|Home | About | Journals | Submit | Contact Us | Français|
Wrote the paper: JRB Perry, TM Frayling, S Cauchi. Statistical analysis: JRB Perry, BF Voight, L Yengo, N Amin, J Dupuis, M Ganser, H Grallert, P Navarro, M Li, L Qi, V Steinthorsdottir, RA Scott, P Almgren. Phenotyping/genotyping: DE Arking, Y Aulchenko, B Balkau, R Benediktsson, RN Bergman, E Boerwinkle, L Bonnycastle, NP Burtt, H Campbell, G Charpentier, FS Collins, C Gieger, T Green, S Hadjadj, AT Hattersley, C Herder, A Hofman, AD Johnson, A Kottgen, P Kraft, Y Labrune, C Langenberg, AK Manning, KL Mohlke, AP Morris, B Oostra, J Pankow, A-K Petersen, PP Pramstaller, I Prokopenko, W Rathmann, W Rayner, M Roden, I Rudan, D Rybin, LJ Scott, G Sigurdsson, R Sladek, G Thorleifsson, U Thorsteinsdottir, J Tuomilehto, AG Uitterlinden, S Vivequin, MN Weedon, AF Wright. Study design: FB Hu, T Illig, L Kao, JB Meigs, JF Wilson, K Stefansson, C van Duijn, D Altschuler, AD Morris, M Boehnke, MI McCarthy, P Froguel, CNA Palmer, NJ Wareham, L Groop, TM Frayling, S Cauchi. Involved in statistical analysis: L Yengo. Involved in the interpretation of results and editing of manuscript: all authors.
Common diseases such as type 2 diabetes are phenotypically heterogeneous. Obesity is a major risk factor for type 2 diabetes, but patients vary appreciably in body mass index. We hypothesized that the genetic predisposition to the disease may be different in lean (BMI<25 Kg/m2) compared to obese cases (BMI≥30 Kg/m2). We performed two case-control genome-wide studies using two accepted cut-offs for defining individuals as overweight or obese. We used 2,112 lean type 2 diabetes cases (BMI<25 kg/m2) or 4,123 obese cases (BMI≥30 kg/m2), and 54,412 un-stratified controls. Replication was performed in 2,881 lean cases or 8,702 obese cases, and 18,957 un-stratified controls. To assess the effects of known signals, we tested the individual and combined effects of SNPs representing 36 type 2 diabetes loci. After combining data from discovery and replication datasets, we identified two signals not previously reported in Europeans. A variant (rs8090011) in the LAMA1 gene was associated with type 2 diabetes in lean cases (P=8.4×10−9, OR=1.13 [95% CI 1.09–1.18]), and this association was stronger than that in obese cases (P=0.04, OR=1.03 [95% CI 1.00–1.06]). A variant in HMG20A—previously identified in South Asians but not Europeans—was associated with type 2 diabetes in obese cases (P=1.3×10−8, OR=1.11 [95% CI 1.07–1.15]), although this association was not significantly stronger than that in lean cases (P=0.02, OR=1.09 [95% CI 1.02–1.17]). For 36 known type 2 diabetes loci, 29 had a larger odds ratio in the lean compared to obese (binomial P=0.0002). In the lean analysis, we observed a weighted per-risk allele OR=1.13 [95% CI 1.10–1.17], P=3.2×10−14. This was larger than the same model fitted in the obese analysis where the OR=1.06 [95% CI 1.05–1.08], P=2.2×10−16. This study provides evidence that stratification of type 2 diabetes cases by BMI may help identify additional risk variants and that lean cases may have a stronger genetic predisposition to type 2 diabetes.
Individuals with Type 2 diabetes (T2D) can present with variable clinical characteristics. It is well known that obesity is a major risk factor for type 2 diabetes, yet patients can vary considerably—there are many lean diabetes patients and many overweight people without diabetes. We hypothesized that the genetic predisposition to the disease may be different in lean (BMI<25 Kg/m2) compared to obese cases (BMI≥30 Kg/m2). Specifically, as lean T2D patients had lower risk than obese patients, they must have been more genetically susceptible. Using genetic data from multiple genome-wide association studies, we tested genetic markers across the genome in 2,112 lean type 2 diabetes cases (BMI<25 kg/m2), 4,123 obese cases (BMI≥30 kg/m2), and 54,412 healthy controls. We confirmed our results in an additional 2,881 lean cases, 8,702 obese cases, and 18,957 healthy controls. Using these data we found differences in genetic enrichment between lean and obese cases, supporting our original hypothesis. We also searched for genetic variants that may be risk factors only in lean or obese patients and found two novel gene regions not previously reported in European individuals. These findings may influence future study design for type 2 diabetes and provide further insight into the biology of the disease.
