Learn more: PMC Disclaimer | PMC Copyright Notice
From the Cover
CD39 deletion exacerbates experimental murine colitis and human polymorphisms increase susceptibility to inflammatory bowel disease
Abstract
CD39/ENTPD1 hydrolyzes proinflammatory nucleotides to generate adenosine. As purinergic mediators have been implicated in intestinal inflammation, we hypothesized that CD39 might protect against inflammatory bowel disease. We studied these possibilities in a mouse model of colitis using mice with global CD39 deletion. We then tested whether human genetic polymorphisms in the CD39 gene might influence susceptibility to Crohn's disease. We induced colitis in mice using Dextran Sodium Sulfate (DSS). Readouts included disease activity scores, histological evidence of injury, and markers of inflammatory activity. We used HapMap cell lines to find SNPs that tag for CD39 expression, and then compared the frequency of subjects with high vs. low CD39-expression genotypes in a case-control cohort for Crohn's disease. Mice null for CD39 were highly susceptible to DSS injury, with heterozygote mice showing an intermediate phenotype compared to wild type (WT). We identified a common SNP that tags CD39 mRNA expression levels in man. The SNP tagging low levels of CD39 expression was associated with increased susceptibility to Crohn's disease in a case-control cohort comprised of 1,748 Crohn's patients and 2,936 controls (P = 0.005–0.0006). Our data indicate that CD39 deficiency exacerbates murine colitis and suggest that CD39 polymorphisms are associated with inflammatory bowel disease in humans.
Inflammatory bowel disease (IBD) results when the complex immune balance in the bowel environment is disrupted. Normally, the immune system suppresses inflammation in the presence of resident gut bacteria or food antigens but can respond rapidly to potentially pathogenic bacteria or viruses with vigorous inflammatory responses (1). In IBD, gut immunity is episodically activated in the absence of a clear “danger signal,” leading to a constellation of symptoms including abdominal pain, diarrhea, gastrointestinal bleeding, venous thrombotic complications, and often malnutrition (1). Strong genetic components contribute to an individual's susceptibility to IBD, particularly Crohn's disease (2).
CD39, also known as ENTPD1, is a vascular and immune cell ectonucleotidase that converts extracellular ATP and ADP to AMP. Since ATP is typically proinflammatory and AMP is rapidly converted by CD73 into the largely anti-inflammatory metabolite adenosine, CD39 tends to promote an anti-inflammatory and immune suppressive milieu. Within the immune system, CD39 is expressed on cells of both the innate and adaptive immune systems, including T cells (particularly T-regulatory cells) (3), B-cells (4), NK cells (5), NKT cells (6), dendritic cells (4), monocytes (5), and macrophages (7).
Studies in our laboratory indicate a particularly important role for CD39 and CD73 in the tandem generation of adenosine with subsequent activation of adenosine A2a receptors (3). Mice null for CD39 have heightened expression of proinflammatory mediators such as IFN gamma, IL-1ß, IL-6, and TNF (8). These mutant mice are also prone to increased injury in a wide range of acute, subacute, and chronic vascular inflammatory models including transplant rejection (9), ischemia-reperfusion (10), and diabetic microvascular disease (11).
Given the prominent role of CD39 in the response to inflammation and adaptive immunity, we asked whether CD39 might participate in regulating the complex immune balance that governs susceptibility to inflammatory bowel disease (IBD). We tested this hypothesis in mice null for CD39 by administering Dextran Sodium Sulfate (DSS), an experimental form of IBD, and characterized these mice clinically, histologically, and for markers of immune activation.
Recent advances in genetics and genomics, such as the HapMap and the advent of genome wide association studies, have generated a wealth of human genetic data. Using data from gene expression profiling studies and Crohn's case-control studies, we are able to identify human SNPs strongly associated with CD39 mRNA expression and have determined the effect of these SNPs on susceptibility to Crohn's disease.
Results
Initial Conditions.
We studied adult wild-type (WT), heterozygous (hz) for CD39, and CD39-null (KO) C57BL6 mice at 16–22 weeks of age. Mice in the treatment groups for DSS (WT 37.9 ± 1.0 g vs. hz 34.9 ± 3.0 g vs. CD39-null 35.4 ± 1.2 g; P > 0.05, n = 5 for each group) had comparable body weights at initiation of the study.
Clinical Course of DSS Colitis.
