A linkage disequilibrium-based statistical test for Genome-Wide Epistatic Selection Scans in structured populations

The quest for signatures of selection using single nucleotide polymorphism (SNP) data has proven efficient to uncover genes involved in conserved and/or adaptive molecular functions, but none of the statistical methods were designed to identify interacting alleles as targets of selective processes. Here, we propose a statistical test aimed at detecting epistatic selection, based on a linkage disequilibrium (LD) measure accounting for population structure and heterogeneous relatedness between individuals. SNP-based (Trv\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{r_v}$$\end{document}) and window-based (TcorPC1v\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{corPC1_v}$$\end{document}) statistics fit a Student distribution, allowing to test the significance of correlation coefficients. As a proof of concept, we use SNP data from the Medicago truncatula symbiotic legume plant and uncover a previously unknown gene coadaptation between the MtSUNN (Super Numeric Nodule) receptor and the MtCLE02 (CLAVATA3-Like) signaling peptide. We also provide experimental evidence supporting a MtSUNN-dependent negative role of MtCLE02 in symbiotic root nodulation. Using human HGDP-CEPH SNP data, our new statistical test uncovers strong LD between SLC24A5 (skin pigmentation) and EDAR (hairs, teeth, sweat glands development) world-wide, which persists after correction for population structure and relatedness in Central South Asian populations. This result suggests that epistatic selection or coselection could have contributed to the phenotypic make-up in some human populations. Applying this approach to genome-wide SNP data will facilitate the identification of coadapted gene networks in model or non-model organisms.