occurs between the time of testing and the actual induction, mutations may occur that decrease expression levels. Our findings support previous recommendations for growth and storage of BL21(λDE3) containing pET plasmids [ 2 , 3 , 6 , 7 ] as the conditions described would limit mutation of the T7 RNA polymerase gene. Specifically, plasmid-containing BL21 (λDE3) grown to saturation in rich media will allow basal expression of the target gene [ 3 ]. Depending on the toxicity of the protein thus produced, these conditions will select for random mutations to the chromosomally encoded T7 RNA polymerase. Our studies also support the use of bacterial strains that contain the T7 lysozyme as the lysozyme is an inhibitor of the T7 RNA polymerase and will thus decrease uninduced expression even further [ 3 ]. We found that the quickest and most reliable method for obtaining optimal protein production was to freshly transform BL21(λDE3) with the desired plasmid, and use a colony directly from the transformation plate for expression studies with only a single subculture of an overnight culture. Should decreased levels of target protein production be observed, the problem can be simply remedied by isolating plasmid DNA from poor expressors and retransforming BL21(λDE3). However, we acknowledge that this approach may not be realistic for very large scale production facilities. Nevertheless, understanding the molecular basis for decreased protein production will facilitate development of techniques that will minimize conditions that may lead to selection and outgrowth of mutant strains. Conclusion A common difficulty with overproduction of proteins using the T7 RNA polymerase based system is decreased target protein synthesis or even lack of detectable protein production. It is clear that synthesis of proteins, even at low basal levels, leads to the loss of induction capability. Effective strategies have been presented previously to avoid loss of expression, all of which are based on preventing basal expression levels. In this report, we demonstrate that a significant underlying mechanism leading to loss of expression is a decrease in functional T7 RNA polymerase, and selection of mutants unable to express the recombinant gene. Materials and methods Bacterial strains and plasmids Escherichia coli B strain BL21(λDE3) was used for overproduction of proteins from plasmids containing T7 promoters and was obtained from Stratagene. All plasmids are derivatives of pET11 (Stratagene). Plasmids encoding SecB (pJW25) [ 5 ], SecA (pT7SecA) [ 9 ], and Tus (pET11Tus) [ 10 ] were generous gifts from Linda Randall, Bill Wickner, and Thomas Hill, respectively. The plasmid encoding proOmpA (pJGGV) was constructed in this laboratory by PCR amplification of the ompA gene from E. coli K12 strain MC4100 [ 11 ] using primers ompA-1 (gacctacccgggcat atgaaaaagacagctatcgc ) and ompA-2 (ggtcatcccgggtgatca ttaagcctgcggctgagttac , underlining indicates regions of homology to the ompA gene). The PCR product was digested with restriction enzymes NdeI and BclI and cloned into pET11c that had been digested with NdeI and BamHI. Standard protocols were used for PCR, restriction digests, ligations, and transformations [ 12 , 13 ]. Plasmid DNA was recovered from strains using a QiaPrep Spin Miniprep kit (Qiagen) following manufacturer's instructions. Induction and analysis of protein production All strains were grown in LB medium [ 14 ]. When plasmid was present, ampicillin was added to a concentration of 100 μg/ml. Cultures were induced for protein production at an A 600 of 0.4 by addition of IPTG to a final concentration of 1 mM. Growth was allowed to continue for 2 hours after addition of IPTG. Uninduced controls were grown the same except no IPTG was added. Cells were lysed by boiling in SDS [ 14 ], and proteins were analyzed by SDS polyacrylamide gel electrophoresis [ 13 ]. For experiments using newly transformed BL21(λDE3), colonies were picked directly from the transformation plate and inoculated into 5 ml LB containing ampicillin for overnight growth. The overnight culture was diluted 1:1000 into fresh LB with ampicillin and grown to an A 600 of 0.4 for induction. Thus, the bacteria were subcultured only once. For continuous subculturing experiments, samples were removed before addition of IPTG and used to inoculate fresh LB plus ampicillin media at a dilution of 1:1000. T7 phage titering Bacteriophage T7 mutant Δ4107 [ 2 ], which was a generous gift from Dr. William Studier, was grown for single plaques on BL21(λDE3) containing various plasmids using standard phage protocols [ 14 ]. PCR amplification of T7 RNA polymerase Genomic DNA was isolated using the Qiagen DNeasy kit. Oligonucleotides used as primers (t7pol1 – gattaacatcgctaagaacg and t7pol2 – gattcatgtcgatgtcttcc) were obtained from Midland Certified Reagents, Midland, TX. PCR was performed using the FailSafe PCR kit (Epicentre Technologies, Madison, WI) following manufacturer's recommendations. PCR products were visualized by agarose gel electrophoresis. Abbreviations used IPTG – isopropyl-β-D-galactoside PCR – polymerase chain reaction LB – Luria-Bertani SDS – sodium dodecyl sulfate Authors' contributions JGGV was responsible for the original observations of inconsistent expression levels and the data for Figure 1 , as well as the agarose gel analysis not shown. AMF performed the experiments for Figure 2 and prepared the manuscript. Both authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545050.xml
533878	A survey of individual preference for colorectal cancer screening technique	Background Due to the low participation in colorectal cancer screening, public preference for colorectal cancer screening modality was determined. Methods A cross-sectional survey was performed of healthy ambulatory adults in a pediatrics primary care office and neighboring church. Overall preference was ranked for each of four colorectal cancer screening modalities: Faecal Occult Blood, Fiberoptic Sigmoidoscopy, Barium Enema and Colonoscopy. Four additional domains of preference also were ranked: suspected discomfort, embarrassment, inconvenience and danger of each exam. Results 80 surveys were analyzed, 57 of which were received from participants who had experienced none of the screening tests. Fecal Occult Blood Testing is significantly preferred over each other screening modality in overall preference and every domain of preference, among all subjects and those who had experienced none of the tests. Conclusions Efforts to increase public participation in colorectal cancer screening may be more effective if undertaken in the context of public perceptions of screening choices.	Background Screening for colorectal cancer lessens the risk of dying from that disease [ 1 ]. Knowledge of this fact has not solved all the problems related to screening. The optimal modality of screening is still the subject of debate [ 1 - 3 ]. More problematic is the very low participation of the general public in recommended screening [ 4 ]. In contrast to breast cancer screening, in which the Healthy People 2000 Goal of the U.S. National Institutes of Health was surpassed, at 64% participation by women over 40 years of age, only 20% of Americans over age 50 had fecal occult blood testing within the past year (This is the best estimate of actual screening, rather than diagnostic endeavors for symptoms for which endoscopy or radiologic imaging might be done.), and 34% had a sigmoidoscopy within the past 5 years [ 5 , 6 ] Even if screening is appropriately performed, it is far from certain that a positive screen will be followed by appropriate diagnostic testing, as has been shown in follow-up surveys of fecal occult blood testing [ 7 ]. Most publication concerning colorectal cancer screening relates to the choice of screening modality; discussing accuracy, efficacy and cost, since the most inexpensive technique, faecal occult blood testing, is inaccurate in the detection of colorectal neoplasia, though effective in significantly diminishing disease specific mortality [ 8 ], and the most accurate technique, colonoscopy, is expensive and not without danger [ 3 ]. The choice is not an easy one for clinicians, much less patients or the asymptomatic public. Therein may lie one of the problems with public participation in screening. Unlike cancers of the breast, cervix, prostate or lung, where a single screening modality dominates current recommendations for each, there are four different and relatively independent screening tests for colorectal cancer that are currently recommended by the American Cancer Society, National Cancer Institute and United States Preventative Services Task Force: faecal occult blood test (FOBT), fiberoptic sigmoidoscopy (FS), barium enema (BE), and colonoscopy (C) [ 1 ]. The absence of a single recommendation may lead from indecision to inaction on the part of clinicians or patients. However the greatest problem related to screening remains the low level of participation by those for whom it is intended: asymptomatic individuals over the age of 50 years with no specific risk factors for colorectal cancer, i.e., no past history of colorectal polyps, cancer, rectal bleeding, colitis, change in bowel habits, iron deficiency anemia, weight loss or a close family member with colorectal cancer. We agree with Dr. Woolf [ 2 ], that strategies to improve public compliance with recommended colorectal cancer screening might be more effective if they include an awareness of what the public thinks about the tests being recommended. Previous studies have not surveyed asymptomatic participants' preference over the whole range of screening choices, focusing instead on symptomatic patients undergoing diagnostic evaluation such as colonoscopy and barium enema [ 9 - 11 ] or patients ailing from extracolonic diseases whose motivation for screening might be very different than the healthy population for whom screening is intended [ 12 - 18 ]. Among these latter studies there has been a general preference noted for FOBT (table 1 ). We have in this report chosen to focus our survey differently and uniquely; first to inform healthy, ambulatory and younger people, and not ailing patients, concerning only the preparation and conduct of each screening test. Secondly, in order to determine how their perceptions of the conduct of each test might affect their participation, participants were then asked to rank not just their overall preference based upon the preparation and conduct of the tests alone, but four other domains of preference for each screening modality: perceived physical discomfort, inconvenience, embarrassment and danger. Test accuracy was not included in the preamble on test performance, first, because we wanted to isolate perceptions of the physical conduct of the screening test, and second, because test accuracy has been part of many of the previous surveys, often presented with considerable bias. Randomized trials of decision aids have also shown that description of a test's ability to detect colorectal cancer has not been successful in increasing participation in screening [ 15 - 17 ]. Lastly, despite the current enthusiasm for screening colonoscopy by organizations that do colonoscopy as the complete screening test [ 19 ], as mentioned above, the choice of screening modality is still regarded as controversial. Methods Participants were a convenience sample of parents or grandparents of children visiting a general pediatrics office (usually for well child visits or minor ailments), personnel working in that office, or parishioners attending a church social gathering, all aged 18 and over. An introductory letter described the purpose of the survey. This was followed by a brief description of the preparation and performance of each commonly used screening test for colorectal neoplasia: faecal occult blood testing (FOBT), fiberoptic sigmoidoscopy (FS), air contrast barium enema (BE) and total colonoscopy (C). The relative accuracy of each exam was not discussed. Six questions followed. The first asked the participant to rank each test in order of overall preference. The second asked the participant to rank each test according to how much that test might cause physical discomfort, the third, inconvenience; the fourth embarrassment, the fifth, the relative danger of the exam. The sixth question asked participants which of the four tests they had previously experienced, along with their gender and age. No further symptom or medical history was obtained and surveys were only numbered consecutively with no personal identifiers. (see appendix for letter and survey) Based upon a related survey concerning subject preference for tests of colonic inflammation [ 20 ], a sample size of 50 individuals was estimated. Eighty four questionnaires were distributed in order to assure receipt of an adequate number of usable responses from individuals who had experienced none of the screening tests. The questionnaire is shown in the Appendix. Analyses Data were analyzed using SPSS 11.0. Analyses focused on comparisons between ranks assigned each test on preference and the other assessed attributes, and included Friedman's test for ranks (to test the hypothesis that ranks differed for different tests) and the Wilcoxon signed-ranks test (to test the hypothesis that pairs of tests were differently ranked.) We also considered whether those rank orders might differ between participants who have and have not received any of these tests, and how gender and age affected preferences. Results 80 of 84 surveys were available for analysis; twenty nine from men and 51 from women. The mean age of the participants was 38.3 years (range 18 – 54 years; St. Dev. 8.19 years; median 40 years). Eight subjects had previously had a colonoscopy, five a barium enema, seven a sigmoidoscopy and 17 had stool collected for various reasons. Fifty seven subjects had experienced none of the screening tests. The mean rankings for preference among the entire sample are presented in Table 2 and among only those individuals who had experienced none of the tests are presented in the Table 3 , score "1" being the most preferred and "4" the least. In each case, mean rankings were found to vary by test (Friedman's test, 3 df), and FOBT was significantly preferred over the second-ranked test (FS) by Wilcoxon signed-ranks test. Median scores were determined for each domain for both the whole survey group and the naive subgroup. For each domain and in each group the results were the same, with ranks of 4,3,2 &1 for C, BE, FS and FOBT respectively, 1 being most preferred, except for embarrassment in both groups in which C and BE each had a median rank 3. The results hold up for each gender subgroup in all cases except that men didn't consider FOBT significantly less inconvenient than FS. Age was not significantly correlated with ranking of FOBT (that is, the ranking given didn't change with age) by Spearman's rho. Rho values ranged from -0.10 to +0.16, none significant. Splitting the groups into ages 18–39 (n = 39) and 40+ (n = 40), the results are the same for both groups except that for those over 40, preference for FOBT vs. FS and inconvenience of FOBT vs. FS did not reach significance by two-tailed test (p = .079 and p = .057 respectively) Discussion A recent review of colorectal cancer screening stated that, "At present there is no preferred CRC screening strategy"[ 1 ]. This presents the perspective of a group of impartial physicians. However from the perspective of those who should take part in CRC screening in the future, a clear preference for FOBT over each other screening modality is expressed in this survey. Each domain of preference similarly ranks FOBT as significantly most preferred. Among previous surveys there are four randomized controlled trials of the use of decision aids that were designed with the intent of altering participation in screening. Three of these presented choices of screening modality or scenario to both intervention and control groups [ 15 - 17 ]. These studies therefore provided information of participant preference for specific screening modality, though again the participants, primary care patients , were quite different from the group reported herein. Only one of those reports offered all four of the screening modalities that we did in our study [ 16 ], the other two offering only a choice between FOBT and FS [ 15 , 16 ]. Nevertheless a uniform preference for FOBT was reported in these studies as well (Table 1 ). None of the test interventions were particularly effective in increasing participation in screening, an endpoint not assessed in our study. The fourth randomized trial randomized non-patients, relatives of gastroenterology patients, to be offered either FS or C and measured differential participation, which was equal in the two groups [ 20 ]. In the survey most similar to the present study, Pignone surveyed 146 patients in a general medicine clinic [ 18 ] and questioned participants after four sequential levels of information were given. Only two screening options were presented, FOBT & FS. Information included in sequence 1) the risk of colorectal cancer, 2) description of the conduct of the test, 3)accuracy of the tests, 4) cost. Previous screening participation was queried but not an exclusion. Less than 5% of those approached refused participation and no data were presented on the screening naïve participants in his sample. FOBT was preferred at each level of investigation, though both tests together were preferred after level 2 (Table 1 ). Participants were also asked for reasons for their preferences. The reasons most often given related to cost, ease of performance and being done alone. Among some physicians there is a growing popularity for the use of definitive diagnostic testing as a screening tool, that is, colonoscopy [ 19 ]. Though expensive and not without danger, reimbursement for the test is declining and the procedure is getting safer. It has obvious theoretical advantages of offering precise diagnostic capabilities, through biopsy, for those with positive screens. Most important, colonoscopy has the best potential for cancer prevention by adenoma removal – which is not possible with any other test [ 22 , 23 ]. This, properly applied, might even result in cost savings in the global cost of caring for colorectal cancer. But the public has to want to participate in this program and there is little evidence in this current survey and previous studies, especially those done in primary care settings [ 13 - 18 ], that this is likely. The concerns expressed herein about safety, embarrassment, inconvenience and discomfort all must be addressed in future efforts to increase screening participation. A potentially significant development related to these issues is that the principal disadvantage of FOBT, its inaccuracy in detecting colorectal neoplasia, might be overcome. Recently developed stool tests show an ability to diagnose cancer with much greater reliability [ 24 ]. Perhaps these gene based stool tests may establish the potential for adenoma discovery by non-invasive testing as well. Competing interests The author(s) declare that they have no competing interests. Authors' contributions RLN conceived of the study, designed the questionnaire and supervised its administration. AS organized the domains of preference and performed the statistical analyses. Both authors participated in the writing of the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC533878.xml
524479	RNA integrity as a quality indicator during the first steps of RNP purifications : A comparison of yeast lysis methods	Background The completion of several genome-sequencing projects has increased our need to assign functions to newly identified genes. The presence of a specific protein domain has been used as the determinant for suggesting a function for these new genes. In the case of proteins that are predicted to interact with mRNA, most RNAs bound by these proteins are still unknown. In yeast, several protocols for the identification of protein-protein interactions in high-throughput analyses have been developed during the last years leading to an increased understanding of cellular proteomics. If any of these protocols or similar approaches shall be used for the identification of mRNA-protein complexes, the integrity of mRNA is a critical factor. Results We compared the effect of different lysis protocols on RNA integrity. We report dramatic differences in RNA stability depending on the method used for yeast cell lysis. Glass bead milling and French Press lead to degraded mRNAs even in the presence of RNase inhibitors. Thus, they are not suitable to purify intact mRNP complexes or to identify specific mRNAs bound to proteins. Conclusion We suggest a novel protocol, grinding deep-frozen cells, for the preparation of protein extracts that contain intact RNAs, as lysis method for the purification of mRNA-protein complexes from yeast cells.	Background Genome analyses of a range of organisms have lead to the identification of an increasing number of putative RNA-binding proteins (RBPs) whose function is still unknown. RBPs have been found to act as integral part of ribonucleoparticles (RNPs) controlling gene expression at different levels [ 1 ]. RNPs are involved in controlling RNA export, RNA stability, RNA subcellular localization and mRNA translation [ 2 ]. It has been proposed that in this context RBPs could act as central coordinators in regulating the expression and fate of specific subsets of RNAs. This model is reminiscent of bacterial operons, where the expression of genes that act in the same pathway is regulated as one unit [ 3 ]. Recently, research has mainly been focused on identifying protein-protein interactions using two-hybrid interactions [ 4 , 5 ], immunopurification [ 6 ] or affinity purification [ 7 ]. So far, only few examples have been reported that were aimed at the identification of mRNA-protein interactions. In yeast, immunopurification has, for example, been used to enrich RNP complexes leading to the identification of 22 mRNAs localized to the bud tip [ 8 ], to the identification of Lhp1p associated mRNAs [ 9 ] and to the identification of mRNA export factor associated transcripts [ 10 ]. There are many examples for affinity purification methods in yeast, but perhaps the one that has been used most extensively is the Tandem Affinity Purification (TAP). TAP consists of two serial affinity purification steps of a protein tagged with a double epitope tag, without affecting the expression level of the protein. It was first described for identifying new protein components of the yeast U1 snRNP [ 11 ] and later used to identify protein-protein interactions in yeast [ 6 ], bacteria [ 12 ], Trypanosoma brucei [ 13 ], Drosophila [ 14 ] and mammals [ 15 ]. It has also been used to describe the set of mRNAs associated with the Puf family of RNA-binding proteins in yeast [ 16 ]. Besides the purification method, the way to lyse cells is also crucial. In yeast, different lysis methods are in use. Glass bead milling has been applied to identify RNAs from immunoprecipitated RNPs [ 9 ]. Both French Press and glass bead milling have been successfully used to characterize protein-protein [ 6 , 11 ] and protein-RNAs interactions [ 16 ]. However, the integrity of the mRNA has not been determined under the conditions used. Here, we show that existing lysis methods lead to extensive mRNA degradation even in the presence of RNase inhibitors. We also present evidence that a third method, grinding deep-frozen cells at ultra-low temperature, can be used to obtain intact mRNAs. Results Glass bead mill lysis leads to degraded RNAs Breaking yeast with a glass bead mill is a common method to produce cell lysates. The principle is based on the physical rupture of the yeast's cell wall and cell due to the friction produced by glass beads rapidly moving through the cell suspension. One of the advantages of this method is the high lysis efficiency. We lysed two different yeast strains using a "bead-beater" bead mill in the presence of RNase inhibitors (100 U/ml SuperaseIn and 20 mM Ribonucleoside Vanadyl Complex, RVC) as described in Methods. We used a strain where Nrp1p, a putative RNA-binding protein that contains one RRM (RNA Recognition Motif) [ 17 ], has been tagged and a wild type strain. As shown in Figure 1A we could enrich the bait protein, Nrp1p, in the TEV eluate as compared to a purification from a control wild type strain performed in parallel. We then analyzed the RNA extracted from the input material at the IgG immunopurification step from both strains by agarose gel electrophoresis. The absence of 25S and 18S rRNAs in the extracts as compared to total RNA extracted by phenol [ 18 ] indicates that RNA was degraded (Figure 1B ). Figure 1 TAP Purification of RNA-binding protein complexes leads to RNA degradation when cells are lysed in a glass bead mill. A , 2 μl (corresponding to ~25 μg protein) of the input material for the IgG immunopurification (input) and the TCA-precipitated material from 200 μl of the TEV eluate from RJY358 (wt, untagged strain) and RJY929 (Nrp1p-TAP) were separated on a precast 4–12% gradient SDS polyacrylamide gel (Invitrogen) and stained with Coomassie. The molecular weight of the protein markers is indicated on the left. The band corresponding to the bait protein (Nrp1p-TAP) is labelled with an asterisk. B , 1 ml (corresponding to ~14 mg protein) of the input material for the IgG immunopurification (input) was phenol:chloroform extracted and 12 μg of the extracted RNA were separated on a 1.2% agarose-formaldehyde gel in the presence of ethidium bromide. As control for intact RNAs, 2 μg of total RNA (P/C lysis RNA) from the same strains prepared with a phenol method [18] were loaded in parallel as control. The position of the 18S and 25S ribosomal RNAs is indicated. C , crude lysate from strains RJY358 (wt, untagged) and RJY933 (Pbp2p-TAP) from two independent experiments were phenol extracted and 8 μg of the RNA were separated on a 1.2% agarose-formaldehyde gel and blotted onto a nylon membrane. 8 μg of total RNA (P/C lysis RNA) from the same strains prepared with a phenol extraction method [18] were loaded in parallel as control for intact RNAs. After methylene blue staining, the membrane was hybridized with a probe against PDA1 mRNA. The positions of the 18S and 25S ribosomal RNAs and the PDA1 mRNA hybridization signal are indicated. The chosen purification method involves several steps: lysis, extract preparation and IgG immunopurification. To rule out in which step RNA degradation takes place, we analysed mRNA integrity in samples taken at different purification steps (Figure 1C and data not shown). Therefore, we used a strain where Pbp2p, a putative RNA-binding protein that contains two KH-type 1 domains (hnRNP K homology domain) [ 19 ], has been tagged and a wild type strain. We found that extensive degradation already occurs during cell lysis (Figure 1C ) as shown by Northern blot against a specific abundant mRNA ( PDA1 ) or by direct staining of ribosomal RNAs after blotting with methylene blue (unbalanced ratio of 25S and 18S ribosomal RNAs in Figure 1C ). Lysis by French Press leads to degraded RNAs A second major lysis method for yeast cells is the French Press. Lysis occurs when the cell suspension is pressed through a small capillary. The pressure difference between the chamber and the capillary ruptures the cell. We lysed cells as described [ 11 ] in the presence of RNase inhibitors (100 U/ml SuperaseIn and 20 mM RVC), and spun the crude lysate at 1200, 20000 and 200000 × g as indicated in Methods. We then analyzed the RNA extracted from the different fractions by Northern blot. As shown in Figure 2 , when analyzing RNA quality by methylen blue staining, RNA degradation is not as obvious as with bead mill lysis (compare the ratio between 25S and 18S ribosomal RNAs in Figure 2 with that in Figure 1C ). However, all different strains tested (wild type, Nrp1p-TAP and Pbp2p-TAP) show a high degree of PDA1 mRNA degradation during cell lysis (crude lysate in Figure 2 ), although subsequent centrifugation steps do not further enhance degradation of this mRNA (S1, S20, S200 or P200 in Figure 2 ). We also tested, if the RNA becomes degraded during the first step of the TAP protocol. We took samples before and after the IgG immunopurification step and checked RNA integrity by Northern blot. As shown in Figure 3 , no further degradation of the RNAs during this step can be detected. Taken together these results lead to the conclusion that most of the RNA degradation observed during French Press or glass bead mill lysis happens during cell rupture despite the presence of RNase inhibitors in the lysis buffer. Figure 2 RNA of cells lysed by French press is degraded. Strains RJY358 (wt, untagged), RJY933 (Pbp2p-TAP) and RJY929 (Nrp1p-TAP) were lysed in a French Press and samples from crude lysate, supernatant of 1200 × g (S1), 20000 × g (S20), 200000 × g (S200) spins and pellet of 200000 × g (P200) spin were phenol extracted. 8 μg of the extracted RNA were loaded onto 1.2% agarose-formaldehyde gels and blotted onto nylon membranes. 8 μg of total RNA (P/C lysis RNA) from the same strains prepared with a phenol extraction method [18] were loaded in parallel as control for intact RNAs. After methylene blue staining, the membranes were hybridized with a probe against PDA1 mRNA. The positions of the 18S and 25S ribosomal RNAs in the methylene blue staining and the PDA1 mRNA hybridization signal are indicated. The nature of the third band that appears in the methylene blue staining on top of the 25S rRNA is unknown. Figure 3 RNA degradation is restricted to the lysis step prior to TAP purification. Strains RJY933 (Pbp2p-TAP), RJY358 (wt, untagged strain) and RJY929 (Nrp1p-TAP) were lysed in a French Press and processed up to the IgG immunopurification as indicated in Methods. Samples from supernatant of 20000 × g (S20) spin and pellet of 200000 × g (P200) spin, as well as input material (IgG input) and flow through (IgG FT) from the IgG immunopurification step were phenol extracted. 8 μg of the extracted RNA were loaded onto 1.2% agarose-formaldehyde gels and blotted onto nylon membranes. 8 μg of total RNA (P/C lysis RNA) from strain RJY933 (Pbp2p-TAP) prepared with a phenol extraction method [18] were loaded in parallel as control for intact RNAs. The positions of the 18S and 25S ribosomal RNAs in the methylene blue staining and the PDA1 mRNA hybridization signal are indicated. Intact RNAs after lysis by mortar grinding We also tested a third described method for cell lysis, which is grinding deep-frozen cells with a mortar [ 20 - 22 ]. This method breaks cells mechanically while keeping proteins and nucleic acids intact due to the lack of enzymatic activity at ultra-low breaking temperatures. After grinding, the extract is thawed in lysis buffer in the presence of RNase inhibitors (10 mM RVC and 100 U/ml SuperaseIn), immediately inactivating deleterious enzymatic activities. First, we checked the integrity of the RNA after cells were lysed by manual grinding in liquid nitrogen. As shown in Figure 4A , RNA in crude lysate from the three strains tested (wild type, Nrp1p-TAP, Pbp2p-TAP) shows little difference from RNA isolated by direct phenol:chloroform extraction as indicated by the 25S/18S ribosomal RNA ratio and hybridization against PDA1 mRNA. We also tested RNA integrity during subsequent centrifugation steps after cell lysis with a motor-driven mortar. As shown in Figure 4B , the RNA from the different samples analyzed show only little degradation as judged by the PDA1 mRNA hybridization signal. Figure 4 RNA degradation is reduced during lysis by grinding. A , Strains RJY358 (wt, untagged strain), RJY929 (Nrp1p-TAP) and RJY933 (Pbp2p-TAP) were manually ground in liquid nitrogen. Samples from crude lysate were phenol extracted and 8 μg of the extracted RNA were loaded onto 1.2% agarose-formaldehyde gels and blotted onto nylon membranes. 8 μg of total RNA (P/C lysis RNA) from strains RJY358 (wt, untagged strain), RJY929 (Nrp1p-TAP) and RJY933 (Pbp2p-TAP) prepared with a phenol extraction method [18] were loaded in parallel as control for intact RNAs. B , Strains RJY358 (wt, untagged strain), RJY929 (Nrp1p-TAP) and RJY933 (Pbp2p-TAP) were ground in a motor-driven mortar in the presence of dry ice. Samples from crude lysate, supernatant of 20000 × g (S20) spin, and pellet of 200000 × g (P200) spin were phenol extracted and 8 μg of the extracted RNA were loaded onto 1.2% agarose-formaldehyde gels and blotted onto nylon membranes. As control for RNA quality, RNA samples from independently lysed cells were loaded on the same gel. The positions of the 18S and 25S ribosomal RNAs in the methylene blue staining and the PDA1 mRNA hybridization signal are indicated. Both RNase inhibitors are needed to keep mRNA intact To optimize the lysis protocol, we also tested if both RNase inhibitors are needed to keep the RNA intact. For this purpose we first used a strain where She2p, an RNA-binding protein required for ASH1 mRNA localization, has been tagged [ 23 ]. She2p is known to bind to specific regions of ASH1 mRNA, which could help to protect the target mRNA. We lysed cells by mortar grinding and thawed them in lysis buffer with a final concentration of 170 U/ml of SuperaseIn as only RNase inhibitor. We took samples from crude lysate, different steps from the differential centrifugation and from the IgG immunopurification steps (input and flow through) and extracted the RNA. As shown in Figure 5A degradation of both PDA1 and ASH1 mRNAs is already detected in the crude lysate. Whereas no further degradation of PDA1 mRNA is detected during subsequent centrifugation, ASH1 mRNA is heavily degraded, indicating that there might not be a protective effect of RNA-binding proteins like She2p. In addition, when the extract was incubated with the IgG-coupled beads for 2 hours degradation becomes evident also for PDA1 mRNA (compare the last two lines in Figure 5A ). Figure 5 Both RNase inhibitors are needed to keep the RNA intact when cells are lysed by mortar grinding. A , Strain RJY1885 (She2p-TAP) was ground in a motor-driven mortar in the presence of dry ice. Samples from crude lysate and supernatants of 1200 × g (S1) and 7400 × g (S7) spins and pellet from 7400 × g spin (P7), as well as input material (IgG input) and flow through (IgG FT) from the IgG immunopurification, were phenol:chloroform extracted as indicated in Methods. 8 μg from the extracted RNA were loaded onto 1.2% agarose-formaldehyde gels. As RNA quality control, 8 μg of total RNA (P/C lysis RNA) extracted from the same strain with a phenol extraction method [18] were loaded in parallel. Membranes were first probed against ASH1 mRNA, stripped and then probed against PDA1 mRNA. The positions of the 25S and 18S ribosomal RNAs are shown, as well as the PDA1 and ASH1 mRNA hybridization signals. B , Sub2p-TAP strain [24] was ground in a motor-driven mortar in the presence of dry ice and thawed in lysis buffer supplemented with 50 U/ml of SuperaseIn and/or 5 mM RVC. Samples from crude lysate and supernatants of 100000 × g (S100) spin and flow through (IgG FT) from the IgG immunopurification, were phenol:chloroform extracted as indicated in Methods. 8 μg from the extracted RNA were loaded onto 1.2% agarose-formaldehyde gels. As RNA quality control, 8 μg of total RNA (P/C lysis RNA) extracted from the same strain with a phenol extraction method [18] were loaded in parallel. The positions of the 25S and 18S ribosomal RNAs are shown, as well as the PDA1 mRNA hybridization signal. We then checked the effect of a reduced concentration of both RNase inhibitors, SuperaseIn and RVC, on RNA integrity. For this purpose we used a strain where Sub2p, an RNA-binding protein required for splicing and mRNA export, is tagged [ 24 ]. Sub2p is more abundant than She2p and believed to bind to mRNAs competent for export [ 25 ]. Like She2p, it could also have a protective function against mRNA degradation that could be more prominent due to its higher abundance. We lysed cells by mortar grinding and thawed them in lysis buffer containing either 5 mM RVC and 50 U/ml SuperaseIn, or 5 mM RVC, or 50 U/ml SuperaseIn. We took samples from crude lysate, supernatant of a 100000 × g spin, as well as input and flow through from the IgG immunopurification. As shown in Figure 5B , no degradation can be observed when RNA is stained with methylen blue and the signals from the 25S and 18S ribosomal RNAs are compared. In contrast, degradation of PDA1 mRNA is visible in crude lysates from samples that have been thawed in lysis buffers supplemented only with SuperaseIn as inhibitor (compare signal of RNA from lines 8 and 1 in Figure 5B ), in concordance with previous results (Figure 5A ). When RVC is present as sole RNase inhibitor (lines 5–7 in Figure 5B and compare with line 1), degradation is reduced. Only when both inhibitors are present in the lysis buffer (lines 2–4 in Figure 5B and compare with line 1) intact RNAs are detected. Discussion Since most RNA-binding proteins fulfil their function in the context of RNA-protein complexes, knowledge of RNAs associated with specific RBPs is essential to elucidate their functions. In order to identify these transcripts, new methods must be developed or existing successful protocols for the identification of protein-protein interactions must be adapted. Although several recent publications have identified RNA partners from RNP-complexes [ 9 , 16 ], there are so far no reports on the quality of the RNAs purified from these complexes. Here we demonstrate that the method used for cell lysis of yeast cells is of great importance for isolation of intact complexes. Whereas standard lysis methods like disruption by French Press or glass bead mill lead to massive RNA degradation (Figures 1 , 2 , 3 ), grinding yeast cells at ultra-low temperatures leaves cellular RNA intact (Figures 4 , 5 ). The major difference between the lysis methods compared in this study is the timing between addition of the RNase inhibitor and target inactivation. In French Press and glass bead milling, a lag between addition and target inactivation occurs because inhibitors require cells to be broken in order to inactivate RNases as the yeast's cell wall acts as a barrier for large molecules. Since RNases are highly active enzymes, this could result in RNA degradation prior to their inactivation by the inhibitor. This lag could explain the degradation observed in samples from crude lysates produced by these lysis methods even in the presence of RNase inhibitors (Figures 1 , 2 , 3 ). This idea is supported by the lack of further degradation during subsequent steps (Figures 2 , 3 ). Although we have only analyzed the first step of the TAP protocol, the IgG immunopurification, the conclusion from this experiment can be extrapolated to immunoprecipitated RNPs as the molecular interactions in both methods are equivalent. In contrast to bead milling or disruption by French Press, the temperature during grinding is kept close to -80°C and enzymatic activities are essentially absent. The results from Figure 4 show only little RNA degradation after thawing ground extracts in buffer supplemented with RNase inhibitors, which supports cell grinding as an RNA integrity conserving method. Our data demonstrating enhanced stability of mRNAs using the grinding method are supported by previous successful purifications of snRNPs with intact snRNAs using grinding as lysis method [ 21 , 22 ]. RNA degradation can severely bias identification of RNA from purified RNPs because only RNA fragments that are protected from degradation would remain in the purified extract. Under these circumstances, the choice of the primer for retrotranscribing the pool of purified RNAs is critical since it is likely that only primers annealing close to or inside the protected sites would result in a cDNA product. Furthermore, when such protected RNA fragments are used to probe micro-arrays, the type of the DNA-array also becomes crucial. Since many RBPs bind to the UTRs of mRNAs and many DNA-microarrays contain only DNA corresponding to the ORF of the genes, detection of bound RNAs by such microarrays could be biased against RBPs binding outside the ORF. This problem would be enhanced using arrays composed of long oligonucleotides since only RNAs whose protected regions overlap with the oligo sequence would be identified. Another point to consider is the low abundance of several RNPs. As many RNA-binding proteins are expressed at low levels, purification of complexes using RNPs as bait might result in low yield of bait protein plus associated RNAs. Under these circumstances, it is mandatory to keep RNA intact at early stages (in the crude lysate), since the following purification steps might become an additional source of RNA degradation. This idea is also supported by our results (Figure 5A ) where, even under optimised lysis conditions, mRNAs become sensitive to degradation during longer incubation times. Conclusions Our data suggest that mechanical breakage of frozen yeast cells at ultra-low temperatures is the lysis method of choice for purification of intact mRNP complexes. Since in recent approaches [ 9 , 16 ] different lysis conditions have been used, a substantial number of protein-associated mRNAs might have been missed. Methods Yeast strains Strains used in this work are isogenic with W303. The relevant genotypes are listed in Table 1 . Table 1 Strains used for this study. Strains used in this study and its relevant genotype. Modified gene is indicated in bold letters. Strain Genotype RJY358 MAT a, ade2-1 , trp1-1 , can1-100 , leu2-3 , 112 , his3-11 , 15 , ura3 RJY929 MAT a, ade2-1 , trp1-1 , can1-100 , leu2-3 , 112 , his3-11 , 15 , ura3 , NRP1 -TAP :: k.l.TRP1 RJY933 MAT a, ade2-1 , trp1-1 , can1-100 , leu2-3 , 112 , his3-11 , 15 , ura3 , PBP2 -TAP :: k.l.TRP1 RJY1885 MAT alpha, ade2-1 , trp1 ::hisG, can1-100 , leu2-3 , 112 , his3-11 , 15 , ura3 , pep4 :: LEU2 , bar1 :: hisG , SHE2 -TAP :: k.l.TRP1 TAP-tagged strains were generated by transformation with PCR products using the lithium acetate method [ 26 ]. The primers used for generating the PCR products from plasmid pBS1479 [ 11 ] are listed on Table 2 . Table 2 Oligonucleotides used in this study. Oligonucleotides used in this study for generating the tagged strains. Oligo name Sequence (5'-3') Pbp2p-TAP forward AATTGATAGATCAAATGCTGAACGTAAAAGAAGGTCGCCCCTCTCCATGGAAAAGAGAAG Pbp2p-TAP reverse GTAGTTTCTGTATTTTTATTTTCTATGTGTTTTTATTGACTAGTACGACTCACTATAGGG Nrp1p-TAP forward TAATAGCGCTTTCGGTAATGGTTTTAATAGTTCAATACGTTGGTCCATGGAAAAGAGAAG Nrp1p-TAP reverse AAATAAAAAATACAATGTGGTTGTGTGAAATTTATTGACCTCGTACGACTCACTATAGGG She2p-TAP forward GTTGTCGCTACTAAATGGCATGACAAATTTGGTAAATTGAAAAACTCCATGGAAAAGAGAAG She2p-TAP reverse GCTATTCATGTATATATATATGTTCTATTAACTAGTGGTACTTATTACGACTCACTATAGGG Lysis methods Yeast cultures were grown at 30°C in YPD media [ 27 ] up to an OD 600 nm /ml of 2–3. Cultures were harvested at 4°C and washed with ice-cold sterile water. Glass bead mill Cell pellets from 10000 OD 600 nm (5 l cultures with an OD 600 nm /ml of 2) were harvested, washed with cold sterile water and immediately lysed or the wet pellet snap-frozen in liquid nitrogen. For lysis, harvested cells were resuspended in 30 ml lysis buffer (10 mM K-Hepes pH 7.9, 10 mM KCl, 1.5 mM MgCl 2 , 0.5 mM DTT, 1 μg/ml aprotinin, 0.8 μg/ml bestatin, 1 μg/ml leupeptin, 0.05 mg/ml pefabloc (Biomol, Germany), 0.7 μg/ml pepstatinA, 100 U/ml SuperaseIn (Ambion, UK), 20 mM Ribonucleoside Vanadyl Complex (RVC, NEB, UK)) and immediately broken. Frozen cells were stored at -80°C until used. Then, they were thawed in 30 ml lysis buffer and immediately lysed. Cells were lysed in a glass bead mill (Bead-beater, Biospec) in 3 cycles (3 min breaking and 5 min cooling). During lysis, cells were kept cold by a water-ice bath. French Press Cell pellets from 5000 OD 600 nm (2.5 l cultures with an OD 600 nm /ml of 2) were harvested, washed with cold sterile water and immediately lysed or the wet pellet snap-frozen in liquid nitrogen. For lysis, harvested cells were resuspended in 15 ml lysis buffer (10 mM K-Hepes pH 7.9, 10 mM KCl, 1.5 mM MgCl 2 , 0.5 mM DTT, 1 μg/ml aprotinin, 0.8 μg/ml bestatin, 1 μg/ml leupeptin, 0.05 mg/ml pefabloc, 0.7 μg/ml pepstatinA, 100 U/ml SuperaseIn, 20 mM RVC) and immediately broken. Frozen cells were stored at -80°C until used. Then, they were thawed in 25 ml lysis buffer and immediately broken. Lysis was performed as described [ 11 ] with a cold chamber. Mortar grinding Cell pellets corresponding to 3–15 l cultures, with an OD 600 nm /ml of 2–3, were frozen by pressing the cell pellet through a 50 ml syringe directly into liquid nitrogen [ 20 ] and stored at -80°C until used. Up to 3 g frozen cells were ground manually in liquid nitrogen. Larger amounts were ground in a motor-driven mortar (Mortar Grinder RM 100, Retsch, Germany) for 15 min at pressure setting 5–7. Mortar and pestle were precooled twice with liquid nitrogen. Crushed dry ice was added continuously during grinding for keeping the mortar cold. Ground material was stored at -80°C until used. Typically, 25 g of ground cells were used for experiments running up to the first TAP immunopurification step, and 1–5 g of ground cells were used for experiments including only differential centrifugation steps. Ground cells were thawed in 1 ml of lysis buffer (40 mM K-Hepes pH 8, 1 mM MgCl 2 , 0.30% Igepal CA-630, 120 mM NaCl, 80 mM KCl, 2 mM EDTA, 1 mM DTT, 1 μg/ml aprotinin, 0.8 μg/ml bestatin, 1 μg/ml leupeptin, 0.05 mg/ml pefabloc, 0.7 μg/ml pepstatinA, 200 U/ml SuperaseIn, 40 mM RVC) per gram. Sample preparation Glass bead mill and French Press After cell lysis, cell extracts were subjected to serial centrifugations at 4°C. First, the crude lysate (CL) was spun 3 min at 1200 × g to pellet cell debris. The total extract (S1) was further spun for 20 min at 7500 × g and the supernatant collected (S7). Salt and buffer concentration were adjusted to IgG binding conditions (20 mM K-Hepes pH 7.4, 5 mM KCl, 0.75 mM MgCl 2 , 100 mM K-acetate, 10 mM Tris-HCl pH 8, 100 mM NaCl, 10 mM Mg-acetate, 1 mM EGTA, 0.1% Igepal CA-630, 0.5 mM DTT, 100 U/ml SuperaseIn, 10 mM RVC, 1 μg/ml aprotinin, 0.8 μg/ml bestatin, 1 μg/ml leupeptin, 0.05 mg/ml pefabloc, 0.7 μg/ml pepstatinA) with the addition of adjusting buffer (50 mM K-Hepes pH 7.4, 140 mM K-acetate, 30 mM Tris-HCl pH 8, 300 mM NaCl, 3 mM EGTA, 0.3% Igepal CA-630, 1 mM DTT, 100 U/ml SuperaseIn, 10 mM RVC, 2 μg/ml aprotinin, 1.6 μg/ml bestatin, 2 μg/ml leupeptin, 0.1 mg/ml pefabloc, 1.4 μg/ml pepstatinA). The resulting sample was again spun for 20 min at 20000 × g and the supernatant (S20) finally spun at 200000 × g for 20 min. The supernatant was collected (S200) and the resulting pellet (P200) was resuspended in 10 ml IgG binding buffer (20 mM K-Hepes pH 7.4, 5 mM KCl, 0.75 mM MgCl2, 100 mM K-acetate, 10 mM Tris-HCl pH 8, 100 mM NaCl, 10 mM Mg-acetate, 1 mM EGTA, 0.1% Igepal CA-630, 0.5 mM DTT, 100 U/ml SuperaseIn, 10 mM RVC, 1 μg/ml aprotinin, 0.8 μg/ml bestatin, 1 μg/ml leupeptin, 0.05 mg/ml pefabloc, 0.7 μg/ml pepstatinA). Mortar grinding After thawing the ground cells in lysis buffer, the sample was subjected to serial centrifugation at 4°C. First, the crude lysate (CL) was spun 3 min at 1200 × g to pellet cell debris. Total extract (S1) was further spun for 20 min at 7500 × g and the supernatant (S7) was collected. The resulting pellet (P7) was resuspended in 10 ml 0.5 × lysis buffer. Supernatant (S7) was again spun for 20 min at 20000 × g and the supernatant (S20) finally spun at 200000 × g for 20 min. The supernatant was collected (S200) and the resulting pellet (P200) was resuspended in 10 ml 0.5 × lysis buffer. For TAP purifications of Sub2p, total extract (S1) was directly centrifuged at 100000 × g for 1 h (S100). The lipid phase was discarded from all fractions. 100–300 μl aliquots from the different fractions were extracted immediately 3–4 times with phenol:chloroform:isoamylalcohol (pH 4), precipitated with sodium acetate and ethanol and stored at -80°C until analyzed. Glycerol (5% final concentration) was then added to those samples used for further TAP purification steps. These samples were snap-frozen in liquid nitrogen and stored at -80°C for a maximum of 2–3 days. Aliquots were taken before freezing and immediately after thawing to control for degradation during freeze-thawing. These aliquots were extracted, precipitated and stored as described above. TAP purification was performed up to the IgG immunopurification step (first step of the TAP protocol) as described [ 11 ], using as IgG binding buffer 20 mM K-Hepes pH 7.4, 5 mM KCl, 0.75 mM MgCl 2 , 100 mM K-acetate, 10 mM Tris-HCl pH 8, 100 mM NaCl, 10 mM Mg-acetate, 1 mM EGTA, 0.1% Igepal CA-630, 0.5 mM DTT, 100 U/ml SuperaseIn, 10 mM RVC, 1 μg/ml aprotinin, 0.8 μg/ml bestatin, 1 μg/ml leupeptin, 0.05 mg/ml pefabloc, 0.7 μg/ml pepstatinA. Samples were taken before (IgG input) and after IgG immunopurification (flow through, IgG FT) and immediately extracted, precipitated and stored at -80°C as described above. Northern blot Total RNA was isolated by direct phenol:chloroform extraction from the same strains used in the study as described [ 18 ], precipitated with sodium acetate and ethanol and kept at -80°C in ethanol until used as quality control. RNA pellets from the different samples were resuspended in DEPC-treated water, ratio A 260 nm /A 280 nm measured and equivalent amounts of RNA were loaded onto 1.2% agarose-formaldehyde gels. Samples loaded on gels for direct visualization were supplemented with 0.2 μg/μl of ethidium bromide in loading buffer. For Northern blot analysis gels were blotted onto nylon membranes, the membrane stained with methylene blue for determining transfer efficiency and probed against ASH1 and/or PDA1 mRNA (see Figure legend for details). List of abbreviations The abbreviations used are: RNP, ribonucleoprotein; RBP, RNA-binding protein; TAP, tandem affinity purification; RVC, ribonucleoside vanadyl complex; RRM, RNA recognition motif; UTR, untranslated region; ORF, open reading frame; DEPC, diethylpyrocarbonate; TCA, trichloroacetic acid; TEV, tobacco etch virus protease. Authors' contributions MLdH carried out the studies and drafted the manuscript. RPJ conceived of the study, and participated in its design and coordination. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524479.xml
520810	Endoplasmic reticulum degradation impedes olfactory G-protein coupled receptor functional expression	Background Research on olfactory G-protein coupled receptors (GPCRs) has been severely impeded by poor functional expression in heterologous systems. Previously, we demonstrated that inefficient olfactory receptor (OR) expression at the plasma membrane is attributable, in part, to degradation of endoplasmic reticulum (ER)-retained ORs by the ubiquitin-proteasome system and sequestration of ORs in ER aggregates that are degraded by autophagy. Thus, experiments were performed to test the hypothesis that attenuation of ER degradation improves OR functional expression in heterologous cells. Results To develop means to increase the functional expression of ORs, we devised an approach to measure activation of the mOREG OR (Unigene # Mm.196680; Olfr73) through coupling to an olfactory cyclic nucleotide-gated cation channel (CNG). This system, which utilizes signal transduction machinery coupled to OR activation in native olfactory sensory neurons, was used to demonstrate that degradation, both by the ubiquitin-proteasome system and autophagy, limits mOREG functional expression. The stimulatory effects of proteasome and autophagy inhibitors on mOREG function required export from the ER and trafficking through the biosynthetic pathway. Conclusions These findings demonstrate that poor functional expression of mOREG in heterologous cells is improved by blocking proteolysis. Inhibition of ER degradation may improve the function of other ORs and assist future efforts to elucidate the molecular basis of odor discrimination.	Background The sense of smell originates in the olfactory epithelium when olfactory receptors (ORs), members of the seven transmembrane domain G-protein coupled receptor (GPCR) superfamily, bind odorant ligands [ 1 , 2 ]. Despite identification of the first constituents of the ~1000 member OR superfamily over a decade ago, efforts to uncover the molecular basis of odor discrimination have been severely limited by the inability to efficiently express ORs at the plasma membrane in heterologous expression systems [ 2 - 6 ]. Recently, we elucidated three specific cellular mechanisms responsible for inefficient OR trafficking to the plasma membrane: ORs are retained within the endoplasmic reticulum (ER) due to inefficient folding and poor coupling to ER export machinery, degraded via the ubiquitin-proteasome system, and sequestered in ER aggregates that are degraded by autophagy [ 7 ]. Thus, we have a clearer understanding of the problems associated with OR expression in heterologous cells. To develop rationale means to improve the functional expression of ORs, an approach was devised to quantitate activation of the mouse mOREG OR (Unigene # Mm.196680; Olfr73), which recognizes the odorant eugenol (spicy, cinnamon-like odor) [ 8 ], following coupling to an olfactory cyclic nucleotide-gated cation channel (CNG) [ 9 ]. Using this assay, we show that degradation by both the ubiquitin-proteasome system and autophagy limits mOREG functional expression. Our results demonstrate for the first time that inhibition of proteolysis can positively modulate OR function. Results and discussion Functional expression of mOREG using a CNG-based assay A cell-based approach was developed to measure functional expression of mOREG in heterologous cells. This system was designed to mimic the signal transduction events involved in OR activation in the olfactory epithelium and utilizes CNG as a cAMP biosensor [ 10 , 11 ]. In olfactory sensory neurons, odorant binding to ORs initiates a signaling cascade involving the heterotrimeric G protein G olf , adenylate cyclase III, and an olfactory CNG, leading ultimately to the sensation of smell [ 1 , 12 ]. Accordingly, we transiently co-expressed mOREG, as an N-terminal fusion protein with the first 20 amino acids of rhodopsin (Rho20-mOREG), with untagged olfactory CNG subunits in HEK293 cells that endogenously express both the heterotrimeric G protein G s , a functional homologue of G olf [ 13 , 14 ], and adenylate cyclase III [ 15 ]. The Rho tag has been shown to facilitate chemosensory GPCR functional expression [ 8 , 16 , 17 ], possibly by enhancing translocation into the ER during protein synthesis. In this system, odorant binding to Rho-mOREG, which couples to endogenous G s and elicits increases in the second messenger cAMP in HEK cells [ 8 , 18 , 19 ], leads to the opening of CNG and the influx of calcium from the extracellular medium. Thus, Rho-mOREG function can be directly correlated with cellular calcium levels. Responses to eugenol, a ligand for mOREG, were detected in cells co-expressing Rho-mOREG and CNG (Fig. 1A and 1B ) but not in cells expressing vector only, Rho-mOREG only, or CNG only (Fig. 1B ). Eugenol activation of Rho-mOREG became evident at 20 sec and increased to a maximal level at 50 sec following odorant application. The EC 50 for eugenol (20.8 +/- 3.4 uM) closely approximated published values (35 uM and 46 uM), thereby validating the utility of our approach [ 8 , 19 ]. Rho-mOREG activation was specific for eugenol as no response was observed when cells were challenged with the control odorants heptanal and octanal, which activate the I7 OR (Fig. 1B ) [ 4 , 16 , 20 ]. Collectively, these data demonstrate that Rho-mOREG can functionally couple to CNG in heterologous cells following odorant stimulation. Figure 1 Functional expression of Rho-mOREG using a CNG-based assay. (A) HEK293 cells transiently transfected with Rho-mOREG and CNG were assayed for increases in intracellular calcium in response to 100 uM eugenol. Images contain ~2000 confluent cells and represent responses at the indicated times following odorant stimulation. Scale bar is 300 um. Responses were observed in ~10% of total cells or ~20% of transfected cells (~50% transfection efficiency). (B) Quantitation of cells expressing vector only (a), Rho-mOREG only (b), CNG only (c), Rho-mOREG and CNG (d and e) following stimulation with 100 uM eugenol (a-d) or 100 uM heptanal plus 100 uM octanal (e). * p < 0.001 compared to cells expressing vector only (a). ER degradation limits mOREG functional expression Previously, we demonstrated that ORs are inefficiently expressed at the plasma membrane of heterologous cells due, in part, to degradation of ER-retained ORs by the ubiquitin-proteasome system and sequestration of ORs in ER aggregates that are degraded by autophagy [ 7 ]. Specifically, inhibition of the proteasome using MG-132 or inhibition of autophagy using 3-methyladenine (3-MA) enhanced Rho-mOREG protein expression 2 to 3-fold by biochemical and cellular analyses [ 7 ]. Therefore, experiments were performed to test the hypothesis that ER degradation limits OR functional expression. These studies used cells stably expressing CNG and transiently expressing Rho-mOREG to achieve more robust responses. As shown in Figure 2 , treatment of cells with 3-MA, a specific inhibitor of autophagy that blocks sequestration of material into autophagosomes [ 21 ], enhanced Rho-mOREG activation by eugenol (compare Fig. 2A and 2B ). Similar results were obtained when cells were treated with MG-132 (compare Fig. 2A and 2C ) or epoxomicin (data not shown), a specific inhibitor of the proteasome [ 22 ]. Autophagy and proteasome inhibitors increased Rho-mOREG functional expression 2 to 3-fold in the linear range of eugenol dose response curves (Fig. 2D and 2F ). Importantly, 3-MA and MG-132 specifically promoted Rho-mOREG function, since responses of the β-adrenergic receptor (β-AR), a GPCR that utilizes the same signal transduction machinery as Rho-mOREG (GPCR-G s -adenylate cyclase) [ 14 ], following isoproterenol challenge were unaffected (Fig. 2E and 2G ). Thus, degradation inhibitors were not modulating the function or trafficking of elements of the signal transduction machinery, including CNG, coupled to Rho-mOREG. Collectively, these data suggest that ER degradation, by autophagy and the ubiquitin-proteasome system, limits Rho-mOREG functional expression. Figure 2 ER degradation limits Rho-mOREG functional expression. CNG cells transiently transfected with Rho-mOREG were assayed for increases in intracellular calcium in response to 10 uM eugenol following treatment with vehicle (control, 0.1% DMSO; A), 10 mM 3-MA (B) or 50 uM MG-132 (C) for 4 h. Images contain ~750–1000 confluent cells and represent responses 60 sec following odorant stimulation. Scale bar is 150 um. Dose-response curves were determined for eugenol (300 nM to 300 uM; D and F) or isoproterenol (100 pM to 100 nM; E and G) in cells treated with vehicle (control), 10 mM 3-MA (D and E) or 50 uM MG-132 (F and G) for 4 h. ~20–25% of total cells or ~40–50% of transfected cells (~50% transfection efficiency) responded to 100 uM eugenol under control conditions. Note that degradation inhibitors specifically increased Rho-mOREG functional responses and had no effect on β-AR function. The EC 50 value for eugenol was larger for control (17.1 +/- 3.0 uM) compared to 3-MA (10.8 +/- 2.6 uM) and MG-132 (8.1 +/- 2.6 uM) treatments, but these differences did not achieve statistical significance. The EC 50 values for isoproterenol for control (3.7 +/- 1.0 nM), 3-MA (4.2 +/- 2.5 nM), and MG-132 (2.6 +/- 0.9 nM) treatments were not significantly different. To determine if increased functional expression of Rho-mOREG following inhibition of ER degradation was attributable to use of an OR fusion protein, we examined the effect of autophagy and proteasome inhibitors on mOREG lacking the Rho tag. Although eugenol activation of Rho-mOREG (EC 50 = 17.1 +/- 3.0 uM; maximum number of responding cells at 300 uM eugenol = 214 +/- 10) was more robust than untagged mOREG (EC 50 = 36.5 +/- 12.9 uM; maximum number of responding cells at 300 uM eugenol = 48 +/- 5; p < 0.05 compared to Rho-mOREG for both EC 50 and maximal number of responding cells), likely due to an established role of the Rho tag in facilitating chemosensory GPCR functional expression [ 8 , 16 , 17 ], MG-132 and 3-MA increased untagged mOREG functional expression 2 to 3-fold (Fig. 3 ), similar to the magnitude observed with Rho-mOREG (Fig. 2 ). Thus, inhibition of ER degradation increased Rho-mOREG and untagged mOREG functional expression, demonstrating that observed effects were not attributable to use of a non-native OR fusion protein. Figure 3 ER degradation limits untagged mOREG functional expression. CNG cells transiently transfected with untagged mOREG were assayed for increases in intracellular calcium in response to increasing concentrations of eugenol (from 3 uM to 1000 uM) following treatment with vehicle (control), 10 mM 3-MA, or 50 uM MG-132 for 4 h. The EC 50 values for eugenol for control (36.5 +/- 12.9 uM), 3-MA (42.6 +/- 6.5 uM), and MG-132 (36.4 +/- 7.5 uM) treatments were not significantly different. CNG cells transfected with vector only (pUC18) and treated with vehicle, 3-MA, or MG-132 did not respond to 1000 uM eugenol. Two seemingly independent mechanisms degrade ORs retained in the ER. First, OR aggregates sequestered in ER subdomains are targeted to lysosomes for degradation by autophagy, and second, misfolded ORs are covalently modified by polyubiquitination and degraded by the proteasome [ 7 ]. To determine if simultaneous inhibition of autophagy and the ubiquitin-proteasome system produced additive effects on Rho-mOREG function, we co-treated cells with 3-MA and MG-132. As shown in Figure 4 , Rho-mOREG function, measured using 10 uM eugenol, a non-saturating concentration near the EC 50 value, was equivalent in cells treated with 3-MA alone, MG-132 alone, or 3-MA plus MG-132. Figure 4 Simultaneous inhibition of autophagy and the proteasome does not produce additive Rho-mOREG functional responses. CNG cells transiently transfected with Rho-mOREG were assayed for increases in intracellular calcium in response to 10 uM eugenol following treatment with vehicle (control), 10 mM 3-MA, 50 uM MG-132, or 10 mM 3-MA plus 50 uM MG-132 for 4 h. Individual or combined treatment with degradation inhibitors yielded similar levels of Rho-mOREG activity. * p < 0.001 compared to cells treated with vehicle. The non-additive effects of 3-MA and MG-132 on Rho-mOREG function suggest one of the following two non-mutually exclusive scenarios: first, OR degradation by autophagy may be linked to OR degradation by the ubiquitin-proteasome system, by a poorly defined mechanism as suggested for other aggregation prone proteins [ 23 - 25 ]; second, in addition to ER degradation, an additional step(s) downstream of proteolysis may limit Rho-mOREG functional expression in heterologous cells. A recent preliminary report described specialized accessory proteins that increase OR surface expression and function [ 26 ]. These proteins could serve as chaperones to package OR cargo into COPII vesicles for export from the ER and/or couple ORs to requisite signal transduction machinery at the plasma membrane, steps that are both downstream of ER degradation. In the absence of necessary accessory proteins, functional expression may not exceed a certain level regardless of the quantity of OR that is diverted from the degradative pathways by autophagy and ubiquitin-proteasome inhibitors. The existence of multiple steps limiting OR functional expression in heterologous cells is further supported by our findings that the function of mOREG, lacking the Rho tag, is less robust than Rho-mOREG. Since the Rho tag may facilitate translocation into the ER during protein synthesis [ 8 , 16 , 17 ], ER translocation could comprise an additional limiting step, upstream of ER degradation, for OR functional expression. ER export and trafficking through the Golgi apparatus are necessary for Rho-mOREG functional expression Inhibition of ER degradation events could permit a pool of Rho-mOREG to achieve an ER export competent conformation and traffic through the Golgi apparatus to the plasma membrane. However, using surface biotinylation, surface immunofluorescence microscopy, flow cytometry, and glycosidase digestion assays, we were unable to demonstrate convincing Rho-mOREG surface expression or visualize a pool of Rho-mOREG containing endoglycosidase H-resistant carbohydrate modifications indicative of transit through the Golgi apparatus following treatment with ER degradation inhibitors (ML and BDM unpublished observations). These results suggested that either a small pool of properly folded Rho-mOREG was expressed at the plasma membrane in quantities below the threshold of the cell biological techniques employed to visualize the receptor, or that a pool of intracellular Rho-mOREG comprised the functionally responsive population in calcium imaging experiments. Notably, numerous studies have documented the functional expression of GPCRs and requisite signal transduction machinery in intracellular compartments, including ER membranes [ 27 - 29 ]. To differentiate between these two possibilities, we adopted a pharmacological approach to selectively and independently block trafficking from compartments in the early secretory pathway, specifically the ER and Golgi apparatus. If inhibition of ER degradation does not increase Rho-mOREG activity under conditions that block export from ER and Golgi compartments, the functionally responsive Rho-mOREG population is likely derived from an internal pool that is required to traffic to the plasma membrane to function. Conversely, if inhibition of ER degradation increases Rho-mOREG activity under conditions that block export from ER and Golgi compartments, the functionally responsive Rho-mOREG population likely resides in an intracellular compartment. To inhibit protein trafficking from the ER, brefeldin A (BFA), which blocks ER export of cargo proteins by inducing collapse of the Golgi stacks into the ER, was used [ 30 ]. As shown in Fig. 5A , BFA completely blocked the enhancement of Rho-mOREG function by 3-MA and MG-132; by contrast, BFA had no effect on functional responses of the β-AR, indicating that BFA was not affecting the function or trafficking of signal transduction machinery, including CNG, coupled to Rho-mOREG. By blocking transport of proteins present in the ER that are in route to the plasma membrane, specifically Rho-mOREG following inhibition of ER degradation, BFA inhibited Rho-mOREG function; β-AR and CNG function were unperturbed since these proteins were already present at the plasma membrane prior to BFA treatment. Since β-AR exhibits a long half-life at the plasma membrane [ 31 ], inhibiting delivery of newly synthesized β-AR by BFA would not adversely affect isoproterenol responses. Thus, ER export is required for increased Rho-mOREG functional expression by degradation inhibitors. Figure 5 ER export and trafficking through the Golgi apparatus are required for Rho-mOREG functional expression. CNG cells transiently transfected with Rho-mOREG were assayed for increases in intracellular calcium in response to 10 uM eugenol, to gauge Rho-mOREG function, or 1 nM isoproterenol, to gauge β-AR function, following treatment with vehicle (0.1% ethanol), BFA (5 ug/ml), monensin (10 uM), or incubation at 20°C for 4 h (A) or 5 min (B). (C) Cells were treated with vehicle or CHX (75 uM) for 4 hr. In all experiments, cells were also co-treated for 4 h with control (0.1% DMSO), 10 mM 3-MA, or 50 uM MG-132 as indicated. Partial inhibition of Rho-mOREG function with BFA or monensin but not 20°C (A, control columns), a temperature that also attenuates endocytic events, is likely attributable to turnover of cell surface Rho-mOREG when the biosynthetic pathway, which normally replenishes the plasma membrane Rho-mOREG pool, is blocked. By contrast, BFA and monensin do not affect function of the β-AR, a control GPCR that exhibits a long half-life at the plasma membrane [31]. * p < 0.005 compared to cells treated with vehicle in the same group. To inhibit protein trafficking from the Golgi apparatus, monensin, an ionophore that disrupts Golgi structure and inhibits Golgi trafficking events, was used [ 32 ]. Monensin, similar to BFA, completely blocked the enhancement of Rho-mOREG function by 3-MA and MG-132 while having no effect on β-AR function (Fig. 5A ). Similar effects were observed when cells were incubated at 20°C (Fig. 5A ), a temperature that arrests protein transport at trans Golgi cisternae [ 33 ]. Importantly, trafficking disrupting agents did not affect Rho-mOREG or β-AR function when acutely applied to cells, further substantiating that results were not attributable to non-specific effects on signal transduction machinery or eugenol binding (Fig. 5B ). Thus, Golgi trafficking events are required for increased Rho-mOREG functional expression by degradation inhibitors. Inhibition of Rho-mOREG function by trafficking disrupting agents, specifically BFA, could be due to activation of the unfolded protein response [ 34 ] and inhibition of Rho-mOREG protein synthesis. To directly address this possibility, experiments were performed to test the effect of the protein synthesis inhibitor cycloheximide (CHX) on functional expression of Rho-mOREG. As shown in Figure 5C , CHX, used at concentrations previously demonstrated to inhibit Rho-mOREG translation [ 7 ], had no effect on increased Rho-mOREG function by 3-MA and MG-132. These data suggest that autophagy and ubiquitin-proteasome inhibitors diverted an existing pool of Rho-mOREG from degradative pathways to the secretory pathway and that effects of degradation inhibitors and trafficking disrupting agents were not attributable to modulation of protein synthesis. Collectively, these data support a model whereby inhibition of ER degradation promotes a small fraction of Rho-mOREG to achieve an ER export competent conformation, thereby satisfying ER quality control processes, and traffic through the secretory pathway to the plasma membrane. Thus, similar to chemical and pharmacological chaperones that promote folding and restore ER export of misfolded GPCR cargo [ 35 - 37 ], agents interfering with ER degradation may promote ER export of GPCR and non-GPCR cargo [ 38 ], trafficking through the biosynthetic pathway, and functional expression at the cell surface. We speculate that the pool of Rho-mOREG expressed at the plasma membrane is below the detection limits of cell biological techniques used to visualize the receptor [ 7 ] but above the detection threshold for calcium imaging methodology used to examine receptor function. Indeed, following pharmacological treatments, improved ΔF508 cystic fibrosis transmembrane conductance regulator functional expression at the plasma membrane is readily detectable by sensitive electrophysiological analyses but neither by cell surface labeling nor by biochemical approaches measuring carbohydrate modifications indicative of transit through the Golgi apparatus [ 39 - 41 ]. Though we favor a model whereby inhibition of ER degradation promotes OR export from the ER and improves OR surface expression, we were unable to obtain cellular and biochemical evidence to support this proposal. We cannot exclude the possibility that inhibition of ER degradation also stabilizes an otherwise labile cofactor or chaperone protein, endogenously expressed by heterologous cells, that modulates OR trafficking in a post-Golgi compartment, stability at the cell surface, and/or function at the plasma membrane [ 42 , 43 ]. Conclusions We have developed an expression system for ORs that utilizes signal transduction machinery coupled to OR activation in native olfactory sensory neurons. Using CNG as a cAMP biosensor to gauge mOREG function, we demonstrate that inhibition of ER degradation, by both autophagy and the ubiquitin-proteasome system, promotes functional expression of Rho-mOREG as well as untagged mOREG. Thus, proteolysis limits mOREG function in heterologous cells. Inhibition of ER degradation may improve the function of other ORs and assist future efforts to elucidate the molecular basis of odor discrimination. Methods Molecular biology Rho20-mOREG expression vector was generated in pRK5 as previously described [ 7 ]. To generate untagged mOREG expression vector, mOREG coding sequence was excised using AscI/NotI and subcloned into a modified pRK5 vector lacking the Rho20 tag. The human CNGA2 and CNGB1b expression constructs encode untagged human CNGA2 and CNGB1b in pEAK10-derived vectors (Edge Biosystems, Gaithersburg, MD) [ 44 ]. For the generation of stable transfectants, CNGA2 was subcloned into pCDNA3.1/zeo (Invitrogen, Carlsbad, CA). The CNGA2 clone contains the C458W and E581M mutations, introduced using the QuickChange Site-Directed Mutagenesis Kit (Stratagene, La Jolla, CA), previously shown to increase cAMP sensitivity in rat CNGA2 [ 11 ]. Compounds, odorants, and ligands BFA, isoproterenol, monensin, and 3-MA were from Sigma (St. Louis, MO); MG-132 was from Calbiochem (San Diego, CA); eugenol, heptanal and octanal were from Aldrich (Milwaukee, WI). Cell culture and transfections HEK293 cells were maintained and transfected as previously described [ 7 ]. For the generation of CNG stable transfectants, cells were transfected with linearized CNGA2 and CNGB1b expression constructs (1:1 ratio) and selected using 50 ug/mL zeocin (Invitrogen) and 0.5 ug/mL puromycin (Calbiochem). Individual colonies were expanded and screened for CNG expression by assaying functional responses to 500 uM eugenol following transient transfection with Rho-mOREG. For functional expression studies, CNG cells were grown in media without selection 72 h prior to experimentation and transiently transfected with mOREG 48 h prior to experimentation. mOREG functional expression Functional expression of mOREG was investigated using calcium imaging methodology as previously described [ 44 ]. Cells seeded in 24 well plates were loaded with the calcium dye fluo-4 acetoxymethyl ester (Molecular Probes, Eugene, OR) 2 d post-transfection using the following conditions: 3 uM dye in 0.5 ml Hanks' balanced salt solution containing divalent cations (HBSS, Invitrogen) for 1 h at room temperature in the dark. Cells were subsequently washed once with 0.5 ml HBSS to remove excess fluo-4, supplemented with 0.25 ml HBSS, and then stimulated with an additional 0.25 ml HBSS containing the appropriate ligand at twice the final concentration. A single ligand was applied to each dish of cells. Typically, 3–4 separate dishes of cells were used for each ligand concentration or condition and experiments were repeated 3–4 times. Thus, individual data points represent the average of 9–16 separate measurements. Changes in intracellular calcium were monitored by fluorescence microscopy using an Axiovert S100 TV inverted microscope with a 10× long working distance Plan Fluor objective (numerical aperture 0.5) and a cooled charge-coupled device camera (Princeton Instruments, Trenton, NJ). Images were acquired using a Lambda DG-4 automated wavelength controller (Sutter Instrument Co., Novato, CA) at 480 nm excitation and 535 nm emission and analyzed using Imaging Workbench 4.0 (Axon Instruments, Union City, CA). Counting the number of cells responding to ligands 60 sec following stimulus addition, when cells had achieved a maximal response, quantitated receptor activity. This established methodology has been used to functionally characterize the human T1R1/T1R3 umami receptor, the human T1R2/T1R3 sweet receptor, and the Drosophila Gr5a trehalose receptor [ 44 , 45 ]. To independently validate this method, we determined that the EC 50 for isoproterenol activation of the β-AR (3.7 +/- 1.0 nM), measured by counting responding cells, closely matched published values (1.7–3.3 nM), measured either by fluorescent intensity measurements or a cAMP accumulation assay [ 10 ]. In addition, the EC 50 for eugenol activation of Rho-mOREG (20.8 +/- 3.4 uM), measured by counting responding cells, closely approximated the published value (35 uM and 46 uM), determined by monitoring fluorescent intensity of responding cells [ 8 , 19 ]. Finally, the EC50s for glutamate activation of mGluR4 and cycloheximide activation of mT2R05 were similar when determined by counting cells or by monitoring fluorescent intensity [ 44 ]. We estimate ~40–50% of transfected CNG cells express functional cell surface Rho-mOREG based on the following points. First, ~20–25% of total cells in a microscopic field respond to maximal doses of eugenol. Second, ~50% of cells are transfected, measured by either co-transfection with red fluorescent protein or by immunolabelling permeabilized cells expressing Rho-mOREG with an anti-Rho antibody. Thus, ~40–50% of transfected cells express sufficient Rho-mOREG at the plasma membrane to elicit a functional response. Statistics Data represent the mean +/- SEM. Statistical significance was determined using an unpaired, two-tailed Student's t-test. Dose-response curves were plotted and EC 50 values were determined using GraphPad Prism v3.02 software. Authors' contributions ML generated the mOREG expression vectors and carried out most of the calcium imaging experiments. LS generated the cells stably transfected with CNG. FE assisted with the calcium imaging experiments. HX generated the CNG expression vectors. BDM coordinated the study and wrote the paper.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC520810.xml
546217	Lymphocyte subsets in hemophilic patients with hepatitis C virus infection with or without human immunodeficiency virus co-infection: a nested cross-sectional study	Background With chronic infection, hepatitis C virus (HCV) RNA can be detected in B cells and associated with B-cell disorders, but these are not well defined. Methods The relationship between HCV infection and lymphocyte subpopulations was evaluated rigorously in 120 asymptomatic hemophilic patients, randomly selected from a prospective cohort study. CD4 + T cells, CD8 + T cells, CD19 + B cells, and CD56 + NK cells were quantified by flow cytometry using cryopreserved peripheral blood mononuclear cells of 24 hemophilic patients in each of five age-matched groups [uninfected; chronic HCV with or without human immunodeficiency virus (HIV); and cleared HCV with or without HIV]. Results As expected, patients with HIV had significantly reduced CD4 + and increased CD8 + T cells. Irrespective of HIV, patients with chronic HCV infection had approximately 25% fewer CD19 + B cells than those without chronic HCV infection. Conclusions These data support the hypothesis that asymptomatic patients with chronic HCV infection have an altered B-lymphocyte population.	Background The natural history of hepatic C virus (HCV) infection may include B-cell diseases. HCV can be detected in peripheral blood mononuclear cells (PBMCs) and may also replicate in these cells [ 1 - 3 ]. Type II cryogloblinemia and some non-Hodgkin lymphomas have been linked to HCV [ 4 ], but links between B cells and asymptomatic HCV infection are poorly defined. One study reported that hemophilic patients co-infected with HCV and human immunodeficiency virus (HIV) had reduced B cells and CD4 + T cells [ 5 ]. To further characterize PBMC immunophenotypes in patients with chronic or cleared HCV infection, we quantified proportions of CD4 + T cells, CD8 + T cells, B cells, and natural killer (NK) cells in PBMCs of carefully selected, well characterized hemophilic patients. Methods Study subjects Patients with hemophilia or another congenital coagulation disorder (herein referred to as "hemophilia") enrolled in the Multicenter Hemophilia Cohort Study [ 6 , 7 ] were categorized into five groups: no HIV or HCV infection (HIV-/HCV-); chronic (viremic) HCV infection without HIV (HIV-/HCV RNA+); chronic (viremic) HCV infection with HIV (HIV+/HCV RNA+); cleared (non-viremic) HCV infection without HIV (HIV-/HCV RNA-); and cleared (non-viremic) HCV infection with HIV (HIV+/HCV RNA-). From each of the five categories, 24 age-matched subjects were randomly selected. Non-viremic subjects were considered to have cleared HCV spontaneously, as they were selected from among all those who, without specific therapy, were negative for HCV RNA in two specimens collected at least 6 months apart. Viremic subjects were selected from among all those with an HCV RNA level above the 67 th percentile of all HCV-positive MHCS patients (described below), all of whom had received HCV-contaminated plasma products many years earlier and thus were considered to have chronic HCV infection. All samples were taken before 1996, the year when highly active antiretroviral therapy (HAART) became available for treatment of HIV infection. HCV and HIV assays HIV status was defined by antibody testing with licensed immunoassays and immunoblot confirmation. HCV antibody status was determined with a commercially available second- or third-generation enzyme immunoassay, with most of the reactive samples confirmed by recombinant immunoblot assay (HCV RIBA2.0 or 3.0, Chiron Corp., Emeryville CA). HCV RNA was detected and viral load was determined with branched-DNA technology [Quantiplex HCV RNA 2.0 Assay (bDNA), Chiron Corp., Emeryville CA] with a lower limit of sensitivity of 200,000 genome equivalents/mL, which is 31,746 international units (IU)/mL. HIV-1 viral load was determined with the HIV Amplicor Monitor (Roche Molecular Diagnostics, Branchburg, NJ). Flow cytometry A vial of 5 million cryopreserved PBMCs, stored in vapor phase LN 2 until testing, was analyzed by flow cytometry for proportions of CD4 + , CD8 + , CD19 + , and CD56 + cells. In a biological safety cabinet, vials were thawed in a 37°C water bath with agitation. To each, 300 units of DNase I (Rnase-free, Roche Molecular Biochemicals) was added. The contents were transferred to a 15 ml polypropylene tube. Thawed cells were slowly diluted with RPMI-1640 media supplemented with 20% fetal calf serum, mixed frequently, then pelleted at 1000 rpm in a Sorvall RT6000 refrigerated centrifuge. The liquid was decanted. Cells were mixed in residual volume by gentle shaking, washed a second time with 4.0 ml of thawing media, then resuspended in 2.0 ml calcium- and magnesium-free Dulbecco's phosphate buffered saline (PBS) containing 4% heat-inactivated human AB serum to block high affinity Fc receptors. After 10 minutes the cells were pelleted in the centrifuge, then resuspended in 0.6 ml of PBS/2% BSA. 100 μl of cell suspension was added to each of five 12 mm × 75 m polypropylene tubes containing three-color combinations of fluorescently tagged monoclonal antibodies (CD45/CD14/CD3, CD45/CD4/CD3, CD45/CD8/CD3, CD45/CD19/CD3 and CD45/CD56/CD3). The lymphocyte gate was defined with light scatter properties and CD45/CD14/CD3. A Coulter XL flow analyzer was used with stops set to collect 5,000 CD3 + lymphocytes in each sample, if possible. Listmode files were analyzed offline using Coulter System II software (version 3.0). Quality control criteria included purity of gated lymphocytes; percentage recovery of lymphocytes in the gate; reproducibility of CD3 between tubes (range <4%); and sum of CD3 + , CD3 - /CD19 + , CD3 - /CD56 + (range 90–110%). Statistical analysis The primary analysis was performed on proportions, rather than absolute levels, of each PBMC subset, to reduce variance that results from counting total lymphocytes in peripheral blood [ 8 ]. The Kruskal-Wallis and Wilcoxon rank sum tests were used to compare the lymphocyte subsets between groups, particularly chronic versus cleared HCV. Statistical analyses were done with the Statistical Analysis System version 6.0 (Cary, NC). Results Characteristics of study subjects Of the 120 patients, 78 had hemophilia A (factor VIII deficiency). Because patients with hemophilia A generally required more intensive clotting factor replacement therapy as well as product (Factor VIII concentrate) that was more infectious for HIV than was Factor IX concentrate used for patients with hemophilia B, hemophilia A was significantly more common in the two groups with HIV co-infection (n = 19 with chronic HCV and n = 21 with cleared HCV) than the three groups without HIV co-infection (n = 11 to 14, p = 0.01). By design, the five groups had similar ages (mean 26.5 years, range 3.4 – 54.8 years, Table 1 ). HCV RNA levels were higher with HIV co-infection (median 4.6 × 10 6 IU/mL) than with HCV only (median 2.4 × 10 6 IU/mL, P < 0.01). Among 48 HIV-infected subjects, 6 (25%) with cleared and 7 (29%) with chronic HCV had clinically defined AIDS (P = 0.75) when their PBMCs were collected for this study. HIV viral load did not differ between the nine with cleared and the eight with chronic HCV who were tested on the date PBMCs were isolated (P = 0.89). Ten of 120 subjects were chronically infected with hepatitis B virus, and the prevalence of chronic HBV infection did not differ among 5 groups (P = 0.12). Only two subjects, both HIV-infected but one with cleared HCV infection and one with chronic HCV infection, had developed ascites, a manifestation of hepatic decompensation, at the time that their PBMCs were collected for this study. Table 1 Age and peripheral blood mononuclear cell subsets (median, interquartile range) by viral infection status. Viral Infection Status a Age and Cell Type HIV- HCV- HIV+ HCV+ HCV RNA- HIV+ HCV+ HCV RNA+ HIV- HCV+ HCV RNA- HIV- HCV+ HCV RNA+ P b (n = 24) (n = 24) (n = 24) (n = 24) (n = 24) Age (years) 24.4 (10.8–38.7) 26.7 (18.6–34.0) 25.6 (20.9–31.0) 22.5 (15.8–30.4) 30.0 (17.0–34.7) 0.80 CD4 + (%) 40.4 (35.5–49.2) 12.4 (2.7–27.0) 15.3 (2.3–24.4) 40.1 (34.4–45.7) 41.4 (36.2–45.8) <0.0001 CD8 + (%) 26.3 (24.4–33.0) 52.3 (42.6–59.5) 55.1 (49.2–62.6) 26.0 (20.4–32.8) 29.1 (22.3–37.6) <0.0001 CD19 + B (%) 15.5 (12.6–21.7) 15.4 (10.5–22.2) 11.4 (4.7–15.7) 17.8 (13.7–22.2) 13.3 (10.7–19.5) 0.04 CD56 + NK (%) 9.0 (6.8–12.9) 7.0 (3.7–9.7) 6.5 (4.1–13.8) 9.7 (6.7–13.5) 8.4 (6.5–12.4) 0.22 a HIV, human immunodeficiency virus; HCV, hepatitis C virus; RNA, presence or absence of HCV viremia, as detected by branched DNA assay (median HCV viral loads 4.6 × 10 6 and 2.4 × 10 6 IU/mL in HIV+ and HIV- subjects, respectively). b ANOVA test for comparison of age and Kruskal-Wallis test for comparisons of cell proportions. CD4 + T cells As expected, subjects with HIV had lower proportions of CD4 + T cells whether they had cleared HCV (12.4% vs. 40.1%) or chronic HCV (15.3% vs. 41.4%, P#0.0001; Table 1 ). Patients with cleared versus chronic HCV, however, had similar proportions of CD4 + T cells, regardless of HIV status (P∃0.82). CD8 + T cells CD8 + T cells differences mirrored those of CD4 + T cells (Table 1 ). Subjects with HIV had higher proportions of CD8 + T cells whether they had cleared HCV (52.3% vs. 26.0%) or chronic HCV (55.1% vs. 29.1%, P#0.0001). Regardless of HIV, subjects with cleared and with chronic HCV had similar proportions of CD8 + T cells (P∃0.30). CD19 + B cells Proportions of CD19 + B cells differed significantly among the five groups (Kruskal-Wallis P = 0.04; Table 1 ). In pairwise comparisons, CD19 + B cells were significantly lower with HIV and chronic HCV infections compared to uninfected subjects (11.4% vs.15.5%, P = 0.03). CD19 + B cells also were lower, albeit not significantly, with chronic HCV than with cleared HCV infection in both HIV-uninfected (13.3% vs. 17.8%, P = 0.09) and HIV-coinfected (11.4% vs. 15.4%, P = 0.08) groups. CD19 + B-cell levels did not differ between uninfected subjects and those with cleared HCV infection (P = 0.63). CD19 + B-cell proportions did not correlate with HCV viral load (Spearman R = 0.06, P = 0.78). CD56 + NK cells Proportions of CD56 + NK cells did not differ significantly among the five groups (Kruskal-Wallis P = 0.22; Table 1 ), suggesting that HCV infection status of hemophilia patients had limited effect on CD56 + NK cells. Discussion Among HIV-infected subjects, we found that CD19 + B-cell proportion was statistically significantly lower, from 15.5% to 11.4%, with chronic HCV infection, a fractional reduction of one-quarter [(15.5–11.4)/15.5 = 0.26]. We found subjects with chronic HCV without HIV co-infection had a reduction of the same magnitude, albeit of marginal statistical significance. HIV infection was more frequent with hemophilia A than with other coagulation disorders [ 6 ], but this did not confound the association of HCV chronicity with lower B-cell proportions. In addition, we observed an approximately one-quarter reduction (25%) in estimated absolute B-cell count in chronic HCV, despite higher variance (data not presented) [ 8 ]. Overall, our results corroborate and are of similar magnitude to those reported from Japan by Yokozaki et al [ 5 ]. Infection of B cells by HCV is controversial. Both B cells and hepatocytes express CD81, a possible receptor for HCV [ 9 ]. If HCV can replicate in B cells, as implied by detection of negative-strand RNA [ 10 , 11 ], then one might postulate HCV-related pathogenesis or apoptosis, as observed in B-cell lines [ 12 ]. B-lymphocytopenia also could occur during initial HCV infection and contribute to HCV chronicity by impairing neutralizing antibody response to HCV envelope 2 (E2) protein's hypervariable region-1 [ 13 ]. We found no differences in proportions of CD56 + NK cells with HCV infection, but this does not negate possible impairment of NK cell function with HCV infection. In vitro , binding of HCV E2 to CD81 inhibits the functions of cross-linked NK cells [ 14 , 15 ]. CD56 + NK cell proportion was reduced among our subjects who cleared HCV despite HIV co-infection, an observation that has no ready explanation and may have appeared by chance. Our study is limited by its cross-sectional design and analysis of PBMCs collected long after primary HCV and HIV infections had occurred. We cannot define sequential PBMC changes in relation to critical virologic events, such as spontaneous clearance of HCV. Furthermore, the cells that we detected in cryopreserved PBMCs may not be representative of those in fresh PBMCs. Still, we did observe the expected associations of CD4 + and CD8 + T cells with HIV infection. These findings, our careful flow cytometry methods, and our rigorous study design, with frequency matching on age and random selection of subjects from a large, well characterized cohort, probably offer a reliable "snapshot" of PBMCs in hemophilic patients infected with HCV alone or co-infected with HIV. Conclusions We found no association of cleared or chronic HCV infection with altered levels of CD4 + or CD8 + T cells, although this does not negate the likely functional importance of HCV-specific T-cell subpopulations in clearance of HCV [ 16 , 17 ]. Subjects with high level, chronic HCV viremia, but not those with cleared HCV infection, had reduced CD19 + B-cell levels. Further studies need to clarify the temporal sequence of B-cell changes with HCV chronicity and associated clinical conditions. Appendix: Collaborating investigators (and institutions) in the Multicenter Hemophilia Cohort Study M. Zhang, J.J. Goedert, T.R. O'Brien, P.S. Rosenberg, C.S. Rabkin, E.A. Engels, D. Whitby (National Cancer Institute, Rockville and Frederick MD); M.E. Eyster (Division of Hematology and Oncology, Pennsylvania State University Medical Center, Hershey PA); B. Konkle (Cardeza Foundation Hemophilia Center, Philadelphia PA); M. Manco-Johnson (Mountain States Regional Hemophilia and Thrombosis Program, University of Colorado, Aurora CO); D. DiMichele, M.W. Hilgartner (Hemophilia Treatment Center, New York Presbyterian Hospital, New York NY); P. Blatt (Christiana Hospital, Newark DE); L.M. Aledort, S. Seremetes (Hemophilia Center, Mount Sinai Medical Center, New York NY); K. Hoots (Gulf States Hemophilia Center, University of Texas at Houston, Houston TX); A.L. Angiolillo, N.L.C. Luban (Hemophilia Center, Children's Hospital National Medical Center, Washington DC); A.Cohen, C.S. Manno (Hemophilia Center, Children's Hospital of Philadelphia, Philadelphia PA); C. Leissinger (Tulane University Medical School, New Orleans LA); G.C. White II (Comprehensive Hemophilia Center, University of North Carolina, Chapel Hill NC); M.M. Lederman, S. Purvis, J. Salkowitz (Case Western Reserve University School of Medicine, Cleveland OH); C.M. Kessler (Georgetown University Medical Center, Washington DC); A. Karafoulidou, T. Mandalaki (Hemophilia Center, Second Regional Blood Transfusion Center, Laikon General Hospital, Athens, Greece); A. Hatzakis, G. Touloumi (National Retrovirus Reference Center, Athens University Medical School, Athens, Greece); W. Schramm, F. Rommel (Medizinische Klinik Innerstadt der Maximilian, Universitaet Muenchen, Munich, Germany); P. de Moerloose (Haemostasis Unit, Hôpital Cantonal Universitaire, Geneva, Switzerland); S. Eichinger (University of Vienna Medical School, Vienna, Austria); K.E. Sherman (University of Cincinnati Medical Center, Cincinnati OH); and B.L. Kroner (Research Triangle Institute, Rockville MD). Disclaimer "The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organization imply endorsement by the U.S. Government." Competing interests The author(s) declare that they have no competing interests. Authors' contributions MZ and JG conceived of the study and drafted the manuscript. MZ performed the random sampling and statistical analyses. JG provided advice on the sampling and statistical analyses, obtained funding, managed the parent cohort, and supervised the virologic testing. TOB provided advice on the design. WK provided advice on and performed the FACS analyses. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC546217.xml
