Mixed evidence for intralocus sexual conflict from male-limited selection in Drosophila melanogaster

Abstract

Sexual conflict over shared traits—intralocus sexual conflict (IaSC)—may be common and consequential, but experimental tests of its relative magnitude are challenging and limited in number. We use a sex-limited selection experiment, designed to subject haplotypes of Drosophila melanogaster to selection for male fitness without opposing selection acting on female fitness. Importantly, we use three novel base populations to compare results with those from the LH_M population, the sole population investigated using this technique. In contrast with previous studies, we find that male fitness of haplotypes subject to male-limited selection (ML populations) are not consistently better than their matched controls when tested in the “wildtype” state. Males from ML lines did not outperform controls in competitive fitness assays, mate choice trials, fecundity induction or sperm offense tests. As predicted, genetic variation for male fitness was reduced, with low fitness haplotypes apparently removed by selection, but this was only surveyed in one replicate population pair and included a potential artefact in the protocol. Female fitness was markedly reduced by carriage of ML haplotypes, as predicted by sexual antagonism. Hence, our results are only partially consistent with the IaSC hypothesis, raising questions about the relative contribution of sexual conflict to the standing genetic variation in these populations and the potential role of artefacts in the protocol that may have obscured our ability to detect IaSC.

Introduction

The evolution of anisogamy is hypothesized to have been a key driver in the origin of sex-specific selection pressures, resulting in different morphological, life-history, reproductive and ecological strategies in a wide range of taxa. To achieve sex-specific adaptation despite a shared genome, a number of genomic mechanisms, such as gene duplications, sex chromosomes, imprinting, and in cis-regulation have evolved in concert with hormones and other signaling systems. Despite these means to facilitate disruptive selection, many sexually-dimorphic traits show positive genetic correlations between the sexes (Poissant et al. 2010). These correlations are expected to act as constraints upon the evolution of further dimorphic responses to selection. This brake on the ability to evolve to separate fitness optima for shared traits results in a conflict between the sexes, often labelled intralocus sexual conflict (IaSC). Anecdotal evidence for IaSC includes the expression of ornaments and armaments in the sex that does not deploy them in sexual selection (e.g. horns in female soay sheep) suggesting the evolution of dimorphism is sometimes incomplete.

There is a longstanding theoretical interest in IaSC and the conditions under which polymorphism for sexually antagonistic alleles can be sustained by selection (e.g. Kidwell et al. 1977; Parker 1979; Lande 1980). For example, Kidwell et al. (1977) showed that the degree of dominance and its sex-specificity plays a key role in determining equilibrium behaviour for alleles with opposing fitness effects upon the sexes. Subsequently Rice (1984) recognized the potential role of sex chromosomes, which present distinctive features in XY and ZW systems: specifically, sex-limitation of the Y/W and haplodiploidy of the X/Z chromosomes. The sex-determining chromosomes (Y/W) have been theorized to attract male/female benefit SA loci and make them sex-limited in the non-recombining region, a central hypothesis for the origin and evolutionary dynamics of these chromosomes (Charlesworth and Charlesworth 2000). Assuming identical allelic dominance parameters in males and females, Rice (1984) predicted an enrichment of SA variation on the haplo-diploid X/Z chromosome. The logic of Rice’s model is that dominant female-benefit alleles can accumulate to appreciable frequencies because of the 2-fold advantage females have in X-expression in populations. On the other hand, recessive male-benefit alleles can reach high frequencies because their deleterious effects upon females are masked in the heterozygous state. However, contrasting theoretical predictions about the SA loci enrichment on the haplodiploid sex chromosome result from differing assumptions about allele dominances (Fry 2010).

Unlike the theory, the empirical investigation of IaSC is a nascent area of investigation. Early experimental evidence for the existence of IaSC was provided by Chippindale et al. (2001) in the form of a negative genetic correlation for reproductive fitness between the two sexes in Drosophila. This measure of intersexual (or cross-sex) genetic correlation for fitness (r_w,g,m-f) has since been used to investigate the existence of IaSC, with negative measures of r_w,g,m-f treated as evidence (cf. Connallon and Matthews 2019). The intersexual genetic correlation for fitness (or components thereof) has been explored using sex-specific parent-offspring data in natural and semi-natural populations, as well as in controlled breeding designs in the lab (reviewed in Bonduriansky and Chenoweth 2009; Poissant et al. 2010; Connallon and Matthews 2019). Besides measurements of intersexual genetic correlations for fitness at the population level, a second key focus of this literature has been on specific traits that are potential fulcrums of sexual antagonism. Through measurements of extant trait values and fitness optima (estimated using selection gradients), a number of dimorphic traits such as body size and development time (among others) have been shown to have SA fitness effects. Numerous taxa appear to exhibit IaSC, including species of fruit flies, seed beetles, birds, ground crickets, meal moths, plants, lizards, snakes, voles, and red deer (reviewed in Bonduriansky and Chenoweth 2009; Cox and Calsbeek 2009; Singh and Punzalan 2018).

