Confirmation of multiple risk Loci and genetic impacts by a genome-wide association study of type 2 diabetes in the Japanese population

Abstract

Objective: To identify novel type 2 diabetes gene variants and confirm previously identified ones, a three-staged genome-wide association study was performed in the Japanese population.

Research design and methods: In the stage 1 scan, we genotyped 519 case and 503 control subjects with 482,625 single nucleotide polymorphism (SNP) markers; in the stage 2 panel comprising 1,110 case subjects and 1,014 control subjects, we assessed 1,456 SNPs (P < 0.0025, stage 1); additionally to direct genotyping, 964 healthy control subjects formed the in silico control panel. Along with genome-wide exploration, we aimed to replicate the disease association of 17 SNPs from 16 candidate loci previously identified in Europeans. The associated and/or replicated loci (23 SNPs; P < 7 x 10(-5) for genome-wide exploration and P < 0.05 for replication) were examined in the stage 3 panel comprising 4,000 case subjects and 12,569 population-based samples, from which 4,889 nondiabetic control subjects were preselected. The 12,569 subjects were used for overall risk assessment in the general population.

Results: Four loci-1 novel with suggestive evidence (PEPD on 19q13, P = 1.4 x 10(-5)) and three previously reported-were identified; the association of CDKAL1, CDKN2A/CDKN2B, and KCNQ1 were confirmed (P < 10(-19)). Moreover, significant associations were replicated in five other candidate loci: TCF7L2, IGF2BP2, SLC30A8, HHEX, and KCNJ11. There was substantial overlap of type 2 diabetes susceptibility genes between the two populations, whereas effect size and explained variance tended to be higher in the Japanese population.

Conclusions: The strength of association was more prominent in the Japanese population than in Europeans for more than half of the confirmed type 2 diabetes loci.

FIG. 1.

Flow chart summarizing the multistage design and study aims. (A high-quality digital representation of this figure is available in the online issue.)

FIG. 2.

Plots of type 2 diabetes association and linkage disequilibrium for regions surrounding CDKAL1 (A), regions near CDKN2A/CDKN2B (B), and KCNQ1 (C). A, B, and C each contain five panels. In the top panels, all genotyped SNPs in the current Japanese GWA scan (that passed the quality control) are plotted with their −log₁₀ (P values) for type 2 diabetes (Cochran-Armitage trend test) against chromosome position (in Mb). Blue and red squares indicate P values for the combined genotypes of stages 1 + 2 and stages 1 + 2 + 3, respectively, whereas vertical bars indicate P values for the stage 1 genotype. In the second panels, −log₁₀ (P values) plots from the DIAGRAM study of populations of European descent are similarly displayed (17). The third panels show the genomic location of RefSeq genes with intron and exon structure (NCBI [National Center for Biotechnology Information] Build 35). The fourth and fifth panels show a WGAViewer (50) plot of linkage disequilibrium (r ²) for all HapMap SNPs across the regions for the HapMap populations—Japanese in Tokyo (JPT) and CEPH subjects from Utah (CEU)—respectively.

FIG. 3.

Comparison of the strength of association for seven confirmed type 2 diabetes loci between Japanese and European-descent populations. For the Japanese population, we estimated ORs and their 95% CIs (red solid squares and vertical lines, respectively) for each locus based on our meta-analysis involving four Japanese case-control studies (supplementary Fig. S2). For populations of European descent, on the other hand, the corresponding values (blue solid squares and vertical lines) were derived from the published data (–10). The association of an SNP with type 2 diabetes is measured by the coefficient of determination (R ²), which represents the ability to detect association signals using the Cochran-Armitage trend test.

FIG. 4.

Estimation of the increase in type 2 diabetes risk from the combination of seven susceptibility variants previously identified and robustly replicated in the current study. We used case and control subjects with complete data from all stages of our study (n = 12,105). First, the risk for the genotypes of an SNP was estimated by logistic regression. Then, the multilocus risk for an individual was assessed as the sum of the risks for his/her genotype at seven SNPs. We simulated a population with 10% prevalence by bootstrap sampling. In the simulated population, we arranged the individuals in the order of their multilocus risk, sorted them into 20 equal-sized groups, and calculated the actual prevalence in each group. Means and 95% CIs of the groupwise prevalence were estimated based on 1,000 bootstrap sampling trials and are plotted in the figure. No significant gene-gene interaction was observed between the seven SNPs by multiple logistic regression analysis. T2D, type 2 diabetes.