The genetic architecture of type 2 diabetes

Abstract

The genetic architecture of common traits, including the number, frequency, and effect sizes of inherited variants that contribute to individual risk, has been long debated. Genome-wide association studies have identified scores of common variants associated with type 2 diabetes, but in aggregate, these explain only a fraction of the heritability of this disease. Here, to test the hypothesis that lower-frequency variants explain much of the remainder, the GoT2D and T2D-GENES consortia performed whole-genome sequencing in 2,657 European individuals with and without diabetes, and exome sequencing in 12,940 individuals from five ancestry groups. To increase statistical power, we expanded the sample size via genotyping and imputation in a further 111,548 subjects. Variants associated with type 2 diabetes after sequencing were overwhelmingly common and most fell within regions previously identified by genome-wide association studies. Comprehensive enumeration of sequence variation is necessary to identify functional alleles that provide important clues to disease pathophysiology, but large-scale sequencing does not support the idea that lower-frequency variants have a major role in predisposition to type 2 diabetes.

Figures

Extended Data Figure 1. Summary of samples and quality control procedures

This figure summarises data generation for whole genome sequencing (GoT2D), exome sequencing (GoT2D and T2D-GENES) and exome array genotyping (DIAGRAM). In addition, GoT2D whole genome sequence data was imputed into GWAS data from 44,414 subjects of European descent.

Extended Data Figure 4. Power for single and aggregate variant association

a-g. Power to detect single-variant association (α=5×10−8) at varying minor allele frequency (x-axis) and allelic odds-ratio (y-axis) for seven effective sample size (Neff) scenarios relevant to the genomes (a-c) and exomes (dg) component of this project. a. variant observed in 2,657 samples (the effective size of the GoT2D integrated panel); b. variant observed in 28,350 samples (the effective size of the imputed data set); c. variant observed in the GoT2D integrated panel and the imputed data set (effective sample size 31,007);d. ancestry-specific variant in 2,000 samples (the size of each of the non-European exome sequence data sets); e. European specific variant in 5,000 samples (the combined size of the European exome sequence data sets);f. variant observed with shared frequency across all ancestry groups in 12,940 samples (the size of the combined exome sequence data set); and g.variant observed in the combined exome array and sequencing data set (effective sample size 82,758). h-i. Power for gene based test of association (SKAT-O) according to liability variance explained. In h, 50% of the variants contribute to disease risk while the remaining 50% have no effect on disease risk; ini., 100% of the variants contribute to disease risk. For each, sample sizes considered are 2,000 (ancestry-specific effects; green) and 12,940 (ancestry-shared effects; blue). Power is shown for two levels of significance (α=2.5×10−6 and α=0.001). From these simulation studies, it is clear that under the optimistic model, where effects are shared across all ethnicities (blue line) and all variants contribute, power is >60% for 1% variance explained and α=2.5×10−6. However, power declines rapidly if either criterion is relaxed.

Extended Data Figure 6. Single variant analyses

Manhattan plot of single-variant analyses generated from a.exome sequence data in 6,504 cases and 6,436 controls of African American, East Asian, European, Hispanic, and South Asian ancestry; b. exome array genotypes in 28,305 cases and 51,549 controls of European ancestry; and c. combined meta-analysis of exome array and exome sequence samples. Coding variants are categorized according to their relationships to the previously reported lead variant from GWAS region. Loci achieving genome-wide significance only in the combined analysis are highlighted in bold. The HNF1A variant reaching genome-wide significance in the combined analysis is a synonymous variant (Thr515Thr). The dashed horizontal line in each panel designates the threshold for genome-wide significance (p<5×10−8).

Extended Data Figure 7. Classification of coding variants according to their relationship to reported lead variants for each GWAS region

The ideogram shows the location of 25 coding variant associations at 16 loci described in the text. The number in each circle corresponds to the number of associated variants at each locus. Variants are grouped into five categories based on inferred relationship with the GWAS lead variant. For some of these categories, the figure includes representative regional association plots based on exome array meta-analysis data from 28,305 cases and 51,549 controls. The locus displayed for each category is designated in bold. The first plot in each panel shows the unconditional association results; middle plot the association results after conditioning on the non-coding GWAS SNP; and the last plot the results after conditioning on the most significantly associated coding variant. Each point represents a SNP in the exome array meta-analysis, plotted with their p-value (on a –log10 scale) as a function of the genomic position (hg19). In each panel, the lead coding variant is represented by the purple symbol. The color-coding of all other SNPs indicates LD with the lead SNP (estimated by European r2 from 1000 Genomes March 2012 reference panel: red r2≥0.8; gold 0.6≤r2<0.8; green 0.4≤r2<0.6; cyan 0.2≤r2<0.4; blue r2<0.2; grey r2unknown). Gene annotations are taken from the University of California Santa Cruz genome browser. GWS: genome-wide significance. *Seven variants, three at ASCC2, and one each atTHADA, TSPAN8, FES andHNF4A did not achieve genome-wide significance themselves, but are included because they fall into genes and/or regions with other significant association signals (see text).

