https://orcid.org/0000-0001-9587-8665
Abstract
Sexual conflict plays a key role in the dynamics of adaptive evolution in sexually reproducing populations, and theory suggests an important role for variance in resource acquisition in generating or masking sexual conflict over fitness and life history traits. Here, I used a quantitative genetic genotype × environment experiment in Drosophila melanogaster to test the theoretical prediction that variance in resource acquisition mediates variation in sex-specific component fitness. Holding larval conditions constant, I found that adult nutritional environments characterized by high protein content resulted in reduced survival of both sexes and lower male reproductive success compared to an environment of lower protein content. Despite reduced mean fitness of both sexes in high protein environments, I found a sex*treatment interaction for the relationship between resource acquisition and fitness; estimates of the adaptive landscape indicate males were furthest from their optimum resource acquisition level in high protein environments, and females were furthest in low protein environments. Expression of genetic variance in resource acquisition and survival was highest for each sex in the environment it was best adapted to, although the treatment effects on expression of genetic variance eroded in the path from resource acquisition to total fitness. Cross-sex genetic correlations were strongly positive for resource acquisition, survival, and total fitness and negative for mating success, although estimation error was high for all. These results demonstrate that environmental effects on resource acquisition can have predictable consequences for the expression of sex-specific genetic variance but also that these effects of resource acquisition can erode through life history.
Introduction
The distribution of genetic variance for fitness and related traits plays an important role in governing the dynamics of adaptive evolution. In sexually reproducing populations comprised of separate sexes, it is the bivariate distribution of male and female fitness that determines the evolutionary rate (Bonduriansky & Chenoweth, 2009; Connallon & Hall, 2016; Lande, 1980). When both sexes harbor substantial genetic variance in fitness, and this variance is concordant—genotypes that have high fitness in one sex also tend to have high fitness when expressed in the other—then there is high potential for adaptive evolution. Alternatively, when genetic variance in fitness and related traits is sex-specific or sexually antagonistic, such that genotypes that have high fitness in one sex have low fitness when expressed in the other (Chippindale et al., 2001; Foerster et al., 2007), then this sexual conflict can impose a major constraint (Lande, 1980) on the evolutionary process.
Simple life history models provide insight into the factors that mediate the expression of genetic variance for fitness and its components. Assuming a life history can be defined by the acquisition of resources and subsequent allocation of resources to traits under selection, then variance in fitness can be decomposed into components related to variance in acquisition and variance in allocation strategy (van Noordwijk & de Jong, 1986). Thus, the key to the expression of genetic variance in fitness is the degree of variance (genetic or environmental) in resource acquisition. This framework applies directly to sexually reproducing populations (Zajitschek & Connallon, 2017), where it is male and female genetic variance in resource acquisition that, in part, determines the degree of sex-specific genetic variance in fitness (Figure 1). There are also reasons to expect resource allocation strategies to be sex-specific (Rowe & Houle, 1996), which could further contribute to generating sex-specific variance in fitness.
A graphical illustration of how (co)variance in resource acquisition and allocation determine cross-sex genetic correlations between fitness components (modified from Zajitschek & Connallon, 2017). Resources are acquired to determine the total resource pool (R) that an individual has available to allocate to a set of traits, zi-n, that are under selection. The genetic variance in resource acquisition and the correlation between male and female resource acquisition has significant consequences on the expression of genetic (co)variance in downstream traits, including rz mf and, most importantly, sex-specific fitness variance and the cross-sex genetic correlation for fitness rw mf. High values of rR mf can swamp sexually antagonistic selection on allocation to z, creating positive rw mf and masking SA. Alternatively, the evolution of sex differences in optimal resource acquisition could reduce rR mf, dramatically increasing sexual antagonism even if selection on allocation to z is not sexually antagonistic. Thus, the model predicts that the evolution of resource acquisition in males and females plays a central role in generating and resolving sexual conflict.
