|Home | About | Journals | Submit | Contact Us | Français|
Dilated cardiomyopathy (DCM) is a major cause of heart failure with a high familial recurrence risk. So far, the genetics of DCM remains largely unresolved. We conducted the first genome-wide association study (GWAS) to identify loci contributing to sporadic DCM.
One thousand one hundred and seventy-nine DCM patients and 1108 controls contributed to the discovery phase. Pools of DNA stratified on disease status, population, age, and gender were constituted and used for testing association of DCM with 517 382 single nucleotide polymorphisms (SNPs). Three DCM-associated SNPs were confirmed by individual genotyping (P < 5.0 10−7), and two of them, rs10927875 and rs2234962, were replicated in independent samples (1165 DCM patients and 1302 controls), with P-values of 0.002 and 0.009, respectively. rs10927875 maps to a region on chromosome 1p36.13 which encompasses several genes among which HSPB7 has been formerly suggested to be implicated in DCM. The second identified locus involves rs2234962, a non-synonymous SNP (c.T757C, p. C151R) located within the sequence of BAG3 on chromosome 10q26. To assess whether coding mutations of BAG3 might cause monogenic forms of the disease, we sequenced BAG3 exons in 168 independent index cases diagnosed with familial DCM and identified four truncating and two missense mutations. Each mutation was heterozygous, present in all genotyped relatives affected by the disease and absent in a control group of 347 healthy individuals, strongly suggesting that these mutations are causing the disease.
This GWAS identified two loci involved in sporadic DCM, one of them probably implicates BAG3. Our results show that rare mutations in BAG3 contribute to monogenic forms of the disease, while common variant(s) in the same gene are implicated in sporadic DCM.
Idiopathic dilated cardiomyopathy (DCM) is a form of heart failure (HF) defined by the presence of left ventricular dilatation and left ventricular systolic dysfunction in the absence of an obvious aetiology, such as coronary artery disease (CAD), hypertension, valvular disease, or congenital defect.1–3 Dilated cardiomyopathy is a major cause of systolic HF and the main indication for heart transplantation.4 It affects ~1/2500 adults and it is more common in men than in women.5
The pathophysiology of DCM is multifactorial with a possible implication of environmental factors and the existence of a strong genetic component attested by a high rate of familial aggregation;3 20–35% of DCM cases having an affected first-degree relative.2,3 Mutations in more than 30 genes have been identified in monogenic forms of DCM, most of them encoding proteins of the cytoskeleton or the sarcomere.6 These genes may also carry variants that are over-represented in sporadic cases;7 however, results of candidate gene association studies have been inconsistent8 and the genetic basis of DCM, whether familial or sporadic, remains largely unresolved.
Over the last few years, genome-wide association studies (GWASs) exploiting the power of high density genotyping arrays have led to the discovery of numerous loci implicated in cardiovascular diseases;9,10 however, no GWAS of DCM has been reported so far, probably because the relatively low prevalence of the disease makes the assembly of large clinically homogeneous cohorts of patients difficult.
Here we report the results of a GWAS based on pools of DNA from patients with sporadic DCM and controls from several European populations and the replication of the findings in two independent populations. One of the replicated loci suggests the implication of the B-cell lymphoma 2-associated athanogene 3 (BAG3) in DCM. The subsequent sequencing of all coding regions of the BAG3 gene in patients with familial DCM identified several mutations that strongly affect the predicted protein structure and are likely to cause the disease.
All participants included in the study were of European origin. Detailed description of recruitment and ascertainment in the various studies is given in the Supplementary information. Ethics committees approved the study protocols, and all participants gave written informed consent. All patients had a diagnosis of idiopathic DCM according to conventional criteria,11 characterized by enlarged left ventricle diameter and low ejection fraction (≤45%) in the absence of causal factors, such as CAD. Only apparently sporadic cases without affected relatives were included.
For the GWAS, DCM patients from three European studies (CARDIGENE, EUROGENE-EHF, and PHRC-DCM)12,13 were assembled and matched with control groups14,15 from the respective populations (see Table 1 and Supplementary information for details).
The DNA pools were stratified on study population as well as gender and age when the numbers permitted. We required that at least 25 samples were mixed in a single pool (Supplementary material online, Table S1). Each pool was constituted twice and each pool replicate was analysed on two arrays independently (Supplementary information).
