INTRODUCTION
Animal taxa that engage in sexual communication typically show high among-species diversity in sexual signals (Andersson, 1994; Coyne and Orr, 2004; Ritchie, 2007; Schaefer and Ruxton, 2015; Wiens and Tuschhoff, 2020). Sexual signals are generally hypothesized to diversify due to directional or disruptive selection (Ritchie, 2007; Schaefer and Ruxton, 2015; West-Eberhard, 2014; Wilkins et al., 2013), although their evolution is still a mystery in many species. Since sexual signals play an important role in the origin and maintenance of species and contribute to biodiversity (Coyne and Orr, 2004), it is important to assess whether there are constraints to their selection response and identify the mechanisms that can mitigate those constraints.
Understanding selection responses in sexual signals is challenging, because signals are often composed of multiple components (Candolin, 2003; Higham and Hebets, 2013; Rowe, 1999). For example, mating songs can vary both in pitch and in rhythm (Blankers et al., 2015; Tanner et al., 2017; Wilkins et al., 2015), color signals can be composed of multiple, functionally distinct patches (Cole and Endler, 2015; Grether et al., 2004), and sex pheromones are often blends of multiple chemical compounds (Ferveur, 2005; Linn et al., 1987). To understand how multicomponent signals evolve, we thus need to consider the selection response in multiple dimensions simultaneously. Signal components can have a shared genetic or developmental basis, or can be subject to correlated selection pressures (Armbruster et al., 2014; Cheverud, 1996). The resulting genetic correlations among signals components can influence how selection on the phenotype translates to changes in the underlying genotypes (Chenoweth and Blows, 2006). To understand how the genotype-phenotype map of sexual signals influences the selection response, we thus need to determine the genetic correlations between the different signal components.
Statistical frameworks in quantitative genetics, in particular the (multivariate) breeder’s equation, allow us to predict and quantitatively understand selection responses in correlated traits. In this framework, the response to selection is a function of the genetic (G) and phenotypic (P) variance-covariance matrix and the selection gradient: selection acts on the P matrix and the resulting response is constrained by the G matrix (Lande, 1979; Lande and Arnold, 1983; Lynch and Walsh, 1998). The difficulty in predicting selection responses of multivariate traits is that selection acting on multiple components may be counterbalancing, e.g. directional selection on one trait, but stabilizing selection on correlated traits, resulting in evolutionary constraints (Barton and Turelli, 1989). Counterbalancing selection is likely prevalent in the evolution of sexual signals, as choosing individuals may favor higher or lower values of some component of the signal, while changes in correlated components may result in reduced mate attraction.
These evolutionary constraints can be overcome if genetic variances and covariances themselves respond to selection, thus reshaping the G matrix (Arnold et al., 2008; Barton and Turelli, 1989; Eroukhmanoff, 2009; Jones et al., 2003; Melo and Marroig, 2014; Revell, 2007; Roff and Fairbairn, 2012). Theory predicts that the G matrix will vary through time, because on one hand selection erodes genetic variance (Barton and Turelli, 1989; Estes and Arnold, 2007), while on the other hand mutation and introgression add new variation, albeit more slowly. Moreover, genetic correlations can respond to selection directly, especially if they result from interactions among unlinked genetic loci affecting the co-expression of multiple traits (Wolf et al., 2005), or from selection acting on correlations directly (Armbruster et al., 2014; Roff and Fairbairn, 2012; Svensson et al., 2021). Interestingly, directional selection can change genetic covariances and increase modularity, meaning that groups (modules) of traits become genetically independent from other modules, in a way that facilitates the phenotypic response to selection (Melo and Marroig, 2014). Empirical work has shown that patterns of covariation can indeed evolve both across populations and species in nature (Bégin and Roff, 2003; Berner et al., 2010; Blankers et al., 2017; Garant et al., 2008; Gosden and Chenoweth, 2014) and during artificial selection experiments (Careau et al., 2015; Hine et al., 2011; Uesugi et al., 2017). However, it is still unclear how changes in the phenotypic selection response are related to changes in genetic (co)variance through time.
In this study, we explored the response in phenotypic means and genetic (co)variances to artificial selection on the female sex pheromone of the moth Heliothis subflexa (Lepidoptera, Noctuidae). Like many other moths, H. subflexa females secrete a sex pheromone blend to which conspecific males are attracted. These sex pheromone blends are species-specific and vary among species in both the presence/absence of components as well as in relative amounts (or ratios) of the components (Cardé and Haynes, 2004; Schneider, 1992).
The sex pheromone blend of H. subflexa females consists of 11 compounds, with the following components that are critically important for conspecific male attraction: (Z)-11-hexadecenal (Z11-16:Ald) as the major sex pheromone component, (Z)-9-hexadecenal (Z9-16:Ald) and (Z)-11-hexadecenol (Z11-16:OH) as the two secondary sex pheromone components, without which H. subflexa males are not attracted (Groot et al., 2007; Vickers, 2002). Interestingly, the acetate esters (Z)-7-hexadecenyl acetate (Z7-16:OAc), (Z)-9-hexadecenyl acetate (Z9-16:OAc), and (Z)-11-hexadecenyl acetate (Z11-16:OAc), from here on referred to as “acetates”, have a dual role: these acetates attract conspecific males, while simultaneously repelling males of H. virescens (Groot et al., 2006; Vickers and Baker, 1997). In geographic regions where H. virescens is present, acetates are more abundant in the H. subflexa pheromone compared to where this species is absent (Groot et al., 2009a). This suggests that the acetates are subject to divergent selection across a geographic cline. The relative amounts of the other components are hypothesized to be under stabilizing selection across the range, as in general in moth pheromone communication the mean blend is preferred over deviations from the mean (Allison and Cardé, 2008; Groot et al., 2010; Kárpáti et al., 2013; Linn et al., 1997; Löfstedt, 1990; Zhu et al., 1997).
To explore the selection response of the acetates, we selected for higher and lower amounts of acetates during 10 generations of truncation selection. Since the acetates vary geographically (Groot et al., 2009a) and have a genetic basis that is partially independent of other components (Groot et al., 2009b), we hypothesized that the relative amount of acetates can evolve in response to univariate selection for higher/lower acetates but that genetic variance will be reduced after selection. However, since acetates also partially share their genetic basis with other components and since all sex pheromone compounds are produced through the same biosynthetic pathway, we also hypothesized that there will be indirect selection responses in other pheromone components, specifically in the unsaturated aldehydes, Z9-16:Ald and Z11-16:Ald, and alcohol, Z11-16:OH (Fig 1). Since some of the correlations among components may reduce the selection response, we also expected genetic covariances to change during selection, specifically in a way that facilitates the phenotypic selection response.