Natural selection and genetic variation for female resistance to harm from males


William R. Rice, Department of Ecology, Evolution, and Marine Biology, University of California, Santa Barbara, CA 93106-9610, USA.
Tel.: +1-805-893-5793; fax: +1-805-893-4724;


The sexual conflict hypothesis predicts that males evolve traits that exploit the higher parental investment of females, which generates selection for females to counter-evolve resistance. In Drosophila melanogaster it is now established that males harm females and that there is genetic variation among males for the degree of this harm. Genetic variation among females for resistance to harm from males, and the corresponding strength of selection on this variation, however, have not been quantified previously. Here we carryout a genome-wide screen for female resistance to harm from males. We estimate that the cost of interactions with males depresses lifetime fecundity of females by 15% (95% CI: 8.2–22.0), that genetic variation for female resistance constitutes 17% of total genetic variation for female adult fitness, and that propensity to remate in response to persistent male courtship is a major factor contributing to genetic variation for female resistance.


Sexual conflict predicts that the sexes interact in a chronic cycle of adaptation and counter-adaptation (Trivers, 1972; Dawkins, 1976; Parker, 1979; Arnqvist & Rowe, 1995; Gowaty, 1997; Rice & Holland, 1997; Chapman et al., 2003). Males evolve traits that increase their fitness, but that also harm females via pleiotropy. Females are expected to respond by evolving counter-adaptations that provide resistance to male traits that harm them. The phenomenon of males harming their mates by reducing female survival and/or fecundity has been documented in a wide diversity of taxa (Dean, 1981; McKinney et al., 1983; Kasule, 1986; Partridge et al., 1987; Arnqvist, 1989; Fowler & Partridge, 1989; Partridge & Fowler, 1990; Burpee & Sakaluk, 1993; Chapman et al., 1995; Rice, 1996; Crudgington & Siva-Jothy, 2000; Moore et al., 2001; Pitnick & Garcia-Gonzalez, 2002), but quantitative assays of genetic variation for traits mediating sexual conflict have been restricted primarily to the Drosophila melanogaster model system. These studies have focused on male harm to their mates rather than female resistance to this harm. In order for sexually antagonistic coevolution to proceed, genetic variation must exist for both male harm and female resistance to this harm. While there is empirical evidence for substantial genetic variation among males for harm to their mates (Civetta & Clark, 2000; Sawby & Huges, 2001) and among females for their influence on male fertilization success in both Drosophila (Clark & Begun, 1998; Clark et al., 1999) and other species (Wilson et al., 1997; Nillsson et al., 2003), only indirect evidence indicates that there may be ample standing genetic variation for female resistance to male harm (Holland & Rice, 1999; Wigby & Chapman, 2004). These latter two studies of experimental evolution found that females resistance to male-induced harm evolved in response to changes in the social environment, but because the experiments were carried out in a novel environments, they did not assess standing heritable variation for female resistance.

Here we extend previous studies by comparing the lifetime fecundity of D. melanogaster females that were continuously exposed to males, to that of females that received minimal exposure to males (i.e. that just sufficient to inseminate them). All measurements were taken under environmental conditions that closely matched those to which the experimental population had adapted for over 300 generations. Our comparison of female fecundity under continuous vs. minimal exposure to males allowed us to measure total male harm to females, including both male behaviour (e.g. persistent courtship and copulation attempts) and the influence of seminal fluid proteins obtained when females mate with more than one male. We specifically assayed: (i) the degree to which all aspects of male harm collectively reduced the lifetime fecundity of females, (ii) the amount of standing genetic variation for female resistance to harm from males, (iii) the degree to which genetic variation for resistance to male harm contributed to total genetic variation among females for their lifetime fecundity, and (iv) the degree to which the act of remating, beyond what is necessary to maintain an ample supply of stored sperm, contributed to reduced harm to female fitness. We found strong selection for females to resist harm from males and that genetic variation among females for resistance is a substantial component of total genome-wide variation for lifetime fitness. These observations support the hypothesis that intersexual conflict drives perpetual antagonistic coevolution between the sexes.


