Simultaneously testing for marginal genetic association and gene-environment interaction

Abstract

In this article, the authors propose to simultaneously test for marginal genetic association and gene-environment interaction to discover single nucleotide polymorphisms that may be involved in gene-environment or gene-treatment interaction. The asymptotic independence of the marginal association estimator and various interaction estimators leads to a simple and flexible way of combining the 2 tests, allowing for exploitation of gene-environment independence in estimating gene-environment interaction. The proposed test differs from the 2-df test proposed by Kraft et al. (Hum Hered. 2007;63(2):111-119) in two respects. First, for the genetic association component, it tests for marginal association, which is often the primary objective in inference, rather than the main effect in a model with gene-environment interaction. Second, the gene-environment testing component can easily exploit putative gene-environment independence using either the case-only estimator or the empirical Bayes estimator, depending on whether the goal is gene-treatment interaction in a randomized trial or gene-environment interaction in an observational study. The use of the proposed joint test is illustrated through simulations and a genetic study (1993-2005) from the Women's Health Initiative.

Figure 1.

Comparison of statistical power for detecting single nucleotide polymorphism (SNP)-environment interaction by means of the simultaneous test, the Kraft 2-df test, and the 2-stage procedures. One thousand simulated data sets were generated with 10,000 SNPs and 2,000 cases and 2,000 controls. Panels A, B, and C show power for a model with genetic main effect γ₂ = 0 and environmental main effect γ₁ = 0 and with OR(G, E) = 0.8, OR(G, E) = 1, and OR(G, E) = 1.2, respectively (OR, odds ratio). MA + EB, the simultaneous test for marginal association (MA) and the empirical Bayes (EB) interaction; MA → EB, the 2-stage test with MA screening followed by the EB interaction testing; Murcray, the 2-stage test with gene-environment correlation screening (12); Hybrid, the 2-stage test using both MA and gene-environment correlation in screening (14); Kraft, the Kraft 2-df test; MA + CC, the simultaneous test for MA and the standard case-control (CC) interaction; MA + CO, the simultaneous test for MA and the case-only (CO) interaction; MA → CO, the 2-stage test with MA screening followed by CO interaction testing.