Although there are clear theoretical expectations for how sexual conflict for fitness is generated and impacts the dynamics of adaptive evolution, the empirical implications for actual organisms are less certain. Estimates of sex-specific genetic variance for fitness vary widely across populations and environmental conditions, even in the same study system (Barker et al., 2010; Gosden & Chenoweth, 2014; Punzalan et al., 2014). Experiments in laboratory organisms have shown that expression of genetic variance for fitness and component fitness can depend on the degree of adaptation to a given environment (Long et al., 2012), with the expectation that most genetic variance is concordant in situations of maladaptation (Connallon & Hall, 2016). However, it is unclear how male and female nutritional environments may mediate variation in the expression of genetic variance. Importantly, many past laboratory manipulations of nutritional environment focus on resource acquisition prior to the onset of the expression of sexual dimorphism (e.g., larval food resources, Bonduriansky, 2007; Bonduriansky & Rowe, 2005; Holman & Jacomb, 2017; Punzalan et al., 2014), making it difficult to assess the potential role that sex-specific resource acquisition may play in mediating sexual conflict given widespread evidence of sex-specific fitness effects of adult diet (Jensen et al., 2015; Maklakov et al., 2008; Reddiex et al., 2013).
Here, I present results of a half-sib genotype × environment (G × E) quantitative genetic experiment in Drosophila melanogaster designed to understand three key questions. First, how does the expression of genetic variance for fitness components change across adult nutritional environments which may differentially affect male and female fitness? Second, can environmental changes in the degree of sex-specific fitness variance be explained by shifts in genetic variance for resource acquisition, as predicted by life history theory? And third, how do these estimates of genetic variance relate to the adaptive landscape for resource acquisition in different nutritional environments? Drosophila flies are ideal for such a test; past work has demonstrated the importance of adult nutrition for both male and female fitness (Camus et al., 2017; Raubenheimer & Simpson, 1997; Reddiex et al., 2013), as well as variable degrees of sexual dimorphism in diet preference (Davies et al., 2018; De Lisle, 2023; Lee et al., 2013). My results illustrate how environmental shifts in the expression of genetic variance for resource acquisition and component fitness can mediate the environmental dependence of sexual conflict.
Materials and methods
Fly rearing and experimental design
Flies were obtained from a large stock population (LHm background; Lund-Hansen et al., 2020; Rice et al., 2005) maintained under standard conditions (25 °C, 12:12 light, and 60% relative humidity), and density controlled each generation. A diagram of the experimental design is provided in Supplementary Figure S1. A paternal half-sib pedigree was set up by haphazardly selecting 60 newly eclosed (virgin) male flies and placing them each individually in a standard fly vial with 3 newly enclosed females for 48 hr. Following this period of mating, each female was then transferred separately to a new individual vial for egg laying for three days, after which females were removed from the vials. Offspring from these matings—which were full siblings with others in the same vial and half-siblings with offspring from the same sire but different dam—were then, upon eclosion, transferred to individual vials containing a single 5 μl microcapillary tube containing one of two liquid diets; a high-yeast diet containing 1:0.7 yeast:sucrose or a high sucrose diet containing 1:7 yeast:sucrose at a constant concentration of 0.1 g/ml. These diet treatments correspond to protein:carbohydrate ratios of approximately 1:2 and 1:16, respectively, representing ends of a continuum of diet variation in D. melanogaster, of which male and female optima are at the extremes (Lee et al., 2008; Reddiex et al., 2013). Vials contained no regular fly food, although they did contain approximately 5 ml of agar solution lacking nutrition to prevent desiccation. Approximately four offspring from each genetic family were used: one male and one female in each respective liquid diet treatment; note that the power to measure additive genetic effects in a half-sib design is related to the number of sires and dams in the study; it is standard to measure only a single full-sib offspring (Lynch & Walsh, 1998). For a handful of families, multiple offspring of the same sex were allocated to the same treatment due to technician error; for these cases, the first value in the dataset was used in mixed models for the estimation of variance components. New microcapillary tubes were added to each vial after 48 and 96 hr. In order to control for the effects of evaporation on measures of liquid food consumption, control vials with agar solution but no flies were also set up to obtain a baseline evaporation rate for each liquid food media (N = 24). After 5 days of exposure to these diet treatments, all living flies were removed, the remaining fluid in microcapillary tubes measured to the nearest 0.01 mm with digital calipers (Mitutoyo series 500 IP67, with an instrumental measurement error of ±0.02 mm), and flies placed individually in new, regular food vials for competitive mating assays. Such measurements of Drosophila microcapillary feeding rates are an established approach for measuring fly food consumption (Garlapow et al., 2015; Reddiex et al., 2013). These competition vials contained standard fly food, the focal fly and two additional flies, a male and a female virgin fly from the same genetic background as the stock population but homozygous for the bw- brown eye mutation (Chippindale et al., 2001). Following 24 hr in these competition vials, all adult flies were removed and resulting offspring allowed to develop in the same set of vials. The number of these offspring with wild type eyes provided a measure of reproductive fitness.
