Quantitative assessment of observed versus predicted responses to selection

Abstract

Although artificial-selection experiments seem well suited to testing our ability to predict evolution, the correspondence between predicted and observed responses is often ambiguous due to the lack of uncertainty estimates. We present equations for assessing prediction error in direct and indirect responses to selection that integrate uncertainty in genetic parameters used for prediction and sampling effects during selection. Using these, we analyzed a selection experiment on floral traits replicated in two taxa of the Dalechampia scandens (Euphorbiaceae) species complex for which G-matrices were obtained from a diallel breeding design. After four episodes of bidirectional selection, direct and indirect responses remained within wide prediction intervals, but appeared different from the predictions. Combined analyses with structural-equation models confirmed that responses were asymmetrical and lower than predicted in both species. We show that genetic drift is likely to be a dominant source of uncertainty in typically-dimensioned selection experiments in plants and a major obstacle to predicting short-term evolutionary trajectories.

Keywords: Dalechampia; G-matrix; Lande equation; artificial selection; breeder's equation; correlated traits; evolvability; indirect selection.