DSS treatment of C57BL6 mice resulted in acute colitis in all mice. We evaluated the clinical severity of the colitis using a disease activity index (DAI) (Fig. 1A), which incorporates weight loss, stool consistency, and GI bleeding (Fig. 1B). CD39-null mice had significantly worse colitis than WT mice starting on day 2 and continuing through day 7. Heterozygote mice were intermediate between WT and CD39-null at all time points, but did not differ significantly from either group at any time point (Fig. 1A).
Experimental assessment of acute DSS colitis. (A) Disease activity index (DAI) correlated with severity of disease and CD39-null mice displayed worse DAI compared to WT DSS colitis. (B) Definition of disease activity index. (C) Hematocrit (Hct) values correlated with the pathological findings. A drop in the Hct occurred in all mice, but WT mice maintained a significant higher Hct than heterozygous or CD39-null mice.
We measured the hematocrit in each group as another indicator of disease severity. There is no difference at baseline in hematocrit (approximately 50) between WT and CD39-null mice (12). However, colitis resulted in decreased hematocrit for both CD39-null (27.8 ± 0.8%) and heterozygous (28.0 ± 0.4%) mice compared to WT mice (33.8 ± 1.0%; P < 0.012 for both comparisons (Fig. 1C)).
Pathology.
Next we examined bowel from mice in each group. Histologically, CD39-null mice displayed severe transmural inflammation with dense wall thickening, profound leukocytic infiltration and ulceration of colonic mucosa (Fig. 2C), when compared to heterozygous (Fig. 2B) and WT mice (Fig. 2A). We also examined myeloperoxidase activity in the colons of the mice as a quantitative index of inflammation. CD39 null mice showed significantly increased MPO activity (P < 0.01) when compared to WT mice, with heterozygous mice demonstrating an intermediate level of inflammation (Fig. 2D).
Acute DSS colitis pathology and myeloperoxidase (MPO) activity. (A) WT mice showed thickening of the colon wall with discrete inflammatory infiltrates, but the architecture and polarity of tissue layers were maintained. (B) DSS caused moderate inflammation in heterozygous mice and (C) severe inflammation in CD39-null mice, affecting the epithelial and subserosal layer in heterozygous and severe inflammation in CD39-null mice, with mononuclear cell infiltrates into the muscularis. (D) Leukocyte recruitment into the colonic wall during DSS colitis was determined by a standard MPO activity assay. DSS treatment was associated with a significant (P < 0.01) increase in MPO activity in CD39-null mice.
Apyrase Rescue.
We pretreated CD39-null mice with parenteral apyrase, a soluble factor with enzymatic activity essentially identical to CD39, before chronic exposure with DSS and used weight as a proxy for colitis severity. Mice receiving apyrase before DSS lost no weight compared to mice receiving only placebo (38.0 ± 1.0 g vs. 37.7 ± 2.3 g), while mice receiving only DSS experienced significant weight loss (32.8 ± 0.8g; P < 0.05 compared to both groups) (Fig. S1, n = 6–10 per group).
Real-Time PCR of CD39 mRNA by Genotype: High Vs. Low ENTPD1 Expressors.
We analyzed available whole-genome expression profiling data in public repositories to generate hypotheses about SNPs in the ENTPD1 gene associated with CD39 mRNA expression levels in humans. Several SNPs in strong linkage dysequilibrium appeared to correlate with CD39 expression, and we chose rs10748643 as the tag SNP because of its proximity to the CD39 promoter (+850 base pairs into intron 1, a region often associated with transcriptional regulation). We directed tested cell lines from HapMap subjects of European ancestry who were either AA (n = 6) or GG (n = 6) at rs10748643 for CD39 mRNA expression. GG homozygotes had mRNA levels 43% higher (P < 0.0004), with no overlap between groups (Fig. 3).
CD39 mRNA expression by genotype at rs10748643. CD39 mRNA expression levels (by real time PCR) were normalized for ribosomal 18S subunit expression level in HapMap lymphocytes. CD39 expression from cell lines from subjects with GG alleles at rs10748643 was higher than expression from subjects with AA alleles. There was no overlap between groups.
High Vs. Low Expressors Across Four Population Groups.
