Inference of causal and pleiotropic effects with multiple weak genetic instruments: application to effect of adiponectin on type 2 diabetes

Abstract

AbstractCurrent methods for Mendelian randomization studies have several limitations: to construct unlinked genetic instruments they can use only one SNP from each clump of exposure-associated SNPs, they require that weak instruments are excluded, and they rely on makeshift procedures for downweighting outliers to allow inference of causality in the presence of pleiotropic effects. This paper describes methods that overcome these limitations. A scalar instrument is constructed from all exposure-associated SNPs in each clump, and inference of causality is based on marginalizing over the distribution of pleiotropic effects.To demonstrate the approach, we tested the effect of circulating levels of adiponectin, encoded byADIPOQ, on the risk of type 2 diabetes. Genotypic instruments were constructed from 24 unlinkedtrans-pQTLs detected in Icelanders using the Somalogic platform and 43 detected in the UK Biobank study using the Olink platform. These instruments were tested for association with type 2 diabetes in a non-overlapping subset of the UK Biobank cohort. In contrast to the results of earlier Mendelian randomization studies of adiponectin that used only acis-pQTL, with multipletrans-pQTLs there was clear evidence for a causal effect: standardized log odds ratio -0.38 (95% CI -0.5 to -0.25) using DeCODE instruments and -0.33 (95% CI -0.43 to -0.23) using UK Biobank instruments.Guidelines for the design of Mendelian randomization studies that recommend exclusion of weak instruments, or restricting the instruments tocis-acting variants where the exposure under study is a gene transcript or gene product, should be reassessed.