Main

Sexual selection is a widespread evolutionary force giving rise to a striking diversity of sights, sounds and smells that filter reproductive success away from less competitive or attractive individuals, frequently at the expense of survival6. Sexual selection will operate to varying degrees whenever sexual reproduction exists, and its significance as a potent force profoundly influencing reproductive fitness of individuals is long established6. In contrast, limited empirical work has been directed at measuring the consequences of sexual selection for the fitness of populations. This lack of attention is surprising for two reasons: first, because population viability is vital for biodiversity maintenance and ecosystem stability, especially under modern anthropogenic stress7,8; and second, because it is predicted that the maintenance of costly sex as the dominant mode of reproduction might only be possible if it allows sexual selection to operate, reducing the universal handicap of mutation load3,4.

Population or lineage health will always suffer at some level from mutation load—the difference in fitness between a (usually theoretical) mutation-free lineage and one that carries a load of deleterious mutations that are segregating in mutation-selection balance7,9. This load exists because new mutations continually arise in all populations every generation, most of which will be deleterious10. Haldane calculated that mutation load would be unexpectedly high11, owing to large numbers of loci across each genome presenting multiple targets to unavoidable mutation rates. Empirical estimates of mutation rate and load are hotly debated7, but we know, for example, that the average human lineage carries hundreds of deleterious loss-of-function mutations12, possibly thousands7. Natural selection will quickly remove mutations with large effects, but load persists through accumulation of mutations that have small individual effects, and/or exist as recessive alleles where their deleterious phenotypes are less frequently exposed to selection. Mutation load can therefore sum to a significant fitness constraint for a population which, because of its dispersed or concealed nature, is difficult to clear7,11. Sex could allow more effective purging of mutation load than asexual reproduction, if there are synergistic epistatic interactions between mutations7, so that their negative fitness impact is greater than strictly additive. However, the evidence that sufficient levels of synergistic epistasis exist remains equivocal7, so alternative explanations for the maintenance of costly sexual reproduction are sought.

Sexual selection could be a key filter against mutation load if, as Darwin acknowledged, “sexual selection will have given its aid to ordinary selection”13. This recognition that sexual selection places an additional, intraspecific filter on adaptive gene flow has been formalized by the idea of ‘genic capture’5, which proposes that reproductive success in the face of competition and choice depends on most or all aspects of an individual’s condition. Thus, sexual selection will act on most loci across the genome, purifying deleterious alleles from individuals within a lineage, and promoting fixation of advantageous ones, via three potentially connected routes. First, if competition or choice promotes non-random mating or fertilization success as an inverse consequence of mutation load, then deleterious dominant and recessive alleles will be under stronger purifying selection in lineages that experience sexual selection3. Second, if mutations are more deleterious in the competing sex than the choosing sex, which would occur if the competing sex suffers amplified reproductive success variance as a result of mutation load, then the equilibrium frequency of both dominant and recessive deleterious mutations will be lower within sexually selected populations, even accounting for synergistic epistasis in the competing sex2. Put simply, fathers should carry fewer mutations than all males3. Third, if female ability to maintain condition for reproductive fitness is under selection from interlocus sexual conflict, then sexual selection could purge mutations through females14,15. These routes for purging load via sexual selection provide theoretical explanations for how costly sex can persist as a dominant mode of reproduction2,3.

