Defining type 2 diabetes polygenic risk scores through colocalization and network-based clustering of metabolic trait genetic associations

Abstract

Background: Type 2 diabetes (T2D) is a heterogeneous and polygenic disease. Previous studies have leveraged the highly polygenic and pleiotropic nature of T2D variants to partition the heterogeneity of T2D, in order to stratify patient risk and gain mechanistic insight. We expanded on these approaches by performing colocalization across GWAS traits while assessing the causality and directionality of genetic associations.

Methods: We applied colocalization between T2D and 20 related metabolic traits, across 243 loci, to obtain inferences of shared casual variants. Network-based unsupervised hierarchical clustering was performed on variant-trait associations. Partitioned polygenic risk scores (PRSs) were generated for each cluster using T2D summary statistics and validated in 21,742 individuals with T2D from 3 cohorts. Inferences of directionality and causality were obtained by applying Mendelian randomization Steiger's Z-test and further validated in a pediatric cohort without diabetes (aged 9-12 years old, n = 3866).

Results: We identified 146 T2D loci that colocalized with at least one metabolic trait locus. T2D variants within these loci were grouped into 5 clusters. The clusters corresponded to the following pathways: obesity, lipodystrophic insulin resistance, liver and lipid metabolism, hepatic glucose metabolism, and beta-cell dysfunction. We observed heterogeneity in associations between PRSs and metabolic measures across clusters. For instance, the lipodystrophic insulin resistance (Beta - 0.08 SD, 95% CI [- 0.10-0.07], p = 6.50 × 10^-32) and beta-cell dysfunction (Beta - 0.10 SD, 95% CI [- 0.12, - 0.08], p = 1.46 × 10^-47) PRSs were associated to lower BMI. Mendelian randomization Steiger analysis indicated that increased T2D risk in these pathways was causally associated to lower BMI. However, the obesity PRS was conversely associated with increased BMI (Beta 0.08 SD, 95% CI 0.06-0.10, p = 8.0 × 10^-33). Analyses within a pediatric cohort supported this finding. Additionally, the lipodystrophic insulin resistance PRS was associated with a higher odds of chronic kidney disease (OR 1.29, 95% CI 1.02-1.62, p = 0.03).

Conclusions: We successfully partitioned T2D genetic variants into phenotypic pathways using a colocalization first approach. Partitioned PRSs were associated to unique metabolic and clinical outcomes indicating successful partitioning of disease heterogeneity. Our work expands on previous approaches by providing stronger inferences of shared causal variants, causality, and directionality of GWAS variant-trait associations.

Keywords: Clustering; Colocalization; Personalized medicine; Polygenic risk score; Type 2 diabetes.

Fig. 1

Genome-wide multi-trait colocalization analysis of T2D and 20 related traits. A Summary of the number of regions across the genome in which T2D colocalizes with at least one related trait. Traits are labelled by overarching phenotype family, e.g., lipids. B A stacked locus plot with an example of the colocalized genetic variant (4:157683685) across T2D, high-density lipoprotein (HDL) cholesterol, triglycerides, waist-to-hip ratio (WHR), and fasting insulin at the PDGFC locus. C Network analysis of T2D genetic variants. Variants were clustered according to their pleiotropic associations with related traits plotted into the network, with nodes representing SNPs and the edges the correlations between SNPs based on trait Z-scores. SNPs that shared similar associations with metabolic traits clustered together. Five clusters were identified relating to insulin resistance, beta-cell deficiency, obesity, hepatic glucose metabolism, and liver and lipid metabolism. List of abbreviations: body mass index (BMI), alanine transaminase (ALT), high-density lipoprotein (HDL), waist-to-hip ratio (WHR), visceral adipose tissue, gamma-glutamyl Transferase (GGT), arm fat ratio, hemoglobin A_1C, leg fat ratio, trunk fat ratio, 2-h glucose tolerance test (2hGlu), low-density lipoprotein (LDL), Homeostatic Model Assessment for Insulin Resistance (HOMA-IR)

Fig. 2

Forest plots of the associations of pathway PRSs with metabolic measures in individuals with type 2 diabetes from three cohorts. A HbA_1C (n = 18,517). B BMI (n = 21,281). C HOMA-IR (Homeostatic Model Assessment for Insulin Resistance) (n = 2241). D HDL cholesterol (n = 19,370). E Triglycerides (n = 20,797). F Alanine transaminase (n = 19,134). G gamma-glutamyl transferase (n = 16,900). H Chronic kidney disease (n = 19,171). I Cardiovascular disease (n = 20,504). Linear regression was conducted for continuous outcomes (A,C,D,E,F,G) and logistic regression for binary (H,I) controlling for age, sex, BMI, and cohort, besides BMI (B) which was controlled for sex and cohort. See Additional file 1: Table S8 for the numbers underlying this figure

Fig. 3

We assessed the causal association between BMI and T2D by comparing the effect sizes of genetic variants mapping to insulin resistance, beta-cell deficiency, and obesity pathway. Regression lines represent causal estimates from Mendelian randomization (MR) methods inverse variance weighted (IVW) regression and Egger regression. Lines represent one standard error. A The effect of genetic variants within the obesity pathway with BMI as exposure and T2D as outcome. B The effect of variants within the IR (insulin resistance) pathway with T2D as exposure and BMI as outcome. C The effect of variants within the beta-cell deficiency pathway with T2D as exposure and BMI as outcome. D Bar plots depicting r-squared of genetic variants on both BMI and T2D with the resulting Steiger Z-test p-values