Base population

The assay for genetic variation for female resistance to harm from their mates was done with a large outbred laboratory population (LHM) that had adapted to laboratory conditions for over 300 generations at the start of these experiments. The LHM base population had been reared on a 14-day, discrete generation cycle in which the flies were sequentially transferred to three consecutive vials. On day 0 eggs were laid in the first set of 56 ‘juvenile competition’ vials, and the hatching offspring remained in this first set of vials through the larval, pupal and early adult stages (at a density of 150–200 per vial). On day 12 the adult flies were mixed among vials and 1792 randomly selected individuals were transferred to a second set of 56 ‘adult competition’ vials (at 16 pairs of adults per vial) where, for 2 days, females competed for a limited supply of live yeast (10 mg) and males competed for fertilization opportunities. Most offspring that began the next generation were produced from matings that occur during this period of time in the adult competition vials (unpublished data). Eighteen hours before the end of the 14-day life cycle, the adult flies were transferred to a third set of 56 vials (‘oviposition vials’, without live yeast), and eggs produced during this time, trimmed to 150–200 per vial, were used to begin the next generation (i.e. at least 8400 eggs started each new generation). Because only the eggs laid during this 18-h period contributed to the next generation, fecundity measured at this time constitutes a female's lifetime fecundity. The protocols used in the assays described below closely matched the timing of events, culture medium, and densities used during the normal culturing protocol of the base population, so that fitness was measured under the environmental conditions to which the flies were adapted.

Screen for female resistance

The rationale for the genetic screen was to compare the lifetime fecundity of females that were given minimal exposure to males during the adult phase of their life (just sufficient to fully inseminate them for the duration of the adult phase of their 14-day life cycle) to that of females who were continuously exposed to males throughout their adult life (the conditions to which the flies had adapted for over 300 generations). By comparing the degree to which lifetime fecundity of females was reduced when males were continuously present, compared with minimal exposure to males, we were able to assess the resistance of different female genotypes to all forms of male-induced harm (i.e. persistent courtship, copulation attempts, the act of copulation itself, and the influences of seminal fluid proteins). Because the post-eclosion phase of the flies’ lifespan is no more than 2–5 days, survival of adults was virtually 100% (unpublished data), and variation among females for adult lifetime fitness was determined by their fecundity during the last 18 h of day 14 of their life cycle. Previous work in our laboratory has shown that there is no genetic correlation between juvenile fitness (egg to adult survival) and adult lifetime fecundity (Chippindale et al., 2001). As a consequence, our measure of a reduction in a female's lifetime fecundity due to interactions with males reflects a reduction in her lifetime fitness.

To accomplish a genome-wide screen for female resistance, we first used cytogenetic cloning to extract and amplify 35 random genomic haplotypes (equivalent to the genome of a gamete) from the LHM base population (see Chippindale et al., 2001 for protocols). These genomic haplotypes span 99% of the genome and include the X chromosome and both major autosomes, but they do not include the dot fourth chromosome. Copies of each genome-wide haplotype were placed in a group of diploid individuals, called a hemiclone, in a manner that caused each copy of the same haplotype to be expressed in a different random genetic background, as described previously (Chippindale et al., 2001). Members of a hemiclone all shared in common the same genomic haplotype, but the remainder of each diploid genome was random.

The only difference from the published protocols for the construction of a hemiclone is that, in the last step used to produce hemiclonal females, males carrying the target genomic haplotype were mated to females from a replica of the LHM base population that carried the autosomal recessive brown-eye marker (i.e. mated to females from the LHM-bw rather than LHM population). This change caused the hemiclonal females to retain a completely wild type phenotype (i.e. they were bw+/bw, so the recessive brown-eye phenotype was not expressed), but because of the heterozygous recessive brown-eye marker, we could detect when females remated with bw/bw males (as described below).

To test for genetic variation among hemiclones of females we first placed a group of virgin females from a single hemiclone (bw+/bw) with red-eyed males (bw+/bw+) from the base population for 90 min at a ratio of three males to one female, which provided the minimal exposure to males needed to ensure that virtually all females were inseminated (Rice, 1996). The red-eyed males used to mate the females were collected 11 days post-egg deposition and aged for 24 h in groups of 30 males/vial prior to being combined with females. The hemiclonal females were collected as virgins 10 days post-egg deposition and aged for 2 days (without live yeast) in two groups, each with 10 females/vial, prior to being combined with the males for initial mating. The timing of these collections caused males and females to be of ages typically experienced during the normal culture of the base population, and they were reared at the same density as the base population (150–200 fertilized eggs per culture vial).

