Phenome-wide analysis of Taiwan Biobank reveals novel glycemia-related loci and genetic risks for diabetes

Abstract

To explore the complex genetic architecture of common diseases and traits, we conducted comprehensive PheWAS of ten diseases and 34 quantitative traits in the community-based Taiwan Biobank (TWB). We identified 995 significantly associated loci with 135 novel loci specific to Taiwanese population. Further analyses highlighted the genetic pleiotropy of loci related to complex disease and associated quantitative traits. Extensive analysis on glycaemic phenotypes (T2D, fasting glucose and HbA_1c) was performed and identified 115 significant loci with four novel genetic variants (HACL1, RAD21, ASH1L and GAK). Transcriptomics data also strengthen the relevancy of the findings to metabolic disorders, thus contributing to better understanding of pathogenesis. In addition, genetic risk scores are constructed and validated for absolute risks prediction of T2D in Taiwanese population. In conclusion, our data-driven approach without a priori hypothesis is useful for novel gene discovery and validation on top of disease risk prediction for unique non-European population.

Fig. 1. Overview of loci identified in this PheWAS and their pleiotropy.

a Number of identified loci for each trait group by trait categories. Color saturation indicates whether the locus has pleiotropic effects. White: trait-specific locus; medium saturation: shared locus within a single trait category; full saturation: shared locus between trait categories. b Fuji plot of the 41 traits having association signals. Each association lead SNP is presented as a dot and arranged by its physical position along the angle starting from the 12 o’clock position. Each line corresponds to a trait indicated in (a) and each lane is colored by the color of trait category. The larger dots indicate pleiotropic association loci. c The number of associated traits for each inter-categorical pleiotropic locus.

Fig. 2. Genetic correlation between binary and quantitative traits.

Pairwise genetic correlations (n = 946) were estimated by bivariate LD score regression (full correlation results are shown in Supplementary Fig. 2). Only correlations between 10 binary traits and 34 quantitative traits (n = 340) are presented in this figure. Positive and negative correlations are colored in blue and red, respectively. The intensity of correlation is indicated by the color saturation. The FDR is calculated by the Benjamini–Hochberg method. Size of the color block represents the FDR of each correlation, and significant correlations (FDR ≤ 0.05) are indicated by asterisks.

Fig. 3. Overlapping loci within glycemic traits.

The upset plot summarizes shared loci of three glycemic traits. The dot-and-line chart in the bottom combination matrix indicates intersections between three traits. For example, the second column from the left indicates a set of loci only associated with HbA_1c, while the middle column indicates loci associated with both HbA_1c and FG. The upper bar chart shows the number of associated loci in each set. The lower left horizontal bar chart represents the number of loci associated with T2D (7), HbA_1c (26), and FG (29).

Fig. 4. 10-year absolute risk of type 2 diabetes for subjects without and with a family history of type 2 diabetes by age and risk profiles.

a Subjects without a family history. b Subjects with family history.

Fig. 5. Regional association plots, identified by comparison of Taiwan Biobank (TWB), Japan Biobank (BBJ) and UK Biobank (UKBB) genome-wide association studies (GWASs).

The X axis represents the position of loci (hg19). The Y axis represents –log₁₀(P). Red dots are P values of variants from TWB. Green dots are p values from the BBJ. Blue dots are P values from UKBB. The red horizontal line represents P = 5 × 10⁻⁸. The black dashed line represents the location of the gene. a GCKR gene; b HACL1 gene; and c GAD2 gene.