After publication of complete HapMap gene expression profiling sets, we examined CD39 mRNA expression in 210 HapMap subjects from four diverse populations to confirm our in vitro findings in a larger set of European subjects and to determine if this expression phenotype could be generalized to people of other ancestries. In Europeans, Africans, Chinese, and Japanese subjects, rs10748643 genotype correlates strongly with CD39 expression. GG is associated with much higher levels than AA (46–77% increase), with carriers of the AG genotype displaying intermediate expression levels (Fig. 4 and Tables 1 and and2).2). In Chinese subjects, the GG genotype was uncommon and AG and GG genotypes were pooled for analysis. For all populations combined, the G allele was associated with higher expression levels with a P value on the order of 10−22 using an additive, linear model (13). These results were confirmed in a second genome wide study, with similarly strong association (10−28) (14).
CD39 expression in 210 subjects across four populations. (A) CD39 mRNA expression in cell lines from subjects with Caucasian/European (CEU), Yoruba (YRI), Chinese (CHB), or Japanese (JPT) ancestry with each rs10748643 genotype. For Chinese subjects, AG and GG were combined because there were only two GG subjects.
Table 1.
HapMap cell lines from 4 ancestral populations
| Genotype CD39 | CEU | YRI | JPT | CHB |
|---|---|---|---|---|
| AA | 9 | 10 | 20 | 26 |
| AG | 30 | 35 | 19 | 17 |
| GG | 21 | 15 | 6 | 2 |
Number of cell lines per genotype in each ancestral population.
Table 2.
CD39 expression in HapMap cell lines
| Relative expression | ||||
|---|---|---|---|---|
| CEU | YRI | JPT | CHB | |
| AA | 1.00 | 1.00 | 1.00 | 1.00 |
| AG | 1.34 | 1.13 | 1.21 | 1.20 |
| GG | 1.77 | 1.46 | 1.61 | 1.72 |
Relative CD39 expression for each population based on genotype at rs10748643.
Association of a SNP Tagging ENTPD1 Expression with Crohn's Disease.
Next, we asked whether this SNP strongly correlated to CD39 expression might influence human susceptibility to colitis, as we saw in our mouse models. We obtained data from the Wellcome Trust Case Control Consortium study in Crohn's disease to test this hypothesis. At rs10748643, there was significant enrichment of the low CD39-expressing AA genotype in Crohn's disease cases, while controls were enriched for the higher CD39-expressing GG allele (Table 3). Using the Wellcome Trust Case Control Consortium (WTCCC) expanded control panel (all without IBD), we noted an increase in statistical significance in additive, genotypic, allelic, and homozygote vs. homozygote models (Table 3). The odds ratio for the A allele was 1.13 (1.04–1.23) or 1.14 (1.06–1.22) using the standard control or expanded control panel, respectively.
Table 3.
Association of rs10748643 with Crohn's disease in a case-control cohort
| AA | AG | GG | Additive: P | Allelic: P, O.R. | Genotypic: P | AA vs GG: P, O.R. | |
|---|---|---|---|---|---|---|---|
| Crohn's cases | 391 (22.4%) | 885 (50.6%) | 472 (27%) | ||||
| Controls | 577 (19.7%) | 1472 (50.1%) | 887 (30.2%) | 0.005 | 0.005, 1.13 (1.04–1.23) | 0.0197 | 0.005, 1.27 (1.07–1.51) |
| Expanded controls | 2108 (19.9%) | 5236 (49.4%) | 3,256 (30.7%) | 0.0006 | 0.0007, 1.13 (1.05–1.22) | 0.0026 | 0.0008, 1.28 (1.11–1.48) |
Frequency of AA, AG, and GG genotypes at rs10748643 in control subjects and patients with Crohn's disease from the Wellcome Trust Case Control Consortium. Expanded controls include additional subjects without Crohn's disease from the WTCCC (see Methods). O.R.=Odds Ratio, with confidence intervals. The additive model is a 1-degree of freedom χ2 test for trend. The allelic and homozygote models are 1-degree of freedom χ2 tests. The genotypic model is a 2-degree of freedom χ2 test.
Discussion
IBD may represent a tipping of the immune balance toward excessive inflammation in genetically susceptible individuals. Crohn's disease in particular has been associated with a skewing toward a Th1/IFN gamma-driven inflammatory response (1). Interestingly, CD39-null mice also exhibit a general tendency toward exaggerated Th1/IFN gamma responses (15).
In the selected mouse model (acute DSS exposure) that recapitulates important cardinal features of colitis in IBD such as diarrhea, weight loss, and gastrointestinal bleeding, we observed that CD39-null mice suffer more severe injury than WT mice. Although heterozygous mice did not differ from either group in our overall index (DAI) of IBD-activity, these mice did show a potential CD39 dose effect in limiting inflammation. DSS-colitis has traditionally been associated with features of both Crohn's disease and ulcerative colitis in rodents, with a bias toward the former.