After being inseminated, the two groups of 10 hemiclonal females were randomly assigned to one or the other of two adult competition environments: (i) ‘males present’ [16 brown-eyed (bw/bw) males per vial] or (ii) ‘males absent’. Both environments contained six brown-eyed competitor females, with which the 10 hemiclonal females (red-eyed) competed for a small amount of live yeast (10 mg) – these densities and food levels matched the normal culturing protocol of the base population (16 males and females per vial, but with the males absent in the minimal male exposure treatment). The brown-eyed competitor females were the same age as the red-eyed hemiclonal females, but they were derived from the LHM-bw replica of the LHM base population. After 48-h in the adult competition vials, the hemiclonal females were individually cultured in separate vials to oviposit for 18 h. The number of progeny produced per family (which constitutes lifetime fitness in the context of the laboratory culture protocol to which these flies were adapted) and whether or not the mother remated (presence of brown-eyed offspring) was then scored 12 days later. Placing each female in an individual vial (instead of in groups of 16/vial) during the oviposition stage of the assay was a necessary departure from the normal protocol used to propagate the base population. The change was needed to determine whether females had remated in the continuous male exposure treatment and was identical in the two experimental treatments, and therefore cannot account for any observed differences between them.

Statistical analysis

We used anova to test for genetic variation among the 35 hemiclones for resistance to male-induced harm to their lifetime fecundity. The statistical model was: lifetime fecundityijkl = μ + treatmenti + hemiclonej + treatment*hemicloneij + replicate[hemiclone]kj + errorijkl, where μ is the grand mean lifetime fecundity, treatmenti is the presence or absence of males during the adult competition phase of the life cycle, replicate[hemiclone]kj = replicate nested within hemiclone is the effect of the kth replicate on the fecundity of the jth hemiclone, treatment*hemicloneij is the treatment by hemiclone interaction, and errorijkl is the residual error, assumed to be normally distributed with mean zero and constant variance. Replicate is nested within hemiclone because all females from the same adult competition vial (in the same replicate) experienced the same, vial-specific environmental effects. Treatment was modelled as a fixed effect and all other factors were modelled as random effects. Lastly, the interaction term between treatment and hemiclone measured the variation among hemiclones for resistance to male-induced harm, i.e. it measured differences among hemiclones in the degree to which continuous exposure to males, compared with the minimal exposure needed for fertilization, reduced the lifetime fecundity of females. This represents a conservative estimate of female resistance because some harm by males may occur from the first mating. Total variation in lifetime fecundity of females was decomposed into its components (among hemiclones, replicates nested within hemiclones, interaction between hemiclone and treatment, and residual) using maximum likelihood methods as implement by the REML option in the JMP statistical software package (SAS Institute®, Cary, NC, USA). Normality and homoscedasticity of the error terms was analysed by distributional analysis of residual error terms.

Previous experiments in our laboratory have demonstrated that female fecundity varies considerably among hemiclones (Chippindale et al., 2001). As a consequence, a significant interaction between treatment and hemiclone in the above anova model would occur even when males influenced all hemiclones to the same proportional degree, i.e. if all females had their fecundity reduced by males by the same percentage. This occurs because the same percentage decline in a low fecundity hemiclone would be less, in the currency of offspring number, than that of a high fitness hemiclone. To eliminate this problem we also analysed the natural logarithm of lifetime fecundity of females. By analysing logarithms of fecundity, a significant interaction between hemiclone and treatment will occur only if some hemiclones have different proportional reductions in their fecundity due to interactions with males.