Empirical research has so far suggested that the D. melanogaster X chromosome does indeed harbor more SA loci than randomly expected (Gibson et al. 2002; Pischedda and Chippindale 2006; Long et al. 2012; Ruzicka et al. 2019; Wong and Holman 2023, but see Lund-Hansen et al. 2020; Abbott et al. 2020). However, Ruzicka and Connallon (2020) suggest that quantitative genetic attempts at identifying SA loci are likely to overestimate the enrichment of SA loci on the X chromosome, due to the asymmetric inheritance pattern of X chromosomes. Further, recently unearthed evidence for sex-specific dominance reversals (Grieshop and Arnqvist 2018; Pearse et al. 2019; Geeta Arun et al. 2021, reviewed in Grieshop et al. 2024) contributes weight to Kidwell et al.’s assumptions, and predictions of random distribution of SA loci across autosomes and sex chromosomes. It is fair to say that generalizations about the genomic distribution of sexually antagonistic loci cannot be made at this point in time, and that continued investigation is warranted.

As a form of balancing selection acting upon the genome constitutively, IaSC is expected to maintain polymorphisms and contribute to genetic variance for traits related to fitness (Prout 2000). However, while the conditions for sexually antagonistic selection are believed widespread, the conditions for the maintenance of genetic variation may or may not be met in nature. Rapid changes in the environment can cause maladaptation, which can align the selection pressures acting on the two sexes, but gradual environmental fluctuations are likely to stabilize antagonistic variation (Connallon and Hall 2016). While early models typically assume single-locus traits, Flintham et al. (2025) further complicate this picture by explicitly considering continuous traits governed by multiple loci. Here, they find that such variation due to sexual antagonism should arise only rarely and often be transient. There is little empirical work directly supporting the maintenance of genetic variance through SA polymorphisms. Ruzicka et al. (2019) show that the SA loci that they identify in a lab population display signatures of balancing selection (measured using Tajima’s D and minor allele frequency measurements) and identify similar signatures in distantly related populations of fruit flies, suggesting stable balancing selection over long periods of evolutionary time at these loci. More recently, Kaufmann et al. (2023) show that artificial sexually antagonistic selection on body size maintains greater genetic variance than sex-limited selection.

In this study, we look to a sex-limited selection protocol (developed by Rice 1996 and pioneered as a tool to study IaSC by Prasad et al. 2007) to address these questions. Using the lack of molecular recombination in D. melanogaster males and an independent population of “clone-generator” (CG) females carrying compound-X chromosomes (X-XY karyotype) and autosomal translocations, we are able to reliably passage haplotypes containing an X chromosome and all major autosomes as “hemiclones” from father to son. Starting with outbred stocks, we used the CG females to capture many (over 1,000) random hemiclones and transmit them to sons, which competed for reproductive success with a new generation of CG females drawn from a separate population. This protocol allowed male–male competition and differential transmission of hemiclones to the next generation via sperm, while removing selection acting upon female function. Sex chromosomes were assumed to be unaffected by reversed-sex inheritance: X chromosomes by paternal transmission via sperm and Y chromosomes by maternal carriage and inheritance via the egg. We predicted that these populations would be released from IaSC, increasing the frequency of male benefit alleles at polymorphic SA loci and thus improving male fitness while lowering the fitness of their daughters, when occasionally produced for assay purposes.

This male-limited (ML) system holds potential advantages over other contemporary systems used in the literature. While pedigree designs allow us to study the standing genetic variance and intersexual correlations for fitness in a population, they tend to require large numbers of sampled lines. To probe the system past fitness and identify SA polymorphisms on the genome, there is a need for complex statistical manipulations, such as the extraction of an “antagonism index” (Innocenti and Morrow 2010; Hill et al. 2017; Ruzicka et al. 2019). Further, it is not amenable to examination of other phenotypic trait variances and corresponding analyses. In contrast, the ML evolution approach 1) provides a direct experimental manipulation rather than a correlational analysis, 2) potentially allows for statistical simplicity and strength at smaller sample sizes through the amalgamation of male-benefit alleles across many loci in the same individuals, increasing signal-noise ratio and 3) allows us to probe adaptive responses in the phenotype when released from divergent sex specific selection pressures.