Common diseases such as type 2 diabetes are highly phenotypically heterogeneous. Few studies have performed genome wide association studies in subsets of patients defined by more stringent phenotypic characteristics. It is possible that reducing the heterogeneity of disease cases may increase power to detect associations over and above the loss of power resulting from reduced numbers. To address these questions we hypothesized that the genetic predisposition to Type 2 diabetes may be different in two strata of cases defined by well-accepted cut-offs for body mass index, the strongest known risk factor for type 2 diabetes.
Genome-wide association (GWA) studies have identified ~50 independent loci robustly associated with type 2 diabetes , , , , , , . These studies have highlighted new candidate pathways involved in the disease , , identified overlap with monogenic forms of the disease , and provided genetic links with correlated phenotypes , .
The GWA studies of type 2 diabetes have not so far provided a greatly improved understanding of the clinical heterogeneity of the disease. Type 2 diabetes cases vary appreciably in their clinical characteristics, particularly age of diagnosis and body mass index (BMI). There is also a group of patients who may present with evidence of an autoimmune component to their diabetes, but who are not insulin dependent . In contrast, the identification of the genetic component to monogenic forms of diabetes has often explained the clinical heterogeneity observed .
Previous studies have provided some evidence of genetic heterogeneity between non-obese and obese type 2 diabetic cases , , , . For example, the variant with the strongest effect on type 2 diabetes risk, in TCF7L2, has a stronger effect in non-obese cases (odds ratio=1.53 [0.37–1.71] compared to obese cases (OR=1.21 [1.09–1.35]) . The effect of FTO variation on type 2 diabetes risk depends on how cases and controls are ascertained by BMI status, but this was expected given FTO's known primary effect on BMI. In the most recent GWA studies of type 2 diabetes , risk variants tended to have stronger effects in non-obese compared to obese individuals – of 30 loci examined, 23 showed stronger associations in non-obese compared to obese individuals.
We designed the present study in an attempt to understand better the genetic heterogeneity of type 2 diabetes. Type 2 diabetes GWA studies tend to be enriched with cases with stronger family histories and lower average BMIs compared to community based studies. Nevertheless, there is a wide spectrum of BMI amongst type 2 diabetes cases used in GWA studies, with more cases being obese than lean. In this study we tested the hypothesis that we would identify new genetic variants by limiting the clinical heterogeneity of type 2 diabetes. By stratifying cases by their BMI status and performing separate GWA studies for each strata of BMI we identified two signals of association not previously reported in the largest GWA studies in Europeans , although one signal has been identified in a South Asian study . In addition we confirmed with additional data that the majority of known type 2 diabetes genetic associations have stronger effects in lean type 2 diabetic cases compared to obese cases.
Descriptions of all cases are available in Table 1, and combined with control details in Tables S1 and S2. Our study was designed to limit the clinical heterogeneity of type 2 diabetes by stratification on BMI, whilst also using the largest sample sizes available:
To test the hypothesis that we would identify new variants associated with type 2 diabetes in different BMI strata, we used the following study design. We used two separate strata of type 2 diabetes cases defined by the two arbitrary, but well established, cut-offs for classifying people as overweight or obese. The first stratum consisted of non-overweight cases, here defined as “lean” (BMI<25 kg/m2). The second strata consisted of obese cases (BMI≥30 kg/m2). For each stratum we used all controls, not selected on BMI to increase statistical power and provide a more robust estimate of the population allele frequency. We did not correct for BMI as BMI was not available in all controls. To check whether or not associations were being driven primarily by effects on BMI we assessed novel variants in an existing GWA studies of BMI using 123,865 individuals from the GIANT consortium . Finally, we performed sensitivity analyses, confirming our findings by stratifying controls by BMI as well as cases.
We chose to include the largest set of studies available. These studies differed in the proportion of total cases defined as lean (8.4–30.4%), the proportion of total cases defined as obese (21.2–77.8%, plus one GWA study, DGDG, that only selected non-obese cases). Some studies were specifically designed as case control studies and some as case-cohort studies, and we note that the extent of phenotyping performed to exclude autoimmune processes was different across studies, ranging from not requiring insulin treatment in the first year of diagnosis and GAD autoantibody negative, to general practitioner diagnosis of type 2 diabetes.