545078	Visualization of comparative genomic analyses by BLAST score ratio	Background The first microbial genome sequence, Haemophilus influenzae , was published in 1995. Since then, more than 400 microbial genome sequences have been completed or commenced. This massive influx of data provides the opportunity to obtain biological insights through comparative genomics. However few tools are available for this scale of comparative analysis. Results The BLAST Score Ratio (BSR) approach, implemented in a Perl script, classifies all putative peptides within three genomes using a measure of similarity based on the ratio of BLAST scores. The output of the BSR analysis enables global visualization of the degree of proteome similarity between all three genomes. Additional output enables the genomic synteny (conserved gene order) between each genome pair to be assessed. Furthermore, we extend this synteny analysis by overlaying BSR data as a color dimension, enabling visualization of the degree of similarity of the peptides being compared. Conclusions Combining the degree of similarity, synteny and annotation will allow rapid identification of conserved genomic regions as well as a number of common genomic rearrangements such as insertions, deletions and inversions. The script and example visualizations are available at: .	Background In the decade since the publication of the Haemophilus influenzae genome sequence in 1995 [ 1 ], 191 microbial genomes have been completed, with another 276 in progress [ 2 ]; as of October 14, 2004). Multiple strains of the same organism, or multiple species of the same genus are being sequenced or have been completed, making comparative genomic analysis possible on an unprecedented scale. As the technology continues to improve, the number of completed microbial genome sequences will further increase – a major challenge of the comparative genomic era is to fully exploit this data. However, the development of tools for analysis of such data sets has not kept pace. BLAST analysis has become a ubiquitous method of interrogating new sequence data, but there are limitations to using BLAST alone as a discriminating tool. Many other methods and individuals use BLAST output E-values as a criterion for data parsing. While this measure may be efficient, the output is often skewed by both the database used for comparison and the length of the match [ 3 ]. Small regions of high similarity can generate an artificially low E-value and negate the global level of similarity exhibited by the sequence. This bias is eliminated when using the BLAST raw score as it is directly derived from the similarity of the match. However the value of the BLAST score varies with the length of the peptide queried, and hence is not suitable alone for comparative analysis using universal cutoffs [ 4 ]. Several other tools utilize the BLAST algorithms to compare nucleotide or peptide sequences from genome projects. The Wellcome Trust Sanger Institute ACT software [ 5 ] can display nucleotide similarity between two genomes based on BLASTN E-value. ACT builds upon Artemis and displays regions of high similarity mapped on the genome annotation. The GenomeComp tool [ 6 ] displays a similar analysis also based on BLASTN E-values to compare genome sequences. NCBI Taxplot, a three-way genome comparison tool based on precomputed protein BLASTP E-values displays a point for each protein in the Reference genome based on the best alignment with proteins in each of the two genomes being compared [ 7 ]. On the other hand, the SimiTri program utilizes BLASTP comparisons of three proteomes and uses the raw BLAST score, not E-values. However, only protein similarity data is represented and no information on the comparative structure of the genomes is provided [ 8 ]. Moreover, the SimiTri program does not address BLAST artifacts derived from the size of the database or the length of the match. This paper describes the BLAST score ratio (BSR) algorithm that enables comparative analysis of multiple proteomes, together with visualization of genome structure (synteny). BSR analysis is a departure from traditional genome scale analyses as it overcomes the limitations of BLAST E-values in comparative studies by normalizing the BLAST raw scores. BSR analysis is a tool for the rapid comparison of complete proteomes of any three genomes, and enables a visual evaluation of the overall degree of similarity of these proteomes and their genomic structure. Implementation We have implemented the BSR algorithm using Perl. The inputs are the predicted proteomes of each of the three genomes under analysis, formatted as multi-FASTA files. An additional file for each of the proteomes is required. This file must contain a unique identifier, matching the FASTA header of the corresponding peptide in the multi-FASTA file, the relative genomic location of the start and stop of the coding region as well as the annotation for each peptide. The user selects one proteome as the "Reference"; the other two proteomes are termed "Query 1 " and "Query 2 " respectively. Initiation of the script results in each of the putative peptides in the Reference proteome being compared to all of the other peptides in the Reference and Query proteomes using NCBI BLASTP. The BSR is then computed as follows. The BLAST raw score for each Reference peptide against itself is stored as the Reference score. Each Reference peptide is then compared to each peptide in the Query 1 and Query 2 proteomes with each best BLAST raw score recorded as Query 1 and Query 2 , respectively (Figure 1 ). The BSR is calculated by dividing the Query score by the Reference score for each Reference peptide (Figure 1A ). Thus, for each peptide in the Reference genome, two numbers are generated, one from each from the best matches in Query 1 and Query 2 , thus normalizing all scores in the range of 0 to 1. A score of 1 indicates a perfect match of the Reference peptide to a Query peptide and score of 0 indicates no BLAST match of the Reference peptide in the Query proteome. The BLAST raw score is used rather than the E-value for the BLASTP results as it more accurately accounts for the length of the similarity between the Reference and Query peptides [ 4 , 9 ]. This normalized pair of numbers can be plotted as coordinates in Cartesian space for each peptide in the Reference proteome, enabling the visualization of the entire Reference proteome in comparison to the two Query proteomes (Figure 1B ). Outputs Following calculation of the BSRs, a number of output files are generated, including both text and graphical formats. The text files are tab-delimited for ease of parsing; filenames are derived from the named proteome files used as input into the script. The R_Q1_Q2.txt (Reference_Query 1 _Query 2 .txt) output contains an ordered list including the Reference peptide unique identifier, annotation, and Reference BLAST raw score, in addition to the unique identifier of the best hits in the Query proteomes, corresponding BLAST raw scores and the calculated BSR. Additionally, four unique files are generated corresponding to the peptides within the four quadrants delineated in Figure 1B . The four quadrants are derived from a BSR threshold value of 0.4, which was empirically determined to represent approximately 30% amino acid identity over approximately 30% of the peptide length, a commonly used threshold for peptide similarity [ 10 ]. This threshold value can be adjusted using the "-C" option (see help file). The graphical output files are viewed with Gnuplot [ 11 ] to reveal the global similarity of the compared genomes as well as the level of conserved genome structure. PostScript and xfig [ 12 ] graphic files are subsequently generated by Gnuplot. The scatter or similarity plot provides an overall view of the level and number of similar and dissimilar proteins in the Reference proteome when compared to the Query proteomes (Figure 1C ). The regions of the graph are color-coded depending on the level of similarity between the three genomes (Figure 1B ). Quadrant A (BSR < 0.4), colored in orange, contains peptides unique to the Reference proteome with little similarity in either of the Query proteomes. Quadrant C (BSR > 0.4), colored Red, contains peptides that have significant similarity in all three compared proteomes. Quadrant B, colored green, contains Reference peptides with similarity to only Query proteome 2, whereas Quadrant D, colored blue, contains Reference peptides that have similarity to only Query proteome 1. Two additional plots, termed synteny plots, are generated, one for comparison of the Reference proteome to each Query proteome, by plotting the genomic location of the Reference peptide on the X-axis and the genomic location of the most similar Query peptide on the Y-axis. This plot alone would demonstrate the level of synteny (conservation of gene order) between the two genomes [ 13 ], however, an additional level of information is included by coloring each point based on the BSR (see legend Figure 2 ). The color provides an additional visual clue to the global level of similarities of the proteomes. For example genomes can be highly syntenic with relatively low levels of proteomic similarity as is shown in Figure 2A and 2B or they may have a high degree of protein similarity and conserved genome structure (Figure 2C ). The Gnuplot, PostScript and xfig outputs allow publication-quality, global visualization of the similarity and synteny of the selected genomes. However these formats do not allow the annotation associated with individual peptides to be viewed interactively. To overcome this limitation, additional XML files for the similarity and synteny plots described above are generated. These files are the input for the freely available GGobi software. GGobi is a data visualization system for viewing high-dimensional data [ 14 ]. The tools provided in the GGobi software package allow the annotation associated with individual points within the similarity and synteny plots to be viewed interactively (Figure 3 ). The GGobi package also allows the expansion of the BSR approach to include more than three genomes or other additional parameters associated with proteomic or genomic data, enabling interactive, user-driven exploration of these complex datasets. The current BSR implementation uses three genomes as input; however, additional genomes can readily be added as new dimensions simply by repeating the analysis with the same Reference genome and varying the Query genomes. Additional non-BSR dimensions are readily included, such as pI or %GC, or factors such as surface localization or some other feature of the peptides of interest. Results Genome structure is often altered during the evolution of species [ 13 ]. Visualization of this structure often lends insight into genome evolution and examination of the various BSR outputs rapidly reveal alterations of the genome structure as well as the overall similarity of the two Query proteomes to the Reference proteome. The genomes of the Order Chlamydiales (Figures 1 , 2 A and 2B ) provide an example of this insight. In Figure 1 a large proportion of the peptides are conserved, with 71.7% of the proteins shared between all three proteomes. If the Query proteomes are further used as the Reference proteome and vice versa we still see a similar trend (data not shown). Additionally, the proteome of C. pneumoniae AR39 (GenBank Accession Number AE002161) is more similar to C. caviae GPIC (GenBank Accession Number AE015925) than C. muridarum strain Nigg (GenBank Accession Number AE002160) as 7.3 % of the proteome is shared between only C. caviae GPIC and C. pneumoniae AR39 compared to only 1.6% between C. caviae GPIC and C. muridarum strain Nigg. Finally, Figure 1 demonstrates that 19.4% of the C. caviae proteome has no significant hit to any of the peptides in the Query proteomes, although many of these peptides (78.2%) are currently annotated as hypothetical. From the analysis in Figure 1 we could conclude that the chlamydial proteomes are extremely similar and suggest that the genome structure will also be similar. However, the synteny plots in Figure 2A and 2B demonstrate that while the chlamydial proteomes exhibit a high degree of similarity, there is significant alteration in the genomic structure. The comparison of the proteomically similar organisms, C. caviae GPIC and C. pneumoniae AR39 reveals that the genomes contain two points of inversion (arrows in Figure 2A ). One of these points of inversion is centered on the terminus of replication. There are more extensive genomic rearrangements between the C. caviae GPIC and C. muridarum strain Nigg genomes (Figure 2B ). The additional color information extends the utility of these synteny plots. While the chlamydial genomes show regions of conserved synteny, as demonstrated by the peptides in the same genomic location forming a line with a slope of 1 or -1, the absolute degree of similarity between the peptides, demonstrated by color indicates divergence. By contrast the synteny plot of two Escherichia coli genomes (Figure 2C ) demonstrates a high level of synteny with a number of unique insertions, however no inversions are present. Moreover the color dimension on this plot reveals that unlike the chlamydial proteome comparisons the E. coli proteomes have a high level of similarity and synteny. In the analysis of the Chlamydial proteomes using BSR score and BLAST E-values approximately 1% of peptides examined have a BSR score > 0.4 and BLAST E-value > 1 × 10 -15 . These peptides were all very small in size (< 70 amino acids) and greater than 50% amino acid identity. This group of peptides is more readily identified by BSR analysis than BLAST E-value, which is artificially low due to the small peptide size. Additionally, peptides that have a BSR score < 0.4 but a BLAST E-value < 1 × 10 -15 correspond 7.8% of the proteome. These represent divergent peptides with an artificially high BLAST E-value score resulting from limited regions of identity. The BSR analysis more accurately classifies these peptides based on the amino acid identity over the entire peptide. As the BSR comparison utilizes a single genome as a reference, the BSR score is calculated using a unidirectional best BLAST hit. However, when the Chlamydial proteomes were compared only one case in over 1000 could be found with a BSR score > 0.4 that was not also a bidirectional best BLAST hit. Conclusions The BSR approach allows rapid evaluation of the level of conservation of any three proteomes and the degree to which the genome structure between the three genomes is similar. While in this report we discuss the applications of this approach to whole genomes, the analysis has been performed on portions of genomes such as genomic or pathogenicity islands, plasmids and phage to identify peptide similarity and regional structure. More genome sequences are being generated from closely related organisms – a trend which shows no sign of abating. The BSR approach has become a crucial tool in our comparative genomics armamentarium and has been utilized in a number of genomic comparisons, revealing regions of similarity and difference between both closely and distantly related organisms [ 10 , 15 , 16 ]. Availability and requirements Project name: BSR.pl Project homepage: Operating System: Unix and MacOS X Programming language: Perl Other requirements: Perl Statistics::Descriptive module License: None Any restrictions to use by non-academics: None List of abbreviations BSR – BLAST score ratio; BLAST – basic local alignment search tool. Authors' contributions DAR, GSAM and JR conceived and implemented the first versions of BSR and prepared the manuscript. All authors have read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545078.xml
514545	Publication bias in situ	Background Publication bias, as typically defined, refers to the decreased likelihood of studies' results being published when they are near the null, not statistically significant, or otherwise "less interesting." But choices about how to analyze the data and which results to report create a publication bias within the published results, a bias I label "publication bias in situ " (PBIS). Discussion PBIS may create much greater bias in the literature than traditionally defined publication bias (the failure to publish any result from a study). The causes of PBIS are well known, consisting of various decisions about reporting that are influenced by the data. But its impact is not generally appreciated, and very little attention is devoted to it. What attention there is consists largely of rules for statistical analysis that are impractical and do not actually reduce the bias in reported estimates. PBIS cannot be reduced by statistical tools because it is not fundamentally a problem of statistics, but rather of non-statistical choices and plain language interpretations. PBIS should be recognized as a phenomenon worthy of study – it is extremely common and probably has a huge impact on results reported in the literature – and there should be greater systematic efforts to identify and reduce it. The paper presents examples, including results of a recent HIV vaccine trial, that show how easily PBIS can have a large impact on reported results, as well as how there can be no simple answer to it. Summary PBIS is a major problem, worthy of substantially more attention than it receives. There are ways to reduce the bias, but they are very seldom employed because they are largely unrecognized.	Background A study is more likely to appear in the literature, and thus be indexed and publicly available, if it shows a strong, statistically significant, or otherwise "better" result. A study that does not show such results has a greater chance of remaining hidden in a file drawer, either because the author (or funder) does not think the result is worth mentioning, or because journals are less interested in publishing studies that find "nothing". This form of publication bias, wherein readers have access to only a biased sample of the full distribution of results, is well studied, particularly in the literature on meta-analysis and other systematic reviews where the problem is most apparent (a primer on the topic can be found at the Cochrane Collaboration webpage [ 1 ]). Even well-read lay people are, at the time of this publication, aware of the problem due to the controversy about pharmaceutical companies selectively releasing research about their products. Far less attention is paid to the bias that occurs when some results from a study are published, but the choice of which results to publish produces bias. As with traditional publication bias, the tendency is to analyze data and choose to present results which are statistically significant, further from the null, or closer to what the researchers believe is the true value. The implications for the literature as a whole, though, are much the same as the file drawer bias. I label this "publication bias in situ " (PBIS) because the biased reporting of study findings exists within each individual research report (with any metaphoric references to cancer – the usual context of the phrase " in situ " in health science – left as an exercise for the reader). Despite the similarities between the file drawer bias and PBIS, there are fundamental differences. In particular, the bias from some studies having no published findings exists only at the level of the whole literature (no particular study can be said to be biased), while PBIS exists within the results reported from a single study (and thus exists in the literature as a whole by aggregation). More practically, PBIS is substantially more difficult to even identify, let alone correct. This paper defines and describes PBIS and identifies some of the choices that create it. The purpose of the paper is not to make novel technical or statistical claims. Indeed, there is probably no single statistical observation here that will not be clear to a skilled data analyst (or, indeed, that could not be explained to anyone competent in middle-school-level math). The literature includes many observations about issues relating to PBIS. Yet the challenge to the validity of the entire health science literature posed by PBIS – arguably a greater issue than conflict of interest, traditional publication bias, or any other commonly discussed threat to the integrity of the literature – has not received the attention it deserves. In the literature about meta-analysis and systematic reviews there is substantial attention to file-drawer publication bias (though most of this considers only bias from reporting statistically significant results, ignoring preferences for reporting more dramatic results or those that agree with authors' or journals' previous reports). But there is considerably less attention to possible picking and choosing which study results to report or statistical methods to use. To the extent that this is considered, it is often bundled with questions about whether ex ante protocols were stated and followed, and is thus put on par with many protocol violations that could be considered mere technicalities. For example, a few years ago, a highly-publicized meta-analysis [ 2 ] called into question the evidence on the benefits of screening mammography by citing apparent problems in the methods of most of the relevant randomized trials. But little was done to distinguish a few of these problems that appeared to be PBIS and others that were minor technical points. The literature and pedagogy related to observational inference, where the potential for PBIS is considerably greater than for well-designed experiments, seems to pay even less attention. Epidemiology textbooks typically discuss some of the methods that can reduce PBIS (e.g., well-defined protocols), but say little or nothing about the possible bias. Indeed, many study and data-analysis methodologies (some of which are noted below) that are typically taught in classes or by apprenticeship seem designed to create PBIS. Despite this, highly-trained experts summarize claims reported in the literature without mention of the likely bias, indicating an unawareness of the major implications of PBIS. In a notable exception, Hahn and colleagues [ 3 , 4 ] discussed some sources of PBIS and argued that they receive too little attention compared to bias from selective publication. Their analyses addressed selective reporting of subgroup results in the context of randomized trials, a topic further discussed below, and reporting a selected subset of multiple measured endpoints. They do not mention the other sources of PBIS discussed below. Unfortunately, their findings do not appear to have had the impact they deserved. It is not clear how common PBIS is or how large the resulting bias, but a few efforts to find it suggest that the potential is great enough that it deserves much more attention. (Labeling might matter: Hahn et al. label the problem "within-study selective reporting." The more dramatic name suggested here, with its emphasis on the bias that results, might catch more attention.) Only by systematically addressing the problem are we likely to substantially affect it. Moreover, as will be discussed, statistical and research methods that ostensibly address some of the sources of PBIS are unsatisfactory, and a systematic attempt should be made to find better solutions. Discussion Many degrees of freedom Reviewers (systematic or otherwise) of the literature can only see what researchers choose to report and highlight in their publications, and that choice can be biased in a number of ways. Researchers have great freedom in deciding exactly what to analyze, how to analyze it, and what to report. All research results are derived from data that can be used to measure many associations. Even the most narrowly focused clinical trial can be analyzed with the endpoint defined in different ways, stratified by age, etc. As with traditionally defined publication bias, the analyses that are deemed unworthy of publication are largely invisible to the scientific and policy community. Among the dimensions of freedom researchers have in deciding what to analyze and report are three choices illustrated by examples in this paper: (1) Which exposures and outcomes to consider in datasets with many variables. (2) Which functional forms to use to represent variables (e.g., how to divide continuous variables into categories). (3) Whether to conduct separate analyses by subgroup, and which subgroup results to emphasize. Making such choices is a legitimate part of research. Indeed, the choices must be made. But when those choices are primarily driven by what produces stronger (or otherwise "better") results, bias is created. This creates a difficult challenge: It is easy to recognize traditional publication bias (paper in journal = no contribution to bias; paper in file cabinet = contribution to bias). But since there is no clearly correct option for any of the above choices (indeed, any particular analysis gives the right answer to some question), there is no clearly wrong choice, and thus no clear way of concluding that a particular choice was biased. Fortunately, as should become apparent from the following analysis, PBIS results less from the choices made and more from what the choices are based on (which can often be inferred) and, to an even greater extent, the generally overlooked issue of how the results are presented (which can be easily observed). Publication bias, either PBIS or the file-drawer effect, can be seen most clearly as an interaction between random errors and researcher choices (e.g., when random sampling error leads to a weaker result, the result is less likely to be reported), creating a bias from what would be unbiased random variation. Systematic errors (confounding, measurement error, etc.) and methods for correcting for them create many additional opportunities for PBIS; however, for simplicity, systematic errors are ignored in this paper. Multiple analyses from the same data Choice (1) in the above list has probably received the most attention in previous literature (for example, debates over whether to correct for multiple hypothesis testing and the appropriateness of data dredging). Despite the disproportionate attention, this choice is probably not the major source of PBIS, but it provides a familiar starting point. Many epidemiologic datasets are characterized by thousands or even millions of possible combinations of exposures, endpoints, and covariates. It is frequently assumed that statistical science tells us the "right" way to deal with this challenge, but current practice (not to mention the confusion of students coming out of epidemiology and biostatistics classes) makes clear that there is substantial disagreement among viewpoints about how to apply statistical methods when dealing with multiple hypotheses or measurements. Further consideration makes it clear that statistical rules cannot actually provide clear answers. At one extreme are viewpoints such as, "We must correct measures of statistical certainty (significance levels for p-values (α-levels) or confidence interval widths) whenever multiple comparisons are made using the same dataset" and even, "statistical analysis can only be legitimate for a short list of pre-specified hypotheses." At the other extreme are viewpoints such as, "regardless of how many comparisons are examined, each can be considered and statistically tested as if it were the only one," and, "it does not matter at all if a hypothesis was proposed after looking at the data." The impasse in this debate seems to stem from both sides attacking straw men, without recognizing that each side has a stronger case some of the time. This can be illustrated with examples. Example: unrelated results from the same dataset A cohort dataset originally used to report the relationship of drinking water source and the occurrence of Helicobactor pylori infection contains data that is later used to look at the relationship of household crowding and performance in school. It is difficult to understand why we would make an adjustment when doing the second analysis because we have already done the first (or, worse, disallow the analysis because it was not pre-specified in the study or because we have already "used up" our .05 worth of α with the H. pylori analysis, and so cannot analyze anything more with this data at all). A logical extension of that argument would be to consider the dataset that contains all quantitative human knowledge (which is logically an epistemologically legitimate definition), and declare that we have to adjust for every statistical analysis ever done, effectively precluding further statistical analysis. Example: multiple comparisons that will support claims of the "same" relationship Researchers investigate the hypothesis that poor nutrition increases the risk of H. pylori infection. The dataset contains dozens of different measures of food and nutrient intake, as is usually the case for nutrition data. This, plus multiple diagnostic tests for H. pylori which are sometimes discordant, creates a large number of statistical comparisons, any of which could be described as supporting the plain language claim, "poor nutrition affects H. pylori status." A typical approach is to find individual comparisons that support the hypothesis, presenting only these comparisons with statistical tests as if each were the only analysis conducted. The claim about the relationship between the particular measure of nutritional status and the particular measure of H. pylori status is accurate, as are the test statistics reported for that association. But the plain language conclusion (which would probably be drawn) was very likely to be supported by some relationship in the data by chance alone, even in the absence of any true underlying association. This fact is obscured by the reported unadjusted tests statistics or confidence intervals. As the second example illustrates, unrestricted picking and choosing of comparisons leads to publication bias. A lot of associations that were not deemed worthy of reporting never appeared in the literature, while the few that were "interesting" did. This problem is well known (though few probably realize that it can lead to hundreds of instances of publication bias, in situ within a single published article, making it a bias of much greater magnitude than the file-drawer effect). The solutions offered by statistical rules – corrections for multiple hypothesis testing or restricting analysis to ex ante hypotheses – is inadequate. Such rules produce absurd implications, noted in the first example. Trying to eliminate the absurdity by exempting from statistical adjustment analyses with disjoint exposures and outcomes, as in the first example, does not work; the second example offers options for disjoint analyses also. Most important (and widely overlooked), correcting for multiple comparisons does not affect the reported biased estimates of effect size; changing test statistics and confidence interval widths does nothing to reduce the bias . This alone shows that the standard statistical corrections for multiple tests do nothing to solve the problem. The other standard method for trying to reduce PBIS, rules that limit analyses to pre-specified hypotheses and protocols, will throw away a lot of potentially valuable findings and is nearly impossible to operationalize because detailed protocols require advanced knowledge that may not exist and can never be specified unambiguously. Frequentist statistical theory cannot offer a solution to this problem because PBIS, like the file-drawer problem, is not a matter of statistics. The second example illustrates where the problem primarily lies: in the plain-language reporting of results. The statistics that describe the relationship between a particular exposure and outcome measure could be exactly right, but the claim about good nutrition (as a generic concept) and infection status (as if we had a gold standard measure), which will likely be emphasized in the paper and its title (and press releases) and will likely stick readers' minds, might be misleading. Consider how the result would be interpreted if there were a table reporting every analyzed comparison, most of which showed little or no association. Most scientifically literate readers would realize the result was not so convincing, even though those same readers seldom think to object when – as is typical – only one or a few results are reported. By contrast, if the researchers in the first example reported the result of the previous study, it would be unlikely to change most readers' assessments. This suggests the simplest partial solution to the problem. By reporting a table of results from other comparisons considered, researchers could report their interesting result (rather than not informing the world due to the lack of a specific ex ante hypothesis or having "used up" the α), but without creating the PBIS that would result otherwise. Indeed, this appears to summarize the most obvious generic rule to reduce PBIS (and publication bias in general): publish everything. An immediate implication of this is that online publications, like this journal, allow researchers to publish less biased articles. Online articles can usually be whatever length is appropriate to report the results (which is of particular value in the health sciences, where paper journals have extremely restrictive length limits), and can include dozens or thousands of alternative analyses in appendices or links to data or software that allow the reader to review still more results. Of course, this opportunity is beneficial only if authors choose to take advantage of it (or editors and reviewers demand that they do). Bias from the choice of functional form The implications of choice (2), the functional form for variables, can be clearly illustrated with a simple example. Consider a study with an exposure variable measured as 10 ordered categories (i.e., values 1,2,...,10, with larger numbers representing greater exposure). Assume researchers wish to analyze the association of a disease endpoint and a dichotomous definition of exposure. If there is no clear cutpoint for defining exposure, there are many options. There are 9 cutpoints that divide the observations into two categories, defining those above the cutpoint to be exposed and those below unexposed. Other options include comparing a group of highest categories to a group of lowest categories, leaving out the middle, such as 8–10 versus 1–3, yielding an additional 36 possibilities. How will the researchers choose a definition of exposure? A typical procedure is to let the data inform the choice: The cutpoint that provides the clearest contrast between the exposed and unexposed is judged to be the right one, the "most sensitive" to the presumed effect. It should be immediately obvious that this procedure will bias the result away from the null. To illustrate this, consider a case-control study with 200 subjects (throughout the examples, subjects are half cases and half non-cases). Calculations for this example and others are based on Monte Carlo simulation of different realizations of the data based on the assumed underlying relationship. All simulations were performed using Crystal Ball (Decisioneering Inc., Denver, Colorado, USA). Assume that each of the 10 exposure categories is equally likely for cases and non-cases. If the researchers consider only the 9 cutpoints that dichotomize the data and choose the cutpoint for "exposed" to get the largest odds ratio (OR), the median result will be 1.5. Since the exposure and disease are not associated, this is clearly an upward bias. For those inclined to focus on statistical significance, the chance of observing a significant positive association at a one-tailed significance level of .025 is 13%. (Of course, for any single definition of exposure, the median OR is 1.0 and the chance of seeing a significant relationship is about 2.5%.) Even if researchers do not analyze their data in every possible way and report the strongest association, any decision to report results that is based on associations in the data (e.g., choosing between a cutpoint at 5 or at 6 based on which produces a stronger association) will create bias. Although this observation should be obvious to anyone with an understanding of statistics, letting the data have some influence on the choice is probably more the rule than the exception among researchers. It is often defended on the grounds that there was no way to know what the "right" cutpoint was before doing the study, and the study data is the only existing answer to the question. This is a legitimate point, but it does not reduce the resulting bias. Less scrupulous researchers – who are trying to support a preferred answer to further a policy agenda or advance their careers – need make no such explanation and can intentionally choose the extreme results. As with the file-drawer effect, results in the literature will tend to show effects greater than the true value. Most important, whichever analysis is reported, the plain-language result will be "we found an association between the exposure and the disease," and so collections of studies that each report an exposure-disease comparison with a greater-than-average association, and will seem to be confirming the same result, even though the comparisons are not the same. Extending this example to illustrate how PBIS can compare to traditional publication bias, assume now that there is a positive association between the exposure and disease. In particular, non-cases are still equally likely to be in each of the ten categories, while cases have respective probabilities for each category of (.069, .072, .077, .084, .093, .102, .112, .121, .131, and .139). These values were chosen so that the true OR is similar, whichever of the 9 cutpoints is chosen (for those interested, the numbers follow a logistic curve). True ORs round to 1.5 for all cutpoints. Consider a collection of studies of varying sizes, with fewer larger studies, as we would typically see in the literature, specifically 100 randomly generated studies (more than would likely exist, but better to illustrate) of random size (drawn from a triangular distribution with modal probability at a minimum value of 100 subjects, diminishing linearly to a maximum of 1600 subjects). Note that to avoid committing the very type of transgression discussed in this paper – repeating an analysis until "good" results are found – the reported results are from the first and only run of the simulation. For a single definition of "exposed" (values >5), a typical result appears in Figure 1 in the form of a funnel plot of study results vs. study size [ 1 , 5 , 6 ]. The results that are statistically significantly different from 1.0 at the two-tailed .05 level are represented by solid dots. The other results (represented by open circles) might never be reported – the simplest form of publication bias – though they could be inferred from the asymmetry of the distribution that would occur if only the significant results were published. Notice that the distribution for all the studies is unbiased and would lead to an estimate very close to 1.5, while a naive summary estimate based only on the statistically significant studies (possibly the only ones published) could almost double the estimated effect size. Figure 1 Traditional publication bias from simulated studies. Simulated results from case control studies with varying populations (half cases, half non-cases) for true odds ratio of 1.5. Solid circles represent statistically significant results at the 2-tailed, .05 level. (Note: x-axis scale chosen for compatibility with other figures.) Compare this to the results for the same 100 studies where the cutpoint is chosen based on the largest OR (Figure 2 ). The distribution is also substantially biased, with PBIS leading to results above the single-definition results of 1.5. A summary estimate of effect size would turn out to have substantially greater bias than would reporting only significant results as in Figure 1 . Notice that though there is a skew, it is much harder to discern a pattern like the asymmetry in Figure 1 that would show a systematic reviewer that the literature is biased. Figure 2 Publication bias in situ from simulated studies. Simulated results from case control studies with varying populations (half cases, half non-cases) for true odds ratio of 1.5. For each simulated study, the largest odds ratio (choosing among 9 different cutpoints for exposure definition) is reported. (Note: one outlier odds ratio estimate not shown.) Unbiased random errors, when combined with picking and choosing functional forms, lead to biases in reported results. A solution to this problem is much less obvious than its existence. The commonly proposed solution of only reporting results for pre-specified functional forms is not satisfactory, because it is difficult to enforce (most every pre-specification has some room for interpretation in retrospect; intentional cheating is difficult to detect; there may be little basis for selecting any particular pre-specified functional form) and it forces us to ignore real unpredicted findings. Sticking to pre-specified analyses is especially unrealistic in research studies that collect data on many different risk factors and outcomes. Simply labeling all results that were not pre-specified with the caveat, "hypothesis generating," accomplishes nothing. If such results were actually treated as not yet "real", the problems of determining exactly which results were pre-specified and the loss of important serendipitous findings are reintroduced. Of course, results with the caveat are very seldom treated as less real than any others in the literature. Moreover, the "generated" hypotheses will never be retested in exactly the form reported, so the label is simply disingenuous. In sum, the proposed solutions to this type of PBIS are no more realistic or satisfactory as solutions than trying to eliminate traditional publication bias by requiring that all studies be adequately powered. A better family of solutions would be to establish a standard of reporting results for alternative variable definitions (perhaps in online appendices). Not only does this directly reduce PBIS by publishing more results, but it provides readers with a choice of results if they prefer different definitions (information that would be discarded by a pre-specified hypothesis rule). If results for every cutpoint from the 100 trials in the example were reported, the results, as pictured in Figure 3 , would be unbiased. Naturally, researchers could emphasize the variable definitions they think best, but by acknowledging other possibilities they would be forced to justify their choice. Figure 3 Publication bias in situ eliminated by reporting all results. Results from Figure 2, but with results for all possible cutpoints reported. In many cases there will not be an obvious short list of variable definitions, but some alternative definitions should be obvious and others could be found in previous literature. A simple, but very useful, improvement would be a standard practice of reporting the closest possible replication of previous published analyses using data from the new study. This would directly address the problem of PBIS resulting from data-driven picking and choosing of functional forms (though it might require the cooperation of previous authors to provide details about what analysis they reported, given the typically abbreviated reporting of methods in published papers – another problem with paltry word limits). Repeating whatever analyses that previous researchers happened to choose is somewhat arbitrary, but each round of new research can also add a new preferred functional form. The key is that results based on previously published functional forms cannot be data driven and, unlike the standard practice of new studies that make different comparisons but describe them with the same plain language, would actually replicate (or fail to replicate) previous results. Bias from the analysis of subgroups In 2003, VaxGen Corporation (Brisbane, California, USA) released results of a large HIV vaccine trial in the United States, one of the highest-profile clinical trials of the year. The disappointing result showed a trivial reduction in incidence among the treatment group compared to the placebo group. But the three non-white racial groups (black, Asian, and "other") each showed a substantial reduction in incidence (Table 1 ). VaxGen reported the overall failure of the trial to the popular and business press [ 7 , 8 ]. A technical report describing the drug and the trial results, written by a third party, appeared later in an indexed journal [ 9 ], though the New York Times articles actually contained more complete study results. VaxGen tried to salvage some hope for the drug by pointing out the results for non-whites, suggesting that maybe it held promise for some populations [ 8 ]. VaxGen's search for a silver lining resulted in rash of criticism from the research community (focused on the reporting of a result that was not a pre-specified hypothesis and the failure to correct for multiple hypothesis testing) and a shareholder lawsuit, alleging that statistically illicit reporting was used to inflate stock prices [ 9 - 12 ]. Table 1 Results of VaxGen HIV vaccine trial total vaccine placebo RR subjects subjects incidence subjects incidence white 4511 3003 179 1508 81 1.11 nonwhite 498 327 12 171 17 0.37 black 314 203 4 111 9 0.24 Asian 73 53 2 20 2 0.38 other 111 71 6 40 6 0.56 total 5009 3330 191 1679 98 0.98 Source: The New York Times [7] and author's calculations. Given the failure to publish the results in a scientific journal, some might argue that VaxGen was guilty of traditional publication bias – not publishing unfavorable results about its products – a charge that is currently being leveled at many drug companies. However, the company actively released fairly complete study results and accompanying analysis to the press, and the results appeared in forums that are more widely read than all scientific journals combined, so the results were clearly not buried in a file drawer. But despite reporting their results, VaxGen was biasing what they published. Singling out of the result for non-whites is a clear case of PBIS. VaxGen did not give equal emphasis to the apparently harmful effect of the vaccine for whites. The company's subsequent report that the different results for whites and non-whites could not be due to chance [ 9 ] does not diminish the apparent bias, since it basically repeats the same information contained in the original (data-driven, biased) reporting of the result for the non-white subgroup. Setting aside some controversy that erupted about systematic errors in the data, what can we say about the results from the perspective of PBIS and chance alone? To look at the result for non-whites, with its one-tailed p-value of .002, ignoring the fact that the subgroup definition was clearly data-driven, would overstate the finding, as suggested in the preceding examples. But a naive correction for multiple hypothesis testing would make the opposite error. Setting aside the possibility that other covariates would have been used to stratify the data had they produced subgroups with positive findings, the combination of the 4 racial groups implies a test of 2 4 -1 = 15 different hypotheses. To adjust for 15 implicit hypotheses makes it very unlikely that any will pass the statistical test, including the population as a whole, even for substantial associations. An alternative is to ask the question, "if the vaccine has no effect, what are the chances of seeing, in any racial group or combination thereof, a result at least as strong as the observed 63% reduction?". Phrased that broadly, simulation shows the answer is about 20%. However, most of the 20% comes from the unstable results for the two smallest groups, Asians and "other". Restricting the analysis to combinations of racial groups that contain black, with or without Asians or other, the probability is only 2.1% (the probability of seeing a 63% reduction by chance alone for any group that includes the whites is vanishingly small). So, what is the right answer? We must return to the observation that there is never a single Right Answer from a study; the quality of an answer always depends on what question was being asked. Did VaxGen find a successful vaccine? Clearly not, as the relative risk for the whole population shows. Should the result for non-whites be considered unlikely to be due to chance (i.e., statistically significant)? It depends on whether you consider it the answer to the question "does the vaccine show a result for non-whites?", in which case the answer is 'yes' (though the effective study size is small), or "does the vaccine show a result for any racial group?" in which case the answer is 'it is fairly likely we would see such a result due to chance alone.' It is worth noting how this illustrates a popular fallacy in data analysis: Frequentist hypothesis testing is not the objective exercise that some think it to be; it depends on subjective decisions about what to test. We might decide to infer that VaxGen would have emphasized the results from any racial subgroup that showed a positive result (and the company did claim the original protocol called for analysis by racial and other subgroups [ 9 ]), and thus that they were answering the latter question. Notice that none of the options that are typically practiced or recommended are satisfying. To just report the subgroup analysis as if it were the only analyzed result obviously leads to bias. (It is worth noting that in a less high-profile research project, that might well have happened, without anyone questioning the result.) But it is not satisfying to suppress the tantalizing findings about non-whites, either because there was not really an ex ante hypothesis that the vaccine would work only in non-whites or because the multiple-hypothesis correction for hundreds of possible racial and other subgroups makes it non-significant. A general rule requiring us to ignore interesting but surprising findings is a huge waste of information. Requiring a data-driven subgroup analysis to be biologically plausible before reporting it offers no solution, since we can usually construct a story to explain whatever associations appear in the data (it has been speculated that some genotypes get a benefit from the vaccine, and the frequency of those genotypes is strongly correlated with race). To offer the "hypothesis generating" caveat would make little difference, scientifically or in the securities market. It is unrealistic to suggest that this "generated" hypothesis will be re-examined given the overall disappointing result. Two studies in Thailand (one completed later [ 13 ], which also found the vaccine ineffective, and another by the U.S. National Institute of Allergy and Infectious Disease that may continue to use the vaccine anyway [ 14 ]) are likely the closest anyone will come to re-examining the hypothesis, but a population of Thais is hardly the same as non-white Americans. Furthermore, this example shows how epistemologically absurd the hypothesis generating caveat is: The result could originally have been considered hypothesis generating. But a few days after the results were released the company claimed that they had an ex ante plan to analyze racial and other subgroups [ 9 ], which would presumably promote the result "hypothesis confirming". However, that claim by the company, accurate or not, was completely uninformative about the effect of the vaccine, telling us nothing about the certainty of the findings, and so cannot legitimately change our conclusions. It does not matter whether the hypothesis was pre-specified. Debating whether the company really proposed the subgroup analysis ex ante , as if that should change our interpretation of the result, seems particularly absurd. When Hahn et al. [ 3 ] observed apparent selective reporting of subgroup analyses, they suggested identifying subgroups in the protocol, keeping that list as short as possible, and implicitly called for reporting results for all pre-specified subgroups. But since every measured covariate creates two, several, or a continuum of possible subgroups, this approach would require ignoring a lot of the results of a study, no matter how interesting they are (as well as severely taxing the imagination of the researchers about which subgroups are the right ones). Since it is unrealistic to expect researchers to not report interesting results (let alone to not even do the analysis that would produce those results) after spending months or years gathering data, we need methods that allow the reporting of results but with less bias. The obvious general solution is to report all subgroup analyses with equal prominence. Any reporting (be that a research paper, abstract, paper title, or press release) that suggests there is a beneficial effect for some people should equally emphasize any apparent harmful effects for other people (and vice versa). The fact that one result is statistically significant and the other is not should be of no consequence. Indeed, selecting which results to report based on statistical significance guarantees there will be publication bias (and, more generally, the inappropriate emphasis on statistical significance may be the source of a large amount of PBIS, but this point must be left for future analyses). The reporting of the multiple subgroups results should be accompanied by statistics similar to those calculated here, instead of standard test statistics, so that readers know the probability that an estimated effect (or test statistic) at least as great as the one found would result from chance alone for any of the subgroup analyses. Such information will allow readers (researchers working on related projects, policy makers, investors) to focus on what they consider to be the answer to their own questions. Summary The opportunities for PBIS, along with the almost universal failure to report research results in ways that avoid it, create the possibility that biased study results are very prevalent in the health science literature. Some of the causes of PBIS are well understood, but the enormity of its implications is largely ignored. PBIS can produce very misleading results, leading to widespread misperceptions and misguided policies. The examples presented here show just a few of the many ways that PBIS can result from random error and researchers' (usually innocent, almost always invisible, possibly quite reasonable) choices. Neither the problem nor the solution lies in the mathematics of data analysis, so answers will not be found by appealing to statistical theory. The critical issue is the completeness of reporting and the plain-language interpretation of results. The simple solutions offered by statisticians are not satisfying or even realistic. All they really let us do is observe that in almost every research report, "the rules" have been violated. This is not helpful. Rather than a right-vs.-wrong view of proper use of statistics that would condemn most of the literature as invalid, we need a realistic way of addressing this problem. The solutions, like the solutions for traditional publication bias, will generally consist of doing a more complete job of reporting what can be reported. List of abbreviations HIV = human immunodeficiency virus OR = odds ratio PBIS = publication bias in situ Competing interests None declared Authors' contributions Single author Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC514545.xml
524492	Cytotoxicity of psammaplin A from a two-sponge association may correlate with the inhibition of DNA replication	Background SV40 DNA replication system is a very useful tool to understand the mechanism of replication, which is a tightly regulated process. Many environmental and cellular factors can induce cell cycle arrest or apoptosis by inhibiting DNA replication. In the course of our search for bioactive metabolites from the marine sponges, psammaplin A was found to have some anticancer properties, the possible mechanism of which was studied. Methods Cell viability was determined by Cell Counting Kit-8 (CCK-8) to count living RAW264.7 cells by combining 2-(2-methoxy-4-nitrophenyl)-3-(4-nitrophenyl)-5-(2,4-disulfophenyl)-2H-tetrazolium (WST-8) and 1-methoxy-phenazine methosulfate (1-methoxy-PMS). The effect of psammaplin A on DNA replication was carried out in SV40 DNA replication system in vitro . The activities of topoisomerase I and polymerase α-primase were measured by the relaxation of superhelical plasmid DNA and the incorporation of [ 3 H]dTTP to the template respectively. The ssDNA binding activity of RPA was assessed by Gel Mobility Shift Assay (GMSA). Results We have found that psammaplin A delivers significant cytotoxic activity against the RAW264.7 cell line. It was also found that psammaplin A could substantially inhibit SV40 DNA replication in vitro , in which polymerase α-primase is one of its main targets. Conclusion Taken together, we suggest that psammaplin A-induced cytotoxicity may correlate with its inhibition on DNA replication. Psammaplin A has the potential to be developed as an anticancer drug.	Background DNA replication in eukaryotic cells is a tightly regulated process [ 1 ]. The regulation of DNA replication is central to understanding the regulation of cell cycle and virus proliferation, events that have a direct impact on our understanding human disease. One critical component of cell cycle regulation is the initiation of DNA replication. The timing of initiation is precisely controlled and is sensitive to both environmental and cellular factors. If DNA replication is blocked by inhibitors or the template is damaged by radiation or other factors, signals are generated that can induce cell cycle arrest or apoptosis [ 2 , 3 ]. Much of what is currently known about the mechanism of DNA replication in eukaryotic cells has come from studying SV40 and related viruses. SV40 virus can use the host replication machinery for its own DNA replication together with the virally encoded SV40 T-antigen. SV40 T-Ag is a multifunctional regulatory protein with numerous biochemical activities, and it has been classified as a member of superfamily III helicase and can unwind dsDNA and RNA [ 4 , 5 ]. All other proteins are supplied by host cells. In replication, replication protein A (RPA) mediates unwinding of SV40 origin-containing DNA in the presence of SV40 T-Ag and the DNA polymerase α-primase complex (pol α-primase) [ 6 , 7 ], which is necessary for the initiation of SV40 DNA replication [ 8 , 9 ]. Psammaplin A is a symmetrical bromotyrosine-derived disulfide dimer that was originally isolated in 1987 from the Psammaplysilla sponge [ 10 ]. Early studies revealed that psammaplin A had general antibacterial and antitumor properties. In 1999, it was found that psammaplin A exhibited significant in vitro antibacterial activity against both Staphylococcus aureus (SA) and methicillin-resistant Staphylococcus aureus (MRSA), which was inferred to be the result of induced bacterial DNA synthesis arrest by psammaplin A through inhibiting DNA gyrase [ 11 ]. Given the increasingly rapid emergence of multi-drug resistant bacterial strains and the corresponding threat to public health, there is significant interest in the development of structurally novel antibacterial agents such as psammaplin A. Additionally, psammaplin A has been reported to exhibit certain inhibition of a number of enzymes including topoisomerase II (topo II) [ 12 ], farnesyl protein transferase [ 13 ], leucine aminopeptidase [ 13 ], and latest reported chitinase [ 14 ]. Among these enzymes, topo II, as one required protein for eukaryotic DNA replication, as well as bacterial DNA gyrase belongs to the topoisomerase family of enzymes responsible for the remolding of DNA topology. Since psammaplin A can inhibit bacterial DNA synthesis through DNA gyrase inhibition, and much of the basic enzymology of the eukaryotic replication fork has close homologies with its prokaryotic counterpart, we wonder whether psammaplin A also can induce eukaryotic DNA replication arrest or not. We have reported that psammaplin A displayed significant cytotoxicity against human lung (A549), ovarian (SK-OV-3), skin (SK-MEL-2), CNS (XF498), and colon (HCT15) cancer cell lines [ 15 ]. In this paper, psammaplin A was found to have dose-dependent cytotoxicity on macrophage cell line. In order to clarify the possible mechanism of the cytotoxicity and also verify our conjecture of its possible action on DNA replication, the effect of psammaplin A on eukaryotic DNA replication was examined by using in vitro SV40 DNA replication system. According to our result that psammaplin A can induce eukaryotic DNA replication arrest through inhibiting some important replication proteins, we suggest that psammaplin A-induced cytotoxicity may correlate with its inhibition on DNA replication, and one of the main target molecules could be DNA polymerase α-primase. Methods Psammaplin A, proteins, cell extracts and DNA Psammaplin A sample was a gift from a Dr. Jung's lab of Pusan National University. SV40 origin-containing circular duplex DNA (pUC-ori + ), SV40 T-Ag, topoisomerase I (topo I), human DNA polymerase α-primase (pol α-primase), replication protein A (RPA), and HeLa extract were prepared as described previously [ 16 ]. Cell lines and chemicals Media for cell culture including HY, DMEM and RPMI were purchased from the Sigma Chemical Co. (St. Louis, MO, USA) and Fetal Calf Serum (FCS) was from Gibco-BRL (Gaithersburg, MD, USA). Cell Counting Kit-8(CCK-8) was purchased from Dojin Laboratories (Kumamoto, Japan). The mouse macrophage cell line RAW264.7 was purchased from Korean Cell Line Bank (Seoul, Korea). Cell viability assay Cell viability was determined by CCK-8 to count living cells by combining WST-8 and 1-Methoxy PMS [ 17 ]. Briefly, macrophage cells (RAW264.7) were seeded into 96 well plates at an initial density of 10 5 cells/well. After incubation with the indicated concentrations of psammaplin A for 12 hr, 10 μl of kit reagent was added and incubated for a further 3 hr. Cell viability was obtained by scanning with a microplate reader at 450 nm. SV40 DNA replication in vitro The reactions were carried out as described previously [ 18 ]. In brief, the reaction mixtures (40 μl) included 40 mM creatine phosphate-di-Tris salt (pH 7.7), 1 μg of creatine kinase, 7 mM MgCl 2 , 0.5 mM DTT, 4 mM ATP, 200 μM UTP, GTP, and CTP, 100 μM dATP, dGTP, and dCTP, 25 μM [ 3 H]dTTP (300 cpm/pmol), 0.6 μg of SV40 T-Ag, 0.23 μg of pUC-ori + , HeLa extracts, and psammaplin A as indicated. The reactions ran at 37°C for 2 hr, after which the acid-insoluble radioactivity was measured [ 18 ]. Topo I assay Topo I was measured by the relaxation of superhelical plasmid DNA [ 19 ]. The 20 μl assay mixture contained 50 mM Tris-HCl (pH 7.5), 120 mM KCl, 10 mM MgCl 2 , 0.5 mM DTT, 0.5 mM EDTA, bovine serum albumin (30 μg/ml), pUC118 (20 μg/ml), topo I (1 unit), and various amount of the psammaplin A. After 30 min at 30°C, the reactions were stopped by the addition of 5 μl of 5% NaDodSO 4 /25% (wt/vol) Ficoll 400 (Pharmacia) containing 0.25 mg of bromophenol blue per ml. The samples were then loaded onto the agarose gel (0.8%) for electrophoresis followed by photography. ssDNA binding assay The assay was performed according to the published procedures [ 7 ]. The reaction mixture (20 μl) contained 50 mM Hepes-KOH (pH 7.5), 150 mM NaCl, 1 mM MgCl 2 , 0.5 mM DTT, 10% glycerol, 50 fmol of 5'- 32 P-labeled oligo(dT) 50 (2200 cpm/fmol), plus the indicated amount of RPA, and was incubated for 15 min at room temperature. The complex was electrophoretically separated on a 5% polyacrylamide gel in 0.5 × TBE (89 mM Tris borate, 2 mM EDTA) at 15 V/cm. The gel was then dried and exposed to X-ray film. Pol α-primase assay DNA pol α-primase activities were assayed as described previously [ 20 ] with the following modifications. Reaction mixtures (30 μl) contained 40 mM creatine phosphate/di-Tris salt, pH 7.7, 1.0 μg of creatine kinase, 7 mM MgCl 2 , 0.5 mM DTT, 6 μg of bovine serum albumin, 4 mM ATP, 33 μM of [ 3 H]dTTP (500 cpm/pmol), 0.1 μg of poly (dA) 4500 : oligo (dT) 25 (20:1), DNA pol α-primase, and psammaplin A as indicated. After incubation at 37°C for 30 min, acid-insoluble radioactivity was determined [ 18 ]. Statistical analysis Values are presented as mean ± SD. Data was initially analyzed by one-way analysis of variance (ANOVA) and comparison of groups was made using Turkey test (SPSS software). Results Effect of psammaplin A on the viability of macrophage cell line As shown in Fig 1 , psammaplin A is a symmetrical bromotyrosine-derived disulfide dimer, which exhibits in vitro antibacterial activity against methicillin-resistant Staphylococcus aureus (MRSA). Psammaplin A is rather interesting owing to its two identical domains which are linked through a disulfide bridge. Macrophage cells are one of the key players in the early innate immune response, and they release inflammatory chemicals known as cytokines when they are activated. This sort of inflammation is not always a good thing, and overactive macrophage cells have been implicated in a number of human diseases, including arthritis and sepsis. When we studied the effect of psammaplin A on the viability of macrophage cell line RAW264.7, a reduced cell count was observed in the psammaplin A-treated cells and this decrease in the number of living cells also showed good dose-dependent (Fig 2 ). Inhibition of SV40 DNA replication in vitro by psammaplin A As it has been mentioned in the background, we wonder that psammaplin A has inhibitory effect on eukaryotic DNA replication or not. To verify this conjecture and also clarify the possible mechanism of psammaplin A-induced cytotoxicity, we examined the effect of psammaplin A on DNA replication using an in vitro SV40 DNA replication system. Addition of increasing amounts of psammaplin A quantitatively inhibited SV40 DNA replication with HeLa cytosolic extract (Fig 3 ). Inhibition of replication by psammaplin A in a cell-free system could be mediated either by damaging the template or by modulating the activity of a protein (or proteins) that is required for replication. The former mechanism is unlikely, because we have directly checked the effect of psammaplin A on DNA and didn't find any detectable damages to the template (data not shown). In order to find what proteins in DNA replication were affected by psammaplin A, we checked topo I activity at first, which plays key roles in DNA replication, transcription, and recombination by forming transient DNA single-strand breaks and acting as DNA strand transferase. In addition, the topoisomerase is now considered to be an important cancer chemotherapeutic target. The inhibitory effect of psammaplin A on the catalytic activity of topo I was shown in Fig 4a . The plasmid DNA was in the superhelical form (lane 1), and topo I relaxed the supercoiled DNA (lane 2). Psammaplin A inhibited the relaxation by topo I strongly at a concentration of 125 μM. Nicolaou and his colleagues reported that the DTT present in many enzyme assays could reduce the disulfide bond of psammaplin A to the corresponding free thiol [ 24 ]. In their experiment without DTT, psammaplin A exhibited no detectable inhibition of bacterial DNA gyrase up to 100 μg/ml. They suggested the weak inhibitory activity observed by the earlier authors could be attributed to the presence of the free thiol rather than the product itself. In our experiments detecting the effect of psammaplin A on topo I, no difference was found in the same gel between the reactions in the presence and absence of 0.5 mM DTT (Fig 4b ). In replication, RPA mediates unwinding of SV40 origin-containing DNA in the presence of SV40 T-Ag and topo I. It interacts with SV40 T-Ag and the DNA pol α-primase complex, which is necessary for the initiation of SV40 DNA replication [ 8 , 9 ]. Here, we examined the effect of psammaplin A on RPA's ssDNA-binding activity. As shown in Fig 5 , RPA formed stable complexes with oligo(dT) 50 , which appeared as two distinct bands in the polyacrylamide gel. The ssDNA binding activity of RPA was inhibited by psammaplin A in a concentration-dependent manner, and 500 μM of psammaplin A totally inhibited the ssDNA-binding activity of RPA. As described above, DNA pol α-primase complex is necessary for the initiation of SV40 DNA replication. To further investigate the inhibitory effect of psammaplin A in replication, we tested psammaplin A for inhibition of pol α-primase activity to see whether it's inhibitory effect on DNA replication correlate with pol α-primase activity. As shown in Fig 6 , the activity of pol α-primase was inhibited by psammaplin A, and 40 μM of psammaplin A inhibited about 94% the activity of pol α-primase. Discussion Psammaplin A has exhibited inhibition on general bacterium, some actinomycetes and fungi, and it also has cytotoxicity toward several cancer cell lines. In our research, we found that psammaplin A deliver significant cytotoxic activity against macrophage cell line RAW264.7. Which process is mostly affected by psammaplin A in cell cycle? What is the mechanism of the inhibition? Enlightened by the inhibitory effects of psammaplin A on bacterial DNA synthesis, bacterial DNA gyrase and eukaryotic topo I, we investigated the effect of psammaplin A on DNA replication using SV40 replication in vitro system attempting to find out the target process and molecules of psammaplin A and have a glimpse on cell cycle regulation. In our study, we found that psammaplin A inhibited SV40 DNA replication in vitro . In order to clarify the inhibition mechanism, further work were carried out. In SV40 DNA replication, three factors, SV40 T-Ag, RPA, and pol α-primase complex, are essential for initiation process. In the presence of topo I, SV40 T-Ag will continue to unwind the DNA to form a highly unwound DNA [ 21 ]. DNA synthesis with three factors and topoisomerase can be quite extensive [ 22 ]. We have suggested that psammaplin A might interfere with some molecules that are required to establish replication forks during the initiation reaction. To address this possibility, we asked whether psammaplin A inhibits topo I, RPA's ssDNA binding activity, and pol α-primase activity. In addition, the topo I is now considered to be important cancer chemotherapeutic target. In mammalian cells, actions of antitopoisomerase drugs on replication, transcription, and other processes ultimately activate pathways of programmed cell death [ 23 ]. Psammaplin A inhibited the DNA relaxation activity of topo I and the ssDNA binding activity of RPA in a dose-dependent manner, and up to 500 μM, psammaplin A can inhibit both the activities of topo I and RPA completely. On the other hand, psammaplin A significantly reduced pol α-primase activity at 40 μM. The above results indicate that major inhibition of SV40 DNA replication by psammaplin A may be due to the inhibition of pol α-primase activity. Here, we cannot rule out the possibility that psammaplin A inhibit the activity of SV40 T-Ag, because it is essential for initiation process. It is puzzling that the DNA pol α-primase, RPA and topo I were readily inhibited by low concentration of psammaplin A whereas the SV40 DNA replication assay still showed about more than 80% DNA replication in the presence of 125 μM psammaplin A. In our opinion, at least three points could account for this discrepancy. First, the cell extract we have used to support in vitro DNA replication system includes a large number of proteins, while in topo I assay, RPA binding assay and pol α-primase assay, purified proteins were used. Many of the proteins in the crude extracts can affect each other by physical or functional interactions. For example, in the process of DNA replication initiation, RPA interacts with T-Ag and DNA pol α-primase, and it is believed that RPA can both stabilize the unwound DNA and stimulate DNA pol α. The universal protein-protein interactions in crude extracts make its working environment quite different from that of purified protein assay system. Second, even for the same protein, for example, pol α-primase, the concentration in the cell extract and in the purified pol α-primase assay is not comparable before any quantification of the replication proteins in the cell extract. Third, due to the active disulfide moiety in the structure of psammaplin A, it is possible that psammaplin A could interact with some particular cellular targets in the crude extract, which may lead to covalent modification of the biological targets and psammaplin A itself. Therefore, the free available concentration of psammaplin A in the crude extract might be different from that in purified protein assay system. Different from the effect of psammaplin A on SV40 DNA in vitro replication, the significant inhibition of psammaplin A on the viability of macrophage RAW264.7 cells occurred at a relatively low concentration. There is the possibility that DNA replication might not be the single or primary event that affected by psammaplin A. It should be evident that the event of DNA replication in a living cell is more complicated than that in the crude extract system because of many existing cell cycle signals. So, it is still vague whether the ability of psammaplin A to inhibit cellular viability is correlated with its ability to inhibit DNA replication. In order to make it clear, it is necessary to check the effects of psammaplin A on other macromolecular synthesis (RNA synthesis and protein synthesis). Although the results in this paper do not clearly define the mechanism of cytotoxicity of psammaplin A, they have convincingly shown that psammaplin A possesses the abilities to inhibit DNA replication and some important replication proteins. Because of the disulfide bridge linking two identical subunits in the structure of psammaplin A, in this study, we also paid attention to the potential reduction effect of DTT present in the topo I assay. Different from the report of Nicolaou [ 24 ], we didn't catch any difference between the reactions in the presence and absence of DTT. There exist two possibilities. One is that in our reaction system, DTT couldn't reduce psammaplin A to the corresponding free thiol. Comparing the in vitro assay conditions in this study, we can find that all these assays were performed in similar conditions (for example: pH 7.5~7.7, 0.5 mM DTT and Mg 2+ ion environment). Given the mildness and tolerance of these reaction mixtures, we guess that psammaplin A is stable in the assays. Of course there is the second possibility that psammaplin A was reduced in the reaction, but the reduction product had nearly the same inhibition effect on topo I as the original compound. Further investigation aimed at this question need to be carried out. Conclusions Based on our results, we suggest that the cytotoxicity of psammaplin A might be related to the inhibitory effect it has on the fundamental cellular process-DNA replication, and one of the main target molecules of psammaplin A could be pol α-primase. Competing interests The authors declare that they have no competing interests. Authors' contributions YHJ and DKK conceived and designed the experiments and wrote the manuscript. YHJ performed all of the experiments. JHJ provided psammaplin A sample. EYA, SHR, JSP, HJY, SY, BJL and DSL participated in the conception, supervision, coordination and guidance of the study and manuscript preparation. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524492.xml