Statistical analysis
To infer the main effects of treatment on component fitness, I used two separate generalized linear models with sex, treatment, and their interaction as main effects. No random effects were included because treatments were applied independently at the level of individual flies; regardless, conclusions remained the same when modeling random effects of genetic family. I assumed binomial error for analysis of survival during the 5-day period of diet exposure and Poisson error for offspring number for those individuals that survived. Analysis of the binomial wt:bw offspring, instead of the number of wt, produced similar conclusions (see Supplementary Material), but I present the analysis of the number of wt offspring as a measure of reproductive success because it is directly related to total fitness, which is the product of survival and reproduction. To infer the main effects on resource acquisition, I used a linear model (Gaussian error) with diet treatment, sex, and their interaction as fixed effects. This model contained fluid consumed, measured as the average of the fluid lost from each microcapillary tube a fly received, corrected for evaporation, as the response variable. Evaporation was corrected for by subtracting the average fluid lost from control fly vials containing microcapillaries with liquid food but lacking flies; separate corrections were performed for the two diet treatments to account for the differential evaporation of the different liquid diets.
I used a series of multiresponse mixed-effects models (Hadfield, 2010) to estimate the two-sex G matrix separately for resource acquisition (Gaussian), survival (threshold), reproductive success (Poisson), and total fitness (Poisson). Total fitness was calculated as the product of survival and reproductive success, and data from all individual flies were included in the study. I fitted each model by Bayesian Markov chain Monte Carlo and used model comparison based on deviance information criteria (DIC) of full (separate G for each nutritional environment) vs reduced (common G) models test the hypothesis that this matrix differs across adult nutritional environments. For survival, a binary trait, I fit a threshold model to estimate variance components but a categorical model for model comparison, as DIC is undefined for the threshold model. These models contained trait, treatment, and trait*treatment interaction as fixed effects. Two G-side random effects were modeled: sire and dam*sire. To assess treatment effects on the G matrix, I fit two models differing in this G-side structure: one with a single sire level covariance matrix and a second with separate matrices across low and high treatments. All G-side random effect covariance matrices were 2 × 2 matrices, with variances in male and female fitness components on the diagonal and between-sex covariance of these fitness components on the off-diagonal. All models also included heterogenous (i.e., sex-specific) residual variance; note that there is no defined covariance at this level as male and female fitness components were measured from different individuals. The residual variance was fixed to unity for the threshold model of survival as this residual variance is undefined. For the cross-sex genetic correlation, rmf, I present pooled (across treatment) estimates from the reduced model for all traits as sampling error was high, making treatment-specific estimates difficult to interpret. For all models, 500,000 iterations of the Markov chain were run, with a burn-in of 100,000 and a thinning interval of 400 to ensure low autocorrelation among samples. Uninformative priors were used. MCMCglmm (Hadfield, 2010) assumes an inverse Wishart for the prior distribution of random effects. For G-side random effects, I assumed V = I(2) and ν = 0.02. Default uninformative priors were assumed for fixed effects.
I used Poisson regression with total fitness (survival during the 5-day diet exposure treatments × reproductive success) as the response to infer phenotypic selection on resource acquisition, measured as the volume of liquid food media consumed by each fly. To assess the statistical significance of treatment effects, I fit a single regression model with resource acquisition, treatment, and sex as fixed effects. This model included all two- and three-way interactions; interactions with resource acquisition indicate differences in phenotypic selection across treatments. This model, which already contained eight fixed-effect parameters, focused on directional selection only to avoid unnecessary complexity, although I estimated nonlinear selection gradients separately by treatment and sex (see below). I modeled absolute fitness as the response variable, noting that the log-linear Poisson model treats fitness on a log scale, analogous to relativizing fitness by the mean within each treatment*sex combination (De Lisle & Svensson, 2017).