An unfortunate consequence of a tight knit research community is that most experimental studies of sexual conflict in Drosophila have been conducted with a single stock population: the LH_M (LH-moderate density) population collected in central California in 1988 by L. Harshman (Rice 1996, 1998; Chippindale et al. 2001; Gibson et al. 2002; Byrne and Rice 2005; Lew and Rice 2005; Lew et al. 2006; Linder and Rice 2005; Pischedda and Chippindale 2006; Prasad et al. 2007; Long and Rice 2007; Bedhomme et al. 2008; Rode and Morrow 2009; Innocenti and Morrow 2010; Abbott et al. 2013; Jiang et al. 2011; Collet et al. 2016; Hill et al. 2017; Ruzicka et al. 2019; Abbott et al. 2020; Lund-Hansen et al. 2020; Manat et al. 2021; Lund-Hansen et al. 2022). While results from Ruzicka et al. (2019) suggest some generalizability from LH_M to the species at large, it is important to affirm this notion by conducting experimental work on different populations. Interestingly, even with LH_M, Jiang et al. (2011) found markedly different results from Rice (1996) for male postcopulatory sexual selection, and Collet et al. (2016) demonstrate that even recently separated populations (~200 generations) of LH_M display differences in sexual antagonism. These disparities may highlight sensitivity of the standing genetic variation to the conditions or duration of maintenance in the laboratory of the very same population (also see Berger et al. 2014, 2016), and call for the investigation of other populations to establish generalities about the nature and level of sexual conflict within the species.

As an example of how trade-offs may be affected by domestication, Service and Rose (1985) proposed that lab maintenance can initially generate positive genetic covariances between fitness components that typically trade off in nature. They argued that under benign conditions, a subset of fortuitously pre-adapted genotypes may rise to fixation, obscuring negative trait correlations. As selective equilibrium is approached, these may re-emerge, revealing latent trade-offs previously masked. Consistent with this view, positive intersexual genetic covariances tend to be observed in recently domesticated populations in crickets (Zajitschek et al. 2007), stickleback fish (Kim and Velando 2016), and fruit flies (Nguyen and Moehring 2017); whereas evidence for negative intersex covariances for fitness is consistently from evolutionarily equilibrated populations—in fruit flies (Chippindale et al. 2001), red flour beetles (Pai and Yan 2002), ground crickets (Fedorka and Mousseau 2004), moths (Lewis et al. 2011), and seed beetles (Berger et al. 2014).

Motivated by 1) a deliberate exclusion of LH_M population to diversify population sampling and 2) the need for a long-domesticated population, we conducted our ML selection experiment on three replicate populations from a different geographic origin (Massachusetts, USA), separated from one another since 1980. We report here on the competitive reproductive fitness (here referred to as CRF) of males and females from selected and control lines. We apply a new breeding design in which we strip test individuals of all artefacts deriving from the selection experiment to assess CRF, and we further examine test males for mating success, fecundity induction, sex ratio drive and sperm offense. Lastly, we use a hemiclonal analysis to study genetic variance for reproductive fitness in males and females from selected and control populations, and use this design to report on the intersex genetic correlation for fitness.

We predicted 1) increased male fitness in selected populations, at the cost of female fitness, 2) masculinization of partially dimorphic traits in both males and females (as seen in Prasad et al. 2007; Abbott et al. 2013), and 3) a reduction of genetic variance for fitness, particularly in males. Previous work suggested that these ML males would be more attractive (Bedhomme et al. 2008), but conflicting results exist for post-copulatory sexual selection. Because sperm and seminal fluid characteristics are presumed to be already male limited traits (Friberg et al. 2005; Bjork et al. 2007; Jiang et al. 2011), we anticipate that the improvement of male fitness will not evolve through improved fecundity induction or sperm competitive ability, other than as a byproduct of increased general vigor (although see Rice 1996, 1998). Previous work reports no precedent of driving X chromosomes, but the selection treatment should strongly select for even weakly driving X chromosomes, should they exist. We also test ML-evolved lines with and without the evolved X chromosome to test the prior claims of it being a “hotspot” for sexually antagonistic loci (e.g. Gibson et al. 2002).

ML selection

Source populations and general maintenance

Our experiments used lines originating from the IV base population, derived from the wild in 1975 in South Amherst (MA, USA). Five populations derived from IV were selected for late reproduction beginning in 1980 (Rose 1984), and our source “CO” populations descend from those postponed aging lines, kept since 1989 on a 4-week discrete generation cycle (~500 generations) in population cages before being adapted for the current experiments. Three CO populations (CO₁, CO_3, CO₅) were arbitrarily chosen from the five replicate CO’s and adapted to shorter generations to facilitate our selection experiment: a 2-week cycle (day 1 to 11 in vials, followed by days 12 to 14 in four litre population cages). We chose these populations because 1) they are laboratory adapted, 2) they are known to harbor genetic variation for life history characters, and 3) other selection experiments on sexual conflict predominantly come from a population (LH_m) kept entirely in vials, potentially limiting the scope of behavioral interactions; the introduction of a cage phase is a departure from previous ML-evolution protocols, potentially increasing environmental complexity. These populations, descended from a common ancestor and kept in parallel for hundreds of generations, offer some control over the effects of drift and divergence under domestication. All populations were maintained under standard laboratory conditions: moderate density (larval density of 80 to 100 larvae/vial), 25°C, 12:12 light:dark cycle on banana/agar/killed-yeast medium, and with a census population size of ~ 1000 haploid genomes per generation.