Descriptions of the participating studies are available in the most recent DIAGRAM manuscript , with summary statistics also presented in Table 1 and in Tables S1 and S2. The two discovery GWA study meta-analyses comprised 2112 lean type 2 diabetes cases or 4123 obese type 2 diabetes cases, compared against up to 54,412 controls. For a subset of SNPs available on the Metabochip (a custom Illumina iSelect SNP array that included the SNPs identified by GWA studies for several diseases and traits including type 2 diabetes loci) we included data from an additional 263 lean type 2 diabetes cases, 1735 obese type 2 diabetes cases, and 3691 controls from the GoDARTs study .
With the exception of the BMI-stratification of cases, the meta-analyses, individual study quality control, and analytical methods were the same as those recently reported . A genomic control inflation factor was calculated for each study for each analysis, and their test statistics were adjusted accordingly. Inverse-variance fixed effect meta-analyses were performed on imputed SNP datasets, testing for an additive genetic effect. All single point effect estimates are given with their [95% confidence intervals (CI)]. Only autosomal SNPs with imputation quality scores >0.5 and a minor allele frequency >1% were included from each study. A SNP was excluded from the meta-analysed dataset if it was present in less than half of the studies. Given the use of two strata, we used a p-value threshold of 2.5×10−8 as the criterion for genome-wide significance.
An additional 4 studies, totalling 2881 lean cases, 8702 obese cases, and 18957 controls were available for de novo genotyping of SNPs (Table S2). For the DGDG replication, all polymorphisms were genotyped using the KASPar system (KBiosciences). For Malmo CC, ADDITION-Ely, and Norfolk Diabetes Case Control Study (NDCCS), Taqman assay genotyping was performed. For all four studies genotyping success rate was >95%, the genotyping error rate was 0% based on re-genotyping of 384 individuals, and all SNPs were in Hardy-Weinberg equilibrium (P>0.05). We re-performed the inverse-variance weighted meta-analysis for the replication SNPs using data from all the discovery and replication datasets.
To test whether or not type 2 diabetes associations could be primarily driven by effects on BMI, we assessed the association of novel SNPs with BMI using data from the GIANT consortium consisting of 123,865 individuals.
There are two possible reasons why a variant may be associated with type 2 diabetes in a stratified sample compared to using all data. First, the variant may have a genuinely larger effect in that stratum compared to the overall sample. Second, chance will influence which SNPs are most strongly associated in different subsets of data. To distinguish between these two possibilities we performed a case only analysis in which we tested whether variants associated with lean or obese type 2 diabetes were also associated with BMI within type 2 diabetes cases. We analysed BMI as a quantitative trait in cases from the GWA studies and meta-analysed the summary statistics. If a variant is genuinely associated with type 2 diabetes with stronger effects in the lean stratum, for example, we would expect the risk allele to be associated with lower BMI within cases. This phenomenon was previously reported for the variant in TCF7L2 .
SNP association statistics on glyacemic traits in healthy individuals were provided by the Meta-Analyses of Glucose and Insulin-related traits Consortium (MAGIC). Phenotypes available were fasting insulin (N=38,238, fasting glucose (N=46,186), beta-cell function (HOMA-B, N=36,466), insulin resistance (HOMA-IR, N=37,037), HbA1C (N=46,368) and 2 hour glucose (N=15,234) after an oral glucose challenge. All traits are naturally log transformed, besides fasting glucose, 2 hour glucose and HbA1c. The studies and methodology for these GWA study data are described in their recent publications , ,  and available online at www.magicinvestigators.org. We also had access to data from joint meta-analyses of SNP and SNPxBMI interaction on fasting glucose (N=58,074), insulin (N=51,570), and 2-hr glucose (N=15,141), also provided by MAGIC (Manning et al, in press).
Identified SNPs were searched against a collected database of expression SNP (eQTL) results including a range of tissues , , , , , , , , , , , , , , , , .
In addition to identifying new loci, we tested the impact of BMI stratification on SNPs previously identified as associated with type 2 diabetes. We calculated the individual SNP association statistics using the lean and obese meta-analyses described above.