The assays were carried out in five sequential statistical blocks (replicates). For reasons that we could not fully control, average female fecundity varied among blocks. This variation in average fecundity caused the variance in female fecundity to vary among blocks. To adjust for this heterogeneity among blocks we transformed all fecundity measures by multiplying each value by the inverse of the ratio of average female fecundity in a block to the global average fecundity across all blocks. This transformation caused all blocks to have the same mean (i.e. the global mean across blocks prior to transformation) and it effectively removed among block heterogeneity of variance. It should be noted, however, that our conclusions regarding genetic variation among hemiclones for resistance to harm from males is completely robust to the specific transformation that we used. This is shown by the fact that when we also tested for the treatment*hemiclone interaction within each of the five independent blocks of the study separately (so that no transformation was needed to homogenize variances among blocks), we found the same statistically significant patterns to be present within every block.

Nondirect effects of males

The two experimental treatments were produced by keeping the density of females constant (16 per vial) while varying the density of males in the adult competition vials (no males in the minimal male exposure treatment and 16 males in the continuous male exposure treatment). The difference in male density between the two treatments can influence female lifetime fecundity directly, by varying the level of male–female interactions – which is the focus of our study – and in a nondirect way, by potentially reducing a limiting resource that controls female fecundity. In a previous study, we found that the resource limiting female fecundity is the small amount of live yeast (10 mg) that we provided on the surface of the killed yeast medium (see online supplementary materials).

As a control to test for the presence of males reducing female fitness in a nondirect manner (due to their consuming live yeast or fouling of the environment) fecundity of females was compared in an unrelated study (see online supplementary materials) when their yeasted adult competition vials were pretreated with an 8 h exposure to (i) 16 males, (ii) 16 females and (iii) no flies of either sex (control). By comparing the pretreatments with males vs. females to the control of no flies present during the pretreatment, we were able to estimate any nondirect harmful effects from males that was not due to persistent courtship, copulation attempts, copulation itself and the influences of seminal fluid. These controls indicated that the presence of males had no measurable nondirect influence on female fecundity (due to their eating or fouling of live yeast in the adult competition vials, see online supplementary materials). As a consequence, any change in the lifetime fecundity of females associated with the experimental treatments (minimal exposure to males vs. males continuously present) was due to direct effects, i.e. persistent courtship, copulation attempts, copulation itself and the influences of seminal fluid.


Females continuously exposed to males had an average reduction in their lifetime fecundity of 15.4% compared with females given only minimal exposure to males (Fig. 1). The 95% confidence interval of the estimate is 8.2–22.0. This result indicates that a substantial proportion of a female's lifetime fecundity is removed owing to antagonistic interactions with males, e.g. kicking, wing-flicking and running away (or decamping) from persistent male courtship, as well as the toxic effects of seminal fluid proteins when remating occurs.

Figure 1.

Average lifetime fecundity of females with minimal vs. continuous exposure to males. Error bars represent 1 standard error.

To determine the degree to which genotype contributes to a female's capacity to resist male-induced harm, we compared female lifetime fecundity of the 35 hemiclones from the two environments (minimal exposure to males vs. males continuously present) using the anova design described above. Significant additive genetic variation was found among hemiclones for lifetime fecundity (Table 1, P < 0.0001). Residual analysis confirmed that the error terms closely matched a normal distribution with homoscedasticity. The treatment*hemiclone interaction term in this model tested for variation among hemiclones in their resistance to harm from males, and it was found to be statistically significant (P < 0.0001). To confirm that this interaction was due to some hemiclones being harmed by a larger proportional reduction in their lifetime fecundity, we also analysed the natural logarithms of lifetime fecundity and found that the interaction term remained statistically significant (Table 1b, P < 0.001). In Fig. 2 we plot, on a log scale, the average lifetime fitness of the 35 hemiclones in environments with minimal and continuous exposure of females to males. Average female lifetime fecundity decreased for all 35 hemiclones in the environment with continuous male exposure, but the variation among slopes, as tested by the interaction term in the anova model (Table 1b), indicates that some hemiclones are harmed more (i.e. a larger proportional reduction of lifetime fecundity) by males than others.

Table 1. anova table for lifetime fecundity among the 35 random hemiclones. The factors hemiclone, replicate[hemiclone] and hemiclone*treatment are the random effects, and treatment is a fixed effect. (a) Lifetime fecundity; (b) natural log of lifetime fecundity.
 SSMS (num)d.f. (num)F-ratioP > F
  1. SS, sum of squares; MS mean square.

