Multi-locus approaches for the measurement of selection on correlated genetic loci

The study of ecological speciation is inherently linked to the study of selection. Methods for estimating phenotypic selection within a generation based on associations between trait values and fitness (e.g., survival) of individuals are established. These methods attempt to disentangle selection acting directly on a trait from indirect selection caused by correlations with other traits via multivariate statistical approaches (i.e., inference of selection gradients). The estimation of selection on genotypic or genomic variation could also benefit from disentangling direct and indirect selection on genetic loci. However, achieving this goal is difficult with genomic data because the number of potentially correlated genetic loci (p) is very large relative to the number of individuals sampled (n). In other words, the number of model parameters exceeds the number of observations (p ≫ n). We present simulations examining the utility of whole genome regression approaches (i.e., Bayesian sparse linear mixed models) for quantifying direct selection in cases where p ≫ n. Such models have been used for genome-wide association mapping and are common in artificial breeding. Our results show they hold promise for studies of natural selection in the wild, and thus of ecological speciation. But we also demonstrate important limitations to the approach and discuss study designs required for more robust inferences.