Abstract
Genome-wide association studies (GWAS) have become an important method for mapping the genetic loci underlying complex phenotypic traits in many species. A crucial issue when performing GWAS is to control for the underlying population structure because not doing so can lead to spurious associations. Population structure is a particularly important issue in nonhuman species since it is often difficult to control for population structure during the study design phase, requiring population structure to be corrected statistically after the data have been collected. It has not yet been established if GWAS is a feasible approach in Saccharomyces cerevisiae, an important model organism and agricultural species. We thus performed an empirical study of statistical methods for controlling for population structure in GWAS using a set of 201 phenotypic traits measured in multiple resequenced strains of S. cerevisiae. We complemented our analysis of real data with an extensive set of simulations. Our main result is that a mixed linear model using the local ancestry of the strain as a covariate is effective at controlling for population structure, consistent with the mosaic structure of many S. cerevisiae strains. We further studied the evolutionary forces acting on the GWAS SNPs and found that SNPs associated with variation in phenotypic traits are enriched for low minor allele frequencies, consistent with the action of negative selection on these SNPs. Despite the effectiveness of local ancestry correction, GWAS remains challenging in highly structured populations, such as S. cerevisiae. Nonetheless, we found that, even after correcting for population structure, there is still sufficient statistical power to recover biologically meaningful associations.