Human 610-quad beadchip arrays were hybridized with DNA pools and scanned on an i-scan instrument (Illumina). After normalization in BeadStudio (Illumina), the data were transferred into the R statistical environment16 for further analyses. Allelic signal intensities within pools were computed from the normalized quantitative signals (x and y) corresponding to the two alleles of each single nucleotide polymorphism (SNP): p(x) = x/(x+y) and p(y) = 1−p(x) (Supplementary methods). Four measurements of x and y were available for each SNP in each pool (two replicates of pools and two arrays/replicate of pool) and were used for statistical modelling. To assess the variation among the four measurements of each pool, we computed the intraclass correlation (IC) of p(x) for each SNP, i.e. the ratio of the between pools variance to the total variance. The median IC was 0.69 with an interquartile range from 0.54 to 0.78.
A mixed linear model (R: LME4) was used to analyse the pooled allelic signals of each SNP, in which p(x) was the dependent variable, disease status, age category, gender, and study population were fixed effects, and the pool identifier was a random effect. In addition, a weight was introduced in the model to account for the number of samples in each pool and the fact that each pool was replicated four times. At the discovery stage, a threshold of P < 5 × 10−7 was adopted for declaring a SNP significant. For in-depth investigation of the genomic context of DCM-associated SNPs, exploration and annotations of loci of interest were carried out using WGAviewer17 and regional association results were plotted with LocusZoom.18
The genomic control inflation factor (λ)19 was computed to see whether after linear adjustment on study population, age, and gender, some population stratification or cryptic relatedness could still bias the distribution of SNP × DCM association statistics. For that purpose, the vector of allelic association P-values from the adjusted linear model was transformed into a vector of 1-df Chi-squared statistics and compared with expected values using the function gcontrol2 from the package R:gap.
For each SNP needing validation, we performed individual level genotyping by TaqMan 5′ nuclease assay technology (Supplementary methods). The association between DCM and individual genotypes was tested using a logistic regression model (R:GLM) assuming an additive allele effect and adjusting on age, gender, and study population as in the analysis of the pools–GWAS data.
To search for mutations in BAG3 in familial DCM, genomic DNA from 168 index patients was sequenced after PCR amplification of the four exons on an ABI 3100 capillary sequencing instrument (Applera). DNAs from other available family members were genotyped for the familial variant by PCR and sequencing. We also sequenced exons 2, 3, and 4 of BAG3 in 364 individuals of European descent without known cardiac disease (Supplementary methods). Moreover, DNA from 95 control subjects of North African origin and 45 control subjects from Turkey were sequenced to check, respectively, for the presence of variants P115S and V468M in non-DCM populations.
A short description of the populations of patients and controls included at the discovery and replication stages is reported in Table 1, and more details are provided as Supplementary information; all included subjects were of European origin. We used strict criteria11 for patient selection to exclude secondary or familial forms of cardiomyopathy. The discovery GWAS was performed on pools of DNA (pools-GWAS). Overall 26 DNA pools were constituted stratified on study population, disease status, gender, and age (see Supplementary material online, Table S1). To improve the precision of effect estimates when comparing pools of DNA from patients and controls, each pool was duplicated and each duplicate was hybridized to two different genotyping arrays. After filtering out copy number variation markers and SNPs with poor signal intensity, 517 382 SNPs were retained in the analysis.
Fourteen SNPs were associated with DCM in the pools–GWAS (P < 5 × 10−7). The values of the genomic control inflation factor (λ) computed on the whole set of SNPs and on the 90% least significant SNPs were 1.042 and 0.994, respectively, indicating that stratification was unlikely to inflate the association statistics in an important way; as a consequence, we only present uncorrected statistics. To check the pools–GWAS results, the 14 DCM-associated SNPs identified in the pools–GWAS were genotyped on individual DNA in the discovery samples (Table 2). In the individual-based analysis, two SNPs, rs2234962 on chromosome 10 [odds ratio (95% CI) associated with minor allele = 0.53 (0.44–0.63), P = 1.1 × 10−13] and rs10927875 on chromosome 1 [OR = 0.71 (0.62–0.81), P = 3.6 × 10−7], remained significantly associated with DCM. A third SNP, rs16983785 on chromosome 21 [OR = 1.79 (1.41–2.27), P = 7.4 × 10−7] was just above the pre-specified threshold (Supplementary material online, Figure S1). These three SNPs were tested in two independent replication studies (Table 1). The pools–GWAS associations were replicated in the two replication studies combined for rs10927875 [OR = 0.82 (0.72–0.93), P = 0.0021], and rs2234962 [OR = 0.82 (0.70–0.95), P = 0.0092]; while for rs16983785, replication was not achieved [OR = 1.25 (0.98–1.61), P = 0.076]. In the data set combining the discovery and replication samples, rs10927875 and rs2234962 were associated with DCM with respective P-values of 9.5 × 10−10 and 4.0 × 10−12. The ORs and 95% CIs for each study group and for men and women separately in the discovery and replication phases are reported in Figure 1 and more detailed information is provided in Supplementary material online, Table S2.
