Key Biological Pathways Causing Type 2 Diabetes Determined By Cluster Analysis

Presentation Number: OR09-1
Date of Presentation: April 4th, 2017

Miriam S Udler*1, Jaegil Kim2, Marcin von Grotthuss2, Jason Flannick2, Gaddy Getz2 and Jose C Florez1
1Massachusetts General Hospital, Boston, MA, 2Broad Institute, Cambridge, MA


Type 2 diabetes (T2D) is thought to be caused by multiple disparate biological pathways ultimately impacting glucose homeostasis. Over the past decade, close to 100 genetic loci have been associated with T2D through genome-wide association studies (GWAS). As these loci are predominantly in non-coding regions of the genome, the mechanisms by which they increase risk of T2D have not been fully elucidated. Therefore, to define key disease processes leading to T2D, we performed cluster analysis of variant-trait associations for the largest set of T2D genetic variants and diabetes-related traits to-date. In the present analysis, 65 T2D genetic variants were selected as associated with T2D at genome-wide significance from published GWAS. Summary statistic results of each variant’s association with 21 diabetes-related traits were collated from published GWAS (e.g. BMI, fasting insulin, lipid values), and the variant-trait association strengths were represented as a 65-by-21 matrix of z-scores. We applied Bayesian non-negative matrix factorization to this matrix, producing clusters of variants and corresponding traits. This approach allows variants to be members of more than one cluster. We confirmed traits were associated with each cluster by performing a fixed-effect, inverse variance meta-analyses of variant-trait associations for each cluster. Our analysis yielded 4 distinct variant-trait clusters, with 5 variants residing in more than one cluster: (i) an “Obesity” cluster contained 2 variants near FTO and MC4R, and was defined by increased body-mass index (BMI) and waist circumference (P<10-8); (ii) an “Insulin Resistance” cluster contained 17 variants and was defined by increased fasting insulin and triglycerides, and decreased high-density lipoprotein (HDL) cholesterol (all with P<10-8); (iii) a “Beta-cell” cluster contained 23 variants and was defined by increased fasting glucose, pro-insulin, height, 2-hour glucose following glucose tolerance test, and decreased BMI and waist circumference (all with P< 10-6); and (iv) a “Lipid” cluster contained 3 variants and was defined by decreased low-density lipoprotein (LDL), total cholesterol, and triglycerides (P < 10-6). Of note, 2 of the variants in the “Lipid” cluster (near GCKR and TM6SF2) have been associated with non-alcoholic fatty liver disease at genome-wide significance. We then tested the variants in each cluster for aggregate risk for coronary artery disease (CAD) in published GWAS. The first 3 clusters were significantly associated with increased risk of CAD (for each P< 10-5), however, the “Lipid” cluster was not (P=0.8), suggesting heterogeneity in genetic pathways shared between T2D and CAD. In summary, in the most comprehensive effort to-date, we have identified 4 clusters of common T2D genetic variants representing distinct biological pathways causing T2D.


Disclosure: JCF: Ad Hoc Consultant, Merck & Co.. Nothing to Disclose: MSU, JK, MV, JF, GG