517716	Protein p16 as a marker of dysplastic and neoplastic alterations in cervical epithelial cells	Background Cervical carcinomas are second most frequent type of women cancer. Success in diagnostics of this disease is due to the use of Pap-test (cytological smear analysis). However Pap-test gives significant portion of both false-positive and false-negative conclusions. Amendments of the diagnostic procedure are desirable. Aetiological role of papillomaviruses in cervical cancer is established while the role of cellular gene alterations in the course of tumor progression is less clear. Several research groups including us have recently named the protein p16 INK4a as a possible diagnostic marker of cervical cancer. To evaluate whether the specificity of p16 INK4a expression in dysplastic and neoplastic cervical epithelium is sufficient for such application we undertook a broader immunochistochemical registration of this protein with a highly p16 INK4a -specific monoclonal antibody. Methods Paraffin-embedded samples of diagnostic biopsies and surgical materials were used. Control group included vaginal smears of healthy women and biopsy samples from patients with cervical ectopia. We examined 197 samples in total. Monoclonal antibody E6H4 (MTM Laboratories, Germany) was used. Results In control samples we did not find any p16 INK4a -positive cells. Overexpression of p16 INK4a was detected in samples of cervical dysplasia (CINs) and carcinomas. The portion of p16 INK4a -positive samples increased in the row: CIN I – CIN II – CIN III – invasive carcinoma. For all stages the samples were found to be heterogeneous with respect to p16 INK4a -expression. Every third of CINs III and one invasive squamous cell carcinoma (out of 21 analyzed) were negative. Conclusions Overexpression of the protein p16 INK4a is typical for dysplastic and neoplastic epithelium of cervix uteri. However p16 INK4a -negative CINs and carcinomas do exist. All stages of CINs and carcinomas analyzed are heterogeneous with respect to p16 INK4a expression. So p16 INK4a -negativity is not a sufficient reason to exclude a patient from the high risk group. As far as normal cervical epithelium is p16 INK4a -negative and the ratio p16 INK4a -positive/ p16 INK4a -negative samples increases at the advanced stages application of immunohisto-/cytochemical test for p16 INK4a may be regarded as a supplementary test for early diagnostics of cervical cancer.	Background Cervical cancer makes up about 10–12% of total women cancers [ 1 , 2 ] with the level of mortality in Russian population 5.0 per 100000 [ 1 ]. The tendency is being observed for the past decades towards reduction of both incidence and mortality. It is mainly due to the population-wide screening protocols in developed countries which allow identifying early asymptomatic forms of cervical carcinomas. However some problems remain to be solved concerning early detection of this type of cancer. The main screening test for cervical cancer is the cytological smear staining technique developed by G. Papanicolaou [ 3 ] and known as Pap test. Despite evident success this test gives a substantial rate of both false-positive and false-negative results. Histological analysis of a biopsy sample, more laborious in preparation and study as compared with that of a cytological smear, is also not absolutely efficient owing to a substantial rate of interobserver discrepancies among expert pathologists examining the same material [ 4 ]. Infection with human papilloma viruses (HPV) belonging to so-called high-risk group is the main risk factor of cervical cancer incidence [ 2 ]. To detect high risk HPVs in epithelial cells of a patient polymerase chain reaction (PCR) has been applied during the past two decades [ 5 ]. However this highly sensitive technique cannot resolve the problem of early cervical carcinoma detection also so far as many early stage lesions regress and epithelial dysplasia (i.e. cervical intraepithelial neoplasms, CINs) and carcinomas appear only in a minor part of the persons in whose epithelium (on a smear) high risk HPVs had been detected [ 2 ]. For the recent years several research groups [ 4 , 6 - 16 ] including us [ 11 ] have dwelt on the protein p16 INK4a for a possible supplementary marker of dysplastic and neoplastic cervical epithelium lesions. This protein belongs to the group of cyclin-dependent kinase Cdk4/6 inhibitors [ 17 ] and is encoded by tumor suppressor gene INK 4a (synonyms: MTS 1 , CDKN2 , INK4a/ARF ). Gene INK4a plays an important role in the regulatory pathway Cdk-Rb-E2F. The product of this gene p16 INK4a prevents pRb phosphorylation by inactivating Cdk4/6; pRb keeps on binding E2F transcription factors and as a result cells stay in G1 phase not passing to DNA replication. In various tumor types INK4a as a bona fide tumor suppressor undergoes homozygous deletions, is inactivated by point mutations, LOH or hypermethylation; p16 INK4a expression is reduced or ceases under such conditions or the protein function may be impaired [ 18 ]. Peculiarity of cervical carcinomas is due to the ability of HPV oncoprotein E7 to interact with pRb and inactivate it [ 19 ]. As a result, the regulatory pathway Cdk-Rb-E2F is disrupted and the status of gene suppressor INK4a and its protein becomes of no importance for a cell so far as they function upstream of the site of breakage. Cells with thus inactivated pRb pass cell cycle checkpoint G1/S without any obstacle. Reciprocallity between status of pRb and that of p16 INK4a commonly found in human permanent cell lines (including cervical cell line cultures) [ 17 , 20 , 21 ] as well as in primary tumor cells [ 9 , 22 , 23 ] served for a logical prerequisite of utilizing p16 INK4a protein as a marker of premalignant and malignant cervical epithelium cells. Functionally active gene RB was shown to be able to negatively regulate the expression of INK4a on a transcriptional level, but details of this negative feed-back loop remain obscure [ 24 ]. To estimate the applicability of p16 INK4a as a marker of dysplastic and neoplastic alterations in cervical epithelium cells we analyzed the expression of this protein utilizing a highly p16 INK4a – specific monoclonal antibody. We examined 197 samples in total. The materials studied included: 1) samples of normal epithelium of healthy women (cytological vaginal smears), 2) diagnostic biopsy samples from patients with cervical ectopia, 3) samples of CINs of various stages, 4) samples of invasive cervical cancer, 5) samples from different normal utery body and cervix tissues from women with gynecological diseases not associated with dysplastic lesions of epithelium (diagnostic biopsies and surgical materials) and 6) samples of cells from 3 cervical carcinoma cell lines. We also compare our data with the results presented by other groups working in similar directions [ 4 , 6 - 10 , 12 - 16 ]. We demonstrate in the present work that overexpression of the protein p16 INK4a is typical for some samples of dysplastic and neoplastic epithelium of cervix uteri. We have not found at least a single sample overexpressing p16 INK4a among control samples. The portion of p16 INK4a -positive samples increases in the following row: CIN I – CIN II – CIN III – invasive carcinoma. However p16 INK4a -negative CINs and carcinomas have also been found. All stages of CINs and carcinomas analyzed turn out to be heterogeneous with respect to p16 INK4a expression: side by side with the samples which express p16 INK4a in 25% of cells or more we detect samples which are stained poorly or lack any staining. So p16 INK4a -negativity does not seem to be a sufficient reason to exclude a patient from the high risk group when results of Pap-test, HPV detection by PCR or histological investigation warn about possibility of poor prognosis. As far as normal cervical epithelium is p16 INK4a -negative and the ratio p16 INK4a -positive/ p16 INK4a -negative samples increases at the advanced stages of CINs and carcinomas application of immunohisto-/cytochemical test for p16 INK4a may be regarded as a supplementary (optional) test for early diagnostics of cervical cancer. Methods Immunohisto-/immunocytochemical study was performed on 197 samples in total. Those included 6 samples from normal epithelium of healthy women (cytological vaginal smears taken during regular examination), as well as the following surgical and diagnostic biopsy materials: 37 samples from patients with cervical ectopia, 113 samples of CINs of different stages including cancer in situ, 26 samples of invasive cervical cancer (21 squamous cell carcinomas and 5 adenocarcinomas), 12 samples of normal tissues from uterus body (myometrium) and cervix from patients with different gynecological diseases, 3 cervical cell line samples taken for positive control (see below). Apart from 9 smears (which included 6 samples from healthy women and 3 samples taken from cell cultures which served as controls) the rest 188 materials were paraffin-embedded histological blocks. The quality of smears turned out to be satisfactory for immunochemical analysis (Fig. 1a and 1d ). Neutral formalin-fixed paraffin embedded samples of biopsies and surgery materials were received from archives of N.N.Blokhin Cancer Research Center (Moscow), P.A. Gertzen Institute of Oncology (Moscow), Central Clinical Hospital (Moscow) and Fourth Clinical Hospital (Poltava, Ukraine). For immunohistochemical analysis 4–5 μm serial sections were transferred on Histobond slides with adhesive layer (SMT Geraetehandel GmbH, Germany). The first section was stained with hematoxylin-eosin for traditional morphological analysis and verification of diagnosis by not less than two independent pathomorphologists. Presence of high risk HPV in cervical CIN and carcinoma cells was verified by PCR [ 25 ]. Some of the squamous cell carcinomas were tested by Southern blot hybridization in addition to PCR. The results of both methods coincided completely. To detect high-risk HPV in adenocarcinomas Hybrid Capture 2 test (enabling to detect but not to discern HPV 16 and HPV 18) was used. The following portions of samples (out of those studied immunochemically) were analysed for high-risk HPV genetic material: CINs I – 9/51, CINs II – 9/32, CINs III – 19/24, invasive squamous cell carcinomas – 21/21, adenocarcinomas – 5/5. This study did not cover all CINs which we studied immunochemically due to the tiny size of most of those samples. Vaginal normal epithelium smears obtained during regular examination from healthy women were transferred on HistoBond slides. Absence of abnormal cells was confirmed in Cytology Department of Cancer Research Center on a parallel slide. Smears were fixed for 5 min in 10% formaldehyde, washed with flowing water, and processed for further development as described for deparaffinized histological slices [ 9 ]. Cells of cervical cell lines were cultured under standard conditions in DMEM supplemented with 10% of foetal bovine serum. For immunocytochemistry cells were taken during culture receeding, dropped in a culture medium on a Histobond slide, air-dried and then processed identically to smears from healthy women. Immunohistochemical staining was carried out using p16 INK4a -specific monoclonal antibodies E6H4 (MTM Laboratories AG, Germany) according to protocol by Klaes et al [ 9 ]. Controls in the course of immunohistochemical studies were as follows Positive controls 1). In the beginning of the study as a whole cells of three cervical cell lines were stained first. They were SiHa (HPV16-positive), C33a and HT-3 (both HPV-negative); all of them had been characterized as expressing the protein p16 INK4a in earlier immunocytochemical studies with the same antibody [ 9 ]. 2). One HT-3 cell slide was stained in parallels with the first series of biopsy materials. 3). Every next group of slides intended for staining compulsorily included the slide with one of the serial sections of the CINIII or invasive carcinoma sample which had been characterized as a p16 INK4a -positive one in our previous studies (on preceeding cuts). The analysis of the given slide series was carried out only if the positivity of the control sample was confirmed. Thus neither slide series in the present study was completely p16 INK4a -negative. Negative controls 1). One more serial cut made from p16 INK4a -expressing material was included into every slide series meant for staining (as in variant 3 of positive controls, – see above) which was processed in a usual way but PBS was applied instead of p16 INK4a -specific antibody. The results were regarded as valid if this slide was negative. 2). As an intrinsic negative control served adjacent to CIN or carcinoma normal tissues including stromal elements. In neither case did we find any staining in these tissues (fig. 1c ). In addition to four categories of staining defined by Klaes et al. [ 9 ],- poor (less than 1% of stained cells,- figure 1b ), sporadic (1–5% of stained cells, – figure 1b ), focal (cell clusters were stained but not more than 25% of cells were positive, – figure 2a ) and diffuse (more than 25% of cells were stained, – figures 1c and 2c,2f,2g,2h ), we formed one more group (negative) of those samples which totally lacked any stained cells (Figures 1a , 2b,2d and 2e ). We regarded those cells as stained in which p16 INK4a was expressed in nuclei and/or in a cytoplasm. Results Various types of p16 INK4a -specific staining of cervix uteri normal, dysplastic and cancer cells are presented in Fig. 1a,1b,1c,1d and Fig. 2a,2b,2c,2d,2e,2f,2g,2h . We did not register any staining in either of 6 smear samples taken from healthy women (Fig. 1a ). An example of a very poor cytoplasmic staining in some separate CIN I cells with a more pronounced (sporadic, both nuclear and cytoplasmic) staining in the adjacent cancer in situ cells is shown in Fig. 1b . In one invasive carcinoma we detected cytoplasmic staining in the predominant majority of cancer cells while but sole nuclei turned out to be stained (Fig. 1c ); the boundary between cancer and normal tissues coincided with the line at which the staining discontinued. As to the cervical cells cultivated in vitro taken for positive controls, in HT-3 cells the specific staining was strongly manifested both in nuclei and in a cytoplasm (Fig. 1d ) while in SiHa and C33a cells it was exclusively cytoplasmic. It is not clear yet why in one and the same cervical cancer sample the protein p16 INK4a (normally showing its activity in nuclei) may be detected but in a cytoplasm in the majority of cells while in a number of cells both in nuclei and cytoplasm, why in some cervical cell cultures it is found solely in cytoplasm and in other cervical cell lines – in nuclei also. We did not find at least a single sample with an exceptionally nuclear staining; similar were the results by other investigators [ 9 ]. With keeping in mind that subcellular location of p16 INK4a -specific staining varies to such a degree we scored as positive every sample in which the staining was expressed either in a cytoplasm or in both nuclei and cytoplasm. The data of the analysis of the cervical epithelium preparations stained with monoclonal antibodies E6H4 are summarized in Table 1 . In all 37 samples from patients with cervical ectopia p16 INK4a -positive cells were observed with low frequencies: 34 samples were negative and 3 ones (8,1%) were stained poorly. In the group of CINs I 63% of the samples were p16 INK4a -negative. In 7 samples (about 14%) sporadic or focal (Fig. 2a ) type of staining was observed. Group of CINs II turned out to be highly heterogeneous in terms of the ratio of p16 INK4a -positive cells as well. Most of these samples (68%, 26 out of 38, Fig. 2b ) did not differ from the samples of normal epithelium, but the rest 12 samples we attributed to poor, sporadic or focal types. In neither sample of CIN I or CIN II did we observe diffuse staining for p16 INK4a . Among the samples of CIN III about 54% (13 out of 24) were attributed to sporadic (Fig 1b ), focal or diffuse (Fig. 2c ) types. However every third CIN III sample was found to be p16 INK4a -negative (Fig. 2d ). As to invasive cancers, only 1 out of 26 samples (3.8%) lacked p16 INK4a -positive cells (Fig. 2e ). One sample expressed the marker poorly. In 24 cases sporadic, focal or diffuse staining was observed. Diffuse staining was registered in about 50% of these samples (Figs. 1c , 2f,2g,2h ). Among 14 samples of invasive carcinomas which had been stored in paraffin blocks for 7 years 1 (7.1%) turned out to be p16 INK4a -negative, 1 (7.1%) stained poorly, 3 (21.4%) stained sporadically, 4 (28.4%)-focally and 5 (35.8%) expressed diffuse staining. We have examined several cases with more than one type of lesion on the same slide. Examples are presented on Figures 1b and 2c . As a rule in combined cases p16 INK4a expression was more pronounced in cells belonging to a more advanced lesion (Fig. 1b ). All the HPV-tested samples of CINs and invasive squamous cell carcinomas were high-risk HPV-positive (data on CINs I and CINs II are shown in the additional file, results with CINs III and squamous cell carcinomas – in Table 2 ). High-risk HPVs were also found in all the samples of adenocarcinomas. In reference group composed of samples from normal uterus body and cervix uteri tissues obtained from patients with gynecological diseases (stromal and glandular tissues of the cervix from patients with cervical ectopia from which squamous cell epithelium had been fully cut off; myometrium of uterus body from the patient after surgery for cervical carcinoma) the results of staining with p16 INK4a -specific antibodies were negative in all 12 cases (Table 3 ). Discussion Immunochemical detection of p16 INK4a by monoclonal antibodies shows that the overwhelming majority of invasive cervical carcinoma samples differ from normal cervical epithelium of healthy women. We found no cases of sporadic, focal or diffuse staining with E6H4 antibodies among 6 vaginal smears from healthy women as well as among 37 samples from patients with cervical ectopia including those in which ectopically localized cervical epithelium koilocytes or condyloma had been registered. In addition, in 11 samples of normal glandular and stromal cervical tissues and one sample of uterus body normal myometrium tissue which were obtained from patients with gynecological disorders p16 INK4a was not expressed at all. The predominant majority of invasive carcinoma samples were both high-risk HPV-positive and p16 INK4a -expressing. However in 2 samples not expressing any p16 INK4a or expressing it poorly high-risk HPV DNA sequences were detected by both PCR and Southern blot hybridization. These data are in a good agreement with those by Klaes et al., who described two samples of p16 INK4a -negative but HPV-positive cervical cancer [ 9 ]. Among CIN III samples which were stained in our experiments poorly or lacked any staining (table 2 , samples 1–6) high risk HPVs were found in 6 out of 6 tested. As to CIN I and CIN II samples expressing p16 INK4a in less than 1% of cells ( Additional file data, samples 1–5 and 10–14, respectively), we did not find any high-risk HPV-negative case in those groups either. Thus there seem to exist dysplastic and neoplastic lesions of cervix uteri which do not overexpress the protein p16 INK4a but harbor high-risk HPV DNA. We confirm the data by Klaes et al. [ 9 ] that prolonged (for 7 years in our case) preservation of cervical carcinoma samples in paraffin blocks does not preclude the material from diffuse staining. Monoclonal antibodies E6H4 originally described by Klaes et al [ 9 ] had been tested by those authors among a number of clones of p16 INK4a -specific monoclonal antibodies. The following commercially available clones had been taken: 1). DCS-50.1/H4-NA29 (Oncogene Research Products, Cambridge, MA), 2). 375P (Biogenex Laboratories, San Ramon, CA), 3). ZJ11 and JC8 (NeoMarkers, Freemont, CA), 4) 05–418 (Upstate Biotechnology, Lake Placid, NY) and 5). J175–405 (PharMingen, San Diego, CA). The results of the comparative experiments by Klaes at al [ 9 ] had demonstrated that clone E6H4 had turned out to be the most specific as inferred on the lowest level of unspecific staining on different tumor cell lines, both HPV-negative and positive. As in our experiments Klaes et al. [ 9 ] registered mainly cytoplasmic staining both in cells of permanent cell lines and in preneoplastic and carcinoma samples; an exclusive nuclear staining was not detected. We observed strong p16 INK4a -positivity (sporadic or focal staining) in a comparatively small number of CIN I and CIN II samples (13–14 %). Among CINs III the portion of such samples increased up to 54,2%, with diffuse staining in every sixth case. The data we have obtained for CINs of all stages rather differ from the results by Klaes et al. [ 9 ]. All of CINs II and CINs III were stained as diffuse by Klaes et al. [ 9 ]. In the present study the group of CINs as a whole is much more heterogeneous. The reasons for those discrepancies are not quite clear yet. The population of Russian and Ukrainian patients whose materials were used in the present study was extremely heterogeneous with respect to age, nationality, etc. We cannot exclude that the discrepancies mentioned may be due to these factors as had been discussed earlier [ 26 ]. Nevertheless the coincidence between our results and the data by Klaes et al [ 9 ] concerning the general trend seems much more important: the increase of p16 INK4a expression in dysplastic and neoplastic cervical epithelium in the course of progression. This trend is especially evident in our study when negative cases only are taken into consideration. All 6 samples were p16 INK4a -negative in control group. Similar was the situation in the reference group: 12/12 negative. Among CINs I, CINs III and invasive carcinomas these indices made up 32/51 (63 %), 8/24 (33%) and 1/26 (4 %) respectively. The data presented herein allow us to conclude that hyperexpression of p16 INK4a when detected immunohistochemically may be regarded as a marker of dysplastic and neoplastic lesions in cervical epithelium. Positive correlation between p16 INK4a expression and morphological stage of the disease on the same slide (when combined cases with both dysplastic and neoplastic lesions were found) also favors this inference. In a number of recent communications immunochemical staining for p16 INK4a was suggested to be performed on cytological smears [ 9 , 13 - 15 ]. Cytological approach has some advantages, so far as it enables a patient to avoid surgical intervention. That is why one may expect in the nearest future immunocytochemical version for detecting p16 INK4a to become rather widely used not only as an addition to surgery but also for regular examinations in outpatient clinics. In this connection it seems important to estimate frequencies of a feasible false-negativity of the test. Keeping in mind that dysplastic lesions of all stages frequently regress and do not convert into invasive cervical carcinomas [ 2 ], we summarized the data by different groups on the frequencies of p16 INK4a -negative invasive carcinoma samples (Table 4 ). We found that p16 INK4a -negative cervical adenocarcinomas had been detected with frequencies which had varied from 0% up to 42,5% while p16 INK4a -negative squamous cell carcinomas – with frequencies that had not exceeded 10,1%. Substantial variability of the data may be due to small numbers of samples analyzed (in some of the studies), to utilization of different types of monoclonal antibodies, to different criteria used by different research groups for the results interpretation, etc. It seems important to realize the proper place of immunocyto-/ immunohistochemical analysis of p16 INK4a expression among previously developed tests for early detection of cervical carcinomas. In this connection the following points deserve mentioning. A pathologist in the course of common histological investigation usually registers some morphologic features such as zones of active mitotic divisions, multipolar mitoses, vicinity of condilomas and other formations. However when performing an immunocytochemical analysis on a smear a cytologist cannot bring in a correlation with morphological structure results of staining with p16 INK4a -specific antibodies in separate cells. According to the results of the present study about 33% of CINs III as well as about 5% of invasive squamous cell carcinomas of cervix uteri do not differ from normal cervical epithelium with respect to p16 INK4a expression. Thus immunochemical staining for p16 INK4a (in both cyto- and histochemical versions) does not seem to be the approach that can help to fully overcome absolutely all existing ambiguities of cervical cancer early diagnostics. As to possible sources of false positivity of immunochemical detection of the protein p16 INK4a the study by Agoff et al [ 16 ] deserve mentioning. According to this communication in 10 out of 10 endometrial biopsy samples studied endometrial cells were positively stained with the p16 INK4a – specific antibody E6H4. So far as endometrial cells may occur on vaginal smears the test does not seem to enable one to fully avoid false positivity. Conclusions Overexpression of the protein p16 INK4a encoded by tumor suppressor gene INK4a is a characteristic of displastic and neoplastic alterations of cervical epithelium. The portion of p16 INK4a -positive samples increases in the following row: CIN I – CIN II – CIN III – invasive carcinoma. All stages of CINs and carcinomas analysed are heterogeneous with respect to p16 INK4a expression: side by side with the samples which expressed p16 INK4a in 25% of cells or more we detected samples which were stained poorly or lacked any staining. According to the data we present herein p16 INK4a -negative cervical neoplasms and carcinomas do exist. Thus the lack of p16 INK4a -positive cells in samples from different types of cervical lesions should not be regarded as a sufficient reason for excluding a patient from the high-risk group. Despite this as far as normal cervical epithelium is p16 INK4a -negative and the ratio p16 INK4a -positive/ p16 INK4a -negative samples increases at the advanced stages of CINs and carcinomas application of immunohisto-/cytochemical test for p16 INK4a may be regarded as an additional (optional) test for early detection of precancerous lesions in cervical epithelium. Competing interests none declared Author's contributions F.K., D.S., G.V. and G.F. contributed to study conception and design; E.K. and A.B. delt with sample obtaining and preliminary diagnosis; Yu.A., G.F. and V.E. were independent pathologists who confirmed the diagnosis; G.V., L.Z. and D.G. performed immunohistochemical staining and analysis; A.B. carried out HPV typing; G.V. drafted the manuscript. All authors read and approved the final version. Pre-publication history The pre-publication history for this paper can be accessed here: Supplementary Material Additional File 1 Data on high risk HPV genome detection by PCR in CIN I and CIN II samples. the data are presented on high-risk HPV genetic material detection in 9 CIN I and 9 CIN II samples. Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517716.xml
544400	Transfer of Health for All policy – What, how and in which direction? A two-case study	Background This article explores the transfer of World Health Organization's (WHO) policy initiative Health for All by the Year 2000 (HFA2000) into national contexts by using the changes in the public health policies of Finland and Portugal from the 1970's onward and the relationship of these changes to WHO policy development as test cases. Finland and Portugal were chosen to be compared as they represent different welfare state types and as the paradigmatic transition from the old to new public health is assumed to be related to the wider welfare state development. Methods The policy transfer approach is used as a conceptual tool to analyze the possible policy changes related to the adaptation of HFA into the national context. To be able to analyze not only the content but also the contextual conditions of policy transfer Kingdon's analytical framework of policy analysis is applied. Conclusions Our analysis suggests that no significant change of health promotion policy resulted from the launch of HFA program neither in Finland nor in Portugal. Instead the changes that occurred in both countries were of incremental nature, in accordance with the earlier policy choices, and the adaptation of HFA program was mainly applied to the areas where there were national traditions.	Introduction The World Health Organization (WHO) launched a policy framework called Health for All by the Year 2000 (HFA2000), in 1978, and has since then been advocating this framework for health policy making to all its member states [ 1 , 2 ]. This paper explores the transfer of HFA policy into national contexts by using the changes in public health policy of Finland and Portugal from the 1970's onward and their relationship to WHO policy development as test cases. Finland and Portugal were chosen to be the cases observed as they represent different welfare state types and as the paradigmatic shift from the old to new public health is assumed to be related to the wider welfare state development. The development of the welfare state constitutes the frame of reference for the analysis of transfer of HFA policy. Policy transfer is a theoretical perspective that has been used to describe the spread of policy ideas from one political setting to another [ 3 ]. Most studies have concentrated on studying the transfer between countries, here the transfer is assumed to be mediated through an international organization (WHO) to its member states. Our aim is to locate the transfer of HFA policy in a broader conceptual framework. This entails clarifying the theoretical and political assumptions inherent in HFA policy as well as studying the transfer process in the historical context of broader welfare state development. In order to analyze the transfer of HFA policy it is necessary to recognize that HFA policy is not a totally coherent health strategy that can be defined in one compact and consensual manner. The ambiguous nature of HFA policy stems from fact that it is constructed in various policy documents drawn up in temporally and contextually different situations. We refer to the following four central documents and the ideal model of public health policy they construct when we speak of HFA policy: the Declaration of Alma-Ata (1978) [ 1 ], Targets in support of the European regional HFA strategy (1985) [ 4 ], the Ottawa Charter (1986) [ 5 ], Health21 for Europe (1999) [ 6 ]. There are points of convergence in the picture of the ideal policy model these documents transfer, but also differences linked to the evolution of temporal macro-political cycles (collapse of colonialism, new international economic order, expansion of the welfare state, collapse of communist regimes in Eastern Europe, globalization and the crisis of the welfare state) or to the regional characteristics (European vs. global). Thus when we speak of HFA we refer to the HFA policy constructed in the aforementioned documents. To be able to analyze the adaptation of HFA in national contexts we have concentrated on examining three aspects of it: primary health care, community approach and healthy public policies. Based on the analysis of these aspects the study aims to explore how since the 1970's a number of essential aspects of health promotion policy have changed in Finland and in Portugal in relation to the ideas of HFA. While a few studies have addressed the spreading of the HFA policy to the member states [ 7 - 11 ], this is quite seldom based on any theory of policy change or policy transfer. Also, while most of these studies have either been descriptive in nature or focused on evaluating national policies in the program level by verifying program's outcomes or situational validity of its objectives, we aim to analyse the policies in a broader societal context by taking into account the societal-level vindication as well as the political context of health policies [ 12 ]. In the policy transfer literature past policies, present policy complexity and the question of policy feasibility are seen as possible policy constraints. Likewise factors such as identical past policies or similar ideology can be seen to facilitate the transfer [ 3 ]. Locating the transfer of HFA policy in the context of existing public health policies and the wider political and social contexts of the countries in question offers one means to identify essential capacities, constraints and conditions for the adaptation of this particular policy innovation. To be able to analyze not only the content ( What was transferred? ) but also the contextual conditions ( How/why did this happen? ) of policy transfer we use Kingdon's (1986) analytical framework of policy analysis [ 13 ]. According to Kingdon, a policy change process is conditioned by three analytically distinct streams: problem, policy and politics stream. Problem stream brings issues to the political agenda, while policy stream, which consists of experts, produces solutions and alternatives to policy problems. From these alternatives the politics stream then determines what, if any, are politically feasible alternatives to be adapted. A window of opportunity is open for a major policy change only if these three different streams of policy making process coexist simultaneously. Policy transfer may occur at different stages of the policy making process. In this paper we will focus mainly on the agenda-setting and policy formulation phases. These phases can be regarded as a valuable starting point for the further development and implementation of HFA at the national level. The policy transfer approach is used as a conceptual tool to analyze the possible policy changes that the adaptation of HFA into the national context may have caused. Method and materials (see figure 1 ) We aim to identify concrete examples of transfer related changes in the content of formal government documents such as laws, reports, strategies and government programs. The detailed analysis of these documents provides some evidence of policy transfer. Non-formal government documents, evaluative reports, studies and relevant discussion are also used as material. The analysis of the policy documents was supported by expert interviews conducted in both of the countries in 2003–2004 for the purposes of this study. Historical reading of the documents can provide evidence about the time frame of policy change. In Portugal the first health strategy was published in 1999 [ 14 ], and thus the primary material for the analysis of governmental health policy before 1999 is government programs [ 15 ]. Finnish health promotion policy and its relevant documents [ 16 - 18 ] have been evaluated twice by an international review group [ 19 , 20 ] and several times by Finnish public health experts and national committees [ 20 ] and thus the analysis of the Finnish case is rather based on these evaluations and reviews than on the programs. Figure 1 The Analytical Framework of the Study Results HFA as a rethinking of public health policy WHO advocated "Health for All" as a rethinking of and challenge for reform in the national public health policies of the member states. HFA was frequently understood as a policy for developing countries focusing on advocacy for linking public health aims with broad social and environmental development policy at the local level, instead of investing in hospital medicine for the elites of the country. The core idea of the Alma-Ata Declaration (1978) was to advocate such a policy under the concept of primary health care . In most OECD countries there already was some organizational form of primary medical care. Many health policy makers thought that HFA does not apply to OECD countries. Others argued that the idea of a comprehensive social and intersectoral health policy under the banner of primary health care also challenged the OECD countries. According to this understanding, Europe, too, was to develop its own HFA policy [ 21 ]. Seven years after Alma-Ata, in 1985, the WHO Regional Committee for Europe adopted its own HFA policy. It advocated a comprehensive, intersectoral and participatory health policy aiming at health gain and equity in health [ 4 ]. The conceptual differences between the Global and the European HFA are significant. The European HFA located primary health care as one aspect of "appropriate care" and to the basic level in the organization of health services. Both the Global and the European HFA argued for a comprehensive and intersectoral health policy. However, the meaning of these concepts is dependent on the context in which they are used. It may be argued that in the European HFA, the context is a welfare state – at that time either state capitalistic or state socialistic – with its numerous institutions and administrative sectors. The context in the Global HFA was a general and broad social and economic development of so-called developing countries. The concept of HFA was accompanied by the introduction of two other challenging concepts: health promotion [ 5 ] and new public health [ 22 ]. All three were rhetorically contrasted to something that was called old or dominant way of thinking and making health policy that was characterized as focusing on hospital and cure, following a biomedical model and applying a narrow understanding of health and determinants of health. Each of these three concepts had their own history and points of reference. For example, health promotion was mainly developed from a critical assessment of the health education of the 1970's [ 23 ]. New public health was advocated as a response to the change in the disease panorama, which meant that instead of hygiene, physical environment and vaccinations the new focus of interventions was to be on the social, cultural and political determinants of lifestyles and health [ 24 ]. HFA advocated an outcome-oriented health policy implemented by a wide range of social and economic institutions instead of focusing on the supply of medical care inputs. The Ottawa Charter on Health Promotion mentions peace, shelter, education, food, income, a stable ecosystem, sustainable resources, social justice and equity as basic prerequisites for health [ 5 ]. In addition to advocating reorientation in health services and the development of personal health skills, the Charter also includes in health promotion foci a wide range of public policies, communities and daily social and physical environments. The WHO Vienna Dialogue (1986) even concluded that the best health promotion policy is a good social policy [ 25 ]. Taking into consideration that OECD had declared, in 1979 [ 26 ], the "Crisis of the Welfare State", we may locate the conceptual innovation of health promotion in OECD countries as advocacy for certain aspect of welfare state reform, as one remedy for the "crisis". The European version of HFA may also be read from a welfare state reform advocacy perspective. The European HFA strongly advocates a broad health policy managed by objectives [ 21 ]. The strong management by objectives advocacy in the European HFA [ 4 , 6 ] links it with the managerialistic reform agenda of the welfare state [e.g. [ 27 , 28 ]]. Managerialism seems to be less prominent in the Ottawa Charter. Rather, it is possible to claim that the Charter has been influenced by ideas to develop welfare states leaning on social and community movements [ 29 ]. For the purposes of this study, we may conclude that there are three related concepts, health for all, health promotion and new public health that have a lot in common in their critique of the hospital focused and biomedically oriented health policy paradigms. They all advocate a broader socio-political orientation for health policy. However, they do not have a common idea of what this broader orientation is. We find it helpful to distinguish at least three different orientations. The first is the broad social and economic development context where the Alma-Ata Declaration located the radical local development program under the concept of primary health care. The second is the managerialistic welfare state reform context where the European HFA located the proposed health policy guided by HFA targets. A third orientation has been developed under the banner of the Ottawa Charter. This third variant may also be located in a welfare state reform context, but in a different idea of reform emphasizing community development. Finland and the HFA challenge Finland experienced an extremely rapid urbanization phase in the 1960's and the 1970's. A large part of the population moved from the rural areas to industrial and service workplaces in the urban centers. The proportion of working people earning their main income from agriculture decreased from 26% to 10% in 1950–1980 [ 30 ]. A significant part – about 7 % of the population – even moved outside the country, to Sweden. Part of the rapid and profound socio-economic change experienced was the development of a Scandinavian type welfare state in Finland. In about 25 years the country developed universal old age, sickness, disability and unemployment benefit systems and started the expansion of a public day care system for children and long term care for the elderly. The existing public education and culture systems were rapidly expanded and a universal health care system was set up. By 1985, it was possible to include Finland in the group of small, prosperous and egalitarian Nordic countries, still somewhat poorer and less generous than the older sisters Sweden and Denmark [ 31 ]. Primary care In the late 1960's and the early 1970's, the challenge of the Finnish health policy was often articulated by asking: "Why does a country with Europe's healthiest children have the sickest middle-aged male population?" International comparative statistics had indicated that the country was at the European top in terms of low child mortality, but the adult population, particularly males, was dying younger than most other West European adult populations. The positive health status of children was understood as an outcome of a universal, strong and preventive maternal, child and school health system. The health system for the adult population was criticized for being too hospital centered [ 32 , 33 ]. The context of rapid socio-economic change, left-center-coalition government and a perspective of rapid overall development of the welfare state was a fertile growing ground for extending the example of universal, public and preventive child and maternal care to the adult population as well. The Primary Health Care Act of 1971 started the building of multi-professional and multi-functional local health centers to carry forward the idea of "people's health work" at the local level [ 34 ]. It took about 25 years to build health centers throughout the whole country. Thus, in Finland the idea was not restricted to demonstration projects or particular regions as in some other countries, from which the idea of health center was learned [ 35 ]. The North Carelia Project, which received widespread international recognition as an example of broad community action for public health initiated by the local health centers [ 36 ], was developed as a demonstration project specifically to reduce the high mortality rate from cardio-vascular diseases, in the rural and less prosperous part of Finland. Given this background, the WHO concept of Primary Health Care as expressed in the Alma-Ata Declaration (1978) was not foreign to Finnish health policy experts. Rather, many of them felt that Finns were pioneers significantly contributing to the development of the WHO policy and demonstrating its applicability also in the Northern hemisphere [ 21 , 37 ]. However, transforming into practice the radical idea of the local health center carrying out "people's health work" was not a simple task. Since the initial expansion phase, the developmental activities and reforms of the health centers have mostly focused on improving the medical cure and care functions [ 19 ]. According to some evaluators, health promotion, community-based prevention and public health have largely been pushed to the margins. The emphasis of main reforms addressing the health centers have focused on the management of diseases, division of labour between health centers and hospitals and the development of the GP function in medical care [ 35 ]. Thus, the radical concept of Alma-Ata was, in practice, transformed into a normalized concept of primary medical care. The community approach The North Carelia project was and continues to be the best known Finnish example of community action for public health. However, the evaluators of Finnish health promotion policy have repeatedly expressed critical assessment of the leadership and implementation of community action at the local level [ 19 , 20 ]. It has not been mentioned as the strong or innovative part of the Finnish health promotion policy. Finland also used to be a dissident in resisting the managerialistic idea advocated by the WHO Regional Office for Europe to manage health promotion policy by setting the policy aims in the form of numerical health improvement targets [ 38 ]. At the beginning of the 2000s, reference to the role and responsibility of local actors and local community is an essential part of health policy rhetoric [ 39 ]. The latest national health promotion programme "Health 2015" [ 18 ] is also built around numerical health improvement targets. Thus, we may conclude that both the managerialistic approach and the community approach to the redesign of health policy in the welfare state have been introduced to national health promotion policy rhetoric. However, at the same time as they are present, the evaluators have indicated that these approaches are not effectively implemented. Healthy public policy One aspect of the rapid expansion of the Finnish welfare state in the early 1970s was the idea of improving people's health through a comprehensive planning system of all public sectors. Health indicators were to be used to provide feedback on the health impact of developments in the various public sectors and policies in these sectors should be adjusted accordingly [ 40 ]. Alcohol taxation and restrictions on its availability had already been used in Finland, mainly to reduce alcohol related criminality and social problems, but now the same policies were motivated primarily by public health concerns [ 41 ]. A comprehensive nutrition policy to change the traditional Finnish diet rich in fatty dairy products and poor in vegetables and fruit was developed. In addition to health education, policies such as shifting the priorities in the subsidies of agricultural products and negotiating changes in the dietary practices of the catering services in the schools and workplaces were used to reduce the consumption of high fat dairy products [ 42 , 43 ]. Tobacco control policies were developed as a flagship of the new health promotion policy applying high excise taxation, restrictions in the availability of tobacco and a ban on advertising. This policy was continuously tightened from the Tobacco Law of 1977 to the late 1990's [ 44 ]. Environmental health was also a rapidly developing sector, both as a part of occupational health and as a part of the development of overall environmental legislation and administration, particularly in the 1980s. The Finnish record on developing policies outside the health sector to promote health has been referred to in placing the country among the forerunners of the Health for All policy in Europe [ 19 , 20 ]. In any case, Finland may be taken as an example of combining ambitious and rapid welfare state building with the ambition of promoting health through the development of the health impact of other policy sectors. It is less obvious that Finland could be taken as an example of how to do this in more mature welfare states. We may, rather, argue that the maturing of the Finnish welfare state from the late 1980's on has been paralleled by growing problems in the development of healthy public policies. The most dramatic example is the dismantling of the traditional Nordic alcohol control policies in the process of redesigning the welfare state under the pressures of European single market legislation and globalization [ 45 ]. The existence and at least partly increasing inequity in health between different socio-economic population groups has also been taken as an indication of the less successful development of healthy public policies [ 46 ]. Paradoxically, the strengthening of the capacity of the sector to promote health has also separated it from the mainstream health promotion policy. Development of environmental policy and policy administration has contributed to the growing distance in policy discourse and policy communities of environmental and public health. The latest international evaluation of the Finnish national health promotion policy [ 20 ] gave a critical assessment of the capacity of health policy makers to assess and influence the policies of other sectors. Portugal and the HFA challenge The Carnation Revolution in 1974 ended a long period of authoritarian rule in Portugal and opened the door to the democratization of the country. As in the other Southern European countries, the democratic Constitution was of a progressive nature while conferring wide economic, social and cultural rights and duties on the citizens [ 47 ]. The Constitution that came into force in 1976 aimed at the creation of a welfare state as a political form of transition to a socialist state and society [ 48 ]. Although the goal of a socialist, classless society was removed from the Constitution in its reform in 1982, the state's responsibilities to guarantee the economic, social and cultural rights of its citizens were left untouched [ 49 ]. Welfare state remained the ultimate goal, but the socialist model was changed to the model of social protection the European Economic Community (EEC) advocated [ 50 , 15 ]. The Southern European welfare state is a relatively recent addition to the conceptual map of European welfare state models. Many southern countries' present day characteristics are related to the legacy of authoritarianism, as well as to the historically strong presence of the Catholic Church [ 51 ]. Leibfried sees the weak institutionalization of constitutional promises of social rights as a characteristic feature of Southern European welfare states [ 47 ]. The term semi- institutionalized welfare state can be used to describe the whole of the Southern European welfare state that has been built up in principle, yet not implemented in practice. On the other hand it is recognized that southern welfare states have during recent decades been catching up the more developed European welfare systems [ 47 , 51 ]. But in spite of the catching-up effect and the overall pressure towards convergence of social policies in the European Union, Southern European countries seem to maintain a relatively distinct type of welfare state [ 52 , 53 ]. Portuguese welfare state development seems to follow the southern pattern, and Portugal is here analyzed from the viewpoint of the Southern European welfare state type. The notions of semi-institutionalization and catching up-effect conceptualize the Southern European welfare state on the one hand as a developing (vs. mature) welfare state and on the other hand as following a different path than the more northern European welfare states [See [ 47 , 52 ]]. The attempts to institutionalize welfare state in Southern Europe occurred simultaneously with the era of welfare state crisis. Consequently, the crisis rhetoric was assumed in Portugal in the initial phases of welfare state development. Thus the welfare state was declared to be in a state of crisis before it actually even existed [ 53 ]. Due to the dynamics of crisis before maturation , welfare state has remained to some extent a semi- institutionalized promise until the present day. The development of Portuguese health policy can be broadly divided into two historical phases that are linked to the general welfare state development. The first period from the beginning of the 18 th century until 1971 was dominated by preventive public health policies. Through general preventive measures, such as sanitary education, environmental sanitation, hygiene, mental hygiene and sickness prevention "sanitary police" ( polícia sanitária ) aimed at governing the health of the nation. Preventive policies were directed towards the collectivity and they benefited the individual citizen only indirectly. Publicly provided health care services were tied to the clientele of social assistance and were only available to poor people until 1971, when the right to health care was legally defined to be the right of every citizen [ 54 ]. The reform bill of Health and Assistance ( Reforma de Saúde e Assistência ) established in 1971 marked the beginning of the second phase of health policies [ 55 ]. The consolidation of the universal right to health care in the Constitution and in the National Health Service (NHS) ( Sistema Nacional de Saúde ) law in 1979 [ 56 ] signified the strengthening of the social citizenship rights and changed not only the nature of health policies, but also the general nature of the Portuguese welfare state. The qualitative change in the welfare policies from the distributive to productive policies happened precisely in the area of health [ 53 ]. Primary care Maternal and child health were already part of health policy during the authoritarian era, and women's and children's health was also included into the primary health care concept established with the Reform of Health and Assistance [ 57 ]. However these programs were limited to the health education and medical monitoring of women's and children's health during and after pregnancy as family planning was prohibited for political and religious reasons until 1974. A right to family planning was legally defined in the Constitution of 1976 [ 58 ]. The integration of family planning into primary health care has widened the scope of maternal and infant health policies in Portugal. Since 1979 Portugal has been collaborating actively with WHO/UNFPA in improving services in family planning [ 21 , 59 ]. In Portuguese health strategy reproductive issues are included in various priority areas. The importance of social policies directed to women, children and family is recognized in the strategy as well as in the government programs. The policies concerned with maternal and child health have developed during the last three decades into policies of reproductive health. The indicators of maternal and child mortality have improved significantly and are on the level of other EU countries [ 60 ]. The reform of Health and Assistance aimed at creating a nationwide network of local level health centers that were supposed to provide primary health care services for the entire population [ 61 ]. Although the full implementation of this reform was hindered due to political and organizational obstacles, it is seen to mark the beginning of a new era of expansion in Portuguese public health policy [ 62 , 63 ]. This reform included most of the principles of primary health care recognized in the Alma-Ata Declaration seven years later [ 63 , 64 ]. The building of a primary health care network was further consolidated in the Constitution and in NHS law. The process of building up a primary health care network was on the Government's health policy agenda from the beginning of the 1970's until 1985 (15). Analysis of scientific texts and reports on the development of Portuguese public health policy as well as the expert interviews conducted for this study in 2003 indicate that although the Declaration of Alma-Ata was used to legitimise the development of the primary health care system – at least on the level of policy stream – the adaptation of the primary health care-concept presented in Alma-Ata did not change the national policy line. A right to health care has been an essential part of the democratization process, strengthening social citizenship. Nevertheless the democratization of health care has not been linear; health was politicized following the creation of the public NHS. The critical welfare state philosophy of the liberal political cycle (1985–1995) affected the content of health policies by favoring privatizations of health care during the term of office of the Social Democrats (centre-right party) [ 63 , 64 ]. Due to continuing political and financial problems in the implementation of NHS, difficulties in access to health care services have persisted as a health policy problem. This situation has in its turn kept the development of the health care system and medical care approach in the center of the problem stream feeding the political agenda. According to some of the public health experts interviewed the clinical, curative approach of health care gained more control in the health sector's internal power sharing during the liberal cycle and at the same time the position of public health declined. The analysis of the government programs proves that at the same time the development of primary health care disappeared from governments' agenda. The crisis period of public health policy lasted a decade (1985–1995) [ 61 ]. However as the institutionalization of health care has signified the permanent centrality of services on the health policy agenda, not even the crisis period of public health did signified a great break in terms of health promotion in policy documents. Indeed some health education campaigns were launched during the crisis period [ 15 ]. The Portuguese Journal of Public Health ( Revista Portuguesa de Saúde Pública ) published a special issue dedicated to HFA in 1988. In the Editorial of the journal it is suggested that HFA2000 should in Portugal have as an objective rather "adequate health care for all" than "health for all" [ 65 ]. The general health service orientation of health promotion and disease prevention is also present on the level of government programs. The clinical, treatment-centered ethos typical of the expansion period of the health care system is dominant in the government programs 1976–2002. Health promotion and disease prevention are conceptualized as activities of primary health care and they are seen to be implemented by the medical and nursing professions [ 15 ]. Concentration on the primary health care element of the HFA-program is not only a Portuguese specialty; other Southern European countries, such as Spain and Greece, have also put weight on the development of primary health care [ 59 ]. The first health strategy is likewise disease-oriented (14 out of 27 of the priority areas are diseases), and since the health service sector is seen as the main actor of health promotion policy, the means are mainly biomedical or educative. Emphasizing rather the individual level than the structural level seems to be a more general Southern European feature in public health policies [ 66 ]. The community approach The Ottawa Charter calls the countries to strengthen community action. However, it does not explicitly define what is meant by the concept of community. In the social policy literature the term community is often understood to refer either to the network of family members, friends and neighborhoods, or to civil society, understood as a complex of social associations and non-governmental organizations. The archetype of Southern European welfare state carries the connotation of the strong and traditional role of community in welfare provision. [ 67 ] However, most comparative studies fail to mention that during the authoritarian era the civil society element of community was repressed, as free associations were prohibited by law. In Portugal only a few religious associations connected to the Catholic Church were approved by the state. Since 1974 the number of associations acting in the field of social and health issues has expanded. [ 68 ] Often the call to strengthen community action is seen from the perspective of the welfare state crisis debates. However, in Portugal the growth of the civil society element of community was not an answer to the welfare state crisis as such, but its growth should be located in the context of the recent liberation from state repression. Yet Sousa Santos [ 69 ] argues that the state restricts true citizen participation and the functioning of those associations created after 1974 as it continues to support conservative religious organizations. In the Portuguese health strategy (1999) private institutions of social solidarity (Instituições Particulares da Solidariedade Social) and non-governmental organizations (Organizações Não-governamentais) are recognized as the main representative categories of community. Strengthening partnerships with these organizations is seen as indispensable for achieving the goals set. Although these organizations are also identified as doing health promotion work, they are mainly actors in curing and caring. The second community level actor identified as relevant for health promotion activity is the local level of public administration. Direct citizen participation (e.g. user/consumer/patients' associations) and the need to cooperate with syndicates and health professionals are also mentioned in the strategy. However, they are not given any significant role in the program implementation. All these community categories identifiable in the health strategy seem to match the current categorizations of community actors and their partners in the social sector [See [ 68 ]]. The fourth category of community action for health promotion manifest in the form of setting-based projects of Healthy Cities (WHO), the Health Promoting Schools- network (WHO & EU) and Healthy Workplaces (EU). The first Healthy City was established in 1995 and now there are 9 cities belonging to the national network of Healthy Cities [ 70 ]. The Health Promoting Schools- network was initiated in 1994 and reaches currently one third of pupils in the public education system [ 71 ]. These projects represent the model of community action that is unique for the domain of health promotion. This kind of community based action model targets the whole population of a certain community, while the traditional actors in social and health sectors concentrate on caring for and curing those who are in need of care. Targeted solidarity of traditional community action is challenged by universal equality dominant in these health promotion projects. The model of community action adapted with these projects introduced new ideology and forms of organization into the sphere of public health. When analyzing the adaptation of HFA in a timeframe it seems that the community level adapted HFA philosophy before the national level. Healthy Public Policies The Ottawa Charter emphasizes the role of policy as a factor promoting healthy choices. In other words, this means that health should be taken into consideration in all public policies. When analyzing the Portuguese development in relation to intersectoral policies, there is action in conventional intersectoral issues, such as tobacco, alcohol and nutrition, but it does not seem that any major development has happened in these policy domains. The project of Healthy Schools and the overall health education campaigns are based on interministerial cooperation and pacts between the Ministry of Health and the Ministry of Education. Intersectoral work is also carried out in the field of drug addiction [ 14 , 15 ]. In this section we focus on one case of public policies, that of sanitation, and observe its development in the welfare state development context. Portuguese public health indicators have shown remarkable improvements during the last three decades. The fact that public health indicators have been improving side by side with general socio- economic indicators has led researchers to conclude that although the creation of NHS and the improved access to health care have influenced the positive evolution of the health status of the Portuguese population, these improvements are greatly connected to general improvements in economic and social conditions, such as education, income and living standard, housing, sanitation, hygiene, and transport infrastructures [ 72 , 73 , 54 ]. These improvements occurred in the context of the expansion of the welfare state. In this process some of the core issues of the ancient sanitary police, such as matters of basic sanitation, have conceptualized more clearly under respective sectoral policies, out of the national health policy agenda. This reflects the administrative differentiation of state functions and sectoral differentiation of respective policies that typifies the expansion of the welfare state. Basic sanitation ( saneamento basico ) has been a priority in Portuguese post- authoritarian development policy, however in the government programs (1976–2002) basic sanitation is not recognized as a priority of health policy. Although in some programs environmental conditions and habitation are seen to influence public health and the welfare of the population, basic sanitation is not explicitly considered either as a health policy problem, or as a goal or means. Basic sanitation is not conceptualized as an issue of health policy, it is not explicitly on the government's health policy agenda, neither is health used as an argument to improve it in other sectors. Only in the XIII Government Program (1995–1999) are water quality and the intersectoral action needed to reach it mentioned in the section dealing with health policy. Apart from this, the issue of basic sanitation has become conceptualized as an issue of renovation of infrastructure, and this discourse has constituted it as a policy of infrastructure and renovation. In the Regional Development Plan (2 nd Community Support Framework 1994–1999) basic sanitation is conceptualized as an issue of environment and no reference is made to health [ 74 ]. In the national health strategy, healthy environments refer to social environments and basic sanitation is not conceptualized as a policy action area. The differentiation of sanitation from the domain of health policy implies that although a change clearly came about in the content of "healthy public policies", it did not happen towards new public health as the improvement of sanitation was not justified by health reasons. Some of the recent documents [ 75 ] imply that in recent years the development has begun to turn in a different direction as issues of basic sanitation are again included in the domain of health policies. Discussion and conclusions "Health for All" was developed as an international synthesis of emerging health policy ideas of the 1970's, sometimes conceptualized as "the new public health". Reflecting both the many roots of the concept and the many different contexts to which it was to be adapted, different interpretations of HFA have coexisted. The Alma-Ata Declaration was adapted to combining new public health with local socio-economic development in the developing countries. The HFA targets of the WHO European Region and the Ottawa Charter combine the new public health with the reform demands of state capitalistic and state socialistic welfare states. The target approach is closer to the managerial reform agenda while the Ottawa approach seems to lean more on the community empowerment agenda. HFA was launched to contribute to the development of national health policies. Thus it may be used as a standard for evaluating national health policies and health promotion policies, as has been done in some studies inspired by the WHO [ 7 - 10 ]. However, understanding HFA as a synthesis of many policy tendencies and allowing different contents for different policy contexts makes such direct comparisons between national policies and WHO documents problematic. In the policy transfer perspective the role of the WHO (or, for that matter, of the EU) may not be that of an international policy leadership but, rather, that of an international policy mediator. We have tried to trace the impact of HFA on the development of the Finnish and Portuguese health policies. The Finnish development of "people's health work" and local health centers was clearly inspired by the same ideas as the primary health care concept of the Alma-Ata Declaration. The Portuguese health policy ideology expressed in the reforms of 1970's also comprehended the ideas of Alma-Ata Declaration. However, neither of these can be seen as a transfer from WHO to the member states. Rather, the Finns claim that the direction of the transfer was from Finland to WHO. The Portuguese primary care concept also had its own national roots, e.g. in the pre-revolution development of maternal and child health. The subsequent development of primary health care in both countries indicates that the Alma-Ata idea of broad primary care tends to contradict the welfare state reforms inspired by the ideas of the New Public Management. This context tends to reduce primary health care to primary medical care. The impact of this change in the welfare state context may be identified both in Finland and Portugal from the 1980's on as well as in comparing the primary care concepts of the Alma-Ata HFA and the HFA targets of the WHO Europe. At the same time, the aim of the Ottawa Charter of reorienting health services towards health promotion does not seem to have guided primary care development in either country. Thus the development of primary care in both countries has been in dialogue with the HFA. However, what primary care means in the framework of HFA has changed over time and the dialogue cannot be simplified into the unidirectional transfer of HFA policy from WHO to member states. Dialogue or interaction are also appropriate concepts to describe the relationship between WHO and the two countries with regard to developing a community approach in health promotion policy. First of all, the different variants of HFA locate "community" in different contexts. In Alma-Ata, community is the totality of local actors without making distinctions between economic, social and health actors or private and public actors. The European HFA target documents [ 4 , 6 ] see community as a partner or a cluster of partners to the health sector and public authorities. The Ottawa Charter seems to be build around the idea of community empowerment and increasingly participative health policy making. The Finnish community approach as expressed in the North Carelia project, in the cooperation of the public health sector with the traditional public health associations and in the emphasis on local public sector action, seems to be quite close to the approach of the European HFA targets. Both the broad community concept of Alma-Ata and the community empowerment approach of Ottawa seem more alien to Finnish health policy strategies. A number of welfare state characterizations [e.g. [ 47 , 67 ]] create expectations that we should find, in Portugal, a strong role of traditional communities strongly linked to the Catholic Church in health promotion policy. Such an expectation may fail to recognize the historical legacy of the authoritarian Salazar regime, which, while keeping close linkage to the Catholic Church, was quite a state centered regime that did not allow strong independent community action. Our analysis indicates that the role of community action in health promotion is not particularly eminent in Portugal, either in governmental health policy documents [ 14 , 15 ] or according to the opinion of public health experts [ 61 , 76 ]. The activity of the Catholic Church and religiously inspired organizations in health promotion is, however, visible [ 77 , 78 ]. But so is also the attempt of the government to conceptualize community action through projects such as Healthy Cities and Health Promoting Schools, where community action is led or arranged by the public authorities. Thus, whatever is meant by the community approach in health promotion policy, Finland and Portugal do not seem to be strong examples of policy development following the initiative of HFA. We could not identify policy transfer other than in participation in the Healthy Cities and other "health settings" projects. Healthy Public Policy was our third focus in health promotion policy. The concept was raised in the European HFA document in 1985. In Alma-Ata the integration of health and other policies is extended much further and no specific concept resembling health public policy is needed. The Finnish health promotion strategy has included a number of public policies outside the health sector, particularly with regard to alcohol, tobacco, nutrition and physical exercise. We could not identify any specific impact of the different HFAs of the WHO on these policies. Rather, there seems to be growing pressure to restrict the use of the impact of other sectors in alcohol control. At the same time, the distance between the mainstream health promotion policy and environmental policy seems to be growing, although the public health impact of environmental policies is obvious. Thus, with the exception of tobacco policy, the idea of healthy public policy may even experience increasing problems, although this is not so far reflected in the development of the health status of the population. The rapid positive development of the health status of the Portuguese population during the last 30 years reflects the rapid improvement of the sanitary conditions as well as of the social determinants of health [ 72 , 73 , 54 ]. Sanitary policy, including both preventive services such as vaccinations and health education, as well as improvement of environmental and housing conditions, has been the most significant aspect of Portuguese healthy public policy. However, the analysis of health policy documents indicates that Portugal has also experienced a distancing of environmental policies and health policies, that is: a trend antagonistic to the ideas of the different version of HFA. Other public policies, including tobacco and alcohol control and nutrition policies are weakly developed in Portugal. Thus, we cannot identify any significant transfer mediated by the WHO in Portugal either. At the beginning of 1970's public health indicators showed that Finland and Portugal were lagging behind the majority of Western European countries in terms of public health indicators. Both defined this distance from the Western European level as the core health policy problem [ 15 ]. This way of defining the policy problem has clearly contributed to the fact that both countries have looked to international organizations and international comparison for their policy development. The Finnish health policy expert community has often referred to WHO and Finland has been an active member of the European region of this organization. In the 1980's, it even took the responsibility for acting as a pilot country for the national development of HFA in Europe [ 37 ]. Thus there has been much interaction between WHO and Finland in health policy development. Our analysis indicates that this interaction cannot be understood as policy transfer and that it has influenced Finnish health policy development much less than is often assumed. For the Portuguese government documents, the EU and the idea of a "European welfare state" has been the reference much more often that the WHO [ 15 ]. However, Portugal has also been in dialogue with the WHO in health policy development, although not to the same extent as Finland. We have also asked what conditions the adoption of HFA policy in the two countries. Our analysis indicates that the phase of welfare state development matters a lot. The ambitious welfare state development period in the late 1960's and the 1970's in Finland was a good basis for adopting the ambitious idea of "people's health work" and setting far-reaching aims for the development of the health impact of all public policies. Much of the Finnish health promotion policy development until the 1990's is rooted in the initiatives of this period. HFA, as expressed in the Alma-Ata Declaration and in the later versions of HFA were taken in Finland as international evidence in support of the policy choices already made in the country. Portugal also had courageous ambitions of developing a European welfare state, after the Carnation Revolution and the call of Alma-Ata was heard in this context. While Finland was fairly successful in building a universalistic institutional welfare state of the Scandinavian type, Portugal seems to have so far ended up in what Leibfried (1992) calls a semi-institutional welfare state. This may be a good explanation for the continuity of health promotion policy in Finland, in contrast to the discontinuity in Portugal which also is reflected in the concept "semi-institutional". Both HFA and the two countries examined have also been influenced by the end of the "Golden Age of the Welfare State" [ 79 ]. The differences between the Alma-Ata approach and those of the Ottawa Charter and European HFA expressed in policy targets is not only the difference between global and Europe or OECD. It is also a difference between the ambitions of the Golden Age and the post-expansion period [ 80 ] of the Welfare State. Now the political agenda is dominated by the idea of reforming the (existing) welfare state. We have linked the Ottawa approach to a reform agenda emphasizing community empowerment and the European HFA targets approach to the more managerialistic reform agenda. While we can identify the impact of the managerialistic agenda in both countries to the reduction of "primary health care" to "primary medical care", we are more hesitant regarding the impact of the community empowerment agenda on the health policy development in the two countries. The development of health promotion policy in the two countries has also been related to changes in politics, particularly to changes in the political composition and orientation of the national governments. In this regard, the Portuguese development has been stormier with a radical regime shift in the Carnation Revolution, starting a new regime inspired by socialist visions, followed by a turn to liberal conservative governments ten years later. The Socialist Party's victory in the elections of 1995 after ten years in opposition signified a change in social policy orientation once again [ 50 , 81 ]. However, there does not seem to be any significant changes in the health promotion policy even if the government has published health strategy. The Finnish political development has been much less stormy. The tradition of coalition governments which normally include both the Social Democratic Party and some parties of the bourgeois side has strengthened continuity rather than radical turns in Finnish health policy. However, within this continuity, an incremental movement from welfare state expansion to post-expansive welfare state reform policy may be identified [ 82 - 84 ]. Our analysis does not give a clear picture of the significance of politics in the adoption of HFA in the two countries. We may assume that the continuity in the Finnish politics has contributed to the continuity in the Finland-WHO dialogue and interaction as expressed, e.g., in the reviews of WHO-teams of Finnish health policy development [ 19 , 20 ]. The more stormy political development in Portugal may also have caused more discontinuity in the WHO relationship. However, the level of interaction was not a direct indicator of significant policy transfer. Rather, our analysis shows that the political context and its changes in countries probably impacts on which version of HFA is adopted. Thus, during the dominance of politics in support of more radical or expansionist welfare state development, Alma-Ata seems to have been the preferred version of HFA, while the European HFA target approach may be more feasible in the post-expansive welfare state politics. In Kingdonian (1995) terms, we may sum up that the health policy problem of both Finland and Portugal, being European laggards in the 1970's, caused them to be open to transfer of policies from abroad. Thus, the "problem stream" was ready for policy transfer. The "policy stream" seems also to have been ready for a certain kind of transfer, but only for those versions or elements of HFA that could be fitted into the specific policy contexts of the countries. HFA as such was not a dynamo of policy change in either country. The "politics stream" changed in both countries so that the window for radical policy changes was closed fairly soon. After that, if any political window was open, it was only for incremental changes in line with the post-expansive welfare state reform agenda. Competing interests This study is a part of a research project called "Finnish National Health Promotion Policy from an International Comparative Perspective", which has been financed by the Academy of Finland. JL acted as a scientist in WHO Centre for Health Policy, Brussels, in 1999. Authors' contributions JL is responsible for the analysis of Finnish policy while LTG has done the analysis on Portugal and drafted the other parts of the article.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544400.xml