rs10927875 is located in an intron of ZBTB17 (zinc finger and BTB domain containing-17 also frequently referred to as MIZ-1) on chromosome 1p36.2–p36.1. The locus of interest is located in a genomic region, covering ~210 kb, that spans several other genes—SPEN (spen homolog, transcriptional regulator), HSPB7 (heat shock 27 kDa protein family, member 7), CLCNKA (chloride channel Ka), and CLCNKB (chloride channel Kb)—and exhibits strong linkage disequilibrium (LD) (Figure 2). Recently, Stark et al.20 reported an association between a SNP in HSPB7 (rs1739843) and DCM, in a study conducted in the group of German patients used for replication in the present study. Because DNA pooling introduces some error that may affect the estimates of relative effects of linked SNPs, we additionally individually genotyped four SNPs in the region (Supplementary material online, Table S3): rs1763601 (a proxy of rs1739843), rs945417 (a proxy of rs1805152), rs945425 (a SNP located in the proximal promoter region of CLCNKA which may generate an alternative splice site), and rs1048261 (a SNP located in the 3′UTR of HSPB7). These SNPs are presented in their genomic context in Supplementary material online, Figure S2. Three of the four additional SNPs exhibited stronger associations with DCM than the lead SNP identified in the pools–GWAS (rs1763601: P = 3.4 × 10−9; rs945417: 1.4 × 10−9; rs945425: 4.2 × 10−7; and rs1048261: 7.9 × 10−9 when compared with P = 1.3 × 10−7 for rs10927875) and, for all these SNPs, the minor allele was associated with a reduced risk of DCM.
To get better insight into the haplotypic structure of the region and its impact on the association of the locus with DCM, haplotypic relative risks were computed.21 As reported in Supplementary material online, Table S4, the five genotyped SNPs determine six common haplotypes. The two major haplotypes CTGCT (0.508 and 0.603 in controls and DCM cases, respectively) and TGCTA (0.301 and 0.230) differ at the five sites (Yin/yang configuration) and are likely to represent ancestral alleles, the four less common haplotypes resulting from recombination between them. In comparison with CTGCT, TGCTA was associated with a reduced risk of DCM [OR = 0.64 (0.53–0.74), P = 3.2 × 10−9].
Among the genes located in the region, HSPB7, which encodes the ‘cardiovascular heat shock protein'22 and exhibits cardiac-specific expression, is a possible candidate for an association with DCM. However, none of the DCM-associated SNPs or tight proxies identified through systematic sequencing by Matkovich et al.23 or genotyped in the present study affects the coding sequence of HSPB7. Moreover, we sequenced the HSPB7 exons in 168 independent index cases diagnosed with familial DCM but were unable to identify any coding variant (data not shown).
