Predicting and Analyzing the Response to Selection on Correlated Characters

The breeder9s equation generally provides robust predictions for the short-term evolution of single characters. When selection targets two or more characters simultaneously, there are often large discrepancies between predicted and observed responses. We assessed how well this standard model predicts responses to bivariate selection on wing color pattern characteristics in the tropical butterfly Bicyclus anynana . In separate laboratory selection experiments, two sets of serially repeated eyespots were subjected to ten generations of concerted and antagonistic selection for either size or color composition. We compared predicted and actual selection responses over successive generations, using the phenotypic data, selection differentials, and estimates of the genetic variance-covariance matrix G. We found differences in the precision of predictions between directions of selection but did not find any evidence of systematic biases in our predictions depending on the direction of selection. Our investigation revealed significant environmental effects on trait evolution across generations. When these were accounted for, predictions using the standard model improved considerably. In the experiment on eyespot size, secondary splitting of selection lines allowed the estimation of changes in G after nine generations of selection. Changes were not in general agreement with expectations from the breeder9s equation. A contour plot of prediction errors across trait space suggests that directional epistasis in the eyespot genotype-phenotype map might occur but estimates of changes in G are too model-dependent to verify whether they agree with that hypothesis. Altogether, our results underscore the need for quantitative genetics to investigate and estimate potential effects of multivariate non-linear genotype-phenotype maps and of environmental effects on G.