551596	Critical role of hnRNP A1 in HTLV-1 replication in human transformed T lymphocytes	Background In this study, we have examined the role of heterogeneous nuclear ribonucleoprotein A1 (hnRNP A1) in viral gene expression in T lymphocytes transformed by HTLV-1. Results We have previously observed that hnRNP A1 (A1) down-modulates the post transcriptional activity of Rex protein of HTLV-1. Here, we tested whether the ectopic expression of a dominant negative mutant (NLS-A1-HA) defective in shuttling activity or knockdown of the hnRNPA1 gene using RNA interference could inhibit Rex-mediated export of viral mRNAs in HTLV-1 producing C91PL T-cells. We show that the expression of NLS-A1-HA does not modify the export of Rex-dependent viral mRNAs. Conversely, inhibiting A1 expression in C91PL cells by RNA interference provoked an increase in the Rex-dependent export of unspliced and singly spliced mRNAs. Surprisingly, we also observed a significant increase in proviral transcription and an accumulation of unspliced mRNAs, suggesting that the splicing process was affected. Finally, A1 knockdown in C91PL cells increased viral production by these cells. Thus, hnRNP A1 is implicated in the modulation of the level of HTLV-1 gene expression in T cells transformed by this human retrovirus. Conclusions These observations provide an insight into a new cellular control of HTLV-1 replication and suggest that hnRNP A1 is likely part of the regulatory mechanisms of the life cycle of this human retrovirus in T cells.	Background The human T cell leukemia/lymphotropic virus type 1 is the etiologic agent of adult T cell leukemia (ATL), an aggressive and fatal leukemia of CD4+ T lymphocytes [ 1 , 2 ] and is also associated with a neurological demyelinating disease, tropical spastic paraparesis (TSP) or HTLV-I associated myelopathy (HAM)[ 3 ]. Infection by HTLV-1 transforms T cells in vitro and in vivo, a process that has been associated with upregulation of specific cellular genes involved in T cell activation and proliferation during the course of viral infection [ 4 - 6 ]. The completion of the replication cycle of HTLV-1 leading to the production of new particles is dependent on two non-structural HTLV-1 encoded regulatory proteins, Tax and Rex, which act at the transcriptional and post-transcriptional levels, respectively [ 7 , 8 ]. The 40-kDa Tax protein trans-activates transcription of the provirus, through its interaction with cellular transcription factors and with Tax response elements present in the 5' long terminal repeat (LTR). The post-transcriptional activity of the 27-kDa Rex protein, an RNA-binding protein, is mediated by its interaction with the Rex response element (XRE) located on the U3/R region of the 3'LTR present on all viral transcripts [ 9 ]. When expressed at a critical threshold, Rex is able to direct the cytoplasmic expression of unspliced gag-pol and singly-spliced env mRNAs, at the expense of the multiply-spiced tax/rex mRNA [ 10 , 11 ]. We have recently reported that heterogeneous nuclear ribonucleoprotein A1 (hnRNP A1) interferes with the binding of Rex to the XRE, thus leading to a functional impairment of this viral protein [ 12 ]. The ubiquitously expressed hnRNP A1 is an abundant nuclear protein that participates in RNA processing, alternative splicing and chromosome maintenance as well as in the nucleocytoplasmic transport of mRNAs [ 13 - 18 ]. This protein contains two RNA-binding domains and a glycine-rich domain implicated in protein-protein interactions. Predominantly located in the nucleus, this cellular protein has the ability to shuttle continuously between the nucleus and the cytoplasm [ 19 - 21 ]. The signal that mediates both nuclear import and export has been identified as a 38-aa sequence, termed M9, located at the C-terminus of hnRNP A1, and is involved in the nucleo-cytoplasmic trafficking of mRNAs [ 22 ]. As indicated above, we have provided evidence that hnRNP A1 impairs the post-transcriptional regulation of HTLV-1 gene expression, by interfering with the binding of Rex to the XRE [ 12 ]. In the present study, we first demonstrate that the mutation of a putative binding site of hnRNP A1 to the XRE leads to an increase of the post-transcriptional activity of Rex. Next, to further address the effect that hnRNP A1 might exert on viral replication in vivo, we elected to investigate its implication in HTLV-1 producing T cells. Two experimental approaches were implemented: impairment of the functional activity of the endogenous hnRNP A1 by ectopic expression of a dominant negative mutant and knockdown of the hnRNPAl gene expression using RNA interference (siRNA). We report that inhibition of hnRNP A1 expression and functionality were achieved, leading to an increase of viral transcription together with an increase of cytoplasmic expression of viral mRNAs and of viral production. These observations by providing insight into a new cellular control of HTLV-I replication, suggest that hnRNP A1 is likely part of the regulatory mechanisms of the life cycle of this human retrovirus. Results A putative hnRNP A1 binding site has been identified, close to the minimal Rex binding site in the stem-loop D of the XRE (Fig 1A ). To further evaluate the role of this binding site in the impairment of the functional activity of Rex, two punctual mutations were performed in the CMV/XRE vector containing the indicator luc gene (Fig 1B ). These mutations modify the UAGGUA sequence into CCGGUA, and the UACCUA sequence into UACCGG, respectively, thus generating the CMV/mutXRE vector. Either vector (CMV/XRE and CMV/mutXRE), or the control vector (CMV 128, containing only the luc gene) were then transiently transfected in Jurkat cells in the absence or in the presence of a Rex-expressing plasmid. It was observed that, in presence of Rex, luc expression in cells transfected with the CMV/mutXRE vector was more than 3-fold higher than that in cells transfected with the CMV/XRE vector (Fig 1C ). These results indicate that the putative hnRNPAl binding site close to the Rex binding site on the SLD sequence in the XRE is directly or indirectly implicated in down-modulating the post-transcriptional activity of Rex. Since the mutations affect a putative binding site for hnRNP A1, these results suggest that hnRNP A1 might be the effector of this down-regulation. To further delineate how this cellular protein perturbs the life cycle of HTLV-1, we elected to investigate its implication in HTLV-1 producing T cells. Two experimental approaches were implemented: impairment of the endogenous hnRNP A1 by ectopic expression of a dominant negative mutant (NLS-A1-HA) defective in shuttling activity and knockdown of the hnRNP A1 gene using RNA interference (RNAi). Figure 1 Functional characterization of HTLV-1 mutated XRE sequence . (A) Schematic representation of the HTLV-1 XRE. On the left, the XRE corresponds to U3 and R sequences within the HTLV-1 long terminal repeat, and consists of four stem-loops. On the right, the predicted secondary structure of the stem-loopD (SLD) with the minimal Rex binding site and the mutations introduced within the putative hnRNP A1 binding site are indicated. (B) Schematic view of the reporter plasmid CMV/XRE. (C) Effect of mutations within the XRE sequence on the Rex trans-activation capacity. Jurkat cells were transfected with 1 μg of the indicated reporter plasmid in the presence or not of Rex expression plasmid (200 ng) and the constitutive internal control tk-renilla luciferase vector (10 ng). Data are expressed as normalized luciferase activity and the error bars represent the standard deviations from three independent experiments. A nucleus-localized shuttling-deficient hnRNP A1 mutant does not affect the post-transcriptional activity of Rex The NLS-A1-HA construct contains the bipartite-basic type NLS of hnRNP K fused in frame with the N-terminus of an HA-tagged hnRNP A1 mutant, which lacked both nuclear import and export activities and inhibits hnRNP A1-dependent mRNA export when microinjected into nuclei of Xenopus laevis oocytes [ 22 , 23 ]. This hnRNP A1 mutant which retains the hnRNP A1 nuclear localization, lacks nuclear export activity [ 24 ]. As such, the nucleus-localized NLS-A1-HA has the potential to compete with wild-type hnRNP A1 for binding to mRNAs, and for its nuclear export. A retroviral vector LXSP-NLS-A1-HA was used to ectopically express this dominant negative mutant in the HTLV-1 transformed C91PL T cells. In these cells, Rex governs the cytoplasmic accumulation of unspliced ( gag/pol ) and singly-spliced ( env ) mRNAs. After a few days of culture in presence of puromycin, immunostaining of the resistant population revealed that about 30% of the cells were displaying HA labelling (Fig. 2 ). Dual immunostaining indicated that both endogenous hnRNP A1 (anti-hnRNP A1, red) and ectopically expressed NLS-A1-HA (anti-HA, green) displayed a nuclear diffuse staining excluding the nucleoli. Figure 2 Expression of a dominant negative mutant of hnRNP A1 in HTLV-1 producing C91PL cells . Confocal microscopy of untransduced (a) or NLS-A1-HA transduced (b)-C91PL cells after dual immunofluorescence staining with anti-HA (green) and anti-hnRNP A1 (red) antibodies; the right panels show the overlay of the green and red staining; We next investigated whether overexpression of this defective hnRNP A1 mutant was interfering with the expression of viral mRNAs. Quantification of the nuclear and the cytoplasmic levels of unspliced gag/pol , singly spliced env and doubly spliced tax/rex mRNAs was performed by RQ-PCR involving pair of primers specific of each viral mRNA (Fig. 3A ). The comparative analysis of the viral mRNAs expression pattern between the control (LXSP) and NLS-A1-HA cells revealed a small increase of unspliced gag/pol and of doubly spliced tax/rex mRNAS in the latter, whereas no modification was observed for the singly spliced env mRNAs (Fig. 3B ). The ratio of nuclear to total RNA and that of cytoplasmic to total RNA allowed to calculate a nuclear export rate (NER). Whereas the cytoplasmic expression of tax/rex mRNAs was slightly enhanced in cells expressing the NLS-A1-HA mutant, the NER of the unspliced and singly spliced mRNAs was not affected (Fig 3C ). As the cytoplasmic expression of these mRNAs is Rex dependent, these results indicate that the ectopic expression of the NLS-A1-HA mutant in C91 PL cells does not interfere with the functionality of Rex. However and surprisingly, a more than 4-fold increase of the p19 gag amount in the supernatant medium of NLS-A1-HA-transduced cells (2786 ± 154 pg/ml) was observed, when compared to the respective control cells (678 ± 104 pg/ml). Taken together, these results indicate that the impairment of the hnRNP A1 functionality might favour the translation of cytoplasmic viral mRNAs. Figure 3 Effect of ectopic expression of a dominant negative mutant of hnRNP A1 in HTLV-1 producing C91PL cells . (A) Primer location on HTLV-1 mRNA; (B) Analysis of the nucleo-cytoplasmic distribution of viral gene expression in NLS-A1- and LXSP- transduced cells. Four days after transduction, mRNAs were extracted from the nuclear and cytoplasmic compartments of each cell type and levels of unspliced ( gag/pol ), singly spliced ( env ) and doubly spliced ( tax/rex ) mRNAs were reverse transcribed and quantified by real-time quantitative PCR (RQ-PCR), by using specific primers. Results are expressed as the amount of nuclear (grey bar) and cytoplasmic (black bar) indicated mRNA relative to β-actin. (C) Evaluation of the nuclear export rate (NER) of Rex-dependent ( gag/pol plus env ) mRNA and of Rex-independent ( tax/rex ) mRNA in NLS-A1- or LXSP- transduced C91PL cells. Numbers are the ratio between cytoplasmic (C) to total (T) RNA and nuclear (N) to total RNA. Efficient inhibition of hnRNP A1 by retrovirus-delivered siRNAs We next evaluated whether HTLV-1 replication is modulated by RNA interference with hnRNP A1 gene expression. To that aim, two oligonucleotides encoding siRNA directed against hnRNP A1, one targeting an RNA sequence located on the 5' end (34-nt after the translation start site), and the other an RNA sequence close to the 3'end (548-nt after translation start site) were each inserted in the pRS retroviral vector [ 25 ], as indicated in Materials and Methods. Both pRS-siRNA+34 and PRS-siRNA+548 vectors, as well as the pRS empty vector were used to produce recombinant retroviral particles used to transduce Jurkat T cells at a multiplicity of infection (m.o.i.) of 5. After four days of puromycin selection to eliminate nontransduced cells, the siRNA mediated-depletion of hnRNP A1 mRNAs was measured by quantitative RT-PCR. While targeting the 5'end (+34) was found inefficient, targeting the 3'end (+548) reduced the level of hnRNP A1 transcripts to 10% of those detected in untransduced Jurkat cells or in Jurkat cells transduced with empty (pRS) retroviral particles (Fig. 4A ). Importantly, the siRNA-mediated reduction in A1 levels did not provoke cell death. Immunoblotting analysis of the PRS-siRNA +548 cells showed a strong reduction of the hnRNP A1 protein level, when compared to that in the pRS-siRNA+34 cells and in control cells (Fig 4B ). Furthermore, the levels of the splicing factor ASF/SF2 were not modified in these cells. These data indicate that expression of hnRNP A1 is specifically repressed in the pRS-siRNA+548-transduced Jurkat cells. Figure 4 RNAi-mediated reduction of hnRNP A1 expression in Jurkat cells . (A) hnRNP A1 mRNA levels in cells transduced with the indicated retroviruses were determined by RQ-PCR. Levels in knockdown cells are given as percent mRNA reduction relative to the level in control cells transduced with empty pRS virus. Standard deviations are from at least three determinations performed in duplicate. (B) Equal amounts of protein from either nontransduced (lane1) or transduced with the indicated virus (lanes 2 to 4) were analyzed by immunoblotting. Actin and ASF/SF2 were used as control. Note that hnRNP A1 was significantly depleted in cells transduced with siRNA+548, whereas ASF/SF2 was not affected. hnRNP A1 depletion in HTLV-1-producing T lymphocytes altered the transcriptional profile and increased the post-transcriptional activity of Rex The above described retroviral vector system was used to mediate the in situ synthesis of siRNAs and to suppress specifically hnRNP A1 gene expression in C91PL cells. Retroviruses produced from pRS-siRNA+548 and from the pRS empty vector were used to transduce these cells with a m.o.i. of 5. Four days after transduction, hnRNP A1 depletion was assessed by quantitative PCR analysis of cytoplasmic mRNAs. In siRNA-transduced C91PL cells, that transcript represented 32% of that in control pRS transduced cells (Fig. 5A ). Interestingly, a western blot analysis of cell lysates further showed that hnRNP A1 was barely detected in siRNA-transduced C91 PL cells, whereas the levels of Rex, or of hnRNP C1/C2 or of actin were found unchanged (Fig. 5B ). Furthermore, a flow cytometry analysis of siRNA-transduced C91PL cells reveals that hnRNP A1 was detected in 6.1% of these cells, whereas it was detected in about 70% of the control cells (Fig. 5C ). Figure 5 Analysis of hnRNP A1 depletion in HTLV-1 producing C91PL cells . (A) Analysis of hnRNP A1 mRNA levels in cells transduced with the indicated retroviruses. Four days after transduction, cytoplasmic RNA were extracted, reverse transcribed with oligo-dT, and levels of hnRNP A1 mRNA were determined by RQ-PCR. (B) Expression of hnRNP A1, Rex and hnRNP C1/C2 was monitored by immunoblotting of total protein extract from C91PL cells transduced with the indicated virus. Equivalent protein loading was confirmed by immunoblotting with an anti-actin antibody. (C) Detection of hnRNP A1 and p19 gag expression in C91PL cells transduced with the indicated virus. Dot plots showing both hnRNP A1 and HTLV-1 gag expressions in one representative experiment. The percentage of cells in each quadrant is indicated. We next investigated whether the decrease in hnRNP A1 expression in C91PL cells was interfering with the expression of viral mRNAs. Real-time quantitative PCR assays were performed to quantify viral mRNAs by using the same primer pairs described above. Results (from two different transduction experiments) assessing the amount of total viral mRNAs (Fig 6A ) revealed that suppression of hnRNP A1 in siRNA-transduced C91PL cells was leading to a significant increase of viral transcription (1.7 to 1.8 fold), when compared to PRS control cells. Then, the analysis of the relative nuclear and cytoplasmic levels of unspliced gag/pol , singly spliced env and doubly spliced tax/rex mRNAs indicated that the expression of unspliced gag/pol mRNA was 2 and 3-fold enhanced respectively in the nucleus and cytoplasm of siRNA-transduced C91PL cells, whereas the expression and the distribution of spliced env mRNAs were not significantly altered (Fig. 6B ). A slight increase of the doubly-spliced tax/rex mRNAs was observed in both compartments. Figure 6 Effect of hnRNP A1 depletion on viral gene expression . (A) Quantification of total viral gene expression in siRNA-transduced C91PL cells by quantitative PCR. Nuclear and cytoplasmic mRNAs were extracted from siRNA (black bars)- or control PRS (white bars)- transduced C91PL cells. Equal amounts of mRNA were reverse transcribed with oligo-dT and subjected to RQ- PCR. Results are expressed as the relative levels of total viral mRNA to cellular β-actin. Error bars indicate standard deviations. (B) Analysis of the nucleo-cytoplasmic expression of viral genes. Four days after transduction, mRNAs were extracted and analyzed as in Fig. 3B. Results are expressed as the amount of nuclear (grey bar) and cytoplasmic (black bar) indicated mRNA relative to β-actin. (C) Evaluation of the nuclear export rate (NER) of Rex-dependent ( gag/pol plus env ) mRNA and of Rex-independent ( tax/rex ) mRNA in PRS- or siRNA- transduced C91 PL cells. These results suggest that inhibition of hnRNP A1 in C91PL cells mainly correlates with a defect in the splicing of genomic mRNAs. The NER of the unspliced and singly spliced mRNAs was significantly higher in siRNA-treated cells than in control cells, whereas the cytoplasmic expression of tax/rex mRNAs, which is Rex-independent was not modified (Fig. 6C ). As the nucleo-cytoplasmic transport of the former is Rex-dependent, these observations propose that the depletion of hnRNPAl correlates with an increase of Rex activity. Finally, whereas a flow cytometry analysis indicated a similar percentage of p19 gag producing cells in siRNA-transduced C91PL cells and in control cells, the quantification of 19 gag in the supernatant medium of siRNA-transduced cells revealed a 1.5-fold increase of the p19 gag amount (1017 ± 26 pg/ml), compared to that in control cells (678 ± 104 pg/ml). Collectively, these data support that the hnRNP A1 depletion in HTLV-1-producing T cells increases viral transcription, is correlated with a defect in the splicing process at the level of the gag/pol transcript and increases the post-transcriptional activity of Rex leading to an increase of viral production. Discussion The ubiquitously expressed hnRNP A1, an RNA-binding protein, is a nucleocytoplasmic shuttling hnRNP that accompanies eukaryotic mRNAs from the active site of transcription to that of translation. As such, hnRNP A1 is involved in a variety of important cellular functions, including RNA splicing, transport, turnover and translation. We have previously shown that hnRNP A1 decreases the post-transcriptional activity of the Rex protein of HTLV-1, by interfering with the binding of the viral protein on its response element, present on the 3' LTR of all viral RNAs. Here we first report that the mutation of a putative binding site of hnRNP A1 in the XRE enhances the functional activity of Rex. This observation obtained through transient transfection experiments, confirms that A1 proteins could antagonize the post-transcriptional activity of Rex, by a competitive mechanism. We have next investigated the role of hnRNP A1 in HTLV-1 transformed C91PL cells, which produce HTLV-1 virions. These express the three differentially spliced (the unspliced gag/pol , the singly spliced env and the doubly spliced tax/rex ) mRNAs, which encode the structural and regulatory proteins. The gag/pol and env mRNAs are dependent on Rex for their cytoplasmic expression. To determine whether hnRNP A1 interferes with viral replication, we first examined the effect of the ectopic expression of an hnRNP A1 mutant (NLS-A1-HA) defective in nuclear export activity. This mutant was previously used to assess the potential role of hnRNP A1 in nucleocytoplasmic shuttling activity in normal and leukemic myelopoiesis. Interestingly it was found that the ectopic expression of this dominant negative form of hnRNP A1 resulted in the downmodulation of the nucleocytoplasmic trafficking of cellular mRNAs that encode proteins affecting the phenotype of normal and transformed myeloid progenitors [ 24 ]. In the present study, we showed that NLS-A1-HA- C91PL cells expressed a higher level of total viral transcripts than that observed in control cells, suggesting that the ectopic expression of this hnRNP A1 mutant correlated with an increased proviral transcription and/or stability of the viral RNA. Furthermore, no modification of the nuclear export rate was observed in the NLS-A1-HA-transduced C91PL cells, indicating that the activity of Rex was not impaired. Finally, as both endogenous hnRNP A1 and the NLS-A1-HA mutant, which are nucleus-localized and consequently able to access the XRE did not decrease the Rex-dependent nucleo-cytoplasmic expression of the viral mRNAs, we should therefore speculate that the simultaneous presence of both types of A1 forbids them to bind the XRE with maximal efficiency. Interestingly, the increase of p19 gag produced by the NLS-A1-HA C91PL cells suggests that the retention of the endogenous hnRNP A1 in the nucleus is favouring an increase in the translation of viral mRNAs We have then proceeded to the knockdown of hnRNP A1 gene using the retrovirus-mediated RNA interference. This system was first validated in transduction experiments performed in Jurkat T cells. A puromycin-selected population of cells was obtained in which a strong overall specific reduction of hnRNP A1 was observed. Note that this hnRNP A1-depleted Jurkat cells were not affected in their growth even for a long time culture (data not shown). This is consistent with other studies showing that si-RNA-mediated reduction in A1 levels did not affect cell division nor provoke cell death in normal cell lines [ 26 ]. We next performed siRNA depletion of hnRNP A1 in C91PL cells and have observed a significant increase in proviral transcription, as demonstrated by the higher level of viral transcripts than that in control cells (Figure 6A ). Furthermore, the level of unspliced transcripts was found to be predominant, compared to the singly-and doubly-spliced transcripts, in the hnRNP A1 depleted cells, pleading for a splicing default (Fig. 6B ). Finally, the increase of the nuclear export of unspliced and singly spliced mRNAs suggests that the knockdown of hnRNP A1 allows a better accessibility of Rex to the XRE and leads to the enhancement of the post- transcriptional activity of Rex. This is in good correlation with the increase in the production of viral particles, as ascertained by the quantification of the p19 gag protein. Since hnRNP A1 has been implicated in nuclear export of cellular mature mRNAs [ 27 ] as well as translational and/or posttranslational events of viral mRNAs (our study), it is possible that its depletion could affect the expression of several transcription and/or splicing factors, leading to an effect, for instance, on the splicing process of viral mRNAs. Of the two experimental approaches used in the present study to apprehend the implication of hnRNP A1 on HTLV-1 replication in in vitro HTLV-1-transformed T-cells, that consisting in the depletion of this cellular protein by RNA interference provides evidence for the role of hnRNP A1 in restraining the viral life cycle at both transcriptional and post-transcriptional levels. We conclude from these findings that down-regulation of hnRNP A1 has an important role on the replicative potential of HTLV-1 in T lymphocytes. Consequently, these data allows us to define hnRNP A1 as a cellular protein endowed with an anti-HTLV-1 activity. Methods pRS construct directing the synthesis of siRNA and Plasmids The vector pRetro-SUPER (pRS) was used to generate biologically active siRNAs from the Pol III H1-RNA gene promoter [ 25 ]. Two annealed 64-bp synthetic oligonucleotides were used: 5'-gatccccAGCAAGAGATGGCTAGTGCttcaagagaGCACTAGCCATCTCTTGCTtttttgga aa-3', and 5'-gatccccCAGCTGAGGAAGCTCTTCAttcaagagaTGAAGAGCTTCCTCAGCTGtttttgga aa-3'. The sequence of each oligonucleotide was designed (Oligoengine) to encode two 19-nt (in capital letters) reverse complements homologous to a portion of hnRNP A1 (nucleotides 34–53 for the first construct, and nucleotides 548–567 for the second one) separated by a 9-nt spacer region, and ending by Bgl II and Hind III sites. Each oligonucleotide was then introduced into pRS resulting in either pRS-siRNA+34 or pRS-siRNA+548 retroviral vectors, respectively. Plasmids pgagpol/MLV and EnvVSV-G were kindly provided by F.L. Cosset (U412-Lyon). LXSP-NLS-A1-HA and empty LXSP retroviral vectors were a kind gift of D. Perrotti and has been described previously [ 23 , 24 ]. For reporter gene analyses, the luciferase plasmid (CMV/XRE) was derived from the reporter plasmid pDM138 containing the CAT gene and the XRE sequences [ 28 ]. It expresses, under the control of the cytomegalovirus promoter, a two-exon, one-intron precursor RNA in which the luc gene and the XRE are located within the intron (see Fig. 1B ). The mutant plasmid (CMV/mutXRE) was generated using a site-directed mutagenesis kit (Stratagene) according to the manufacturer's instructions, and with the following primer, 5'-AAAGCCCTGTCAAAACAGGAAATGGCAAGCGCTTCATCCAGCC-3'. This construct was verified by DNA sequencing before use in transfection. The rex- expression plasmid, containing the wild type Rex sequence under the control of the cytomegalovirus promoter, was a gift from B.C. Cullen. Cell culture and DNA transfection Jurkat lymphoblastoid T-cells were incubated at 37°C in a 5% CO2 atmosphere, in RPMI-1640 medium (Invitrogen) supplemented with 10% heat-inactivated fetal calf serum (FCS) and 20 IU/ml penicillin, 20 μg/ml streptomycin. The HTLV-1-transformed T-cell line, C91PL [ 29 ] was cultured in complete RPMI medium. The human epithelial 293T cells and the human rhabdomyosarcoma TE cellswere cultured in Dulbecco's minimum eagle medium (DMEM, Invitrogen) supplemented with 10% FCS and 20 IU/ml penicillin, 20 μg/ml streptomycin. These cells seeded at 1.2 × 10 5 cells per well of a 12-well plate were transfected using the calcium phosphate coprecipitation technique [ 30 ]. Jurkat cells were transfected by using the X-treme GENE Q2 transfection reagent (Roche Molecular Biochemicals) according to the manufacturer's indications. The amount of plasmid used in each transfection assay is indicated in the figure legends. To assess the efficiency of the transfection assay, 10 ng of the tk-renilla Luciferase plasmid (Promega) were co-transfected in each assay. Cells were harvested 24 h after transfection, resuspended in 100 μl of passive lysis buffer (Promega) and assayed for both firefly and renilla luciferases by using a Dual-Luciferase Reporter assay system (Promega). Preparation of viral stocks and transduction of T cells Fresh viral stocks were prepared by transfecting 293T cells (seeded at 5 × 10 5 cell/well of a 6-well plate) with 2 μg of pRS or pRS-siRNA together with 1 μg of pgag-pol/MLV and 0,45 μg of env/VSV-G with ExGen 500 reagent (Euromedex). Twelve hours later, the cells were washed once with PBS, and newly produced virions were harvested over 24 h in 1,5 ml of fresh medium. Viral supernatants were clarifed by passage through a 0.45-μm syringe filter and aliquots were stored at -80°C. Titers of virus stocks were determined by infecting rhabdomyosarcoma human TE cells (60% confluent) with serially diluted viral stocks. After infection, cells were split and plated in the presence of puromycin (5 μg/ml); puromycin-resistant colonies were scored after 7 days. Virus titers generally ranged from 3 to 5 × 10 5 transducing units per ml. Transduction of Jurkat or of C91 PL T cells with retroviral vectors was carried out as followed: briefly, cells (1 × 10 6 ) plated in a 24-well plate were infected at a multiplicity of infection (moi) of 5 with viral stocks in a final volume of 1.0 ml containing 4 μg of polybrene/ml, for 18 h and allowed to recover for 24 hr with fresh medium. When necessary, transduced cells were selected with puromycin 4–5 μg/ml for 4 days and maintained in culture for long time period with 1 μg/ml puromycin. RNA isolation and real time quantitative RT-PCR Nuclear and cytoplasmic RNAs were extracted from 2 × 10 6 cells by using an Rneasy RNA-preparation kit (Qiagen) according to the manufacturer's instructions. To reduce the amount of DNA originating from lysis, samples were treated with Rnase-free Dnase (10 U/μl, Boehringer) for 30 min at 20°C and then for 15 min at 65°C. 500 ng of RNA sample were reverse transcribed by using oligo(dT)12–18 and Superscript II (Life Technologies, Inc.). Reverse transcription was performed for 50 min at 42°C. The total cDNA volume of 20 μl was frozen until real-time quantitative PCR was performed. After thawing for PCR experiments, the cDNA was diluted in distilled water and 2 μl of diluted cDNA was used for each PCR reaction. The realtime quantitative PCR (RQ-PCR) was performed in special lightcycler capillaries (Roche) with a lightcycler Instrument (Roche), by using the LightCycler-FastStart reaction Mix SYBR-Green kit (Roche). The following specific primers were used to detect: hnRNP A1, sense 5'-AAGCAATTTTGGAGGTGGTG-3' and antisens, 5'-ATAGCCACCTTGGTTTCGTG-3', gag/pol HTLV-1 sense, 5'-CCCTCCAGTTACGATTTCCA-3' and antisens, 5'-GGCTTGGGTTTGGATGAGTA-3', env HTLV-1 sense, 5'-CTGTGGTGCCTCCTGAACT-3' and antisens, 5'-AAAGTGGCGAGAAACTTACCC-3', pXIII sense, 5'-ATCCCGTGGAGACTCCTCAA-3' and antisens, 5'-CCAAACACGTAGACTGGGTATCC-3'. β-actin sense,5'-TGAGCTGCGTGTGGCTCC-3' and antisens: 5'-GGCATGGGGGAGGGCATACC-3'. The thermal cycling conditions consisted of 40 cycles at 95°C for 10 sec, 61°C for 5 sec, 72°C for 10 sec. The fluorescence signal increase of SYBR-GREEN was automatically detected during the 72°C phase of the PCR. Omission of reverse transcriptase in the RT-PCR protocol led to a failure of target gene amplification in the positive controls. Light cycler PCR data were analyzed using LightCycler Data software (Idaho Technology). The software first normalizes each sample by background subtraction of initial cycles. A fluorescence threshold is then set at 5% full scale, and the software determines the cycle number at which each sample reached this threshold. The fluorescence threshold cycle number correlates inversely with the log of initial template concentration. β-actin transcript levels were used to normalize the amount of cDNA in each sample. Melting curve profiles were used to confirm amplification of specific transcripts. Immunoblotting Cells were washed and harvested in ice-cold PBS containing protease inhibitors (complete mini EDTA-free, Roche Molecular Biochemicals). Cells were lysed in RIPA buffer (150 mM NaCI, 50 mM Tris-HCI pH 8.0, 0.5% deoxycholate, 0.1% SDS, 0.5% Nonidet P-40, protease inhibitors, 80 U/ml endonuclease) and incubated for 30 min at 4°C. After centrifugation at 12,000 rpm for 10 min at 4°C, the supernatant was assayed for protein content by Bradford assay (Bio-Rad). Equal amounts of proteins were separated by SDS/PAGE. Cells were lysed in Laemmli buffer and equal amounts of proteins were subjected to 12% SDS-PAGE. They were subsequently blotted onto nitrocellulose membrane (BA, Schleicher & Schuell). The membrane was then blocked overnight at 4°C in blocking buffer (PBS and 0.1% Tween-20) supplemented 10% non-fat powdered milk and probed with the appropriate antibody diluted in blocking buffer plus 10% non-fat powdered milk. The following antibodies were used: rabbit anti-actin (Sigma), mouse anti-ASF/SF2 (gift from Dr. J. Stevenin) mouse monoclonal anti-hnRNP A1 and anti-hnRNP C antibodies (4B10 and 4F4, respectively; gifts from G. Dreyfuss), followed with an anti-rabbit (Immunotech, France) or anti-mouse (Dako) Immunoglobulin G-horse radish peroxidase-conjugated antibody. Blots were then developed using an enhanced chemiluminescence detection system (Renaissance, NEN, Life Science Products). Bands were visualized by using Hyperfilm (Amersham Pharmacia Biotech). Flow cytometric analysis and Immunostaining Cells (5 × 10 5 ) were washed twice with PBS, resuspended in 3% (vol/vol) paraformaldehyde/PBS for 45 min at room temperature, and permeabilized with 0.5% Triton X-100/PBS for 5 min. After washing with PBS, the cells were incubated with specific antibodies (4B10) diluted in 1% BSA/PBS for 1 h. Cells were washed twice with PBS and were then incubated with FITC-conjugated goat anti-mouse, PE-conjugated goat anti-rabbit in 1% BSA/PBS for 40 min. Cells were washed three times with PBS and resuspended in a 2% paraformaldehyde/PBS solution. The fluorescence intensity was measured on a FACScan instrument (Becton Dickinson Labware, Mountain View, Calif;). The integrated fluorescence of the gated population was measured, and data from 10,000 analyzed events were collected. For immunostaining, C91PL cells were centrifuged on cytoslides using a cytospin (Thermo Shandon, Pittsburgh, PA), fixed on slides with 3.7% paraformaldehyde for 15 min at room temperature, and permeabilized with 0.5% Triton X100 for 5 min in 4°C. The samples were saturated with PBS containing 0.5% gelatin and 0.25% bovine serum albumin for 1 h and stained for 1 h with a 1/100 dilution of a rabbit polyclonal serum directed against HA (Y11 from Santa Cruz Biotechnology) (NLS-A1-HA staining) or 1/1000 dilution of mouse monoclonal antibodies (4B10) (hnRNP A1 staining) in the same saturation solution. The samples were then washed three times with PBS containing 0.25% gelatin and incubated for 1 h with a 1/100 dilution of the following secondary antibodies: goat anti-rabbit immunoglobulin G conjugated to fluorescein isothiocyanate (green color for HA) and goat anti-mouse immunoglobulin G conjugated to lissamine rhodamine sulfchloride (red color for hnRNP A1) (Jackson Immunoresearch). The samples were washed three times in PBS with 0.25% gelatin and mounted for analysis on a Zeiss LSM 510 laser scanning confocal microscope. ELISA p19gag was measured in culture medium using the RETROTEK HTLV p19 Antigen ELISA kit (Zeptometrix). Medium of the cell culture was centrifuged at low speed to remove the cell debris, and filtrated through a 0,45-μm filter. The amount of Gag protein was quantified in the resultant supernatant according to the manufacturer procedure. Results are expressed as pg/ml of p19 protein and are the mean of two different experiments, each point tested in quadruplicate. Competing interests The author(s) declare that they have no competing interests.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC551596.xml
514551	DMBT1 expression is down-regulated in breast cancer	Background We studied the expression of DMBT1 (deleted in malignant brain tumor 1), a putative tumor suppressor gene, in normal, proliferative, and malignant breast epithelium and its possible relation to cell cycle. Methods Sections from 17 benign lesions and 55 carcinomas were immunostained with anti DMBT1 antibody (DMBTh12) and sections from 36 samples, were double-stained also with anti MCM5, one of the 6 pre-replicative complex proteins with cell proliferation-licensing functions. DMBT1 gene expression at mRNA level was assessed by RT-PCR in frozen tissues samples from 39 patients. Results Normal glands and hyperplastic epithelium in benign lesions displayed a luminal polarized DMBTh12 immunoreactivity. Normal and hyperplastic epithelium adjacent to carcinomas showed a loss of polarization, with immunostaining present in basal and perinuclear cytoplasmic compartments. DMBT1 protein expression was down-regulated in the cancerous lesions compared to the normal and/or hyperplastic epithelium adjacent to carcinomas (3/55 positive carcinomas versus 33/42 positive normal/hyperplastic epithelia; p = 0.0001). In 72% of cases RT-PCR confirmed immunohistochemical results. Most of normal and hyperplastic mammary cells positive with DMBTh12 were also MCM5-positive. Conclusions The redistribution and up-regulation of DMBT1 in normal and hyperplastic tissues flanking malignant tumours and its down-regulation in carcinomas suggests a potential role in breast cancer. Moreover, the concomitant expression of DMTB1 and MCM5 suggests its possible association with the cell-cycle regulation.	Background DMBT1 (Deleted in Malignant Brain Tumor) at chromosome 10q25.3–26.1, was considered a tumour suppressor gene because frequent deletions and/or lack of expression had been found in malignant tumours of the brain [ 1 - 3 ], the gastrointestinal tract [ 4 , 5 ], and the lung [ 6 , 7 ]. However, recent evidence has shown that DMBT1 is not a typical tumour suppressor gene because it is up-regulated in normal or inflamed epithelium adjacent to lung carcinomas as well as in some primary lung cancers and glioblastomas [ 8 ]. It was also shown that mutations in combination with loss of heterozygosity do not play a role in DMBT1 inactivation in many tumours including breast cancer [ 9 , 10 ] DMBT1 is a multi-functional protein related to the Mac-2 binding protein and mucins that have a protective function and mediate cell-extracellular matrix interactions [ 8 , 11 ]. DMBT1 GP340 (glycoprotein-340) and DMBT1 SAG (Salivary Agglutinin) are DMBT1 variants, present in the respiratory tract and the oral cavity, respectively, that have been linked to innate host defense and epithelial regeneration [ 3 , 12 - 16 ]. In the respiratory tract, DMBT1 GP340 is involved in innate host defence by its interaction with the lectins surfactant protein D and A (Sp-D, Sp-A) and by its ability to stimulate alveolar macrophage migration [ 12 , 13 , 17 ]. Another lectin, mannan binding protein (MBP), has been shown to have an anti tumour effect against gliomas and colorectal carcinomas in vitro and in vivo [ 18 , 19 ]. While this anti tumor property remains to be demonstrated for Sp-A and Sp-D, we have previously reported the presence of surfactant protein A (Sp-A) in the ductal epithelium of the normal breast and its variable expression in mammary carcinomas. We found that in the normal ductal epithelium Sp-A immunoreactivity is present on the luminal side, whereas in carcinomas it is present both on the luminal side as well as on the baso-lateral surfaces. In carcinomas, the Sp-A immunoreactivity is not only redistributed but also decreased as the histological tumour grade increases [ 20 ]. Remarkably, this resembles the changes of DMBT1 expression and localisation observed in other tumor types originating from mono-layered epithelia such as primary esophageal adeno-carcinomas and lung carcinomas [ 5 , 8 ]. Hensin, the rabbit homologue of DMBT1, triggers epithelial terminal differentiation by mediating cell-extracellular matrix (ECM) interactions, and this may apply also to DMBT1 [ 3 , 21 - 24 ]. CRP-ductin, DMBT1 homologue in the mouse, and a novel homologue in the rat ( DMBT1 4,7 kb ), have been linked to initiation of cell proliferation, differentiation and repair [ 25 , 26 ]. It has been proposed that translocation of DMBT1 to the ECM triggers cellular differentiation, whereas a loss of expression of DMBT1 favours tumor cell growth [ 8 ]. These findings led us to explore whether similar principles apply to breast epithelium. Here we examined the expression of DMBT1 in the morphologically normal, hyperplastic, and neoplastic breast epithelium. We also examined in morphologically normal epithelium a possible association of DMBT1 expression with that of an early proliferative marker such as MCM5 (Minichromosome Maintenance protein 5). MCM5 is one of the 6 pre-replicative complex proteins (MCM 2–7) [ 27 ] and its expression was compared with that of Ki 67 proliferation index. Methods Tissue and cell-lines From the files of the Surgical Pathology at S.Paolo University Hospital, we retrieved paraffin blocks of 72 consecutive breast lesions surgically resected. This material included 16 cases of fibrocystic disease (FD), 1 case of gynecomastia (GM), 10 samples of ductal carcinoma in situ (DCIS) and 45 cases of infiltrating carcinoma (IC) (36 ductal, 2 lobular, 2 mixed lobular and ductal, 2 mucinous, 1 apocrine, 1 cribriform, 1 papillary). RT-PCR (Reverse-Transcriptase Polymerase Chain Reaction) was performed on 39 samples: i.e. 37 samples out of the 55 carcinomas used for the immunohistochemical studies, 1 sample of normal breast tissue that had been removed with a neoplastic lesion and 1 sample of fibrocystic disease. All 39 samples had been quickly frozen in liquid nitrogen after surgical resection and stored at -80°C. As confirmed by histopathologic examination, in all tumor specimens the amount of tumor cells equalled or exceeded 80% of the overall sample. As controls for each test a frozen sample of the same case positive with immunohistochemical method was used. We also analysed by RT-PCR two breast cancer cell lines (Hs578T and T-47D) obtained from the American Type Culture Collection and maintained as recommended. Immunohistochemistry Sections from blocks containing the predominant lesion were incubated with anti DMBT1h12 monoclonal antibody [ 3 ] as well as anti Ki67 Mib1 (Dako Cytomation DK). Double labelling with anti- DMBT1 and anti MCM5 monoclonal antibodies (Novocastra Laboratories Ltd, Newcastle upon Tyne, U.K.) was performed on 17 benign lesions and a subset of 19 in situ and invasive carcinomas where residual normal breast tissue was well represented. On these selected samples a single incubation with anti MCM5 and anti Ki 67 antibody was performed in order to compare the two respective immunoreactivities. Antibodies were diluted as follows: anti-DMBT1h12, 1:500; anti MCM5, 1:40; anti Ki 67 Mib1, 1:100. For anti- DMBT1 and Ki 67 Abs, sections were pre-treated for antigen retrieval in boiling 0.01 M sodium citrate buffer pH 6, in microwave (MW) oven for 12 minutes. DMBT1 and MCM5 double labelling was performed in two steps following a modification of the method of Lan et al. [ 28 ], briefly: 1) first antigen retrieval 6 minutes in 0.5 M EDTA pH8 in MW oven three times, followed by 1 h incubation with anti DMBT1h12. The reaction was developed with alkaline phosphatase using Fucsin + (Dako) as chromogen. 2) same antigen retrieval followed by over-night incubation with anti MCM5 Ab and development with peroxydase and diaminobenzidine (DAB) as chromogen. All slides were incubated in an automated slides stainer (Biogenex, S.Ramon, CA, U.S.A.); reaction products were visualized by En Vision product (Dako). Negative and positive controls were added in each experiment. Three investigators (AM, PB, PGN) evaluated separately DMBTh12 immunopositivity of morphologically normal, hyperplastic, and malignant cells. Immunopositivity was scored in 4 distinct categories according to the percentage of positive cells: 0 = negative, 1 = < 10%, 2 = 10%–50% and 3 = > 50%. MCM5 and Ki67 positivity in morphologically normal epithelium was determined with a semi-quantitative method: percentage of positive glandular structure and number of positive cells observed in each positive gland divided number of glands. RNA extraction Total RNA was extracted from breast specimens using Trizol reagent, (Invitrogen s.r.l., S. Giuliano Milanese, Italy) according to the manufacturer's instructions. Total RNA was quantified and checked for purity by UV spectrophotometry. cDNA synthesis Reverse transcription of RNA was done in a final volume of 20 μl containing 1x buffer (50 mM KCl, 15 mM Tris-HCl, pH 8.3), 500 μM each deoxynucleotide triphosphate, 5 mM MgCl 2 , 20 units of RNase inhibitor, 50 units of MuLV Reverse Transcriptase, 2.5 μM Random hexamers and 1 μg of total RNA. All reagents were purchased by the supplayer (Applied Biosystems, Foster City, CA, USA). The samples were incubated at 25°C for 10 min, 42°C for 60 min and at 95°C for 5 min. PCR amplification Polymerase chain reaction (PCR) of the cDNA was performed in a final volume of 50 μl containing all four dNTPs (each at 200 μM), 1X buffer, 1.5 mM MgCl 2 , 1.25 units of Taq Gold DNA polymerase and each primer at 0.3 μM. The PCR amplification of DMBT1 was carried out in two steps, following a nested procedure. In the first step we used 2 μl of cDNA from the reverse transcription and in the second 2 μl of the first PCR product as template. Primers and thermal condition were used according to Mollenhauer et al [ 3 ]. The specificity of the PCR product was confirmed by sequencing with BigDyeTM terminator chemistry using the genetic analyser ABI Prism 310 (Applied Biosystems), β-2-microglobulin (as control for sample integrity) was amplified in the amplification mixture described above and using the following primers: 5'-TCTGGGTTTCATCCATCCGA-3' and 5'-CCCCAAATTCTAAGCAGAGTATGTAA-3'. The thermal cycling parameters were: initial step of 10 min at 94°C, 39 cycles at 94°C for 30 sec, 58°C for 30 sec, 72 °C for 30 sec, and a final step for 5 min at 72°C. All the PCR products were separated by electrophoresis on a 1.5% agarose gel containing ethidium bromide and visualized under UV transillumination. Amplification product of 602 bp (β-2-microglobulin) and 435 bp ( DMBT1 ) were obtained. Data analysis Statistical differences between immunopositive cases/total number of cases were calculated by Fisher's exact test. p value less than 0.05 were considered statistically significant. T-test for independent values was used to evaluate statistically significant differences between immunoreactivity of normal and hyperplastic epithelium in benign, pre-invasive and invasive lesions. Results Immunohistochemistry Different patterns of immunopositivity were observed in the histologically normal and hyperplastic breast tissue, present in the sections of benign and malignant lesions: a) single supra-nuclear, intra-cytoplasmic dot polarized toward luminal surface [Figure 1a ]; b) a strong surface decoration of luminal cells [Figure 1b ], c) intra-luminal immuno-reactive material [Figure 1c ], d) multiple, randomly distributed peri-nuclear, irregular dots [Figure 1d ], e) marked supra-nuclear and/or baso-lateral accumulation of reaction product [Figure 1e ] and f) strong diffuse granular intra-cytoplasmic decoration [Figure 1f ]. Morphologically normal ducts showed the pattern of immunopositivity described in section a). In addition to pattern a), the epithelium in benign lesions showed immunoreactivity patterns b) and c). Hyperplastic tissues flanking carcinoma frequently showed depolarization of immuno-reaction of patterns d), e), and f). Normal and hyperplastic mammary structures displayed a variable DMBTh12 immunoreactivity: some ducts were decorated while others were negative. Moreover a variable percentage of cells was immunopositive. This heterogeneity suggests that there is no constitutive expression of DMBT1 at protein level in the mammary gland epithelium. In tumour areas, only 3 out of 55 carcinomas showed variable degrees of immunopositivity with anti-DMBTh12 Ab: in one case there was a widespread and strong decoration of numerous tumour cells (immunopositivity score = 3) with irregularly distributed dots [Figure 1g ]; in the second case only few cells had cytoplasmic positivity (immunopositivity score = 1) [Figure 1h ] and in the third case only weak granular decoration in a small clusters of cells was present (immunopositivity score = 1). All other carcinomas were completely negative (immunopositivity score = 0) [Figure 1i ]. Down-regulation of DMBT1 expression in the cancerous lesions compared to the normal and/or hyperplastic epithelium was statistically significant for both, ductal carcinoma in situ (DCIS) (1/10 versus 33/42 normal and/or hyperplastic epithelia; p = 0.0001) and invasive carcinoma (IC) (2/45 versus 33/42 normal and/or hyperplastic epithelia; p < 0.0001). The results of DMBT1 expression at protein and mRNA level are summarized in table 1 . According to the 4 category scoring system, there was no statistically significant difference of DMBT1 immunoreactive cells between the normal and the hyperplastic epithelium that flanked benign lesions, in situ and invasive carcinomas (Table 2 ). Double immunostaining with DMBTh12 Ab and anti MCM5 Ab revealed a co-expression of the two antigens, respectively, in the cytoplasm and the nucleus of the same cells in morphologically normal breast epithelium (Figure 2a,2b ). Conversely, glands that were negative with anti DMBT1 were negative also with anti MCM5 antibody (Figure 2b ). DMBT1 and MCM5 positive or negative glands were morphologically undistinguishable. In comparison to MCM5, very few morphologically normal epithelial cells labelled by Anti DMBT1 Ab, expressed Ki67 (Figure 3a,3b,3c ). 30–50% of glands were positive with anti MCM5 Ab with a mean number of 4,5 positive cells per positive glandular structure versus less than 10% of positive glands with a mean number of 1,46 positive cell per positive glandular structure with anti Ki 67 Ab. As expected for a pre-replicative marker, anti MCM5 Ab decorated diffusely all benign proliferative and malignant lesions. The percentage of Ki 67 positive cells of carcinomas varied and was assessed as part of the protocol for the surgical pathology diagnosis of breast neoplasia together with other prognostic parameters i.e. hormonal receptors status and Her2 Neu positivity. These data did not correlate with DMBT1 expression. Expression analysis by RT-PCR In 28/39 cases (72%) RT-PCR confirmed immunohistochemical results, whereas 11 tumours that were negative with anti-DMBTh12 revealed PCR-products after RT-PCR amplification (Table 1 ). In two of these discordant cases we could document the presence in the sample of residual normal epithelial cells. DMBT1 mRNA was detected in both the breast carcinoma cell lines analysed: Hs 578T (Figure 4 ) and T47D (not shown). The RT-PCR positivity of samples did not correlate with histological parameters or with IHC expression of hormonal receptors, proliferation index Ki 67 and Her 2/Neu. Discussion In the present study we investigated DMBT1 expression by immunohistochemistry and RT-PCR in benign and malignant breast epithelial lesions. By immunohistochemistry we found a variable DMBTh12 positivity in the morphologically normal and hyperplastic epithelium adjacent to benign (88%) and malignant (78%) lesions. Among 55 carcinomas we could detect immunopositivity only in 3 tumours (5%); in one case of infiltrating carcinoma the immunopositivity was widespread with a pattern similar to that seen in hyperplastic lesions near carcinoma; in the other two cases (one carcinoma in situ and one infiltrating ductal carcinoma), it was limited to few tumour cells. DMBT1 expression was detected by RT-PCR in 13 of 35 (37%) infiltrating carcinomas and in 2 breast cancer cell lines. Eleven tumours, that were negative with anti-DMBTh12, revealed PCR-products after RT-PCR amplification. This discordance between IHC and RT-PCR results may be due to residual normal epithelial cells present in the sample. Furthermore other possible reasons of this discrepancy could be: 1) the greater sensitivity of molecular technique like RT-PCR in detecting very low expression levels of DMBT1 that were below detection by immunohistochemistry, 2) sampling problems i.e. the presence of few positive tumour cells only in the frozen samples or 3) few immunoreactive cells have been missed by microscopic examination and finally 4) DMBT1 protein could have an altered epitope not recognized by its paratope in the antibody. By either immunohistochemistry or molecular technique we found no correlation between DMBT1 expression and tumor histotype, grade, hormonal receptors status, proliferation index, Her2 Neu expression or lymph-nodal metastases. In a recent paper on breast carcinogenesis, qualitatively similar data were reported for DMBT1 in human breast cancer, but an indirect weak correlation between the degree of differentiation of the breast carcinomas and the extent of the immunoreactivity was found [ 10 ]. The discrepancy in the results between our and this study may be due to sampling problems, the size of the sample and heterogeneous expression of DMBT1 in breast cancer or to other poorly-defined factors such as race, age or hormonal status of the population under study. Mollenhauer et al. point out that DMBT1 expression may be up-regulated in pathophysiological conditions such as inflammation, liver injury or after carcinogen administration and it is down-regulated in tumors to achieve a growth advantage in a dynamic and complex model. Also according to these authors, the possibility that DMBT1 loss may be a consequence rather than a primary event in carcinogenesis is unlikely (10). Although the normal mammary gland has been reported to have low DMBT1 expression by RT-PCR [ 12 ], we found variable DMBTh12 immunopositivity in breast epithelium flanking benign and malignant lesions. Immunopositive epithelial cells in our material appeared either morphologically normal and indistinguishable from immunonegative cells or hypertrophic with abundant cytoplasm and large elongated nuclei oriented perpendicularly to the basement membrane [Figure 1i ]. Moreover, DMBTh12 immunoreactivity was also maintained in hyperplastic epithelium flanking carcinomas. This resembles the DMBT1 expression patterns observed in the human respiratory tract and in associated inflammation and carcinomas [ 8 ] and suggest that the variable DMBT1 expression in the normal mammary gland epithelium may result from the necessity of DMBT1 induction. Indeed, in parallel in vivo studies, it was found that DMBT1 is strongly up-regulated in the mammary gland tissue of the mouse shortly after carcinogen-administration and prior to the onset of histo-pathologically visible symptoms [ 10 ]. It has been reported that Sp-D and Sp-A are ligands of DMBT1 products [ 29 ]. We recently demonstrated that Sp-A, in breast carcinoma tissues, has a pattern of expression resembling that of its interaction partner [ 20 ]; thus, we hypothesize that co-ordinated induction of both SP-A and DMBT1 in breast tissue may be part of a protective response as suggested by Kang et al [ 30 ] for DMBT1 and Sp-D in pulmonary epithelium. While this interaction may be attributed to lumenally secreted DMBT1 , the depolarisation of the staining patterns indicates that DMBT1 may also be secreted to other destinations, i.e. the ECM, by the mammary gland epithelial cells. In this regard, the second putative function of DMBT1 , i.e. triggering of epithelial differentiation, is noteworthy. It may prevent hyperproliferative epithelial cells from out-of control proliferation and progression to the tumor state. This view is supported by two present findings. The correlation between MCM5 and DMBT1 expression may point to a coupling to cell-cycle processes: basolateral relocation and hyper-expression of DMBT1 at the ECM may provide cells that are licensed to proliferate with protection from unregulated proliferation. On the other hand loss of DMBT1 expression in a significant numbers of DCISs and ICs is consistent with the notion that DMBT1 inactivation may offer growth advantages to neoplastic cells [ 10 ]. The loss of DMBT1 expression appears to take place early, i.e. already at the time of transition from hyperplastic lesions to carcinoma, as it has been reported in oesophageal squamous cell carcinomas [ 5 ]. The exact mechanisms for DMBT1 inactivation remain unclear. An inactivation by mutation appears not to be a major mechanism for DMBT1 inactivation in tumors [ 10 , 31 ] The studies of Munoz and co-workers suggest that epigenetic silencing of either DMBT1 or genes that regulate DMBT1 expression takes place in some tumors [ 32 ]. Accordingly, except for direct epigenetic silencing of the DMBT1 promotor, loss of expression or mutation of DMBT1 -regulatory genes might represent alternative candidate mechanisms. Conclusions We have shown that DMBT1 expression and cellular location in breast epithelium varies in normal, hyperplastic and neoplastic cells. In normal or hyperplastic tissue flanking carcinomas it is up-regulated and polarized toward glandular lumen or the basolateral membrane while it is down-regulated in carcinomas. These findings closely resemble the patterns observed for its interaction partner SP-A in an earlier study. Additionally we have shown that DMBT1 is co-expressed with MCM5 in normal and hyperplastic tissue. We hypothesize that hyper-expression and basolateral relocation of DMBT1 in these epithelia prevents the out of control growth that characterizes neoplastic epithelium. List of abbreviations used DMBT1: Deleted in Malignant Brain Tumor 1, MCM5: Minichromosome Maintenance protein, 5 RT-PCR: Reverse Transcriptase Polymerase Chain Reaction GP 340: Glycoprotein 340 SAG: salivary Agglutinin Sp-A, Sp-D: Surfactant Protein A, D MBP: Mannan Binding Protein ECM: Extracellular Matrix FD: Fibrocystic Disease GM: gynecomastia DCIS: Ductal Carcinoma In Situ IC: Invasive Carcinoma MW: Micro Wave EDTA Ethylenediaminetetraacetic Acid DAB: Diaminobenzidine Competing interests None declared. Autors Contributions PB designed the study, participated in the evaluation of immunohistochemical results and drafted the manuscript. PGN participated in the evaluation of immunohistochemical results and performed statistical analysis. JM participated in sequence alignment, criticism and produced DMBTh12 antibody. AP participated in sequence alignment and data evaluation. CP Carried out RT-PCR. AM participated in the evaluation of histologic samples and immunohistochemical results. GB participated in data evaluation and sequence alignment GC participated in study design and data evaluation. SB participated in study design, sequence alignment, criticism and coordinated molecular study. GGP participated in study design, criticism and general revision of paper. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC514551.xml
368170	Genome Sequence of the Intracellular Bacterium Wolbachia	null	Wolbachia have a thing against males. A member of one of the most diverse groups of bacteria, called Proteobacteria, this parasitic “endosymbiont” lives inside the reproductive cells of a wide variety of the nearly 1 million species of arthropods, including insects, spiders, and crustaceans. It has also been found in worms. Wolbachia 's preferred habitat is the cytoplasm of its host's gametes. Since sperm have very little cytoplasm, Wolbachia seek out the company of females, securing its survival by hitching a ride to the next generation in the cytoplasm of the mother's eggs. Wolbachia 's effects range from beneficial to pathological, depending on which species infects which invertebrate host, but since most species are not beneficial, Wolbachia infections often turn out badly if the host is male. On the other hand, if female, the host could very well live longer, produce more eggs, and have higher hatching rates than its noninfected cousins—thereby facilitating Wolbachia 's transmission from mother to offspring. Wolbachia have evolved an impressive repertoire of “reproductive parasitic” strategies to adapt its host's physiology to its own advantage. One strategy involves inducing “cytoplasmic incompatibility” between sperm and egg, which in effect uses infected males to keep uninfected females from producing viable offspring. Another causes infected females to reproduce asexually, creating a new generation of infected clones. Another turns developing male embryos into females. And, in a pinch, some Wolbachia simply kill developing males. The biochemical mechanisms that trigger different strategies in different hosts are unclear, however, in part because it's so far been impossible to cultivate sufficient quantities of these obligate endosymbionts (that is, intracellular species that cannot survive outside their host). But now that Scott O'Neill, Jonathan Eisen, and colleagues have sequenced the complete genome of one strain of Wolbachia pipientis , scientists investigating the biology and evolution of Wolbachia –host interactions have a valuable new research tool. The strain they sequenced, W. pipientis w Mel, lives inside the fruitfly Drosophila melanogaster , the favorite model organism of geneticists for nearly 100 years. This strain causes cytoplasmic incompatibility in its host. Transmission electron micrograph of Wolbachia within an insect cell (Image courtesy of Scott O'Neill) The structure of the w Mel genome, the O'Neill and Eisen groups note, is strikingly different from any other obligate intracellular species. While its genome is compact, it nonetheless contains large amounts of repetitive DNA and “mobile” DNA elements. Mobile genetic elements, as the name implies, are DNA sequences that move around the genome and are often acquired from other species. Most of the repetitive and mobile elements in Wolbachia do not appear in other α-Proteobacteria species and were probably introduced some time after Wolbachia split off from its evolutionary ancestors. Wolbachia , unlike other obligate intracellular bacteria, seem quite amenable to incorporating foreign DNA, which the authors speculate was introduced by the bacteria-infecting virus called phage. Analysis of the Wolbachia genome sheds light on the mechanisms that might help the parasite manipulate the host cell's physiology to its own advantage. One likely bacterial weapon for host exploitation is the abundance of predicted genes encoding ankyrin repeat domains, amino acid sequences characteristic of proteins important for protein–protein interactions in eukaryotes (organisms with nuclei, which bacteria lack). In bacteria, ankyrin repeats might regulate host cell-cycle pathways, which one wasp-infecting Wolbachia strain modifies to induce cytoplasmic incompatibility. Other molecular interactions between w Mel and its host, the researchers propose, might also rely on proteins with these ankyrin repeats. The Wolbachia genome also provides insight into mitochondrial evolution. It is widely believed that these intracellular energy-metabolizing centers were once free-living bacteria belonging to the α-Proteobacteria group, though it's not clear which branch of the α-Proteobacteria tree they inhabit. Complete genome analysis of various α-Proteobacteria—including w Mel, the first non- Rickettsia species sequenced in the Rickettsiales group—provides no evidence that mitochondria are more related to Rickettsia species than to Wolbachia , as was previously thought. In fact, further analysis failed to consistently connect mitochondria to any particular species or group within the α-Proteobacteria. While the information hidden in the Wolbachia genome seems to raise as many issues as it settles, biologists studying a wide range of problems—from the evolution and biology of Wolbachia and endosymbiont–host interactions to the origin of mitochondria—have a valuable new tool to explore their questions. The Wolbachia genome will also provide important molecular guidance for efforts to suppress insect pests and control filariasis, a human disease caused by worms. Since beneficial Wolbachia live in both insect and worm, applying antibiotics to target the Wolbachia will ultimately kill the insect pest and infecting worm, which both depend on the bacteria to survive.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC368170.xml
521071	Seasonal ovulatory activity exists in tropical Creole female goats and Black Belly ewes subjected to a temperate photoperiod	Background Seasonality of ovulatory activity is observed in European sheep and goat breeds, whereas tropical breeds show almost continuous ovulatory activity. It is not known if these tropical breeds are sensitive or not to temperate photoperiod. This study was therefore designed to determine whether tropical Creole goats and Black-Belly ewes are sensitive to temperate photoperiod. Two groups of adult females in each species, either progeny or directly born from imported embryos, were used and maintained in light-proof rooms under simulated temperate (8 to 16 h of light per day) or tropical (11 – 13 h) photoperiods. Ovulatory activity was determined by blood progesterone assays for more than two years. The experiment lasted 33 months in goats and 25 months in ewes. Results Marked seasonality of ovulatory activity appeared in the temperate group of Creole female goats. The percentage of female goats experiencing at least one ovulation per month dramatically decreased from May to September for the three years (0%, 27% and 0%, respectively). Tropical female goats demonstrated much less seasonality, as the percentage of goats experiencing at least one ovulation per month never went below 56%. These differences were significant. Both groups of temperate and tropical Black-Belly ewes experienced a marked seasonality in their ovulatory activity, with only a slightly significant difference between groups. The percentage of ewes experiencing at least one ovulation per month dropped dramatically in April and rose again in August (tropical ewes) or September (temperate ewes). The percentage of ewes experiencing at least one ovulation per month never went below 8% and 17% (for tropical and temperate ewes respectively) during the spring and summer months. Conclusions An important seasonality in ovulatory activity of tropical Creole goats was observed when females were exposed to a simulated temperate photoperiod. An unexpected finding was that Black-Belly ewes and, to a lesser extent, Creole goats exposed to a simulated tropical photoperiod also showed seasonality in their ovulatory activity. Such results indicate that both species are capable of showing seasonality under the photoperiodic changes of the temperate zone even though they do not originate from these regions.	Background Seasonality of reproduction is a common feature in sheep and goat breeds of temperate latitudes [ 1 , 2 ] and seems to have been present for millennia in the sheep and goat breeding systems [ 3 ]. The annual breeding season begins in summer in Ile-de-France ewes and in autumn in Alpine goats and ends in winter in both species, resulting in a marked seasonality in birth dates of lambs and kids. In goats and sheep, this seasonality is under photoperiodic control. In experimental conditions, long days inhibit and short days stimulate sexual activity (goats: [ 4 - 6 ] sheep: [ 1 ]). However, under natural conditions of temperate countries, goats, as well as sheep [ 7 , 8 ], probably have an endogenous rhythm that is synchronized by photoperiod such that breeding occurs during autumn/winter and anovulation (anestrus) occurs during spring/summer. When transferred to equatorial conditions (12 h of light per day, with a limited control of temperature change amplitudes), Suffolk ewes (a European breed) cycled at irregular intervals with no clear anovulatory season [ 9 ]. In contrast, when transferred to tropical photoperiodic conditions where the annual amplitude of photoperiodic changes exists but is lower than in temperate regimen, Alpine goats do not greatly modify the seasonal characteristics of their breeding season, and long periods of anestrus and anovulation are still present during spring and summer as in control females maintained under temperate photoperiod [ 10 ]. Local breeds of sheep [ 11 - 15 ] and goats [ 12 , 16 , 17 ] under their native tropical conditions, are either non-seasonal breeders or exhibit only a weak seasonality of reproduction. The females of these breeds ovulate and exhibit estrus almost the whole year round, even though short periods of anovulation and anestrus are detected in some females. Two main hypotheses can be raised to explain the near-absence of seasonality: either the females are insensitive to photoperiod, or the amplitude of the photoperiodic changes is too small. It is thus interesting to determine whether absence of seasonality persists when females of these breeds are subjected to major annual changes in the amplitude of day length, the prevailing conditions in temperate regions, or whether seasonality appears as it does in most temperate breeds. In the present experiment, seasonal ovulatory activities were assessed in tropical Creole goats and Black-Belly ewes. These two breeds originate from the Carribean Islands, where they have been bred for 3–4 centuries, and constitute the progeny of African tropical breeds (see Methods). The animals of the present experiment were imported into Europe and experimentally subjected for more than two years in light-proof rooms to an annual photoperiodic regimen simulating that of temperate regions (TE group), and compared to females under tropical cycle simulating that of a tropical region (TR group). Results Ovulatory activity in Creole goats Ovulatory activity demonstrated marked differences between experimental groups over the course of the experiment. Individual ovulatory activity is presented in Figure 1 . Marked seasonality of ovulatory activity appeared in female goats of the TE group; the seasonal inactivity occurred from May to September for the three years of the study (Figures 1 & 2 ). The percentage of female goats experiencing at least one ovulation per month dramatically decreased from May to September for the three years (0%, 27% and 0%, respectively). All female goats experienced an anovulatory period during the first spring/summer season (1990), only one of them continued its ovulatory activity in 1991, and all of them stopped again in May 1992. In contrast, female goats exposed to the TR photoperiodic cycle demonstrated much less seasonality as the percentage of goats experiencing at least one ovulation per month never went below 56% (minima for the three years of study: 56%, 56%, 57%). Two female goats cycled continuously during the course of the experiment, one female goat in mid-1990 and three in mid-1991 continued their ovulatory activity during spring and summer, and four females were still cycling at the end of the experiment (June 1992). These differences between the two groups led to significant differences in some, but not all, parameters. The percentage of goats experiencing at least one ovulation per month (Figure 2 ) was significantly lower in the TE than in the TR group in May (P < 0.05), June, July, August and September 1990 (P < 0.001), tended to be lower in May 1991 (P < 0.10), and was lower again in May (P < 0.05) and June 1992 (P < 0.001). In both groups the females which stopped their ovulatory activity did so at roughly the same date (April-May) in the 3 years of the experiment (Table 1 ). TR goats began their first breeding season significantly earlier than TE goats in 1990, but this difference did not appear in the second breeding season (Table 1 ). Variances of the dates of end of the 2 nd and 3 rd breeding seasons were significantly higher in TR goats (Table 1 ). The duration of the anovulatory period in 1990 was significantly shorter in TR than in TE goats and the duration of the 1990–1991 breeding season was significantly longer in TR than in TE goats (Table 2 ). The 1991 anovulatory period and the 1991–1992 breeding season did not differ between groups (Table 2 ). Variances of the duration of the 2 nd anovulatory season and of the 3 rd breeding season were significantly higher in TR goats (Table 2 ). Ovulatory activity in Black-Belly ewes Ovulatory activity demonstrated marked variations in both experimental groups over the course of the experiment. Individual ovulatory activity is presented in Figure 3 . From October to March of the first year, all females were cycling in both groups (100% of ewes showed at least one ovulation per month; Figure 3 & 4 ). In the TR group, ovulatory activity dropped in April, remained low (2 ewes cycling) in May and June, then rose again in July and August to reach its maximum from September to April of the next year; minimum activity was observed again from May to August, before a maximal activity in September and October. Ewes of the TE group roughly followed the same pattern, with a later onset of cyclicity in September of the first year (% of ewes cycling in August P < 0.001); and a later end in May-June of the second year (% of ewes cycling in May P < 0.01). The percentage of cycling females never went below 17% (2 ewes cycling). One ewe in the TE group never stopped cycling during the course of the experiment. Mean dates of the end of the breeding season did not differ between TR and TE ewe groups in the first and second year (April, Table 3 ). The onset of the breeding season occurred earlier in TR than in TE for the first year, but not for the second year (Table 3 ). Thus, duration of the first and second anestrous seasons and/or duration of the sexual season did not differ between TR and TE ewe groups (Table 4 ). Variances of the durations of anovulatory and of breeding seasons were significantly higher in TE ewes (Table 4 ). Discussion The two tropical breeds of Creole goats and Black-Belly ewes used in the present study demonstrated clear seasonal breeding activity with a definite cut-off when maintained under the simulated extensive photoperiodic variations of temperate areas. As many other goat and sheep breeds originating and raised under a temperate climate, these two breeds imported from the tropics displayed cessation of ovulatory activity in spring and summer, i.e. the usual months for anovulation and anestrus in a temperate climate. Even though it appears that their anovulatory season seemed shorter than European breeds of goats [e.g. Alpine, [ 10 ]] and sheep [e.g.. Ile-de-France, [ 18 ]], almost all Black-Belly ewes stopped their ovulatory activity for about 4 months and Creole goats for 2.5 months. Black-Belly ewes stopped their ovulatory activity late in the year (second half of April) as compared to the majority of breeds, for example Ile-de-France breed [mid-January; [ 18 ]) or the majority of British breeds [ 19 ]. On the other hand, they started their breeding season later that these breeds, showing a more "primitive" type of breeding season, similar to those displayed by the Moufflon [ 20 ], Romanov [ 21 ] or Icelandic [ 22 ] breeds of sheep. A similar observation could be made for the end of the breeding season of the Creole goats maintained under simulated temperate photoperiod: they stopped late in the season (April-May) compared to temperate breeds [February, [ 10 ]]. However, this was not true for the onset of the breeding season which generally started in September-October in both goat breeds. The control group of goats maintained under simulated tropical photoperiodic variations displayed significantly less seasonality. The percentage of goats showing at least one ovulation per month was significantly higher in May and June for 2 years out of 3, and did not drop to 0 as it did in the temperate group of goats. A relatively high number of female goats did not experience an anovulatory season either at all or during some years of the experiment, and those that did showed a significantly shorter anovulatory period during the first year of the experiment. Thus, the comparison between Creole female goats maintained under simulated temperate photoperiod and control females placed under simulated tropical photoperiod leads to the conclusion that their breeding season is sensitive to large photoperiodic variations. In contrast, Black-belly ewes maintained under simulated tropical photoperiodic variations did not differ greatly from those maintained under simulated temperate photoperiod. The two groups of ewes and goats used here differed in various parameters, some of which could explain the photoperiod x species interaction observed here: (a) Recipient ewes in which embryos were implanted in autumn were maintained under natural photoperiod. It is known that light changes during pregnancy may affect the progeny's photoperiod sensitivity, especially regarding the onset of puberty in sheep [ 23 - 25 ] and in rodents [ 26 ]. This was not the case in the goats, as these experimental animals were the 3 nd or 4 th progeny of females imported as embryos. (b) Experimental ewes were artificially raised in TE vs TR photoperiod from birth to 6 months old from the start of the experiment, whereas female goats were raised under simulated tropical photoperiod until the start of the experiment at one or two years old. (c) Black-Belly ewes were included in the experiment at 6 months of age, whereas Creole goats were 12 and 24 months old at the start of the experiment. These three main differences between the experiments in female goats and ewes could possibly explain this photoperiod x species interaction. However, it may also come from a real difference of sensitivity to non-photoperiodic factors (such as temperature changes or activity, see later) between the two species. The spontaneous ovulatory activity demonstrated in the present experiment by the Black-Belly ewes from the TR group were very different from those registered earlier in their natural breeding conditions on the island of Martinique in the tropics, where they cycled all year round [ 15 ]. The Creole goats from the simulated tropical photoperiod did not display here the same results as females of the same flock raised in their natural breeding conditions on the island of Guadeloupe in the tropics, where they also cycled almost continuously [ 17 ]. This unexpected difference suggests that other external cues may act in combination with photoperiod to inhibit breeding activity. The fact that all animals simultaneously stopped cycling in spring, suggests that an external physical cue could be involved. The temperature of the light-proof rooms in which the experiments were performed was not controlled and the high and low-amplitude variations of air temperature of the tropics were not applied to our experimental animals. Such a cue may interact with photoperiod and enhance the negative effects of the limited photoperiodic changes, which do not appear in normal tropical conditions. To our knowledge, very few experiments have been carried out in sheep and/or goats to determine the role played by low temperature in the appearance of seasonality. It has been demonstrated that inversion of the temperature rhythm does not entrain ovulatory activity in ewes of a European breed maintained under an equatorial photoperiodic schedule [ 27 ] and that low temperatures in the summer time may induce an 8-week advance in the onset of the annual breeding season in dark-faced ewes [ 28 ]. In Suffolk ewes, a seasonal breed, the maintenance of ewes under an equatorial regimen with a limited but efficient control of temperature change amplitude, induced cycles at irregular intervals with no clear anovulatory season [ 9 ] However, in other species it has been clearly demonstrated that the combination of photoperiod and temperature is responsible for seasonal changes in reproductive activity [ 29 , 30 ]. Thus, it is possible that, in the absence of a major cue (photoperiod), seasonal ovulatory activity of females of the TR groups has been entrained by temperature. In Syrian hamsters, exercise by access to a running wheel can completely inhibit the short-day induced regression of the testis [ 31 , 32 ]. Experimental Creole goats and Black Belly ewes of the present experiment were raised in light-proof rooms where physical exercise was limited, whereas in their original management conditions where the initial observations were done [ 15 , 17 ], animals were maintained at pasture. Thus, it is possible that in their original management conditions, physical exercise prevented the inhibitory effects of the 13 hours of light that was observed in our experimental light-proof rooms in the TR groups of goats and ewes. The sheep and goat breeds used here are local breeds of the Carribean islands. Even though their presence in these islands is associated with human history, they are considered as the progeny of tropical but not European ancestors, because they originate from the West coast of tropical Africa (see Methods). In these areas all sheep and goat breeds are considered as non or low seasonal breeders. Thus, the sensitivity of the Creole goats and Black-Belly ewes observed here under temperate photoperiod could be hypothetized as a true sensitivity of these breeds, not a simple inheritance of a trait coming from European ancestors. Conclusions A marked seasonality in the ovulatory activity of tropical Creole goats and Black-Belly ewes was induced when females were exposed to a simulated temperate photoperiod. Unexpectedly, and differing from the results obtained in their original breeding location, Black-Belly ewes and, to a lesser extent, Creole goats exposed to a simulated tropical photoperiod also showed significant seasonality in their ovulatory activity. Such results indicate that both species are capable of showing seasonality under the the photoperiodic changes of the temperate zone even though they do not originate from these regions. Methods Production of experimental animals from deep-frozen embryos Experimental animals from both species were produced after importation of deep-frozen embryos. The embryos were thawed and re-implanted into recipient females of Saanen goats for Creole embryos (1983) and of Ile-de-France sheep for Black-Belly embryos (1997). Embryos from donor Creole goats were collected as previously described [ 33 ] from 4 genetic families considered as representative of the native population of the Caribbean island of Guadeloupe (French West Indies). This breed has been raised in Guadeloupe for several centuries and probably originates from the West coast of Africa which it was imported from in the 17 th -18 th centuries [ 34 - 37 ]. Creole goats from Guadeloupe have many similarities with the "West African dwarf goat" regarding their size, coat color, fertility and prolificacy, growth rate and horn shapes [ 34 , 37 ]. Common genetic markers were found between Creole goat from Guadeloupe and West African goats [ 36 ], which reinforced the hypothesis of an African origin for the Creole goat. The first generation of animals, originating from deep-frozen imported embryos, was not used in the experiment. They were raised, with their progeny, under tropical photoperiodic conditions in light-proof buildings, as described later. The 3 rd and 4 th generations, constituting a sufficient number of animals, made up the two experimental groups. Embryos from donor Black-Belly ewes were collected using the technique described by Heyman et al. [ 38 ]. The donor females belonged to an INRA flock raised in Guadeloupe and were part of the 6 different families constituting this flock, bred from genitors from Martinique (F.W.I.) [ 39 , 40 ] and Barbados. This flock was considered to present characteristic production traits of the Black-Belly sheep population of the Caribbean [ 41 ]. As for goats, Black-Belly sheep is considered to have an African origin about 3 to 4 centuries ago [ 11 , 42 ]. This is confirmed by their phenotypic characteristics of hair sheep (i.e. not wool sheep), including performance traits [ 43 ]. After checking the absence of Blue-tongue virus in the collection media, embryos were re-implanted into Ile-de-France ewe lambs, 2 embryos inserted per recipient ewe. In the case of sheep, the first generation originating from deep-frozen embryos was used directly in the present experiment. After birth, all lambs were immediately placed under artificial suckling conditions, in the two experimental groups in light-proof rooms under tropical or temperate photoperiod, until the start of blood sampling for progesterone determinations. Animals and feeding conditions Both experiments were performed at the INRA Station near Tours (latitude 47°25 North). - Thirty three Creole female goats were used from October 1989 when the animals were one (n = 15 females) and two (n = 18 females) years old, for 33 months to June 1992 . They were divided into two groups (n = 17 TR and 16 TE) in visual and tactile contact, with entire and vasectomized Creole bucks, but separated by a fence. Each group was maintained in a separate light-proof room throughout the experiment. Feeding conditions were constant throughout the experiment. The animals were fed once daily with a diet of 240 g of barley, 60 g of wheat, 700 g of meadow hay and 300 g of barley straw. No flushing was performed. They had free access to water and to mineral blocks containing oligoelements and vitamins. - Twenty four Black Belly ewes were used from September 1998 when the animals were 6 months old, for 25 months, to October 2000. They were divided into two groups (n= 12 TR and 12 TE) in visual and tactile contact, with entire Black Belly rams, but separated by a fence. Each group was maintained in a separate light-proof room throughout the experiment. Feeding conditions were constant throughout the experiment. The animals were fed once daily with a diet of 150 g of corn, 110 g of barley, 45 g of dehydrated protein complement and 400 g of hay. No flushing was performed. They had free access to water and to mineral blocks containing oligoelements and vitamins. Photoperiodic treatments Within each species, one group was subjected to the large photoperiodic changes prevailing at the 45° North latitude (8 to 16 h of light per day from winter to summer solstices); this group was called the temperate group (TE). The other group was subjected to the limited photoperiodic changes prevailing at the 16° North latitude (11 to 13 h of light per day from winter to summer solstices); this group was called the tropical group (TR). In all rooms, photoperiod was regulated by an electric clock that operated bulbs providing an intensity of 300 lux, lateral to the animals' eyes. The photoperiod was adjusted by 15 min shifts (more or less frequent depending on the slope of the natural changes in daylength) to produce a complete photoperiodic cycle every 365 days. The four rooms were adjacent and of the same size (30 m 2 ). Temperature was not controlled but variations were parallel to those monitored outside but with a lower amplitude (minimum +1°C in January, maximum +29°C in August). Measurements and samplings Liveweight of Creole goats at the beginning of the experiment was 24.6 (± 3.4, sd) kg in group TE and 25.5 (± 3.0) kg in group TR. Liveweight was measured monthly and showed a regular increase until the end of the experiment (55.7 ± 7.5 and 56.2 ± 5.2 kg for TE and TR respectively). Liveweight of Black-Belly at the beginning of the experiment was 33.2 (± 2.5 sd) kg in group TE and 31.9 (± 2.7) kg in group TR. Liveweight was measured monthly and showed a regular increase until the end of the experiment (49.5 ± 5.9 and 47.0 ± 5.8 kg for TE and TR respectively). Ovulatory activity was assessed twice weekly in goats up to end October 1991 and once weekly thereafter; and once weekly in ewes, using blood samples for the progesterone radioimmunoassay. A rapid assay was performed using the technique described by Terqui and Thimonier [ 44 ]. When progesterone concentration was lower than 1.0 ng per ml of plasma in female goats and 0.75 ng/ ml of plasma in ewes, the female was considered to be in the follicular phase of the cycle or in anovulation. Definitions and analysis of results, statistical tests The first occurrence of a positive Progesterone sample was considered as the date of the first ovulation of the season, and the last occurrence of a positive Progesterone sample was considered as the date of the last ovulation of the season. Mean duration of ovulatory activity is the number of days between first and last ovulation in the same breeding season. Mean duration of anovulation is the number of days between last ovulation in a breeding season and first ovulation of the next season. Females which did not present cessations of their ovulatory activity were not included in the calculations of duration of the breeding seasons or duration of the anovulatory periods. In September and October 2000, individual blood samplings were stopped in ewes that had resumed their ovulatory activity. Mean dates of onset and end of the breeding season, durations of the breeding season and of anovulatory activity were compared between groups using an unpaired t-test. Variances were compared with F-Tests. Percentages of females showing at least one ovulation per month were analyzed using the Chi 2 method. (Statview ® , Abacus Concept, Berkeley, Ca, USA). All procedures were performed in accordance with French legal requirements and with the Ministry of Agriculture authorization for animal experimentation nb A37801 . Authors' contributions The authors contributed equally to this work. PC conceived the study, and was responsible for its design and coordination. AD followed the experiment in goats and DC the experiment in sheep. YC and GA performed all the embryo transfer procedures, and were in charge of pathological analyses. AD, DC and PC analysed the data. PC drafted the manuscript. All authors read, corrected and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC521071.xml
