Genetic correlations among traits can alter the evolutionary trajectory of a trait across the adaptive landscape because traits are indirectly subjected to the selection pressures aimed at correlated traits (Maynard Smith et al. 1985; Falconer and Mackay 1996). Drawing on the idea that linked genes can interfere with the most rapid fixation of beneficial traits, correlated traits are often thought of as impeding the evolutionary approach to an optimum trait value (Lande 1979; Clark 1987; Mitchell-Olds 1996; Roff and Fairbairn 2006). However, if both traits are advantageous under the same circumstances, genetic correlations can potentially accelerate adaptation. Such synergistic selection is thought to lead to the erasure of the correlation as genetic variation declines (Falconer and Mackay 1996), but such correlations may be observed prior to fixation. Furthermore, it is possible for a correlation itself to have adaptive value; that is, the fitness of a particular value of one trait may depend on the value of another trait (Brodie 1992; Sinervo and Svensson 2002).
The consequences genetic correlations can have for phenotypic evolution have driven interest in the mechanisms leading to genetic correlations. Traits can be correlated for multiple reasons, including pleiotropy, linkage due to chance, or linkage due to selection (Houle 1991; Armbruster and Schwaegerle 1996; Lynch and Walsh 1998; Roff and Fairbairn 2006). Pleiotropy results in correlations between traits because the same gene(s) underlie the correlated traits (Lande 1980; McKay et al. 2003). For example, a single gene can be an upstream member of multiple genetic pathways, with allelic variation in that gene or its regulation leading to consequences in distinct downstream outcomes. Correlations between traits can also occur by chance if genes controlling the traits are physically linked in the genome (Lande 1976), limiting the dissolution of allelic associations by recombination. Lastly, traits can be correlated if there is selection for the maintenance of a suite of phenotypic traits (Lande 1984). Determining the extent to which these different mechanisms contribute to trait correlations is an important foundation for understanding the evolution of complex traits.
Drawing on classic work by Fisher (1958) and Maynard Smith (1978), evolutionary biologists have been fascinated by recombination because of its apparent costliness and potential to change the dynamics of adaptation (reviewed in West et al. 1999; Lehtonen et al. 2012; Meirmans et al. 2012). Recombination influences genetic correlations among traits differently depending on the mechanism that leads to correlations. Both the chance correlations due to physical linkage and adaptive correlations exist in a state of linkage disequilibrium. If recombination is frequent enough in populations with standing genetic variation for the traits in question, both will consequently be limited. Theoretical and empirical work has shown that recombination breaks up physically linked genes helping to eliminate interference via indirect effects of selection (Falconer and Mackay 1996; Colegrave 2002). These two mechanisms differ in that under physical linkage, correlations will be transitory and are not expected to have any consistency across ecologically similar populations. In contrast, correlations driven by selection will exist as a balance between recombination and selection, and should take the same basic structure across ecologically similar populations. Unlike correlations arising from disequilibrium, correlations arising from pleiotropy will exist irrespective of recombination and selection.
Work by Baer and Lynch (2003) provides a framework for differentiating between correlations maintained by different mechanisms. By looking at the correlations that appear within and between populations, they distinguished between pleiotropy and selection. This was done by assaying body size on clones from two populations of Daphnia in a laboratory common garden, and then selecting the largest and smallest clones for a subsequent life history assay. Baer and Lynch (2003) expected a similar pattern of correlation between large and small body size within populations to that found between populations if correlations were maintained because of pleiotropy. However, they concluded that selection was maintaining correlations between traits in their populations because correlation patterns within the two populations they studied differed from the trait associations between the two populations' trait averages.
Here, we apply an analogous approach by examining correlated responses to clonal selection on body size in four different populations of Daphnia pulicaria (Fig. 1). We focus on ecologically similar populations with historically different levels of sex (Cáceres and Tessier 2004). This design allows us to explore how recombination affects the response to direct and indirect selection via genetic correlations and evaluate the potential mechanisms for how these correlations are being maintained. First, we examine the direct response of body size to clonal selection for large and small body size at maturity. Second, we examine correlated responses in life history traits in the selected clones in the context of differences in historical frequencies of sex. In particular, we ask (1) whether historical frequency of sex influences the magnitude of correlated responses to clonal selection; and (2) whether the slope of genetic correlations is uniform across populations.
Predicted correlated responses
The mechanisms which cause correlations to evolve lead to contrasting predictions in our system. First, consider the distinction between pleiotropy and linkage-based mechanisms. If genetic correlations are driven by pleiotropies, we predict that correlated responses will be unaffected by the historical frequency of sex. If this were true, and trait correlations were governed by pleiotropies of the same set of genes drawing on a common pool of allelic variation, we would expect the slope of the association to be similar across populations. If, however, the association between two traits was driven by allelic variation at one pleiotropic gene (or set of genes) in one population, and by allelic variation at different pleiotropic genes in another population, the slopes could differ. Of course, if genes with pleiotropic effects on two traits were fixed within a population, there would be no correlated response to selection.
If genetic correlations are driven by linkage, we expect that populations with different frequencies of sex will differ in their correlated responses. However, specific predictions depend on the underlying distribution of allelic variants. We can consider a simplification of the problem where two traits, A and B, are governed by variation at two linked loci, a and b. In one form of allelic variation, alternate alleles at loci a and b lead to different slopes of association between traits A and B. If allelic variation takes this form, we expect there to be no net trait correlation within a population, and thus our clonal selection on body size would yield no response in a second trait.
Alternatively, alternate alleles at loci a and b may lead to trait variation that slides along a common slope. Such variation would lead to the observation of genetic correlations among traits within a population. In this case, correlations that arise out of chance linkage are expected to be quickly eroded away by recombination in populations with relatively high frequencies of recombination. Therefore, correlated responses to selection would be stronger in populations with lower frequencies of sex. As these associations arise out of chance events, there is no force driving correlations to be similar in different populations, and therefore we expect slopes of association to differ among populations.
The situation under correlations which arise from linkage due to selection is more complicated. As with chance linkage, any correlations which do arise are going to be eroded by recombination, and we might expect this to result in populations with high frequencies of recombination to have smaller genetic correlations between traits than in populations with lower recombination. However, this expectation ignores the role of recombination in novel allele combinations with high fitness, and it is possible that the creative potential of sex outweighs the erosive effects of sex. If the net effect of sex is beneficial, and a particular correlation holds a selective advantage, sex would be creating high fitness clones that adhere closely to the optimal trait correlation. In taxa with the potential for asexual reproduction, such as Daphnia, these fit clones can spread through demographic expansion, shielding highly favored combinations from sexual erosion.
If the erosion of favorable combinations outweighs their origin, the net effect of sex is detrimental. This may occur, for example, if the capacity for asexual demographic expansion is limited or if few of the potential combinations of alleles are in fact advantageous. In this case, the predicted differences between populations with high and low frequencies of sex are qualitatively similar to the situation under chance linkage: high-sex populations should have smaller correlated responses than low-sex populations. However, the magnitude of the difference would be smaller than in the chance linkage situation, because selection is still countering the erosion via sex to some degree.