Given the lack of coding variant in HSPB7, we envisaged that sequence variations in the HSPB7 region might be related to the DCM risk through an effect on gene expression. To investigate this possibility, we took advantage of genome-wide expression data generated using RNA from circulating monocytes in the Gutenberg Heart Study24 and from circulating monocytes and in vitro-derived macrophages in the Cardiogenics study.25 In both data sets, HSPB7 expression was not detectable in monocytes. Conversely, in the Cardiogenics resource, the gene was weakly expressed in macrophages and several SNPs were associated in cis with its expression, the strongest signal was observed with rs945425 (P = 6.1 × 10−57). The same SNPs were also, but less strongly, associated with the expression of the nearby CLCNKA gene (P = 2.2 × 10−26 with rs945425) (see Supplementary material online, Table S5). These results might indicate a role of HSPB7 or CLCNKA genes through mediation of expression in macrophages. On the other hand, Cappola et al.26 recently showed that the rs1739843 HSPB7 SNP, which is in strong LD with rs945425, had no effect on HSPB7 mRNA expression or splicing in left ventricular myocardium tissue. The authors further suggested that the variant responsible for modulating the risk of heart failure was more likely to be a non-synonymous polymorphism in the nearby CLCNKA gene, which altered the activity of the encoded chloride channel. However, the mechanism linking this variant to heart failure remains speculative since this channel is not expressed in the heart. The association of SNPs with CLCNKA and HSPB7 gene expression found in macrophages but not in the myocardium tissue suggests that these eQTL may be cell specific or may depend on specific activation mechanisms. However, the associations observed in macrophages must be interpreted with caution because the HSPB7 probe present on the Illumina array targets a polymorphic sequence and the two CLCNKA probes may also target polymorphic sequences (see legend of Supplementary material online, Table S5 for more details). The strong LD reflected by the Yin/yang haplotypic structure in the region could thus account for the observed co-association of the signals generated by the polymorphic probes and several SNPs. Functional studies will be needed to assess whether the HSPB7 and CLCNKA eQTLs in circulating macrophages may have any relevance for DCM. The other genes in the region (SPEN, ZBTB17, CLCNKB) were either not detected in monocytes or macrophages or their expression was not or only weakly influenced by the DCM-associated SNPs in macrophages (Supplementary material online, Table S5). Importantly, given the extensive LD in the region, it cannot be excluded that variant(s) located at a distance of the initially identified markers may implicate unforeseen genetic regulatory mechanisms.
rs2234962 is a non-synonymous SNP (c.T757C, p.C151R) located within the coding sequence of BAG3 on chromosome 10q25.2–q26.2 (Figures 3 and 5). Our results show that the minor allele of this variant is associated with a reduced risk of DCM. The cysteine to arginine substitution at Position 151 of the BAG3 protein is predicted to be probably damaging by the Polyphen tool.27 As rs2234962 could tag functional SNPs absent from the genotyping array used in our study, we examined the LD of this SNP with other variants in the region and observed that only two SNPs in HapMap release 22 were in tight LD (r2 > 0.80) with the lead SNP and both were intronic with no evidence that they could be functional.
BAG3 presents several non-synonymous SNPs that are referenced in the HapMap database. To complement our analysis of the locus, we selected two of these SNPs, rs3858340 (c.C1526T, p.P407L) and rs35434411 (c.G518A, p.R71Q), on the basis of a MAF > 5% in the HAPMAP CEU population, and individually genotyped them in the discovery cohorts. The rs35434411 was relatively rare and it was not associated with DCM status (MAF 0.032 and 0.029 in patients and controls, respectively). Conversely, rs3858340 was significantly associated with DCM (P = 3.6 × 10−5, MAF of 0.12 and 0.081 in patients and controls, respectively). Multiple logistic regression analysis indicated that both rs2234962 and rs3858340 were independently associated with DCM (P = 3.1 × 10−12 and P = 3.6 × 10−3, respectively). In the discovery cohort, these two SNPs were in complete LD (D′ = −1) defining three haplotypes, C757-C1526 (R-P), T757-C1526 (C-P), and T757-T1526 (C-L). According to the Forest plots in Figure 4, the presence of an arginine (R) at Position 151 and a proline (P) at Position 407 of BAG3 is associated with a reduced risk of DCM.
Finally, simultaneously testing the associations of HSPB7 and BAG3 individual genotypes with DCM in the discovery samples showed that the effect of both SNPs was additive (rs945417: P < 2.5 × 10−8, rs2234962: P < 5.6 × 10−13), with no evidence of interaction (P = 0.7).
Given the apparent involvement of non-synonymous BAG3 SNPs in sporadic DCM, we investigated whether mutations in BAG3 might contribute to monogenic forms of DCM. The plausibility of such implication was reinforced by a previously reported linkage between markers in the chromosome 10q25–26 region, including the BAG3 locus, with familial DCM.28 The four BAG3 exons and intron–exon boundaries were sequenced in 168 independent index cases from DCM families of mainly European origin. A total of 19 molecular variants were detected (Figure 5 and Supplementary material online, Table S6); among them, six were known SNPs referenced in dbSNP, four were present in a control group of 347 healthy individuals of European descent in which they were tested, and three were unlikely to be deleterious (see legend of Supplementary material online, Table S6). The six remaining variants include four truncating and two missense mutations (Supplementary material online, Figure S3). Two of the BAG3 mutations were observed in relatively large families (six and five proven mutation carriers, respectively) and the four others were observed in small families (one to two mutation carriers in each of these families) (Supplementary material online, Figure S4). The presence of DCM was observed in 13/18 mutation carriers (72%) and possible DCM in 3 (LV dilatation and mild LVEF < 60% in 3, congestive heart failure in 2). Two mutation carriers had a normal cardiac examination (a 7-year-old male and a 33-year-old female). Overall, the predicted severity of the mutations, their pattern of distribution in affected families, and their absence in a sample of 347 healthy individuals suggested that they may be disease causing.