To assess the effects of combining information from all known type 2 diabetes SNPs, we next used a single study, the GoDARTs  study, independent from the discovery GWA studies. In GoDARTs there were a total of 263 lean type 2 diabetes cases, 1735 obese type 2 diabetes cases, and 3691 controls. Known SNPs (N=36 on the metabochip) were defined as those reaching genome-wide significance in studies using samples of European descent (excluding FTO due to primary effect on BMI, and DUSP9 not present on the chip) , , . We also combined the 36 SNPs into a single allele count model. This analysis consisted of a logistic regression model comparing the count of an individual's type 2 diabetes risk alleles, against case-control status. Each risk allele count was weighted by the point estimate effect size of that SNP from the DIAGRAM meta-analysis . We repeated this analysis using stratified controls (BMI<25 kg/m2 versus lean cases and BMI≥30 kg/m2 versus obese cases) instead of all controls. Finally, individuals were binned into quintiles based on their weighted allele score and per-quintile odds ratios calculated.
Three independent association signals reached P<2.5×10−8 in the lean case genome wide meta-analysis (Table 2). Two represented previously reported loci - TCF7L2 (OR=1.58 [1.47–1.68], P=2×10−40) and CDKAL1 (OR=1.26 [1.17–1.35], P=7×10−10). One novel locus reached genome-wide significance, lead SNP positioned ~25 kb from the HLA-DQA2 gene (OR=1.3 [1.19–1.42], P=1×10−8). Three further independent signals reached P<5×10−7, two of which were previously identified (SNPs in or near ADCY5, OR=1.25 [1.15–1.35] P=6×10−8, and SLC30A8, OR=1.23 [1.15–1.33] P=4×10−8) and one of which was novel (SNPs in LAMA1, OR=1.22 [1.12–1.30] P=1×10−7). Rs numbers are given in Table 2.
In the obese case genome wide meta-analysis, five signals reached genome-wide significance (Table 2), all in or near known loci TCF7L2, FTO, CDKAL1, HHEX, and IGF2BP2. A further three signals reached P<5×10−7; SNPs in or near the MC4R gene (previously associated with BMI), and two other signals; in HMG20A (previously reported in South Asians -OR=1.14 [1.09–1.19] P=2×10−7) and in ANKS1A (OR=1.3 [1.18–1.43] P=5×10−7).
We sought to replicate the signals reaching P<5×10−7 not previously reported in Europeans. SNPs representing the LAMA1 (rs8090011), HLA-DQA2 (rs3916765), HMG20A (rs7178572), and ANKS1A (rs16896390) signals were genotyped in up to 2,881 lean cases, 8,702 obese cases and 18,957 control individuals. Combined discovery and follow-up association statistics for these SNPs are shown in Table 2. In the lean case analysis, the LAMA1 variant was associated with type 2 diabetes (combined P=8.4×10−9, OR=1.13 [1.09–1.18], total lean cases N=4,993, controls=70,515) compared to an OR=1.03 [1.00–1.06] in the obese case analysis (Figure 1 and Figure 2). In the obese case analysis, the HMG20A signal was associated with type 2 diabetes (combined P=1.3×10−8, OR=1.11 [1.07–1.15], total obese cases N=8,583, controls=62,063) compared to an OR=1.09 [1.02–1.17], P=0.015, in the lean analysis (Figure 3 and Figure 4). In previously published studies including 8,130 cases not stratified by BMI , the LAMA1 and HMG20A variants reached only nominal levels of significance of P=0.002 (OR=1.07 [1.03–1.12]) and P=0.003, OR=1.07 [1.02–1.12] respectively (both in the same directions as reported here).
Considering a random-effects model  for both LAMA1 and HMG20A signals gave similar evidence for association (LAMA1 lean analysis: P=5×10−10, obese analysis: P=0.02; HMG20A lean analysis: P=0.04, obese analysis: P=2.7×10−8). Evidence for association at the HLA-DQA2 and ANKS1A signals was reduced when follow-up data were included.
We next attempted to understand further the associations between SNPs in the LAMA1 and HMG20A loci and lean and obese type 2 diabetes cases respectively. Our study design, together with the associations between the FTO and MC4R variants in the obese strata, suggested that variants that primarily operate through BMI could drive our newly identified associations. We therefore assessed the two signals in the existing GWA studies of BMI performed by the GIANT study and consisting of 123,865 individuals . The LAMA1 SNP was not associated with BMI (P=0.19) whilst the type 2 diabetes risk allele at the HMG20A SNP was nominally associated with increased BMI (P=0.02).