Selection treatment

After 24 generations of 2-week propagation, an ML line and matched-control (MC) line were derived from each of the three replicate populations described above. These were correspondingly labelled ML₁-MC₁, ML₃-MC_3, and ML₅-MC₅. ML selection lines were established nearly identically to the protocol described in Prasad et al. (2007) (Fig. S1a). Briefly, 1,000 males from each replicate base population were mated to 1,000 CG females that carried a compound X(C(1)DX, y, f), a Y chromosome and a homozygous-viable translocation of the two major autosomes (T(2: 3)rdgc st in ri p^p). Using the genetic markers for eye color, the absence of molecular recombination in males, and the aneuploid mortality experienced by animals inheriting uneven combinations of translocated autosomes, we selected 1,050 sons carrying a haploid genome (or “hemiclone”) from sires each generation on day 10 post oviposition (PO) under light CO₂ anesthesia. These males were combined on day 12 PO with virgin CG females of matched age in 3 cages (333 to 334 males and females in each 4 L plastic cage) to allow hemiclones to compete for mates. Eggs were collected from these cages on day 14 at a density of 320 to 400 eggs per vial, to achieve a realized larval density of 80 to 100 per vial after accounting for aneuploid viability costs (3/4 of the eggs were inviable due to the CG genetics).

To match these conditions in controls, 500 males (1,000 haploid genomes) from each MC line were collected under light CO₂ anesthesia each generation on day 10 PO and combined on day 12 with virgin MC females in 2 cages (250 pairs per cage). Eggs were collected from these lines at a density of 80 to 100 per vial, with 20 vials dedicated to collection of virgin females in the next generation and 20 to male collections. MC males thus experienced the same sociosexual conditions and handling procedures as the ML males did.

Recombination box

In the absence of molecular recombination in males, and the artificial prevention of independent assortment of chromosomes, the ML selection regime is not only sex-limited, but also clonal in propagation, potentially exposing it to higher levels of genetic hitchhiking. We followed the system described in Prasad et al. (2007) and cycled a subset (5%) of the ML hemiclones through a two-generation cross, which moved ML-evolved chromosomes into females for a generation, allowing recombination to occur, before sending their sons back into the selection treatment. The specific cytogenetics of these two crosses are described in Rice (1996) and Prasad et al. (2007), but, briefly, in the first generation, 50 ML males from each ML population were crossed to virgin CG females carrying a dominant eye color marker (bw^D). Brown-eyed sons (carrying the ML haploid genomes) were crossed to control (MC) females, producing hybrid ML/MC daughters where recombination could occur. Fifty wildtype sons of these females were added into the ML cages each generation (Fig. S1b).

Each generation after the first selection generation, instead of control females, females from the second cross (RC2) were collected as virgins and retained as dams. Thus, although ML selection was suspended in this cross, the repeated influx of ML-selected chromosomes reduced this setback. For example, after 4 generations of backcrossing, these females are expected to be ~ 95% ML in genome content, increasing incrementally each generation. This crossing scheme was meant to limit genetic drift but also provided a means by which to derive ML-evolved genomic haplotypes for experimentation, as noted below.

Experimental animals for fitness assays

We sought to isolate “innate” changes in sex-specific fitness from adaptation specifically to the unusual genetics of the ML selection regime. Specifically, the reversed inheritance of sex chromosomes, the foreign cytoplasm, the maternally inherited Y chromosome, and the translocated autosomes present potential artefacts in the ML selection treatment. We therefore used males and females produced from the second recombination cross (RC2) in the ML treatment described above. As noted, after many generations of backcrossing ML genetic content into the “recombination box”, RC2-derived flies were expected to carry selected chromosomes with normal inheritance and no genetic markers. Thus to create the actual test animals, we used a crossing design between RC2 and MC flies.