Following support for treatment*sex differences in phenotypic selection, I estimated standardized directional and quadratic selection gradients separately for each treatment*sex combination. To estimate standardized selection gradients directly from the log-linear model, following Morrissey and Goudie (2022), I calculated linear and quadratic gradients as,
where and are the corresponding linear and (doubled; Stinchcombe et al., 2008) quadratic terms from a Poisson regression. I fit separate Poisson regressions to estimate and as the distribution of resource acquisition values was skewed (Lande & Arnold, 1983). I mean-centered and variance standardized resource acquisition separately for each sex*treatment group. Standard errors for and were obtained by resampling a multivariate normal distribution centered at the estimates of and with covariance equal to the information matrix from the fitted quadratic Poisson GLM (Morrissey & Goudie, 2022). Because all sex*treatment combinations showed a mixture of directional and stabilizing selection, it was appropriate to calculate the distance from the optimum phenotype, taken as (Phillips & Arnold, 1989), which gives the location of the optimum in units of phenotypic standard deviations from the mean. Full dataset and R script are available at https://doi.org/10.5281/zenodo.10707240.
Results
The final dataset consisted of fitness measures from 534 individual flies (Figure 2A and B), with fitness measures for both sexes obtained for a total of 125 (low protein) or 122 (high protein) genetic families. Fitness data for both sexes in both environments was obtained for 113 families. Survival was lower for both sexes in the high protein environment (GLM z = 2.302, p = 0.021) and higher overall for males than for females in both environments (GLM z = 3.66, p = 0.000246; Figure 2C); there was no evidence of a sex*environment interaction for survival (GLM z = 0.012, p = 0.99). Survival estimates were generally similar to other studies of nutritional effects in D. melanogaster (Lee et al., 2008). There were no main effects of treatment on the number of offspring produced in a competitive mating assay (GLM z = 1.608, p = 0.108), although there was a significant sex*treatment interaction (GLM z = 3.997, p = 6.41 × 10−5); males, suffered reduced mating success in high protein environments relative to low protein environments, although this effect was not observed in females (Figure 2D). Total fitness was thus lower in high-protein environments than in low-protein environments (Figure 2A and B). There was a sex*environment interaction for resource acquisition, measured as the volume of liquid media consumed in microliters (controlling for evaporation); females consumed more food in the low protein treatment than males, while both sexes showed similar, lower levels of food intake in the high protein treatment (LM sex*treatment t = −2.246, p = 0.0251; treatment t = 6.659, p = 6.9 × 10−11; Figure 3).
Total and component fitness under low and high protein adult nutritional environments. Panels (A) and (B) show distributions of total fitness for females and males in both nutritional environments. Survival was higher for males and in low protein environments (C). Fecundity was lower for males in high-protein environments than in low-protein environments but did not differ for females (the sex*treatment interaction was significant). Panels (C) and (D) show marginal effects, and 89% CIs from GLMs.
Food consumption in low and high protein nutritional environments. Both sexes consumed more liquid food media in low-protein conditions than in high-protein conditions, although there was little evidence of a sex difference in consumption in high-protein environments. Points are marginal effects and 89% CIs from a linear model.
Genetic variance, as captured in estimates of the univariate two-sex G matrix, differed substantially for resource acquisition and survival, but there was little evidence for treatment effects on G for reproductive success or total fitness (see Table 1). Female genetic variance in resource acquisition was higher in high-protein environments than in low-protein environments, while the opposite was true for males, which showed elevated levels of genetic variance in low-protein environments (Figure 4). Although there was statistical support for environmental differences in G for survival, the effects were weaker (Figure 4). There was little statistical support for environmental differences in G for reproductive success or overall fitness (Table 1; Figure 4). Pooled (by treatment) estimates of the cross-sex genetic correlation, rmf, were strongly positive for resource acquisition, survival, and total fitness and negative for mating success, although confidence limits for all estimates ranged from strongly negative to strongly positive (Table 2).
