Impact of type 2 diabetes susceptibility variants on quantitative glycemic traits reveals mechanistic heterogeneity

Abstract

Patients with established type 2 diabetes display both β-cell dysfunction and insulin resistance. To define fundamental processes leading to the diabetic state, we examined the relationship between type 2 diabetes risk variants at 37 established susceptibility loci, and indices of proinsulin processing, insulin secretion, and insulin sensitivity. We included data from up to 58,614 nondiabetic subjects with basal measures and 17,327 with dynamic measures. We used additive genetic models with adjustment for sex, age, and BMI, followed by fixed-effects, inverse-variance meta-analyses. Cluster analyses grouped risk loci into five major categories based on their relationship to these continuous glycemic phenotypes. The first cluster (PPARG, KLF14, IRS1, GCKR) was characterized by primary effects on insulin sensitivity. The second cluster (MTNR1B, GCK) featured risk alleles associated with reduced insulin secretion and fasting hyperglycemia. ARAP1 constituted a third cluster characterized by defects in insulin processing. A fourth cluster (TCF7L2, SLC30A8, HHEX/IDE, CDKAL1, CDKN2A/2B) was defined by loci influencing insulin processing and secretion without a detectable change in fasting glucose levels. The final group contained 20 risk loci with no clear-cut associations to continuous glycemic traits. By assembling extensive data on continuous glycemic traits, we have exposed the diverse mechanisms whereby type 2 diabetes risk variants impact disease predisposition.

Figure 1

Cluster analysis of effects of 36 type 2 diabetes loci on principal physiologic traits. Clustering of traits with meta-analysis results from at least 10,000 individuals (principal traits). The existence of five clusters was revealed using two clustering approaches. A: Complete linkage dendrogram of type 2 diabetes SNPs with P values (%) indicating the robustness of each branching event (shown in red). We named the clusters as HG loci linked to reduced BC function after glucose stimulation, IR loci with a primary effect on IR at basal measurements, PI locus linked to decreased fasting PI, BC loci associated with defective BC function, and UC loci with no apparent impact on glycemic measures. Strong support exists for the baseline branching notes (strength P ≥ 0.84), whereas branching of IR from the BC-UC clade shows lesser evidence for support (strength P = 0.64). B: Calinski index computed on the centroid-based clustering of type 2 diabetes SNPs provides further evidence for the existence of five locus groups.

Figure 2

Scatter plots of standardized allelic effect size estimates for selected trait pairs. In each scatter plot, loci were assigned to the groups defined from the cluster analysis of principal traits (groups highlighted by different colors). A: Insulinogenic index vs. FI: this plot highlights the effects of loci linked to IR (PPARG, KLF14, IRS1, and GCKR) with respect to FI and insulinogenic index. B: Insulinogenic index vs. FG: the plot reveals the largest impact of HG loci (MTNR1B and GCK) on FG driven by reduced BC function. Large negative effects on insulinogenic index are also seen for CDKAL1 and HHEX/IDE, but with very modest effects on FG. C: HOMA-B vs. HOMA-IR: the plot shows the separation of the BC, HG, and IR clusters. Cluster group colors are as follows: HG, orange; IR, green; PI, pink; BC, red; UC, blue. Loci named in the box are coded numerically within the respective scatter plot.