In CD39-null mice, injury spanned the bowel wall, a typical and critically important feature of Crohn's disease that is not seen in classic ulcerative colitis (16). Examination of large bowel in the treated mouse groups revealed substantially more severe histological lesions in the null mice, with loss of crypts and a far more vigorous inflammatory response. Increased MPO activity in CD39-null mice provided further evidence of this heightened inflammatory response, emphasizing the role of infiltrating monocyte-macrophages in this disease process. Treatment of CD39-null mice with soluble apyrase could reverse weight loss from chronic DSS treatment, implicating the nucleotide hydrolyzing capacity of CD39 as the functional protective mechanism.
Immune system dysregulation is generally considered the major cause of IBD, but there is a growing appreciation for the role of the vasculature in modifying this response (17–19). CD39 is both the dominant immune cell and vascular ectonucleotidase, so its protective effect against DSS colitis in mice may be due to either of these two roles, or perhaps both. CD39 is important for Treg cell function through generation of adenosine (3), and we would propose this function as the most likely factor in our model. However, recently others have shown that CD39 may specifically modify vascular regulation in murine colitis (20). We expect tissue specific deletion of CD39 in immune cells or vascular cells will answer this question definitively in the future.
In mice, several experimental models have been published showing involvement of purinergic signaling in IBD and toxin-induced colitis. The ultimate products of ectonucleotidases are nucleosides/adenosine that activate type 1 A2A adenosine receptors on immune and vascular cells, markedly attenuating intestinal inflammation in some animal models of IBD (21, 22). CD39 is not expressed on epithelial surfaces but rather has similar tissue distribution to the A2A receptor, and the similar response to colitis in CD39- and A2A-null mice suggests a functional interaction. A2A activation protects against colitis in these studies, associated with decreased production of inflammatory cytokines such as TNF-α and IFN gamma. Notably, these cytokines are upregulated in CD39-null mice (23). In contrast, conflicting recent reports show paradoxically that activation of another adenosine receptor, A2B, may be either pathogenic or of benefit in mouse colitis models (24–26). A2B is a lower affinity adenosine receptor that is also expressed on the epithelial surface and has differing effects to the A2A receptor.
Quite recently, investigators have shown that luminal ATP produced by gut bacteria plays a critical role in Th17 T-cell immune deviation in the lamina propria (27). Administration of ATP worsened and luminal administration of apyrase ameliorated colitis in mouse models (27). These findings suggest that more efficient ATP scavenging in the gut environment by the host organism might protect against colitis. The integral role of the ATP receptor P2×7 in the inflammasome complex (28), a critical coordinator of innate immunity, also suggests a potential candidate pathway that may mediate this effect. Other studies have suggested roles for various P2 receptors in colitis, particularly P2×3, but little consensus has emerged (29–32).
We sought further evidence for relevance of the purinergic system to human IBD by identifying SNPs associated with high or low CD39 expression. We took advantage of the fact that CD39 was initially described as a B-cell activation marker (4), and that B-lymphoblast cell lines are available for all HapMap subjects, along with genotyping at several million SNPs. Initially we identified rs10748643 due to its proximity to the promoter, but because of high LD between this SNP and several dozen others, we are not able to determine if rs10748643 is the functional SNP or a SNP that tags the functional SNP. In either case, it does appear to distinguish high versus low CD39 expressors. We suspect the GG genotype leads to an increase in expression in Caucasians of closer to the 77% calculated from our bioinformatic studies than the 43% observed in our in vitro work, due to chance inclusion of the highest CD39-expressing AA cell line in the CEU HapMap collection in our abbreviated panel of six AA cell lines. Nonetheless, all six AA cell lines had lower CD39 mRNA expression than all six GG cell lines tested. The correlation with this set of SNPs tagged by rs10748643 and CD39 expression was consistent in cell lines across four population groups, although the frequency of the A vs. G alleles varied widely between ancestral groups. Previously, we have shown that CD39 expression levels in B-lymphoblasts correlate well with ATPase and ADPase functional activity (33).