Few experiments have tested whether sexual selection removes deleterious mutations, yielding inconsistent findings. After introduction of mutations, heightened sexual selection rescued fitness faster in Rhizoglyphus robini mites16, Drosophila melanogaster flies17 and Onthophagus taurus beetles18. However, further work showed that sexual selection reduced mutation loads only for a subset of fitness traits in Drosophila19. More detailed studies found no evidence that sexual selection could purge deleterious alleles from experimentally evolving lineages20, or restore fitness after mutation load had been induced, even after 60 generations21; in fact, reproductive fitness within mutated lines became higher when sexual selection had been minimized21. These inconsistent results could be explained by interlocus sexual conflict22: while sexual selection might play a beneficial role in purging mutation load, direct short-term constraints on population productivity may also arise when female fitness is constrained by conflicting adaptations that promote only male reproductive potential22. Thus, short-term fitness costs arising simultaneously from interlocus sexual conflict could confound the measurement of longer-term fitness benefits arising from sexual selection7. To avoid this problem, we assayed mutation load after populations (from the same ancestry) had experienced almost 7 years of sole variation in the intensity of sexual selection (Extended Data Figs 1 and 2). Although no evidence for carry-over effects of interlocus sexual conflict exists in our model under these conditions15, we also removed potential trans-generational effects by enforcing two generations of monogamous reproduction before beginning to assay fitness. To expose mutation load in each population, we enforced inbreeding using sib × sib pairings, tracking fitness changes down multiple family lines (Extended Data Fig. 3). Because mutations will more likely persist as recessive alleles that are less frequently exposed to selection, compared with dominant wild types, fitness depression is a normal consequence of inbreeding as homozygosity increases the expression frequency of deleterious recessives through partial dominance23. By tracking extinction rates and fitness declines down inbreeding lineages perpetuated by monogamous sibling pairings, we tested the hypothesis that sex allows sexual selection to generate significant benefits through the purging of mutation load2,3, while avoiding the concurrent confound of interlocus sexual conflict22.

Experimental evolution lines began in 2005, with two different regimes (A and B) that both exposed replicate lineages to either ‘strong’ or ‘weak’ treatments of sexual selection, providing parallel independent experiments to measure whether sexual selection purges mutation load. Regime A exposed populations to 54 generations of divergent adult operational sex ratios that were either male-biased (10 females to 90 males) versus female-biased (90 females to 10 males), while regime B allowed polyandry (1 female to 5 males) versus enforced monogamy (1 female to 1 male) for 45 generations. All other conditions among lines and treatments within a regime were kept identical, including equalizing theoretical effective population sizes (see Methods, Extended Data Figs 2 and 4 and Supplementary Information for details confirming that heterozygosity was identical between treatments within either regime). The only difference between ‘strong’ and ‘weak’ treatments was therefore intensity of sexual selection during adult reproduction.

Our findings clearly showed for both regimes that lineages derived from populations experiencing evolutionary histories of strong sexual selection resisted extinction and maintained fitness more effectively when mutation load was exposed (Figs 1 and 2 and Extended Data Figs 5, 6, 7; all means presented in this article are arithmetic means ± standard errors). Over the 3-year extinction assay, families derived from male-biased populations survived 44% longer than families from female-biased populations (mean number of generations to extinction = 9.24 (±1.29 s.e.m.) versus 6.46 (±0.15); Fig. 1a). Families from polyandrous histories survived 37% longer on average than those from monogamous treatments (8.50 (±1.02) generations versus 6.21 (±0.46); Fig. 1b). When we combined data from regimes A and B into one analysis (incorporating sexual selection as a fixed factor variable), the history of sexual selection remained a significant predictor of number of generations to extinction (z = −3.43, P < 0.001), and there was no difference in extinction rates between regimes A and B (z = −0.51, P = 0.611), nor a significant interaction (z = 0.44, P = 0.660), revealing a consistent effect of sexual selection on extinction rates (Extended Data Fig. 5c). Overall, families derived from populations evolving under histories of strong sexual selection survived for 40% longer than those derived from weak sexual selection histories, giving an average survival time under mutation load exposure of 8.87 (±0.37) versus 6.33 (±0.13) generations respectively. All 108 initial families derived from weak sexual selection histories ceased to produce offspring beyond the 10th generation of inbreeding, whereas 8 of the 108 families from the strong sexual selection histories were still producing offspring after 20 generations of inbreeding.

Figure 1: Extinction trajectories under increasing inbreeding differ between family lines derived from strong (red squares) versus weak (blue circles) sexual selection histories.
figure 1

Each generation mean presents the average (±s.e.m.) proportion of surviving families for three independent lines per treatment. a, Regime A: male-biased (red) versus female-biased (blue) sexual selection histories. Each line is represented by 28 initial families (n = 84 total families for either treatment). b, Regime B: polyandrous (red) versus monogamous (blue) selection histories. Each line is represented by eight initial families (n = 24 total families for either treatment). Using parametric accelerated failure-time survival models with sexual selection treatment as a fixed effect, and incorporating correlated data within lines using a generalized estimating equation approach, we identified significantly lower extinction rates in populations that had previously experienced strong histories of sexual selection (Extended Data Fig. 5a, regime A: z = 3.40, P < 0.001; Extended Data Fig. 5b, regime B: z = 2.81, P = 0.005).

