A Comparison of Genome-Wide Association Studies in Multi-Environment Trials

Abstract

In genome-wide association studies (GWAS), trait variance is crucial to statistically test phenotypic-marker associations. The aim of this study was to compare the accuracy of different GWAS approaches with and without genotype-by-environment (GE) variance in the statistical model. Single- and multi-environment models, including molecular markers or pedigree information to account for genetic correlations, were compared. Data comprise wheat lines genotyped with 1279 molecular markers and phenotypically assessed in four environments. We compared the following approaches: M1–Single-environment model fitted using pedigree information to explain genetic correlations; M2–Single-environment model with correlations among lines estimated from molecular similarity; M3–Multi-environment model including pedigree information; and M4–Multi-environment model with molecular similarity. For each model, variance components, marker scores, and best linear unbiased predictor (BLUP) of genotypic effects were obtained. Multi-environment models were more efficient to test marker effects.