Model comparison of multiresponse mixed-effects models to assess environmental differences in the two-sex G.
| Trait | DIC.Full | DIC.Reduced | ∆ DIC |
|---|---|---|---|
| Resource acquisition | 1,238.189 | 1,241.938 | −3.748 |
| Survival | 658.466 | 660.736 | −2.269 |
| Reproduction | 2,206.747 | 2,206.853 | −0.105 |
| Fitness | 2,314.078 | 2,315.134 | −1.055 |
| Trait | DIC.Full | DIC.Reduced | ∆ DIC |
|---|---|---|---|
| Resource acquisition | 1,238.189 | 1,241.938 | −3.748 |
| Survival | 658.466 | 660.736 | −2.269 |
| Reproduction | 2,206.747 | 2,206.853 | −0.105 |
| Fitness | 2,314.078 | 2,315.134 | −1.055 |
Model comparison of multiresponse mixed-effects models to assess environmental differences in the two-sex G.
| Trait | DIC.Full | DIC.Reduced | ∆ DIC |
|---|---|---|---|
| Resource acquisition | 1,238.189 | 1,241.938 | −3.748 |
| Survival | 658.466 | 660.736 | −2.269 |
| Reproduction | 2,206.747 | 2,206.853 | −0.105 |
| Fitness | 2,314.078 | 2,315.134 | −1.055 |
| Trait | DIC.Full | DIC.Reduced | ∆ DIC |
|---|---|---|---|
| Resource acquisition | 1,238.189 | 1,241.938 | −3.748 |
| Survival | 658.466 | 660.736 | −2.269 |
| Reproduction | 2,206.747 | 2,206.853 | −0.105 |
| Fitness | 2,314.078 | 2,315.134 | −1.055 |
Estimates of cross-sex genetic correlations.
| Trait | rmf | HPD interval |
|---|---|---|
| Resource acquisition | 0.63 | (−0.53, 0.92) |
| Survival | 0.91 | (−0.63, 0.98) |
| Reproduction | −0.847 | (−0.95, 0.86) |
| Fitness | 0.97 | (−0.89, 0.99) |
| Trait | rmf | HPD interval |
|---|---|---|
| Resource acquisition | 0.63 | (−0.53, 0.92) |
| Survival | 0.91 | (−0.63, 0.98) |
| Reproduction | −0.847 | (−0.95, 0.86) |
| Fitness | 0.97 | (−0.89, 0.99) |
Note. Shown are the posterior mode and 95% highest posterior density interval from the reduced Bayesian mixed model for each trait.
Estimates of cross-sex genetic correlations.
| Trait | rmf | HPD interval |
|---|---|---|
| Resource acquisition | 0.63 | (−0.53, 0.92) |
| Survival | 0.91 | (−0.63, 0.98) |
| Reproduction | −0.847 | (−0.95, 0.86) |
| Fitness | 0.97 | (−0.89, 0.99) |
| Trait | rmf | HPD interval |
|---|---|---|
| Resource acquisition | 0.63 | (−0.53, 0.92) |
| Survival | 0.91 | (−0.63, 0.98) |
| Reproduction | −0.847 | (−0.95, 0.86) |
| Fitness | 0.97 | (−0.89, 0.99) |
Note. Shown are the posterior mode and 95% highest posterior density interval from the reduced Bayesian mixed model for each trait.
Posterior distributions of genetic variances for fitness components. Distributions show either the raw posterior (for the cross-sex covariances in the top row) or the posterior on a log scale (for sex-specific variances) of the sire variance component for bivariate mixed models fit separately for each trait/fitness component. There was statistical support for treatment effects on the sire covariance matrix for resource acquisition and survival but not for reproductive success or total fitness. Genetic correlations are presented in Table 2.
Selection on resource acquisition differed significantly by sex and environment (sex*environment*resource acquisition interaction, Poisson GLM z = −2.52, p = 0.011), although overall, there was positive directional selection on resource acquisition and negative quadratic curvature (stabilizing selection) of the fitness surface (Table 3), captured both in Poisson GLMs (Figure 5) and in estimates of standardized selection gradients calculated separately by sex and treatment (Table 4). Although each sex was displaced below its optimum in each environment (i.e., there was positive directional selection; Figure 5), males were closer to their optimum in the low protein environment, whereas females were closer to their optimum resource acquisition level in the high protein environment (Figure 6). That is, there is evidence of sex-specific adaptation to the different nutritional environments consistent with past work on nutritional requirements of insects, although it should be noted that sampling error for the location of the optimum was high despite statistical support for sex and treatment differences in selection.