If the associations at the LAMA1 and HMG20A loci are genuinely stronger in one strata of diabetic cases compared to the other, we should observe an association of those variants with BMI within cases only. This phenomenon has previously been reported for the variants in TCF7L2. The LAMA1 type 2 diabetes risk allele was associated with lower BMI within cases alone (P=2×10−6 when analysing BMI as a quantitative trait in 26,366 cases), a result consistent with its association being stronger in the lean case analysis. The HMG20A risk allele showed no evidence of association (P>0.05).
Next we used data from MAGIC to assess potential roles of variants in normal glycaemia. The SNP representing the novel LAMA1 association showed no association with fasting glucose (P=0.48, beta(se)=0.0027(0.004) N=46,186), fasting insulin (P=0.87, beta(se)=0.0006(0.004) N=38,238), HbA1C (P=0.19, beta(se)=0.005(0.004) N=46,368), 2-hour glucose response (P=0.43, beta(se)=−0.016(0.02), N=15,234), or any of the SNP×BMI-interaction models. However, LAMA1 isn't unique amongst type 2 diabetes loci in showing no effect on glycemic traits in the MAGIC study.
The HMG20A diabetes risk allele was associated with higher fasting glucose (P=0.04, beta(se)=0.008(0.004), N=46,186), higher HbA1C (P=0.002, beta(se)=0.01(0.004), N=46,368) and higher fasting glucose after accounting for BMI and SNPxBMI interaction (P=0.008, N=58,074).
In an attempt to gain further insight into likely functional genes in the LAMA1 and HMG20A loci, we tested the lead SNPs at for association in a number of eQTL datasets. Tissues tested included various blood, brain, liver and fat samples (see Methods). Only ‘cis’ associations were considered (eQTL effects on a transcript within 1 Mb of the signal SNPs). The rs7178572 SNP in the HMG20A region was significantly associated with mRNA expression levels of HMG20A in the liver (P=4×10−5), supported by two separate expression probes, and was the strongest known regional SNP for both the liver eQTL and type 2 diabetes. No other study-wide significant results were observed (N=14 tissues, 24 datasets/analyses).
For each of 36 published type 2 diabetes loci (identified in European studies and available on the metabochip) we compared the effect sizes between the lean and obese GWA study meta-analyses (Table 3). Among the 36 independent variants, 29 had a larger point estimate odds ratio in the lean analysis compared to the obese analysis (binomial test of 29/36 versus 50% under the null hypothesis of no difference, P=0.0002). We next assessed the combined effect of these SNPs in a case control study independent of the GWA studies - GoDARTs (Figure 5). In the lean stratum, we observed a weighted per-risk allele OR=1.13 [1.10–1.17], P=3.2×10−14. This was larger than the same model fitted in the obese strata where the OR=1.06 [1.05–1.08], P=2.2×10−16. Results were very similar when stratifying the controls as well as the cases by BMI: lean weighted per risk-allele OR=1.13 [1.09–1.17]; obese weighted per risk-allele OR=1.08 [1.05–1.10] (heterogeneity of odds ratios P=0.036). We also observed a difference between lean and obese cases when removing controls and fitting a regression model of lean cases vs obese cases (P=0.0001). None of these 36 variants were associated with BMI in 28,000–32,000 individuals from GIANT , .
We next divided the case/control samples into risk quintiles, based on the number of risk alleles they carry, weighted by the relative effect sizes of those alleles from the larger DIAGRAM meta-analysis. The risk of being in each quintile relative to the median quintile is shown in Figure 6. For the lean group, we observed an OR=2.1 [1.47–3.01] for the quintile of individuals carrying the most risk alleles compared to the middle quintile. This effect was larger than that in the obese group where the equivalent OR=1.37 [1.15–1.64].
We have confirmed our hypothesis that it is possible to identify genetic associations in previously tested samples by constraining the phenotypic heterogeneity of disease cases. By stratifying type 2 diabetes into two well accepted definitions of lean and obese cases, we identified and replicated one locus in each BMI stratum, each previously unreported in European studies: a signal in the LAMA1 gene in the lean stratum and a signal in the HMG20A gene in the obese stratum. Lack of evidence for association with BMI for these two signals in 123,000 individuals  argues that these associations are not driven by a primary association with BMI.