We examined the effect of rs10748643 in data obtained from a carefully collected and phenotyped case-control study of Crohn's disease in people of European ancestry. We found that control subjects were more likely to have the high CD39-expressing GG genotype, and Crohn's patients were more likely to have the low-expressing AA genotype. Although the effect size was modest in terms of odds ratio (about 1.27 when comparing homozygote groups), the high frequency of the risk (A) allele, about 40% in Caucasians, means that the effect of this variant on populations as a whole (population attributable risk) may be more important than variants in other genes with higher relative risks but very low frequencies. Conversely, there may be rare CD39 variants with greater effects on expression level and more impact on relative risk.
Although the P value of our variant is also modest, we think this result is best appreciated from the perspective of Bayesian probability. Rather than ask 500,000 simultaneous tests with equal prior probability of significance, we first identified a gene with functional importance in the specific disease process in an animal model. Secondly, we found a SNP that strongly correlates with expression of that functional gene. Given these first two pieces of evidence, the prior probability of this SNP's effect on Crohn's disease is markedly increased compared to a randomly chosen SNP. In this Bayesian context, a P value of 0.005 (or 0.0007 using the expanded controls) provides strong evidence of an effect of this SNP in human Crohn's disease.
While many studies identify genetically modified mice susceptible to an injury model, and many others find SNPs associated with human disease, it is uncommon when both functional data in mice and studies demonstrating relevance in humans are in alignment. We believe that evidence supporting both of these criteria makes our study particularly interesting, especially since the SNP we identified definitively alters gene expression levels. Our data are particularly intriguing given the links between inflammation and thrombosis in IBD (34) and the observation that CD39 was first characterized as a vascular thromboregulatory factor (35). Our future aim is to further define the pathogenetic mechanisms underlying the role of CD39 in IBD, conclusively identify the specific functional SNP and molecular gene regulation mechanisms, and analyze more patients to rigorously establish the genetic association of this CD39 polymorphism with Crohn's disease in humans.
Materials and Methods
Animals.
Pathogen-free C57BL6 CD39-null and matched wild-type (WT) mice were studied in accordance with standard institutional animal welfare guidelines. The derivation and characterization of CD39-null mice have been described elsewhere (12). Genotyping of the CD39 alleles was performed as previously described (12). Animal care and experiments were carried out under the guidelines and protocols approved by the Animal Care and Use Committee, Beth Israel Deaconess Medical Center, Harvard University, Boston.
Induction and Clinical Assessment of Experimental (Acute) Colitis.
Dextran Sodium Sulfate (DSS) colitis. CD39-null (n = 8), heterozygous for CD39 (n = 11), and WT (n = 12) mice were equally treated with 3% DSS in standard drinking water. There was no restriction regarding the dose of DSS solution that was provided ad libitum for 7 consecutive days. Control mice received standard drinking water. Body weights, hemoccult, gross blood, and stool consistency were analyzed on a daily basis. Disease activity index (DAI) was calculated by scoring percent weight loss, intestinal bleeding [no blood, occult blood (hemoccult +), or gross blood], and stool consistency (normal stool, loose stool, or diarrhea), as previously described (36). The clinical features were scored separately and then correlated with a histological score: DAI = (body weight loss) + (diarrhea score) + (rectal bleeding score) (Fig. 1B).
Chronic DSS Colitis.
Sixteen- to eighteen-week-old CD39 null mice were further studied with a longer course of DSS, with or without apyrase to reconstitute CD39 function. CD39-null mice were treated with four 1-week intervals of DSS in standard drinking water followed by 1 week off DSS treatment. The total duration of the treatment was 7 weeks. We treated with parenteral apyrase and compared three different conditions: CD39-null mice with DSS (n = 10), CD39-null mice with DSS and apyrase (n = 10), and CD39-null mice receiving only apyrase (n = 6). We used grade-VII apyrase (derived from S tuberosum) purchased from Sigma at a dose of 2.5 U/g twice a day by i.p. injection. This apyrase has comparable (approximately1:1) ATPase and ADPase activity, paralleling the nucleotide preference of CD39.
Histological Assessment of Colitis.
Specimens were embedded and snap-frozen as previously described (37) and stored at −80 °C. Four- to five-micrometer tissue sections were stained with hematoxylin and eosin. Stained sections were examined for evidence of colitis as previously published, (38) using as criteria the presence of lymphocytic infiltration, macrophages or polymorphonuclear cells, elongation and/or distortion of crypts, crypt abscesses, reduction in goblet cell number, ulceration, and edema formation of the colon wall.
Colonic Myeloperoxidase (MPO) Activity.