Poisson GLM to assess treatment effects on sex-specific selection on resource acquisition.
| Effect | Estimate | SE | Z | p |
|---|---|---|---|---|
| (Intercept) | 1.818759 | 0.057736 | 31.501 | <2 × 10−16 |
| Resource acquisition | 0.415489 | 0.045644 | 9.103 | <2 × 10−16 |
| Sex | 0.689204 | 0.071722 | 9.609 | <2 × 10−16 |
| Environment | 0.234847 | 0.084261 | 2.787 | 0.005318 |
| Sex*Resource acquisition | 0.009804 | 0.057808 | 0.17 | 0.865325 |
| Treatment*Resource acquisition | −0.088139 | 0.056758 | −1.553 | 0.120447 |
| Sex*Treatment | 0.339636 | 0.100289 | 3.387 | 0.000708 |
| Sex*Treatment*Resource acquisition | −0.177709 | 0.070416 | −2.524 | 0.011612 |
| Effect | Estimate | SE | Z | p |
|---|---|---|---|---|
| (Intercept) | 1.818759 | 0.057736 | 31.501 | <2 × 10−16 |
| Resource acquisition | 0.415489 | 0.045644 | 9.103 | <2 × 10−16 |
| Sex | 0.689204 | 0.071722 | 9.609 | <2 × 10−16 |
| Environment | 0.234847 | 0.084261 | 2.787 | 0.005318 |
| Sex*Resource acquisition | 0.009804 | 0.057808 | 0.17 | 0.865325 |
| Treatment*Resource acquisition | −0.088139 | 0.056758 | −1.553 | 0.120447 |
| Sex*Treatment | 0.339636 | 0.100289 | 3.387 | 0.000708 |
| Sex*Treatment*Resource acquisition | −0.177709 | 0.070416 | −2.524 | 0.011612 |
Poisson GLM to assess treatment effects on sex-specific selection on resource acquisition.
| Effect | Estimate | SE | Z | p |
|---|---|---|---|---|
| (Intercept) | 1.818759 | 0.057736 | 31.501 | <2 × 10−16 |
| Resource acquisition | 0.415489 | 0.045644 | 9.103 | <2 × 10−16 |
| Sex | 0.689204 | 0.071722 | 9.609 | <2 × 10−16 |
| Environment | 0.234847 | 0.084261 | 2.787 | 0.005318 |
| Sex*Resource acquisition | 0.009804 | 0.057808 | 0.17 | 0.865325 |
| Treatment*Resource acquisition | −0.088139 | 0.056758 | −1.553 | 0.120447 |
| Sex*Treatment | 0.339636 | 0.100289 | 3.387 | 0.000708 |
| Sex*Treatment*Resource acquisition | −0.177709 | 0.070416 | −2.524 | 0.011612 |
| Effect | Estimate | SE | Z | p |
|---|---|---|---|---|
| (Intercept) | 1.818759 | 0.057736 | 31.501 | <2 × 10−16 |
| Resource acquisition | 0.415489 | 0.045644 | 9.103 | <2 × 10−16 |
| Sex | 0.689204 | 0.071722 | 9.609 | <2 × 10−16 |
| Environment | 0.234847 | 0.084261 | 2.787 | 0.005318 |
| Sex*Resource acquisition | 0.009804 | 0.057808 | 0.17 | 0.865325 |
| Treatment*Resource acquisition | −0.088139 | 0.056758 | −1.553 | 0.120447 |
| Sex*Treatment | 0.339636 | 0.100289 | 3.387 | 0.000708 |
| Sex*Treatment*Resource acquisition | −0.177709 | 0.070416 | −2.524 | 0.011612 |
Standardized selection gradients for each sex*environment combination.
| Sex | Environment | β | SE | γ | SE |
|---|---|---|---|---|---|
| M | Low protein | 0.090509 | 0.014874 | −0.278421 | 0.011012 |
| M | High protein | 0.149633 | 0.015545 | −0.373672 | 0.017224 |
| F | Low protein | 0.147805 | 0.020614 | −0.37612 | 0.024444 |
| F | High protein | 0.143732 | 0.021335 | −0.45732 | 0.033975 |
| Sex | Environment | β | SE | γ | SE |
|---|---|---|---|---|---|
| M | Low protein | 0.090509 | 0.014874 | −0.278421 | 0.011012 |
| M | High protein | 0.149633 | 0.015545 | −0.373672 | 0.017224 |
| F | Low protein | 0.147805 | 0.020614 | −0.37612 | 0.024444 |
| F | High protein | 0.143732 | 0.021335 | −0.45732 | 0.033975 |
Standardized selection gradients for each sex*environment combination.