 Hemiclone* treatment9058.1266.415342.5563<0.0001
 Replicate [hemiclone]55700.2403.6251383.8729<0.0001
 Hemiclone* treatment9.180330.27001342.04940.0003
 Replicate [hemiclone]55.53730.402441383.0546<0.0001
Figure 2.

Interaction plot (log scale) of the 35 hemiclones depicting mean lifetime fecundity of females when males are continuously present vs. minimal exposure to males.

To estimate the proportion of heritable variation of lifetime fecundity that is attributable to resistance to harm from males, we decomposed the variation of female fecundity into its components (Table 2a). Hemiclone and its interaction with treatment (which measures resistance to harm from males) both contribute significantly to total variation for lifetime fecundity, i.e. the confidence intervals for both factors do not overlap zero. These two factors combine to estimate the genetic variation among hemiclones for fecundity. Collectively 13.60% (11.23 + 2.37%) of the total variation for lifetime fecundity was estimated to be genetic variation among hemiclones, and 17.42% (2.37/13.60%) of this genetic variation for fecundity was estimated to be due to differences among hemiclones in their resistance to harm from males. Using the log-transformed data, these estimates are 11.73% (10.1 + 1.72%) and 14.7% (1.72/11.73%) , respectively. In summary, genetic variation for resistance to male harm is a small but significant proportion of total phenotypic variation [i.e. the heritability of female resistance is about 2% and its estimated additive genetic coefficient of variation is 4.85% (1.5% for the log-transformed data)] and a substantial proportion [17.4% (14.7% for the log-transformed data)] of additive genetic variation for lifetime fecundity among hemiclones.

Table 2.  Decomposition of variance components for lifetime fecundity among the 35 random hemiclones. The factors hemiclone, replicate[hemiclone] and hemiclone*treatment are random effects and treatment is a fixed effect. (a) Lifetime fecundity of females; (b) natural log of lifetime fecundity of females.
 Variance componentSE95% Lower95% Upper% Total
 Hemiclone* treatment3.2831.3241.6998.8202.370
 Replicate [hemiclone]15.2542.55711.26221.83111.013
 Residual104.340   75.329
 Total138.513   100.00
 Hemiclone* treatment0.00280.00140.00130.00971.722
 Replicate [hemiclone]0.01390.00260.00990.02078.407
 Residual0.1318   79.860
 Total0.1651   100.00

Because seminal fluid contributes to harm from males to their mates (Chapman et al., 1995) we assayed for variation among hemiclones for remating rate and then correlated this variation with lifetime female fecundity. There was substantial heterogeneity among hemiclones for remate rate [Fig. 3b, anova, P < 0.0001, remate rate is the proportion of females that mated again (i.e. remated) at least once during the 48 h period the flies spent within the adult competition vials] with some hemiclones being nearly monogamous (i.e. most females did not remate) and other highly promiscuous (i.e. most females remated at least one time). Average remate rate was negatively correlated with the percentage change in lifetime fecundity (reduction was relative to the minimal male exposure treatment) among the 35 hemiclones when males were continuously present (Fig. 3a, Pearson's product–moment correlation =−0.479, P < 0.001). This correlation would be noncausative, however, if low fitness hemiclones were less vigorous and therefore more likely to remate in response to persistent male courtship. To test this noncausal interpretation, we correlated fecundity of females from the minimal-exposure treatment with remate rate from the males-continuously present treatment (Fig. 4b) and compared this with the same correlation when both fecundity and remate rate were estimated from the males-continuously present treatment. No significant correlation with average remate rate was observed when we analysed fecundities from the minimal-exposure treatment (Pearson's product–moment correlation = −0.077, n.s.) but a negative correlation was observed when we analysed fecundities from the males-continuously present treatment (Pearson's product–moment correlation = −0.374, P < 0.05), indicating that elevated remating rate is probably causally responsible for reducing lifetime fecundity of females.

Figure 3.

(a) The percentage change in female fitness [Δ = (average lifetime fecundity of females with continuous exposure to males − average lifetime fecundity with minimal exposure to males)/average lifetime fecundity of females with minimal exposure to males] becomes more negative when remating occurs more frequently. (b) The distribution of remate rate (proportion of hemiclonal females that mate more than once) among 35 genome-wide hemiclones.