PowerPoint slide

Source data

Full size image
Figure 2: Reproductive fitness declines under increasing inbreeding of families derived from strong (red squares) versus weak (blue circles) sexual selection histories differ in magnitude and rate.
figure 2

a, Regime A: male-biased (red) versus female-biased (blue) selection histories. b, Regime B: polyandrous (red) versus monogamous (blue) histories. Each generation mean presents the average (±s.e.m.) number of offspring produced under standardized conditions for three independent lines per treatment. Lines are represented by eight initial families (n = 24 total families per treatment), and two breeding pairs per family. Average fitness under identical conditions but without inbreeding (‘Ref’) plotted for reference. Using generalized linear mixed models, and accounting for overdispersion as well as nesting non-independent replicate families within lines as random effects, we found in both regimes that inbreeding was a highly significant predictor of fitness (regime A: z = −21.34, P < 0.001; regime B: z = −21.17, P < 0.001), and that sexual selection treatment history was also a significant predictor of overall differences in fitness (regime A: z = −3.75, P < 0.001; regime B: z = −1.97, P = 0.048; see Extended Data Table 1 and Extended Data Figs 6 and 7).

PowerPoint slide

Source data

Full size image

Declines in reproductive fitness exhibited similar patterns to the extinction rates they underpinned. Having removed sexual selection history via enforced monogamy in regime B, baseline reproductive fitness even without inbreeding is substantially reduced (Fig. 2b shows that a polyandrous history improves baseline fitness by ∼30%). Most importantly, significant interactions between inbreeding and fitness decline revealed that fitness declines were much faster for families derived from treatments experiencing weak sexual selection histories in both regimes (z = 13.82 and 10.56 for A and B respectively, P < 0.001 for both; Fig. 2, Extended Data Table 1 and Extended Data Figs 6 and 7). To control for the possibility that failure-to-mate could have created differences between treatments, we repeated these analyses including only pairs that produced offspring: identical patterns of significant interactions between sexual selection history, inbreeding and fitness decline remained (z = 5.73 and 6.36 for A and B respectively, P < 0.001 for both; Extended Data Figs 6c, d and 7c, d).

Results from this 10-year experiment provide compelling empirical support for the complementary models given in refs 2 and 3, which argued that costs of sex1 could be offset by population genetic benefits derived from sexual selection. The most obvious mechanism explaining these differences is that heightened sexual selection, via ‘genic capture’4,5, more effectively strips out, or prevents, mutation load from becoming fixed in a population, strengthening its ability to withstand stress. Hemizygous selection could make sex-linked loci especially prone to this process24, but since the T. castaneum X chromosome makes up only ∼6% of the genome25, we expect purging of load to have occurred on the autosomes too. There is emerging evidence that sexual selection can profoundly shape the genome, with feminized patterns of sex-linked expression in the transcriptomes of both sexes after experimental evolution under monogamy in Drosophila26. Our findings indicate that sexual selection also acts to purge mutation load, even in populations that we expect to be adapting close to their natural fitness peaks27. Stronger sexual selection will drive greater variance in reproductive success, so that the average father should carry fewer deleterious mutations than the average male3, and perhaps also the average mother if interlocus sexual conflict constrains female fitness15.

Within the promiscuous Tribolium model28 (Extended Data Fig. 8 and Methods), both pre- and post-copulatory processes for winning or controlling fertilizations will operate through broad behaviours and physiology within the whole organism, down to competition and choice at the gamete level. Genic capture predicts that individual success across this continuum will depend on overall condition and genes at many loci5. Under monogamy, where sexual selection has been removed, populations suffer the constraints of sexual reproduction, but none of the benefits of sexual selection, explaining the lowered baseline fitness even without inbreeding, as well as more rapid fitness declines and heightened extinction rates under inbreeding (Figs 1 and 2 and Extended Data Figs 5, 6, 7). Although we exposed mutation load for experimental measurement through inbreeding, the intense and diverse demands from selection across multiple generations in the natural environment, where populations are more likely to be displaced from their fitness peaks27, are likely to expose even greater fitness differentials due to load variance. After only one generation of inbreeding, for example, mean fitness between strong versus weak sexual selection histories differs by 20% and 40% in regimes A and B respectively (Fig. 2). Our results indicate, as demands rise upon populations that have been depleted and fragmented, and displaced further from their fitness peaks27 via increasing anthropogenic stress29, that sexual selection could be an important force protecting species or populations from the extinction vortex30.

