Univariate responses in multicomponent signals
In nature, correlational selection may be quite common (Svensson et al., 2021) and sexual signals are often found to be subject to multivariate stabilizing and directional selection (Bentsen et al., 2006; Blankers et al., 2015; Brooks et al., 2005; Devigili et al., 2015; Fisher et al., 2009; Gerhardt and Brooks, 2009; Hine et al., 2011; Oh and Shaw, 2013; Ryan and Rand, 2003; Tanner et al., 2017). In our experiment, we applied a univariate selection gradient on the relative amount of Z11-16:OAc to consider the evolvability of multivariate traits to univariate selection gradients. We justified this univariate selection based on the found geographic variation in the relative amount of Z11:16OAc in H. subflexa . Females are likely specifically selected for higher rates of acetates to avoid heterospecific mate attraction in regions where the congener H. virescens occurs (Groot et al., 2009a).
Our observed selection response shows that there are no phenotypic or genetic constraints to the evolution of Z11-16:OAc (or the other acetates), as we reached similar or even higher levels of divergence compared to those observed across the range of H. subflexa . After 10 generations in our selection experiment, the high line individuals had on average almost 20% more Z11-16:OAc compared to the low line (Fig 2E), while H. subflexa populations in the eastern US were found to have blends with up to 10% more Z11-16:OAc compared to populations in the south-western US and Mexico (Groot et al., 2009a). The selection response that we found was also surprisingly univariate, as only the three acetate esters and none of the other components showed divergence between the high and low lines. This indicates that the acetates can evolve independently from the other components. This is surprising, both in the context of the observed genetic correlations in the starting population and in later generations (Fig 4) and in the context of what is known about the shared biochemical pathways across the pheromone components (Groot et al. 2009a; Fig 1). However, it is not an uncommon result of artificial selection experiments (Hill and Caballero, 1992; Saltz et al., 2017), including for sexual signals (Ritchie and Kyriacou, 1996). Thus, genetic correlations among sexual signal components may quickly dissolve in the absence of correlational selection, which likely facilitates their evolution on short time scales in response to directional selection.
We acknowledge that our study does not include replicate lines in the experiment. This is because each selection line consisted of hundreds of single pair matings per generation, as H. subflexa is highly sensitive to inbreeding. We thus needed to maintain large populations of individuals that are mated in pairs and also individually reared to avoid cannibalism, which makes these experiments very laborious. Therefore, we chose to rear two lines in separate directions rather than two replicates of only the high or low line. The fact that we found parallel results between the lines in the selection response both in phenotypic means and genetic covariances lends confidence to the patterns observed. However, we do avoid drawing conclusions based on differences between the lines, because these can both reflect differential selection effects or sampling variance.