UNPREDICTABILITY OF CORRELATED RESPONSE TO SELECTION: LINKAGE AND INITIAL FREQUENCY ALSO MATTER

In a recent paper, Gromko (1995) showed using computer simulations that pleiotropy and sampling interact to generate variation in correlated response to selection. His simulations demonstrated that different combinations of pleiotropic effects could lead to the same genetic correlation value, yet, as long as the population size and correlation value were large enough, result in significantly different variance of correlated responses after 10 generations of selection. We extended those results using Alan Roberston's "reparameterization" of selection processes in finite populations. As for direct selection response, a satisfactory description of the correlated response and its variability can be expressed in terms of Nih, t/N, and NL, where N is the effective population size, i is the selection intensity in standard units, h² is the heritability of the selected and correlated traits, t is the number of generations, and L is the length of the chromosome in map units. For a given number of loci, there exits an Nih threshold under which differences between pleiotropic systems will not be detected. For values of Nih above this threshold, the higher Nih is the smaller t/N needs to be for significant differences in correlated responses between pleiotropic systems to be observed. A large number of loci or tight linkage increases the unpredictability of the correlated response, but if Nih is large enough, it does not absolutely prevent significant differences between pleiotropic systems to occur. On the other hand, a small initial allele frequency of the favorable allele would tend to cancel the differences between pleiotropic systems, even for large values of Nih and t/N. Finally, epistasis decreases the overall variability of the correlated response, but generally preserves the difference between pleiotropic systems. Thus, Gromko's conclusions on the unpredictability of correlated response due to variable pleiotropy seem fairly robust, at least in the long term.