The difficulties of predicting evolutionary response to selection

Abstract

There has been a long debate about the difficulties of predicting evolutionary change despite knowledge of selection and genetic variance. One reason might be the stochastic effects of sexual reproduction creating a variance of offspring genotypes and thus phenotypes. This was tested by means of an explicit genetic individual-based simulation with one trait determined by 50 loci. After 100 generations of weak fluctuating selection an experiment was performed where the optimum was displaced one standard deviation. The response and the predicted response were then compared for 100 populations. The simulation shows two things, first there is a considerable variation in response between populations. This was also found when the same population was replicated many times with the same selection. Second, using the individual-based estimates of genetic variance seriously overestimates the predicted response, while an approach using the variance of mean pair breeding values gave essentially unbiased predictions. However, in all cases there was a lack of precision. Hence, the "missing response" is, at least partly, due to an overestimation of predicted response and an inevitable variance around the predicted value.