Phenotypically, DCM patients carrying a mutation showed isolated DCM, without associated conduction defect or skeletal myopathy and a normal serum creatine kinase level in all but one patient (Supplementary material online, Table S7). Four patients had heart transplantation and five relatives, whose DNA was not available, died prematurely from cardiac cause with a previous diagnosis of DCM.
Our GWAS and familial results implicating BAG3 in DCM may constitute an example of a gene harbouring common susceptibility variants implicated in sporadic forms of the disease and rare mutations responsible for monogenic forms.
We report the results of the first GWAS conducted in patients with HF due to DCM. For the discovery phase of this study, we used a DNA-pooling approach and identified 14 SNPs at different loci showing some evidence of association with DCM. Three of these associations were confirmed by individual genotyping and two of them were replicated in independent DCM patients and controls.
The first DCM-associated SNP is located in a region exhibiting strong LD that encompasses the HSPB7 and the CLCNKA gene, two potential candidate genes. Recently, Matkovich et al.23 identified several SNPs in the HSPB7 sequence in patients with sporadic cardiomyopathy. According to the HapMap data,29 some of these polymorphisms are in strong LD with the SNPs investigated in the present study, suggesting that the variants reported in both studies correspond to the same Yin/yang haplotypic structure. Yin/yang haplotypes are commonly observed in the human genome.30 Such configuration has already been reported for the CLCNKA and CLCNKB genes31 and it is likely that the rs10927887 exonic CLCNKA SNP recently suggested to be responsible for modulating the risk of heart failure26 also belongs to this ying/yang haplotype, given its strong LD with the rs1739843 intronic HSPB7 SNP. Although functional experiments demonstrated a 50% loss of function of the CLCKNA chloride channel associated with rs10927887,26 linking this mechanism with risk of heart failure remains speculative. On the other hand, the association of several CMD-associated SNPs with HSPB7 gene expression in macrophages suggests that HSPB7 might be the culprit gene. HSPB7 is also called cardiovascular HSP as a consequence of its selective expression in cardiovascular tissues,22 and it belongs to the small HSP (sHSP) family,32 whose members bind denatured proteins. The physiological response of muscle fibres to stress involves sHSPs and HSPs of high molecular mass, such as HSP70. An example is provided by HSPB5 (αB-crystallin); a dominant mutation in this gene causes a severe form of desmin-related cardiomyopathy characterized by impaired autophagy and accumulation of misfolded proteins.33 In humans, genetic variants in HSPB7 have been reported to be associated with advanced heart failure and systolic dysfunction of unspecific origin.23,34
The second DCM-associated SNP affects the sequence of the protein encoded by BAG3. Even if suggestive evidence supporting a role for the C151R and P407L common variants has been presented in this report, extensive work is required to better characterize the exact contribution of all common BAG3 SNPs on the disease. BAG3 is a member of a conserved family of cyto-protective co-chaperone proteins containing a conserved domain able to interact with HSC70/HSP70 and sHSPs proteins. BAG3 is involved in numerous activities including macro-autophagic protein degradation in aging cells.35 BAG3 is mainly expressed in striated muscle and colocalizes with Z-disc (a sarcomeric protein assembly essential for actin anchoring in striated muscles). Following normal muscle development, Bag3-deficient mice present a progressive myopathy with Z-disc disruption and develop a fulminant myopathy characterized by non-inflammatory myofibrillar degeneration with apoptotic features.36 Autophagic degradation appears essential for maintaining Z-disc integrity and muscle contractility and Bag3 plays an essential role in this process which also involves small Hsps, such as HspB8.37
In humans, a mutation in BAG3 (P209L), which affects a conserved I-P-V motif of the BAG3 protein known to play an important role in the interaction of BAG3 with sHSP,38 causes severe dominant childhood muscular dystrophy with cardiomyopathy.39 In the present study, six mutations that affect the BAG domain of BAG3, either by substituting highly conserved amino acids or by deleting the whole domain, were identified. Surprisingly, none of the carriers of the mutations presented with myofibrillar disorders as reported for the P209L mutation, suggesting that different mutations in the BAG3 gene could result in different phenotypes. In Bag3 knock-out mice, fulminant myopathy and cardiomyopathy develop in homozygous −/− but not in heterozygous animals,36 while in Drosophilia, the Bag3 ortholog starvin deletion is associated with locomotion decline and myofibre sarcomeric disorganization even in heterozygous animal.37 Further experiments are clearly needed to better understand the pathophysiology of DCM caused by BAG3 mutations.
