Fine-mapping type 2 diabetes loci to single-variant resolution using high-density imputation and islet-specific epigenome maps

Abstract

We expanded GWAS discovery for type 2 diabetes (T2D) by combining data from 898,130 European-descent individuals (9% cases), after imputation to high-density reference panels. With these data, we (i) extend the inventory of T2D-risk variants (243 loci, 135 newly implicated in T2D predisposition, comprising 403 distinct association signals); (ii) enrich discovery of lower-frequency risk alleles (80 index variants with minor allele frequency <5%, 14 with estimated allelic odds ratio >2); (iii) substantially improve fine-mapping of causal variants (at 51 signals, one variant accounted for >80% posterior probability of association (PPA)); (iv) extend fine-mapping through integration of tissue-specific epigenomic information (islet regulatory annotations extend the number of variants with PPA >80% to 73); (v) highlight validated therapeutic targets (18 genes with associations attributable to coding variants); and (vi) demonstrate enhanced potential for clinical translation (genome-wide chip heritability explains 18% of T2D risk; individuals in the extremes of a T2D polygenic risk score differ more than ninefold in prevalence).

Figure 1. Manhattan plots of the sex-combined BMI-unadjusted and adjusted meta-analysis for T2D.

a, Manhattan plot (top panel) of genome-wide association results for T2D without BMI adjustment from meta-analysis of up to 71,124 cases and 824,006 controls. The association p-value (on -log₁₀ scale) for each SNP (y-axis) is plotted against the genomic position (NCBI Build 37; x-axis). Association signals that reached genome-wide significance (p<5x10^-8) are shown in purple if novel. b, Manhattan plot (bottom panel) of genome-wide association results for T2D with BMI adjustment from meta-analysis of up to 50,409 cases and 523,897 controls. Novel association signals that reached genome-wide significance (p<5x10^-8) only in the BMI-unadjusted analysis are shown in orange.

Figure 2. Comparison of estimated T2D effect size between BMI-adjusted and unadjusted models.

Z-score for each of the 403 distinct signals from BMI-unadjusted analysis (50,791 cases and 526,121 controls; x-axis) is plotted against the z-score from the BMI-adjusted analysis (50,402 cases and 523,888 controls; y-axis). Variants that display higher T2D effect size in BMI-adjusted analysis are shown in red and variants with higher T2D effects in BMI-unadjusted analysis are shown in blue. Diameter of the circle is proportional to -log₁₀ heterogeneity p-value.

Figure 3. Summary of fine-mapped associations.

a, Distinct association signals. A single signal at 151 loci, and 2-10 signals at 92. b, Number of variants in genetic and functional 99% credible sets. Eighteen and 23 signals were mapped to a single variant in genetic and functional credible sets, respectively. c, Distribution of the posterior probability of association of the variants in credible sets.

Figure 4. Comparison of fine-mapping resolution at 83 distinct signals.

The number of variants included in the 99% credible set for each of the 83 distinct signals constructed using meta-analysis of GWAS data imputed using the 1000G multi-ethnic reference panel (26,676 T2D cases and 132,532 controls) (x-axis; logarithmic scale) is plotted against those (y-axis; logarithmic scale) derived using HRC-based imputation (74,124 T2D cases and 824,006 controls). Inset presents the same plot but with linear scales.

Figure 5. The relationship between effect size and minor allele frequency.

Conditional and joint analysis effect size (y-axis) and minor allele frequency (x-axis) for 403 conditionally independent SNPs. Previously-reported T2D associated variants are shown in green and novel variants are shown in purple. Stars and circles represent the “strongest regional lead at a locus” and “lead variants for secondary signals”, respectively.

Figure 6. Comparison of posterior probability of association for each variant with and without incorporation enrichment information.

Posterior probability of association from genetic credible sets (y-axis) and fGWAS analysis (x-axis) for each variant included in the 99% credible sets.