Extended Data Figure 9. Exclusion of synthetic associations and construction of credible causal variant sets at T2D GWAS loci

Ten T2D GWAS loci were selected for synthetic association testing (p<0.001; Methods). a, The effect size observed at the GWAS index SNV (sequence data) before (navy blue) and after (light blue, grey) conditioning on candidate rare and low-frequency (MAF<5%) variants which could produce synthetic association. b, Example of synthetic association exclusion at the TCF7L2 locus. c, Credible sets for T2D GWAS loci where credible set consisted of <80 variants displaying the proportion of credible set variants present in the HapMap and 1000G catalogs.

Extended Data Figure 10

Genome enrichment analysis in GoT2D whole genome sequence data (n=2,657) a, Functional annotation categories were defined using transcription, chromatin state and transcription factor binding data from GENCODE, ENCODE and other studies. b, T2D association statistics for variants at each T2D locus were jointly modelled with functional annotation using fgwas. In the resulting model we identified enrichment of coding exons (CDS), transcription factor binding sites (TFBS), mature adipose active enhancers and promoters (hASC-t4 EnhA, TssA), pancreatic islet active and weak enhancers (HI EnhA, EnhWk), pre-adipose active and weak enhancers (hASC-t1 EnhA, EnhWk), embryonic stem cell active promoters (H1-hESC TssA) and 5’ UTR. Dots represent enrichment estimates and horizontal lines the 95% confidence intervals. c, At the CCND2 locus, three variants not present in HapMap2 have a combined 90% posterior probability of being causal (rs4238013, rs3217801, rs73040004). One of these variants, rs3217801, is a 2-bp indel that overlaps an islet enhancer element.

Extended Data Figure 11. Low frequency variants in exome array data

Results from meta-analysis of 43,045 low-frequency and common coding variants on the exome array (assayed in 79,854 European subjects). a. Observed allelic ORs as a property of allele MAF. Variants missing in >8 cohorts or polymorphic in only one cohort were excluded. Colored lines represent contours for liability variance explained. Regions shaded grey denote ranges of OR and MAF consistent with 80% power (in this case, at α=5×10−7) to detect single-variant associations in this data set (given the observed range of missing data). Variants with a black collar are those highlighted by a bounding analysis as having a probability>0.8 of having LVE>0.1%; b. Distribution of each variant in the MAF/OR space was computed by assuming T2D prevalence of 8% and a beta and normal distribution for MAF and OR respectively. Probability is obtained by integrating the joint MAF-OR distributions over ranges of LVE; c. Single variant association, liability and bounding results for the known T2D GWAS variants on the exome array (Methods).

Figure 1. Ascertainment of variants and single-variant results

a, Sensitivity of low-coverage genome sequence data to detect SNVs in the deep exome sequence data, relative to other variant catalogs. Points represent results for a specific minor allele count. All results assume OR=1 for all variants, unless stated otherwise. Manhattan plots of single-variant association analyses for: b, sequence data alone (1,326 cases and 1,331 controls) and c, meta-analysis of sequence and imputed data (total of 14,297 cases and 32,774 controls).

Figure 2. Association between T2D and variants in genes for Mendelian forms of diabetes

a, p-values of aggregate association for variants from 6,504 T2D cases and 6,436 controls in three sets of Mendelian diabetes genes, for five variant “masks” (Methods). Dotted line:p=0.05. b, Estimated T2D odds ratio (OR) for carriers of variants in each gene-set and mask. Error bars: one standard error. c, Estimated ORs (bars, left axis) and p-values (dots, right axis) for carriers of variants in the PTV+NSstrict mask for each gene. Error bars: one standard error. Red: OR > 1; blue: OR < 1; dotted line:p=0.05.

Figure 3. Empirical T2D association results compared to results under different simulated disease models

Observed number of rare and low-frequency (MAFT and degree of coupling τ between the causal effects of variants and the selective pressure against them. Simulated data were generated to match GoT2D imputation quality as a function of MAF (Methods).

EXTENDED METHODS REFERENCES