Methods

Experimental evolution lines

Beetles were of the widely used Georgia-1 (GA1) ‘wild type’ strain, originally collected from stored corn in 1980, and since cultured by the Beeman laboratory (United States Department of Agriculture), maintained under standard conditions. Adult virgin beetles from the same ancestral GA1 population were randomly allocated to begin their respective treatments within both regimes. Each generation, male and female pupae were separated and placed in fresh fodder (organic flour, yeast (10:1) and oats) for 10 days to allow adult emergence and sexual maturation15,28,31. Then, using the controlled sex ratios in the different sexual selection treatments, mature adults were placed in ad libitum fodder for 7 days to compete, choose, mate, oviposit and therefore reproduce under divergent intensities of sexual selection15,28,31. After 7 days, adults were removed and the eggs and larvae (typically about 70 offspring per female) were left to develop under standardized conditions with equal offspring densities by maintaining ad libitum levels of food in proportion to the number of offspring. These conditions were maintained for about 7 years, allowing the application of strong versus weak sexual selection in a total of 12 independent lines through sole variation in the adult operational sex ratio, while equalizing effective population size within regimes. Two parallel experimental evolution regimes were run, regimes A and B, which both applied treatments that created contrasting divergences in sexual selection intensities (see Extended Data Fig. 2 for visual details). Under regime A, the male-biased treatment provided 10 females with choice among 90 males, and simultaneous competition between the 90 males to fertilize the 10 females, whereas in female-biased treatments the reverse scenarios applied. Six independent lines (three per treatment) were maintained for 54 generations under regime A. Regime B generated divergence by enforcing either monogamy (20 replicate pairs per line), or allowing reproduction to be achieved through competition between five males under polyandry (12 replicate groups per line). Thus, polyandry provided each female with a choice among five males, and competition between five males for just one female, whereas monogamy completely removed all female choice and male–male competition. Six independent lines (three per treatment) were maintained for 45 generations under regime B. After each adult reproduction period and sexual selection treatment, offspring from replicate families within each independent line were pooled under standardized densities for larval development and genetic mixing to the next generation.

Quantifying sexual selection intensity

Divergence in the strength of sexual selection between the different treatments was estimated using a male mating potential assay, in which males from the ancestral GA1 control stock population were each provided with a series of 84 unmated females. Individual were sexed as pupae and matings took place after 10 days of adult maturation (as per the sexual selection treatments). Males were placed individually in microcentrifuge tubes with approximately 1 ml of flour topped with oats; 24 h before their trial they were identified with a small dot of white correction fluid on their thorax15,28,31. Females were placed in groups of six per Petri dish with 7 g of flour topped with oats. Mating assays were performed by placing individual males into Petri dishes containing standard fodder and six females, moving males to a new group of six females every 12 h for 7 days (12 × 7 = 84 females given to each male over the week). After each 12 h mating period, females were placed singly into Petri dishes with standard fodder for 7 days to lay eggs. Any offspring were then allowed to develop for 35 days before being frozen, and their presence noted to score successful matings. Eleven males were assayed, and the average number of successfully fertilized females across this 1 week period per male was 50 (±3 s.e.m.); Extended Data Fig. 8), providing extreme divergence in potential levels of female polyandry (and therefore sexual selection) between strong and weak sexual selection treatments. Thus, in regime A the 10 females at each adult generation in the male-biased treatment were potentially exposed to 90 × 50 = 4,500 successful matings, or 450 matings per female per week involving all 90 males. Whereas the 90 females in the female-biased treatment potentially experienced 500 matings (10 × 50), or 5.6 per female per week, enabling each female to be mated, on average, by 5.6 of the males available. In the monogamous treatment, we removed sexual selection altogether as only one male was available for mating, while the polyandrous treatment generated a fivefold increase in potential for pre- and post-copulatory male–male competition and female choice through the availability of five males, all of which could mate with the female.

