Identifying Pleiotropic Effects: A Two-Stage Approach Using Genome-Wide Association Meta-Analysis Data

Abstract

Pleiotropic effects occur when a single genetic variant independently influences multiple phenotypes. In genetic epidemiological studies, multiple endo-phenotypes or correlated traits are commonly tested separately in a univariate statistical framework to identify associations with genetic determinants. Subsequently, a simple look-up of overlapping univariate results is applied to identify pleiotropic genetic effects. However, this strategy offers limited power to detect pleiotropy. In contrast, combining correlated traits into a composite test provides a powerful approach for detecting pleiotropic genes. Here, we propose a two-stage approach to identify potential pleiotropic effects by utilizing aggregated results from large-scale genome-wide association (GWAS) meta-analyses. In the first stage, we developed two novel approaches (direct linear combining, dLC; and empirical combining, eLC) combining correlated univariate test statistics to screen potential pleiotropic variants on a genome-wide scale, using either individual-level or aggregated data. Our simulations indicated that dLC and eLC outperform other popular multivariate approaches (such as principal component analysis (PCA), multivariate analysis of variance (MANOVA), canonical correlation (CCA), generalized estimation equations (GEE), linear mixed effects models (LME) and OBrien combining approach). In particular, eLC provides a notable increase in power when the genetic variant exhibits both protective and deleterious effects. In the second stage, we developed a unique approach, conditional pleiotropy testing (cPLT), to examine pleiotropic effects using individual-level data for candidate variants identified in Stage 1. Simulation demonstrated reduced type 1 error for cPLT in identifying pleiotropic genetic variants compared to the typical conditional strategy. We validated our two-stage approach by performing a bivariate GWA study on two correlated quantitative traits, high-density lipoprotein (HDL) and triglycerides (TG), in the Genetic Analysis Workshop 16 (GAW16) simulation dataset. In summary, the proposed two-stage approach allows us to leverage aggregated summary statistics from univariate GWAS and improves the power to identify potential pleiotropy while maintaining valid false-positive rates.\n\nAuthor SummaryPleiotropy, occurring when a single genetic variant contributes to multiple phenotypes, remains difficult to identify in genome-wide association studies (GWAS). To leverage data for multiple phenotypes and incorporate univariate GWAS summary results, we propose a novel two-stage approach for discovering potential pleiotropic variants. In the first stage, two novel combining approaches were developed to screen potential pleiotropic variants on a genome-wide scale. Simulations demonstrated the superior statistical power of these approaches over other multivariate methods. In the second stage, our approach was used to identify potential pleiotropy in the candidate marker sets generated from the first stage. The proposed two-stage approach was applied to the GAW16 simulation dataset to discover pleiotropic variants associated with high-density lipoprotein and triglycerides. In summary, we demonstrate that the proposed two-stage approach can be applied as a viable and robust strategy to accommodate phenotypic and genetic heterogeneity for discovering potential pleiotropy on genome-wide scale.