Is type 2 diabetes a causal risk factor for coronary artery disease? Multivariate mendelian randomization to test causal relationships among cardiometabolic traits. R. Do1, 2, M. Daly2,3, B. Neale2,3, S. Kathiresan1,2 1) Center for Human Genetic Research, Massachusetts General Hospital, Boston, MA, USA; 2) Broad Institute, Cambridge, MA, USA; 3) Analytic and Translational Genetic Unit, Massachusetts General Hospital, Boston, MA, USA.
Observational epidemiological studies have established correlations among cardiometabolic traits such as type 2 diabetes (T2D), body mass index (BMI) and coronary artery disease (CAD); however, causal inference based on these correlations is challenging. We have developed and implemented a human genetics approach, multivariate mendelian randomization (MMR), that leverages genetic effect sizes of common SNPs to dissect causal influences amongst a set of correlated traits. We first examined the epidemiological association between all pairs of ten cardiometabolic traits, including plasma lipids (low density lipoprotein cholesterol (LDL-C), high density lipoprotein cholesterol (HDL-C), triglycerides), BMI, T2D, systolic blood pressure, fasting insulin, fasting glucose, C-reactive protein and CAD. In 9,610 European Americans from the Atherosclerosis Risk in Communities study, we observed widespread correlations between the ten traits (76 out of 90 comparisons are significantly correlated with P<0.05). Using MMR, we were able to infer causal relationships among the ten traits. Since our approach requires a set of SNPs where effects on the biomarkers of interest are precisely measured, we leveraged estimates of effects from published large-scale genome-wide association studies for each trait. From this analysis, we highlight three sets of observations. First, we observed that, across 32 SNPs associated with BMI at genome-wide significance, the BMI-increasing allele was correlated with effects on five other traits: lower HDL-C (beta=-0.33, P=8.6x10-6); higher triglycerides (beta=0.23, P=5.9x10-9), higher risk for T2D (beta=0.85, P=8.8x10-9), higher fasting insulin (beta=0.17, P=4.4x10-7) and higher C-reactive protein (beta=0.37, P=6.7x10-8). Second, across 19 SNPs associated with fasting insulin at genome-wide significance, the fasting-insulin-raising allele was correlated with effects on two traits: lower HDL-C (beta=-1.28, P=1.1x10-5) and higher triglycerides (beta=0.87, P=2.9x10-5). Finally, in a single model including effect sizes of all nine risk factors and CAD, we observed only four significant predictors for CAD including the strength of a SNPs effect on LDL-C (beta=0.45, P=1.9x10-19), on triglycerides (beta=0.31, P=4.1x10-6), on blood pressure (beta=0.35, P=0.003), and T2D (beta=0.13, P=1.1x10-4). These results suggest that as few as four factors causally relate to CAD and that MMR can be used to infer causal relationships between correlated traits.