| Sex | Environment | β | SE | γ | SE |
|---|---|---|---|---|---|
| M | Low protein | 0.090509 | 0.014874 | −0.278421 | 0.011012 |
| M | High protein | 0.149633 | 0.015545 | −0.373672 | 0.017224 |
| F | Low protein | 0.147805 | 0.020614 | −0.37612 | 0.024444 |
| F | High protein | 0.143732 | 0.021335 | −0.45732 | 0.033975 |
| Sex | Environment | β | SE | γ | SE |
|---|---|---|---|---|---|
| M | Low protein | 0.090509 | 0.014874 | −0.278421 | 0.011012 |
| M | High protein | 0.149633 | 0.015545 | −0.373672 | 0.017224 |
| F | Low protein | 0.147805 | 0.020614 | −0.37612 | 0.024444 |
| F | High protein | 0.143732 | 0.021335 | −0.45732 | 0.033975 |
Individual fitness surfaces for male and female resource acquisition in low- and high-protein environments. Fitted lines show quadratic Poisson regressions fitted separately for each sex*treatment combination. Similar conclusions on the shape of the fitness surfaces were obtained from cubic splines, suggesting the quadratic approximation is appropriate.
Distance from the optimum for each sex in each nutritional environment. Distance estimated as , where variance standardized selection gradients were calculated from regression coefficients from Poisson regression (see text). Standard errors were calculated empirically by resampling the information matrix from the fitted Poisson regressions.
Discussion
Life history theory suggests that variance in fitness reflects a combination of variance in individual resource acquisition along with variance in how resources are allocated to traits under selection (van Noordwijk & de Jong, 1986). For sexually reproducing organisms, patterns of sex-specific resource acquisition thus play a potentially important role in mediating the potential for sexual conflict for fitness (Zajitschek & Connallon, 2017). Here, I used a half-sib experiment in D. melanogaster to understand how the adult nutritional environment mediates the expression of resource acquisition variance, component fitness variance, and phenotypic selection in males and females. I found that each sex expresses greater genetic variance in resource acquisition rate in the nutritional environment they are best adapted to, as measured by distance from the phenotypic optimum resource acquisition level, and that this pattern of variance in resource acquisition manifests a similar pattern of genetic variance in survival. However, these treatment effects on genetic variance erode when examining reproductive fitness, with little evidence of treatment effects on genetic variance in mating success or total fitness. These data suggest that the nutritional environment can have important and predictable effects on the expression of genetic variance in fitness components.
I found that fitness was lower, and selection on resource acquisition was stronger for both sexes in the high protein nutritional environment. The observation of reduced survival for both sexes and reduced competitive mating success for males in high-protein environments are generally consistent with past work on the nutritional fitness effects in flies and other insects (Camus et al., 2017, 2018; Garlapow et al., 2015; Jensen et al., 2015; Maklakov et al., 2008; Reddiex et al., 2013). In both environments, the selection of male and female resource acquisition can be characterized as a combination of directional and stabilizing selection. This indicates that not only is resource acquisition under positive directional selection, as generally expected, but also that an optimum level of resource acquisition exists within the vicinity of the phenotypic distribution. Despite consistent treatment differences in the strength of directional selection, male mean resource acquisition was closer to the optimum in the low-protein nutritional environment and further in high-protein, while the opposite was true for females. This suggests that males are better adapted to low-protein nutritional environments and females to high-protein environments, which is consistent with past work in D. melanogaster (Camus et al., 2017; Reddiex et al., 2013) that used different experimental designs to assess selection on resource acquisition.
I found no evidence of a consistent male bias in genetic variance, as has sometimes been reported for phenotypic traits (Wyman & Rowe, 2014) and fitness (Singh & Agrawal, 2022). Estimates of cross-sex genetic correlations for all traits were high, either strongly positive (resource acquisition, survival, and total fitness) or negative (reproductive success). This pattern suggests genetic variance is largely sexually concordant for both fitness and traits under natural selection, such as resource acquisition and survival, while the negative estimate of rmf for competitive mating success suggests the presence of sexually antagonistic genetic variance for this fitness component consistent with past work (Chippindale et al., 2001). However, it is noteworthy that these genetic correlations are estimated with substantial error, with confidence regions widely overlapping zero. Nonetheless, the point estimates of rmf are consistent with a scenario of elevated sexual conflict for mating success, with concordant variance for resource acquisition and survival.