There are two reasons why previously undetected genetic associations may be observed in stratified data. First chance, in this context “sampling error”, may occur – new signals may reach statistical thresholds in subsets of data due to a combination of real association and chance. Second, the signal may represent genuine heterogeneity. The enrichment of the LAMA1 signal in lean type 2 diabetes cases compared to obese cases is likely to be a real effect but the enrichment of the HMG20A signal in obese cases is more likely to be due to chance. Whilst we observed some regression to the mean (or “winner's curse”) for the LAMA1 signal, the effects remained different in lean compared to obese cases in the replication samples alone (Figure 1). In addition, the LAMA1 type 2 diabetes risk allele was associated with lower BMI within cases alone (P=2×10−6 when testing BMI as a quantitative trait in cases) – a similar result was previously reported for the TCF7L2 risk allele . In contrast there is no evidence that the HMG20A signal is stronger in obese replication strata compared to lean replication strata (Figure 3) and there was no association with increased BMI within cases alone (P>0.05 when testing BMI as a quantitative trait in cases).
The LAMA1 signal falls in a recombination block within the LAMA1 gene (Figure 2), with the lead SNP positioned within intron 61. Searching for correlated SNPs (r2>0.5) using 1000 Genomes Project data identified only additional intronic SNPs. Previous cell biology studies support a role for LAMA1, encoding laminin-1, in diabetes etiology - inhibition of LAMA1 expression reduced glucose-stimulated secretion in INS1E cells . Several studies observed the beneficial effects of laminin-1, and extracellular matrix (highly enriched with laminin-1) preparations on pancreatic islet development and function , , , , , . Laminin-1 is expressed in intra-islet capillaries  and a role for laminin receptor 1 was proposed in angiogenesis .
The confidence in the HMG20A association is enhanced by several lines of evidence from other studies. The HMG20A signal was previously identified in a GWA study of South Asian individuals  and was nominally associated with fasting glucose (P=0.04, N=46,186) and HbA1C (P=0.002, N=46,368) in non-diabetic individuals analysed by the MAGIC consortium. The association with fasting glucose became stronger when adjusting for BMI in an interaction model (P=0.008).
We initially discovered a genome-wide significant signal near the HLA-DQA2 locus, which subsequently failed to replicate (rs3916765, P=1×10−6). This variant is not in the same gene or in linkage disequilibrium with previously reported associations between HLA loci and type 2 diabetes , . Concerned with the prospect of this association being due to auto-immune diabetes case admixture, we assessed the association of the strongest known type 1 diabetes signals in our lean meta-analysis. None of these showed any significant evidence of association – including the lead signals from the WTCCC type 1 diabetes study in the HLA region (rs3129941, P=0.08), or near the INS (rs3842748, P=0.64) or PTPN22 (rs2476601, P=0.38) genes.
This study has provided the most robust evidence to date that lean type 2 diabetic cases are likely to carry a disproportionately high load of known type 2 diabetes risk alleles. More than 80% (29/36) of type 2 diabetes variants established in Europeans had stronger effects in lean compared to obese cases and the odds ratio for the 20% of lean cases carrying the most risk alleles was more than twice that of the 20% of obese cases carrying the most risk alleles. The corollary of these findings is that obese cases on average carry a disproportionately low load of confirmed type 2 diabetes risk variants, but their diabetes risk will likely be more heavily influenced by their genetic and environmental predisposition to gaining weight in adulthood.
Despite this enrichment of stronger effects in lean versus obese cases, analyses focused only on lean cases is not a more powerful study design compared to using all cases. For each of the known loci tested, the power gained by increased effect sizes is easily offset by the reduced power of having a case sample size of ~25%. Nevertheless our data indicate that, given limited resources, recruitment strategies that target leaner type 2 diabetes cases will have more power than those that target a similar number of cases but without enrichment for lower BMI.
There are several limitations to our study. First, the use of an unstratified control group made testing the significance of differences between lean and obese cases difficult in the context of a genome wide meta-analysis. However, several lines of evidence support our conclusions that lean individuals are enriched for known type 2 diabetes genetic effects. This evidence includes: the very large differences between the upper and lower 95% confidence intervals of the weighted per allele effects in lean and obese, the consistency of the weighted per allele results when stratifying controls as well as cases, and the 80/20 proportion of SNPs showing stronger effects in lean compared to obese individuals respectively. Second, after stratifying by BMI, we did not use other criteria to reduce the clinical heterogeneity of type 2 diabetes. Of note, cases within the BMI strata differed appreciably in their age at diagnosis and the degree to which autoimmune or monogenic diabetes had been excluded. Instead, having stratified by BMI, we opted to use the largest available sample sizes. It is possible that a small number of monogenic or autoimmune forms of diabetes amongst our cases could have reduced our power to detect novel variants. Further studies may help refine how known and novel diabetes signals operate in more clinically homogenous settings. Finally, known type 2 diabetes signals are likely to account for only a small fraction of all risk variants that exist in the genome and any inferences we make are limited to the known signals.