Effective population sizes

To avoid differential inbreeding that could have subsequently influenced our extinction assay, we equalized the theoretical effective population size, Ne, for mixed adult sex ratios () (ref. 32) in our divergent sexual selection treatments; thus, in regime A Ne = 36 for both male- and female-biased treatments, and in regime B Ne = 40 for both monogamous and polyandrous treatments. It is important to note that strong sexual selection treatments with male-biased or polyandrous structures may translate into reduced realized Ne due to male success skew under strong sexual selection (where competition and choice could allow reproductive success by a smaller subset of males in the breeding population). Thus, our experimental design is conservative to any influence of Ne, because we test the prediction that lineages with histories of strong sexual selection (and therefore potentially lower Ne) should demonstrate resistance to extinction and reduced rates of fitness decline under inbreeding (because they have more effectively purged mutation load). If Ne is lower in these strong sexual selection populations, we would expect this to result in reduced heterozygosity at the start of inbreeding, resulting in lower initial fitness and therefore faster extinction rates (yet we observe the converse in Figs 1 and 2). As an additional, third, check for Ne confounds, we also directly measured heterozygosity and allelic richness in our lines to establish that differential inbreeding had not occurred under different sexual selection treatments (methods outlined below). By screening 628 individuals representing the different sexual selection treatments in both regimes A and B at 13 microsatellite loci (Supplementary Information), we were able to confirm that heterozygosity and allelic richness showed no differences between ‘weak’ or ‘strong’ sexual selection histories (Extended Data Fig. 4).

Exposing mutation load

Having been subjected to experimental evolution that applied strong versus weak sexual selection under equal effective population sizes, we then exposed the mutation load carried in these populations by inbreeding down multiple replicate family lines to expose deleterious recessives, and tracking fitness decline to extinction as a result of inbreeding depression through partial dominance23. No statistical methods were used to predetermine sample size, which was fundamentally controlled by the number of independent selection lines that had been maintained for 6 to 7 years. At the end of experimental evolution, we created multiple full-sib families using monogamous crosses of randomly chosen unmated males and females within each independent selection line, which simultaneously removed any transgenerational interlocus sexual conflict effects, and from which full-sib offspring were then used to continue a total of 216 family lines down increasing inbreeding coefficients and homozygosity (Extended Data Fig. 3). There were thus two generations of monogamy applied before fitness assays began, eliminating any carry-over effects of sexual selection or interlocus conflict. As levels of inbreeding increased, we tracked extinction rates and fitness declines for multiple families within each of the 12 independent lines from the original sexual selection treatments. In regime A, we created 28 full-sib families in each independent line (Ntotal = 28families × 3independent lines × 2sexual selection treatments = 168 families), allowing extinction rates to be measured within each of the three independent lines in either of the two sexual selection treatments. When sibling pairs failed to produce female and male offspring that survived to adulthood to enable continuation of the line, it was recorded as extinct (see Extended Data Fig. 3). In addition to extinction, we also measured reproductive fitness in a subset of these families, assayed as the average number of offspring produced by two randomly chosen sib × sib pairs within eight of the families per line (Extended Data Fig. 3). Identical protocols for regime B were followed, except that both extinction rates and average fitness were measured across eight families per independent line (Ntotal = 8families × 3independent lines × 2sexual selection treatments = 48 families). Blinding was not performed as the protocol did not permit biasing to affect results. Inbreeding was continued for 20 generations of sib × sib matings (which operated over 3 years), by which time Wright’s inbreeding coefficient F had increased to 0.986. After 20 generations of inbreeding, 208 of the 216 initial lines had gone extinct, with the eight survivors all derived from strong sexual selection histories.

