Development of a statistical framework for association mapping in recurrently selected populations
University of Delaware
The genetic architecture underlying response to artificial selection has not been widely characterized so far. Elsewhere, an experimental framework has been proposed that combines association mapping and selection mapping (analysis of allele frequency change) to characterize the genetic basis of crop improvement in recurrently selected populations. However, the approach has not yet been applied. This study concerns the performance of the association mapping component of the framework based on extensive computer simulation. Populations subjected to short-term recurrent selection have an evident basis for coancestry (family-level genetic relatedness), which needs to be controlled for during association analysis to limit false positive associations and minimize bias in the estimation of allele effects. Armed with the knowledge of a population’s founders, a founder-adjusted identity-by-state (FAIBS) estimator was proposed herein for calculating coefficients of coancestry specifically in closed breeding populations under recurrent selection. Weights based on IBS information of the founders were used to arrive at probabilities of identity-bydescent for the derived population. A whole genome simulator was developed in R to mimic multi-generational plant breeding populations along with their genotypic and phenotypic values. Forward simulations were used to generate founders of a population based on demographic and evolutionary parameters specific to maize. Using a specified single nucleotide polymorphism (SNP) allele-frequency distribution and under different selection intensities and trait heritabilities, closed-breeding populations under recurrent selection for a fixed number of generations were generated from these founders. The SNP genotypic information for "metapopulations" formed from samples drawn across generations was used to compare the performance of FA-IBS and two other estimators of coancestry (simple IBS and Loiselle) in terms of accuracy, precision, and correlation with the expected coancestry, along with the power of QTL detection and proportion of type-I errors. Results showed that selection had a marked impact on coancestry and on association mapping inference. Relative to a defined expectation, FA-IBS and IBS produced similar results while Loiselle produced quite different results. Using FA-IBS, a maximum of ~85% power was observed in selected populations for QTL that explained the greatest amount (15-18%) of genetic variation. However, selection also caused a corresponding increase in false positive associations relative to populations under pure drift. In general, association mapping was found to be applicable to closed breeding populations under recurrent selection as long as the change in coancestry due to selection was accounted for. With further improvements in accuracy of coancestry estimates and consequently in model fit, better control over false positive associations could be achieved for selected populations.