Chaperone-assisted degradation or selective autophagy appears to be an essential mechanism for the preservation of cellular structures in striated muscles that is evolutionarily conserved from flies to human and implicates Bag3, Hsc70, and small Hsps.37,40 As far as we know, no mention of interaction between Bag3 and HspB7 has been made in the literature, but the possible implication of these two proteins in human DCM may point to a common important pathophysiological pathway implicating an inappropriate handling of degraded proteins within striated myocytes. Alternatively, in the absence of identification of the functional variant(s) in the region with strong LD around HSPB7, an effect on the DCM risk that is independent of HSPB7 cannot be ruled out, an alternative candidate being CLCNKA.26
Even when a very strict methodology is applied, estimates of allelic effects based on DNA pools are affected by variation in the quantity and concentration of individual DNAs composing the pools and by quantification errors.41 The reduced specificity of pools–GWAS compared with individual-based GWAS may be compensated by individual-based genotyping of the most significant hits in the same samples. As shown in Table 2, associations observed in the pools–GWAS were more frequently reproduced by individual genotyping when several SNPs in LD at the same locus were associated with the disease. The observation made at the ZBTB17/HSPB7/CLCNKA locus is interesting in this regard, because the strongest hits were not the same in the pools–GWAS and in the individual genotyping. As a consequence of errors inherent to the DNA pooling approach, pools–GWAS also exhibit a reduced sensitivity, implying that our study may have missed some associations that would have been identified by an individual-based GWAS. Nevertheless, this approach is efficient41–43 and it allowed us to replicate the only locus identified so far for sporadic DCM and to identify another locus whose implication in this disease was unsuspected.
Among the six mutations detected in family index cases, 5 were observed in C151C homozygous individuals while one was observed in a C151R heterozygous individual. Resequencing of BAG3 in all sporadic DCM cases will be needed to assess whether rare causal mutations may occur preferentially in association with the C151 allele and contribute to the association observed between the C151R SNP and sporadic DCM.
This pools–GWAS identified two loci associated with sporadic DCM. The most likely candidate gene at the chromosome 10 locus is BAG3, whereas HSPB7 and CLCNKA are two potential candidates at the chromosome 1 locus. Beyond DCM, it will be of interest to investigate the implication of the these loci in other forms of heart failure and muscle dysfunction, as well as in the age-related changes that affect the heart and muscle physiology.
This work was supported by grants from Assistance Publique—Hôpitaux de Paris (PHRC programme hospitalier de recherche clinique AOM04141 & AOM95082), the ‘Fondation LEDUCQ' (Eurogene Heart Failure network and CAERUS network), the Société Française de Cardiologie/Fédération Française de Cardiologie (2006), the CONNY-MAEVA charitable foundation. The GWAS was supported by the French ‘Agence Nationale pour la Recherche' (ANR Project ANR-06-BLAN-0264 CSD 8, ‘POOLCA') and the European Union/FP7 (grant 241924; INHERITANCE). UK replication study was supported by UK National Institute for Health Research Biomedical Research Unit and Biomedical Research Centre funding, the British Heart Foundation, the ‘Fondation LEDUCQ’, and the Medical Research Council UK. The UK Blood Services collection of Common Controls (UKBS collection), part of the Wellcome Trust Case-Control Consortium, was funded by the Wellcome Trust (076113/C104/Z), and the Juvenile Diabetes Research Foundation (WT06l858). The German replication study was supported by the EU (Eugeneheart, IP 01883), the University Hospital of Regensburg and the German Research Foundation (DFG).
Conflict of interest: none declared.