In conclusion, we report associations with the LAMA1 and HMG20A (not previously associated at genome-wide significance in Europeans) gene regions with type 2 diabetes risk. We have demonstrated that lean diabetic cases are enriched for known type 2 diabetes risk alleles compared to obese cases. This enrichment is consistent with the observation that many of the variants with the strongest effects on diabetes are associated with reduced beta cell function . At the opposite end of the spectrum, obese cases presumably need fewer diabetes risk variants to push them towards diabetes, as they are already under strain from the physiological impact of obesity and insulin resistance. These data suggest a disease model where type 2 diabetes cases lie across a continuous distribution with regards to genetic/environmental risk, and beta-cell dysfunction versus insulin resistance aetiologies.
Summary characteristics of discovery GWA studies cohorts. Eurospan represents a single cohort in the main text, however is split into its component studies in this table. n/a=not applicable.
Summary characteristics of replication cohorts.
Full study acknowledgements.
Full study acknowledgments are available in Text S1.
The authors have declared that no competing interests exist.
JRB Perry is supported by the Wellcome Trust as a Sir Henry Wellcome Postdoctoral Research Fellow (092447/Z/10/Z). This work was partially funded by grants from the Wellcome Trust 083270/Z/07/Z and MRC G0601261. The Atherosclerosis Risk in Communities Study is carried out as a collaborative study supported by National Heart, Lung, and Blood Institute contracts (HHSN268201100005C, HHSN268201100006C, HHSN268201100007C, HHSN268201100008C, HHSN268201100009C, HHSN268201100010C, HHSN268201100011C, and HHSN268201100012C), R01HL087641, R01HL59367, and R01HL086694; National Human Genome Research Institute contract U01HG004402; National Institutes of Health contract HHSN268200625226C; and grants DK062370 and DK072193. ARIC: Infrastructure was partly supported by Grant Number UL1RR025005, a component of the National Institutes of Health and NIH Roadmap for Medical Research. Work at Lund University diabetes centre was funded by several grants from the Swedish Research Council (LG prject grant, Linné, Exodiab). Norfolk Diabetes Case-Control and ADDITION-Ely Studies: The work on Ely, ADDITION, and EPIC-Norfolk studies was funded by support from the Wellcome Trust and MRC. The Norfolk Diabetes study is funded by the MRC with support from NHS Research and Development and the Wellcome Trust. This research was conducted in part using data and resources from the Framingham Heart Study of the National Heart, Lung, and Blood Institute of the National Institutes of Health and Boston University School of Medicine. The analyses reflect intellectual input and resource development from the Framingham Heart Study investigators participating in the SNP Health Association Resource (SHARe) project. This work was partially supported by the National Heart, Lung, and Blood Institute's Framingham Heart Study (Contract No. N01-HC-25195) and its contract with Affymetrix for genotyping services (Contract No. N02-HL-6-4278). A portion of this research utilized the Linux Cluster for Genetic Analysis (LinGA-II), funded by the Robert Dawson Evans Endowment of the Department of Medicine at Boston University School of Medicine and Boston Medical Center. Also supported by National Institute for Diabetes and Digestive and Kidney Diseases (NIDDK) R01 DK078616 to JB Meigs and J Dupuis, and NIDDK K24 DK080140 to JB Meigs. The research performed at deCODE Genetics was part funded through the European Community's Seventh Framework Programme (FP7/2007-2013), ENGAGE project, grant agreement HEALTH-F4-2007- 201413. The DGDG study was supported by the French Government (Agence Nationale de la Recherche), the French Region of Nord Pas De Calais (Contrat de Projets État-Région), Programme Hospitalier de Recherche Clinique (French Ministry of Health), and the following charities: Association Française des Diabétiques, Programme National de Recherche sur le Diabète, Association de Langue Française pour l'Etude du Diabète et des Maladies Métaboliques, Association Diabète Risque Vasculaire (Paris, France), and Groupe d'Etude des Maladies Métaboliques et Systémiques. This