A Mixed Model Approach to Genome-Wide Association Studies for Selection Signatures, with Application to Mice Bred for Voluntary Exercise Behavior

Selection experiments and experimental evolution provide unique opportunities to study the genetics of adaptation because the target and intensity of selection are known relatively precisely. In contrast to natural selection, where populations are never strictly "replicated," experimental evolution routinely includes replicate lines so that selection signatures-genomic regions showing excessive differentiation between treatments-can be separated from possible founder effects, genetic drift, and multiple adaptive solutions. We developed a mouse model with four lines within a high running (HR) selection treatment and four nonselected controls (C). At generation 61, we sampled 10 mice of each line and used the Mega Mouse Universal Genotyping Array to obtain single nucleotide polymorphism (SNP) data for 25,318 SNPs for each individual. Using an advanced mixed model procedure developed in this study, we identified 152 markers that were significantly different in frequency between the two selection treatments. They occurred on all chromosomes except 1, 2, 8, 13, and 19, and showed a variety of patterns in terms of fixation (or the lack thereof) in the four HR and four C lines. Importantly, none were fixed for alternative alleles between the two selection treatments. The current state-of-the-art regularized F test applied after pooling DNA samples for each line failed to detect any markers. We conclude that when SNP or sequence data are available from individuals, the mixed model methodology is recommended for selection signature detection. As sequencing at the individual level becomes increasingly feasible, the new methodology may be routinely applied for detection of selection.