Sexual selection is thought to be responsible for the rapid divergent evolution of male genitalia with several studies detecting multivariate sexual selection on genital form. However, in most cases, selection is only estimated during a single episode of selection, which provides an incomplete view of net selection on genital traits. Here we estimate the strength and form of multivariate selection on the genitalia arch of Drosophila simulans when mating occurs in the absence of a competitor and during sperm competition, in both sperm defence and offense roles (i.e. when mating first and last). We found that the strength of sexual selection on the genital arch was strongest during non-competitive mating and weakest during sperm offense. However, the direction of selection was similar across selection episodes with no evidence for antagonistic selection. Overall, selection was not particularly strong despite genitals clearly evolving rapidly in this species.

Geometric morphometric analysis was used to quantify the size and shape of the posterior and ventral lobe of the genital arch (i.e. Centroid size (CS), Relative Warp (RW) 1, 2, 3 & 4). The independent variables in our multivariate selection analysis were Wing Size (i.e. WS), CS, RW1, RW2, RW3 & RW4.

We used standard multivariate selection analysis to estimate linear and nonlinear sexual selection on male body size (i.e. wing length, WL), genital size (CS) and shape (RW1, RW2, RW3 & RW4) during a non-competitive mating with a virgin female or competitive fertilization with a twice mated female during sperm defence (i.e. P1) or offense (i.e. P2). Male fitness was assigned a continuous fitness score that was the total number of offspring sired.

For each bout of selection, we transformed the response variables to relative fitness by dividing individual scores by the mean fitness of the population and standardized the male phenotypic traits to zero means and unit variances (Lande and Arnold 1983). We then fitted separate, linear and polynomial regression models for each of the three bouts of selection to estimate linear and nonlinear (i.e. quadratic and correlational) selection gradients for male size and genital size and shape during non-competitive (β_V and γ_V) and competitive (β_P1, β_P2, γ_P1 and γ_P2) mating (Lande and Arnold 1983, see Hunt et al. 2009 for details). All quadratic selection gradients were doubled as stabilizing or disruptive selection is underestimated by a factor of 0.5 (Stinchcombe et al. 2008).