Several caveats to the interpretation of these data are warranted. First, with fitness estimates from just over 500 flies from 60 sires, there was relatively limited power to infer treatment effects on genetic variance components, particularly cross-sex genetic correlations, which are notoriously challenging to estimate (Bonduriansky & Chenoweth, 2009). Second, while fecundity was assessed in a competitive mating assay, thus capturing a measure of competitive reproductive success, this assay was performed only once, and so the contribution of late-life fecundity (Lee et al., 2008; Maklakov et al., 2008) was missed. Although this could be important for females, the assessment of reproductive fitness at approximately 5 days posteclosure is consistent with the evolutionary history of LHm flies. Third, feeding rates were determined by fluid loss from microcapillary tubes, a method subject to error arising from evaporation (De Lisle, 2023), although evaporation was controlled for in calculations of resource acquisition.
Data and theory suggest that both selection and standing genetic variance may depend fundamentally on the location of a population relative to a phenotypic optimum. For example, when both sexes are maladapted and thus far from their sex-specific optima, selection and genetic variance are expected to be sexually concordant (Connallon & Hall, 2016), an expectation that seems to be supported by a range of empirical data from the lab and wild (De Lisle et al., 2018; Long et al., 2012). In this study, I was able to assess the effects of adaptation when both sexes are relatively close to their optima. Selection in this experiment was largely sexually concordant, although subtle differences in the strength of directional selection, the curvature of quadratic selection, and differences in male and female mean resource acquisition result in a situation where the sexes differ in the distance from their phenotypic optima in each environment, ultimately revealing that each sex is best adapted to a different nutritional environment. Estimates of genetic variance further reveal that variance in resource acquisition is highest for each sex in the environment they are best adapted to. Although it is unclear why this may be the case, it is noteworthy that the same pattern holds for genetic variance in survival and for female mating success and total fitness. This finding of a concordance between variance in resource acquisition and component fitness is consistent with life history theory (van Noordwijk & de Jong, 1986; Zajitschek & Connallon, 2017), and the breakdown in the path to total fitness suggests other traits not associated with adult resource acquisition play a role in generating fitness variance. In D. melanogaster, this is to be expected, in part because most resource acquisition occurs at the larval stage.
This study also highlights the importance of directly inferring features of the adaptive landscape in making inferences about the degree of adaptation. Both sexes suffered reduced mean fitness in high protein environments, although this environmental effect on mean fitness does not entail a further displacement from the optimum resource acquisition level for females. Thus, although mean fitness was reduced, this reduction in mean fitness does not correspond to a large displacement of both sexes from the optimum resource acquisition level. This sort of discrepancy illustrates the utility of phenotypic selection analysis to infer features of the adaptive landscape (Arnold et al., 2001; Lande & Arnold, 1983), even in laboratory studies of genetic variance.
As a proof of concept, these results partially support the theoretical expectation that sex-specific genetic variance in fitness components is related to variance in resource acquisition, as well as the environment in which genetic variance is expressed (Hoffman & Merilä, 1999). More generally, the findings of environmental effects on sex-specific resource acquisition and sex-specific selection add to a growing body of work (Arbuthnott et al., 2014; De Lisle, 2019, 2023) demonstrating the role that ecological factors, such as diet and nutrition, can play in sex-specific adaptation.
Data availability
Data and code are provided available at https://doi.org/10.5281/zenodo.10707240
Funding
This work was funded by grants from the Swedish Research Council (VR grant number 2019-03706 to S.P.D.), the Crafoord Foundation (20220602), Formas (grant 2021-01096), and the Royal Physiographical Society of Lund (Kungl Fysiografiska Sällskapet i Lund grants 42305, 41593) to S.P.D.
Acknowledgments
I thank K. Lund-Hansen for patient advice on Drosophila and A. Singh, E. Svensson, and J. Abbott for sharing knowledge and resources. Petronella Romberg kindly assisted in setting up the experiment.
Conflicts of interest
None declared.