528727	Weight loss dynamics during combined fluoxetine and olanzapine treatment	Background Fluoxetine and olanzapine combination therapy is rapidly becoming an effective strategy for managing symptoms of treatment-resistant depression. Determining drug-drug interactions, drug metabolism and pharmacokinetics is of particular interest for revealing potential liabilities associated with drug augmentation in special patient populations. In the current studies, we chronically administered fluoxetine and olanzapine in non-stressed rats to extend our previous findings regarding body weight dynamics. Results Chronic fluoxetine (10 mg/kg) and olanzapine (5 mg/kg and 0.5 mg/kg) treatment decreased weight gain irrespective of olanzapine dosing. At the 10 mg/kg and 5 mg/kg dose, respectively, fluoxetine and olanzapine also significantly reduced food and water consumption. This pharmacodynamic event-related effect, however, was not observed at the 10 mg/kg and 0.5 mg/kg dosing paradigm suggesting differences in tolerability rates as a function of olanzapine dose. The decrease in weight gain was not associated with apparent changes in glucose metabolism as vehicle- and drug-treated rats showed undistinguishable serum glucose levels. The combination of fluoxetine and olanzapine in rats yielded drug plasma concentrations that fell within an expected therapeutic range for these drugs in psychiatric patients. Conclusions These data suggest that fluoxetine and olanzapine treatment decreases weight gain in rats; a pharmacodynamic event-related effect that differs considerably from what is observed in the clinical condition. The possibility of mismatched models regarding body weight changes during drug augmentation therapy should be seriously considered.	Background Treatment-resistant depression is a serious issue in psychiatry as a significant number of affected individuals show an inadequate response to single antidepressant therapy. An emerging strategy to achieve maximum mood stabilization for treatment-resistant depression, bipolar illness and depression with psychotic features is the augmentation of fluoxetine (Prozac) with novel anti-psychotic agents such as olanzapine (Zyprexa). Indeed, a number of clinical trials have suggested that such an augmentation strategy offers superior efficacy for treating resistant major depression when compared with either fluoxetine or olanzapine alone [ 1 - 3 ]. Despite the apparent clinical benefits of this drug strategy, little is known about the mechanisms by which fluoxetine plus olanzapine actually function to relieve depression. The limited literature on this issue suggests that drug augmentation therapy, at least in the rat brain, is likely to be more complicated and perhaps more indirect than a simplistic version of fluoxetine or olanzapine would imply [ 4 - 7 ]. For instance, whereas fluoxetine and olanzapine alone activate several signaling pathways involved in cell survival and plasticity [ 8 - 10 ], drug augmentation therapy reduces the levels of certain transcription factors (e.g., cAMP response element binding protein and FOS-like proteins) implicated in the chemical circuitry (e.g., prefrontal cortex and hippocampus) underlying emotional behaviors [ 5 ]. Consequently, it is conceivable that fluoxeine plus olanzapine treatment is effective against treatment-resistant depression due to their combined actions on numerous brain regions and various interconnected intracellular signaling pathways that ultimately promote some type of prophylactic effect. We have previously shown that sub-chronic (i.e., 7 days) administration of fluoxetine plus olanzapine results in a significant reduction of weight gain in rats [ 5 ]. This finding is of significant interest as fluoxetine and olanzapine alone have distinct and opposite effects on body weight dynamics in both rodents and humans. For example fluoxetine often reduces food intake and thus body weight in rats during sub-chronic and chronic (i.e., 21 days) drug regimens [ 11 ], an effect apparently mediated by fluoxetine impact on serotonin (5-HT) signaling pathways [ 12 ]. In sharp contrast, treatment with olanzapine is associated with significant weight gain in schizophrenic patients, a serious side effect that may increase the risk for type II diabetes and may also lead to treatment non-compliance [ 13 , 14 ]. In this case it is thought that olanzapine's particular affinity for 5-HT (5-HT 2A ), dopamine (DA; D 2–4 ), acetylcholine muscarinic (ACh; M 1 –M 5 ) and histamine (H 1 ) receptors distributed widely in limbic neural circuits may somehow account for the pharmacological basis of olanzapine-induced weight gain [ 15 ]. Needless to say, understanding body weight dynamics in relation to drug augmentation therapy is of critical importance if we are going to gain further knowledge on the mechanisms of therapeutic action and side effect profile of anti-depressant medications. In this regard, appetite disturbances are noted in many medicated depressed patients and several peptide transmitters implicated in feeding behavior co-exist in the hypothalamus and may therefore be involved in the onset of affective states [ 16 ]. In this study, we have examined in more detail the effects of fluoxetine (10 mg/kg) plus olanzapine treatment on rat body weight during the time course of 18 days under two olanzapine doses: 5 and 0.5 mg/kg. In addition, we have measured blood levels of these two drugs using gas-chromatography-mass spectrometry (GC-MS) to assess their combined pharmacology and their correlation to body weight dynamics. Results All rats tolerated the fluoxetine plus olanzapine regimen well. There were no mortalities as a result of 18 days of drug administration in any of the rat groups tested. The only apparent untoward side effect was tissue necrosis in the peritoneum of rats injected with fluoxetine plus 5 or 0.5 mg/kg olanzapine (Fig. 1 ). Thus, fluoxetine appears to produce focal necrotising vasculitis within the site of injection. The necrotic properties of the above antidepressant have previously been reported [ 17 ]. Olanzapine, on the other hand, does not produce tissue necrosis in the peritoneal cavities of rats when administered alone (data not shown). Figure 1 Tissue necrosis during chronic fluoxetine (fluox) and olanzapine (olanz) treatment. This figure depicts equally excised peritoneal cavities of males injected IP with either a vehicle-solution (cyclodextrin) or the above drug combination pattern for 18 consecutive days. Note the extent of tissue damage (~1 cm wide) at the site of drug administration. Focal necrosis was evident in drug-treated rats irrespective of olanzapine dosing. All rats showed a steady increase in body weight during the 18 days of cyclodextrin or fluoxetine plus olanzapine treatment. However, fluoxetine in combination with olanzapine significantly retarded this growth rate (P ≤ 0.001) when compared with the vehicle-treated group (Fig. 2 ). This weight loss, beginning on day 7 of treatment, was observed equally in both the 5 and 0.5 mg/kg olanzapine-treated groups. At this time, rats administered with cyclodextrin showed body weights of 275.6 ± 3.7 g, whereas fluoxetine plus 5 mg/kg olanzapine-treated animals showed weights of 248.1 ± 4.4 g (at the 95% confidence interval for differences between means: 13.8 to 41.2, P ≤ 0.001). Along the same lines on day 7, rats treated with fluoxetine plus 0.5 mg/kg olanzapine showed body weights of 254.1 ± 3.7 g. In contrast, their vehicle-treated cohorts showed weights of 277.8 ± 7.3 g (at the 95% confidence interval for differences between means: 6.90 to 40.41, P ≤ 0.01). The magnitude of this difference in body weight increased further on day 14 and was firmly established by day 18 of drug augmentation therapy (Fig. 3 ). Thus, fluoxetine treatment, irrespective of olanzapine's ability to cause weight gain, produces a gradual and considerable weight loss in male rats. In general, these findings are consistent with mono-therapy studies where chronic olanzapine treatment invariably leads to weight loss in rodents [ 18 , 19 ]. Figure 2 Body weight changes during chronic fluoxetine (fluox, 10 mg/kg) and olanzapine (olanz, 5 mg/kg) treatment. Rat body weights were recorded before and after drug augmentation therapy. Data represent means ± SEM. N = 5–7 animals per group. P ≤ 0.05 when compared with drug-treated rats. NS = not significant. Figure 3 Changes in food and water intake during chronic fluoxetine (fluox, 10 mg/kg) and olanzapine (olanz, 5 mg/kg) treatment. Rats under this combined drug regimen showed a significant reduction in the consumption of nutrients and fluids at day 10 and 12 of drug therapy, respectively. Data represent means ± SEM. N = 5–7 animals per group. P ≤ 0.05 when compared with drug-treated rats. The fact that fluoxetine plus olanzapine treatment for 18 days retards the continuous weight gain observed in cyclodextrin-exposed rats suggests at least two testable possibilities. First, rats exposed to drug augmentation therapy might be eating less than their cyclodextrin-treated cohorts. Second, administration of fluoxetine plus olanzapine might be altering glucose metabolism of drug-treated animals. To test the first possibility, we measured average food intake over a 12 hr period of the dark cycle in rats treated with fluoxetine plus 5 mg/kg olanzapine. On day 10 of drug treatment, rats exposed to this drug augmentation regimen ate significantly less (t 10 = 5.5, P ≤ 0.001) than vehicle-treated animals (Fig. 4 ). Interestingly, the same group of rats also showed a significant reduction in water intake (t 10 = 6.7, P ≤ 0.01) when compared with cyclodextrin-treated animals (Fig. 4 ). Thus, rats exposed to fluoxetine plus 5 mg/kg olanzapine are eating (~32%) and drinking (~38%) less at day 10 and 12 of drug treatment, respectively. Figure 4 Body weight changes during chronic fluoxetine (fluox, 10 mg/kg) and olanzapine (olanz, 0.5 mg/kg) treatment . Rat body weights were recorded before and after drug augmentation therapy. Data represent means ± SEM. N = 5–7 animals per group. P ≤ 0.05 when compared with drug-treated rats. NS = not significant. In contrast to the above findings, animals injected first with fluoxetine and then 15 min later with 0.5 mg/kg olanzapine did not show differences in food (t 10 = 2.2, P ≥ 0.05) or water (t 10 = -1.2, P ≥ 0.05) consumption during the dark cycle when compared with cyclodextrin-treated rats (Fig. 5 ). Thus, although fluoxetine plus 0.5 mg/kg olanzapine-treated rats show a significant and progressive weight loss at days 7, 14 and 18, this weight loss is not associated with reductions in food or water consumption. This finding suggests that chronic olanzapine treatment is apparently modifying feeding behavior in rats, an effect particularly conspicuous at the 5 mg/kg dose. To test the second possibility, that chronic fluoxetine plus olanzapine is altering the metabolism of drug-treated animals, we measured their blood glucose levels under fasting conditions (Fig. 6 ). Glucose levels at the time of sacrifice were not significantly different between cyclodextrin- and drug-treated animals at either the 5 mg/kg olanzapine dose (t 10 = -0.73, P ≥ 0.05) or the lower 0.5 mg/kg olanzapine dose (t 10 = -1.4, P ≥ 0.05). Thus, changes in glucose metabolism are not the proximate cause affecting the differential body weight dynamics, nor the differential consumption of food and water among vehicle- and drug-treated rats. Along the same lines, levels of the hormone leptin did not differ between cyclodextrin- and drug-treated animals (data not shown), thus suggesting that chronic fluoxetine and olanzapine drug therapy does not affect leptin messages under fasted conditions in male rats. Figure 5 No changes in food consumption or water intake during chronic fluoxetine (fluox, 10 mg/kg) and olanzapine (olanz, 0.5 mg/kg) treatment. Rats under this combined drug regimen did not show an apparent reduction in the consumption of nutrients and fluids at day 10 and 12 of drug therapy, respectively. Data represent means ± SEM. N = 5–7 animals per group. NS = not significant. Figure 6 No changes in fasting glucose levels after chronic fluoxetine (fluox, 10 mg/kg) and olanzapine (olanz, 5 mg/g or 0.5 mg/kg) dosing. Rats under the depicted drug regimens did not show overt differences in glucose metabolism. Data represent means ± SEM. N = 5–7 animals per group. NS = not significant. Non-fasting glucose levels in rats are typically in the range of 155–242 mg/dL. We next assessed the pharmacokinetic profile of fluoxetine and olanzapine in rats treated with the above drug combination pattern (Table 1 ). At doses of 10 mg/kg and 5 mg/kg respectively, fluoxetine plasma levels ranged from 62 ng/mL to 476 ng/mL. In contrast norfluoxetine levels ranged from 292 ng/mL to 1175 ng/mL. Norfluoxetine is the only identified active metabolite of fluoxetine; it is formed through N-demethylation of the parent molecule. Plasma concentrations of olanzapine ranged from 74 ng/mL to 301 ng/mL. At doses of 10 mg/kg fluoxetine and 0.5 mg/kg olanzapine, levels of the antidepressant drug ranged from 276 ng/mL to 576 ng/mL, whereas those for norfluoxetine ranged from 266 ng/mL to 966 ng/mL. Olanzapine levels at this low dose were in the range of 34 ng/mL and 65 ng/mL. When inter-group comparisons of drug plasma concentrations were made between fluoxetine (10 mg/kg) in combination with olanzapine at either the 5 mg/kg or the 0.5 mg/kg dose range, differences in mean values of the two groups did not vary enough (P ≥ 0.05) to reject the possibility of random sampling variability. A similar statistical trend was observed for norfluoxetine levels after 18 consecutive days of drug treatment; there was not a statistically significant difference (P ≥ 0.05) between the two groups. However, differences in the median values between 5 mg/kg and 0.5 mg/kg olanzapine doses were greater than would be expected by chance (P ≤ 0.001). As expected, rats injected IP with cyclodextrin showed no traces of either fluoxetine or olanzapine levels in plasma (≤5 ng/mL). In general, our GC-MS measurements detect pharmacological and relevant levels of both fluoxetine and olanzapine in rats, a finding consistent with previous reports [ 4 ]. Table 1 Plasma concentrations of fluoxetine, norfluoxetine and olanzapine after 18 consecutive days of drug augmentation therapy. Drug Measured Fluoxetine Dose (10 mg/kg) Olanzapine Dose (5 mg/kg) Fluoxetine 344.1 ± 54.2 (ng/mL) Norfluoxetine 695.4 ± 118.4 (ng/mL) Olanzapine 178.5 ± 34 (ng/mL) Drug Measured Fluoxetine Dose (10 mg/kg) Olanzapine Dose (0.5 mg/kg) Fluoxetine 410.0 ± 36.6 (ng/mL) Norfluoxetine 501.5 ± 114.7 (ng/mL) Olanzapine 46.7 ± 4.4 (ng/mL) The fact that regardless of olanzapine dosing both rat groups lost equal amounts of body weight is indicative that these two phenomena (i.e., relative drug levels and body weight dynamics) may not be causally related. No discernible changes in brain structure or integrity were found, as assessed by Nissl staining and stereological cell counts conducted within the rat hypothalamus (data not shown). Values are means ± SEM. N = 7 per dosing group. * P ≤ 0.01 when compared with appropriate olanzapine dose. Discussion The present study shows that 18 days of concomitant fluoxetine and olanzapine treatment leads to a significant decrease of weight gain in rats. Given that the above drug combination is particularly effective in treatment-resistant depression, our findings are of interest for revealing potential liabilities associated with its therapeutic use. Here, our data suggest a possible pharmacodynamic event-related effect regarding the action of two psychoactive drugs over time. Indeed, there is a well-established relationship between clinically effective drugs, appetite control and weight changes across diverse patient populations [ 20 ]. For instance, weight gain appears to be correlated positively with clinical responses to anti-psychotic medication [ 21 , 22 ]. The combination of fluoxetine and olanzapine in our studies produced weight loss irrespective of anti-psychotic drug dosing. That is, fluoxetine at a fixed dose of 10 mg/kg administered concomitantly with either 5 mg/kg or 0.5 mg/kg olanzapine yielded an approximately 20% mean reduction in body weight for both doses. Therefore, body weight changes associated with the above drug combination are more likely due to the effects of olanzapine or its metabolic pathways (see below). Indeed, this hypothesis is further supported by the fact that although rats treated with 5 mg/kg olanzapine were eating and drinking less than animals injected with a smaller dose (i.e., 0.5 mg/kg), body weight outcome was nevertheless similar for all drug-treated rats. In this regard, it is conceivable that the suppressed consumption of food and water observed in animals injected IP with 5 mg/kg olanzapine might have been the result of malaise, or at least the result of aversion to the hedonic aspects of food and water. However this possibility does not explain, in general, the sustained and consistent decrease in weight gain for all rats treated with fluoxetine and olanzapine. Changes in glucose metabolism were also ruled out as a causal role for the reduction in weight gain and food intake as both vehicle-and drug-treated animals showed undistinguishable serum glucose levels during fasting. Further studies of these questions will yield insight into centrally acting peptides and/or peripherally acting thermogenic mechanisms underlying decreases in weight gain in adult rats. Placing our results in the framework of clinical situations, decreases in rat weight gain as a result of fluoxetine and olanzapine treatment do not mirror the profile occurring across diverse patient populations. There is evidence that long-term fluoxetine plus olanzapine treatment frequently leads to weight gain in individuals with major depressive disorders with and without treatment-resistant depression [ 23 ]. Further, high doses of fluoxetine do not appear to counteract the weight gain often induced by atypical anti-psychotics such as olanzapine [ 24 ]. The stark disparity between rat and human studies regarding body weight dynamics raises the possibility of mismatched models for revealing certain liabilities associated with fluoxetine and olanzapine therapy in mood disorders. It is conceivable, for instance, that rats might be more sensitive to the anorectic effects of fluoxetine than humans. Fluoxetine is known to produce anorectic effects that often lead to a decrease in weight gain; a phenomenon observed equally at the experimental and clinical level [ 11 , 25 , 26 ]. Alternatively, olanzapine metabolism may differ significantly in rats as indirectly suggested by previous reports [ 27 , 28 ]. In this context, olanzapine is metabolized to its 10- and 4'-N-glucuronides, with the 10-N-glucuronide being the most abundant metabolite in humans [ 15 , 29 ]. As the pharmacokinetic and pharmacodynamic profile of olanzapine in rats is relatively obscure [ 27 ], it is possible that changes in glucuronidation metabolism in rodents may have impacted the ability of the parent drug to influence heterogeneous population of cells associated with body weight dynamics. From these statements, one might conclude that our findings are not clinically significant and perhaps of limited value for additional investigation. Although the animal data indeed do not support the clinical situation, the above findings could harbor important information as to how species-specific differences limit drug-drug interactions or body weigh regulation, lessons that could influence subsequent studies regarding fluoxetine and olanzapine therapy in more defined experimental settings. In the present study, measurements of fluoxetine, norfluoxetine and olanzapine plasma concentrations were made to assess their pharmacology after 18 days of combined drug exposure. In general, drug plasma levels fell within the expected therapeutic range typically observed in psychiatric patients. For instance, after 30 days of dosing at 40 mg/day, plasma levels of fluoxetine are in the range of 90–300 ng/mL across diverse patient populations [ 15 ]. In our animal studies, at a dose of 10 mg/kg (IP), mean plasma concentrations achieved were in the range of 300–400 ng/mL after 18 days of combined drug treatment. Oral doses of olanzapine at 20 mg/day often yield plasma levels of 20–100 ng/mL in healthy volunteers and in patients with schizophrenia [ 30 ]. Concentrations ≥80 ng/mL are considered threshold for the occurrence of adverse effects. In our present study, at a dose of 5 mg/kg olanzapine, mean plasma levels achieved of the anti-psychotic drug were ~178 ng/mL. The relatively high levels of olanzapine may help explain in part the hypophagic and adipsic phenomena experienced by rats at this particular dosing. Interpreted in this way, olanzapine concentrations ≥80 ng/mL (as in our studies) reached a threshold for the onset of malaise or taste aversion effects. In contrast, animals exposed to a 0.5 mg/kg olanzapine dose showed optimal therapeutic range of olanzapine plasma levels (~47 ng/mL) and normal feeding and drinking behaviors. It should be noted that the dosing paradigm implemented in our current studies yielded fluoxetine, norfluoxetine and olanzapine plasma concentrations similar to those reported by Zhang et al [ 4 ] under an acute experimental design. Therefore it is possible that little or no significant metabolic interactions between fluoxetine and olanzapine combination treatment occurs in rats as a function of chronic drug exposure. This possibility has merit as no clinically significant metabolic interactions are also reported during combined fluoxetine and olanzapine therapy [ 29 ]. Placing the current data in the framework of the growing body of experimental and clinical evidence, it is unlikely that drug-drug interactions modify the pharmacological profile of fluoxetine and olanzapine when the two psychoactive agents are administered concomitantly to experimental animal models. Conclusions Combination therapies of anti-depressant and anti-psychotic drugs are increasingly used for treatment-resistant mood disorders. Here, we have provided further evidence that fluoxetine and olanzapine have pharmacodynamic event-related effects on body weight dynamics [ 5 ]. In rats, these effects are manifested in the form of anorexia or perhaps anhedonia to food and water. Of interest, anorectic phenomena are also observed in rats chronically treated with valproic acid and lithium [ 31 ]; both valproic acid and lithium are widely touted as effective prophylactic agents for manic-depressive illness [ 32 ]. It is quite probable therefore that augmentation therapy of several mood stabilizers is associated with weight loss in rats, whereas the same combination drug pattern results in weight gain in special patient populations. This disparity adds a new level of complexity to the issue of body weight changes associated with psychopharmacology [ 19 ], and indicates species-specific variations in this phenomenon. In our particular case, adjusting olanzapine dosing to rat studies from 5 mg/kg to 0.5 mg/kg should preclude malaise bouts and/or taste aversion effects. In addition, the above dosage modification should be considered for achieving clinically therapeutic anti-psychotic plasma levels. If such a dosing paradigm is overlooked, it may lead to erroneous conclusions regarding mechanisms of medication action and side effect profile during drug augmentation therapy. Methods Animals and drug administration Adult male Long-Evans rats (Harlan, Indianapolis, IN) were used in all experiments described herein. Prior to any drug treatment, all rats were handled for 5 days to minimize non-specific stress. Rats were then randomly assigned to the various experimental groups and cage mates received the same drug treatment. Animals were group-housed, 2–3 per cage under a 12 hr light:dark cycle (light on 0700) and allowed ad libitum access to food and water, except when noted (see below). For the chronic drug regimen, rats were injected intraperitoneally (IP) first with fluoxetine (10 mg/kg) followed 15 min later by olanzapine (5 or 0.5 mg/kg) to decrease potential pharmacokinetic interactions. Fluoxetine was dissolved in 5% γ-cyclodextrin, whereas olanzapine was dissolved in 12% γ-cyclodextrin (cyclodextrin was used to improve the stability and bioavailability of poorly soluble drugs). Dosages of the two drugs were chosen according to each drug's in vivo potencies for affecting 5-HT, DA, ACh and H systems [ 15 , 33 , 34 ], and also from pharmacological doses reported in the literature [ 4 , 8 , 11 , 35 , 36 ]. Doses of drugs are expressed as their respective salts. Control animals received 5% and 12% γ-cyclodextrin injections (1 ml/kg) at 15 min intervals so that this group was given the vehicle-solution at the same times as the fluoxetine plus olanzapine experimental group. All injections were administered between 1000 and 1100 hr of the light cycle. All aspects of the following experiments were carried out in accordance with the NIH Guide for the Care and Use of Laboratory Animals and with approval from the NYIT IACUC. Experimental procedures Rats were injected with fluoxetine plus olanzapine or their respective vehicle-solutions for 18 consecutive days and body weights recorded before (day 0) and after (day 18) drug treatment. In addition, body weights were also recorded 7 and 14 days after drug treatment to adjust for dosage. To determine average food intake over a 12 hr period, control-vehicle and experimental rats were given pre-measured food pellets (25 g/rat) on day 10 and subsequent consumption was recorded on the next day. To keep track of food spillage, cage bedding was randomly separated to assess degree of unconsumed food. A similar procedure was instituted to assess average water intake over a 12 hr period: both groups were given pre-measured tap water (100 mL/rat) on day 12 and subsequent consumption was recorded on the next day. On the last day of injections (day 18), rats were either decapitated or perfused under deep chemical anesthesia (ketamine/xylazine/acepromazine, 60 mg/kg) with 4% paraformaldehyde. Trunk blood or blood collected from cardiac puncture was collected in centrifuge tubes containing either no EDTA or a 0.3 M EDTA (pH 7.4) solution. Blood samples with 0.3 M EDTA were centrifuged at 10,000 RPM for 10 min and the collected serum frozen at -80°C until determination of glucose levels. Blood collected without EDTA was also frozen at -80°C until determination of drug plasma concentrations by GC-MS. Glucose measurements To determine relative glucose levels, all animals were fasted for 12 hr on the last day of injections (i.e., day 18). Glucose serum levels were determined using the Life-Scan One Touch Basic Meter (Johnson & Johnson, New Brunswick, NJ). In brief, 10 μl of serum was pipetted as a free-flowing drop onto each One Touch Test Strip and read in the glucose meter for 45 sec. Fasting glucose levels are presented as means ± SEM in mg/dL. Analysis for fluoxetine, norfluoxetine and olanzapine by GC-MS All solvents used for drug analyses were HPLC grade. Chloroform and 1-chlorobutane were obtained from Burdick & Jackson (Muskegon, MI). Ethyl Acetate and acetonitrile were obtained from EMD (Gibbstown, NJ). Ammonium Hydroxide, ACS certified, was obtained from Fisher (Fairlawn, NJ). Ascorbic acid was obtained from Sigma (St. Louis, MO), whereas trifluoroacetic anhydride was obtained from Pierce (Rockford, IL). Working solutions containing olanzapine, fluoxetine and norfluoxetine at 10 ng/μL, 1.0 ng/μL, and 0.1 ng/μL, respectively were prepared in methanol. The working solutions used to prepare the calibration standards and controls were derived from different sources of reference material (e.g., fluoxetine) or different weighing of the same reference material (e.g., olanzapine). Calibration standards and controls were prepared by adding appropriate amounts of working solutions to clean, separate silanized 16 × 100 mm culture tubes that contained 1 mL of blank bovine blood and 0.2 mL of 2.5% ascorbic acid. The calibration standards ranged from 1 ng/mL to 1000 ng/mL. The controls were prepared at 35 ng/mL, 100 ng/mL, and 650 ng/mL. A 0.25 mL volume of each blood sample was transferred to clean silanized 16 × 100 mm culture tubes. To bring the sample preparation to a volume of 1 mL, a 0.75 mL volume of Milli Q H 2 O was added to each tube. The samples were therefore a 4-fold dilution compared with the standards and controls. A 0.2 mL volume of 2.5% ascorbic acid was added to each sample in a preparation tube. Sample, standards, and controls were extracted by a liquid/liquid procedure. Eighty ng of fluoxetine-d 6 (80 μL of 1 ng/μL fluoxetine-d 6 in Milli Q H 2 O) and 80 ng of clozapine (80 μL of 1 ng/μL clozapine in methanol) were added to each tube and the tubes were then briefly vortexed. A 0.1 mL volume of the concentrated ammonium hydroxide and a 4 mL volume of 1-chlorobutane: acetonitrile (4:1) was then added to each tube. A clean teflon-lined screw cap was placed on each tube. The tubes were mixed 20 min using a reciprocating shaker and centrifuged at 2000 rpm for 10 min using an IEC centrifuge (Needham, MA). Using clean, separate glass Pasteur pipettes, the upper organic layer from each tube was transferred to clean, separate 13 × 100 mm culture tubes. The organic layer was evaporated to dryness under a stream of air at 40°C using a Turbo Vap evaporator (Zymark Corporation, Hopkinton, MA). For derivatization, a 0.1 mL volume of chloroform and a 0.1 mL volume of trifluoroacetic anhydride were added to each tube. Clean teflon-lined screw caps were then placed on each tube. The tubes were heated for 20 min at 70°C using a dry block heater. After heating, the tubes were removed from the heater and allowed to cool at room temperature. The caps were then removed and the tubes evaporated using the same conditions as described above. Derivatized extracts were reconstituted with 100 μL of ethyl acetate and were then transferred to clean, separate auto-sampler vials. The GC-MS system consisted of an Agilent 6890 gas chromatograph and an Agilent 5973 MSD mass spectrometer (Palo Alto, CA). The data system consisted of a Hewlett-Packard X A 6/400 computer and Agilent Chemstation software. For chromatographic separation, a ZB-5, 30 meter × 0.25 mm id, 0.25 μm capillary column (Phenomenex, Torrance, CA) was used. The carrier gas was ultra high purity helium at a flow rate of 1.0 mL/min. The injection port temperature was 260°C and the transfer line temperature was 300°C. The column oven temperature program was 125°C, held at this temperature for 0.2 min, and then increased to 300°C. Positive chemical ionization was used for the mass spectrometry analysis. The ion source temperature was 200°C and ammonia was used as a reagent gas. Selected ion monitoring was used and the following ions (m/z) were analyzed: norfluoxetine: 409, fluoxetine: 423, fluoxetine-d 6 : 429, olanzapine: 409, clozapine: 423. For fluoxetine, norfluoxetine, and fluoxetine-d 6 , the protonated ammonia adducts of the molecule ions were monitored. For olanzapine and clozapine, the protonated molecule ions were also monitored. Fluoxetine-d 6 was used as the internal standard for fluoxetine and norfluoxetine. Clozapine was used as the internal standard for olanzapine. The limit of quantification for fluoxetine, norfluoxetine and olanzapine was 5 ng/mL. Data are presented as the means ± SEM. Data analysis Statistical comparisons in body weight and glucose levels were carried out using one-way ANOVA or two-tailed t tests where appropriate. Plasma levels of fluoxetine and olanzapine were analyzed by a Student's t -test. The probability level interpreted as significant was P ≤ 0.05. Authors' contributions JAP and JMC participated in the in vivo studies and in the biochemical assays. BHH and JMH participated in the design of the studies and performed the statistical analysis. GT drafted the manuscript, conceived the study and participated in its design and coordination. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC528727.xml
538252	Which is the best lipid-modifying strategy in metabolic syndrome and diabetes: fibrates, statins or both?	Although less clinical intervention studies have been performed with fibrates than with statins, there are evidences indicating that fibrates may reduce risk of cardiovascular events. The potential clinical benefit of the fenofibrate will be specified by the ongoing Fenofibrate Intervention and Event Lowering in Diabetes (FIELD) study, which rationale, methods and aims have been just published. Controlled clinical trials show similar or even greater cardiovascular benefits from statins-based therapy in patient subgroups with diabetes compared with overall study populations. Therefore, statins are the drug of first choice for aggressive lipid lowering actions and reducing risk of coronary artery disease in these patients. However, current therapeutic use of statins as monotherapy is still leaving many patients with mixed atherogenic dyslipidemia at high risk for coronary events. A combination statin/fibrate therapy may be often necessary to control all lipid abnormalities in patients with metabolic syndrome and diabetes adequately, since fibrates provide additional important benefits, particularly on triglyceride and HDL-cholesterol levels. Thus, this combined therapy concentrates on all the components of the mixed dyslipidemia that often occurs in persons with diabetes or metabolic syndrome, and may be expected to reduce cardiovascular morbidity and mortality. Safety concerns about some fibrates such as gemfibrozil may lead to exaggerate precautions regarding fibrate administration and therefore diminish the use of the seagents. However, other fibrates, such as bezafibrate and fenofibrate appear to be safer and better tolerated. We believe that a proper co-administration of statins and fibrates, selected on basis of their safety, could be more effective in achieving a comprehensive lipid control as compared with monotherapy.	Due to their beneficial effects on glucose and lipid metabolism, peroxisome proliferator activated receptors (PPAR's) alpha agonists (fibrates) are good potential candidates for reducing the risk of myocardial infarction (MI) in subjects with metabolic syndrome [ 1 - 3 ]. Although less clinical intervention studies have been performed with fibrates than with statins, there are evidences indicating that fibrates may reduce risk of cardiovascular disease and particularly non-fatal MI [ 4 - 10 ]. Interestingly, reduction of cardiovascular disease with one of the fibric acid derivates – gemfibrozil – was more pronounced in patients displaying baseline characteristics very similar to metabolic syndrome definitions [ 4 , 5 ]. There have been no direct head-to-head comparisons of a statin with a fibrate in any clinical endpoint trial. However, compared with statins, fibrates appear to more selectively target the therapeutic goals in obese individuals with features of insulin resistance and metabolic syndrome (i.e. with near-goal low-density lipoprotein (LDL)-cholesterol and inappropriate high-density lipoprotein (HDL)-cholesterol and triglyceride levels). The primary-prevention trial Helsinki Heart Study showed that treatment with gemfibrozil led to a significant reduction in major cardiovascular events [ 4 ]. Regarding secondary prevention, in the VAHIT study (Veterans Affairs High-density lipoprotein cholesterol Intervention Trial) – which included 30% of diabetic patients – gemfibrozil reduced the occurrence of major cardiovascular events by 22 % [ 5 ]. Similarly, reduction of cardiovascular disease with gemfribrozil was more pronounced in patients displaying above three of the features of metabolic syndrome [ 11 , 12 ]. In two previous small studies bezafibrate decreased the rate of progression of coronary atherosclerosis and decreased coronary event rate [ 6 , 7 ]. In another large trial in 1568 men with lower extremity arterial disease with a relatively short follow-up period, bezafibrate reduced the severity of intermittent claudication for up to three years. [ 8 ]. Ingeneral, the incidence of coronary heart disease in patients on bezafibrate has tended to be lower , but this tendency did not reach statistical significance. However, bezafibrate had significantly reduced the incidence of non-fatal coronary events, particularly in those aged <65 years at entry, in whom all coronary events may also be reduced [ 8 ]. In the Bezafibrate Infarction Prevention (BIP) study an overall trend of a 9.4% reduction of the incidence of primary end point (fatal or non-fatal myocardial infarction or sudden death) was observed. The reduction in the primary end point in 459 patients with high baseline triglycerides (≥200 mg/dL) was significant [ 9 ]. These results are consistent with studies in experimental models showing that pre-treatment of rats with the PPAR-alpha agonist clofibrate causes a significant reduction in induced myocardial infarct size of 43% [ 13 ]. Recently, reduced incidence of type 2 diabetes in patients with impaired fasting glucose level on bezafibrate has been demonstrated [ 14 ]. The potential clinical benefit of the other widespread fibric acid derivative, fenofibrate, on the reduction of cardiovascular disease is still unknown and will be specified by the ongoing Fenofibrate Intervention and Event Lowering in Diabetes (FIELD) study, which rationale, methods and aims have been just published [ 15 ]. It will be the largest (approximately 10000 patients) ever conducted fibrate-based controlled clinical trial in diabetic patients. The results are expected for 2005. An added strength of this trial is its ability to examine important clinical outcomes across diverse ethnic and gender subgroups. The results of this study will clarify whether the beneficial lipid-modifying effects of micronised fenofibrate lead to a reduction of cardiovascular morbidity and mortality. Despite increasing use of statins, a significant number of coronary events still occur and many of such events take place in patients presenting with the metabolic syndrome. Whereas statins remain the drug of choice for patients who need to achieve the LDL-cholesterol goal, fibrate therapy may represent the alternative intervention for subjects with atherogenic dyslipidemia typical for metabolic syndrome and an LDL-cholesterol already close to goal values. In addition, the concomitant use of fibrates seems to be attractive in patients whose LDL-cholesterol is controlled by statin therapy but whose HDL-cholesterol and/or triglyceride levels are still inappropriate [ 16 - 19 ]. This strategy will be tested in the ongoing Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial [ 20 ]. The factor that dominates in overweight-related metabolic syndrome is the permanent elevation of plasma free fatty acids (FFA) and the predominant utilization of lipids by the muscle inducing a diminution of glucose uptake and insulin resistance. Currently, an insulin-resistant state – as the key phase of metabolic syndrome – constitutes the major risk factor for development of macrovascular complications [ 21 - 23 ]. On the basis of the current concept of the evolution of adipogenesis via PPAR modulation toward insulin resistance and atherothrombotic macrovascular complications (including MI), the decreasing of plasma FFA and improving of insulin sensitization by PPAR agonists seems to be a logical and valuable goal for therapy. It is important to note that on a whole-body level, lipid and glucose metabolisms interact intimately. Briefly, PPAR alpha is activated by fibric acids (e.g. bezafibrate) and form heterodimers with the 9-cis retinoic acid receptor (RXR). These heterodimers bind to peroxisome proliferator response elements, which are located in numerous gene promoters and increase the level of the expression of mRNAs encoded by PPAR alpha target genes. Bezafibrate reduces triglyceride plasma levels through increases in the expression of genes involved in fatty acid-beta oxidation and by decrease in apolipoprotein C-III gene expression. Fibric acids increase HDL-cholesterol partly by increasing apolipoprotein A-I and apolipoprotein A-II gene expression. Their triglyceride-lowering and HDL-cholesterol raising effects lead to decreased systemic availability of fatty acid, diminished fatty acid uptake by muscle with improvement of insulin sensitization and reduced plasma glucose level [ 24 - 28 ]. Evidence also suggests that there is a 'fibrate effect' that mediates the reduction in CHD risk beyond the favorable impact of these agents on HDL-cholesterol levels. This last notion is consistent with the pleiotropic effects of fibrates which are known to be related to their mechanisms of action [ 29 ]. Being PPAR alpha ligands, fibrates have a significant impact on the synthesis of several apolipoproteins (apo) and enzymes of lipoprotein metabolism as well as on the expression of several genes involved in fibrinolysis and inflammation. Such changes contribute to improve the catabolism of triglyceride-rich lipoproteins, leading to a substantial increase in HDL-cholesterol levels accompanied by a shift in the size and density of LDL particles (from small, dense LDL particles to larger, more buoyant cholesteryl ester-rich LDL). Controlled clinical trials show similar or even greater cardiovascular benefits from statins-based therapy in patient subgroups with diabetes, impaired fasting glucose, and metabolic syndrome, compared with overall study populations. Therefore, statins are the drug of first choice for aggressive lipid lowering actions and reducing risk of coronary artery disease in these patients. However, current therapeutic use of statins as monotherapy is still leaving many patients with mixed atherogenic dyslipidemia at high risk for coronary events. A combination statin/fibrate therapy may be often necessary to control all lipid abnormalities in patients with metabolic syndrome and diabetes adequately, since fibrates provide additional important benefits, particularly on triglyceride and HDL-C levels. Thus, this combined therapy concentrates on all the components of the mixed dyslipidemia that often occurs in persons with diabetes or metabolic syndrome, and may be expected to reduce cardiovascular morbidity and mortality. Safety concerns about some fibrates such as gemfibrozil may lead to exaggerate precautions regarding fibrate administration and therefore diminish the use of the seagents. However, other fibrates such as bezafibrate and fenofibrate appear to be safer and better tolerated [ 30 - 36 ]. We believe that a proper co-administration of statins and fibrates, selected on basis of their safety, could be more effective in achieving a comprehensive lipid control as compared with monotherapy. Competing interests The author(s) declare that they have no competing interests.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC538252.xml
548951	Timing the Brain: Mental Chronometry as a Tool in Neuroscience	Mental chronometry, which has origins dating back over a century, seeks to measure the time course of mental operations in the human nervous system	How do we relate human thought processes to measurable events in the brain? Mental chronometry, which has origins that date back more than a century, seeks to measure the time course of mental operations in the human nervous system [ 1 ]. From the late 1800s until 1950, the field was built almost entirely around a single method: measuring and comparing people's reaction times during simple cognitive tasks. As far back as 1868, Franciscus Donders [ 2 ] subtracted the time taken to make a single response to an unvarying stimulus—what he called an instructed reflex—from the time it took to make the same response to one of two events, obtaining the time required to discriminate between the two stimuli. Further, he subtracted the time to discriminate two stimuli from a situation in which there were also two possible responses in order to obtain the time required for choice. In the 1950s, studies of reaction time were combined with the then-developing mathematical theory of information [ 3 ] to address issues such as the maximum transmission rate of the human nervous system [ 4 ] and how coding in the brain of stimuli and their responses could influence these limits [ 5 ]. These studies revealed that reaction time alone was not sufficient to elucidate the exact processes by which the brain achieved the human ability to process information. However, when combined with other methods, the latency of responding can help connect brain studies to the behavior of humans in real situations. Recording the average event-related electrical potentials from scalp electrodes became a research tool in the 1960s, with the advent of analog and then digital computers to accomplish the recording and averaging. It became clear that components of the event-related potential could be systematically related to sensory and motor stages of information processing. For instance, a visual stimulus was found to evoke a short-latency scalp response from the primary visual cortex at about 60 milliseconds, followed by positive and negative voltage changes in neighboring visual areas. Similarly, scalp recorded potentials from the frontal cortex could be recorded in relation to motor activity. It was now possible to observe some of the sensory and motor stages that were inferred from Donders's subtractive method (see [ 6 ] for a review). Saul Sternberg [ 7 ] developed a much-improved method for dividing reaction time into successive or serial stages, called the additive factors method. Subjects were asked to determine whether or not a probe digit had been present in a just previously presented series of digits. Sternberg argued that the time to complete the task could be divided into a sensory stage, dependent on stimulus parameters such as the intensity or clarity of the probe; a comparison stage, dependent only on the number of items in memory; and a response stage that reflected the difficulty of the specified response. Factors that influenced one stage (e.g., stimulus intensity) would be additive with those that influenced another stage (e.g., motor output). With this simple framework, it was now possible to determine at which stage(s) a new factor (e.g., nicotine, sleep deprivation, or Parkinson's disease) had its influence. In the 1950s, the advent of microelectrode recordings of single neurons from anesthetized monkeys allowed for an even finer resolution of neurophysiological processes and seemed to provide support for the view that the brain does indeed process information in serial stages. Hubel and Wiesel [ 8 ] argued that successive levels of the visual system could be seen as accomplishing successive analyses of input. The microelectrode strategy was quickly adopted to alert animals, making it apparent that higher level brain areas involved in operations upon input might feedback their influences on earlier processing stages [ 9 , 10 ]. These control systems, often called attention, posed something of a problem for completely serial views of information processing. However, they also provided evidence of localized brain areas within the parietal lobe of the monkey that could be systematically related to processing operations involved in attention—which were then being investigated by mental chronometry in patients with parietal and other lesions [ 11 ]. In the late 1980s, neuroimaging experiments made possible the examination of activity in localized brain areas, first through the use of injected radionuclides detected by positron emission tomography (PET) [ 12 ] and later through the use of an externally imposed magnetic field in functional magnetic resonance imaging (fMRI) [ 13 ]. Over the last ten years, fMRI has improved in spatial and temporal resolution and can now provide evidence of quite specific brain areas, in the millimeter range, that are involved in cognitive tasks. Most studies have shown a small number of widely distributed brain areas that must be orchestrated to carry out a cognitive task. Although, as in all sciences, there are disagreements, the convergence of results in areas of attention and language seem to me particularly impressive. When the fMRI method for localization is brought together with methods that can accurately measure timing of the same activity (i.e., electrical or magnetic event-related potentials) they can provide considerable insight into the nature of thought. Consider the simple task of deciding whether a presented digit is above or below five [ 14 ]. Dehaene argued that the task could be divided into four stages. The first involves obtaining the identity of the probe input (encoding), the second making a comparison against the stored representation of the digit five, the third selecting a response, and lastly, checking the output for error. According to additive factor theory, a variable that effects overall reaction time by varying the time to complete one stage will be additive with the effects of variables that affect other stages. The input or encoding stage was varied by using either Arabic or spelled digits (e.g., “3” or “three”). The comparison stage was varied by comparing digits close to five (e.g., six) with those far from five (e.g., nine). The response stage was varied by specifying a response from either the dominant or non-dominant hand and error monitoring was examined by comparing error with correct trials. Each of these variables influenced only the appropriate stage and was additive in its effect with each of the other variables (see Figure 1 ). Figure 1 Reaction Time for Various Conditions People were asked to judge whether a presented digit was greater or less than five. The time to respond (reaction time) varied systematically as a function of notation (Arabic digits vs. spelledout numbers), distance (closer or farther in sequence from five), and responding hand. The three effects are additive, indicating the likelihood of serial stages of cognitive processing. (Adapted from [ 14 ]) Moreover, scalp-recorded, event-related potentials showed a separate scalp distribution and latency for each variable [ 14 ]. Subsequent fMRI data has confirmed and increased the precision of the anatomy proposed for each of these stages. Of course, not all human activities involve a set of exhaustive and independent serial stages that can be shown to add up to the overall reaction time. While tasks like the number comparison discussed above can be usefully divided into stages, some components may deal with simultaneous operations and may be limited only by a total capacity of central mechanisms. We know that many situations involve parallel processing and feedback loops at many levels. Sternberg has attempted to apply a modified version of additive factor theory to brain systems using neuroimaging that allows for some of these possibilities [ 15 , 16 ]. Figure 2 Regions of the Brain Involved in a Number Comparison Task Derived from EEG and fMRI Studies The regions represented correspond to those showing effects of notation used for the numbers (pink and hatched), distance from the test number (orange), choice of hand (red), and errors (purple). (Illustration: Giovanni Maki; adapted from [ 18 ]) Laboratory studies often use the simultaneous execution of two different tasks (dual tasks) to simulate the more realistic situations where humans time-share activities. In this issue of PLoS Biology , Sigman and Dehaene [ 17 ] provide a model that further extends the additive factor logic to the dual task situation. They propose that for tasks that can be broken down into three consecutive stages—perception, decision based on noisy integration of evidence, and response—the perceptual and motor stages can operate both simultaneously with and independently of stages of another task and are thus easily amenable to additive factor analysis. The decision stage, however, appears to represent a kind of “cognitive bottleneck” for which the reaction times of the two tasks become interdependent. The model adds considerably to the range of situations to which an additive factor approach can be applied, allowing investigators to seek more information about how new variables influence hidden processing stages. Many cognitive and emotional tasks studied with neuroimaging have implicated a small number of brain areas that are consistently active. Mental chronometry plays a role in suggesting the cognitive operations that each of these areas performs and how they are organized in real time. The toolkit of new techniques provides the basis for further tests to evaluate whether a chronometric model reveals a crucial set of connected computations (circuit) for carrying out the task. For example, using a magnetic pulse delivered outside the skull, it is now possible to induce a reversible lesion at the time of a particular computation to determine whether the specific computation assigned to a given area is truly needed to carry out the task. Studies using diffusion tensor imaging can examine whether there are large-scale connections between neural areas posited by a particular model. In describing the links between brain and behavior, mental chronometry is still a cornerstone that binds psychology to the techniques of neuroscience.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548951.xml
529439	A touchdown nucleic acid amplification protocol as an alternative to culture backup for immunofluorescence in the routine diagnosis of acute viral respiratory tract infections	Background Immunofluorescence and virus culture are the main methods used to diagnose acute respiratory virus infections. Diagnosing these infections using nucleic acid amplification presents technical challenges, one of which is facilitating the different optimal annealing temperatures needed for each virus. To overcome this problem we developed a diagnostic molecular strip which combined a generic nested touchdown protocol with in-house primer master-mixes that could recognise 12 common respiratory viruses. Results Over an 18 month period a total of 222 specimens were tested by both immunofluorescence and the molecular strip. The specimens came from 103 males (median age 3.5 y), 80 females (median age 9 y) and 5 quality assurance scheme specimens. Viruses were recovered from a number of specimen types including broncho-alveolar lavage, nasopharyngeal secretions, sputa, post-mortem lung tissue and combined throat and nasal swabs. Viral detection by IF was poor in sputa and respiratory swabs. A total of 99 viruses were detected in the study from 79 patients and 4 quality control specimens: 31 by immunofluorescence and 99 using the molecular strip. The strip consistently out-performed immunofluorescence with no loss of diagnostic specificity. Conclusions The touchdown protocol with pre-dispensed primer master-mixes was suitable for replacing virus culture for the diagnosis of respiratory viruses which were negative by immunofluorescence. Results by immunofluorescence were available after an average of 4–12 hours while molecular strip results were available within 24 hours, considerably faster than viral culture. The combined strip and touchdown protocol proved to be a convenient and reliable method of testing for multiple viruses in a routine setting.	Background Acute respiratory tract infections are major causes of morbidity and mortality. In 2000, lower respiratory tract infections were globally the number one infectious cause of disability adjusted life-years [ 1 ]. The commonest respiratory viruses that cause acute upper and lower respiratory tract infections and which are routinely tested for in most virus diagnostic laboratories are: influenza A virus (FLA); influenza B virus (FLB); respiratory syncytial virus (RSV); parainfluenza virus type 1 (PF1); parainfluenza virus type 2 (PF2); parainfluenza virus type 3 (PF3) and adenovirus (ADV). Additionally, human rhinoviruses (HRV) and coronavirus 229E (CoV-229E) are also linked to acute respiratory infection but less commonly included in laboratory reports; human metapneumovirus (hMPV) is not yet part of most United Kingdom virus laboratory test repertoires (personal feed-back from the United Kingdom Clinical Virology Network). As part of service development it was necessary to provide an alternative to virus culture for testing immunofluorescence negative respiratory specimens. Historically and indeed currently immunofluorescence [ 2 ] and virus culture [ 3 , 4 ] are the main methods used to diagnose acute respiratory virus infections. Culture is accepted as more sensitive than immunofluorescence but slower and therefore less useful for direct patient management decisions. Using a standard culture technique [ 4 ] for the culture of respiratory viruses our median reporting times for culture positive and culture negative specimens were 6 days (based on 407 specimens) and 7 days (based on 2159 specimens) respectively; virus identification by this technique required the use of monoclonal antibody staining of the cell monolayer in addition to observation for viral cytopathic effect. We therefore wished to develop a test capable of reporting on immunofluorescence negative specimens within a 24 hour period. Increasingly however, the sensitivity of nucleic acid amplification techniques for diagnosis has become recognised [ 5 - 10 ]. However widespread concerns about contamination issues [ 11 , 12 ] and perceived cost [ 13 ] have slowed their widespread adoption. An added problem for acute respiratory tract infections is the relatively large number of viruses that need to be accounted for, a problem which presents specific technical challenges. One such challenge is the different optimal annealing temperatures of the primer sets for each prospective virus target. The ABI PRISM 7000 real-time facility from Applied Biosystems addresses this by using bundled software to design primer/probe combinations that use a common amplification protocol. However this approach is compromised by the inability of software to allow for target heterogeneity. In addition it does not allow users to adopt clinically validated primer sets from the literature. To address these problems we adopted an alternative approach through the development of a generic touchdown amplification protocol. Touchdown protocols involve a pre-designed stepped reduction in the annealing temperature used for primer-to-template binding, which introduces a competitive advantage for specific base-pair priming over non-specific priming [ 14 ]. A detailed knowledge of the optimum annealing temperature is therefore not required. The study protocol was empirically constructed and proved robust when applied to a large range of respiratory viral and bacterial targets, without compromising individual test sensitivity. It was designed for use with in-house primer master-mixes that recognise 12 common respiratory viruses. Before deciding on the layout of the molecular strip, as described in the methods, we undertook a wide range of preliminary validation steps for each primer set. The complexity of the strip makes it impossible to fully evaluate using the classical approach of applying an individual gold standard to each virus type. Classically this approach works well where a single target is under investigation [ 15 ]. However although the strip is putatively designed to identify 12 viruses, the actual number of individual types targeted is over one hundred and sixty because of the inclusion of generic primer sets for HRV [ 16 ] and ADV [ 17 ] respectively. The classical approach is further compounded for viruses (a) that cannot be grown or grown easily; (b) for which commercial IF sera are not available; (c) for which specimen panels are not available. We therefore adopted a phased validation, culminating in the present study. Sensitivity was ascribed by undertaking copy number determination on cloned targets and these ranged form 6 × 10 3 copies per ml for human rhinovirus type 1b to 4.2 × 10 3 copies/ml for RSV-A. Specificity was ascribed through reproducibility, i.e. specimens which were repeatedly positive, following our standard clinical reporting algorithm [ 6 ], were regarded as true positives; a similar approach was recently described for hMPV [ 18 ]. In addition amplicon sequencing was used as an initial specificity check. The primers sets were tested on clinical respiratory specimens arising from a number of ethically approved studies. These included respiratory specimens from patients: (a) with chronic obstructive pulmonary disease; (b) with acute asthma; (c) on assisted ventilation in intensive care. They were also tested on respiratory specimens collected as part of an influenza spotter program as well as on laboratory specimens of known virus reactivity. To test the feasibility of its routine use we needed to clinically validate its performance in a routine setting on specimens tested in parallel with our standard immunofluorescence protocol for the diagnosis of acute virus respiratory infections. Although the routine immunofluoresence panel lacked capacity for the detection of rhinoviruses, human metapneumovirus and CoV-229E, these were included on the strip for clinical reasons during the period of the study. These findings and their implications are reported. Results Patients and specimens A total of 99 viruses were detected in 84/222 specimens from a total of 79/183 patients and 4/5 National External Quality Assurance Scheme (NEQAS) controls; immunofluorescence did not detect the parainfluenza virus type 2 virus in one of the NEQAS specimens. Viruses were detected in all of the specimen types processed. The molecular strip detected virus in: 16/36 (44.4%) broncho-alveolar lavages, 62/120 (51.6%) nasopharyngeal secretions, 11/35 (31.4%) sputa and 10/31 (32.2%) combined throat and nasal swabs. Immunofluorescence detected virus in: 6/36 (16.6%) broncho-alveolar lavages, 23/120 (19.1%) nasopharyngeal secretions, 1/35 (2.8%) sputa and 1/31 (3.2%) combined throat and nasal swabs. The median age of male and female patients where virus was detected was 3 y (range 2 weeks – 79 years) and 4 y (5 weeks – 81 years) respectively. Sixteen viruses were detected in 14/27 (51.8%) specimens, confirming a respiratory virus in 12 out of 24 (50%) patients investigated in general practice. Seventy-nine viruses were detected in 70/191 (36.6%) specimens, confirming a respiratory virus in 67 out of 159 (42.1%) patients investigated in hospital. Of the 16 viruses detected in specimens from the community, PCR detected all 16 in contrast to a single identification, influenza A (H3), by immunofluorescence. Nested PCR PCR identified one or more viruses in specimens from 84 of the 183 patients and the 4 NEQAS positive specimens, detecting a total of 99 viruses as shown in Table 1 . The viruses detected were: influenza A (H3) virus (17); influenza A (H1) virus (4); influenza B virus (2); human rhinovirus (39); adenovirus (22); parainfluenza virus type 2 (1); parainfluenza virus type 3 (10); respiratory syncytial virus type A (2); respiratory syncytial virus type B (2);. No parainfluenza virus type 1, coronavirus 229E or human metapneumovirus were detected. Dual infections were detected in 11/79 (13.9%) patients. The dual infections were: influenza A (H3) and adenovirus (4); influenza A (H3) virus and rhinovirus (2); influenza A (H1) and adenovirus (1); adenovirus and rhinovirus (3); respiratory syncytial virus type B and rhinovirus (1). Nine patients had more than one specimen taken on the same day in which a virus was detected in at least one specimen by PCR. For 5 of the patients the same virus was detected in each of the 2 specimens. The viruses identified were rhinovirus (3), adenovirus (1) and parainfluenza type 3 (1); the latter was also immunofluorescence positive. In 2 cases a rhinovirus was detected in only one of the specimens. As part of a separate rhinovirus validation protocol one of these specimens was subjected to retesting coupled with limited sequencing of the 5' non-coding region amplicon which confirmed the presence of a rhinovirus sequence. Additionally, premature twin boys admitted to intensive care on the same day with severe bronchiolitis, both had evidence of acute rhinovirus infection by PCR. Limited sequencing of the 5' non-coding region of these viruses as part of the rhinovirus validation protocol confirmed the presence of an identical sequence of rhinovirus in both specimens. Immunofluorescence Immunofluorescence identified a virus in specimens from 28 of the 183 patients and 3/4 NEQAS positive specimens (16.4%), detecting a total of 31 viruses as shown in Table 1 . The viruses detected were: influenza A virus (15); influenza B virus (1); parainfluenza virus type 3 (8); respiratory syncytial virus (4); adenovirus (3). No parainfluenza virus types 1 or 2 were detected including a NEQAS mock parainfluenza virus type 2 infection which was recorded as negative. No dual infections were detected. One patient had 2 specimens taken on the same day in which the same virus, parainfluenza type 3, was detected. Discussion Although touchdown PCR has been used successfully to help overcome some of the uncertainties associated with the thermal amplification of microbial nucleic acid targets [ 19 - 22 ], its use in this study has extended its role further and in so doing brought closer the goal of undertaking molecular diagnostics in a routine setting. Previously its main impact has been seen where multiplexing [ 23 , 24 ] or degenerate primers have been needed [ 25 - 27 ] and where the problems of choosing correct annealing temperatures are at their most difficult. In this study the large number of targets is the main problem encountered. Using an empirical approach a series of amplification steps linked to a stepped reduction in annealing temperature from 55°C to 46°C was constructed. This proved extremely resilient when used with a wide range of primer sets and included the apparent anomaly of putting adenovirus through an initial reverse transcription step to stream line all of the targets on to a single strip; we have previously reported this approach for testing group F adenovirus alongside norovirus, astrovirus and rotavirus [ 28 ]. The touchdown surprisingly out-performed individual amplification protocols optimised for annealing temperature and thus proved suitable for use on the diverse range of respiratory viruses addressed in the study. Where multiple viral targets are sought in clinical practice, we believe that it is only feasible to correlate the performance of the new assay in a head-to-head comparison with that already in routine use. Unfortunately for many clinical laboratories there is an elusion of testing for a wider range of viruses than is the case, by the inoculation of cell lines with a theoretical ability to grow the respective viruses. The annual reports of most clinical laboratories of one of the commonest human respiratory viruses, human rhinovirus, is an example of this; using the touchdown protocol we now report approximately 450 HRV infections per annum. The under reporting of adenovirus by standard methods [ 17 ] and the paucity of hMPV reporting, further underlines this assertion. The ability to simultaneously validate the performance of multiple molecular primer sets in a routine clinical setting is a major accomplishment of the current methodological approach. The results demonstrated that a range of primers from both the medical literature and from in-house development could be amplified with a single generic touchdown protocol. It therefore confirmed the feasibility of directly incorporating primer sets into a standard operating procedure without the necessity for the individual optimisation of cycling parameters. As such the study results should facilitate primer selection and formal critical evaluation as here described. As an example of this enhanced flexibility we have recently replaced the primer sets for influenza A H1 and H3 (with respective copy number sensitivities of 8 × 10 3 and 2 × 10 3 copies per ml) with a generic matrix set (copy number sensitivity of 1 × 10 3 copies per ml). The use of strips containing pre-dispensed mastermixes facilitates their use in a routine setting where laboratory personnel have only to thaw the strip and add the specimen extract. We make and aliquot for routine use a large range of multi-reaction mastermixes which are repeatedly subjected to freeze-thaw cycles as required on a daily basis. Provided the mixes are handled on ice, they remain extremely stable, over many months if so required. However the strip is designed for a single use only and thus only goes through a single freeze-thaw cycle. Mix stability is not a problem and the single positive control is used only to confirm that the touchdown amplification cycle has run successfully. Because the technique of using nested amplification followed by running agarose gel electrophoresis is relatively cumbersome, it was important to evaluate how the complete protocol, inclusive of report generation, would perform when introduced into a routine line-managed diagnostic setting. Over the 18 months of the study the technique fitted in well to the demands of routine service. Central to this was the use of pre-dispensed and quality checked primer master-mixes which allowed the molecular strip to be adapted for use in a routine laboratory. The study confirmed that a broad based molecular approach was feasible as an alternative to virus culture to support immunofluorescence in the diagnosis of respiratory viruses. The overall superior performance of the strip and the missed NEQAS specimen by immunofluorescence underlines the need for a more sensitive back-up for negative specimens. While nested protocols must be regarded as a pragmatic, interim solution until perfected single round systems are available, the format of the strip reduces the concern most attached to nested formats, i.e. false positive results. In our experience there is little evidence to support contamination arising from environmental sources and that the two major points of contamination in a nested system are (a) cross-contamination during manual extraction and (b) contamination of second round adjacent wells with product from a first round positive amplification. The use of the QIAGEN BioRobot for the extraction of all specimens reduced the former while the nature of the strip prevents the latter, since all the wells have separate mixes (Table 2 ). With both nested and non-nested assays the most critical requirements for reliable results are the use of well trained, appropriately skilled and knowledgeable staff, operating in a managed environment. As with any service, test performance must stand up to both external and internal quality assurance and in this regard we welcome the new respiratory quality control panel soon to be made available from Quality Control for Molecular Diagnostics (QCMD), Glasgow. The results obtained were very encouraging. Although the strip was constructed to detect a wider range of viruses than immunofluorescence, over the period of validation it almost doubled (59 versus 31) the number of viruses that could have been detected by immunofluorescence, including a positive NEQAS specimen which was missed by immunofluorescence. Of this group of viruses the detection of adenovirus showed the most dramatic increase, an observation we have also previously made in a separate study [ 17 ] and which we continue to see both in routine respiratory specimens and in a number of respiratory studies. Similar to HRV viruses we believe these common infections are underdiagnosed by the standard techniques of immunofluorescence and culture. They are the second commonest virus, after HRV, that we observe in mixed infections and it is self-evident that these additional infections are at a level below the detection thresholds of standard methods. Their clinical significance when detected at these lower copy numbers remains to be determined. As mentioned in the introduction a factor which often impacts negatively on a laboratory's decision to use molecular diagnostics is one of cost. It is worth considering that no matter which assay is chosen for use, it will attract the same overheads needed to provide the infrastructure of a laboratory set-up i.e. building, utilities, staff and equipment. In this regard there are no cheap tests and to use reagent costs as the sole factor in determining which assay to use is somewhat perverse. While the reagent costs of the strip are higher that commercial immunofluorescence reagents by a factor of 3, including extraction, this would undoubtedly narrow if immunofluorescence were capable of closing the pathogen gaps that currently exist e.g. HRV, hMPV. Currently using this approach, we have been able to replace both immunofluorescence and viral culture and this ability makes molecular diagnostics a more cost effective method for diagnosing viral infections. Taking into account the superior range, sensitivity, ability to quantify and speed of molecular techniques it is incredible how little they are used in routine laboratories. With the advent of SARS and the threat of avian influenza, this deficit is now beginning to disturb health care planners at the highest level. Because specimen sampling was not contiguous seasonal peaks were not detected, accounting for the small numbers of respiratory syncytial virus detected and the lack of detection of human metapneumoviruses, parainfluenza virus type 1 and coronavirus 229E; subsequent (unpublished) data from the routine use of the molecular strip support an important role for human metapneumovirus in acute respiratory infections and the sporadic nature of infections caused by parainfluenza type 1 and coronavirus 229E. Several interesting observations need highlighting. First, for immunofluorescence to perform reliably it was essential that a good nasopharyngeal specimen was available. The use of throat and/or nasal swabs with immunofluorescence alone is inappropriate. Second, immunofluorescence was very poor at detecting viruses from patients in the community, again almost certainly because of the universal use of swabs in that setting. Third, the rapid results of immunofluorescence were complemented by the touchdown protocol which can report definitive results within 24 hours, considerably faster than culture. Fourth, the molecular strip was better at detecting multiple infections. Even allowing for the inability of immunofluorescence to detect rhinoviruses, it should have detected the mixed adenovirus and influenza virus infections. Although immunofluorescence is capable of diagnosing dual infections, its routine use along with culture probably grossly underestimates their prevalence. The most plausible explanation is that the molecular technique detects infections where one of the viruses is below the detection threshold of immunofluorescence. These low level viruses are either just starting or more likely reaching the end of an infectious episode (latency is less likely) and this raises the previously unaddressed question of their role in viral respiratory pathogenesis. Fifth, the extent of rhinovirus infections was very significant. Their clinical significance ranged from acting as a definitive respiratory pathogen to a less certain role when acting as the most frequently detected co-pathogen in mixed infections. Conclusions In conclusion the use of the touchdown protocol with pre-dispensed and quality checked primer master-mixes was suitable for replacing virus culture for the diagnosis of respiratory viruses for immunofluorescence negative specimens. Immunofluorescence results were available after an average of 4–12 hours while molecular strip results were available within 24 hours, considerably faster than viral culture. The combined strip and touchdown protocol is a convenient and reliable method of testing for multiple viruses in a routine setting. Its generic nature makes it especially useful for introducing test repertoire modifications e.g. incorporating primers for the newly identified coronaviruses SARS-CoV and HCoV-NL63. Methods Patients and specimens A total of 222 specimens were included in the validation between January 2002 and June 2003, including 14 from an influenza surveillance scheme. The specimens were collected from 183 patients including: 103 male, median age 3.5 y (7 m – 84 y); 80 female patients, median age 9 y (7 m – 84 y); both male and female ages were skewed towards the lower age ranges, and 5 national external quality assurance scheme (NEQAS) specimens (4 positive, 1 negative). One hundred and fifty-nine patients were in hospital and 24 were in the community at the time of sampling. Specimens tested consisted of a wide range of specimens including: broncho-alveolar lavage (36), nasopharyngeal secretions (120), sputum (35) and combined throat and nasal swabs (31). Immunofluorescence Nasopharyngeal secretions, broncho-alveolar lavage and sputum specimens were received in dry sterile containers at ambient temperature. Upon receipt they were re-suspended in 2 ml of virus transport medium (VTM) consisting of phosphate buffered saline pH 7.1, bovine serum albumin 7.5 μg/ml, penicillin G sodium 1000 units/ml, streptomycin sulphate 1000 μg/ml and amphotericin B 2.5 μg/ml. Throat and nasal swabs were received in 2 ml of VTM and vortexed on arrival to release cells attached to the fibres of the swab. An aliquot of 410 μl was taken off for extraction after which the specimens were centrifuged at 2600 g for 5 min and the resulting cell deposits air-dried on glass multi-well slides and fixed in acetone prior to testing. Immunofluorescence was set up on the respiratory specimens using commercial reagents according to the manufacturer's instructions, and was able to detect: influenza A, influenza B, respiratory syncytial virus, adenovirus, parainfluenza type 1, parainfluenza type 2 and parainfluenza type 3 (Dako diagnostics, Ely, UK). Specimen extraction A volume of 200 μl of the respiratory specimen suspension was extracted on a QIAGEN BioRobot 9604 using the Blood and Body Fluid Vacuum Protocol of the QIAamp DNA Blood Kit (Qiagen Ltd., Crawley, England, U.K). This protocol allows the co-extraction of both RNA and DNA simultaneously. Nested PCR Simultaneous amplification of all targets was facilitated by using a standard 8 well multi-well PCR strip to which all mixes were pre-dispensed and stored frozen; this format is referred to in the paper as the "respiratory strip" because of the respiratory nature of the targets. The respiratory strip targeted the following 12 common respiratory viruses: influenza A (H3), influenza A (H1), influenza B, respiratory syncytial virus type A, respiratory syncytial virus type B, adenovirus, coronavirus 229E, parainfluenza virus type 1, parainfluenza virus type 2, parainfluenza virus type 3, human rhinovirus and human metapneumovirus. The final configuration of the single and multiplex primer mixes in the 8 well strip are shown in Table 2 . The primer sets used were taken mainly from published studies [ 16 , 29 - 31 ] but also included primer sets validated in-house after modification or de-novo design, including those for influenza A (H1), influenza A (H3) and the generic adenovirus primers [ 17 ]. The primers, gene targets and expected product sizes following amplification are shown in Table 3 . Each primer master-mix was made-up and titrated against a known positive control before being aliquoted and dispensed into its respective well of the 8-well microtube strip. The strips were stored frozen at -20°C until used. A positive control was also aliquoted and stored separately at -20°C until used. For the duration of the study the positive control was the cloned target of parainfluenza virus type 1; a negative control was not deemed necessary. First round volumes were made-up in Access RT-PCR buffer (Promega, Southampton, England, U.K) and in the final 10 μl volume contained the following reagent amounts: 1.5 mM MgSO 4 , 1 unit AMV reverse transcriptase, 1 unit Tfl DNA polymerase, 0.2 mM each deoxynucleoside triphosphate (dATP, dCTP, dGTP, dTTP) and 1 μM outer primers. Second round volumes were made-up in Taq Buffer B (Promega) and in the final 10 μl volume contained the following final amounts: 10 mM Tris-HCl (pH 9.0), 3.5 mM MgCl 2 , 50 mM KCl, 0.1% Triton X-100, 0.2 mM of each deoxynucleoside triphosphate (dATP, dCTP, dGTP, dTTP), 0.25 units of Taq DNA polymerase (Promega) and 0.2 μM inner primers. First round amplification was performed on 2 μl of extract added to 8 μl of first round primer master-mix per well. Second round amplification was performed on 0.2 μl of the first round reaction added to 9.8 μl of second round primer master-mix per well; a multi-channel pipette facilitated the transfer of the 8 volumes in one step. The positive control was run on the eighth well of each strip. The second round products were run on ethidium bromide stained 2% agarose gels and photographed. Specimens were reported positive when respectively the correct size bands and the positive control bands were present. Touchdown amplification protocol Amplification was carried out on a range of thermal cyclers including the Applied Biosystems GeneAmp 2400 and 9700 series and a DNA Engine Tetrad PTC 225 (MJ Research, USA). The first and second round amplification protocols consisted of 36 identical cycles with the exception that (a) a reverse transcription step of 48°C-10 min preceded the first round and (b) a hot-start preceded the second round by transferring the strip directly from ice to a thermal cycler held at 94°C. The touchdown protocol consisted of 6 steps as follows: (a) initial denaturation (94°C-2 min); (b) 3 cycles of denaturation (94°C-30 s), annealing (55°C-30 s) and extension (72°C-30 s); (c) 3 cycles of denaturation (94°C-30 s), annealing (52°C-30 s) and extension (72°C-30 s); (d) 20 cycles of denaturation (94°C-30 s), annealing (49°C-30 s) and extension (72°C-30 s); (e) 10 cycles of denaturation (94°C-30 s), annealing (46°C-30 s) and extension (72°C-30 s); (f) 72°C for 5 mins. Authors' contributions PVC: Touchdown and molecular strip design and manuscript preparation. GMO: Early application of touchdown cycling to respiratory samples. HJO'N: Study protocol and manuscript preparation. CMcC: Study protocol and manuscript preparation. DDEO: Routine application of touchdown protocol during study. FM: Routine application of touchdown protocol during study. SJM: Routine application of touchdown protocol during study. SAF: Primer-mastermix manufacture and quality control. DEW: Early application of touchdown cycling to respiratory samples and manuscript preparation. MF: Routine application of immunofluorescence protocol during study. JS: Design and validation of primers for human metapneumonavirus detection.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529439.xml
546203	A primary care, multi-disciplinary disease management program for opioid-treated patients with chronic non-cancer pain and a high burden of psychiatric comorbidity	Background Chronic non-cancer pain is a common problem that is often accompanied by psychiatric comorbidity and disability. The effectiveness of a multi-disciplinary pain management program was tested in a 3 month before and after trial. Methods Providers in an academic general medicine clinic referred patients with chronic non-cancer pain for participation in a program that combined the skills of internists, clinical pharmacists, and a psychiatrist. Patients were either receiving opioids or being considered for opioid therapy. The intervention consisted of structured clinical assessments, monthly follow-up, pain contracts, medication titration, and psychiatric consultation. Pain, mood, and function were assessed at baseline and 3 months using the Brief Pain Inventory (BPI), the Center for Epidemiological Studies-Depression Scale scale (CESD) and the Pain Disability Index (PDI). Patients were monitored for substance misuse. Results Eighty-five patients were enrolled. Mean age was 51 years, 60% were male, 78% were Caucasian, and 93% were receiving opioids. Baseline average pain was 6.5 on an 11 point scale. The average CESD score was 24.0, and the mean PDI score was 47.0. Sixty-three patients (73%) completed 3 month follow-up. Fifteen withdrew from the program after identification of substance misuse. Among those completing 3 month follow-up, the average pain score improved to 5.5 (p = 0.003). The mean PDI score improved to 39.3 (p < 0.001). Mean CESD score was reduced to 18.0 (p < 0.001), and the proportion of depressed patients fell from 79% to 54% (p = 0.003). Substance misuse was identified in 27 patients (32%). Conclusions A primary care disease management program improved pain, depression, and disability scores over three months in a cohort of opioid-treated patients with chronic non-cancer pain. Substance misuse and depression were common, and many patients who had substance misuse identified left the program when they were no longer prescribed opioids. Effective care of patients with chronic pain should include rigorous assessment and treatment of these comorbid disorders and intensive efforts to insure follow up.	