Identifying gene-environment interactions across genome-wide, twin, and polygenic risk score approaches

Until recently, many researchers have been hesitant to conduct genome-wide gene-environment interaction (GxE) research due to perceptions of low rates of statistical power and skepticism from controversial findings from the existing literature. Nevertheless, twin and polygenic risk score (PRS) studies suggest that GxE is pervasive and may have a large impact on complex genetic traits. Our goal in this paper is to demonstrate that consistent findings emerge from twin, PRS, and genome-wide approaches to identify GxE, subject to the known limitations for each method.

We conducted a series of simulation studies, generating dataset that can be used in twin, PRS and GWAS analyses.

We highlight a high degree of consistency across approaches, with each method detecting GxE. Specifically, genome-wide approaches identify individual variants that interact with an environmental moderator, but struggle with low statistical power when a trait is highly polygenic. Alternatively, aggregating genome-wide effects from a discovery sample into a PRS in the target sample increases the ability to detect broad genetic effects. However, if the statistical power in the discovery sample is low, the associations with the PRS tend to underestimate the genetic signal. This is true for both genetic main and interaction effects. Finally, twin studies are generally robust to differences in polygenicity as well as the underlying distributions of the genetic main and interaction effects. The ability of all three methods to robustly identify genomic moderation emphasizes the fact that multiple valid ways to detect GxE exist that stem from the same basic assumptions about the genetic architecture of complex traits.