From a full-factorial crossing design, 4 types of male offspring can be obtained (Fig. S2a). First, crosses with MC sires and RC2 dams produce males carrying a single complete haploid genome from the ML treatment (ML “single dose”); ML_SD males were the wildtype equivalent of the ML selection treatment and were hypothesized to be better competitors than controls. Crosses between RC2 sires and MC dams produce males carrying a control X chromosome and a single set of ML autosomes; these males were labelled ML_SD(a) (single dose autosomes) and used to help deduce the role of the autosome versus X-linked adaptation by the ML males. From crossing RC2 males and females, we can also derive flies carrying completely ML type genomes—i.e. homozygous for ML autosomes—labelled ML_DD (ML “double dose”) for which we had no clear predictions. Fourth, MC x MC crosses yielded control males and females. Finally, both directions of the MC x ML hybrid cross produced females carrying a complete ML haploid genome heterozygous with control for all three major chromosomes.

Experimental animals for other assays

While CRF assays were conducted using the fully factorial design described above, some of the experiments reported below predated that design and were handled slightly differently; these differences are denoted using the variable “Treatment” in analyses.

Both experimental and control animals bred for mate choice, sperm competition, body size and development time assays were produced using a cross analogous to the second recombination cross used for generating RC2 males in the selection protocol (Fig. S2b). First, ML and MC male haplotypes were “captured” using the dominantly-marked (bw^D) CG females. Sons from these crosses carrying the MC or ML haplotypes were crossed to virgin MC or RC2 females, respectively, to produce test flies. Here too, as before, the treatments are labelled as per the system above (MC, ML_DD, ML_SD, ML_SD(a)). While mostly identical to the design used for fitness assays, the male test animals from this design carry Y chromosomes from clone generator lines that may have been exposed to feminization or mutation accumulation effects from maternal carriage.

Hemiclone lines

To investigate genetic variance for fitness in these populations, we created hemiclonal lines (as described in Chippindale et al. 2001) using single haploid ML-1 and MC-1 genomes randomly sampled through crosses with CG females. Each individual haplotype was amplified in numbers in sons using crosses with CG (bw^D) females to create 40 hemiclone lines for each population. At the same time, we isolated the X chromosome from each hemiclone and amplified its numbers using lines carrying a balancer (recombination-suppressing) X chromosome and MC-1 autosomes; this allowed us to transmit the X from the dam in a cross to daughters (Fig. S3). These sons and daughters respectively were then crossed to one another to produce test animals carrying ML and MC haplotypes, along with control derived autosomes. Females derived were “homozygous” for the X chromosome, having two Xs from the same line, and males carried the Y chromosome from the CG (bw^D), much like the test animals from the mate choice and sperm competition assays.

General handling and competitor lines

Across all assays, test animals were reared at a larval density of 80 to 100 animals per vial. Virgin flies were collected 6 to 8 h from eclosion using light anesthesia between days 8 and 10 PO. Non-virgin test animals were collected on days 10 to 11 PO.

CRF, mate choice and sperm competition assays required the use of phenotypically marked flies to identify successful mates and measure competitive siring/damming success by distinguishing the progeny phenotype. To create these competitor populations, we backcrossed recessive eye color markers known to be relatively benign into control-derived lines for at least five generations. Two such populations were created, both by combining flies from the three replicate populations of the MCs. In the first, we introduced pink-peach (p^p) and, in the second, the brown (bw¹) eye color marker. Collectively, we refer to these recessive competitor lines as Control _recessive (“Cr”).

Assays

CRF assays

CRF for was assayed for both males and females in competition with the recessive-marked Cr stocks in equal numbers. Test females were collected as virgins and test males were collected on day 10 to 11 PO. On day 12, 30 MC or ML flies were combined with 30 Cr individuals of the same sex, and 50 Cr of the opposite sex. These flies were held together in ‘mini cages’ (ventilated 6 oz bottles) with ad libitum yeast between days 12 and 14 PO, to mimic the handling of the selection treatments. For each population, we assessed 6 such competition arenas (labelled contests in analysis) by collecting 6 vials of eggs and counting offspring proportions to estimate CRF.

Male CRF was assayed thrice, in generations 50, 64, and 70 of the selection experiment. In generations 50 and 64, only ML_SD and MC animals were tested. All four genetic configurations were tested in generation 70. Whereas previously we had waited for most or all of the flies to eclose, with males potentially mating before collection, for the experiment conducted in the 70th generation, males were collected as virgins. Female CRF was assayed twice, in generations 48 and 50 of the selection experiment. On both occasions, we only tested ML_SD and MC animals.

Hemiclonal analysis of fitness

Conducted in generation 78, CRF assays for hemiclonal analysis were designed very similarly to the CRF assays described above. The competition size was reduced in response to the increase in number of lines (76) being studied. Here, we combined 4 test males with 4 Cr males and 6 Cr females. For female fitness, 4 virgin test animals were combined with 3 Cr males. Rather than CRF, female productivity was assayed as a proxy for fitness. In both sexes, up to 5 such contests were set up for each line, depending on the availability of usable test animals. Complete hemiclonal data for both sexes was collected from 76 lines out of an initially sampled 80.