Background Chronic, non-cancer pain, defined as pain of greater than 3 months duration, is a common, important health issue. The prevalence of chronic pain ranges from 20% to 60% [ 1 ]. The prevalance of low back pain is greater than 30%[ 2 ], and the prevalance of migraine is approximately 15%[ 3 , 4 ]. Pain disorders, including headache, back pain, arthritis and other musculoskeletal pain, are estimated to cost the United States economy $61 billion per year in lost productive time [ 5 ]. It is frequently asserted that chronic non-cancer pain is undertreated [ 6 - 8 ]. The optimal approach to treating chronic pain is controversial [ 9 - 13 ]. The realization that acute pain and cancer pain were often undertreated led to a liberalization of opioid use in these populations in the late 1980's and throughout the 1990's [ 6 , 14 - 16 ]. Subsequently, many advocates and practitioners called for an increase in the use of opioids for patients with chronic non-cancer pain as well [ 14 , 17 , 18 ]. Other experts assert that use of opioids for chronic pain should be limited [ 10 , 19 ]). used with caution [ 20 ], or eschewed altogether due to potential for substance misuse and lack of proven efficacy [ 13 ]. In large part, uncertainty about how to best manage patients with chronic pain stems from a lack of research in this area [ 21 ]. The limited available research comes from studies performed in specialty practice settings [ 11 , 22 - 25 ]; few studies have examined the care of chronic pain in primary care [ 26 , 27 ]. Unlike other common chronic diseases such as congestive heart failure or diabetes, there are few clinical trials to inform practice guidelines. Chronic pain management is complex. Like other chronic diseases, chronic pain is multi-faceted, and it is attended by its own set of comorbidities. It is complicated by substantial psychological and functional impairment in the forms of depression, disability, and loss of livelihood [ 28 - 30 ]. It is also costly. Patients with depression and back pain or migraine incur 3 to 4 times higher medical costs than patients with these pain conditions alone [ 31 ]. Generalists, even if well-versed in the biopsychosocial model of disease, often feel unprepared to manage chronic pain. They may lack the training in using certain pharmacological regimens, such as combining chronic opioids with psychiatric and pain modulating agents [ 27 , 32 ]. They may fear regulatory scrutiny and sanction when prescribing opioids, and they may be wary of fostering opioid dependence, misuse or addiction [ 33 , 34 ]. These barriers result in variations and inconsistencies in care that leave both patients and providers frustrated. We developed a disease management program to improve the management of chronic pain in opioid-treated patients attending an academic general medicine practice. Traditional models of office-based care focus on diagnonsis and acute management of medical problems in the context of a single provider-patient relationship. In contrast, disease management emphasizes: (1.) The use of multi-disciplinary teams providing integrated care; (2.) Evidence-based algorithms; (3.) Interval visits to monitor response to therapy; and (4.) Information systems that permit tracking of outcomes and adjustment of therapy [ 35 - 39 ]. We applied these principles in developing and implementing an intensive, structured, and coordinated program to improve the management of patient with chronic pain. We focused our program on patients treated with opioid medications and attempted to provide an environment that would allow the safe and effective use of these drugs. We sought to determine if this program could improve pain, functional, and psychiatric outcomes in a 3 month uncontrolled trial. Methods Development of the program In designing the program, we reviewed existing research, contacted experts, and conducted informal assessments of the main barriers to effective pain management in our practice [ 40 ]. Barriers existed at the provider and patient levels. These included part-time providers, frequent provider turnaround, physicians in training, and a socio-economically-disadvantaged, geographically dispersed patient population with multiple comorbidities. We applied lessons and systems from existing disease state management programs in diabetes, anticoagulation, and heart failure in our practice [ 41 ]. Patient recruitment Patients were eligible if they had pain of greater than 3 months duration and were either taking or considering opioid therapy. Attending and resident physicians were encouraged to refer patients if they were having difficulty managing their pain or if they suspected misuse of opioid medications. We publicized the program through educational conferences with residents and attendings and through informal communication within our practice. Baseline assessment After obtaining informed consent, a research assistant administered a comprehensive baseline assessment to gather socio-demographic data and review the medical history with an emphasis on the medical management of pain. Validated measures of pain, disability and psychological status were included. Using an 11 point scale, the Brief Pain Inventory (BPI) asked patients to rate their pain at the time of the interview and at its worst, least, and average over the past month [ 42 , 43 ]. The 7 item Pain Disability Index (PDI), a measure of pain related disability, asked patients to rate the degree of disability on a 10 point scale [ 44 - 47 ]. Higher scores indicate higher disability and the scale discriminates between high and low levels of disability. To assess depression, we used the Center for Epidemiological Studies-Depression Scale (CESD) [ 48 ]. This twenty item tool rates affective symptoms on a scale of 0 to 3. Intervention Patients were managed by a multi-disciplinary team consisting of the patient's primary care physician, a clinical pharmacist, a program assistant with training in health behavior, and a psychiatrist with sub-specialization in pain management who saw patients with the team one half-day per week. One nurse was dedicated to checking-in study patients and obtaining urine specimens. At entry, all patients signed a Medication Contract (Appendix A) [ 49 ] and provided a sample of urine for toxicological testing (UTS) [ 50 , 51 ]. The Medication Contract specified the conditions under which opioids would or would not be prescribed. The clinical pharmacist or psychiatrist modified or titrated a patient's pain medications in consultation with the primary care physician. During titration of medications, patients returned at one month intervals. Medical management adhered to published guidelines and expert opinion on the management of chronic, non-cancer pain [ 17 , 52 - 54 ]. The principles of management were: • Longer-acting opioids (long-acting morphine, fentanyl patches, methadone, sustained-release oxycodone) were initiated in patients who had been receiving short-acting agents that were only partially effective. • Short-acting, potent opioids (usually oxycodone preparations) were prescribed for breakthrough pain. • Longer-acting opioids were titrated at interval visits. • Less costly, generic medications (e.g. methadone and long-acting morphine) were preferred over proprietary products of equal or lesser efficacy [ 55 ]. • Tricyclic anti-depressants, gabapentin, and other agents were used adjunctively, especially for neuropathic pain. To address psychiatric comorbidity, patients with depression and other complex psychiatric conditions (e.g. psychotic depression and bipolar disorder with substance misuse) received psychiatric evaluation. Depression was diagnosed based on clinical interview and CESD score. In addition, primary care physicians could request psychiatric consultation on patients who had unaddressed psychiatric problems. The clinical pharmacist, psychiatrist and primary care physicians used CESD scores to guide treatment of patients scoring in the depressed range. The program, however, did not employ a strict protocol for depression treatment. As defined in our medication agreement with participants, we prospectively monitored substance misuse through clinical history, review of medications, communication with pharmacies and providers, and urine toxicological screening (UTS). Medications were documented in the electronic medical record and our program database. Discrepencies and inconsistencies were discussed with the patient's primary provider. We contacted a patient's pharmacy to verify procurement of medications, and, if substance misuse was suspected, we contacted additional pharmacies to ascertain whether or not a patient was receiving opioids from multiple sources. A UTS was obtained at each visit and was correlated with the patient's reported history of medication use. In collaboration with our institution's toxicologist, results of the UTS were verified using the appropriate confirmatory assays. For example, the presence or absence of "opiates" on the UTS was confirmed using gas chromatography. In addition, all positive results for amphetamines were confirmed with gas chromatography due to the possibility of assay interference from other medications [ 50 , 51 ]. We defined serious substance misuse prospectively as any of the following: 1. Cocaine or amphetamines detected on UTS; 2. Procurement of opioids from more than one provider on a regular basis ("doctor collecting"); 3. Diversion of opioids; 4. UTS negative on at least two occasions for prescribed opioids in the context of a reported history that the patient was taking the medication as prescribed (We considered repeatedly "negative" urines as an indicator of possible diversion.); 5. UTS positive on at least two occasions for opioids not prescribed by our practice (an inappropriate or inconsistent urine). A positive cannabinoid finding on UTS was not defined as serious substance misuse for the purposes of our study, but we counseled patients to refrain from marijuana use. Patients were advised at entry into our program (and in the Medication Contract) that serious violations of the contract would result in discontinuation of opioids. Past instances of serious misuse were not subject to sanction. We constituted a formal practice-wide committee to evaluate and respond to suspected misuse. It consisted of the practice director, two attending physicians, a clinical pharmacist, two resident physicians, and a nurse. The committee deliberated through secure email and considered the violations defined above. Patients committing serious substance misuse were offered referral to substance abuse experts at our institution. In most cases, opioid therapy would be reconsidered in 6 months if the patient participated in substance abuse counseling (The practice policy addressing substance misuse is included as Appendix B.). Reassessment At 3 month follow-up, a research assistant reassessed each patient's clinical status. Pain, disability, and mood scores were re-measured using the instruments previously described. The research assistant was not blinded to study participation status. Analysis Descriptive statistics are reported as means and percents. Paired t-tests were used to compare changes in pain, disability and depression scores from baseline to 3 month follow-up. McNemar's test was used to measure differences in proportions of patients receiving treatment for depression at 0 and 3 months. We also compared changes in pain scores based on changes in opioid dose. All analyses were performed using Stata 7.0 (College Station, TX). The research protocol was approved by the University of North Carolina School of Medicine Committee on the Protection of the Rights of Human Subjects. The funding sources had no role in the collection or interpretation of the results. Results Between December 2002 and May 2003, 85 patients agreed to participate in the study. Table 1 presents the baseline demographic characteristics of the study participants. All patients completed baseline assessment and 63 (73%) completed the 3 month assessment. Of the 22 patients who did not complete 3 month assessment, 15 did not return after a serious violation of the medication contract led to the discontinuation of opioids, 4 patients were lost to follow up, and 3 changed their venue of primary care. There were no important differences in baseline demographic, pain, depression, or disability scores between completers (n = 63) and all non-completers (n == 22), although some differences emerged among non-completers who committed substance misuse (n = 15) (Table 2 .). Table 1 Univariate Analysis N = 85 Mean age, y (± SD) 51 (9.6) Range 27–76 Male, % 60 White Race, % 78 Marital Status, % Married 49 Stable relationship 7 Unmarried 44 Disabled, % 65 Education, % Not high school graduate 38 HS graduate 28 Some college 34 Income <$20,000/yr, % 83 Medicare or Medicaid, % 58 Uninsured,% 29 History of Smoking, % 87 H/O Alcohol Use, % 75 H/O Substance Use, % 44 H/O Depression, % 51 Table 2 Characteristics of Study Completers and Non-Completers Characteristic Completers (N = 63) Non-Completers (N = 22) P-Value Non-Completers with Substance Misuse (N = 15) P-Value Age, y 51 49 0.422 48 0.215 % Male 62 55 0.544 67 0.732 % White 81 68 0.216 60 0.083 % High School Graduate 62 62 0.975 64 0.890 % Disabled 66 63 0.815 54 0.405 % Uninsured 29 32 0.774 33 0.716 % Substance Use 40 55 0.226 67 0.059 CESD 24 27 0.416 31 0.050 PDI 47 41 0.141 46 0.810 Pain Scores Worst in last month 9.2 9.3 0.727 9.2 0.946 Least in last month 4.6 4.4 0.768 4.7 0.884 Average in last month 6.5 6.5 0.925 6.6 0.861 Current pain 6.8 7.3 0.361 7.5 0.279 Patient characteristics The average age of patients was 51 years, 60% were male, and 78% were white, most (83%) had an income less than $20,000 per year and 65% were disabled. Forty-four percent had a history of illicit substance use (e.g. marijuana, cocaine, amphetamines). All patients had pain of at least 3 months duration and 90% had pain for greater than 1 year. At baseline 93% were receiving opioids. At 3 month follow up, 97% were receiving opioids. Table 3 presents the principal pain types. The lumbar spine was the most frequently involved primary site. Overall, axial spine pain accounted for 49% of the primary pain reported by patients. Myofascial pain and polyarticular arthritis were also frequently represented. Patients in the "Other" category commonly had mixed etiologies of pain, often attributable to previous trauma or surgery. One chronic headache patient is included in this category. Table 3 Primary Pain Type (N = 85) Number (%) Spine 42 (49) Lumbar 30 (35) Cervical 7 (8) Thoracic 5 (6) Diffuse (Fibromyalgia, Chronic fatigue syndrome) 13 (15) Polyarthritis 8 (9) Knee 5 (6) Abdomen 4 (5) Diffuse neuropathic 4 (5) Elbow & Hip 2 (2) Other 7 (8) Effect of the intervention Table 4 presents the effect of the intervention on pain, functional status and depression. Baseline results reveal high pain scores. The worst pain was 9.2, the least pain was 4.6, the average pain was 6.5, and current pain was 6.8. The average PDI score, 47.0, suggested substantial disability. There was a high prevalence of depression. The mean CESD score, 24.0, falls in the "severely depressed" category of the scale. Table 4 Pre and Post Intervention (N = 63) Pre Post Improvement, % P-Value* Pain at worst in the last month & 9.2 8.1 12 <0.001 Pain at least during the last month 4.6 3.9 15 0.038 Pain on average during the last month 6.5 5.5 15 0.003 Pain right now 6.8 5.8 15 0.014 Pain Disability Index 47.0 39.3 16 <0.001 CESD 24.0 18.0 25 <0.001 % CESD in depression range: Conventional cutoffs £ 79.4 54.0 32 0.003 Chronic pain cuttoffs ¶ 38.1 23.8 37 0.049 % Depression medication 44.4 52.4 15 0.059 * Paired t-test except where indicated McNemar's test & Score 1–3 is mild pain; 4–6, moderate pain; 7–10, severe pain £ Score of ≥ 15 ¶ Score of ≥ 27 At 3 month follow up, BPI pain scores improved by 12% to 15%, and all reductions were statistically significant. The mean depression score improved from 24.0 to 18.0 (p < 0.001), and the proportion of patients scoring in the depression range decreased from 79% to 54% (p = 0.003). We did not correct for multiple comparisons because of the exploratory nature of our analyses. Some authors have demonstrated that the conventional cutoffs for the CESD (depression ≥ 15; severe depression ≥ 22) may lack specificity in patients with chronic pain and have proposed a CESD threshold of 27 for diagnosing depression in this population [ 56 ]. Using this threshold, the proportion of patients scoring in the depressed range decreased from 38% to 24% (p = 0.049). Pharmacologically, depression was undertreated at baseline. The proportion of depressed patients receiving anti-depressants increased from 44% at baseline to 52% at 3 months (p= 0.059). Relationship between pain and opioid dosing The mean daily opioid dose in milligram equivalents of morphine increased from 72 mg per day to 91 mg per day (A milligram morphine equivalent approximates a milligram of oxycodone.). Forty-eight percent of patients had their opioid dose increased over 3 months. In these patients, the mean opioid equivalent increased from 53 mg to 105 mg per day. No clear relationship emerged between opioid dosing and improvements in pain, disability, and depression scores, after adjusting for baseline pain, disability and depression (Table 5 .). Table 5 Effect of Opioid Increase on Pain (N = 63) Opioids Increased P-Value Yes (n = 30) No (n = 33) Δ Pain at worst in the last month 1.40 0.99 0.37* Δ Pain at least during the last month 0.80 0.52 0.66* Δ Pain on average during the last month 0.96 0.94 0.93* Δ Pain right now 1.14 0.87 0.70* Δ Pain Disability Index 8.34 7.03 0.63 ¶ Δ CESD 5.21 6.71 0.74 * Adjusted for baseline pain ¶Adjusted for baseline PDI Adjusted for baseline CESD Substance misuse Twenty-seven patients (32%) committed some form of serious substance misuse (Table 6 .). Although we confirmed only one instance of diversion, we suspect that patients with repeatedly negative UTS's or inconsistent UTS's may have been diverting their medications. Substance misusers accounted for the preponderance of subjects who dropped out of the study. Table 2 compares selected baseline characteristics between substance misusers who did not complete three months and subjects who completed the trial. Although the numbers are small, there is a trend toward greater representation of non-white race, history of illicit substance use, worse depression, and higher pain scores at baseline assessment among substance misusers who did not complete the trial. Table 6 Substance Misuse (N = 27) Misuse Number (%) Stimulants on UTS 13 (15) Cocaine 11 (14) Amphetamines 2 (2) Diversion 1 (1) Doctor collecting 3(3) Inappropriate/Inconsistent UTS 2 (2) Negative ("Clean") Urines 7 (8) Prescription adulteration 1 (1) Discussion We found that a multi-disciplinary, primary care-based, disease management program can improve pain, depression and disability scores in opioid-treated patients with chronic pain in a 3 month uncontrolled trial. The improvements across all outcomes support an improved quality of life resulting from the intervention. These improvements were obtained using an approach that balanced the potential benefits and adverse effects of opioids. We hyothesize that these improvements resulted from the combined effects of systematizing pain management and treating depression. Improvements appear independent of opioid dosing. The clinical significance of the 12% to 15% improvement in pain scores is unclear. Uncontrolled trials in specialty pain clinics have reported a 20% to 25% reduction in pain scores [ 57 ]. Some research suggests that a 30% decrease in pain scores (i.e. about 2 points) represents clinically significant relief of pain [ 58 ], but the issue of how to interpret pain scales clinically is not resolved. The improvement in depression scores was clinically important and may reflect combined effects of intensification of pharmacological management for depression and pain and the systematization of care. Although the reciprocal relationship between pain and depression has been established in previous studies, the effect that treating one condition has on the other has not been well-assessed [ 29 ]. One recent study demonstrated that the presence of severe pain predicted a poor response to antidepressant therapy, and thus it is plausible that intensifying pain management would have a beneficial effect on depression [ 59 ]. Clearly, the improvements in depression scores seen in this study cannot be attributed to increasing anti-depression pharmacological therapy alone because the proportion of patients treated with anti-depressants increased from 44% to 52% only. We have since added a structured treatment algorithm to increase the use of anti-depressants. It is important to note that there was a statistically signficant trend toward greater depression among substance misusers who did not complete the trial; thus, it is possible that the trial overestimates the effect of multi-disciplinary management on depression outcomes. To our knowledge, this is the first study to prospectively examine the effects of multi-discplinary pain management on the outcomes of pain, disability, depression, and substance misuse in an academic primary care practice caring for a wide range of patients. Previous studies conducted in a military clinic[ 60 ] and a health maintenance organization [ 26 ] demonstrated improved pain and functional scores with systematic, multidisciplinary intervention. The military trial was uncontrolled and enrolled referred patients into a specialty clinic. The HMO trial was conducted in a primary care setting. It was controlled, and did evaluate pain, function and mood outcomes. Neither study systematically monitored substance misuse. We documented a high prevalence of substance misuse (32%). We did not assess addiction per se. The prevalence of substance misuse and addiction in patients receiving chronic opioids is unclear and depends on the populaton under study. Some authors have asserted that addiction and substance misuse are uncommon consequences of opioid use for pain. One widely cited reference estimated the prevalence of addiction at approximately 4 in 10,000 treated patients [ 61 ]. Others have reported prevalences of addiction ranging from 3% to 17% [ 62 , 63 ]. A recent retrospective study in a primary care setting documented a high prevalence of opioid misuse: 24% in a resident physician clinic and 31% in a Veterans Administration outpatient clinic[ 64 ], but the criteria used to determine the prevalence of opioid misuse were limited by chart review. Some behaviors defined as indicators of opioid misuse (e.g. lost or stolen medications, requests for early refills) could be construed as indicators of inadequately treated pain (i.e. "pseudoaddiction"[ 65 ]) and not substance misuse or addiction. Opioid misuse not only complicates the management of pain in the individual patient, but has negative societal consequences as well, especially when opioids are diverted from their intended use [ 66 - 68 ]. Several states have documented increases in unintentional deaths from opioids, especially diverted methadone [ 69 - 71 ]. National surveys demonstrate dramatic increases in the non-medical use of OxyContin ® and other prescription drugs among teens and young adults [ 72 , 73 ]. The trauma literature has documented recent increases in opioid use among patients admitted to trauma centers [ 74 ]. In response, there have been state and national initiatives to reduce prescription drug misuse [ 75 , 76 ]. Though high rates of substance misuse are a source of concern, our program may serve as an example for how care can be organized to reduce misuse without eschewing the benefits of opioid medications. Although the prevalence of substance misuse in our study population is higher than reported in clinical trials of opioids, these trials have occurred in specialty settings with selected populations. They excluded patients with a history of substance misuse, or have not systematically monitored patients for substance misuse [ 11 , 12 , 23 , 66 , 77 ]. In addition, they commonly excluded patients with psychiatric illness (including depression) which is a strong predictor of substance misuse [ 23 , 66 , 78 - 80 ]. Patients in our program had a high baseline prevalence of depression, previous substance and alcohol abuse, and other psychiatric disorders (Table 1 .). The strong relationship between mental illness and substance abuse disorders is well known and thus the high prevalence of substance misuse is not entirely unexpected [ 81 ]. Previous studies of mental illness have documented a high coexisting prevalence of substance and alcohol misuse: 32% with unipolar depression; 61% with bipolar depression; 47% with schizophrenia; 84% with personality disorders: and 24% with anxiety disorders [ 78 , 82 ]. We specifically sought out patients whose pain management was difficult for providers or in whom substance misuse was suspected. Many had established or suspected psychiatric diagnoses. How to identify chronic pain patients at risk for drug misuse and to treat their pain remains a challenge [ 83 - 89 ]. The pattern of substance misuse in our population often suggested polysubstance abuse; this places patients at especially high risk of morbidity and mortality [ 90 ]. Although patients committing substance misuse were offered substance abuse treatment referral, only two followed through and most did not return to our program. The option of pain management without the use of opioid analgesics was offered to all patients who committed substance misuse. The difficulty in obtaining mental health and substance abuse treatment services is a pressing public health issue and a topic of national debate in the United States [ 81 ]. In our sample there was a clear trend toward increased comorbid depression among substance misusers who did not complete three month follow up (Table 2 .). Despite the availability of on-site psychiatric consultation, our program was not successful in retaining and managing a challenging subset of patients with substance misuse and depression. We are aware that some of our substance misusing patients migrated to other practices in order to obtain opioids and other controlled substances. Our program implemented policies to prevent migration of patients within our practice (Appendix B.) and the University of North Carolina Health Care System. The cornerstone of these policies was meticulous documentation in an electronic medical record that is accessible to all physicians at our medical center and to hospitals and physicians affiliated with our system in the surrounding communities. In general, though, we have no direct control over migration that occurs outside of our practice and our health care system. In order to curtail migration and "doctor shopping," some states have implemented centralized monitoring systems for opioids and other controlled substances. North Carolina is currently exploring the feasibility of such a system. A description of the operational state monitoring programs is available online through the United States Drug Enforcement Agency Diversion Control Program website [ 91 ]. We believe that our results may be generalizable to other academic primary care practices that serve diverse patient populations with a high burden of medical and psychiatric comorbidity. The etiologies and sites of pain were similar to those reported in population-based epidemiological surveys and clinical trials in primary care, except that headache was under-represented in our population [ 1 , 12 , 92 ]. The results replicate epidemiological research that demonstrates a strong interaction between pain and the psychiatric comorbidities of depression and substance misuse [ 87 , 88 ]. Our study may be more applicable to the general medical setting than previous trials examining the effects of opioids on chronic pain because we did not exclude patients with serious psychiatric comorbidity or those suspected of substance abuse. To be effective, pain management should encompass more than pharmacological management directed at pain scores; it should address a variety of behavioral and psychosocial factors that contribute to suffering [ 86 , 93 ]. Our study has several limitations. It was uncontrolled and of relatively short duration. The research assistants were not blinded to pre- and post-treatment assessments. The improvements could reflect secular trends, although the chronic nature of our patients' symptoms and disability makes this less likely, and improvements were noted across all of the pre-specified outcomes. Not all patients receiving chronic opioids in our practice were referred. Thus, we may have over-estimated the prevalence of substance misuse because providers were more likely to refer "problem" patients in whom they suspected opioid misuse. Another limitation of our study is its individualized nature. We did not adhere to strict algorithms for diagnosis and treatment and did not test a single intervention. The evidence-base for managing chronic pain in the general medicine setting is limited and the multi-modal nature of our intervention was by necessity empirical and exploratory. As such, we decided to allow more latitude and individualization in treatment choice. We have used our experience and the data collected to develop more robust algorithms to guide the management of pain and depression and to make psychiatric referral when appropriate. As a corollary to our multi-modal approach, it is difficult to ascertain if the improvements derived from pain medications, intensification of depression therapy, or simply participation in an organized program. Improvements and changes in behavior that occur as a result of becoming a target of special interest in a program are often referred to as a Hawthorne effect [ 94 , 95 ]. Conclusions In a 3 month trial conducted in an academic primary care setting, a systematic, multi-disciplinary approach to chronic pain management that included the use of opioids and tools to prevent misuse was effective in improving pain, depression, and function scores. Future efforts will be directed at examining their durability and promoting their sustainability. A randomized control trial would determine whether these are real effects or represent a secular trend. Comorbid depression and substance misuse were common. Efforts will also be made to further characterize the interaction of these and other comorbid psychiatric conditions with chronic pain. Chronic pain patients with substance abuse are a challenging subset of patients who could benefit from new research and initiatives to mitigate the risk of abuse while ameliorating pain control. Competing interests The author(s) declare that they have no competing interests. Authors' contributions PC developed the study designand intervention, performed the statistical analyses, and drafted the manuscript.TI developed the study designand intervention, administered surveys and directed pharmacologic management, assisted in editing and revising themanuscript.KF participated in developing the study design, administering surveys, data and projectmanagement, and editing and revising the manuscript. SP participated in study design, developing pharmacologic protocols, performing psychiatric evaluations, and editing and revising the manuscript.TM participated in developing the study design and editing and revising the manuscript.JP assisted in administering surveys, data management and collection, and editing and revising the manuscript.RM participated in the development of the study design, data management, editing and revising the manuscript.MB participated in the development of the study design, editing and revising the manuscript. DD provided statistical analytical support and assisted in the drafting, editing, and revising of the document.MP developed the study design, supervised the overall conduction of the study, participated in data analysis, and assisted in the drafting, editing and revising of the manuscript. Appendix A: Medication Contract Patient Name ____________________________ Diagnosis ______________________________ Physician Name __________________________ Telephone Number _______________________ I agree to abide by the following guidelines for managing my prescription for opiate pain medications: 1. I will only request and receive opiate (narcotic) pain medications from Dr. _________ or from his/her designee in the Internal Medicine Clinic Pain Service. I agree to inform any other physicians participating in my care of this agreement. If another physician wishes to suggest changes in pain management, they can contact Dr. ___________ during regular business hours, but no changes will be made without such contact. 2. Dr. ____________ and I have agreed that I will receive the following: medicine ______________, directions __________ quantity _____, per ___ days, medicine ______________, directions __________ quantity _____, per ___ days, medicine ______________, directions __________ quantity _____, per ___ days. I will not request refills prior to this date. I understand that if my medicines are lost or stolen, they will not be refilled prior to the next refill date. If I use up my supply of medication before the date of the next refill, I understand that my doctor will not provide extra medication. I further understand that I may suffer symptoms of withdrawal. I will inform my doctor in a timely manner if I miss taking a dose of my medication, have an increased need for the pain medication, or have difficulty taking the medication as prescribed. If I find that the current dose of pain medication is no longer adequate, I will discuss this situation with my doctor at a scheduled visit. 3. I agree to use _________________________________________________Pharmacy, located at _____________________________________________________________, telephone ________________________, for filling prescriptions for all of my pain medicine. 4. I will bring all unused pain medicine to every office visit, including all current prescription vials. 5. While this contract is in effect I will not abuse alcohol or other illicit drugs. As a part of this program, urine drug screening will occur at enrollment and may be required at future visits. 6. I will not sell or share opiate medications. 7. If I violate the terms of this contract, I understand that my doctor and other doctors in the Internal Medicine Clinic will no longer prescribe opiate medications for me. If this occurs, I understand that I may receive care elsewhere or continue with my current doctor and not receive opiate medicines. If I change doctors, I agree to allow my current physician to contact my new physician to transfer medical information including information about chronic pain treatment. 8. I understand that my doctor may verify whether or not I have a history of criminal drug convictions. Patient Signature ________________________________________________ (print name) ________________________________________________ Physician signature ________________________________________________ Date ________________________________________________ Appendix B UNC General Internal Medicine Practice Pain Review Committee: Policy on Serious Opioid or Controlled Substance Misuse & Misconduct Definition The Pain Review Committee defines serious misuse or misconduct with regard to opioid medications and other controlled substances as any of the following: 1. Forgery or alteration of prescriptions. 2. Use of cocaine or other stimulant medications (e.g. amphetamines) concurrently with prescribed opioids and their detection on urine toxicological testing. Stimulant abuse is strongly indicative of polysubstance abuse. 3. Diversion of opioids or controlled substances. 4. Doctor collecting or shopping: Procuring controlled substances from more than one provider and misrepresenting the fact. This is a felony in North Carolina. 5. Negative or "clean" urines: The absence of prescribed opioids from urine toxicological testing on at least two occasions in the context of a history that the patient is taking the medication as directed. This finding suggests diversion and/or substituting a separate urine sample. 6. Inappropriate or inconsistent urines: The presence on urine toxicological testing of opioids or other controlled substances (excluding cannabinoids) not prescribed by our clinic or pain program on at least two occasions without a reasonable explanation. This finding suggests polysubstance abuse, doctor collecting, or drug bartering (a form of diversion). Procedure Serious misuse or misconduct is a special category of misuse. It results in the immediate discontinuation of opioids in the Internal Medicine Clinic. The Pain Review Committee will address instances of serious misuse on an expedited basis. Individual cases will not require the review of the entire committee. The following procedure applies: • The specific violation will be documented in the electronic medical record. • Instances of serious misuse discovered by the pain management team will be discussed with the patient's primary care provider (PCP). • The provider who discovers the violation will report it to the Clinic Director, Dr. Thomas Miller, or designee on the Pain Review Committee. (The designee will be either Dr. Paul Chelminski or Dr. Timothy Ives.) The designee will inform Dr. Miller of the violation and make a recommendation to Dr. Thomas Miller and the entire Pain Review Committee for the immediate discontinuation of opioids. • Dr. Miller will make the final disposition on the recommendation. • Committee members will receive communication about recommendations and disposition through email. Committee members may recommend alternative sanctions. • The PCP will be informed of the disposition. • The patient will receive verbal and written notice of disposition. Sanctions A. Serious misuse or misconduct will lead to one of two possible sanctions: 1. Forgery or alteration of prescriptions and diversion will result in immediate and permanent discontinuation of opioids and/or other controlled substances. The clinic director will decide whether instances of forgery or diversion also merit dismissal from the clinic. 2. Stimulant use, doctor collecting/shopping, negative urines, or inappropriate urines will result in immediate discontinuation of opioids and/or other controlled substances with possible re-evaluation in six months for a first violation. The Committee will stipulate substance abuse counseling as a condition for re-evaluation. B. Two serious violations of the medication contract will result in permanent discontinuation of controlled substances. Provider Issues 1. The Pain Review Committee cannot compel attending physicians to cease prescribing opioids or other controlled substances; however, providers who continue to prescribe these medications must understand that this practice may jeopardize their DEA license and/or expose them to regulatory and even criminal investigation. If the attending continues to prescribe opioids after a recommendation of discontinuation by the Committee, the patient is not eligible for re-enrolment in the General Internal Medicine Pain Program after six months. 2. The responsibility for stewardship and teaching of resident physicians requires special oversight of residents' patients who receive controlled substances. The residency program has an obligation to promote appropriate clinical practice and protect residents from practices that may jeopardize their professional status. In addition, residents prescribe scheduled substances under the authority of the hospital's DEA number, and the inappropriate prescription of opioids may expose the hospital to sanction. If the committee recommends discontinuation of opioids for the patient of a resident, the committee will instruct the resident that he or she can no longer prescribe scheduled medications for this patient. Likewise, other residents in the practice are not authorized to prescribe opioids to patients of residents or attendings who have had opioids discontinued. Communication of Decisions 1. The patient should be informed of discontinuation verbally. Usually, this responsibility will fall to the PCP, but in certain instances it may fall to the person discovering the violation (e.g. the pharmacist who sees the patient in clinic for follow up in the pain program). 2. Dr. Miller will send the patient a registered letter. 3. The PCP will be copied on the letter. 4. A copy of the letter will be entered into the permanent electronic medical record as a Phone Message. 5. The patient's problem list on the electronic medical record will contain the entry "VIOLATION OF MEDICATION CONTRACT" and will be annotated "PATIENT VIOLATED THE MEDICATION CONTRACT SIGNED WITH GENERAL MEDICINE, AND WAS DISMISSED FROM NARCOTICS – SEE [date] CIS NOTE." Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC546203.xml
551541	An ontology for cell types	An ontology for cell types that covers the prokaryotic, fungal, animal and plant worlds is described. It includes over 680 cell types. These cell types are classified under several generic categories and are organized as a directed acyclic graph.	Background One of the most challenging problems now facing the model organism databases is the formal description of phenotypic data. While some databases, for example those for mouse ( Mus musculus ) [ 1 ], corn ( Zea mays ) [ 2 ] and fruit fly ( Drosophila melanogaster ) [ 3 ], include a rich heritage of data describing the phenotypes of mutants, and some progress is being made to bring these data into a well structured computable representation [ 3 - 5 ], the annotation of these phenotypes is hampered by a lack of structured information describing a variety of other biological objects, including cell types. A structured vocabulary of cell types is also required by databases for the description of other biological objects, such as gene-expression data. In addition, using the same concepts for the description of these data in all of these databases would facilitate interoperability among them. To address these needs, we have developed an ontology that describes the cell types of the major model organisms, both animal and plant. Its use will allow a biologist to query a single database with such questions as: list all of the cell types in mouse that express the Notch gene and all of the cell types in Drosophila and Caenorhabditis elegans that express the closest homolog of this gene; list all of the genes in mouse, rat, human and zebrafish that are expressed in the cell type Schwann_cell; CL:0000218; list all of the genes in D. melanogaster and C. elegans that have a mutant phenotype in the cell types that develop from the cell type myoblast; CL:0000056. The use of the cell ontology will thereby promote the de facto integration of data from diverse databases. Since the development of the Gene Ontology (GO) for the annotation of attributes of gene products [ 6 ], many ontologies have been developed in the model organism informatics community. Several of these are available, in a choice of common formats, from the Open Biological Ontologies (OBO) site [ 7 ]. They include comprehensive developmental and anatomical ontologies for many model organisms (for example, mouse, Drosophila , Arabidopsis thaliana and C. elegans ), and ontologies for mouse pathology and human disease. There are several other ontologies that include cell types such as Systematized Nomenclature of Medicine (SNOMED) [ 8 ], the Foundational Model of Anatomy (FMA) [ 9 ], the anatomy ontologies used in model organism databases at the OBO site [ 7 ], vocabularies used by the resources that hold cell lines such as the American Type Cell Collection (ATCC) or the European Collection of Cell Cultures (ECACC) [ 10 , 11 ], and others [ 12 , 13 ]. Our approach for handling cell types differs from that adopted by these resources. First, SNOMED, FMA and the species-specific anatomy ontologies explicitly assume that the cell types they include are associated with one particular organism. Their identifiers cannot therefore be used to annotate cell types from other organisms, even if these cell types are essentially identical to those in the organism-specific ontologies. Second, these resources, together with those that hold cell lines (for example, ECACC and ATCC), tend to define cell types as constituents of tissues rather than provide phenotypic information about their attributes - the knowledge that they encapsulate is severely limited. Third, some ontologies do not have publicly available identifiers for each term; hence they cannot be used for general annotation [ 10 , 11 ]. The Plant Ontology [ 14 ] provides a cell type node that shares some of the organizing principles of our cell ontology, but it is limited to those cell types found in plants. For all these reasons, we set out to produce an organism-independent ontology of cell types based on their properties (such as functional, histological and lineage classes) and report here the availability on the Open Biological Ontologies site [ 7 ] of this ontology, which incorporates the cell types possessed by a broad range of phyla and is defined by a rich set of criteria. Results The ontology The first design decision was whether we should attempt to integrate cell types from all phyla within a single ontology or build independent ontologies for different taxonomic groups. The former has the great advantage of facilitating de facto integration of data from diverse databases, as described above. This approach does, however, pose conceptual problems: for example, are a mammalian 'muscle_cell' and a nematode 'muscle_cell' homologous? In this particular example we have little doubt that the answer is 'yes'; both of these cell types are evolutionary descendants of the first metazoan's 'muscle_cell'. In other cases, however, matters are not quite as straightforward, a plant 'hair_cell', a 'hair_cell' of the mammalian cochlea and an insect 'hair_cell' are probably not homologous, despite some similarities in their functions and genes expressed within them [ 15 ]. Despite these problems in building an 'integrated' cell-type ontology, the advantages, were we to succeed, outweigh them, and we have therefore taken this approach to develop a single ontology that integrates cell types from different phyla. The ontology consists of concepts or terms (nodes) that are linked by two types of relationships (edges). This means that the ontology appears as a complex hierarchy (technically known as a directed acyclic graph, or DAG) where a given term (or concept) may not only have several children, but also several parents. The parent and child terms are connected to each other by is_a and develops_from relationships. The former is a subsumption relationship, in which the child term is a more restrictive concept than its parent (thus chondrocyte is_a mesenchyme_cell). The latter is used to code developmental lineage relationships between concepts, for example that a hepatocyte develops_from a mesenchymal_cell. The is_a relationship implies inheritance, so that any properties of the parent concept are inherited by its children; the develops_from concept carries no inheritance implications. The rules for building the ontology are the same as those defined by the GO Consortium. That is, each concept in the Cell Ontology has an identifier with the syntax CL:nnnnnnn, where nnnnnnn is a unique integer, and CL identifies the Cell Ontology, (concepts should always be cited with their full identifier when being used in the context of a database). In addition, if there are precisely equivalent terms in other databases, for example in the Fungal Anatomy [ 16 ], Arabidopsis [ 17 ], Plant Ontology [ 14 ] or FlyBase databases [ 3 ], then the unique identifiers from these databases are included in the Cell Ontology. Most concepts in the Cell Ontology are provided with free-text definitions and may have one or more synonyms. Within the context of this ontology, synonyms are precise; a concept and its synonym can be exchanged without changing the concept's meaning. We use the same stratagem as does the GO when we have concepts that are lexically identical but have different meanings in different communities [ 18 ]. Thus, it is far from obvious that vertebrate and invertebrate pigment cells are homologous and these concepts are therefore described as pigment_cell_(sensu_Vertebrata) and pigment_cell_(sensu_Nematoda_and_Protostoma, respectively. The two top-level nodes of the Cell Ontology are cell_in_vivo and experimentally_modified_cell. The former includes cell types that occur in nature, the latter those that are experimentally derived, including cell lines and such constructs as protoplasts. Experimentally derived cells are under-represented in the current version of the ontology. Naturally occurring cells are classified both by organism-independent categories and by organism (animal cells, plant cells, prokaryotic cells). The organism-independent classification of cells follows several different criteria that include: 'function' (for example, electrically_excitable_cell, secretory_cell, photosynthetic_cell), histology (for example, epthelial_cell, mesenchyme_cell), lineage (for example, ectodermal_cell, endodermal_cell) and ploidy (for example, haploid_cell, polyploid_cell). The present version of the Cell Ontology has an average 'depth' of about 10 nodes. The richness of the ontology can be illustrated by example (Figure 1 ). Kupffer cells are specialized vertebrate macrophages of the reticuloendothelial system. They function to filter small foreign particles (including bacteria) and old reticulocytes from the blood. In the Cell Ontology they are to be found by their function (they are a type of defensive_cell), by their lineage (they are derived from a mesodermal_cell derived from a hematopoietic_stem_cell, itself a type of stem_cell), by their morphology (they are a type of circulating_cell) and by their organism (they are a type of animal_cell). Discussion Ontologies in bioinformatics are intended to capture and formalize a domain of knowledge, and the ontology reported here attempts to do this within the domain of cell types. It is designed to be useful in the sense that a researcher should be able to find, in a rapid and intuitive way, any cell type in any of the major model organisms and, having found it, learn a considerable amount about that cell type and its relationships to other biological objects. A core feature of the ontology, and one that differentiates it from other resources that contain cell types such as SNOMED and the FMA [ 8 , 9 ], and the Drosophila and Arabidopsis ontologies [ 3 , 17 ], is that the cell ontology explicitly sets out to include cell types from all the major model organisms within a common framework. In addition, it also seeks to incorporate a great deal of phenotypic information about these cell types and is thus far more comprehensive in its cellular detail than these other resources. The intention is that the new cell-type ontology should provide organism-independent knowledge as well as cell-type unique identifiers (ID) that can be incorporated into any database holding cell-type-associated knowledge. The formalized structure of the ontology, together with its set of unique IDs, will allow curators to incorporate cell-type data into their databases, integrate the data with the knowledge encapsulated in the ontology, and use the IDs to interoperate with other databases. While we expect such bioinformatics applications to be its immediate use, we hope that, in the longer term, all biologists will find the ontology useful. The expected short-term use of the ontology will thus be in cataloguing phenotypes and gene expression patterns. Indeed, it is quite surprising that those who work with model organisms still lack the bioinformatics resources needed to catalogue, archive and access the details of the phenotypes emerging from mutant screens and natural variations. A robust representation of normal and mutant phenotypes in all of the model organisms will require ontologies for a wide range of macroscopic properties (pathology, anatomy, abnormal quantifiers, and so on) and we view the cell ontology as a component of this programme that should be useful in cataloguing phenotypes (and other attributes) associated with cell types. In the long term we expect that molecular biology and biological databases will move beyond being gene-centric and that biological mechanisms will be studied at a more integrated level. Cells are the biological units with which tissues and organs and organ systems are built. A rich and explicit description of cell types across phyla that are adapted by biological databases will help facilitate this transition. Finally, it should be pointed out that, like many such resources, this ontology is not complete: although it contains all the common cell types, there will certainly be some that have been omitted. Most importantly, although many of the cell types are fully described by function, morphology, organism, and so on, others are inadequately described and more relationships need to be made. A particular weakness is the fact that the category identified as experimentally_modified_cell has yet to be populated, and doing this will involve consideration of the various cell lines held in the major collections. As with other community resources, community input is essential for the development and maintenance of the Cell Ontology; biologists with comments and additions are therefore welcome to contribute to the ontology and should contact the curator ashburner@ebi.ac.uk . Materials and methods The ontology includes the major cell types from the major model organisms (for example, human, mouse, Drosophila , Caenorhabditis , zebrafish, Dictyostelium discoideum , Arabidopsis , fungi and prokaryotes). These cell types have been collated from our own knowledge, from major textbooks (for example [ 20 - 22 ]), from the embryo and anatomy ontologies available on the OBO site [ 7 ], and from colleagues (who are thanked in the acknowledgements). The ontology currently holds some 680 cell types, together with their synonyms and, in most cases, text definitions. The ontology was constructed using the open source Java tool OBO-Edit (previously known as DAG-Edit) [ 23 ], which is convenient for building ontologies that are consistent with the GO formalism. The resulting ontology is available in both the GO 'flat-file' format [ 24 ] and the newly defined 'OBO format' [ 25 ], and can easily be viewed using the OBO-Edit or the COBrA open source Java tool [ 26 ]. Availability The Cell Ontology is available from the OBO site [ 19 ]. Following the cell.obo link will take the user to a page in which the current version of the Ontology, and archived older versions, can be viewed (view) or downloaded (download). Differences between the current and previous version can be seen by following the Diff to link.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC551541.xml
368158	Preferential Duplication of Conserved Proteins in Eukaryotic Genomes	A central goal in genome biology is to understand the origin and maintenance of genic diversity. Over evolutionary time, each gene's contribution to the genic content of an organism depends not only on its probability of long-term survival, but also on its propensity to generate duplicates that are themselves capable of long-term survival. In this study we investigate which types of genes are likely to generate functional and persistent duplicates. We demonstrate that genes that have generated duplicates in the C. elegans and S. cerevisiae genomes were 25%–50% more constrained prior to duplication than the genes that failed to leave duplicates. We further show that conserved genes have been consistently prolific in generating duplicates for hundreds of millions of years in these two species. These findings reveal one way in which gene duplication shapes the content of eukaryotic genomes. Our finding that the set of duplicate genes is biased has important implications for genome-scale studies.	Introduction Gene duplication is the most important source of new genes and consequently a vital source of genetic novelty ( Ohno 1970 ). Recently, the availability of completely sequenced genomes has sparked renewed attention in this subject at the genome scale. Most genomic studies of gene duplication have focused on the mechanisms responsible for generating duplicate genes, the consequences of gene duplication for genetic redundancy, or the effect that duplication has on the molecular evolution of the genes involved ( Seoighe and Wolfe 1999 ; Lynch and Conery 2000 ; Dermitzakis and Clark 2001 ; Van de Peer et al. 2001 ; Gu et al. 2002 , Gu et al. 2003 ; Kitami and Nadeau 2002 ; Kondrashov et al. 2002 ; Nembaware et al. 2002 ). Comparatively less attention has been devoted to the essential question of whether some genes are more likely to give rise to functional and persistent duplicates than others and thus contribute more to the gene content of eukaryotic genomes (but see Kondrashov et al. 2002 ; Nembaware et al. 2002 ). Investigating this aspect of gene duplication will not only help answer questions about gene content—such as why certain proteins duplicate to generate multigene families while others remain in single copy—but will provide insight into the process of duplication itself. Each of the three steps leading to the generation of preserved gene duplicates, including their (1) mutational generation, (2) fixation in a population, and (3) preservation through a period when they may be functionally redundant, may favor some genes over others. For example, gene duplicates that lead to an advantageous increase in gene dosage will be preferentially fixed by positive selection, as has been observed in bacteria and Saccharomyces cerevisiae ( Romero and Palacios 1997 ; Brown et al. 1998 ; Dunham et al. 2002 ). For other genes, for which stoichiometry is important, the converse may be true: gene duplication may be strongly deleterious ( Gerik et al. 1997 ), and while such duplications may commonly arise in single individuals, they are unlikely to become fixed in the population. The step of preservation also has a great potential to create a bias in the types of genes that duplicate since the vast majority of duplicate gene copies that arise in a population are rapidly lost to nonfunctionalizing mutations ( Lynch and Conery 2000 ). Theoretical accounts of duplicate gene preservation make various predictions about the types of genes that will be preserved following duplication. Specifically, these models predict that genes with a larger number of cis -regulatory regions, expressed in many tissues ( Lynch et al. 2001 ) or encoding multidomain proteins ( Gibson and Spring 1998 ; Stoltzfus 1999 ), will be preferentially preserved. By investigating the molecular attributes of the types of genes that duplicate, we may be able to validate these predictions and determine which steps in the process of duplication act as a selective sieve, promoting the duplication of some genes and hindering the duplication of others. Beyond providing information about the mechanisms of duplication, data about the biases in which genes duplicate will serve as an essential baseline for other genome-scale studies in this field. For example, recent work has argued that gene duplication leads to a relaxation of selection and consequently an elevation in the rate of molecular evolution for the duplicated genes ( Kondrashov et al. 2002 ; Nembaware et al. 2002 ). In support of this argument, these studies compared the evolutionary rate of genes that had duplicated to the rate of genes that were in single copy. A higher rate of evolution for the genes with duplicates was taken to support their hypothesis. One problem with this approach is that it is based on the assumption that the set of genes that generate duplicates is not biased with respect to the genes' rate of evolution. Indeed, if the genes that duplicate had higher rates of evolution prior to duplication, this would invalidate the above conclusions. Similarly, any study that reveals differences between the properties of duplicate genes and those in single copy ( Kondrashov et al. 2002 ; Nembaware et al. 2002 ; Gu et al. 2003 ) should hesitate to conclude that these differences are caused by duplication per se without considering the biases in the attributes of the genes that lead to duplicates. In some cases, the authors themselves acknowledge this problem (e.g., Kondrashov et al. 2002 ; Gu 2003 ). For these reasons we chose to investigate a bias in the molecular attributes of the genes that duplicate. One very informative gene attribute is the rate of protein evolution defined as the number of nonsynonymous substitutions per nonsynonymous site in a given time ( K A ). This measure of protein evolution has been shown to be related to several important properties of genes, including dispensability, level of expression, and the number of protein–protein interactions ( Hirsh and Fraser 2001 ; Pal et al. 2001 ; Fraser et al. 2002 ). We chose to compare the rates of evolution of the genes that have given rise to observable duplicates in the well-studied genomes of S. cerevisiae and Caenorhabditis elegans with those that have not. Such a comparison is not straightforward since gene duplication itself may affect the rate of molecular evolution ( Lynch and Conery 2000 ; Kondrashov et al. 2002 ). To avoid this problem, we did not use the rate of evolution of each singleton and duplicate pair in S. cerevisiae and C. elegans (the “study genes”), but instead measured evolutionary rates in two distantly related outgroup species, Drosophila melanogaster and Anopheles gambiae (such a pair of orthologs is referred to as a “representative pair”). Because evolutionary rates for a particular gene are highly correlated in diverse lineages ( Bromham and Penny 2003 ), we reasoned that the nonsynonymous divergence between the members of each representative pair would be a good proxy for the rate of evolution of the study genes in a way that is unaffected by the process of duplication ( Figure 1 ). Our results reveal that the genes that have duplicated in the genomes of S. cerevisiae and C. elegans appear to be a biased set of slowly evolving genes and that slowly evolving genes have been consistently prolific in generating duplicates for hundreds of millions of years in these lineages. Figure 1 The Approach Used to Estimate the Rate of Evolution for Duplicate and Singleton Genes For each duplicate (gray lines) and singleton (black lines) gene/pair in S. cerevisiae and C. elegans , unduplicated orthologs were identified in D. melanogaster and A. gambiae . The K A between this representative pair of orthologs was taken as an estimate of the rate of evolution of duplicate and singleton genes in the study species that is independent of the effects of duplication on molecular evolution. Results Evolutionary Rates of Duplicate and Singleton Genes The number of duplicate pairs and singleton genes identified in the genomes of S. cerevisiae and C. elegans and the number of representative pairs of these genes found in D. melanogaster and A. gambiae are provided in Table 1 . Our comparison of the nonsynonymous divergence between orthologs of these two classes of genes revealed that representative pairs of duplicates in both S. cerevisiae and C. elegans have much slower rates of evolution (Mann–Whitney U test, p < 0.001 for both) ( Figure 2 ). The representative pairs of the duplicated genes in S. cerevisiae evolve on average more than 50% slower than the representative pairs of singletons (0.192 versus 0.302), while in C. elegans the difference exceeds 25% (0.230 versus 0.296). Figure 2 A Comparison of the Evolutionary Rates of Duplicate and Singleton Genes The average rate of nonsynonymous evolution ( K A ) for representative pairs of duplicate and singleton genes in the two study organisms S. cerevisiae (A) and C. elegans (B) is shown. Representative pairs of duplicate genes evolve significantly more slowly in both study organisms (Mann–Whitney U test, p < 0.001). Table 1 Number of Genes/Pairs Identified in the Study Organism and Number of Orthologs of These Genes Found in D. melanogaster and A. gambiae In addition to estimating rates of evolution for representative pairs of the two classes of genes, we also attempted to quantify structural protein evolution by computing the number of gaps per basepair in the alignments of the representative pairs. We reasoned that this measure is likely to be a monotonic proxy for the number of indels that have occurred in the evolution of a protein since the split of A. gambiae and D. melanogaster . Results from this analysis echoed those of the K A comparisons: representative pairs of duplicate genes in both S. cerevisiae and C. elegans are much less likely to have accumulated insertions or deletions than representative pairs of singletons (Mann–Whitney U test, p < 0.0001 for both) ( Figure 3 ). Figure 3 A Comparison of the Rate of Structural Evolution for Duplicate and Singleton Genes For each representative pair, the number of gaps per aligned nucleotide was calculated. For both S. cerevisiae (A) and C. elegans (B), representative pairs of duplicates have significantly fewer insertions per basepair than representative pairs for singletons (Mann–Whitney U test, p < 0.0001 for both). To further validate these conclusions, we wanted to test several potential sources of error in our analysis of both K A and the indel rate. First, some of the orthologs identified in D. melanogaster and A. gambiae have undergone duplication in these lineages. This could both affect their rates of evolution, as discussed above, and also lead to the identification of the slowest evolving paralog in D. melanogaster and A. gambiae for the representative pairs of study genes. The latter effect can lead to an artificially low estimate of the evolutionary rates. To test for this possibility, we repeated our analysis using only representative pairs that have not duplicated in either D. melanogaster or A. gambiae . Although this analysis included substantially fewer genes (see Materials and Methods ), the results remained unchanged and strongly statistically significant (Mann–Whitney U test, p < 0.005 for both organisms). Second, we wanted to make sure that the bias is not due to the peculiarly slow evolution of duplicate genes in multigene families. A reanalysis for only those duplicated genes (in the study organisms) with no other paralogs in the genome revealed very similar results (data not shown). Third, it is possible that our conservative definition of “singleton” may have artificially biased the set of singletons towards rapidly evolving genes. This could be true if slowly evolving singleton genes tend to possess anciently conserved, widely shared protein domains. By generating homology to other genes, these domains may make these singletons fall below the conservative E -value cutoff that we used. To test this possibility, we relaxed our criteria for singleton genes to include all those genes with no E -value less than 10 –10 . The average rate of evolution for this group of singleton genes was no different than for the former set (data not shown). Biased Mutation Cannot Explain the Lower K A of Duplicates The simplest interpretation of these data is that the genes generating preserved duplicates are a biased set of constrained, slowly evolving proteins. An alternative explanation is that representative pairs of singletons are found in genomic regions with a higher mutation rate than are representative pairs of duplicates—although there is no a priori reason why this should be true. One way of testing this possibility is to compare the number of synonymous nucleotide substitutions per synonymous site ( K S ) for the representative pairs of the two classes of genes. This measure is customarily used as a proxy for mutation rate because substitutions at synonymous sites are generally thought to be selectively neutral. However, in many genes, especially those expressed at high levels, synonymous sites appear to be under selection, as evidenced by codon bias. For such genes, the rate of synonymous evolution will underestimate the rate of mutation ( Sharp et al. 1988 ; Shields et al. 1988 ; Sharp and Li 1989 ; Li 1997 ). Given that previous reports have suggested that duplicate genes are expressed at particularly high levels in S. cerevisiae ( Seoighe and Wolfe 1999 ), their rate of synonymous evolution should be lower than that of singletons even in the absence of mutational differences. To overcome this complication, we computed the partial correlation coefficients between each of the three factors: codon bias (measured by the codon adaptation index [CAI] [ Sharp and Li 1987 ] in D. melanogaster) , gene class (whether the representative pair was for a duplicate or singleton), and K S (between representative pairs). Our results, presented in Table 2 , reveal that, as expected, representative pairs of S. cerevisiae duplicate genes have a lower K S than representative pairs of singleton genes (Spearman Correlation column), but that this correlation disappears when we control for codon bias (Partial Correlation Coefficient column). Thus, in the case of S. cerevisiae, the higher codon bias of the slowly evolving representative pairs completely accounts for the differences in K S between the two groups. For C. elegans, the K S of the representative pairs for duplicate genes is in fact marginally higher than that for singleton genes, and this slight trend remains when codon bias is taken into account. Thus, mutational differences cannot account for the differences in the rate of protein evolution in either S. cerevisiae or C. elegans . Table 2 Correlation Coefficients and Partial Correlation Coefficients for the Three Factors Gene Class (Duplicate or Singleton), CAI (in D. melanogaster ), and K S (of Representative Pairs) Significance was tested for the direct and partial correlation coefficients using the statistics and , respectively, where n is the sample size, m is the number of variables held constant, and r is the rank correlation coefficient ( Sokal and Rohlf 1995 ) a For this parameter, representative pairs were given a value of either 0 (for a singleton) or 1 (for a duplicate) NS, nonsignificant; , p = 0.05; , p = 0.01; *, p = 0.001 Codon Bias and the Rate of Evolution of Duplicate Genes We can also use the level of codon bias to gain additional insight into the potential reasons for the generation and maintenance of duplicate copies of conserved genes. Codon bias is a proxy for the level of expression ( Akashi 2001 ), while the level of expression is a good predictor of the rate of protein evolution ( Pal et al. 2001 ; Krylov et al. 2003 ). To determine whether the reason for the slow evolution of duplication-prone genes is their higher level of expression, we performed a partial correlation analysis similar to the analysis of K S above. Table 3 shows Spearman rank and partial rank correlations between pairs of the three variables gene class (singleton or duplicate study gene), CAI (in D. melanogaster ), and K A (of the representative pairs). This analysis revealed some important differences in how the duplication bias is generated in S. cerevisiae and C. elegans . Table 3 Correlation Coefficients and Partial Correlation Coefficients for the Three Factors Gene Class, CAI (in D. melanogaster ), and K A (of Representative Pairs) Significance was tested for the direct and partial correlation coefficients using the statistics and , respectively, where n is the sample size, m is the number of variables held constant, and r is the rank correlation coefficient ( Sokal and Rohlf 1995 ) a For this parameter, representative pairs were given a value of either 0 (for a singleton) or 1 (for a duplicate) NS, nonsignificant; , p = 0.05; , p = 0.01; *, p = 0.001 First, both direct and partial correlations for S. cerevisiae show that the CAI of the representative pairs of duplicates is greater than that of the representative pairs of singleton genes. This indicates that the genes leading to preserved duplicates in S. cerevisiae tend to be unusually highly expressed ( p < 0.001). In contrast, for C. elegans, duplicate genes do not appear to be biased towards highly expressed genes ( p > 0.1). This difference may reflect a disparity in the mutational generation, fixation, or preservation of duplicates in these two organisms. This analysis also reveals that when codon bias is held constant, the relationship between K A and gene class persists in both organisms. In the case of C. elegans , the correlation coefficient between gene class and K A remains nearly identical when CAI is held constant. For S. cerevisiae, the partial correlation coefficient between K A and gene class does decrease when CAI is held constant (but remains highly significant), implying that the slower evolution of representative pairs of the duplicated genes in S. cerevisiae is partly mediated by preferential duplication of highly expressed genes. To validate these conclusions, we repeated the same analysis using CAI values in the study organisms rather than in D. melanogaster . This analysis revealed very similar results (data not shown). Time Uniformity of the Bias To determine whether conserved genes have been preferentially duplicated throughout the history of the S. cerevisiae and C. elegans lineages, we plotted the evolutionary rate of representative pairs and the average CAI (both in D. melanogaster and in the study organisms) for duplicate pairs of different age classes (where age is measured by K S between the duplicate study genes) ( Figure 4 ). While large K S estimates are subject to a large amount of error (such that estimates of K S above 2 are typically unreliable), this analysis captures the uniformity of the bias in these lineages. For both organisms, slowly evolving genes appear to have led to the duplicate genes in all age classes (covering hundreds of millions of years). For C. elegans, both the evolutionary rates of the representative pairs and their CAI values remain virtually constant for duplicated genes of all ages. In addition, the CAI values for the duplicate pairs of different ages in C. elegans are very similar to the CAI values for singletons—the only exception is a slight elevation in the CAI for duplicate pairs in the K S range from 1 to 1.5. By contrast, the plot for S. cerevisiae reveals that young duplicate genes ( K S < 2.0) tend to have representative pairs with a lower K A than those of older pairs, and this trend is paralleled by the elevated CAI of these young duplicate pairs. Figure 4 The Codon Bias and Rate of Evolution of Genes Leading to Duplicates over the Evolutionary History of S. cerevisiae and C. elegans For both S. cerevisiae (A) and C. elegans (B), moving averages of nonsynonymous substitutions per site ( K A , in dark gray), codon bias in the study organism (measured with CAI, in black), and codon bias of the representative ortholog in D. melanogaster (CAI, in light gray) are plotted against the number of synonymous substitutions per site ( K S ) between duplicate pairs. The bin size is 15, and standard error bars are shown. Dashed lines represent the average CAI of singleton genes and the average K A of representative pairs of singleton genes. A problem for interpreting this trend in S. cerevisiae is that duplicate pairs with a high codon bias are expected to have a depressed value of K S , as discussed above, and thus will appear younger than they really are. To overcome this problem, we corrected K S estimates for S. cerevisiae genes based on their CAI using a simple approach recently developed for this species (see Materials and Methods ) (A. Hirsh, H. Fraser, and D. Wall, personal communication). After correcting K S estimates, the plots of K A and CAI shift slightly ( Figure 5 ), but the trends remain. We can further see that the duplicate pairs with the unusually high CAI and the unusually low K A of the representative pairs have corrected K S less than 2.0. It is intriguing that this age range matches the estimated time of the whole-genome duplication in the S. cerevisiae lineage ( K S , approximately1.0; 80 million years ago) ( Wolfe and Shields 1997 ; Pal et al. 2001 ). If the set of genes preserved after polyploidization in S. cerevisiae was biased towards highly expressed genes, this could explain the heterogeneity in both K A and CAI and could explain why duplicate genes in C. elegans , an organism that has likely not undergone a whole-genome duplication, were not enriched for genes with a high level of expression. With respect to this hypothesis, it is interesting to note that for young duplicate genes ( K S < 2), K A estimates for representative pairs of duplicate genes in S. cerevisiae are much lower than for duplicate genes in C. elegans , whereas for older duplicate genes ( K S > 2), the K A estimates are roughly equivalent in both S. cerevisiae and C. elegans . Figure 5 Correcting for Synonymous Substitutions Reveals That S. cerevisiae Genes That Have Recently Duplicated Have a Higher Codon Bias and Slower Rate of Evolution Than Those That Duplicated in the Ancient Past For duplicate genes in S. cerevisiae , moving averages of the number of nonsynonymous substitutions per nonsynonymous site of representative pairs ( K A , in dark gray), the codon bias in S. cerevisiae (CAI, in black), and the codon bias of representative pairs in D. melanogaster (CAI, in light gray) are plotted against the adjusted number of synonymous substitutions per site (see Materials and Methods ) between duplicate pairs. The bin size is 15, and standard error bars are shown. Lines with broad dashes show the respective averages for singleton genes in S. cerevisiae, and the line with short dashes shows the average K A for representative pairs of duplicate genes in C. elegans . Other studies have noted that ribosomal subunit proteins were particularly prolific in generating duplicate pairs via polyploidization in S. cerevisiae ( Seoighe and Wolfe 1999 ). Indeed, these genes account for 49 of the duplicate pairs in our study. To determine whether this group is responsible for the depressed rates of evolution of young duplicate pairs, we plotted CAI and K A versus K S without ribosomal proteins ( Figure 6 ). The plot reveals that without ribosomal proteins, young duplicate genes possess rates of evolution comparable to those of other age classes and more similar to the values found for duplicate genes in C. elegans . Thus, the overrepresentation of duplicate ribosomal proteins following the polyploidization event in S. cerevisiae appears to explain the low rates of evolution of young duplicate genes in this species. Even with these ribosomal genes removed, however, younger genes have much higher CAI values. Figure 6 After Removing Ribosomal Genes, the Magnitude of the Bias towards the Slower Evolution of Duplicate Genes Is Similar in Both S. cerevisiae and C. elegans For nonribosomal duplicate genes in S. cerevisiae , moving averages of the number of nonsynonymous substitutions per nonsynonymous site of representative pairs ( K A , in dark gray), the codon bias in S. cerevisiae (CAI, in black), and the codon bias of representative pairs in D. melanogaster (CAI, in light gray) are plotted against the adjusted number of synonymous substitutions per site (see Materials and Methods ) between duplicate pairs. The bin size is 15, and standard error bars are shown. Lines with broad dashes show the respective averages for singleton genes in S. cerevisiae, and the line with short dashes shows the average K A for representative pairs of duplicate genes in C. elegans . Discussion Most genome-scale studies of duplicate genes have focused either on the mechanisms of duplication or on the consequences of duplication at the molecular or organismal level. In this study we ask a different type of question: namely, which types of genes are more likely to duplicate than others? The method we use—identifying duplicate genes in one organism and obtaining evolutionary rate measurements from two outgroup species (see Figure 1 )—allows us to compare the evolutionary rate of genes that have duplicated to that of those that have not. Importantly, it allows us to do this without confounding the effect the duplication itself has on the rate of molecular evolution ( Lynch and Conery 2000 ; Kondrashov et al. 2002 ). Our data reveal that genes that have duplicated in the genomes of S. cerevisiae and C. elegans have much slower rates of amino acid substitution, as well as lower rates of insertion and deletion, on average than those that have remained in single copy. To strengthen this conclusion, we tested several potential sources of error in our estimates of rates of evolution for the two classes of genes. We found that none of the potential complications—including the effect of duplication within the lineages of D. melanogaster and A. gambiae , duplications predating the split of the studied lineage and the outgroups, the especially slow evolution of multigene families, the operational definitions of duplicate and singleton genes, or the possibility of mutational differences—appear to affect our estimates of evolutionary rates of the two gene classes. We have also attempted to ascertain whether conserved genes have been generating duplications in a persistent fashion or whether this bias was generated at a particular time in the history of the two studied genomes. Our analysis demonstrates that both lineages have experienced a consistent and very similar level of bias over hundreds of millions of years. In addition, there has been a recent duplication of particularly slowly evolving genes in the yeast genome, coinciding roughly with the time of the postulated genome duplication in this lineage. Importantly, the consistency of the pattern over such long evolutionary periods of time in such diverse lineages suggests that the preferential generation or retention of duplicates of slowly evolving genes might be a general feature of eukaryotic evolution. Why do conserved, slowly evolving genes have a proclivity to generate duplicates? In order to answer this question, it is important to determine which of the three steps of duplication—mutation, fixation, or preservation—are responsible for this trend. As discussed above, both fixation and preservation have the potential to create a bias in the types of genes that duplicate. The step of fixation could generate a bias either because (1) many of the genes that are duplicated in a single individual are deleterious and thus are quickly removed from the population or (2) many of the duplicate genes that reach fixation in a population do so because of positive selection for the duplicate copy, rather than reaching fixation neutrally by genetic drift. For the first mechanism to work, increases in the dosage of slowly evolving genes must be less likely to have deleterious consequences to organismal fitness than increases in the dosage of more rapidly evolving genes. Recent empirical work, however, has shown that the opposite might be true. In particular, data from yeast have shown that less dispensable, slowly evolving genes are more likely to be haploinsufficient than dispensable genes ( Papp et al. 2003 ). This implies that changes in dosage of slowly evolving genes may have greater fitness consequences in general. The second mechanism by which fixation may generate the bias is more tenable. This mechanism requires that many duplicate genes fix by positive selection and that duplicates of slowly evolving genes do so with higher likelihood. Examples from S. cerevisiae and bacteria ( Romero and Palacios 1997 ; Brown et al. 1998 ; Dunham et al. 2002 ) support the possibility that duplications of genes can lead to beneficial increases in dosage and can be fixed by positive selection. One set of genes that may be especially likely to lead to beneficial increases in dosage following duplication are genes that are already required at high expression levels. It is interesting in this regard that many highly expressed genes have recently duplicated in S. cerevisiae (see Figure 5 ) ( Seoighe and Wolfe 1999 ) and that the preferential duplication of genes with a high codon bias accounts partially for the bias that we observe in S. cerevisiae (see Table 3 ). While the preferential duplication of highly expressed genes is not observed for C. elegans , it is possible that duplications of slowly evolving genes are also likely to lead to beneficial increases in dosage for some other, yet unknown, reason. The step of preservation also has the potential to generate the bias we observe since (1) many of the duplicate gene copies that arise in a population are lost quickly to nonfunctionalizing mutations ( Lynch and Conery 2000 ) and (2) several models of duplicate gene preservation suggest that slowly evolving genes may have an increased likelihood of being preserved. In particular, these models predict the preferential preservation of genes with many cis -regulatory regions, expressed in many tissues ( Lynch et al. 2001 ), or of genes that encode multidomain proteins ( Gibson and Spring 1998 ; Stoltzfus 1999 ). Because the higher level and the greater breadth of expression, as well as the larger number of protein interactions, correlate with the slower rate of protein evolution ( Duret and Mouchiroud 2000 ; Pal et al. 2001 ; Fraser et al. 2002 ), these models predict preferential preservation of slowly evolving genes. If the step of preservation accounts for the slower evolution of duplicate genes, one prediction is that the rates of evolution of newly arisen gene duplicates should be higher than the rates of older gene duplicates and closer to the rates of evolution of singletons. Our data do not reveal any such trend for either S. cerevisiae or C. elegans (see Figure 4 ). The negative result, however, may simply reflect a lack of statistical power. The higher evolution rates of newly arisen gene duplicates should only be apparent for very young duplicate pairs. Indeed, the average half-life of a duplicate pair may be as short as 5 million years ( Lynch and Conery 2000 ), corresponding to a K S of approximately 0.05. There are very few such pairs in our dataset. It is unclear whether fixation, preservation, or both of these steps together cause the bias towards the preferential duplication of slowly evolving genes. The relative importance of these two steps depends largely on the frequency with which duplicate genes are fixed by positive selection. If the vast majority of duplicate genes are initially redundant and fix by genetic drift, as assumed in many models of gene duplication ( Ohno 1970 ; Force et al. 1999 ; Lynch and Force 2000 ; Lynch et al. 2001 ), fixation cannot explain the bias. If, on the other hand, duplicate genes often fix by positive selection ( Kondrashov et al. 2002 ), the step of fixation may be dominant in generating the bias inthe types of genes that duplicate. The relative frequency with which duplicate genes fix because of positive selection and genetic drift remains to be established. Beyond providing insight into the mechanisms of gene duplication, the bias has important consequences for the content of eukaryotic proteomes. If conserved, slowly evolving genes consistently generate preserved duplicate copies of themselves, proteomes will tend to become enriched for these genes over the course of evolution. This prediction is especially interesting in relationship to recent complementary work ( Krylov et al. 2003 ) that shows that genes with a slow rate of evolution, a low dispensability, and a high level of expression are less likely to be lost over the course of evolution. Taken together, these two studies predict that slowly evolving genes should be the main sources of genes in eukaryotic genomes. It is also noteworthy that the two results are not independent. If slowly evolving genes are more likely to duplicate to form multigene families, then they should be less likely to be lost from a particular lineage, since this would entail the loss of many distinct genetic copies. The extent to which this effect explains the preferential loss of fast evolving genes remains to be determined. The mere existence of this bias is very important for the interpretation of genomic-level studies of gene duplication. For example, some recent studies have argued that two general consequences of gene duplication are (1) an increased rate of evolution for the duplicated genes immediately following duplication (e.g., Kondrashov et al. 2002 ) and (2) increased functional redundancy at the genetic level ( Gu et al. 2003 ). To make their arguments, both of these studies compare duplicate and singleton genes within a single organism under the assumption that the types of genes that duplicate are unbiased with respect to the molecular attribute of interest (note that a correction for this problem has been attempted before by separating genes into functional classes [e.g., Kondrashov et al. 2002 ]). The study presented here shows that this assumption is not valid. Duplicate genes are, in fact, a very biased set of genes, at least with respect to their rate of evolution. Interestingly, in the case of the studies just mentioned, the bias that we observed makes the conclusions conservative. Indeed, the bias that we observed may explain why other studies have failed to find the expected higher rate of evolution for genes that have recently undergone duplication (e.g., Kitami and Nadeau 2002 ). The preferential duplication of conserved genes, combined with the increased rate of evolution following duplication, may lead to no measurable difference in the rate of evolution between singleton and duplicate genes. In general, any genome-scale study that attempts to assess the effects of duplication on molecular evolution should consider the prior distribution of the molecular attributes of the genes that lead to duplicates. Materials and Methods Identification of duplicate and singleton genes and their orthologs The gene and protein sequences of S. cerevisiae, C. elegans, D. melanogaster, and A. gambiae were downloaded from GenBank (Bethesda, Maryland, United States) at http://www.ncbi.nlm.nih.gov/Ftp/index.html . To identify duplicate and singleton genes, a reciprocal protein BLAST ( Altschul et al. 1997 ) was performed on the proteomes of the two study organisms using default parameters and simple sequence filtering. Singleton genes were conservatively defined as those genes without a hit with an E -value of less than 0.1, following previous studies ( Gu et al. 2003 ). Duplicate pairs in these genomes of S. cerevisiae and C. elegans were identified as reciprocal best hits with an E -value of less than 10 –10 in both directions that could be aligned over greater than 60% of their lengths. Orthologs were identified as reciprocal best BLAST hits between two organisms using the same criteria: E -values of less than 10 –10 and alignable over greater than 60% of the gene lengths. In the case of duplicate pairs, the same criteria were used, except that both duplicates needed to hit the same gene in the outgroup species and the duplicate genes needed to be the top two best hits in the reciprocal blast. To identify representative pairs for each singleton and duplicate gene, we first identified an ortholog in D. melanogaster and then identified the ortholog of this gene in A. gambiae . Obtaining K A and indel measurements for representative pairs To obtain the nucleotide alignments for each representative pair, we obtained the BLASTP alignment of the two orthologs, removed gaps in these alignments by trimming back from both ends of each gap until an anchor pair was found (following the method described in Conery and Lynch [2001] ), and then replaced the amino acid alignment with the respective nucleotide sequence. Based on these alignments, we used the PAML software package ( Yang 1997 ) to estimate the number of synonymous and nonsynonymous substitutions per site. The number of gaps per nucleotide length of each alignment was also recorded and used as a proxy for the number of indels that have occurred during the divergence of D. melanogaster and A. gambiae . To test whether including duplicate pairs and singleton genes with representative pairs possessing paralogs in the D. melanogaster and A. gambiae lineages biased our results, we reanalyzed the distributions of nonsynonymous rates of evolution and number of indels after removing these genes. For both C. elegans and S. cerevisiae , we eliminated representative pairs with paralogs with a BLAST E -value less than 10 –10 in either of the outgroup genomes (leaving 60 duplicates and 225 singletons and 48 duplicates and 530 singletons, respectively) and eliminated all representative pairs with paralogs with an E -value of less than 0.1 (leaving only 38 duplicates and 114 singletons and 29 duplicates and 318 singletons, respectively). Results from the reanalysis revealed significant trends similar to those found when using all representative pairs. Obtaining CAI values and correcting K S We obtained CAI values for genes in the D. melanogaster , S. cerevisiae , and C. elegans genomes using the program CodonW (available from ftp://molbiol.ox.ac.uk/Win95.codonW.zip ; written by John Peden, now at Oxagen [www.oxagen.co.uk], and originally developed in the laboratory of Paul Sharp, Department of Genetics at the University of Nottingham, United Kingdom). The table used to calculate CAI for S. cerevisiae is the standard table included in the package. We obtained the appropriate codon usage tables for C. elegans and D. melanogaster from studies by Duret and Mouchiroud (1999 ) and Carbone et al. (2003 ), respectively. For duplicate genes in S. cerevisiae, we used CAI values of each pair to help obtain a better relative estimate of their ages. This was necessary because duplicate pairs with a high codon bias effectively have fewer neutral synonymous sites, resulting in the gross underestimation of their age based on K S alone ( Sharp et al. 1988 ; Shields et al. 1988 ; Sharp and Li 1989 ; Li 1997 ). A recent study has shown that the number of synonymous substitutions expected for genes with a given codon bias in S. cerevisiae is given by K S = rt (1 – c ), where r is the rate of synonymous substitution in genes with no codon bias, t is time, and c is codon bias as measured by CAI (A. Hirsh, H. Fraser, and D. Wall, personal communication). Rearranging this equation yields the formula K S ′ = rt = K S /(1 – c ), which we used to obtain corrected estimates of the age of duplicate pairs in S. cerevisiae . No such correction was made for C. elegans genes because they were not shown to have a significantly higher codon bias than singleton genes and because no simple means of correction is presently known.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC368158.xml