study was also supported in part by a grant from the European Union (Integrated Project EuroDia LSHM-CT-2006-518153 in the Framework Programme 6 [FP6] of the European Community). The D.E.S.I.R. study was supported by the Caisse Nationale d'Assurance Maladie des Travailleurs Salariés, Lilly, Novartis Pharma and Sanofi-Aventis, Institut National de la Santé et de la Recherche Médicale (INSERM) (Réseaux en Santé Publique, Interactions entre les déterminants de la santé, Cohortes Santé TGIR 2008), Association Diabète Risque Vasculaire, Fédération Française de Cardiologie, Fondation de France, Association de Langue Francaise pour l'Etude du Diabete et des Maladies Metaboliques, Office National Interprofessionnel des Vins, Ardix Medical, Bayer Diagnostics, Becton Dickinson, Cardionics, Merck Santé, Novo Nordisk, Pierre Fabre, Roche and Topcon. The D.E.S.I.R. Study Group: INSERM 1018: B Balkau, P Ducimetière, E Eschwège; INSERM U367: F Alhenc-Gelas; Centre Hospitalier Universitaire D'Angers: Y Gallois, A Girault; Bichat Hospital: F Fumeron, M Marre, R Roussel; CHU de Rennes: F Bonnet; CNRS UMR8199, Lille: P Froguel; Medical Examination Services: Alençon, Angers, Blois, Caen, Chartres, Chateauroux, Cholet, Le Mans, Orléans and Tours; Research Institute for General Medicine: J Cogneau; General practitioners of the region; Cross-Regional Institute for Health: C Born, E Caces, M Cailleau, JG Moreau, F Rakotozafy, J Tichet, S. Vol. We are grateful to all patients for participation in the genetic study. We also thank Marianne Deweirder, Frédéric Allegaert (UMR CNRS 8199, Genomic and Metabolic Disease, Lille, France) for their technical assistance and their precious management of DNA samples. This work was partially funded by grants from the Wellcome Trust 083270/Z/07/Z and MRC G0601261. This work was presented as a poster at the American Diabetes Association's scientific sessions June 2011. The NHS/HPFS T2D GWA study (U01HG004399) is a component of a collaborative project that includes 13 other GWA studies funded as part of the Gene Environment-Association Studies (GENEVA) under the NIH Genes, Environment and Health Initiative (GEI) (U01HG004738, U01HG004422, U01HG004402, U01HG004729, U01HG004726, U01HG004735, U01HG004415, U01HG004436, U01HG004423, U01HG004728, RFAHG006033) with additional support from individual NIH (NIDCR:U01DE018993, U01DE018903; NIAAA: U10AA008401, NIDA: P01CA089392,R01DA013423; NCI: CA63464, CA54281, CA136792, Z01CP010200). EUROSPAN cohorts were supported by the European Union framework program 6 EUROSPAN project (contract no. LSHG-CT-2006-018947). The ERF study was supported by grants from the NWO, Erasmus MC and the Centre for Medical Systems Biology (CMSB). We are grateful to all patients and their relatives, general practitioners and neurologists for their contributions and to P Veraart for her help in genealogy, Jeannette Vergeer for the supervision of the laboratory work and P Snijders for his help in data collection. MICROS: The MICROS study is part of the genomic health care program ‘GenNova’ and was carried out in three villages of the Val Venosta on the populations of Stelvio, Vallelunga and Martello. In South Tyrol, the study was supported by the Ministry of Health and Department of Educational Assistance, University and Research of the Autonomous Province of Bolzano and the South Tyrolean Sparkasse Foundation. The VIS study in the Croatian island of Vis was supported through the grants from the Medical Research Council UK and Ministry of Science, Education and Sport of the Republic of Croatia (number 108-1080315-0302). The research within the KORA study was partially funded by the German Center for Diabetes Research (DZD), the Helmholtz Zentrum München, Neuherberg, Germany, and supported by grants from the German Federal Ministry of Education and Research the Federal Ministry of Health, the Ministry of Innovation, Science, Research, and Technology of the state North Rhine-Westphalia, the German National Genome Research Network (NGFN), and the Munich Center of Health Sciences (MC Health) as part of LMUinnovativ. The research of I Prokopenko is funded in part through the European Community's Seventh Framework Programme (FP7/2007-2013), ENGAGE project, grant agreement HEALTH-F4-2007- 201413. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.