Mating success assays

We assayed female mate choice (or male mating success) in generation 66. Following the crossing design, test males were collected on day 10 to 11 post-eclosion. On day 12 PO, single test males were combined with a Cr male, and a Cr virgin female. When one of the two possible pairs began copulation, the excess male was aspirated out by gently removing the foam plug without disturbing the pair in copula. After the mating ended and animals separated, the remaining male was also removed, and the female was left undisturbed to oviposit for 48 h under ad-libitum yeast conditions. The time of introduction (observation start time), amplexus formation (mating start time) and separation (mating end time) were noted, which provided us with measures of mating latency and mating duration. The offspring sired in each vial were sexed and enumerated to test for sex ratio drive effects and male fecundity induction, respectively. The eye color of the offspring in each vial was noted to identify the sire that successfully mated. For each treatment (within each replicate pair of lines), we assessed 50 such binary choice vials.

Sperm offense assays

Sperm offense (or P2) was assayed (in generation 74) by mating Cr females first to Cr males and subsequently exposing them to males from either the MC or ML populations. On day 12, 12 virgin Cr females with ad-libitum yeast nutrition were combined in vials with Cr males in vials. Matings were observed carefully and almost all females were observed to be in amplexus synchronously and quickly. In all vials, at least 10 females were directly observed to have been mating synchronously. An hour after these matings ended, the flies were gently anaesthetized with CO₂, and males were discarded. Using a control treatment where no second male was provided, we found that this system resulted in about 97% of females being mated and fertile from this first mating round. Although unlikely, especially as recently mated females tend to be refractory to mating attempts and few potential secondary matings were observed, the group setup did not completely preclude double-mating; however this was consistent across all the treatments because Cr females were randomly assigned. On day 13, 10 ML or MC “target” males were combined with females. Second matings were observed carefully to ensure that no single female was multiply mated at this stage. To achieve this, we observed all the vials to note when matings began and continued observing the vials for a period of ~ 3 h after this point to note the number of matings seen, at the end of which the males were discarded to ensure that the first mated female did not return to a receptive state. We then randomly selected 10 females and transferred them to individual test tubes with ad libitum yeast to oviposit for 24 h, after which they were discarded. Due to the arbitrary cut-off in the second mating period, dictated by the first pair to copulate in each vial, we did not compare the number of matings achieved by target males in each treatment, instead assessing the proportion of offspring sired where a second mating has successfully taken place. Offspring from each tube were phenotyped by eye color and enumerated. For each treatment (within each replicate pair of lines), we assessed 30 such competition vials.

Statistical analysis

Statistical analysis was conducted in R version 4.2.3—“Shortstop Beagle” (R Core Team 2025). Data were visually assessed for residual normality and using Bartlett’s test for residual heterogeneity. Where generalized linear models were used, residual dispersion was tested using the DHARMa package (Hartig 2022).

Linear mixed models were employed to analyse CRF. Treatment and replicate population were used as fixed factors in the analysis to explain the proportion of offspring sired/dammed by the target individuals. To avoid pseudo-replication, contest was included as a random factor. Where significant differences were identified, we employed planned contrasts between the treatments in a single replicate population pair with a corrected alpha value.

In the hemiclonal analysis of fitness, we analysed genetic variance for fitness in each selection regime, for each sex, using a random effects model that explained the proportion of offspring sired/dammed by the target individuals, using the hemiclonal line as a random factor. The proportion of variance explained by the random factor line was then compared between selection treatments using an F test. Fitness data from the hemiclonal analysis was also analysed using a mixed model, where selection was used as a fixed factor. Here, we did not include replicate as in the case of CRF, as the study was conducted only on replicate population pair 1. Here too, line was included as a random factor. Additionally, mean estimates of male fitness and female fitness for each line from each selection regime were analysed using a Pearson correlation test to test if the ML population displayed a different intersex genetic correlation for fitness from the control. The correlation coefficients were compared against one another using a Fisher’s Z transformation.

To study the mating success of target males in female mate choice trials, we used a generalized linear model with a binomial distribution. The proportion of mating successes for target animals was explained using selection and replicate as fixed factors. Likewise, log (mating latency), mating duration, fecundity induction (offspring count), and sex drive (offspring sex ratio) were analysed using a Gaussian distribution for tubes, with selection and replicate as fixed factors.

For sperm offense, the extreme left skewed distribution of P2 estimates created a substantial challenge in appropriately transforming the data to fit a model. Instead, we chose to analyse the proportion of offspring sired by the second male in two steps. First, to create a data transformation that enabled the analysis, we subtracted the proportion of offspring sire from 1. This inverted measure was used to separate the dataset in two parts—values of zero (where all the offspring are sired by the target male) and non-zero values. The number of zeroes attributable to each treatment was analysed as a generalized linear model with a binomial distribution, using treatment and replicate as fixed factors. Non-zero values were log transformed and modelled independently using a linear model with the same combination of fixed factors.