524486	Branched-chain amino acid aminotransferase and methionine formation in Mycobacterium tuberculosis	Background Tuberculosis remains a major world-wide health threat which demands the discovery and characterisation of new drug targets in order to develop future antimycobacterials. The regeneration of methionine consumed during polyamine biosynthesis is an important pathway present in many microorganisms. The final step of this pathway, the conversion of ketomethiobutyrate to methionine, can be performed by aspartate, tyrosine, or branched-chain amino acid aminotransferases depending on the particular species examined. Results The gene encoding for branched-chain amino acid aminotransferase in Mycobacterium tuberculosis H37Rv has been cloned, expressed, and characterised. The enzyme was found to be a member of the aminotransferase IIIa subfamily, and closely related to the corresponding aminotransferase in Bacillus subtilis , but not to that found in B. anthracis or B. cereus . The amino donor preference for the formation of methionine from ketomethiobutyrate was for isoleucine, leucine, valine, glutamate, and phenylalanine. The enzyme catalysed branched-chain amino acid and ketomethiobutyrate transamination with a Km of 1.77 – 7.44 mM and a Vmax of 2.17 – 5.70 μmol/min/mg protein, and transamination of ketoglutarate with a Km of 5.79 – 6.95 mM and a Vmax of 11.82 – 14.35 μmol/min/mg protein. Aminooxy compounds were examined as potential enzyme inhibitors, with O-benzylhydroxylamine, O-t-butylhydroxylamine, carboxymethoxylamine, and O-allylhydroxylamine yielding mixed-type inhibition with Ki values of 8.20 – 21.61 μM. These same compounds were examined as antimycobacterial agents against M. tuberculosis and a lower biohazard M. marinum model system, and were found to completely prevent cell growth. O-Allylhydroxylamine was the most effective growth inhibitor with an MIC of 78 μM against M. marinum and one of 156 μM against M. tuberculosis . Conclusion Methionine formation from ketomethiobutyrate is catalysed by a branched-chain amino acid aminotransferase in M. tuberculosis . This enzyme can be inhibited by selected aminooxy compounds, which also have effectiveness in preventing cell growth in culture. These compounds represent a starting point for the synthesis of branched-chain aminotransferase inhibitors with higher activity and lower toxicity.	Background Tuberculosis remains one of the leading causes of worldwide mortality and morbidity, infecting an estimated 8 million people annually with approximately 2 million deaths [ 1 ]. The situation regarding the control of tuberculosis has significantly worsened over the last decades, with the spread of multidrug resistant strains. In the absence of an effective vaccine for tuberculosis, there is an urgent need for the development of novel antimycobacterial agents. The study of mycobacterial biochemistry assists this development through the identification and characterization of cellular enzymes amenable to therapeutic inhibition. Polyamine synthesis and its associated methionine (Met) regeneration pathway (Figure 1 ) are known to be potential drug targets in a variety of microorganisms [ 2 - 4 ]. The synthesis of polyamines is essential during periods of DNA replication, although the exact physiological role of these compounds remains unclear [ 3 ]. The production of spermidine from putrescine, or spermine from spermidine, consumes the amino acid Met in a 1:1 stoichiometry yielding methylthioadenosine (MTA) as a byproduct. As Met biosynthesis is energetically expensive, and many organisms lack the ability to synthesize the amino acid, a unique pathway exists which recycles Met from MTA. To date, the entire pathway has only been fully characterised in the Gram-negative bacterium Klebsiella pneumoniae [ 5 - 11 ] and the Gram-positive bacterium Bacillus subtilis [ 12 - 14 ] Selected individual enzymes active in the pathway have been studied in a wide variety of eukaryotic and prokaryotic organisms [ 7 , 15 - 20 ]. For Mycobacterium spp ., only methionine adenosyltransferase has been cloned, expressed, and fully characterised [ 21 ]. The final step in Met regeneration is the transamination of ketomethiobutyrate (KMTB) by an aminotransferase. The specific aminotransferase responsible for the reaction has been identified and characterised in a number of microorganisms, including malaria, African trypanosomes, K. pneumoniae , B. subtilis , and B. anthracis [ 7 , 16 , 17 ]. In the lower eukaryotes Plasmodium falciparum , Trypanosoma brucei brucei , Giardia intestinalis , and Crithidia fasciculata , this reaction is catalysed by the subfamily Ia enzyme aspartate aminotransferase [ 17 ]. In K. pneumoniae , however, the reaction was performed by the close homologue tyrosine aminotransferase, which is also a member of subfamily Ia [ 7 ]. Gram-positive bacteria and archaea appear to lack any subfamily Ia homologues in their genomes, and B. subtilis , B. cereus , and B. anthracis were recently found to catalyse Met regeneration via a branched-chain amino acid aminotransferase (BCAT) [ 16 ]. This enzyme is a member of family III, along with D-amino acid aminotransferase (DAAT), and is unrelated structurally to family I enzymes [ 22 ]. Intriguingly, B. subtilis and B. cereus / B. anthracis utilised BCAT enzymes from separate subfamilies (IIIa vs. IIIb respectively). As Mycobacterium spp. also appear to have no subfamily Ia aminotransferase sequences ([ 16 ], and data not shown), it would be expected that M. tuberculosis also catalyses the conversion of KMTB to Met via a BCAT. In this paper, we report the identification, cloning, and functional expression of a single BCAT from M. tuberculosis . In addition, this enzyme has been demonstrated to actively catalyse Met formation and is subject to inhibition by a variety of aminooxy compounds. Results Branched-chain amino acid aminotransferase in M. tuberculosis The complete, published genome of M. tuberculosis H37Rv was found to contain a single gene with a very high sequence homology to either B. subtilis YbgE or YwaA, which are both known to be subfamily IIIa BCATs [ 16 , 23 ]. In contrast, the tuberculosis genome did not contain a homologue to B. subtilis YheM, B. cereus BCAT, or B. anthracis BCAT, which are all subfamily IIIb aminotransferases [ 16 ]. This relationship can be clearly seen in Figure 2A , where selected family III aminotransferases have been aligned and an unrooted tree constructed. The putative M. tuberculosis BCAT gene, Rv2210c, has not been previously cloned, expressed, or characterised. It is interesting to note that the M. tuberculosis genome contains a single BCAT homologue and no obvious DAAT homologue. Examination of complete and incomplete genome projects for Mycobacterium spp . uncovered a single gene in M. leprae , M. bovis , M. marinum , M. ulcerans , M. avium , and M. smegmatis with an extremely high identity to Rv2110c. Together, with other subfamily IIIa aminotransferases, the putative mycobacterial sequences were aligned and a cladogram constructed (Figure 2B ). The M. tuberculosis and M. bovis sequences were identical, as were the M. marinum and M. ulcerans sequences. Aside from M. bovis , all the mycobacterial BCAT sequences were found to be 85 – 88% identical to the M. tuberculosis sequence. However, the tuberculosis sequence was 57% identical to the putative BCAT from Streptomyces coelicolor and 45% identical to B. subtilis YbgE. Figure 2B highlights the fact that the mycobacterial BCAT sequences are more closely related to eukaryotic enzymes than to most other bacterial homologues. There was little sequence conservation with enzymes found in subfamily IIIb, with only 27% identity to the E. coli BCAT, 18% to the B. anthracis BCAT, and 15% to B. subtilis YheM. The low level of sequence conservation outside of the genus can be seen in the alignment of selected BCAT sequences shown in Figure 3 . Only 19 residues are completely conserved across even this small sequence sampling. Interestingly, of the residues found by X-ray crystallography to be important in substrate binding to the E. coli BCAT [ 24 ], only K228(K159) and T339(T257) were conserved across the 13 sequences in Figure 3 . The residues in parentheses represent the corresponding position in the E. coli BCAT. Of these two residues, K228(K159) is the PLP binding site and would be expected to be invariant. If one excludes the only DAAT in Figure 3 , then Y91(Y31), F96(36), Y233(164), and A340(A258) can be added to this conserved list of residues important for substrate binding in the E. coli BCAT. Clearly, sequence conservation is very low across family III. Expression and characterization of the branched-chain amino acid aminotransferase The putative M. tuberculosis BCAT was cloned as a deca-histidine fusion protein for expression in E. coli . To prevent complete inclusion of the recombinant protein, it was necessary to induce expression with a relatively low concentration of IPTG (0.1 mM) at 20°C for 20 hr. Under these conditions, sufficient soluble material was produced and purified over Ni 2+ affinity columns (Figure 4 ). Assay of the eluted material with 2 mM each of ADEFGHIKLNQRSTVWY and 1 mM KMTB resulted in appreciable Met production (data not shown), demonstrating that the enzyme was active and catalysed Met formation. The purified enzyme was screened against 2 mM of each individual amino acid and 1 mM KMTB to determine the amino donor range for Met regeneration. Isoleucine, leucine, and valine were found to be the most effective substrates (Figure 5 ), while glutamate and phenylalanine were also active as amino donors. Tyrosine and tryptophan were found to have a much lesser ability to transaminate KMTB and all other amino acids were inactive. The five most active amino donors were more closely examined in order to determine their kinetic parameters (Table 1 ). The Km for Leu, Ile, and Val ranged from 1.77 – 2.85 mM, while that for Glu was 9.53 mM and Phe 7.44 mM. The Vmax for all five amino acids was similar at 2.17 – 5.70 μmol/min/mg protein. KMTB was found to have a Km of 4.20 mM. The enzyme was also examined for branched-chain amino acid and KG aminotransfer in order characterise the "classic" reactions associated with a BCAT (Table 1 ). The Km of the substrates was found to be similar, while the Vmax ranged from 11.82 – 14.35 μmol/min/mg protein. Therefore, the tuberculosis BCAT catalyses aminotransfer of KG about 3 times more readily than KMTB. This result is similar to that seen with the B. subtilis BCAT, which also transaminates KG at a higher rate than KMTB [ 16 ]. Inhibition studies Thirteen aminooxy compounds were assayed for inhibitory effects on the tuberculosis BCAT. The enzyme was incubated with 2.0 mM leucine, 1.0 mM KMTB and 0.1 or 1.0 mM inhibitor to assay for the effect on Met regeneration (Figure 6 ). With the exception of O-trimethylsilylhydroxylamine, all of the compounds inhibited Met formation to some extent. The four most active compounds at 0.1 mM were O-allylhydroxylamine, carboxymethoxylamine, O-benzylhydroxylamine, and O-t-butylhydroxylamine, and these inhibitors were further examined in order to determine K i values (Table 2 ). For all four compounds, the inhibition data was not consistent with a simple competitive or uncompetitive model, but fit very well with a model of mixed mode inhibition [ 25 ]. The competitive component of inhibition yielded a K ic of 8.20 – 21.61 μM, while the uncompetitive component gave a K iu of 84.08 – 386 μM. Therefore, the inhibition of the tuberculosis BCAT by these four aminooxy compounds is primarily competitive. These four inhibitors and canaline, an aminooxy analogue of ornithine that has been demonstrated to be an effective aminotransferase inhibitor in other systems [ 16 , 17 , 26 - 28 ], were screened against M. tuberculosis and M. marinum in vitro to determine potential antimicrobial activity. M. marinum is a close relative of M. tuberculosis that causes a similar disease in fish, grows faster than M. tuberculosis in culture, and does not cause serious infections in humans [ 29 ]. As such, it is an excellent surrogate for the initial screening of antimycobacterial agents, and we wished to validate its use for aminooxy compounds. All the inhibitors were found to have some degree of antimycobacterial activity (Table 3 ), with MIC values ranging from 78 μM – 10 mM and IC 50 values of 8.49 μM – 467 μM. The best inhibitor was found to be O-allylhydroxylamine. While O-t-butylhydroxylamine and O-benzylhydroxylamine appeared to be the best enzyme inhibitors, they were significantly less effective than O-allylhydroxylamine as growth inhibitors. Unlike other organisms examined to date [ 27 , 30 ], canaline was not a particularly good inhibitor of both enzyme activity and cell growth. The inhibition results for M. tuberculosis and M. marinum were very similar, with MIC results being identical or within 1 dilution. In addition, M. tuberculosis was found to have an MIC of 2 μg/ml for streptomycin while M. marinum had one of 8 μg/ml. Discussion The specific aminotransferase involved in the formation of Met from KMTB has been examined in a number of eukaryotic and prokaryotic organisms [ 7 , 16 , 17 ]. However, within the low-GC content Gram-positive bacteria, only B. subtilis , B. cereus , and B. anthracis have been studied [ 16 ]. In all of these Bacillus spp. , a BCAT has been found to be responsible for catalysing the reaction, with B. subtilis and B. cereus / B. anthracis utilising enzymes from different aminotransferase subfamilies. No member of the high-GC content Gram-positive bacteria has been previously examined. Like B. subtilis , M. tuberculosis has been found to catalyse Met regeneration using a subfamily IIIa aminotransferase. In fact, the kinetic parameters for the two aminotransferases were almost identical. The M. tuberculosis BCAT had Km values of 1.77 – 2.85 mM and Vmax values of 2.58 – 4.28 μmol/min/mg protein for branched-chain amino acids and KMTB, while the B. subtilis YbgE had the corresponding values of 2.36 – 3.20 mM and 1.84 – 2.03 μmol/min/mg protein [ 16 ]. For branched-chain amino acids and KG, the values were 5.79 – 6.16 mM and 11.82 – 14.35 μmol/min/mg protein for the M. tuberculosis BCAT, and 2.82 – 3.99 mM and 13.93 – 16.61 μmol/min/mg protein for B. subtilis YbgE [ 16 ]. Therefore, a 45% sequence identity between the two enzymes is sufficient to conserve both the substrate range and kinetic properties of the BCATs. Structural information is only available for the E. coli BCAT (IlvE) and the human mitochondrial BCAT [ 24 , 24 , 31 ], but the key residues involved in substrate specificity appear to be conserved in the M. tuberculosis BCAT. However, while the human mitochondrial BCAT is also a family IIIa aminotransferase, there are some clear differences when compared to the M. tuberculosis enzyme. The human enzyme will not accept aromatic amino acids, whereas the tuberculosis BCAT would use phenylalanine as an amino donor. In addition, the human enzyme contains the redox-active motif CXXC at positions 311–314 (positions 341–344 in Figure 3 ) which is essential for maintaining activity, while the tuberculosis BCAT lacks these residues. Structural analysis of the M. tuberculosis and/or B. subtilis enzymes would clarify these issues. The M. tuberculosis BCAT was also screened with a variety of aminooxy compounds as potential inhibitors. These compounds are known aminotransferase inhibitors and act by forming a stable Schiff-base with the PLP cofactor [ 32 ]. Unlike previous studies [ 7 , 16 , 17 , 33 ], canaline was not found to be one of the better inhibitors of aminotransferase activity. Instead, O-benzylhydroxylamine, O-t-butylhydroxylamine, carboxymethoxylamine, and O-allylhydroxylamine were the most efficient inhibitors of Met formation from KMTB. In addition, these compounds demonstrated mixed type inhibition with a lower Ki for the competitive component. This result contrasts with that previously found for canaline with the Bacillus spp. enzymes, where inhibition was uncompetitive [ 16 ]. It may be possible that this difference may be due to the structure of the inhibitors, as canaline is a γ-substituted amino acid analogue, while the present inhibitors are α-substituted or non-amino acid analogues. Essentially, the inhibitors examined in this study do not present an α-amino group suitable for participation in the transamination reaction whereas canaline does. Further screening of the inhibitors against M. tuberculosis and M. marinum in vitro demonstrated that the compounds can act as effective antimycobacterial agents. The close correspondence of the MIC values for M. tuberculosis and M. marinum validates the use of the latter organism as a more rapid and safe initial screen of the antimycobacterial properties of aminooxy compounds. The MIC values found for streptomycin against these two organisms was also found to be consistent with previously published values [ 34 ]. M. marinum can thus be used to quickly test a larger number of potential inhibitors, with M. tuberculosis used as a follow up for more promising candidates. Interestingly, there was no direct correlation between the Ki of the compounds against recombinant M. tuberculosis BCAT and the MIC/IC 50 against cell growth. It is possible that there may be differences in the uptake rate of the various compounds into viable cells. Alternatively, the most effective growth inhibitors act by inhibiting other PLP-dependent enzymes in addition to BCAT. In any case, O-allylhydroxylamine was the most effective antimycobacterial agent with an MIC of 78 μM against M. marinum and 156 μM against M. tuberculosis . Unfortunately, the compound is corrosive, and is thus unsuitable for further in vivo study. However, the structure of the compound might provide the basis for the design of less toxic, more active structural analogues. In future studies, it will be necessary to examine the effect of potential inhibitors on human BCAT, in order to better assess the potential for host toxicity. Any further development of aminooxy compounds as antimycobacterial agents will depend on discovering a selective inhibitor for the microbial enzyme. Several older studies have been conducted on the antimicrobial effect of aminooxy compounds, with M. tuberculosis included amongst the organisms tested [ 35 - 39 ]. From these papers, the only compound in common with the present study was carboxymethoxylamine, which was found to have an MIC of 313 μM (present data), 910 μM [ 35 ], 170 – 686 μM [ 36 ], or 170 μM [ 37 ]. Given the variety of media used in these studies for determining the MIC value, the results are quite consistent. The variety of non-commercially available aminooxy compounds synthesized and tested in these older studies included aminooxy acids, aminooxy amides, aminooxy hydroxamic acids, aminooxy hydrazides, aminooxy alkanes, and aminooxy guanidines. Several of these compounds were very effective growth inhibitors in vitro, with MIC values as low as 0.30 μM against M. tuberculosis . One of the compounds has been administered to mice, with favourable, albeit sparsely detailed, results with regard to toxicity and in vivo antitubercular effect [ 38 ]. While it is unclear what effect these inhibitors would have against the M. tuberculosis BCAT, it would appear to be possible to design more effective, less toxic aminooxy compounds for use against M. tuberculosis . Several interesting findings arose during the course of this investigation. First, while M. tuberculosis has only the one branched-chain aminotransferase, it does contain a coding sequence (Rv0858c) with a high similarity to the B. subtilis ykrV gene product. YkrV was found to be a subfamily If aminotransferase and could also catalyse the conversion of KMTB to Met using glutamine as the only effective amino donor [ 16 ]. Therefore, it is possible that the Rv0858c gene product might be capable of KMTB transamination. It should be stressed that while the recombinant B. subtilis YkrV could transaminate KMTB with glutamine, B. subtilis cell homogenates did not produce Met from KMTB when supplemented with glutamine [ 16 ]. Similarly, cell homogenates of M. smegmatis grown in Middlebrook 7H9 incomplete medium were only able to produce Met from KMTB when supplemented with valine, isoleucine, leucine, glutamate, or phenylalanine, as was seen for the recombinant M. tuberculosis BCAT in figure 5 (data not shown). M. tuberculosis was found to contain no putative gene product with significant homology to a DAAT. In fact, the organism appeared to contain no subfamily IIIb aminotransferases. The physiological significance of a lack of a DAAT is unclear, but many organisms do not contain a homologue of this enzyme. With DAAT, there might be a diminished capacity to catabolise D-amino acids for energy, although the same reactions could be performed by a D-amino acid oxidase. M. tuberculosis is known to be reliant on carbohydrate catabolism during the active growth phase and lipid metabolism during the chronic, dormant phase [ 40 ]. Therefore, the lack of a DAAT might be reflective of a lifestyle where protein and peptide catabolism is relatively unimportant. Similarly, M. tuberculosis was found to lack clearly identifiable homologues of several enzymes in the Met regeneration pathway. The most glaring omission is the lack of an S-adenosylmethionine decarboxylase (SAMdc) homologue (see Figure 1 ). M. tuberculosis contains the preceding enzyme, methionine adenosyltransferase [ 21 ], and has an easily identifiable homologue for the succeeding enzyme, spermidine synthase [ 23 ]. Therefore, M. tuberculosis must catalyse SAMdc activity via another enzyme in order to be able to synthesize polyamines. A previous study has demonstrated SAMdc activity in M. bovis homogenates, but has not identified the enzyme responsible [ 41 ]. Resolution of this issue is critical for a more complete understanding of polyamine biosynthesis in tuberculosis, and may yield a novel enzyme as an additional drug target. The M. tuberculosis genome also appears to be missing homologues of the enzymes converting methylthioribose to KMTB (see Figure 1 ). However, outside of K. pneumoniae and B. subtilis , these enzymes have not been well studied, and, between these two organisms, there are key differences in the enzymes catalyzing several steps [ 12 , 20 ]. In silico analyses have suggested that Pseduomonas aeruginosa , Xylella fastidiosa , Leptospira interrogans , and Thermoanaerobacter tengcongensis have readily identifiable, complete Met regeneration pathways [ 42 ]. However, the presence or absence of the pathway in a variety of prokaryotic and eukaryotic organisms remains to be determined by functional analysis. Therefore, there is much left to examine before concluding that M. tuberculosis contains neither homologues nor analogues to these Met recycling enzymes. However, even in the absence of a complete Met salvage pathway, M. tuberculosis , as an intracellular pathogen, might utilise exogenous KMTB as a Met source. Conclusions Branched-chain amino acid aminotransferase has been cloned and characterised from M. tuberculosis . This enzyme was found to be responsible for the formation of methionine from ketomethiobutyrate, and could be inhibited in vitro by a series of aminooxy compounds. Several of these compounds were found to be effective inhibitors of M. tuberculosis or M. marinum growth in culture, with MIC values as low as 156 μM and 78 μM respectively. These studies demonstrate the importance branched-chain amino acid and methionine metabolism to the survival of mycobacteria, and open up the potential for the development of more potent and less toxic aminooxy inhibitors of the branched-chain aminotransferase. Methods Cells and reagents M. tuberculosis H37Rv and M. marinum Aronson (ATCC927) were cultured in liquid Middlebrook 7H9 complete medium or on Middlebrook 7H10 plates at 37°C for M. tuberculosis or 30°C for M. marinum . All substrates and inhibitors were obtained from Sigma-Aldrich (Oakville, ON, Canada). Cloning and functional expression Genomic DNA was isolated from M. tuberculosis by vortexing packed cells in a minimal volume of 50 mM Tris-HCl pH 8.0/10 mM EDTA/100 mM NaCl containing 500 μm acid washed glass beads (Sigma). After allowing the glass beads to settle, the supernatant was added to an equal volume of 10 mM Tris-HCl pH 8.0/100 mM NaCl/25 mM EDTA/0.5% w/v sodium dodecyl sulfate/0.1 mg/ml proteinase K and incubated for 1 hr at 37°C with occasional gentle mixing. The mixture was then subjected to extraction with phenol:chloroform:isoamyl alcohol (25:24:1), and the DNA ethanol precipitated. The sequence of the putative M. tuberculosis BCAT gene was discovered by a BLAST search of the complete M. tuberculosis H37Rv genome using the B. subtilis YbgE, YwaA, or YheM gene products as the query proteins [ 16 , 23 , 43 ]. The single resulting putative BCAT gene was used to construct oligonucleotide primers for PCR amplification. The 5' primer was TCGAGGCGGCCGCAAATGACCAGCGGCTCCCTTCA and incorporated a NotI restriction site and an in-frame start codon. The 3' primer was ATCGAGCTCGAGTTACCCCAGCCGCGCCATCCAG and incorporated a XhoI restriction site and an in-frame stop codon. The BCAT gene was then amplified using a 5:1 mixture of Taq:Pfu polymerases (Promega; Madison, WI, USA) and the following program: 1 cycle of 95°C for 1.5 min; 30 cycles of 95°C for 1 min, 55°C for 1 min, and 72°C for 1 min; and 1 cycle of 72°C for 10 min. The resulting PCR product was excised from a 1% agarose gel and recovered using the Qiaex II kit (Qiagen; Mississauga, ON, Canada). The purified product was digested with NotI and XhoI and ligated into a similarly digested pET 19 m (a modification by us of pET19b (Novagen; Madison, WI, USA) to incorporate extra restriction sites in the multiple cloning site) using a Rapid Ligation kit (Fermentas; Burlington, ON, Canada). The recombinant plasmid was then transformed into Escherichia coli XL10 cells (Stratagene; La Jolla, CA, USA) and was subsequently recovered using the Qiaspin miniprep kit (Qiagen). Positive clones were determined by digesting the plasmid with NotI and XhoI to confirm the presence of the insert on a 1% agarose gel. The sequence of the insert was confirmed by using the Big-Dye cycle sequencing kit (ABI; Foster City, CA, USA) and an ABI Prism 310 genetic analyser. The plasmid from positive clones was transformed into E. coli BL21(DE3) CodonPlus RIL cells (Stratagene) for functional expression. Cells were grown in liquid LB medium containing 50 μg/ml ampicillin and 50 μg/ml chloramphenicol at 37°C and 250 rpm until the culture reached an A 600 nm of 0.6–0.8. The culture was then cooled to 20°C for 30 min at 250 rpm before the addition of 0.1 mM isopropylthiogalactopyranoside (IPTG) and an additional 20 hr of incubation at 20°C and 250 rpm. The culture was centrifuged at 3500 × g for 20 min at 4°C, and the cell pellet resuspended in 50 mM HEPES (pH 7.4)/750 mM NaCl and frozen at -20°C. The resuspended cells were then thawed, sonicated on ice, centrifuged at 3000 × g for 20 min at 4°C, and the supernatant loaded onto a 1.6 × 9.5 Chelating-Sepharose-FF column (Amersham Biosciences; Baie d'Urfe, QC, Canada) charged with NiSO 4 . The column was washed with 50 mM HEPES (pH 7.4)/750 mM NaCl and 50 mM HEPES (pH 7.4)/750 mM NaCl/80 mM imidazole, before elution with 50 mM HEPES (pH 7.4)/750 mM NaCl/800 mM imidazole. Fractions containing the recombinant protein were pooled and concentrated to less than 3.0 ml using a 30 kDa molecular mass cut-off filter (Pall Filtron; Mississauga, ON, Canada). The concentrated enzyme was then dialysed against 50 mM HEPES (pH 7.4)/1 mM dithiothreitol/1 mM EDTA/trace pyridoxal-5-phosphate (PLP) overnight at 4°C. The concentrated enzymes were stored at 4°C for several days, or with 20% v/v glycerol at -20°C for several weeks, without appreciable loss of activity. Recombinant protein samples were examined by electrophoresis on 10% SDS polyacrylamide gels followed by Coomassie Brilliant Blue R250 staining. Protein concentration was measured using the Bio-Rad reagent (Bio-Rad; Mississauga, ON, Canada). Enzyme assays and inhibition studies Aminotransferase activities were assayed by an HPLC method [ 17 ]. 5 or 10 μl of recombinant enzyme was added to 100 μl of substrate mix (100 mM PO 4 (pH 7.4)/50 μM PLP/various concentrations of amino acid/various concentrations of keto acid) and incubated for 30 min at 37°C. The samples were then stored at -20°C until analysis by HPLC. All samples were analysed by pre-column derivatisation and reverse-phase HPLC. 10 μl of sample was mixed with 50 μl of 400 mM borate pH 10.5 and then with 10 μl of 10 mg/ml o-phthalaldehyde/12 μl/ml mercaptopropionate/400 mM borate pH 10.5 prior to the injection of 7.0 μl onto a 2.1 × 200 mm ODS-AA column (Agilent; Mississauga, ON, Canada). The column was eluted using 2.72 mg/ml sodium acetate pH 7.2/0.018% v/v triethylamine/0.3% v/v tetrahydrofuran as Buffer A and 2.72 mg/ml sodium acetate pH 7.2/40% v/v methanol/40% v/v acetonitrile as Buffer B with a linear gradient of 0 – 17% B over 16 min followed by a linear gradient of 17–100% B over 1 min and 6.0 min at 100% B. The flow rate was 0.45 ml/min from 0 – 16 min and 0.80 ml/min from 17–30 min. The elution of derivatised amino acids was monitored at 338 nm and fluorometrically with an excitation of 338 nm and an emission of 450 nm. All separations were performed on an Agilent 1100 HPLC equipped with an autosampler, variable wavelength ultraviolet/visible spectrophotometric detector, fluorescence detector, and Chemstation operating system. The amino donor range for Met regeneration was determined by incubating 2 mM of each individual amino acid and 1 mM KMTB, followed by HPLC for Met quantification. Amino acids which were effective amino donors were further studied at 0.1 – 10 mM amino acid and 10 mM KMTB to determine the kinetic constants. Similar assays were performed with 0.1 – 10 mM KMTB and 10 mM Leu. Replacement of KMTB with ketoglutarate (KG) in these experiments and subsequent HPLC analysis of Glu formation allowed for the determination of BCAT activity. The apparent Km and Vmax values for each substrate were assessed by non-linear curve fitting using the Scientist software programmed with the Michaelis-Menton equation (Micromath; Salt Lake City, UT, USA). Initial inhibition studies screened 13 aminooxy compounds against M. tuberculosis BCAT using 2.0 mM Leu/1.0 mM KMTB/0.1 or 1.0 mM inhibitor in the enzyme incubation. Inhibitors which demonstrated better than 50% reduction of activity at the 0.1 mM concentration were further studied for the determination of K i values. These reactions involved 0.5, 1.0, 2.0, or 3.0 mM Leu and 1.0 mM KMTB in the reaction mixture together with 0, 25, 50, 75, 100, 150, 200 μM of inhibitor. The K i values were determined by non-linear curve fitting with the Scientist software programmed with competitive, uncompetitive, and mixed inhibition equations [ 25 ]. In vitro growth inhibition studies were performed on M. tuberculosis and M. marinum using the most effective enzyme inhibitors. Cultures at mid-log growth in Middlebrook 7H9 complete medium was diluted to 2 × 10 5 cfu/ml and 100 μl added to 96 well microtitre plates containing 100 μl of doubling dilutions of each inhibitor. The final inhibitor concentration ranged from 10 mM – 298 pM. Positive and negative controls consisted of 100 μl Middlebrook 7H9 medium replacing the inhibitor or cells respectively. The plates were incubated at 30°C for 8 days ( M. marinum ) or 37°C for 14 days ( M. tuberculosis ) with no agitation before measurement of growth at A 650 nm using a Molecular Devices 96-well spectrophotometer (Sunnyvale, CA, USA). The MIC was determined as the lowest dilution that completely prevented microbial growth and the IC 50 was determined by non-linear curve fitting with the Scientist software programmed with the Chou equation [ 44 ]. Phylogenetic analysis Additional BCAT and DAAT sequences were obtained from GenBank [ 38 ]. Mycobacterium spp. BCAT sequences from preliminary genome projects were made available from The Institute for Genomic Research for M. smegmatis and M. avium , from The Sanger Centre for M. marinum , and from The Institut Pasteur for M. ulcerans . These sequences were aligned using the Clustal algorithm and the BLOSUM sequence substitution table in the ClustalX program [ 46 ]. Aligned sequences were viewed using the Bioedit program [ 47 ] and were then used with the ProtDist component of the PHYLIP [ 48 ] to construct a distance matrix that was the basis for tree construction using the neighbour-joining method [ 49 ]. All trees were visualised using Treeview [ 50 ]. Authors' contributions ESV performed the cloning, expression, and characterisation of the enzyme, and assisted in writing the manuscript. CLR assisted in the cloning and expression experiments. MHK performed the M. marinum experiments. BJB conceived the study, performed the M. tuberculosis experiments, and wrote the manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524486.xml
406394	Planting the Seeds of a New Paradigm	RNA-mediated gene silencing has emerged in recent years as an important mechanism for regulating gene expression. Some of the key discoveries have been made in plants	Although the word ‘revolution’ should not be used lightly in science, there is no other way to describe the recent explosion in our awareness and understanding of RNA-mediated gene silencing pathways. The central player in RNA-mediated gene silencing is a double-stranded RNA (dsRNA) that is chopped into tiny RNAs by the enzyme Dicer. The tiny RNAs associate with various silencing effector complexes and attach to homologous target sequences (RNA or DNA) by basepairing. Depending on the protein composition of the effector complex and the nature of the target sequence, the outcome can be either mRNA degradation, translational repression, or genome modification, all of which silence gene expression ( Figure 1 ). Present in plants, animals, and many fungi, RNA-mediated gene silencing pathways have essential roles in development, chromosome structure, and virus resistance. Although the mechanistic details are still under investigation, RNA-mediated silencing has already provided a powerful tool for studying gene function and spawned a fledgling industry that aims to develop novel RNA-based therapeutics to treat human diseases ( Robinson 2004 ). Figure 1 RNA-Mediated Silencing Short RNAs derived from Dicer cleavage of dsRNA are incorporated into multiprotein effector complexes, such as RISC and RITS (RNA-induced initiation of TGS) ( Verdel et al. 2004 ) to target mRNA degradation (RNAi/PTGS), translation inhibition, or TGS and genome modifications. ARGONAUTE (AGO) proteins (the name comes from a plant mutant [ Bohmert et al. 1998 ]) bind short RNAs and ‘shepherd’ them to appropriate effector complexes ( Carmell et al. 2002 ). siRNAs originate from perfect RNA duplexes, which can be produced by RDR activity on ssRNA templates; miRNAs originate from imperfect RNA hairpins that are encoded in intergenic regions of plant and animal genomes. Functions are shown at the bottom. In addition to roles in transgene silencing, both TGS and RNAi/PTGS control genome parasites called transposons ( Flavell 1994 ; Plasterk 2002 ). Genome modifications (DNA and histone methylation) can potentially be targeted by short RNAs that basepair to DNA or to nascent RNA synthesized from the target gene ( Grewal and Moazed 2003 ). Target nucleic acids are shown in blue, short RNAs in red, proteins and enzyme complexes as ovals. Many biologists first learned of RNA-mediated gene silencing in 1998 following the discovery, in the nematode worm Caenorhabditis elegans ( Fire et al. 1998 ), of a process called RNA interference (RNAi), in which dsRNA triggers sequence-specific mRNA degradation. The roots of RNA-mediated silencing, however, can be traced back 15 years, when a handful of botanical labs stumbled across strange cases of gene silencing in transgenic plants. To highlight the many seminal contributions of plant scientists to the field, we offer here a personal perspective on the origins and history of RNA-mediated gene silencing in plants. Early Silencing Phenomena Starting in the late 1980s, biologists working with transgenic plants found themselves confronted with a ‘bewildering array’ of unanticipated gene silencing phenomena ( Martienssen and Richards 1995 ). Most intriguing were cases in which silencing seemed to be triggered by DNA or RNA sequence interactions, which could occur between two separate transgenes that shared sequence homology or between a transgene and homologous plant gene. Several early examples supplied the prototypes for two types of RNA-mediated gene silencing that are recognized today. In one type, silencing results from a block in mRNA synthesis (transcriptional gene silencing [TGS]); in the second type, silencing results from mRNA degradation (posttranscriptional gene silencing [PTGS]) ( Figure 1 ). TGS was revealed when two different transgene complexes were introduced in sequential steps into the tobacco genome. Each complex encoded different proteins, but contained identical gene regulatory regions (promoters). Unexpectedly, the first transgene complex, which was stably active on its own, often became silenced in the presence of the second ( Figure 2 ). The promoters of the silenced transgenes acquired DNA methylation, a genome modification frequently associated with silencing. Silencing and methylation were reversed when the transgene complexes segregated from each other in progeny, suggesting that interactions between the common promoter regions triggered silencing and methylation ( Matzke et al. 1989 ; Park et al. 1996 ). Figure 2 Early Examples of Gene Silencing in Transgenic Plants TGS: Normally when two plants harboring separate transgenes encoding resistance to kanamycin (kan) or hygromycin (hyg), respectively, are crossed, 50% of the progeny are resistant to the individual antibiotics and 25% are resistant to a combination of both (top). In cases of silencing, expression of the KAN marker is extinguished in the presence of the HYG marker, as indicated by only 25% kan resistance and no double resistance (middle). PTGS: Transformation of wild-type petunia (bottom left) with a transgene encoding a pigment protein can lead to loss of pigment (white areas) owing to cosuppression of the transgene and homologous endogenous plant gene. (Photos on the left and in the middle were provided by Jan Kooter and on the right were provided by Natalie Doetsch and Rich Jorgensen.) PTGS was discovered in two ways. One involved experiments to evaluate antisense suppression, a promising approach at the time for selectively silencing plant gene expression. In theory, antisense RNA encoded by a transgene should basepair to the complementary mRNA of a plant gene, preventing its translation into protein. Although the control ‘sense’ transgene RNAs are unable to basepair to mRNA and hence should not induce silencing, they often inexplicably did ( Smith et al. 1990 ). In another type of experiment, efforts to enhance floral coloration in petunia by overexpressing a transgene encoding a protein involved in pigment synthesis led paradoxically to partial or complete loss of color ( Figure 2 ). This resulted from coordinate silencing (‘cosuppression’) of both the transgene and the homologous plant gene ( Napoli et al. 1990 ; Van der Krol et al. 1990 ), later shown to occur at the posttranscriptional level ( De Carvalho et al. 1992 ; Van Blokland et al. 1994 ) A related phenomenon, called quelling, was observed in the filamentous fungus Neurospora crassa ( Romano and Macino 1992 ). Similarly to TGS, PTGS was often associated with DNA methylation of transgene sequences ( Ingelbrecht et al. 1994 ). Two influential papers appeared in the early 1990s. One reported the discovery of RNA-directed DNA methylation in transgenic tobacco plants ( Wassenegger et al. 1994 ). This was the earliest demonstration of RNA-induced modification of DNA, a process that we return to below. A second study showed that plant RNA viruses could be both initiators and targets of PTGS. Plants expressing a transgene encoding a truncated viral coat protein became resistant to the corresponding virus, a state achieved by mutual degradation of viral RNA and transgene mRNA ( Lindbo et al. 1993 ). In addition to forging a link between RNA virus resistance and PTGS, this study included a remarkably prescient model for PTGS that featured an RNA-dependent RNA polymerase (RDR), small RNAs, and dsRNA, all of which were later found to be important for the RNAi. PTGS was subsequently shown in 1997 to protect plants naturally from virus infection ( Covey et al. 1997 ; Ratcliff et al. 1997 ). Transgene PTGS thus tapped into a preexisting natural mechanism for combating viruses. To recap: by 1998—the year in which RNAi was reported—plant scientists had documented sequence-specific RNA degradation (PTGS), sequence-specific DNA methylation that triggered TGS, and RNA-directed DNA methylation. They had also proposed models for PTGS involving dsRNA ( Lindbo et al. 1993 ; Metzlaff et al. 1997 ), small RNAs, and RDR ( Lindbo et al. 1993 ). RNAi RNAi was discovered in experiments designed to compare the silencing activity of single-stranded RNAs (ssRNAs) (antisense or sense) with their dsRNA hybrid. While only marginal silencing of a target gene was achieved after injecting worms with the individual strands, injection of a sense–antisense mixture resulted in potent and specific silencing ( Fire et al. 1998 ). This unequivocally fingered dsRNA as the trigger of silencing. Shortly thereafter, dsRNA was shown to provoke gene silencing in other organisms, including plants ( Waterhouse et al. 1998 ). Indeed, the relatedness of RNAi, PTGS, and quelling was confirmed when genetic analyses in worms, plants, and Neurospora identified common components in the respective silencing pathways ( Denli and Hannon 2003 ). This included the aforementioned RDR, which can synthesize dsRNA from ssRNA templates (see Figure 1 ). PTGS is now accepted as the plant equivalent of RNAi. The discovery of RNAi established a requirement for dsRNA in silencing, but details of the mechanism remained unclear. In 1999, plant scientists studying PTGS provided a crucial clue when they detected small (approximately 25 nucleotide-long) RNAs corresponding to silenced target genes in transgenic plants ( Hamilton and Baulcombe 1999 ). They proposed that the small RNAs provided the all-important specificity determinant for silencing. Consistent with this, a rapid succession of studies in Drosophila systems demonstrated that 21–23 nucleotide ‘short interfering'RNAs (siRNAs), derived from cutting longer dsRNA, can guide mRNA cleavage ( Zamore et al. 2000 ; Elbashir et al. 2001 ); identified RISC (RNA-induced silencing complex), a nuclease that associates with small RNAs and executes target mRNA cleavage ( Hammond et al. 2000 ); and identified Dicer, the enzyme that chops dsRNA into short RNAs ( Bernstein et al. 2001 ) (see Figure 1 ). RNAi/PTGS was detected originally in experiments involving transgenes, injected RNAs, or viruses. Did the RNAi machinery also generate small RNAs for host gene regulation? Strikingly, the newly discovered siRNAs were the same size as several ‘small temporal’ RNAs, first identified in 1993 as important regulators of developmental timing in worms ( Lee et al. 1993 ; Reinhart et al. 2000 ). Everything came together in 2001 when heroic cloning efforts unearthed dozens of natural small RNAs 21–25 nucleotides in length, first from worms and flies and later from plants and mammals ( Lai 2003 ; Bartel 2004 ). Similar to siRNAs, the natural small RNAs, dubbed microRNAs (miRNAs), arise from Dicer processing of dsRNA precursors and are incorporated into RISC ( Denli and Hannon 2003 ). In many cases, miRNAs effect silencing by basepairing to the 3′ ends of target mRNAs and repressing translation (see Figure 1 ). miRNAs are now recognized as key regulators of plant and animal development. Identifying their target genes and full range of action are areas of intense research ( Lai 2003 ; Bartel 2004 ). Up until 2002, RNAi/PTGS and miRNAs were the most avidly studied aspects of RNA-mediated gene silencing. The next major advance, however, abruptly turned attention back to RNA-guided modifications of the genome. By 2001, plant scientists working on RNA-directed DNA methylation and TGS had demonstrated a requirement for dsRNAs that are processed to short RNAs, reinforcing a mechanistic link to PTGS ( Mette et al. 2000 ; Sijen et al. 2001 ). This established the principle of RNA-guided genome modifications, but the generality of this process was uncertain because not all organisms methylate their DNA. Widespread acceptance came with the discovery in 2002 of RNAimediated heterchromatin assembly in fission yeast ( Hall et al. 2002 ; Volpe et al. 2002 ). This silencing pathway uses short RNAs produced by Dicer and other RNAi components to direct methylation of DNA-associated proteins (histones), thus generating condensed, transcriptionally silent chromosome regions (heterochromatin) (see Figure 1 ). Targets of this pathway include centromeres, which are essential for normal chromosome segregation. The RNAi-dependent heterochromatin pathway has been found in plants ( Zilberman et al. 2003 ) and Drosophila ( Pal-Bhadra et al. 2004 ) and likely represents a general means for creating condensed, silent chromosome domains. More Lessons from Plants Plant scientists can chalk up other ‘firsts’ in RNA-mediated gene silencing. Systemic silencing, in which a silencing signal (short RNA or dsRNA) moves from cell to cell and through the vascular system to induce silencing at distant sites, was initially detected in plants in 1997 ( Palauqui et al. 1997 ; Voinnet and Baulcombe 1997 ) and later in worms ( Fire et al. 1998 ), although not yet in Drosophila or mammals. Viral proteins that suppress silencing by disarming the PTGS-based antiviral defense mechanism were discovered by plant virologists in 1998 ( Anandalakshmi et al. 1998 ; Béclin et al. 1998 ; Brigneti et al. 1998 ; Kasschau and Carrington 1998 ). One of these, the p19 protein of tombusviruses, acts as a size-selective caliper to sequester short RNAs from the silencing machinery ( Vargason et al. 2003 ). A recent study suggests that animal viruses encode suppressors of RNA-mediated silencing ( Li et al. 2004 ). Although RNA-mediated gene silencing pathways are evolutionarily conserved, there are various elaborations in different organisms. For example, the plant Arabidopsis has four Dicer-like (DCL) proteins, in contrast to mammals and worms, whose genomes encode only one Dicer protein ( Schauer et al. 2002 ). The RDR family has also expanded in Arabidopsis to include at least three active members. An important goal has been to determine the functions of individual family members. Previous studies in Arabidopsis have shown that DCL1 is needed for processing miRNA precursors important for plant development ( Park et al. 2002 ; Reinhart et al. 2002 ), but not for siRNAs active in RNAi ( Finnegan et al. 2003 ). The paper by Xie et al. (2004) in this issue of PLoS Biology delineates distinct functions for DCL2, DCL3, and RDR2. Nuclear-localized DCL3 acts with RDR2 to generate short RNAs that elicit DNA and histone modifications; DCL2 produces short RNAs active in antiviral defense in the cytoplasm of cells. This study illustrates nicely how RNA silencing components have diversified in plants to carry out specialized functions. By identifying small RNAs as agents of gene silencing that act at multiple levels throughout the cell, molecular biologists have created a new paradigm for eukaryotic gene regulation. Plant scientists have figured prominently in RNA-mediated silencing research. Instrumental to their success was the early ability to produce large numbers of transgenic plants, which displayed a rich variety of gene silencing phenomena that were amenable to analysis. The agricultural biotechnology industry provided incentives to find ways to stabilize transgene expression and use transgenic approaches to modulate plant gene expression and to genetically engineer virus resistance. As exemplified by the petunia cosuppression experiments, nonessential plant pigments provide conspicuous visual markers that vividly reveal gene silencing. The history of gene silencing research shows once again that plants offer outstanding experimental systems for elucidating general biological principles.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC406394.xml
523841	Oats Intolerance in Celiac Disease	null	Most patients with celiac disease can eliminate their symptoms—at a price: life-long adherence to a gluten-free diet. This means no wheat, rye, barley, and, until recently, no oats. Then some recent studies suggested that oats did not cause the intestinal inflammation characteristic of the disease, and thus oats are now often included in the celiac disease diet. This is good news for patients coping with severe restrictions on what they can and must not eat, but a study by Ludvig Sollid and colleagues in this issue of PLoS Medicine suggests that oats are not safe in all cases. The celiac diet excludes many cereal products (Photo: National Cancer Institute) Like other chronic inflammatory diseases, celiac disease is caused by a complex interplay between genetic and environmental factors, but it is better understood than most. Long believed to be a relatively rare disorder, it is now thought to affect about one in 250 people worldwide. Clinical symptoms are present in less than half of patients and vary considerably. Genetically, almost all patients have one of two predisposing HLA molecules, which determine the context in which their immune system encounters foreign antigens, including gluten proteins found in wheat and other cereals. In individuals with celiac disease, the immune system mounts an abnormal response to gluten, which is characterized by gluten-reactive intestinal T cells and by inflammation and compromised function of the small intestine. Ludvig Sollid and colleagues applied the current understanding of celiac disease and a range of molecular pathology tools to studying the response to oats of nine patients with celiac disease. The nine patients were not a random sample: all of them had been eating oats, and four of them had shown clinical symptoms after oats ingestion. The goal of the study was to characterize the intestinal T cell response to oats in these patients, and to relate it to clinical symptoms and intestinal biopsy results. All patients were on a gluten-free diet and ate oats that were free of contamination by other cereals. Three of the four patients who had reported problems after eating oats showed intestinal inflammation typical of celiac disease, and Sollid and colleagues studied intestinal T cells from these three patients. Two of the five patients who seemed to tolerate oats also had oats-reactive intestinal T cells. Functional study of these T cells showed that they were restricted to celiac-disease-associated HLA molecules and that they recognized two peptides derived from oat avenin that are very similar to peptides of gluten. Taken together, the findings show that intolerance to oats exists at least in some patients with celiac disease, and that those patients have the same molecular reaction to oats that other patients have to wheat, barley, or rye. However, identical reactions were also seen in two of the patients who were clinically tolerant to oats. The authors suggest that these reactions could develop into symptomatic disease after some time delay, but there is no proof that the presence of oats-reactive T cells is an indicator of future symptoms or even of enhanced susceptibility to clinical oats intolerance. Oats are not safe for all patients with celiac disease, but future studies are needed to determine the frequency of oats intolerance.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC523841.xml
314483	Science and Technology Communication for Development	The Science and Development Network (SciDev.Net) is a source of online news and analysis about the role of science and technology in meeting the needs of the developing world	Developing countries today face a wide range of needs, from more secure food supplies to cheap and effective medicines. One factor that almost all these needs have in common is that satisfying them adequately will not occur without the use of modern science. These same countries also face a range of political dilemmas—such as whether to accept the growing of genetically modified crops or how to adapt to the impact of climate change—that also require a knowledge of science, albeit of a slightly different nature. From both points of view, development can be characterised as the process of putting scientific and technical knowledge into practice. Conversely, it is important that building the capacity to absorb and make use of scientific and technical knowledge must be placed at the heart of the development aid efforts if these are to be successful in achieving their goals. But knowledge will not reach those who can benefit from it unless it has been effectively communicated to individuals with the power and skills to put it into practice, whether those are government officials and decision-makers, community groups and their representatives, or even nongovernmental organisations. Formal education, of course, has a key function here. But so does the informal education provided through the media. Furthermore, information and communications technologies (ICTs) have an important role to play in this process by reducing or eliminating the transactional (nonproduction) costs of communicating knowledge about science and technology. At the same time, it is important that the practice of science communication reflects the fact that it takes place in social context. In other words, it is not just a question of conveying information, but also of engaging the potential users of that information. The need is to encourage dialogue and eventually to empower those to whom the information is being provided so that this information can be applied in a practical and useful way. SciDev.Net It was with this in mind that the Science and Development Network (SciDev.Net) was launched in December 2001 as a source, through its Web site ( www.scidev.net ), of online news and analysis about the role of science and technology in meeting the needs of the developing world. Much of the material we use is taken from the science journals Nature and Science , both of which provide us with free access for up to four articles each week, the selection being based on a decision about which articles—ranging from news items or editorials to full scientific papers—are directly relevant to the needs of developing countries. In addition, other news articles are contributed by staff writers and a growing team of correspondents, including science journalists in South Africa, India, Tanzania, Brazil, Colombia, and China. We also summarise and link to relevant news stories, feature items, and opinion articles from media outlets around the world. An equally important part of the Web site are our dossiers. Dossiers are comprehensive, online guides to topical issues at the interface of science, technology, and society. Each dossier is compiled using advice from an international panel of experts and provides authoritative insight into a topic at the heart of international debate. So far, we have produced six dossiers on the following topics: intellectual property rights, climate change, indigenous knowledge, genetically modified crops, the ethics of biomedical research, and the brain drain. Those currently in the works focus on HIV/AIDS, research policy, genomics, and biodiversity. Target Audience Our target audience is not easy to define, as it is made up of individuals from many different groups, from university researchers and teachers to government officials and aid agency staff. In broad terms, however, it can be taken to include all those with a professional or personal interest in the interactions between science, technology, and development. The number of registrants has risen steadily since the Web site was launched in December 2001 and now stands at more than 7,000. About half of our registered users come from the developing countries themselves. Of these, 43% come from Latin America—a reflection of various factors, including the size of its scientific community and its relatively high level of Internet connectivity. Sub-Saharan Africa comprise about 25%, and South Asia (mainly India) makes up 20%. For those with an interest in science activities in a particular geographical region, we link together coverage into so-called regional gateways on the Web site. At present, there are four of these—covering Latin America, the Middle East, South and East Asia, and sub-Saharan Africa. Each gateway therefore has a strong regional identity, and in some cases this is reinforced by the translation of material—either in full text or as a summary—into local languages. The Latin American gateway, for example, carries three languages, and material can appear in one, two, or all three of these, with summaries in the others. Channels of Communication Reflecting a commitment to the idea that scientific knowledge will only be effectively communicated if individuals are adequately trained to carry out this task, we are engaged not only in presenting information on our Web site, but also in helping to generate a capacity within developing countries to improve channels of communication for this knowledge. One way in which we do this is through workshops on science communication. Some of these have a broad focus, seeking to help stimulate a national or regional debate on the importance of science communication, how it can be best carried out, and the challenges it faces. Such, for example, was the goal of two roundtables on the theme of “science communication and sustainable development” organised in August 2002 during the World Summit on Sustainable Development in Johannesburg, South Africa. Similarly, a meeting launching our activities in Latin America was held in Sao Paulo in May this year under the title “Science Communication and Development in Latin America.” Other events to encourage scientific communication have been more specific. In particular, we organised (jointly with the United Nations Educational, Scientific, and Cultural Organisation) a five-day workshop for African women communicators in Kampala, Uganda, in April 2003 on the use of ICTs to report on the science of HIV/AIDS ( Figure 1 ). A similar event took place in Chennai, India, in November 2003, with participants from seven Asian nations. Figure 1 A Workshop on the Use of ICTs to Report on the Science of HIV/AIDS This workshop was held in Kampala, Uganda, in April 2003. (Photograph: SciDev.Net.) Partnerships Building the communication channels that will allow science to be fully integrated into the development process at all levels—from governments down to communities—is a major challenge that involves many participants. SciDev.Net does not pretend to be more than one actor and is keen to establish partnerships with others in this field, convinced both of the synergies that can be gained through collaboration and of the contribution that ICTs can make to this (for example, by sharing information and links between Web sites). One of the major challenges that we face is to build up our role as a platform for the voice of the developing world on the issues that we cover. We already do this to a certain degree through our correspondents, our regional networks of contributors, and the participation of scientists from developing countries who are on the advisory panels for our dossiers. But we are very aware that more effort is needed. Despite this, we already feel that we are beginning to make a mark. A staff member of the United States' National Academy of Sciences recently informed us: “We hosted a group of visitors from seven African scientific academies here in Washington last week, and they spoke highly of the SciDev project … you are becoming a truly global newsletter!”	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC314483.xml
515303	Dialysis Disequilibrium Syndrome: Brain death following hemodialysis for metabolic acidosis and acute renal failure – A case report	Background Dialysis disequilibrium syndrome (DDS) is the clinical phenomenon of acute neurologic symptoms attributed to cerebral edema that occurs during or following intermittent hemodialysis (HD). We describe a case of DDS-induced cerebral edema that resulted in irreversible brain injury and death following acute HD and review the relevant literature of the association of DDS and HD. Case Presentation A 22-year-old male with obstructive uropathy presented to hospital with severe sepsis syndrome secondary to pneumonia. Laboratory investigations included a pH of 6.95, PaCO2 10 mmHg, HCO3 2 mmol/L, serum sodium 132 mmol/L, serum osmolality 330 mosmol/kg, and urea 130 mg/dL (46.7 mmol/L). Diagnostic imaging demonstrated multifocal pneumonia, bilateral hydronephrosis and bladder wall thickening. During HD the patient became progressively obtunded. Repeat laboratory investigations showed pH 7.36, HCO3 19 mmol/L, potassium 1.8 mmol/L, and urea 38.4 mg/dL (13.7 mmol/L) (urea-reduction-ratio 71%). Following HD, spontaneous movements were absent with no pupillary or brainstem reflexes. Head CT-scan showed diffuse cerebral edema with effacement of basal cisterns and generalized loss of gray-white differentiation. Brain death was declared. Conclusions Death is a rare consequence of DDS in adults following HD. Several features may have predisposed this patient to DDS including: central nervous system adaptations from chronic kidney disease with efficient serum urea removal and correction of serum hyperosmolality; severe cerebral intracellular acidosis; relative hypercapnea; and post-HD hemodynamic instability with compounded cerebral ischemia.	Background Acute renal failure requiring hemodialysis (HD) is a common clinical problem in critically ill patients that is independently associated with increased mortality[ 1 ]. Dialysis disequilibrium syndrome (DDS) is the clinical phenomenon of acute central nervous system dysfunction attributed to cerebral edema that occurs during or following HD. The precise epidemiology of DDS is poorly defined[ 2 ]. Review of MEDLINE (January 1966 – March 2004) suggested that DDS in critically ill patients has rarely been reported[ 3 , 4 ]. We report a case of DDS-induced cerebral edema that resulted in irreversible brain injury and death following acute HD. Further, we review the relevant literature of the association of DDS and HD. Case presentation A 22-year-old homosexual male presented to hospital with progressive dyspnea, productive cough, generalized malaise and fever. He had a known history of intravenous cocaine abuse and recent serology in prior 3 months was negative for human immunodeficiency virus (HIV). Results of a physical examination showed signs of tachypnea, tachycardia, accessory muscle use and left lung base crackles. Tympanic temperature was 34.7°C. The remainder of the examination was unremarkable except for urethral meatus stenosis. Initial laboratory investigations are presented in Table 1 . Arterial blood gases showed pH of 6.95, PaCO 2 10 mmHg, PaO 2 109 mmHg, HCO 3 2 mmol/L, and lactate 0.6 mmol/L consistent with high anion gap metabolic acidosis with respiratory compensation. Serum creatinine and blood urea nitrogen were 587 μmol/L and 46.7 mmol/L, respectively. Toxicology and drug screen was negative. The metabolic acidosis was partially accounted for by acute renal failure with retained unmeasured anions and ketonemia. Urinalysis showed pyuria. Electrocardiogram (ECG) showed normal sinus rhythm. Table 1 Laboratory values at prior to and following initiation of hemodialysis in the intensive care unit. Laboratory test Pre-dialysis Value Post-dialysis Values Reference range Hemoglobin 96 78 137–180 g/L White cell count 25.8 16.1 4.0–11.0 × 10 9 /L Band count 3.1 - 0.0–1.3 × 10 9 /L Platelets 603 486 150–400 × 10 9 /L Sodium 132 132 133–145 mmol/L Potassium 3.1 1.8 3.5–5.0 mmol/L Chloride 107 93 98–111 mmol/L Bicarbonate 2 19 21–31 mmol/L Glucose 6.3 9.0 3.6–11.1 mmol/L Magnesium 0.88 0.57 0.65–1.15 mmol/L Osmolality 330 - 280–300 mosmol/kg Urea 46.7 13.7 3.0–7.6 mmol/L Creatinine 537 - 61–111 μmol/L Lactate 0.6 1.2 < 2.0 mmol/L Serum ketones 2+ - Undetected Anion gap 23 20 12–14 Osmolar gap 14.5 - 0–10 Chest radiograph revealed right middle lobe and lingular patchy opacification. An abdomino-pelvic CT scan demonstrated moderate to severe bilateral hydronephrosis, bladder wall thickening with multiple diverticuli, and retroperitoneal streaking consistent with acute infection. A provisional diagnosis of severe sepsis was made with multiple potential foci of infection. The patient was given empiric ceftriaxone, metronidazole and vancomycin. Sputum specimen cultured heavy methicillin-sensitive Staphylococcus aureus , blood cultures were positive for S. aureus , Escherichia coli , and Group B Streptococcus . Urine cultured greater than 10 8 CFU/L of multiple gram positive and negative organisms. The patient was admitted to the intensive care unit (ICU). The metabolic acidosis persisted (pH 7.00) a despite 100 mEq of 8.4% sodium bicarbonate bolus and infusion of three liters of normal bicarbonate solution (150 mEq of 8.4% sodium bicarbonate in 1000 mL D5W). The patient had a suprapubic bladder catheter inserted by angiography. However, due to concern the patient remained oliguric following 4 L crystalloid resuscitation, hemodialysis was organized. Hemodialysis parameters included: F160 membrane (surface area 1.5 m 2 and KUf 50 mL/hr/mmHg), dialysate sodium 136 mmol/L, potassium 3 mmol/L, calcium 1.25 mmol/L, bicarbonate 40 mmol/L, and Q D 500 mL/min, Q B 250–300 mL/min via a 25 cm left femoral double-lumen Uldall catheter. The patient had 71.5 L of blood processed over four hours with no fluid removal. Although the patient was alert and appropriate (Glasgow Coma Scale 15) with tachycardia and stable normal range blood pressure before the initiation of dialysis, he was demonstrating an increased work of breathing and oxygen requirements suggestive of worsening sepsis syndrome. Approximately 2.5 hrs after start of dialysis the patient became rapidly unresponsive prompting intubation for airway protection. At completion of HD and over the subsequent 4 hours the patient's neurologic status deteriorated with evidence of loss of all brainstem reflexes. Head CT-scan is shown in Figure 1 . Figure 1 Computerized Tomography (CT) head showing diffuse cerebral edema with effacement of basal cisterns and generalized loss of gray-white differentiation Repeat laboratory investigations immediately following hemodialysis revealed a pH 7.36, HCO3 19 mmol/L, sodium 132 mmol/L, potassium 1.8 mmol/L, and urea 13.7 mmol/L (urea-reduction-ratio was 71%) (Table 1 ). The patient rapidly progressed to refractory shock and multi-organ dysfunction Diagnosis of brain death was declared independently by an intensivist and a neurologist. At autopsy, the brain showed evidence of diffuse cerebral edema. Cardiac assessment showed left ventricular enlargement consistent with systemic hypertension likely as a result of chronic kidney disease. Both lungs showed patchy acute bronchopneumonia with edema and congestion. Both kidneys appeared grossly pyonephrotic with dilated, thickened ureters and suggested the presence of acute on chronic pyelonephritis. The meatal aperture was scarred and stenosed. Discussion The immediate indication for renal replacement therapy was correction of refractory metabolic acidosis in the setting of oliguria; however, following initiation of HD this patient developed irreversible symptoms consistent with DDS. DDS occurs most commonly following initiation of chronic HD for patients with end-stage renal disease[ 2 ]. Patients with pre-existing neurologic disease, such as head trauma, stroke or malignant hypertension, may be at greater risk for developing DDS[ 5 , 6 ]. The precise epidemiology of DDS is poorly defined and may be under-reported due to the wide spectrum of clinical manifestations. Mild symptoms such as headache, nausea, blurred vision, muscle cramps, disorientation, anorexia, restlessness, hypertension and dizziness are common during or following HD and may be attributed to DDS[ 2 ]. More severe symptoms consistent with central nervous system dysfunction such as seizures, central pontine myelinolysis, coma and death are rare[ 7 ]. The temporal profile for DDS is not well described. DDS has been credited for acute electroencephalographic (EEG) abnormalities and structural changes on diagnostic imaging following rapid hemodialysis [ 8 - 10 ]. Likewise, brain MRI studies immediately following hemodialysis in chronic dialysis patients have shown quantitative increases in brain volume consistent with cerebral edema[ 11 ]. The pathogenesis remains debated and incompletely understood; however, two central hypotheses have emerged. First, acute urea removal occurs more slowly across the blood-brain barrier than from plasma, generating a 'reverse osmotic gradient' promoting water movement into the brain and cerebral edema[ 12 ]. Absolute increases in brain water content have been demonstrated in a rat model of uremia undergoing rapid hemodialysis that was accounted for by an increase in the ratio of brain to plasma urea[ 13 , 14 ]. Down-regulation of central nervous system urea transporters have been proposed as a mechanism contributing to the delay in urea clearance from the brain[ 15 ]. The second hypothesis states that the increased osmolality of the extracellular fluid in uremia stimulates an adaptive accumulation of intracellular organic osmolytes to limit cerebral cell dehydration[ 16 ]. During hemodialysis, retention of these organic osmolytes contributes to a paradoxical reduction in intracellular pH resulting in increased brain osmolality and cerebral edema[ 17 , 18 ]. The patient in this case unfortunately may have been susceptible to both proposed pathophysiologic mechanisms. The likely and under-appreciated presence of pre-existing kidney disease (chronic obstructive nephropathy and pyelonephritis) with an increased serum osmolality would have resulted in adaptive changes in the central nervous system. Ensuing hemodialysis correction of the plasma metabolic acidosis may have eclipsed a more severe cerebral intracellular acidosis. Further, urea clearance by hemodialysis was efficient at approximately 70% and probably generated a sufficient plasma-to-brain urea gradient for development of cerebral edema, intracranial hypertension and DDS. A less efficient initial course of hemodialysis would have diminished the osmolar gradient of urea across the central nervous system reducing the likelihood of symptoms of DDS. Other variables may have contributed. The patient was compensating for the severe metabolic acidosis by hyperventilation (PaCO 2 10 mmHg); however, initial post-intubation PaCO 2 was 42 mmHg. Rapid elevations in PaCO 2 can alter cerebral autoregulation resulting in exacerbated intracranial hypertension[ 19 ]. Concomitant sepsis syndrome with poly-microbial bacteremia resulting in widespread immune activation may alter blood-brain-barrier permeability and contribute to cerebral edema[ 20 , 21 ]. These factors likely contributed to an increased risk for DDS-induced cerebral edema. The symptoms of DDS have been ameliorated by several interventions targeted to reduce the hemodialysis-induced plasma-to-brain osmotic gradient promoting cerebral edema[ 22 ]. A similar case of severe DDS requiring intubation was prevented from recurring during subsequent hemodialysis by use of modified dialysate containing 10.1 mmol/L of urea[ 23 ]. Likewise, the administration of intravenous mannitol and hyperventilation reversed a case of severe DDS-induced central nervous system dysfunction in a patient undergoing initial hemodialysis for acute renal failure[ 3 ]. Conversely, sodium profiling, high sodium or hyperglycemic dialysate have been attempted with variable results[ 24 , 25 ]. Prevention of DDS is traditionally the mainstay of therapy, particularly during initiation of hemodialysis in new patients. Despite the absence of evidence-based guidelines, the conventional aim is for a gradual clearance of urea. This can be accomplished with intermittent hemodialysis by use of a smaller, less efficient dialyzer and by reducing the duration of initial dialysis to approximately 2 hours with targeted lower blood flow rates of 150–200 mL/min, use of sustained low-efficiency dialysis (SLED), or initiation of continuous renal replacement therapy (CRRT) with more gradual and stable clearance of urea[ 2 , 26 - 28 ]. As a result, DDS has not been reported with the use of SLED or CRRT in critically ill patients. By providing a shorter, less efficient trial of initial hemodialysis, the severe DDS and brain death in this case may have been altogether prevented. In summary, the precise epidemiology and pathophysiology of DDS remain unclear. Although DDS usually presents in end-stage renal disease patients undergoing initial therapy, critically ill patients may represent a unique population where co-existing illnesses such as sepsis, brain injury or other central nervous system disease, multiorgan dysfunction, and need for sedation can present obstacles for prompt diagnosis of DDS. Furthermore, for similar reasons, critically ill patients may have increased susceptibility to DDS conditions. Competing interests None declared. Authors' contributions SMB wrote and revised the manuscript. ADP, MH, PJEB, KBL and CJD provided critique of successive drafts of the manuscript. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC515303.xml
517500	Race/ethnicity and disease severity in IgA nephropathy	Background Relatively few U.S.-based studies in chronic kidney disease have focused on Asian/Pacific Islanders. Clinical reports suggest that Asian/Pacific Islanders are more likely to be affected by IgA nephropathy (IgAN), and that the severity of disease is increased in these populations. Methods To explore whether these observations are borne out in a multi-ethnic, tertiary care renal pathology practice, we examined clinical and pathologic data on 298 patients with primary glomerular lesions (IgAN, focal segmental glomerulosclerosis, membranous nephropathy and minimal change disease) at the University of California San Francisco Medical Center from November 1994 through May 2001. Pathologic assessment of native kidney biopsies with IgAN was conducted using Haas' classification system. Results Among individuals with IgAN (N = 149), 89 (60%) were male, 57 (38%) white, 53 (36%) Asian/Pacific Islander, 29 (19%) Hispanic, 4 (3%) African American and 6 (4%) were of other or unknown ethnicity. The mean age was 37 ± 14 years and median serum creatinine 1.7 mg/dL. Sixty-six patients (44%) exhibited nephrotic range proteinuria at the time of kidney biopsy. The distributions of age, gender, mean serum creatinine, and presence or absence of nephrotic proteinuria and/or hypertension at the time of kidney biopsy were not significantly different among white, Hispanic, and Asian/Pacific Islander groups. Of the 124 native kidney biopsies with IgAN, 10 (8%) cases were classified into Haas subclass I, 12 (10%) subclass II, 23 (18%) subclass III, 30 (25%) subclass IV, and 49 (40%) subclass V. The distribution of Haas subclass did not differ significantly by race/ethnicity. In comparison, among the random sample of patients with non-IgAN glomerular lesions (N = 149), 77 (52%) patients were male, 51 (34%) white, 42 (28%) Asian/Pacific Islander, 25 (17%) Hispanic, and 30 (20%) were African American. Conclusions With the caveats of referral and biopsy biases, the race/ethnicity distribution of IgAN differs significantly from that of other major glomerulonephridities. However, among individuals undergoing native kidney biopsy, we see no evidence of a race/ethnicity association with severity of disease in IgAN by clinical and IgAN-specific histopathologic criteria. Further studies are needed to identify populations at higher risk for progressive disease in IgAN.	Background IgA nephropathy (IgAN) is the most common form of glomerulonephritis (GN) worldwide [ 1 ]. Approximately 20–30% of individuals with IgAN develop end-stage renal disease (ESRD) by 10–20 years following diagnosis [ 2 ]. Despite the increasing recognition of IgAN as a significant cause of chronic kidney disease (CKD), no specific therapies for IgAN have been developed. Glucocorticoids and fish oil have been recommended for selected patients with IgAN based on controversial results from small clinical trials [ 3 - 5 ]. Many patients with IgAN receive no specific therapy. Clinical reports suggest that individuals of Asian/Pacific Islander heritage are more likely to be affected by IgAN than whites, African Americans, and persons of Hispanic descent. Reports from U.S. centers have generally compared results of white and African American IgAN patients, with little or no available information on U.S. patients of Asian/Pacific Islander heritage. Studies from Japan and China have reported that more individuals with ESRD in these countries had IgAN, implying that IgAN may have a more severe disease course in certain Asian populations [ 6 , 7 ]. To explore whether IgAN was more common and severe among Asian/Pacific Islanders in our population, we examined clinical and pathologic data on 149 patients with IgAN and a random sample of 149 patients with other primary glomerular lesions (focal segmental glomerulosclerosis, membranous nephropathy and minimal change disease) at the University of California San Francisco (UCSF) Medical Center. Methods The records of 183 percutaneous native and transplant kidney biopsies with a diagnosis of IgAN received between November 1994 and May 2001 at the Renal Pathology Laboratory at UCSF were reviewed. Baseline demographic and clinical data included age, gender, race or ethnicity, history of kidney transplant, date of biopsy, and serum creatinine concentration at the approximate time of biopsy. In addition, the presence or absence of heavy proteinuria (≥ 3.0 g/day, with or without the nephrotic syndrome) and the presence or absence of hypertension at the time of biopsy were recorded. Major ethnic groups included white, Asian/Pacific Islander, Hispanic, African American, and other/unknown. Ethnicity was determined using information from patient health insurance forms and history provided at the time of biopsy. Any case in which Henoch-Schönlein purpura, systemic lupus erythematosus or chronic liver disease were considered likely diagnoses were excluded, as were cases of IgAN superimposed on a systemic disease involving the kidney (e.g., diabetic nephropathy). Two examiners unaware of the clinical data independently reviewed the biopsies. Biopsies displaying fewer than six glomeruli by light microscopy or insufficient immunofluorescence staining, as defined below, were also excluded. Fifteen biopsies were additionally excluded due to incomplete recovery of microscopic slides from files. Biopsies from 149 patients, including 25 kidney transplant recipients, satisfied the criteria for inclusion and provided the basis for the IgAN analytic sample. Aside from IgAN, the most commonly diagnosed primary glomerular lesions at our institution over the same time period were focal segmental glomerulosclerosis (N = 314), membranous nephropathy (N = 197) and minimal change disease (N = 147). To establish baseline race/ethnicity prevalences for our region and referral population, we collected demographic data on a computer-generated random sample of individuals with non-IgAN glomerular disease (N = 149), stratified by kidney transplant. Pathologic assessment Pathologic assessment of the IgAN native kidney biopsies was performed based on Haas' IgA nephropathy classification system [ 8 ]. All cases included in the study also met the following criteria: (1) immunofluorescence studies showing at least 2+ (scale 0 to 3+) mesangial deposition of IgA, with IgA comprising the dominant immunoglobulin deposited in the glomeruli, and (2) electron microscopy (EM) studies showing the presence of mesangial deposits. Statistical analysis Demographic and clinical data are reported as mean ± standard deviation, medians with interquartile ranges, and proportions with 95% confidence limits. Inter-ethnicity comparisons were performed using the Cochrane-Mantel-Haenzsel χ 2 test for categorical variables, and analysis of variance (general linear models) or Kruskal-Wallis test for continuous variables. Two-tailed P-values <0.05 were considered statistically significant. SAS version 8.2 was used for all statistical analyses (SAS Institute, Cary, NC, USA). Results Patient clinical characteristics for the IgAN group are summarized in Table 1 . Eighty-nine (60%) patients were male and 60 (40%) were female. Fifty-seven (38%) patients were white, 53 (36%) Asian/Pacific Islander, 29 (19%) Hispanic, 4 (3%) African American and 6 (4%) were of other or unknown race/ethnicity. The mean age of the IgAN patients at the time of kidney biopsy was 37 ± 14 years. Among the three main ethnic groups (whites, Hispanics, and Asian/Pacific Islanders), Hispanic patients tended to be slightly younger at the time of biopsy compared with whites and Asian/Pacific Islanders. The distributions of age and gender, however, did not differ significantly among white, Hispanic, and Asian/Pacific Islander groups. Table 1 Summary characteristics: IgA nephropathy Characteristic a Total (N = 149) White (N = 57) Asian/PI b (N = 53) Hispanic (N = 29) African American (N = 4) Unknown (N = 6) P -value c Mean age (yr) 37 ± 14 38 ± 15 37 ± 15 34 ± 14 26 ± 22 36 ± 24 0.54 Mean SCr (mg/dL) 3.1 ± 3.8 3.0 ± 2.6 3.1 ± 4.5 2.9 ± 4.2 6.3 ± 6.2 2.7 ± 2.9 0.96 Male, N (%) 89 (60) 37 (65) 29 (55) 16 (55) 3 (75) 4 (67) 0.50 Proteinuria ≥ 3g/d, N (%) 66 (44) 28 (49) 20 (37) 13 (45) 3 (75) 2 (33) 0.08 Hypertension, N (%) 74 (50) 28 (49) 28 (53) 14 (48) 2 (50) 3 (33) 0.86 Transplant, N (%) 25 (17) 4 (7) 13 (24) 8 (28) 0 (0) 0 (0) 0.15 Haas subclass d , N (%) I 10 (8) 4 (8) 5 (13) 0 (0) 0 (0) 1 (17) 0.76 II 12 (10) 5 (9) 4 (10) 2 (10) 0 (0) 1 (17) III 23 (19) 10 (19) 8 (20) 2 (10) 1 (25) 2 (32) IV 30 (24) 13 (24) 8 (20) 7 (33) 1 (25) 1 (17) V 49 (40) 21 (40) 15 (37) 10 (47) 2 (50) 1 (17) a Values represent mean ± standard deviation. b Asian/Pacific Islander c P -value refers to overall NOVA, Kruskal-Wallis test, or χ 2 for comparison of white, Asian/Pacific Islander and Hispanic groups. d Haas subclass assessment performed on native kidney biopsies (N = 124). The median serum creatinine (SCr) of the IgAN cohort was 1.7 mg/dL (interquartile range 1.1–3.4 mg/dL). The median SCr of the African American group was signficantly higher (5.0 mg/dL) than the other ethnic groups; however, these calculations were based on a small sample size (N = 4) due to the low prevalence (3%) of African Americans with in our IgAN cohort. Median serum creatinine concentrations were not significantly different among white, Hispanic, and Asian/Pacific Islander groups ( P = 0.64). Sixty-six patients (44%) exhibited heavy (≥ 3 g/d) proteinuria, and 74 (50%) had documented hypertension (systolic blood pressure ≥ 140 or diastolic blood pressure ≥ 90 mm Hg) at the time of kidney biopsy. The fractions of patients with heavy proteinuria and hypertension at the time of kidney biopsy were not significantly different among white, Hispanic, and Asian/Pacific Islander groups. Of the 124 native kidney biopsies, the majority of cases (64%) fell into Haas subclasses IV or V, which are known independent predictors of progressive disease and poor renal outcomes [ 2 , 8 ]. Only 22 biopsies (18%) were graded as Haas subclasses I or II, reflecting a relatively high threshold for kidney biopsy in our referral region. The distribution of Haas subclass did not differ significantly among race/ethnicity groups. Table 2 shows the demographic characteristics of the IgAN and non-IgAN groups. Among the random sample of patients (N = 149) with non-IgAN primary glomerulopathies, 67 (45%) patients had focal segmental glomerulosclerosis (FSGS), 58 (39%) membranous nephropathy, and 24 (16%) minimal change disease. Seventy-seven (52%) patients were male, 51 (34%) white, 42 (28%) Asian/Pacific Islander, 25 (17%) Hispanic, and 30 (20%) were African American. In contrast to previous reports, the distribution of gender did not differ significantly between the IgAN and non-IgAN groups (P = 0.16) [ 8 ]. Patients in the non-IgAN group were, however, significantly older (mean age 42 ± 21 years vs. 37 ± 14 years, P = 0.006) compared to patients with IgAN. In addition, the distribution of race/ethnicity differed significantly between the two groups ( P < 0.001). This association of IgAN and distribution of race/ethnicity persisted even when stratified by kidney transplant ( P < 0.001 for native kidney comparison, P = 0.006 for kidney transplant recipient comparison). Table 2 Patient demographics: All glomerular lesions Characteristic a IgA nephropathy (N = 149) Non-IgAN glomerular lesions b (N = 149) P -value c Mean age (years) 37 ± 14 42 ± 21 0.006 Male, N (%) 89 (60) 77 (52) 0.16 Transplant, N (%) 25 (17) 25 (17) 1.00 Race/ethnicity, N (%) White 57 (38) 51 (34) <0.001 African American 4 (3) 30 (20) Asian/PI d 53 (36) 42 (28) Hispanic 29 (19) 25 (17) Other/Unknown 6 (4) 1 (1) a Values represent mean ± standard deviation. b Focal segmental glomerulosclerosis, membranous nephropathy, and minimal change disease. c P -value refers to overall ANOVA, Kruskal-Wallis test, or χ 2 . d Asian/Pacific Islander Discussion In a biopsy series of 244 patients with IgAN, Haas found fewer African Americans (in a major urban setting), similar to that noted in other U.S.-based studies of IgAN [ 9 - 11 ]. While limited by the size of certain ethnic groups in the study population, Haas found no significant difference in renal survival associated with "white race, black race or Hispanic origin" [ 8 ]. The reason for a lower prevalence of IgAN in African Americans relative to other kidney diseases remains unclear. The frequency of IgAN in African Americans does not appear to be influenced by the higher prevalence of the IgA2 allotype among this group [ 12 ]. In our study, the fraction of biopsies in subclasses I and II (18%) was similar to that observed by Haas (23%). However, we observed a higher proportion of biopsies in subclasses IV and V (64% vs. 31%), and a lower proportion of biopsies in subclass III (19% vs. 45%) compared with Haas, possibly reflecting a temporal trend towards a higher biopsy threshold along with intergrader measurement bias. In a study reviewing the pattern of glomerulonephritis in Singapore over the past two decades, Woo and colleagues reported that IgAN was the most common primary GN occurring in Singapore (42% of all primary GNs in the first decade and 45% in the second decade) [ 13 ]. In our biopsy population during the same period that we studied, IgAN represented 12.8% of all biopsies with primary glomerular diseases and 8.4% of all biopsies (excluding transplant biopsies for non-glomerular diseases). In China, Li reported that IgAN was the leading cause of ESRD, accounting for approximately 18% of patients [ 6 ]. In a national survey of Japanese patients with ESRD, Koyama et al. reported that 28% of new dialysis patients had IgAN listed as their primary cause of ESRD. Moreover, due to the number of additional biopsies showing chronic glomerulonephritis without immunofluorescent microscopic descriptions in the survey, the authors estimated that possibly 40% of newly registered dialysis patients in Japan might have had CKD from IgAN. In contrast, only 0.8% of incident ESRD patients in the U.S. have documented or suspected IgAN [ 14 ]. Katznelson and Cecka, using data from the United Network for Organ Sharing (UNOS), have also reported a higher incidence of IgAN and chronic glomerulonephritis causing ESRD in Asian/Pacific Islander American recipients of renal allografts between 1991 and 1996 [ 15 ]. In contrast, based on smaller biopsy series, a striking variation in prevalence rates of IgAN has been reported from Europe and South America. In the UK, for example, Ballardie and colleagues noted comparatively low prevalence rates of IgAN in a predominantly white population (Manchester, England) in the early 1970's. In the subsequent 15-year period, however, these investigators reported a phenomenal rise in the observed incidence of IgAN (accounting for 31% of all glomerulopathies in 1986), which the investigators felt more likely reflected a higher frequency of detection rather than true rise in disease incidence. Similar prevalence rates have also been documented from isolated white populations in Finland and southern Italy [ 16 , 17 ]. In contrast, few studies have addressed the epidemiology of IgAN in Latin America. In a small Brazilian single-center cohort (N = 205) of primary glomerular diseases, Mazzarolo et al. reported relatively modest prevalence rates (10.2%) of IgAN [ 18 ]. A larger series (N = 1,263) of renal biopsies from Peru noted much lower prevalence rates of IgAN, which accounted for only 0.9% of all glomerular lesions over a 10-year period at a central reference renal pathology laboratory in Lima [ 19 ]. These differences may be partially attributed to increased screening and disparities in the indication for kidney biopsy. In Japan and South Korea, for example, school-aged children undergo annual screening for urinary abnormalities; kidney biopsy is subsequently recommended for children with evidence of proteinuria or hematuria [ 20 ]. More comprehensive yearly health exams are further performed on full-time salaried employees throughout Japan, Singapore, and Hong Kong, making detection more likely in these ethnic groups. Furthermore, a significant reporting bias may also contribute to the higher reported prevalence rates of IgAN in Asian/Pacific Islanders, e.g., in the study by Koyama et al., only 502 (7%) of the approximate 6800 patients diagnosed with IgAN had undergone a confirmatory kidney biopsy [ 7 ]. Although the etiology of IgAN remains unknown, there exists a strong suspicion for an environmental antigen trigger combined with a genetic susceptibility factor. Along these lines, several hypotheses have been proposed to account for the reportedly higher prevalence of IgAN in Asian/Pacific Islanders. With respect to potential dietary antigens, Wakai et al. found that high intake of rice and n-6 polyunsaturated fatty acids (PUFA) were associated with an increased risk of IgAN [ 21 ]. Recent reports from Japan have also suggested a potential role of H. parainfluenzae as a causative agent of IgAN in Japanese children and adults. Such claims are supported by studies showing the glomerular deposition of outer membrane H. parainfluenzae antigens and greater levels of plasma IgA1 antibody against OMHP in Japanese patients with IgAN (compared to Japanese patients with other renal diseases) [ 22 , 23 ]. Whether Japanese, or Asian/Pacific Islanders in general, have higher rates of H. parainfluenzae colonization and/or infection has yet to be established. The presence of either hypertension or proteinuria ≥ 3.0 g/24 hrs at the time of diagnosis significantly correlated with worsened renal survival in IgAN, even when controlling for serum creatinine at the time of kidney biopsy [ 2 ]. We found no difference in the distribution of Haas subclass, hypertension and nephrotic proteinuria among Caucasians, Asian/Pacific Islanders, and Hispanics. Despite ongoing investigative efforts, scant data are available regarding genetic markers that may predispose individuals to progressive disease from IgAN. Recent immunogenetic studies have suggested a potential role for the T-cell receptor (TCR) in the development of immune-mediated diseases. Deenitchina and colleagues found that genetic polymorphism of the TCR constant alpha chain was associated with progression of CKD in a cohort of Japanese patients with IgAN. Although promising, such polymorphisms of the TCR gene have yet to be evaluated in large, prospective studies or by genetic analysis of familial IgAN [ 24 ]. Our results contest the assertion that IgAN follows a more severe course in individuals of Asian/Pacific Islander descent. One reason for the similar disease severity of IgAN in our study population may stem in part from the large subpopulation of Filipino patients comprising our Asian/Pacific Islander cohort. It is unclear whether certain subpopulations of Asian/Pacific Islanders, including Filipinos, exhibit IgAN prevalence rates similar to those documented by Koyama and Woo. Anecdotal reports from Thailand and India documenting prevalence rates of 4–9% suggest that IgAN may not have the same epidemiology among all southeast Asians [ 1 ]. Despite having higher incidence rates of ESRD than the U.S. white population, the Asian/Pacific Islanders remain a largely unstudied group, for whom more comprehensive data collection is warranted. There are several important limitations to this report. As with any single-center biopsy series, we may have been underpowered to detect a clinically significant difference due to the limited sample size (type II error). In addition, racial admixture may have also confounded the results, as we were unable to subclassify patients in the Asian/Pacific Islander group or account for the growing population of bi- or multi-ethnic individuals in our population. Furthermore, due to the study's case control design, and breadth of our referral base (northern California and Hawaii), we were unable to control for the criteria for kidney biopsy. As a result, a biopsy bias may have confounded our results. In other words, Asian/Pacific Islander patients in our referral base with mild to moderate proteinuria and/or hematuria might have been given a presumptive diagnosis of IgAN without nephrology referral or confirmatory kidney biopsy. With regard to disease prevalence, these potential referral and biopsy biases based on race/ethnicity are largely conservative in nature, and would have biased our results towards the null. Finally, we have included data from a modest-sized IgAN transplant population (N = 25), the donor demographics of which were unavailable at the time of the study. However, the association of race/ethnicity and distribution of glomerular lesion persisted, even when stratified by kidney transplant, and thus our overall conclusions remained the same. In addition, a small European study of donor-recipient pairs (average follow-up 7 years) has shown that when a donor kidney with asymptomatic IgA deposits is transplanted into a recipient with ESRD secondary to a disease other than IgAN, the IgA immune deposits in the donor kidney are rapidly removed [ 25 ]. Conclusions In conclusion, with the caveats of referral bias and biopsy bias, the race/ethnicity distribution of IgAN differs significantly from that of other major glomerulonephridities. However, among individuals undergoing native kidney biopsy, we see no evidence of a race/ethnicity association with severity of disease in IgAN by clinical and IgAN-specific histopathologic criteria. Further studies are needed to identify populations at higher risk for progressive disease in IgA nephropathy. Competing interests None declared. Authors' contribution YH designed the study, collected and analyzed the data, and drafted the manuscript. GC supervised the study design, analyzed the data, and edited the manuscript. EF graded the IgAN histopathologic slides. JO collected, reviewed and graded all histopathologic data, and edited the manuscript. All authors approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517500.xml