Colon samples were obtained from control and DSS treated animals, and prepared as previously decribed (39). The samples were thawed for MPO activity assay determination according to the o-dianisidine method as previously described (39). In brief, tissue samples were thawed, weighed, suspended in 50 mM potassium phosphate buffer (Kpi), pH 6.0, containing 0.5% hexadecyltrimethylammonium bromide buffer (0.1 g/20 mL Kpi), and homogenized. After sonication the sample was microcentrifuged. The reaction was started by mixing and incubating the supernatant in 50 mM Kpi, 20 mg/mL o-dianisidine dihydrochloride, and 20 mM hydrogen peroxide. The reaction was stopped by adding 2% sodium azide. The change of the absorbance was monitored at 460 nm for 10 min. MPO activity was expressed as the amount of enzyme necessary to produce a change in absorbance of 1.0 per min/g wet weight of colonic tissues.
Cell Culture.
Lymphoblasts from HapMap participants of European (CEU) ancestry were purchased from Coriell Institute (www.coriell.org). Genotypes for each cell line were obtained from www.hapmap.org. Cells were grown in RPMI 1640 media with 2 mM L-glutamine, 15% FBS, penicillin-streptomycin mix, and 10 mM HEPES buffer. Cells were seeded at 200,000 per mL and harvested 24–48 h later in log-phase growth.
Real-Time PCR.
mRNA was isolated using RNeasy (Qiagen) according to the manufacturer's instructions. mRNA (0.5 μg) was reversed transcribed to cDNA using the Taqman Reverse Transcription Kit (Applied Biosystems). Probe-primer sets for ENTPD1 and the 18S ribosomal subunit (loading control) were obtained from Applied Biosystems. Real-time PCR was performed with an Applied Biosystems 7900 system. ENTPD1 mRNA expression was normalized using 18S expression. mRNA was quantified twice from each cell line and averaged.
Bio-Informatic Assessment of ENTPD1 Expression.
Expression data for 210 HapMap cell lines was downloaded from the Wellcome Trust Sanger Institute (http://www.sanger.ac.uk/humgen/genevar/). The methods for both genotyping and expression profiling of these cell lines have been described in detail (13). ENTPD1 expression level for each genotype was calculated. We used log2-transformed data for tests of statistical significance.
Crohn's Disease Genetics.
Genotypes at rs10748643 for both control and Crohn's patients in the Wellcome Trust Case Control Consortium (WTCCC) were downloaded from http://www.wtccc.org.uk/. Data were available for 1,748 Crohn's cases and 2,936 apparently healthy controls. We conducted a second analysis of the data using an expanded control panel (10,600 subjects) that included the healthy controls plus additional subjects from the Wellcome Trust case cohorts. We included hypertension, coronary artery disease, bipolar disorder, and type 2 diabetes cohorts that are not expected to have disease SNPs in common with Crohn's disease, but excluded subjects from autoimmune disease cohorts (type 1 diabetes and rheumatoid arthritis) that are likely to have disease SNPs that overlap with Crohn's disease (40).
Statistical Analysis.
Statistical analysis was performed with Microsoft Excel or Graphpad Prism 4 software. For mouse data, we used ANOVA with Newman-Keuls posthoc test for parametric analyses between three or more groups and the Kruskal-Wallis test with Dunn's posthoc test for nonparametric analyses. For cell line data, we used unpaired Student t-test for comparison between groups. For Crohn's genetic data, we analyzed the data under a genotypic model (two-degree of freedom χ2 test), an allelic model (one-degree of freedom χ2 test), a homozygote vs. homozygote model (1- degree of freedom χ2 test), and an additive model (one-degree of freedom χ2 test for linear trend).
Acknowledgments.
We thank Mark Daly for advice and suggestions; the Wellcome Trust and the authors of the Wellcome Trust Case Control Consortium study (40) for making their data publicly available; Stranger et al. (13) for making their gene expression data publicly available; and the National Institute of General Medical Sciences and the Coriell Cell Repository for their service in providing human cell lines for use by the research community. This work was supported by National Institutes of Health Grants K08 DK076868, HL57307, HL63972, and HL076540 (to D.J.F. and S.C.R.) and German Research Foundation Grants DFG KU 1957/1-1 and DFG KU 1957/3-1 (to B.M.K.).
Footnotes
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
This article contains supporting information online at www.pnas.org/cgi/content/full/0902869106/DCSupplemental.