Results

CRF

Male CRF

In generations 50 and 64, we found no effect of treatment, replicate or their interaction on male CRF for the ML_SD treatment, which is most representative of the selection treatment (TableA1a). In generation 50, ML_SD displayed a CRF of 0.34 +/− 0.01 (mean +/− 1.96*se) compared to 0.32 +/− 0.01 for MC (Fig.  1a). In generation 64, these figures were 0.34 +/− 0.02 and 0.32 +/− 0.02, respectively (Fig.  1b).

Fig. 1.

The proportion of offspring (CRF) of target MC (red) & ML_SD (blue) males in a) generation 50, and b) 64; target females in c) generation 48, and d) 50. ML lines were tested as whole (X, II, III) haplotypes in a single copy.

In generation 70, we found a significant effect of treatment on reproductive fitness (P < 0.001), and no effect of replicate or their interaction term (although the interaction term approached the α = 0.05 threshold) (Table S2). Excluding replicate (and the corresponding interaction term), we conducted pairwise contrasts between all 4 treatments to find that the only significantly different pairs were ML_SD > ML_DD (Tukey adjusted P = 0.006, Table S3) and MC > ML_DD (Tukey adjusted P = 0.01). CRF for each treatment (Fig.  2.2) was as follows: MC (0.45+/−0.02), ML_SD (0.47+/−0.02), ML_SD(a) (0.42+/−0.02), ML_DD (0.37+/−0.02).

Fig. 2.

The proportion of offspring sired by target males in generation 70. ML lines were tested as whole (X, II, III) haplotypes in a single copy (ML_SD—blue), with just the autosomes in single copy (ML_SD(a)—purple) or with all autosomes from the ML populations, a “double-dose” (ML_DD—green); see text for details.

Female CRF

In generation 48 we found a significant effect of selection treatment (P < 0.0001) on CRF, and no effect of replicate or their interaction term on CRF (Table S9). ML_SD females displayed a CRF of 0.53+/−0.01, and MC females 0.63+/−0.01 (Fig.  1c).

In generation 50, we found a significant effect of selection treatment (P < 0.0001), replicate (P < 0.0001) and the interaction term (P < 0.001) on CRF (Table S9). Here too, ML_SD females displayed reduced CRF compared to MC animals. Contrasts between treatments within each replicate revealed that the difference between ML_SD and MC females of replicate 5 was not significant (P = 0.074, Table S10) at the adjusted α = 0.017 threshold for 3 comparisons. Here too, the ML_SD mean was found to be lower than that of MC. Overall, ML females displayed a CRF of 0.51+/−0.02, and MC females 0.63+/−0.01 (Fig.  1d).

Hemiclonal analysis of fitness

Male fitness

We quantified genetic variance for fitness for both ML (0.0509) and MC (0.1771) males as the proportion of variance attributed to “line” in a random effects model. Using an F(37, 37) test, we found that these variances were significantly different from one another (P = 0.0001) (Table S4). Male fitness analysed using selection as a fixed factor and line as a random factor revealed an effect of selection (P < 0.0001) (Table S6). This model did violate the Bartlett’s test for homogeneity of variances, but as shown above, this is likely due to the reduction of heritable variance for fitness in the MLs. Overall, ML males displayed a CRF of 0.55 +/− 0.03, compared to MC males 0.45 +/− 0.04 (Fig.  3).

Fig. 3.

Heritable variance for male CRF. Histograms represent actual measurements from hemiclonal lines from ML (blue) & MC (red) treatments. Mean, variance for each treatment is depicted using dotted curves. Solid distributions depict heritable variance.

Female productivity

Likewise, we quantified genetic variance for productivity in ML (0.467) and MC (0.424) females. These were not significantly different from one another in a F(37,37) test (P = 0.39) (Table S5). Female productivity analysed using selection as a fixed factor and line as a random factor revealed no effect of selection (P = 0.29) (Table S6). ML dam productivity was 17.69 +/− 2.05 offspring and for MCs it was 16.09 +/− 1.92 offspring.

Intersexual genetic correlation for fitness

Intersex genetic correlations for fitness for both the ML (ρ = 0.0247) and MC (ρ = 0.11) population were not significantly different from 0 (Fig.  4, Table S7). Likewise, they were not significantly different from one another (P = 0.34) (Table S8).

Fig. 4.

Intersex genetic correlations plotted between female productivity and male CRF for each hemiclonal line from ML (blue) and MC (red) treatments. A type II linear regression is plotted for each treatment. Shaded region depicts permutationally estimated 95% confidence interval.