Figure 4.

(a) Adult fitness of the 35 hemiclones declines as average remate rate increases in the treatment with continuous male exposure (b) but not with minimal male exposure.


There has been considerable recent interest in the process of sexually antagonistic coevolution, yet no study before this one has estimated the degree to which intersexual conflict contributes to total fitness variation for either sex. The finding that interactions with males reduced female adult fitness by 15.4% (95% CI: 8.2–22.0) indicates that males produce a substantial lag-load in females (i.e. a reduction in average population fitness owing to lack of counter-adaptation). This lag-load will select for females that have increased resistance to male-induced harm and provide fuel for an arms race between the sexes. It is important to point out that the observed lag-load in females will select for the evolution of female resistance even if females recouped some or all of the cost through indirect benefits, i.e. by producing offspring of higher genetic quality. In this case females would be selected to reduce the costs of interacting with males to the degree that costs and indirect benefits can be decoupled genetically.

In order for an arms race to ensue between the sexes there must be additive genetic variance for both male harm and female resistance to this harm. Past studies have documented genetic variation among males for their harm to female survival (Civetta & Clark, 2000; Sawby & Huges, 2001). This study documents low but extant standing genetic variation for female resistance in the currency of lifetime adult fitness. Past work in our laboratory has shown that adult fitness contributes about 85% of the genetic variation for total net fitness among females, with no genetic correlation between juvenile and adult fitness (Chippindale et al., 2001). We estimate that approximately 14% of total phenotypic variation among females for lifetime fecundity is due to is additive genetic variation among hemiclones and about 17% of this is due to additive genetic variation for resistance to harm from males. Although the heritability and genetic coefficient of variation of female resistance is relatively small (about 2 and 4%, respectively) this is to be expected because chronic selection for female resistance will continually erode any standing additive genetic variation for this trait. Collectively, these data indicate that there is substantial selection on females to resist male-induced harm with low but ample genetic variation among females to respond to this selection to fuel an arms race with males. The observed low heritability of female resistance would be expected to impede a response to selection, but not stop it in our large LHM base population. Because the standing genetic variation for resistance is a nontrivial proportion of its mean value (additive coefficient of variation for resistance is 4.9% among hemiclones, which would be approximately 10% among diploid individuals, see below), a gradual response to selection is nonetheless expected (Houle, 1992).

Additive fitness variation among hemiclones is a conservative estimate of fitness variation among diploid genomes because only half of the diploid genome of an individual is assayed. Nonetheless, this conservative aspect of genetic variation among hemiclones may be offset by the presence of nonadditive genetic variation due to epistatic interactions among nonallelic genes embedded in the same genomic haplotype. However, theoretical and empirical work suggest that any contribution of epistatic variance is likely to a small portion of total fitness variation (for discussion see Tachida & Cockerham, 1988).

The fact that female resistance is strongly negatively associated with female remate rate suggests that reluctance to remate is an important factor contributing to female resistance to harm from males. This association would not be causative if lower-fitness females were less vigorous and therefore less capable of warding off persistent male courtship. However, the fact that correlation between female fecundity and remate rate is absent in the ‘minimal-exposure’ treatment while present in the ‘continuous-exposure’ treatment supports the conclusion of causal relationship between enhanced remating and lowered adult fitness of females.

Overall this study documents strong selection for female resistance to harm from males and low but ample genetic variation to respond to this selection. This study indicates that the requisite conditions are met, at least on the female side, for an intersexual arms race to be in progress in our LHM base population. More generally, this study supports the general hypothesis that a perpetual arms race between the sexes may cause many reproductive traits, and the genes that code for them, to evolve at an accelerated rate.


We thank A. Stewart, E. Morrow, N. Orteiza, A. Kaplan, E. Cunningham and T. Lew for technical assistance. This work was supported by two grants (DEB-0128780 & DEB-0410112) from the US National Science Foundation to WRR and a summer fellowship from the UCSB College of Creative Studies to JEL.

Supplementary material

A PDF file with more information on this subject is available to download from the following website: