Genome-Wide Association Analysis of Adaptation Using Environmentally Predicted Traits

Current methods for studying the genetic basis of adaptation evaluate genetic associations with ecologically relevant traits or single environmental variables, under the implicit assumption that natural selection imposes correlations between phenotypes, environments and genotypes. In practice, observed trait and environmental data are manifestations of unknown selective forces and are only indirectly associated with adaptive genetic variation. In theory, improved estimation of these forces could enable more powerful detection of loci under selection. Here we present an approach in which we approximate adaptive variation by modeling phenotypes as a function of the environment and using the predicted trait in multivariate and univariate genome-wide association analysis (GWAS). Based on computer simulations and published flowering time data from the model plant Arabidopsis thaliana, we find that environmentally predicted traits lead to higher recovery of functional loci in multivariate GWAS and are more strongly correlated to allele frequencies at adaptive loci than individual environmental variables. Our results provide an example of the use of environmental data to obtain independent and meaningful information on adaptive genetic variation.Author Summary: Finding genes involved in adaptation to the environment has long been of interest to evolutionary biologists and ecologists. Most commonly, researchers look for loci whose differences in allelic state correlate with differences in a particular trait or environmental variable such as temperature. The implicit assumption behind such methods is that natural selection by the environment will shape variation in adaptive traits through associated changes in allele frequencies. This means that both environmental and phenotypic variation are relevant for detecting adaptive genes, although we have incomplete knowledge of how the two types of variation relate to adaptation. Here we present a method that aims to identify adaptive genes by combining phenotypic and environmental data. We first predict trait variation from a set of environmental variables as a way to extract the most biologically relevant information from the environment and then look for genes associated with both the predicted and observed trait. Using simulations and published data from the model plant Arabidopsis thaliana, we show that this approach may find adaptive genes more effectively compared to existing methods. We also demonstrate that predicted traits can be used to identify relevant loci in individuals for which no phenotypic data is available.