Baseline fitness

At four time-points throughout the extinction assay (parental, F7, F15 and F21 generations), we assayed ‘baseline’ fitness (without inbreeding) of standard monogamous crosses from the different sexual selection treatments, using identical protocols to fitness measures applied for the inbred crosses. Pupae from each of the three independent lines within each of the four different selection treatments were separated into single-sex groups and isolated for 10 days to allow adult eclosion and sexual maturation. Twenty randomly chosen male–female pairs were then established for each of the three independent lines within each of the four sexual selection treatments. Each pair was placed into a 7 ml plastic vial, with ad libitum food and oats, and allowed to mate and oviposit for 7 days. After this, the flour from each vial, containing the eggs laid during the 7 day mating period, was transferred to a Petri dish containing a further 10 g of fodder, and the eggs allowed to develop to adult eclosion (∼35 days). Fitness was scored as the total number of adults produced from each cross across 7 days of mating and oviposition. Data from previous work15 where male–female pairs were allowed to interact for 7 days shows (a) that the first week of oviposition produces >25% of the total female reproductive fitness for such male–female interactions, and (b) that female fitness over the first week significantly correlates with total female reproductive fitness (r = 0.48, P = 0.008, n = 29 pairs). Average baseline fitness values are presented for reference in Fig. 2a, b (‘Ref’).

Extinction analyses

Extinction rates were analysed using the ‘survival’ package33 in R version 3.1.0 (ref. 34). To assess whether sexual selection history influenced extinction rate after 20 generations of sib × sib inbreeding for either regimes A or B, we analysed generation to extinction using a parametric accelerated failure-time survival model with Weibull baseline hazard distribution, taking into account right-censored data from families that were still alive at the end of the experiment. Generation of extinction was modelled with sexual selection treatment as a fixed effect, and the shape and scale of the underlying Weibull hazard distribution was allowed to vary by treatment. We also ran a combined analysis where both experimental evolution regimes were incorporated into the model, with level of sexual selection (strong versus weak), comparison (regime A or B) and their interaction as fixed effects. In all survival analyses, correlated data within lines were incorporated using a generalized estimating equation approach. Akaike’s information criteria and graphical interpretation of the complementary log–log survival plots were used to confirm appropriateness of the specified Weibull distribution; no violations of model assumptions were detected. Model fits are presented in Extended Data Fig. 5a–c.

Reproductive fitness analyses

We tested for a relationship between fitness and generation of inbreeding by fitting generalized linear mixed models (GLMMs) using the glmmADMB package35 in R 3.0.3 (ref. 34). We assumed a negative binomial error structure which added the parameter (k) to the variance mean relationship, allowing us to account for the overdispersion in our data introduced by a large number of zero fitness observations. GlmmADMB has two options for fitting the relationship between the mean and the variance. These are family = ‘nbinom1’, which assumes the variance = k × mean, and family = ‘nbinom2’, which assumes the variance = mean(1 + mean/k). We fitted models with each of these options in turn and then selected the parameterization that provided the lowest AIC score and therefore accounted for the greatest amount of variance in our data set. To account for the nested nature of the experimental design and the non-independence between replicate families, random effects were included in the model as replicate nested within family nested within line. Separate GLMMs were fitted to the regime A (male-biased versus female-biased) comparison and the regime B (polyandry versus monogamy) comparison (Extended Data Figs 6 and 7). To ensure that failure-to-mate was not a reason for differences in fitness declines between treatments, we repeated the analyses including only those families that produced some offspring each generation (Extended Data Figs 6c, d and 7c, d).

Microsatellite analyses

A total of 628 individuals from the 12 independent lines representing all four sexual selection treatments (strong versus weak in regimes A and B) were genotyped at 13 loci conforming to Hardy–Weinberg equilibrium (Supplementary Information). Scored genotypes for all individuals were analysed in Arlequin version 3.5 (ref. 36) to obtain allelic richness and observed and expected heterozygosity. We tested for differences in heterozygosity across selection treatments using linear mixed models, implemented in lme4 version 1.1-6 (ref. 37) package in R34, with significance testing performed using the package lmerTest version 2.0-6 (ref. 38). Heterozygosity was included as a response variable with selection treatment as a fixed factor and locus as a random factor. To assess whether independent line identity had a significant effect on genetic diversity beyond that of selection treatment, a second mixed model was tested with the same variables plus line identity as an additional random factor. The two models were then compared using likelihood ratio tests. All model residuals were tested for normality and no violations of model assumptions were found. Estimated heterozygosity from the linear mixed effect models and the associated standard error is plotted in Extended Data Fig. 4. Microsatellite data are available from the Dryad Digital Repository at http://dx.doi.org/10.5061/dryad.86750.

Code availability

Scripts of analyses and code used for figure production are available upon request from M.J.G.G.