538285	Caregivers' practices, knowledge and beliefs of antibiotics in paediatric upper respiratorytract infections in Trinidad and Tobago: a cross-sectional study	Background Antibiotic overuse and misuse for upper respiratory tract infections in children is widespread and fuelled by public attitudes and expectations. This study assessed knowledge, beliefs, and practices regarding antibiotic use for these paediatric infections among children's caregivers' in Trinidad and Tobago in the English speaking Caribbean. Methods In a cross-sectional observational study, by random survey children's adult caregivers gave a telephone interview from November 1998 to January 1999. On a pilot-tested evaluation instrument, respondents provided information about their knowledge and beliefs of antibiotics, and their use of these agents to treat recent episodes (< previous 30 days) of upper respiratory tract infections in children under their care. Caregivers were scored on an antibiotic knowledge test and divided based on their score. Differences between those with high and low scores were compared using the chi-square test. Results Of the 417 caregivers, 70% were female and between 18–40 years, 77% were educated to high school and beyond and 43% lived in urban areas. Two hundred and forty nine (60%) respondents scored high (≥ 12) on antibiotic knowledge and 149 (34%) had used antibiotics in the preceding year. More caregivers with a high knowledge score had private health insurance (33%), (p < 0.02), high school education (57%) (p < 0.002), and had used antibiotics in the preceding year (p < 0.008) and within the last 30 days (p < 0.05). Caregivers with high scores were less likely to demand antibiotics (p < 0.05) or keep them at home (p < 0.001), but more likely to self-treat with antibiotics (p < 0.001). Caregivers administered antibiotics in 241/288 (84%) self-assessed severe episodes of infection (p < 0.001) and in 59/126 (43%) cough and cold episodes without visiting a health clinic or private physician (p < 0.05). Conclusions In Trinidad and Tobago, caregivers scoring low on antibiotic knowledge have erroneous beliefs and use antibiotics inappropriately. Children in their care receive antibiotics for upper respiratory tract infections without visiting a health clinic or a physician. Educational interventions in the community on the consequences of inappropriate antibiotic use in children are recommended. Our findings emphasise the need to address information, training, legislation and education at all levels of the drug delivery system towards discouraging self-medication with antibiotics in children.	Background Acute respiratory infections and diarrhoeal diseases are the leading causes of childhood mortality, resulting in 25–33% of all deaths in children in developing countries. Worldwide, antibiotics are the most commonly prescribed and abused drugs for upper respiratory tract infections (URTIs) [ 1 , 2 ]. In surveys on antibiotic use, about 20 % of prescriptions were inappropriate [ 3 ] and a high proportion of patients received antibiotics during clinic visits [ 4 ]. A significant force driving the occurrence and the spread of antibiotic resistance is the inappropriate use of antibiotics in primary and ambulatory care settings. Streptococcus pneumoniae , Hemophilus influenzae and Moraxella catarrhalis , the most common bacterial pathogens causing acute bacterial rhinosinusitis (ABRS) in children, in both the developed and developing countries, have already demonstrated resistance to the first line antibiotics [ 5 - 9 ]. The Sinus and Allergy Health Partnership (SAHP) guidelines for the treatment of ABRS observe antibiotics prescribed for ABRS are not only ineffective, but may contribute to the development of antibiotic-resistant bacterial infections [ 10 ] which is supported by the increasing resistance of Streptococcus pneumoniae and the increasing prevalence of strains resistant to the beta-lactams and co-trimoxazole. Unnecessary antibiotic use in viral respiratory illnesses in humans is a key factor influencing the emergence and spread of resistant pneumococci. Inappropriate antibiotic use may be consequent to misdiagnosis of the illness (viral and bacterial URTIs present with similar symptoms), patients' expectations, and their demands which induce physicians to prescribe antibiotics [ 11 ]. In the United States, a higher proportion of infections due to penicillin resistant pnuemococci among young children and whites have been attributed to an overuse of antibiotics [ 12 , 13 ]. Epidemiological studies demonstrate recent antibiotic use is strongly associated with carriage of resistant pneumococci in the community and the individual, and in patients with invasive pneumococcal disease, recent antibiotic use has been associated with increased risk of infection with a resistant strain [ 14 ]. Previous community-oriented studies suggest an irrational use of antibiotics, particularly in the developing and lesser-developed countries. Factors influencing antibiotic resistance are the higher incidence of infectious diseases in children, the lack of access to health care, costs and poor regulatory controls on the use of prescription drugs such as antibiotics, coupled with low antibiotic knowledge prompting increased self medication with these drugs [ 15 - 17 ]. As the first health-decision makers in deciding when to initiate antibiotic treatment and limit their unnecessary use, mothers and caregivers must have the appropriate knowledge to enable correct decisions. This first Caribbean study investigated the knowledge, beliefs and practices of children's caregivers in Trinidad and Tobago regarding antibiotic utilisation and explored these beliefs in self-administration of antibiotics in childhood URTIs. Gaining a better understanding of caregivers' management of childhood URTIs and factors that influence their use of antibiotics will allow appropriate educational interventions and reduce unnecessary antibiotic use in children. Methods Setting Trinidad and Tobago, a twin island republic in the Caribbean located off the Venezuelan coast with a population of 1.3 million people is the second largest country in the English speaking Caribbean. About 74% of the country's population live in the urban areas and 60% of the population is aged between 15–64 years. The literacy rate in Trinidad and Tobago is 98% and at least 70% of the country's adult population has completed secondary (high) school to the pre-university level. Medical care in Trinidad and Tobago is publicly financed through three (3) regional health authorities (North West, South West and Eastern regions) in Trinidad, and one in Tobago. Secondary and tertiary care are provided at one general hospital in Port-of-Spain, one in San Fernando (1,245 beds), and at two county hospitals in Trinidad (111 beds), and at one hospital in Tobago (96 beds), besides institutions for specialised services. Primary health care is provided at 82 health centers in Trinidad and 19 in Tobago. The ratio of the population to a health center ranges from < 3000 per center in Tobago to > 21,000 per center in the north-west region. There are 33 private hospitals and approximately 45% of the population preferentially uses the private sector services. Private general practitioners are concentrated in the cities and larger towns and the estimated ratio of physician per population is 7.5 per 10, 000 inhabitants [ 18 ]. Design This prospective cross sectional observational study was conducted in Trinidad and Tobago from November 1998 to January 1999 in randomly selected subjects interviewed over the telephone. The study design and methods have been described previously [ 19 ]. Briefly, a sample size of 800 participants with a working telephone was calculated based on 80% power to detect a difference of at least 3% use of antibiotics giving an error of 0.05. This being the first such telephone survey in the country, with no experience of the rejection rate, 1,600 telephone numbers were randomly obtained from a sampling frame of 167,272 telephone subscribers of The Telecommunication Services of Trinidad and Tobago, the only telephone service provider in the country. Of 824 respondents, 753 agreed to participate with a response rate of 91.4%. At the outset of the study participants were questioned if they were caring for a child ≤ 12 years and the term 'antibiotic' was explained in a simple sentence : "Antibiotics are drugs that are prescribed for the treatment of diseases caused by germs" . The pilot-tested questionnaire consisted of 42 items in three parts, designed to investigate knowledge, beliefs and practices of antibiotics [ 19 ]. The first part on the caregiver's demographic data included the employment status, health insurance and educational background. The second part inquired about caregivers' knowledge and beliefs. To determine antibiotic knowledge participants were asked to identify 4 antibiotics from a list of 8 commonly used drugs, and could attain a maximum score of 16. Caregivers' beliefs about antibiotics were determined using the following three questions: 1. Do you think antibiotics can cure all infections? 2. Do you believe antibiotics are free from side-effects? 3. Do you think antibiotics are generally safe? From our earlier report and the pilot project of the current study respondents differentiated their understanding of 'side-effects' and 'safety'. The former was associated with unwanted disturbances from drug therapy on the quality of life and the latter was associated with life-threatening issues like organ toxicity and death. The third part of the questionnaire ascertained caregivers' practices of antibiotic use. They were asked about symptoms which children in their care had in the past 30 days, their assessment of the symptom severity, whether they sought medical assistance and if they administered any antibiotic to the child. Respondents were not asked to name the antibiotic. Information on suspected side-effects or allergic reactions was excluded from the final questionnaire as these responses in the pilot study were uncertain and subject to memory recall. Analysis Four hundred and seventeen of the 753 respondents were adult caregivers with children in the family and their responses were analysed. An antibiotic knowledge score was created based on the caregivers' responses to eight common drugs, which included four antibiotics. A high Antibiotic Knowledge Score (AKS) was defined as that at or above the median score. Data were analysed using SPSS version 11.0 (Chicago), and associations were determined by the Chi square test for antibiotic knowledge and caregiver's education, beliefs and practices and recent and past antibiotic use for URTIs. Results The majority of respondents was female (70%), ≥ 31 years, (72%) and had completed high school (pre-university) education (77%). There was a marginally high (57%) representation from the rural area (Table 1 ), but residence showed no relation to the determinants of the study. The ratio of respondents with African and Indian heritage bore close similarity to the ethnic profile of the population in Trinidad and Tobago. Seventy one percent of caregivers with children did not have private health insurance. Comparable proportions of caregivers, 35% and 34% respectively reported using antibiotics recently (< 30 days) and in the past (within the last one year). The median antibiotic knowledge score was 12 and was determined by the correct identification of penicillin, tetracycline, 'Augmentin', and 'Bactrim' as antibiotics from a list of 8 common drugs. Two hundred and forty-nine (60%) caregivers scored at or above the median score. The significant predictors of high antibiotic knowledge in caregivers were those who were employed, had private health insurance and, high school education. More caregivers with a high AKS had used antibiotics recently (p < 0.037) and in the past (p < 0.008) (Table 2 ). Table 1 Characteristics of children's caregivers in Trinidad and Tobago. Characteristics Caregivers with children (n = 417) Number of people in the house (mean +/- SD) 4.89 +/- 1.9 Number of children (mean +/- SD) 2.26 +/- 1.38 Gender: Males 126 (30)% Females 291 (70%)* Age group 18–30 yrs 112 (28%) 31–40 yrs 165 (42%) > 41 yrs 129 (30%) Residence Rural 235 (57%) Urban 176 (43%) Ethnicity African 142 (35%) Indian 158 (39%) Others 110 (29%) Education Primary school 96 (23%) High school 228 (55%)* College 90 (22%) Health Insurance With Private Health insurance 114 (29%) Without Private Health insurance 281 (71%)* Antibiotics used in previous year 143 (34%) Antibiotics used in last 30 days 149 (35%) * Significant at p < 0.05 Table 2 Caregivers' Antibiotic Knowledge Score (AKS) and associated factors Factors AKS <12(n = 168) (%) AKS ≥ 12(n = 249) (%) p VALUE Gender Males 51 (30) 51 (30) 0.96 Females 117 (70) 117 (70) Residence Rural 75 (46) 101 (41) 0.29 Urban 88 (54) 147 (59) Health Insurance Has insurance 35 (22) 79 (33) 0.02* Does not have 122 (78) 159 (67) Age Groups 18–30 yrs 55 (34) 57 (23) 0.07 31–40 yrs 60 (37) 105 (43) >41 yrs 47 (29) 82 (34) Ethnicity African 57 (35) 85 (35) 0.70 Asian 67 (41) 91(37) Others 41 (24) 69 (28) Education Primary 53 (37) 43 (17) 0.002* High School 87 (52) 141 (57) Tertiary Education 28 (18) 62 (26) Employment Employed 69 (43) 130 (54) 0.09 Self Employed 28 (18) 40 (17) Housewife/retired/unemployed 62 (39) 71 (29) Recent and past antibiotic use Used in the last 12 months 45 (27) 98 (40) 0.008* Not used in the last 12 months 118 (70) 150 (60) Used in the last 30 days 50 (30) 99 (40) 0.037* Not used in the last 30 days 122 (73) 150 (60) Significant at (p < 0.05) A majority of respondents had correct beliefs regarding whether 'antibiotics cure all infections' (54% [227/417]) and 'antibiotics are free from side effects' (61% [253/417]). Few (11% [49/417]) respondents believed antibiotics are generally safe. and some caregivers (18%–24%) remained non-responsive to all questions. The AKS of caregivers did not influence their beliefs (Table 3 ). Eighty six caregivers (22%) admitted to demanding antibiotics from a doctor. More caregivers (28%) with a low AKS demanded antibiotic prescriptions (p < 0.05) and kept these drugs at home (33%) (p < 0.001), to treat illnesses. Self-initiation of treatment for URTIs with antibiotics was more frequent (p < 0.05) among caregivers who had a high AKS. Caregiver's knowledge scores were not associated with the use of antibiotics given by relatives and / or friends or with compliance with the course whether recommended by the pharmacist or the doctor. (Table 3 ). Table 3 Antibiotic beliefs and practices of Caregivers in Trinidad FACTORS AKS <12 (%) AKS ≥ 12 (%) p value 1. Antibiotic Beliefs: a. Cure all infections Agree 48 (37) 63 (30) 0.17 Disagree 81 (63) 146 (70) b. Free from side -effects Agree 21 (18) 44 (21) 0.30 Disagree 99 (82) 153 (79) c. Generally safe Agree 21 (17) 28 (13) 0.40 Disagree 105 (83) 183 (87) 2. Antibiotic Practices Demands from doctor for URTI in children Yes 41 (28) 45 (19) 0.05* No 106 (72) 188 (81) Keeps at home Yes 53 (33) 45 (19) 0.001 No 107 (67) 195 (91) Self treatment Yes 33 (12) 69 (32) No 154 (88) 150 (68) 0.001 Given by friends and relatives Yes 16 (10) 25 (11) No 142 (90) 213 (89) 0.9 Compliance Yes 89 (63) 152 (71) No 52 (37) 63 (29) 0.13 Significant at p < 0.05, * Significant at p < 0.001 Caregivers reported as many as 450 episodes of URTIs in children within the previous 30 days (1.07 episodes per family) and 149 (35%) caregivers self-administered antibiotics in 64% (288/450) of these episodes. Cough and cold was the most frequently reported URTI symptom (48%, 214/450) followed by fever (28%,128/450) and sore throat (24%,108 / 450). Caregivers were more likely to give children antibiotics when they perceived URTIs to be severe [241/288 (84%)] (p < 0.001), and administered these drugs for the common cold [112/136, (82%)] fever [75/87 (83%)] and sore throat [54/65 (86%)] (Table 4 ). In 16% of episodes which caregivers deemed to be of mild severity, they self-administered antibiotics. Children received antibiotics without visiting a health clinic or a private physician for 126 (44%) URTI episodes, and more frequently for the common cold 59 (43.7%) compared with fever and sore throat (p < 0.05). Table 4 Antibiotic administration by caregivers for severe URTIs (n = 288) and visits to health provider URTI Episodes Assessed severe by caregiver (%) Visited a health clinic/private physician (%) Yes No Total Yes No Total Cough and Cold 112(82) 24(18) 136 77 (57) 59(43)* 136 Fever 75 (83) 12(17) 87 55 (63) 32(37) 87 Sore throat 54 (86) 11(14) 65 30 (46) 35(54) 65 Total 241(84)** 47 (16) 288 162(56) 126(44) 288 * Significant at p < 0.05 ** Significant at p < 0.001 Discussion This cross sectional study in Trinidad and Tobago determined the antibiotic knowledge of children's caregivers and the influence of this knowledge on their beliefs and use of these agents for URTIs in children under their care. We found high school education and higher socio-economic status (income permitted private health insurance) was significantly associated with higher knowledge scores. Similar associations between knowledge and antibiotic use were reported in a study in the Indian state of Kerala [ 20 ]. In the present study more caregivers who scored high on antibiotic knowledge, had used antibiotics (recently and in the past) compared with those who attained a low knowledge score. A significant proportion of caregivers in the present study had misconceptions that could contribute to the inappropriate use of antibiotics. Equal proportions of caregivers with high and low knowledge scores believed that antibiotics cure all infections and are free from side-effects. Even though URTIs are generally of viral aetiology [ 21 , 22 ], these mistaken beliefs may have steered antibiotic abuse from self treatment or over the counter demands at the pharmacy which are fostered from easy availability of these drugs at community pharmacies in Trinidad and Tobago [ 19 ]. In a survey from the United States, 48% of paediatricians reported parents do pressure them to prescribe antibiotics [ 23 ], and 78% of the sample believed educating parents on appropriate indications for antibiotic use was the single most important factor to promote suitable prescribing, suggesting effective communication between physicians and parents may reduce inappropriate antibiotic prescribing. Practices such as demanding a prescription for antibiotics from a physician, and keeping antibiotics at home (hoarding) were higher in caregivers with a low AKS. In Israel Shlomo et al [ 24 ] found lower education was a predictor of parents' expectations to receive antibiotics for URTIs and in Trinidad, Mohan et al reported that general practitioners attributed antibiotic over-prescribing in general practice to parents' demands [ 25 ]. A proclivity to demand antibiotics was associated with decreased knowledge and in children from insured families higher rates of antibiotic use were associated with low antibiotic knowledge and a tendency to demand antibiotics [ 26 ]. In Hong Kong educated respondents and working guardians had higher knowledge scores, and those who knew the viral aetiology of URTIs were less likely to demand antibiotics [ 27 ]. In the present study, the rate of self-treatment with antibiotics by caregivers, was higher in those who had a high AKS (32%vs12%). This may be consequent to caregivers needing to report for work following quicker recovery of children whom they care for. Braun and Fowles found a correlation between the expectation to get antibiotic treatment and parents' occupation and parents who worked full time had higher expectations to get antibiotic treatment, assuming perhaps that antibiotics shorten disease duration and allow an earlier return to work [ 28 ]. In Trinidad and Tobago at least 25% of the population demand a prescription for antibiotics from a doctor and 21% keep antibiotics in the house for emergency purposes [ 19 ]. Educational campaigns for the public can correct the widespread misconceptions on antibiotic use and storage. Earlier in describing the prescribing practices of Caribbean physicians we reported that respiratory tract infection was the most frequent reason for antibiotic prescriptions by physicians in the English and Dutch speaking Caribbean [ 29 ]. The influence of caregivers' (parents and relatives) knowledge on antibiotic use in children with URTIs has not been studied in any Caribbean region. Proportionately more children received antibiotics from caregivers for severe episodes of cold and cough, than for sore-throat and fever. Even for what they considered mild episodes (16%) of URTIs, caregivers administered antibiotics which as is current practice, probably obtained from community pharmacies on request [ 19 ]. In Malta parents gave antibiotics to their children without a prescription particularly for sore throat and the community pharmacy made the drugs available [ 30 ]. Caregivers in our study treated 44% of URTIs in children with antibiotics without consulting a physician or attending a health facility. We did not ascertain if the child suffered any unwanted effects of the drug, which information may have been important to discuss the consequences of freely giving antibiotics to children. We believe caregivers with health insurance, education beyond the primary level and a high AKS had obtained antibiotics informally at community pharmacies and those with a low knowledge score initiated medication with antibiotics from the home storage or given them by relatives and friends. A call for strict vigilance and enforced controls regarding 'over-the-counter' availability of antibiotics without a physician's prescription, despite being controlled drugs in Trinidad and Tobago, has been made in an earlier report [ 19 ]. Inappropriate antibiotic use is a common practice in the out patient setting in a clinic, or a physician's office [ 31 , 32 ], and has been attributed to the combination of time pressures of outpatient practice, diagnostic uncertainty, and physicians' misconceptions of patient expectations [ 33 ]. The cost and time spent for a visit to the health center or a physician's office and a genuine concern about the children's health could have pushed caregivers in Trinidad and Tobago to purchase antibiotics without a prescription which is a widespread practice here, and initiate treatment for common childhood respiratory tract illnesses. In Africa, Asia and Latin America antibiotics are obtained from pharmacies, hospitals and even from untrained vendors at the market place [ 21 , 34 - 36 ]. In Bavi, Vietnamese children were treated with antibiotics frequently by caregivers without physician consultation, resulting in a high prevalence of multi drug-resistant strains (MDR) among respiratory pathogens [ 37 ]. Contributing to the existence of the reservoir of MDR genes among bacterial pathogens undermines the effectiveness and success of antibiotic therapy in childhood respiratory tract infections in low-income countries. Conclusions Inappropriate antibiotic use for paediatric URTIs in Trinidad and Tobago may have been facilitated by low knowledge, erroneous beliefs and easy availability of these drugs without the required prescription, at a retail pharmacy. A combination of education and communication to combat patients' expectations for treatment, and the physicians' appropriate prescription for antibiotics can halt inappropriate antibiotic use in children. Using narrow-spectrum antibiotics, promoting dialogue with caregivers to discuss symptom relief and antibiotic resistance, and encouraging active management of the child's illness with follow-up calls is recommended. Pharmacists have a serious responsibility not to dispense these agents without prescriptions and to discourage patients from obtaining these drugs for self-treatment. Widespread educational campaigns targeting the general public in Trinidad and Tobago, particularly parents and caregivers of young children should focus on the difference in bacterial and viral infections and the futility of treating viral infections with antibiotics. A multidisciplinary approach to rational antibiotic use, dispensing these drugs as 'prescription only medicine' and educating the public can halt inappropriate use and contain resistance. List of abbreviations ABRS = Acute Bacterial Rhinosinusitis AKS = Antibiotic Knowledge Score SAHR = Sinus and Allergy Health Partnership URTI = Upper Respiratory Tract Infection Competing interests The author(s) declare that they have no competing interests. Authors'contributions PP conceptualised the study and drafted the protocol, NP did data collection, and contributed to the draft manuscript, LMPP prepared the final manuscript. All authors contributed to the statistical analysis and the literature search. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC538285.xml
544831	Transport lattice models of heat transport in skin with spatially heterogeneous, temperature-dependent perfusion	Background Investigation of bioheat transfer problems requires the evaluation of temporal and spatial distributions of temperature. This class of problems has been traditionally addressed using the Pennes bioheat equation. Transport of heat by conduction, and by temperature-dependent, spatially heterogeneous blood perfusion is modeled here using a transport lattice approach. Methods We represent heat transport processes by using a lattice that represents the Pennes bioheat equation in perfused tissues, and diffusion in nonperfused regions. The three layer skin model has a nonperfused viable epidermis, and deeper regions of dermis and subcutaneous tissue with perfusion that is constant or temperature-dependent. Two cases are considered: (1) surface contact heating and (2) spatially distributed heating. The model is relevant to the prediction of the transient and steady state temperature rise for different methods of power deposition within the skin. Accumulated thermal damage is estimated by using an Arrhenius type rate equation at locations where viable tissue temperature exceeds 42°C. Prediction of spatial temperature distributions is also illustrated with a two-dimensional model of skin created from a histological image. Results The transport lattice approach was validated by comparison with an analytical solution for a slab with homogeneous thermal properties and spatially distributed uniform sink held at constant temperatures at the ends. For typical transcutaneous blood gas sensing conditions the estimated damage is small, even with prolonged skin contact to a 45°C surface. Spatial heterogeneity in skin thermal properties leads to a non-uniform temperature distribution during a 10 GHz electromagnetic field exposure. A realistic two-dimensional model of the skin shows that tissue heterogeneity does not lead to a significant local temperature increase when heated by a hot wire tip. Conclusions The heat transport system model of the skin was solved by exploiting the mathematical analogy between local thermal models and local electrical (charge transport) models, thereby allowing robust, circuit simulation software to obtain solutions to Kirchhoff's laws for the system model. Transport lattices allow systematic introduction of realistic geometry and spatially heterogeneous heat transport mechanisms. Local representations for both simple, passive functions and more complex local models can be easily and intuitively included into the system model of a tissue.	Background Heat transfer in biological systems is relevant in many diagnostic and therapeutic applications that involve changes in temperature. For example, in hyperthermia the tissue temperature is elevated to 42–43°C using microwave [ 1 , 2 ], ultrasound [ 3 ], or laser light [ 4 ]. There has also been long standing interest in thermal properties of skin [ 5 ] in order to understand conditions leading to thermal damage (burns) to skin, usually involving contact of skin to hot objects [ 6 ], in which local thermal conduction and heat capacity are dominant. Investigation of such bioheat transfer problems requires the evaluation of temporal and spatial distributions of temperature. This class of problems has been traditionally addressed using the Pennes bioheat equation [ 7 , 8 ]. Here we show that a transport lattice approach [ 9 ] can solve bioheat problems. This method is illustrated by solving models for skin contact heating used in transcutaneous blood gas monitoring and for spatially distributed heating due to 10 GHz microwave radiation. Contact heating is used in transcutaneous blood gas monitoring, in which oxygen is transported out of the vasodilated capillary bed to a surface mounted oxygen sensor. Heating is used to achieve vasodilation. In 1851 it was already known that "skin breathing" occurs, in which oxygen diffuses out of ambient air into the body, supplying of order 1% of the body's oxygen uptake [ 10 ]. Typically the ambient air temperature even with clothing insulation causes the skin surface temperature to be significantly cooler than body core temperature. Much later, in 1957, it was shown that elevated skin temperature caused an outward diffusive flux of oxygen, so that oxygen could be measured at the surface of the skin [ 11 ]. The basic idea is that contact surface heating results in heat transport into the body, such that the outer portion of the dermis (the site of the outermost blood capillaries) experiences a significant increase in perfusion. This temperature-dependent perfusion "arterializes" the blood content of the capillaries, such that the oxygen concentration is closer to that of arterial blood, because to a good approximation capillary flow increases faster than oxygen transport out of the capillaries. Initial clinical demonstration with neonates occurred in 1969 when a polarographic electrode placed on the head was used to measure oxygen partial pressure [ 12 ]. Since the early studies significant development has taken place [ 13 - 16 ]. A basic issue of safety is involved, as sensor contact is often prolonged (1–8 h) in which a heated sensor (typically 45°C) with contacting material of high thermal diffusivity is placed against the skin. Spatially distributed heating of skin and deeper tissue by electromagnetic fields and ultrasound is also of established interest [ 17 - 20 ]. Microwave electromagnetic radiation is incident on tissue under a variety of exposure conditions. As an example, we consider 10 GHz microwave exposure. In this case, the penetration depth is approximately 3 mm so that most of the power is deposited within the outer region of the skin. Accurate prediction of the temperature distribution in skin exposed to microwave radiation is important in understanding both beneficial and harmful effects [ 21 - 23 ]. In hyperthermia, tissue is heated to enhance the effect of conventional radio- or chemotherapy. By delivering thermal energy, the tissue is stimulated to increase the blood flow by thermoregulation in order to remove the excess heat. The common method to produce local heating in the human body is the use of electromagnetic waves. Many of the bioheat transfer problems have been modeled using the Pennes equation, which accounts for the ability of tissue to remove heat by both passive conduction (diffusion) and perfusion of tissue by blood. Perfusion is defined as the nonvectorial volumetric blood flow per tissue volume in a region that contains sufficient capillaries that an average flow description is considered reasonable. Most tissues, including much of the skin and brain, are highly perfused, with a perfusion coefficient denoted by ω (traditionally with units of 100 ml/100 g min = 1 ml g -1 min -1 ). Alternatively, ω can be replaced by ω m , the nondirectional mass flow associated with perfusion. Perfusion is valid on the spatial scale of ~100 μ m. The contributions of heat conduction and perfusion are combined in the Pennes bioheat equation [ 7 , 8 ], which we use in a form [ 24 ] that employs ω m (SRI units of kgm -3 s -1 ), Here, ρ , c , k are the density, specific heat and thermal conductivity of tissue, respectively and c b , is the specific heat of blood, ρ b is the density of blood, T is local tissue temperature, T a is a reference temperature (arterial blood), t is time, Q m is the metabolic heat production per volume, and P ( z , t ) is the heat deposited per volume due to spatially distributed heating. In this general form, ω m is a function of temperature to include the specific case of temperature dependent perfusion. Vascularized tissue generally experiences increased perfusion as temperature increases [ 25 , 26 ]. Because of thermoregulation skin blood flow rises 15 fold to 100 ml l00 g min -1 , often with a time lag of minutes. Prediction of heat transport has long been carried out by both analytical and numerical methods [ 27 , 28 ]. The temperature rise for constant (temperature independent) perfusion has been predicted by traditional analytical methods based on Eq. 1, which can be solved analytically for simple geometries [ 22 , 29 ] or by finite element models for more realistic, complicated tissue geometry [ 30 - 32 ]. Models which include temperature-dependent increases in perfusion are more difficult to solve, but the case of a linear temperature dependence have been described using analytical expressions [ 33 ] and numerical simulations [ 24 ]. The bioheat transfer (Eq. 1) has been used in a wide range of applications to describe heat transport in blood perfused tissues, [ 34 ] and solved by a variety of methods. An adaptive finite element method was used to optimize the nonlinear bioheat equation for optimizing regional hyperthermia [ 35 ]. Two-dimensional biothermal models of ultrasound applicators based on the bioheat equation were solved by finite difference equation [ 36 ]. The boundary element and finite difference methods have also been used to solve the bioheat equation [ 37 - 41 ]. Recently, closed form analytical solutions to the bioheat transport problems with space and transient heating were reported using Green's function method [ 42 ]. Here we show that the transport lattice approach can be used to model transport of heat by conduction and temperature-dependent blood perfusion. This method employs a network of locally interacting transport, storage and source models that are solved as a system model by Kirchhoff's laws. Although Kirchhoff's laws can be used to describe transport of heat (and of molecules), usually charge transport is treated. Indeed, there is an extensive literature and robust methodology for solving large electric circuits [ 43 ]. For this reason, we use electrical circuits which provide mathematical analogs to heat transport and storage. Transport lattices allow systematic introduction of realistic geometry and spatially heterogeneous heat transport mechanisms. One attribute of a transport lattice model is that local representations for both simple, passive functions (e.g. heat storage via fixed heat capacitance and thermal conduction via fixed thermal conductivity) and more complex local models (e.g. nonlinear, temperature-dependent perfusion and spatially non-uniform perfusion in which the time lag of perfusion onset can be selected) can be easily and intuitively included into the system model of a tissue. It is a fundamentally modular approach in which local models can be introduced or removed. Methods Here we extend the transport lattice modeling approach previously demonstrated for electrical fields and currents in single and multiple cells [ 9 ] and supra-electroporation of cells by submicrosecond pulses [ 44 ] to describe heat transport within a multilayered skin model. A related approach has been described for analysis of calorimeters to measure specific heat of liquids [ 45 ]. Circuit analysis has long been used to solve problems that can be described by differential equations [ 45 - 49 ]. Here we use a modular approach in which the skin is represented by three layers, each with many interconnected local models that account for the local heat storage (heat capacity) and local transport by both conduction and perfusion (Fig. 1 ). The different parameters employed in the model and their values are listed in Table 1 . We model two cases of skin heating: surface contact heating and spatially distributed heating. Figure 1 Transport lattice method – Geometry with transport, source and storage models . The one-dimensional model of human skin (top) is represented by a lattice of conduction models ( M c ) and source, storage and sink models ( M s ). The subscripts denote the different layers of the model, namely, a: air, e: epidermis, d: dermis and s: subcutaneous tissue. The equivalent circuit models are shown for each layer: R c represents heat conduction, R p represents heat removal by perfusion (not present in epidermis) and C represents heat storage. The conduction model, M c , is represented by R c while the storage, perfusion and power input model, M s , is represented by the combination of R p , C and I (Table 2). (a) Surface Contact Heating : The surface temperature T s is elevated from 33°C to 45°C at t = 0. (b) Spatially Distributed Heating : A layer of air contacting the skin is added to the model with the air temperature ( T air ) held at 25°C. The local microwave power dissipation is represented by the current source (I) at each node. The arterial reference temperature T a is represented by a common node. A ladder-like network of variable resistors, R p , represents the temperature dependent perfusion in dermis and subcutaneous tissue. Table 1 Thermal and electrical property values assigned to different layers of skin. Air t a thickness 500 μ m N a number of lattice elements 100 ℓ lattice node spacing 5 μ m k a thermal conductivity 0.0263 Wm -1 ° C -1 ρ a density 1.3 kg m -3 c a specific heat 1004 Jkg -1 ° C -1 Epidermis t e thickness 80 μ m [52] N e lattice elements 80 ℓ lattice node spacing 1 μ m σ e electrical conductivity 8.01 Sm -1 [70] ε e relative permittivity 31.3 [70] η e penetration depth 3.8 mm [70] λ e wavelength 5.2 mm [70] k e thermal conductivity 0.23 Wm -1 ° C -1 [52] ρ e density 1200 kg m -3 [52] c e specific heat 3590 Jkg -1 ° C -1 [52] ω e perfusion rate 0 m 3 s -1 m -3 tissue [52] Dermis t d thickness 2000 μ m [52] N d lattice elements 100 ℓ lattice node spacing 20 μ m σ d electrical conductivity 8.01 Sm -1 [70] ε d relative permittivity 31.3 [70] η d penetration depth 3.8 mm [70] λ d wavelength 5.2 mm [70] k d thermal conductivity 0.45 Wm -1 ° C -1 [52] ρ d density 1200 kg m -3 [52] c d specific heat 3300 Jkg -1 ° C -1 [52] ω d perfusion rate 1.25 × l0 -3 m 3 s -1 m -3 tissue [52] Subcutaneous Tissue t s thickness 18000 μ m [52] N s lattice elements 100 ℓ lattice node spacing 180 μ m σ f electrical conductivity 0.585 Sm -1 [70] ε f relative permittivity 4.60 [70] η f penetration depth 19.6 mm [70] λ f wavelength 13.9 mm [70] k s thermal conductivity 0.19 Wm -1 ° C -1 [52] ρ s density 1000 kg m -3 [52] c s specific heat 2675 Jkg -1 ° C -1 [52] ω s perfusion rate 1.25 × 10 -3 m 3 s -1 m -3 tissue [52] Blood c b specific heat 3770 Jkg -1 ° C -1 [52] ρ b density 1060 kg m -3 [52] Surface contact heating The case of a fixed skin surface temperature is relevant to transcutaneous blood gas sensors, in which a skin-contacting sensor with controlled temperature, and a local source of heat of up to 45°C are employed to significantly increase perfusion within the outer capillary bed [ 50 ], thereby "arterializing" the capillary blood. This situation also represents heating at the skin surface by a heat source, or the skin contacting a hot object with a large thermal diffusivity, such as in thermal injury [ 51 , 52 ]. Surface heating may be either essentially constant (long duration) or transient (short duration). The latter is relevant to laser pulse application or flash skin burns. We model surface contact heating by considering step heating of skin surface to different temperatures at t = 0. The core temperature is assumed to be constant at the ambient temperature (T a ). The boundary conditions are shown in Fig. 1 . Spatially distributed heating Spatially distributed heating occurs in skin exposed to penetrating, dissipative radiation such as microwave, ultrasound and laser light [ 51 , 53 ]. These heating methods often involve an exponentially decaying power transmission accompanied by reflection at the interface of regions with different electrical properties. We consider a uniform plane wave incident normally upon the skin surface, with a layer of air included to model the reflection at the skin/air interface. We also account for interference from reflections at the interface of dermis and subcutaneous fat and at the skin/air interface. The average absorbed power density, P ( z , t ), in epidermis and dermis (of thickness d = t e + t d ), in the range 0 < z ≤ d is given by where and and the average absorbed power density in the subcutaneous fat layer ( z > d ) is given by where where P (0, t ) is the power density incident on the skin surface at time t , E (0, t ) is the corresponding electric field amplitude, ( z , t ) is the propagating electric field in the epidermis and dermis, E ( d , t ) is the electric field at the dermis-subcutaneous fat interface, η d and η s are the penetration depths for dermis and subcutaneous fat, λ d and λ s are the wavelengths in dermis and subcutaneous fat, and Z a , Z d and Z s are the intrinsic impedances of air, dermis, and subcutaneous fat, respectively. Note that the incident power, P (0, t ), is expressed as an area density whereas the absorbed power density, P ( z , t ), in the skin is expressed as a volume density. The summation of ( z , t ) in Eq. 2 was carried out to 10 terms, although only the first two terms are significant. The reflection (Γ) and transmission ( T ) coefficients at the skin/air ( sa ) and for skin/fat ( sf ) interfaces and the intrinsic impedances are given by [ 54 ] where σ d and σ s are the conductivities of dermis and subcutaneous fat and ε d and ε s are the permittivities of dermis and fat, respectively (listed in Table 1 ), μ 0 is the permeability of free space, ε 0 us the permittivity of free space, and f = 10 10 Hz. Circuit model of heat conduction A motivation for the transport lattice for heat conduction is the electrical equivalence of heat transport (a diffusion process [ 27 ]). We consider the well known equivalence of electrical and thermal conduction. Heat conduction is described using a thermal resistance, R , which relates the heat flow per unit time Q to the temperature difference Δ T as Q = (1/ R )Δ T . In the case of heat conduction across a cube of thickness ℓ and area ℓ 2 , R c = ℓ/( k ℓ 2 ) = ( k ℓ) -1 where k is the thermal conductivity of the slab material. Heat storage is described by the thermal capacitance, C , which for a slab is C = ρ c p ℓ·ℓ 2 = ρ c p ℓ 3 where ρ is the density of the slab material and c p is its specific heat. The associated thermal relaxation time is τ Q = Q / = ( ρ c p ℓ 3 )/( k ℓ) = ℓ 2 / α , where α = k /( ρ c p ) is the thermal diffusivity. The heat conduction models for different layers of skin are shown in Fig. 1 as R c and C with subscripts identifying the particular skin layer. Circuit model of perfusion Pennes bioheat equation provides an approximate description of heat transport by tissue conduction and by blood flow using a local temperature dependent conduction path to perfusing blood. This additional heat removal is proportional to the local temperature difference T - T a . Here, local heat removal by perfusion is described by a thermal resistor, R p = (ℓ 3 ω m c b ρ b ), connected to a reference node at ambient temperature (Fig. 1 ) where c b is specific heat of blood, ρ b is the density of blood. Circuit model of surface heat loss Unoccluded skin often transports heat across its outer surface via a combination of conduction into a boundary layer of air, convective movement of air, and black or gray body radiation. Because our emphasis here is on conduction and perfusion, we lump these surface transport mechanisms into a single heat transfer coefficient. The numerical value of this coefficient was determined by requiring the initial skin surface temperature to be T s = 33°C (before contacting the skin to a heated surface or applying microwave radiation). The surface heat loss for microwave heating is represented by a series of conduction models ( R ca in Fig. 1 ). For contact heating the surface is initially set to 33°C and then raised to 45°C at t = 0. Circuit model of spatially distributed power deposition Spatially distributed power deposition from 10 GHz radiation is modeled by representing Eqs. 2 and 3 by an equivalent local current (heat flux) source, I z = P ( z , t )ℓ 3 , at each node (Fig. 1 ). That is, each node has a local power input based on Eqs. 2 and 3 multiplied by the local volume. Metabolic heat generation Metabolic heat generation can also be represented by local sources, but these are set to zero in the present models. In a transport lattice model an additional heat source can be introduced at each node to represent heat generation by metabolism. Here, metabolism is assumed to maintain the baseline temperature at a constant value equal to the arterial blood temperature. However, metabolism could also be made a function of temperature. Thermal damage to tissue An Arrhenius rate constant relationship is widely used to estimate cumulative thermal damage associated with burns of tissue, including skin [ 55 - 60 ]. This is equivalent to describing the conversion of a native form molecule to a denatured form by overcoming an energy barrier [ 61 ]. The Arrhenius rate constant-based expression for accumulation of irreversible thermal damage describes the process in terms of a rate at which the native form of a molecule moves to a transition state atop the energy barrier and then a final, denatured state. This simple description assumes that a single damage process, with Ω a dimensionless indicator of accrued tissue damage [ 56 , 62 , 63 ]: where A = 2.9 × 10 37 s -1 is the attempt rate, Δ E = 2.4 × 10 5 J mol -1 is the effective activation energy, ℜ = 8.31 J mol -1 K -1 is the universal gas constant, and T ( z , t ) is the absolute temperature at a given location (here depth). According to Lee and co-workers [ 57 , 58 ], the approximate threshold for the onset of thermal damage is 42°C. We therefore estimate the accumulated thermal damage using where z is the depth into the tissue and t exp is the duration of the exposure. The cumulative damage index, Ω, has been related to tissue damage but can also be interpreted as the fraction of hypothetical indicator molecules that are denatured. Complete epidermal necrosis corresponds to Ω = 1. Although the heat induced damage to skin involves many processes, Eq. 7 is a simple model with zeroth-order kinetics [ 61 ]. Initial and boundary conditions Surface heating The temperature of the surface node is elevated to the indicated temperature at t = 0 for a specified duration. The temperature of the core node deep in the skin is held constant at the ambient temperature of 37°C. In the 2-D case, all the lattice nodes at the skin surface are elevated to the indicated initial temperature at t = 0. Spatially distributed heating The far (left) boundary of the air layer away from the skin is held at 25°C while the core temperature is fixed at 37°C. The thermal current sources with different values is a function of z (Eq. 2), representing power deposition at different nodes, are turned on at t = 0 for a specified duration. This accounts for the spatially distributed power dissipation. Transport lattice solution The transport lattice method employs locally interacting functional models to describe heat conduction, heat sources, removal of heat by perfusion and heat storage that are solved by Kirchhoff's Laws. We use electrical circuits which are mathematically analogous to the thermal processes (Fig. 1 ). The resulting electrical circuits were solved by Kirchhoff's laws using Berkeley SPICE version 3f5 [ 43 , 47 ], yielding currents and voltages of lattice elements. The voltages are converted to equivalent temperatures and displayed as temperature plots and images using Matlab (MathWorks, Natick, MA). A Pentium based computer (2 GHz CPU, 4 GB RAM) was used to obtain the solutions. Results We demonstrate the use of a transport lattice approach to solve bioheat problems, using surface contact heating and spatially distributed heating of skin as illustrations. Method Validation The transport lattice approach is validated theoretically by comparison to an one-dimensional analytical solution of the perfusion equation for a single medium with a homogeneous sink (equivalent to uniform perfusion). In this validation case, the surface of the medium was instantaneously changed to T 1 (= 45°C) while the core was maintained at T 2 (= 37°C). The initial condition assumed that the tissue temperature throughout was 37°C. The steady-state analytical solution to the bioheat equation (Eq. 1) for these conditions is the same as a spatially distributed uniform sink given by the equation [ 27 ] Here L is the length of the model geometry (= 10 cm), ρ b is the density of blood, c b is the specific heat of blood, ω m is the perfusion rate and k is the thermal conductivity of the tissue. The analytical result of Eq. 8 is compared with solutions of a one-dimensional transport lattice model with dermis tissue values assigned to the local models. For validation comparisons, the tissue (10 cm in length) is represented by a lattice with different number of nodes. The transport lattice temperature profiles agree remarkably well with Eq. 8 for different perfusion levels (Fig. 2 ). The performance of the transport lattice method was quantified by the maximum deviation of the transport lattice temperature profile from the analytical result normalized by ( T 1 -37). As seen in Fig. 2 , the numerical error is less than 1% when the geometry is represnted by only 40 nodes and becomes progressively better as more nodes are used. Figure 2 Validation using one-dimensional geometry . The validation model is a homogeneous section of a material (here dermis) with uniform perfusion (homogeneous sink), 10 cm in length and 100 μ m × 100 μ m in area. The surface of the tissue was elevated to 45°C at t = 0 s. The perfusion level (in ml/100 g min) was varied as shown in inset. The solid line represents the analytical solution (Eq. 8) and the symbols represent the transport lattice solution. Top: Steady-state temperature as a function of depth. The 10 cm long tissue is represented by 100 lattice elements, but the temperature profile is shown only to the depth of 6 cm. Bottom: The 10 cm long tissue is now represented by different number of lattice elements. Maximum % deviation as a function of the number of nodes used to represent the tissue is shown. The deviation is computed as the maximum discrepancy between the simulated temperature and the corresponding analytical value normalized to the step increase in temperature (here = 8°C). Surface Contact Heating – Transcutaneous Application In this case, the skin surface temperature is approximated as increasing instantaneously from 33°C to 45°C at t = 100 s. This situation is encountered in transcutaneous blood gas sensing and, for more extreme heating, in skin burns due to a hot metal object and flash fire exposure. Temporal distribution The spatial distribution of temperature and the resulting tissue damage from surface contact heating is shown in Fig. 3 . The steady-state temperature distribution shows an exponential fall off with spatial decay constants dependent on the thermal properties of different layers of skin. Accumulated tissue damage is shown for different perfusion levels in Fig. 3 . As expected intuitively, when the surface contact temperature is elevated, outer layers experience more damage than deeper regions of skin. For typical transcutaneous blood gas sensing conditions the estimated damage is small, even with prolonged skin contact to a 45°C surface. Figure 3 Temperature distribution with surface contact heating . The surface of the skin was elevated from 33°C to 45°C at t = 0 s for 3600 s. This approximates contact heating in which a metal (large thermal diffusivity) -encased heater with controlled temperature is held against the skin. Top: the temperature distribution as a function of skin depth with 10 ml/100 g min perfusion is shown for four different time points (inset in s). The four curves show the temperature profile before the application, immediately after the application, before the removal, and after the removal of the surface heating. Bottom: Tissue damage indicator predicted for the transcutaneous heating for four different perfusion levels (inset in ml/100 g min). Temperature-dependent perfusion Experiments have shown that heat stress causes a temperature-dependent response of the vasculature in tissues [ 64 ]. The blood flow in skin and muscle increases significantly for temperatures up to 43°C. Here temperature dependent blood perfusion in dermis and subcutaneous tissue is represented by ω 0 (1 + γ T ) where ω 0 is the baseline perfusion and γ is the linear coefficient of temperature dependence. Figure 4 shows the temporal distribution of temperature close to skin surface for different values of γ for 1-hr exposure. As expected, increased perfusion causes a decline in local temperature. The accumulated tissue damage (Fig. 4 ) is also lower if the blood perfusion has a higher temperature coefficient, because the temperature rise is constrained. Figure 4 Temperature-dependent perfusion distributions for contact heating . The surface of the skin was elevated to 45°C at t = 100 s for 1 h. The perfusion level was dependent on local temperature with a temperature coefficient shown in inset. The basal perfusion rate was 10 ml/100 g min. Top: Temperature of skin close to surface as a function of time. Bottom: Tissue damage as a function of depth integrated over time (only the damage for two smallest perfusion values are discernible, hence the curves for higher perfusion rates are not seen in the figure). Spatially Distributed Heating by Microwave Exposure The spatially distributed heating case illustrated here relates to heat generation (power dissipation) decaying exponentially with the distance within each skin tissue layer. We analyzed the case of an exposure to 10 GHz microwave for 3 s duration (a short-duration and high power microwave [HPM] exposure [ 21 ]). Applied power level Figure 5 shows the change in skin surface temperature over time for different incident power levels. The skin is exposed to a 1 to 10 W cm -2 10 GHz pulse for 3 s. The layer of air farthest from the skin was set at 25°C and the core (2 cm below the surface) was set to 37°C. This resulted in the skin/air interface having a steady-state temperature of 34°C before the microwave exposure. The skin/air interface has a power transmission coefficient (\| T sa \| 2 Re { Z a / Z e }) of 0.49 at 10 GHz. Applying 10 GHz microwave results in an essentially linear rise in temperature, in agreement with prediction using other methods. When the input power level is less than 5 W cm -2 , the peak surface temperature is less than 42°C. When the microwave exposure is turned off, relaxation of the skin temperature occurs over a time scale of several seconds. Onset of tissue damage occurs when the local tissue temperature reaches 42°C. The distribution of tissue damage with depth is shown for different power densities (Fig. 5 ). Even for an incident power density as high as 10 W cm -2 , the accumulated tissue damage for a 3 s exposure is far less than 1, even in the epidermis and dermis. Because of the large difference in the conductivity and permittivity of the dermis and subcutaneous fat, over 20% of the power deposited at the interface is reflected back into the dermis resulting in reduced power deposition in the fat layer. For power densities 5 W cm -2 and less, the tissue temperature remains less than 42°C and the tissue damage indicator is negligible throughout the skin. This suggests that the tissue suffers no damage during this exposure. Figure 5 Effect of applied power level for spatially distributed heating . Response to a 10 GHz microwave pulse of 3 s duration with four different power densities (inset in W cm -2 ). The layer of air farthest from the skin (2 mm) was at 25°C, the skin surface was at 34°C before the pulse was applied and the core temperature was fixed at 37°C. The blood perfusion level was assumed to be 10 ml/100 g min. Top: Surface temperature as a function of time. Bottom: Tissue damage indicator, Ω, as a function of depth. Only the two highest levels of power generate noteworthy values of Ω (the plots for lower power levels are, therefore, not visible in the figure). Perfusion level The peak surface temperature is shown in Fig. 6 for different blood flow rates. The basal perfusion levels in dermis and subcutaneous tissue were varied from 2.5 ml/100 g min to 20 ml/100 g min. The surface temperature distribution was nearly identical for this range of blood flow rate, a level of 20 ml/l00 g min is already at the high end of physiologic range for skin. This is consistent with the same skin temperature increases at different blood flow rates at 100 GHz reported by Nelson et al. [ 21 ]. Figure 6 Effect of blood perfusion level . A 10 GHz microwave pulse of 3 s duration with power density of 5 W cm -2 was considered for illustrative purposes. Blood perfusion levels in units of ml/100 g min are shown in inset. Skin surface temperature as a function of time is shown for these different perfusion levels. The different plots essentially overlap, showing that blood perfusion has negligible effect on temperature distribution in the case of a 3-sec 10 GHz exposure. Temperature distribution dynamics Change in the spatial temperature distribution over time due to a 10 GHz pulse is shown in Fig. 7 . The temperature of the outer layers of skin is below the core temperature of 37°C before the microwave exposure. During the pulse, the temperature of epidermis and dermis layers increases rapidly compared to deeper subcutaneous tissue. The temperature in the subcutaneous fat layer does not increase appreciably from its initial temperature because only a fraction of the incident power is transmitted into this region of the skin, and although heat absorbed in outer layers is removed partially by conduction, heat in the outer layers is mainly intercepted and removed by perfusion. Figure 7 Temperature distribution dynamics . A 10 GHz microwave pulse of 3 s duration with a power density of 5 W cm -2 was applied at t = 1 s. The layer of air farthest from the skin (2 mm) was set to 25°C, the skin surface was at 34°C before the RF field was applied and the core temperature (here 20 mm deep) was at 37°C. The temperature-independent blood perfusion level was assumed to be 10 ml/100 g min. Temperature change from baseline as a function of distance from skin surface is shown for different time points (2, 3 and 4 s). Skin heterogeneity The local elevated temperature at the interface of dermis and subcutaneous tissue observed in the spatial distribution of temperature during a 10 GHz exposure is due to different thermal properties of the homogeneous slabs that comprise the model. The effect of skin heterogeneity on temperature distribution is shown in Fig. 8 . The specific heat and thermal conductivity of epidermis and subcutaneous tissue were varied relative to each other using a range of published values. In agreement with qualitative expectations, the temperature distribution prior to the end of microwave pulse shows that the larger the difference between the specific heat of the two layers, the larger the magnitude of the locally elevated temperature. However, most of the temperature increase is confined to the epidermis and dermis, as most of the incident power is deposited in those layers. It is expected that at higher frequencies, the temperature distribution in subcutaneous layers will be uniform because most of the RF energy will be deposited closer to the surface of the skin as the penetration depth diminishes. Figure 8 Effect of skin layer parameters on temperature distribution for 10 GHz exposure . Temperature change due to a 10 GHz pulse of 3 s duration with an incident power density of 5 W cm -2 . The layer of air farthest from the skin was at 25°C, the skin surface was at 34°C before this RF field was applied and the core temperature was at 37°C. The blood perfusion level was assumed to be 10 ml/100 g min. The thermal conductivity and specific heat of dermis and subcutaneous tissue were varied relative to each other as shown in the inset. Two-dimensional temperature distribution The use of transport lattice approach for predicting heat transport in spatially heterogeneous structures is further illustrated with a two-dimensional model of the skin. The model is derived from an image of a histological section of skin (Fig. 9 ). The temperature distribution from a thermally insulated wire (20 μ m diameter, 60 μ m length) with a hot tip that is inserted into the epidermis is also modeled. The model assumes that the tip of the metal wire ( k w = 200 W m- 1 °C -1 ; ρ w = 8900 kg m -3 ; c w = 383 J kg- 1 °C -1 ) is enclosed in a thermally insulating material ( k p = 0.15 W m -1 °C -1 ; ρ p = 1200 kg m -3 ; c p = 2010 J kg -1 °C -1 ). The skin model contains stratum corneum, epidermis and dermis. As before, the core temperature (37°C) is fixed at 2 cm from the skin surface by extending the subcutaneous layer. Before heating the wire conducts heat outwardly to the air, consistent with the isotherms. The temperature at the hot wire tip is increased to 45°C at t = 10 s. The temperature contours at different time points is shown in Fig. 9 . The heterogeneity in skin structure is seen in the temperature contours immediately after the tip is heated, but then diminishes with time because the thermal properties of different skin regions differ only slightly. As intuitively expected, the thermal contours show a temperature gradient into the surrounding air when the wire tip is heated. Figure 9 Two-dimensional distribution of tissue temperature . Top Left: Image of skin cross-section used in generating the simulation geometry. Scale bar: 50 μ m . (Top Right). The Cartesian grid (ℓ = 1 μ m ) is superimposed on the model geometry to create the simulation geometry. Stratum corneum was assigned the same thermal properties as the epidermis. The hot wire is seen as dark blue region with a red tip. Center: A part of the 216 × 188 lattice is shown. At each node, five functional models are connected, four conduction models (M c ) to the neighboring nodes and a fifth model (M s ) representing heat storage and perfusion-transported heat to a reference temperature, T a . Depending on the type of underlying tissue (or air), the transport model at any node is one of four models shown in Fig. 1. The tip of the hot wire was elevated to 45°C at t = 10 s for 50 s with a rise and fall time of 5 s. Bottom Panels: Temperature contours at different time points: before the hot wire tip temperature is raised; just before the hot wire is turned off; immediately after the hot wire is turned off; 10 s later. Discussion Temperature-dependent perfusion Both in vivo and in vitro studies have shown that the tissue response to heat stress is strongly temperature-dependent [ 33 , 64 , 65 ]. When heated to 41–43°C, temperatures that are commonly used in clinical hyperther-mia, the blood flow in normal tissues increases significantly [ 35 ]. In order to demonstrate the use of transport lattice approach to model temperature-dependent perfusion, we considered the simplest case of perfusion varying linearly with temperature. The perfusion was assumed to include a temperature-independent basal component and a temperature-dependent vasodilatory component. As shown in Fig. 4 , increased perfusion resulting from temperature dependence results in a lower peak temperature close to the skin surface. An increase in perfusion causes greater heat loss from the tissue into the blood, thus reducing the peak temperature. In addition, the temperature decays faster for a larger temperature coefficient after the removal of external heat source. The coefficient of temperature dependence could also utilize a non-linear function of temperature in the transport lattice method. This could reflect a decrease in perfusion at temperatures over 45°C resulting from heat-induced damage to blood capillaries. A more comprehensive non-linear temperature dependent perfusion model has been applied in modeling hyperthermia. Tompkins et al. [ 66 ] used temperature-dependent models to show that blood perfusions initially increase with tissue temperature and then decrease at higher temperatures. Erdmann et al. [ 35 ] employed a Gaussian profile for temperature-dependence of perfusion increase between 37°C and 45°C, and a plateau for temperatures above 45°C. Our simpler linear dependence of perfusion on temperature is intended to demonstrate the use of a transport lattice method for heat transport in skin. Skin heterogeneity We present a modular approach to modeling in which the skin is represented by three homogeneous layers, each with many interconnected local, steady state models that account for the local heat storage (heat capacity), local heat dissipation (local microwave heating) and local transport by both conduction and perfusion (Fig. 1 ). The existence of different thermal properties in adjacent layers of the model is particularly important for spatially distributed heat sources that extend through skin layers. In the case of a 10 GHz microwave radiation, the penetration depth is approximately 3 mm. Therefore, an exposure to 10 GHz radiation will cause a non-uniform temperature distribution within the skin. As shown in Fig. 8 , differences in thermal conductivity and specific heat of different layers of the skin create different temperature profiles, especially in the subcutaneous fat layer. However, with higher frequency RF radiation, the penetration depth decreases, and the difference in temperature profiles in the skin will diminish. Tissue damage Prolonged exposure to elevated temperatures can cause tissue damage by, for example, protein alteration or denaturation, often followed by recognizable changes in the optical properties of tissue [ 67 ]. The rate of the transition from natural to denatured states is governed approximately by the Arrhenius rate equation (Eq. 7). The lipid bilayer components of the cells are most vulnerable to thermal damage because they are held together only by forces of hydration [ 68 ]. Exposure to ambient microwave fields has been shown to cause tissue damage. The rate of tissue heating has a large dependence on the density of dipoles, resulting in a much slower microwave heating in fatty tissues [ 69 ]. When skin is exposed to a 10 GHz pulse of 3 s duration, the tissue damage indicator near the skin surface may be as high as 0.08, which suggests some damage at high power levels (Fig. 5 ). This exposure generates a surface temperature of approximately 51°C. Human pain perception studies have shown that the threshold for perception corresponds to a significantly lower mean skin temperature of 44°C [ 22 ]. Thus, a relatively non-damaging exposure might cause significant pain. Conclusions Transport of heat by conduction, and by temperature-dependent, spatially heterogeneous blood perfusion, is predicted using a transport lattice model. This approach uses interconnected, local, steady state models for transport and storage, to together represent the Pennes bioheat equation. The thermal system model of the skin was solved by exploiting the mathematical analogy between local thermal models and local electrical (charge transport) models, thereby allowing robust, circuit simulation software to obtain solutions to Kirchhoff's laws for the system model. The skin model has a nonperfused viable epidermis, and deeper regions of dermis and subcutaneous tissue with perfusion that was constant or temperature-dependent. Spatially distributed heating and surface heating cases were considered. Accumulated thermal damage was estimated by using an Arrhenius type relation at locations where viable tissue temperature exceeds 42°C. Prediction of spatial temperature distributions was also illustrated with a two-dimensional model of skin created from an image. Validation of the transport lattice approach using experimental data is necessary for practical application of this method. Authors' Contributions TRG constructed and solved the several transport lattice models and wrote much of the manuscript. DAS computed the reflected and transmitted power in the skin layers, contributed to construction and solution of the models, and to writing of the manuscript. GTM provided guidance and advice with respect to thermal modeling, and helped write the manuscript. JCW conceived the local transport lattice model for solving the bioheat equation, provided overall guidance and helped with interpretation of results and writing the manuscript. All authors read the final manuscript. Table 2 Definition of model parameters used in the transport lattice simulations. The parameter values were obtained from the sources cited in the rightmost column. Air R ca conduction model = ( k a ℓ) -1 C a storage model = ρ c a ℓ 3 Epidermis R ce conduction model = ( k e ℓ) -1 C e storage model = ρ c e ℓ 3 I e distributed local power = P ( z , t )ℓ 3 Dermis R cd conduction model = ( k d ℓ) -1 C d storage model = ρ c d ℓ 3 R pd conduction model = ℓ 3 ω m c b ρ b I d distributed local power = P ( z , t )ℓ 3 Subcutaneous Tissue R cs conduction model = ( k s ℓ) -1 C s storage model = ρ c s ℓ 3 R ps conduction model = ℓ 3 ω m c b ρ b I s distributed local power = P ( z , t )ℓ 3	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544831.xml
544945	Why do we need a new journal in autoimmunity?	A new online Journal of Autoimmune Diseases is created as an independent open access journal. In addition to the obvious advantages of the open access, the Journal will practice a double-blind reviewing of the manuscripts, which means that both the reviewers and the authors remain anonymous to each other. We believe that such a policy will reduce the influence of personal and other non-scientific factors on the reviewer's decision making.	Autoimmune diseases are common and can affect virtually every organ in the body. They range from organ specific diseases such as thyroiditis or type 1 diabetes to life-threatening multi-system diseases such as systemic lupus erythematosus and the systemic vasculitides. Clinicians from every field of medicine may encounter these patients and both generalists and specialists need to keep up to date with clinical and experimental developments in autoimmunity. The advent of targeted therapies for example the anti-TNF-α agents for the treatment of rheumatoid arthritis exemplify the application of basic science research that can lead to effective therapies. Inspired by the publisher, BioMed Central, a group of enthusiastic professionals gathered to launch the Journal of Autoimmune Diseases . There are several journals devoted to autoimmune diseases already, with at least four having the word "autoimmunity" in the title. In addition, a number of immunological journals and journals that specialise in particular diseases publish papers on autoimmunity. So is there a need for yet another journal? The Editorial Board of Journal of Autoimmune Diseases certainly thinks so. This Journal is one of the new breed of online-only journals which are proving extremely successful. But it is not the magic word "internet' that makes the difference. Two distinct features provided by, BioMed Central and the Editorial Board offer special advantages to readers and authors. Open Access The aim of publishing is to share information with the community. Authors are also keen to know that their work is being read. One of the syndromes of authorship has been the little card politely requesting a reprint. This card is sometimes forgotten or filed in the circular filing cabinet thus inducing considerable guilt both in the paper's author and the recipient. The era of reprint requests may be drawing to a close since most of the journals can now be found on-line. However, the access to a paper on-line often requires a subscription or a purchase of an individual article ("pay-per-view"). The high publication costs of printed journals deprive their publishers of the generosity of complete free access, but this tends to limit the dissemination of research. An alternative approach provided by BioMed Central is to make access totally free for everybody, as part of their Open Access policy [ 1 ]. It means that Journal of Autoimmune Diseases is universally and freely available online to everyone, its authors retain copyright, and it is archived in internationally recognised free repositories such as PubMed Central [ 2 ]; e-Depot [ 3 ]; Potsdam [ 4 ] and INIST [ 5 ]. Journal of Autoimmune Diseases enables scientists from countries and institutions with limited funds to read the same material as wealthier ones [ 6 ]. This has become possible because lower publication costs can be covered by the authors. The advantage for the authors is also obvious: access to their work is much more freely available throughout the world [ 7 ]. Double-blinded peer review The Journal of Autoimmune Diseases aims to provide a high standard of double-blinded peer review, in which the reviewer's name is not disclosed to the authors, and the authors remain anonymous to the reviewers. A few journals already use this system, so why do we believe that this will benefit the scientific community? We have all come across situations where an honest, laborious piece of work was hard to publish. Some journals reject up to 95% of manuscripts [ 8 ], and this includes good papers. The problem here is that the decision made by a reviewer is not always based solely on the scientific merit of the paper. This may cause either of two antiscientific consequences, unfair rejection or unfair acceptance. It is the nature of human beings that if a reviewer does not like you personally , a good paper is rejected or additional possibly unnecessary experiments are required which will take a year or two. Unfair rejection can occur due to a conflict of interest, be it personal or financial. For instance, an individual may be rejected on the basis of ethnicity [ 9 ], so may his paper be. An unfair acceptance takes place when the reviewer is benevolent to a weak paper that has an outstanding name as the last author. Some authors may successfully exploit this weakness of a reviewer's human being by placing the famous name intentionally. This results in a ghost authorship of which the celebrity may be unaware! One of us remembers an anecdotal conversation he heard a few years ago in one of the universities: Dorit (a secretary to the Professor): Robert? There is a new paper of yours I see on Medline. Do you wish me to update your publication list? Professor: Who are the other authors? (Dorit reads 20 last and first names) Professor: Hm-m-m...I don't know any of these people. What is the title there? (Dorit reads the title) Professor: I don't remember even discussing anything like that or being consulted... Dorit: Should I ignore it then? Professor: Could you print it out for me? I will read it first. If the paper is good enough...Well, then I'll have nothing to do but to add it to my list of publications. What are the remedies then? Two policies that could help to overcome these difficulties are completely open peer review and completely anonymous, or double-blind peer review. In open peer review [ 10 ], the reviewers' names are disclosed to the authors and vice-versa, ensuring accountability. Obviously, this might help against unfair rejections, but not against unfair acceptance. In addition, there is a danger that younger reviewers will be intimidated and the political power of the established will be increased [ 8 ]. We believe that a double-blinded review process will be much more effective in helping to avoid rejection of good papers and acceptance of unsound manuscripts for subjective, political or other non-scientific reasons. Journal of Autoimmune Diseases We hope to attract clinical and basic science reports from leading and innovative authors. We call for established authors to publish with us in order to be easily accessible by all other scientists, to stop fighting with the windmills of the elite journals, to write detailed and clear papers in the unlimited space provided by internet. We would also strongly encourage more junior authors who are making their way in this field to consider publishing high quality science in the Journal of Autoimmune Diseases – we all need that important first paper to get going. The Editorial Board has been carefully selected from leading authorities in their own fields to help achieve our ambitious aim of developing a high class international journal that is accessible by all who have internet access and we look forward to receiving your submissions.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544945.xml
340950	Dissecting the Urge to Create	A neuropsychiatrist reviews "The Midnight Disease: The drive to write, writer's block and the creative brain"	Creative human beings are the torch-bearers of civilization. How does their creativity arise? What causes some minds/brains to achieve awe-inspiring artistic or scientific achievements? We cannot help but be fascinated by the fact that Shakespeare—a merchant's son with “small Latin and less Greek”—could emerge from the “nowhere” of rural Stratford to create the richest literary treasure in the English language. We wonder how Michelangelo—a stonecutter's son who also came from a rural nowhere—found within himself the vision to see the shape of David in a block of discarded marble or the apolcalyptic fresco of The Last Judgment on the wall of the Sistine Chapel. What genetic influences shaped their brains to create—and to create these very specific wondrous things? How did their environments promote or impede them? Would Michelangelo have been great without the patronage of the Medicis or the competitive edge induced by Leonardo? Great art and great science are indeed often forged in the smithy of pain—with the fire fueled by self-doubt, obsessive preoccupation, sorrow, depression, competition, or economic needs. The Midnight Disease: The Drive to Write, Writer's Block, and the Creative Brain by Alice Weaver Flaherty unites two intrinsically fascinating domains of knowledge—the workings of the brain and the nature of creativity. Its author, a neurologist who has also become a writer by virtue of having published her first nonacademic book, draws on her knowledge of neuroscience, her medical career as a clinician, and her experiences as a patient. Early in the book, she describes her own hospitalization for manic-depressive illness, a disclosure that implicitly places her in the pantheon of other artists who have suffered from serious mental illness and provides her with lustre-by-association. The result of all these juxtapositions is, however, a somewhat disconcerting blend of pop-science and pop-confessional genres. The author frequently talks to us in the first person, but one is not quite sure which person (the neuroscientist, the doctor, or the patient) is actually speaking. In other words, this book has a jarring lack of a strong single voice, despite a knack for often finding a fine turn-of-phrase or a clever word choice. Given that the book topic is promising and that the author can often write very well, it is dismaying that this book is not better than it is. It is written for the intelligent lay public, many of whom avidly collect and read “brain books” to expand their minds. Most painful is the fact that this book is filled with factual errors, glib and misleading generalizations, and careless misstatements. Perhaps most shocking and most erroneous, we are told (by a neurologist!) that “The tips of the temporal lobe can be lopped off without much changing a person's behavior.” HM, the most famous patient to receive bilateral temporal lobectomy, remains frozen in a past linked to a never-changing present because he lost the capacity to retain new memories. Temporal lobe syndromes are discussed more accurately later in the book, but that is a weak excuse for this early error. We are also told that “manic depression is a genetically transmitted syndrome” (when, in fact, no replicable genetic loci have yet been identified), that “a very high proportion of manic depressives become writers” (the lifetime prevalence rate of bipolar disorder is approximately 1%, and only a tiny proportion of that 1% are writers), and that “electrophysiology, because it is dangerous, is rarely performed” (electrophysiology tools—e.g., the study of evoked potentials or electroencephalograms—are noninvasive and frequently used; recordings of the activity of individual neurons with electrodes placed in the gray matter are indeed rare, but nothing from the context suggests that this particular type of electrophysiology is being discussed). There are many more such careless misstatements. The intelligent lay reader deserves better than this. The book raises and addresses a variety of interesting questions that have intrigued many thoughtful people for more than two millennia. What is the nature of creativity? What is the difference between skill and creativity? What is the relation between mental illness and creativity? Is creativity inhibited when mental illnesses are treated? What is the relation between mind and brain? The book also addresses some unique and interesting twists on these questions. Its focus is the domain of writing, drawing from the author's own experience of a compulsion to write, or hypergraphia, following a pyschic break. What is the relationship between hypergraphia and the brain? Between writer's block and the brain? Are these problems always pathological, or do they sometimes enhance creativity? Does that college student who can't finish a term paper have a “disease”? Can “mind-expanding” drugs that affect the brain enhance creativity? In short, The Midnight Disease raises many important questions, but fails to address them completely and accurately. There is much more to learn, and much more to say, about the nature of creativity, its origins in the mind/brain and in the human genome, and its boundaries with health and disease.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC340950.xml

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.