Mate choice assays

Mate choice

We find no effect of selection, replicate or their interaction upon the male chosen for mating, mating latency, or mating duration (Tables S11, S12 and S13). Across all replicates however, we note a consistent direction of difference in mate choice success. ML males were more successful, with greater differences in replicates 1 (61.7% success for ML vs 50.0% MC) and 3 (56.5% vs 46.5%), but also in 5 (44.7% vs 40.0%) (Fig.  5a). It is possible that there exists an effect the study was not sufficiently powered to capture.

Fig. 5.

a) The proportion of target MLSD (blue) or MC (red) males gaining mating with females relative to a marked competitor line and b) post-copulatory reproductive success as second of two mates (i.e. P2 or sperm offence). ML lines were tested as whole (X, II, III) haplotypes in a single copy (ML_SD—blue), with just the autosomes in single copy (ML_SD(a)—purple) or with all autosomes from the ML populations, a “double-dose” (ML_DD—green); see text for details.

MLs on average showed a mating latency of 13.96 +/− 4.14 min and a mating duration of 18.43 +/− 1.00 min. For the MCs, these figures were 14.93 +/− 4.33 min and 17.95 +/− 1.48 min, respectively (Fig.  6a, b). The model analysing mating duration deviated from homogeneity of variances as tested by Bartlett’s test, suggesting a potentially flawed analysis. However, from data visualization it appears that there are no real differences in this dataset, as suggested by the model.

Fig. 6.

Measures of a) latency to mating (log transformed), b) duration of matings, c) brood sex ratios, and d) fecundity induction for target ML_SD (blue) and MC (red) males.

Fecundity initiation, sex ratio drivers

We find no effect of selection, replicate or their interaction on fecundity induction, or sex ratio drivers (Tables S14, S15). Sex ratio for all the populations hovered around the expected 0.5 mark. ML male offspring were at 0.51 +/− 0.02 and for MCs it was 0.50 +/− 0.03 (Fig.  6c). ML males induced a fecundity of 51.09 +/− 2.94 offspring in partners, while MC males achieved 50.94 +/− 3.21 (Fig.  6d).

Sperm offense

We find no effect of selection treatment, replicate or their interaction on either of the sequential steps of analysis on the P2 component of sperm competition (Tables S16, S17). Proportion of offspring sired by each treatment as a second male was as follows: MC (0.90+/−0.03), ML_DD (0.85+/−0.03), ML_SD(a) (0.89+/−0.03), and ML_SD (0.90+/−0.02) (Fig.  5b).

Discussion

Sex-limited evolution has been a powerful tool in the study of sexual conflict in D. melanogaster but, to our knowledge, has only been used with a single base population—the LH_M population originating in central California. Our study, employing populations from a different geographic origin (Massachusetts, USA), fails to find strong evidence for a key proof of principle prediction of this system: the generic improvement of male fitness in response to the ML selection treatment. We find only inconsistent and potentially conditional increases in male CRF, and no measurable improvements in several of its components. On the other hand, through a hemiclonal analysis of genetic variance in one population pair, we found a significant fitness advantage to the ML-selected population, coupled with a marked reduction in heritable variation for male fitness in the selected population relative to its paired control. Moreover, also consistent with IaSC, we found that the fitness of females carrying ML-evolved haplotypes had declined sharply relative to controls, as predicted by selection upon sexually antagonistic variation. Our results compel us to consider that selection acted as predicted, but within the complex genetic milieu of the selection treatment employed. This, along with the evolutionary history of the populations used in the selection experiment, may have contributed to the mixed findings reported herein.

Previous applications of ML selection have documented fitness gains over short evolutionary time-frames, suggesting ample standing genetic variation in the LH_M population. Rice (1996) primarily attributed increased male fitness to the adaptation of males to the non-evolving “clone generator” females used to create sex-limited gene transmission. The females, drawn each generation from a separate stock, could not evolve in response to their mates and potentially acted as a “non-responding target”. Hence males that could gain matings with these specific females, or favorably influence their paternity after mating, would be selected. In a subsequent paper (Rice 1998) the potential role of IaSC in the response was recognized, with females expressing ML-evolved hemiclones showing longer development time, which was interpreted as a correlate of reduced fitness. Later work (Prasad et al. 2007) as well as our own recent (in prep) results also suggest that development time can evolve in the direction of extant sexual dimorphism under ML-selection, with both sexes slowing. While Prasad et al. (2007) showed evidence for improvement in male fitness from the application of the ML selection regime and unambiguous declines in female fitness, male fitness increases were not tested in the absence of “target” CG females. These results highlight the potential difficulties of discriminating between IaSC and fitness gains through specialization representative of interlocus sexual conflict.

To convincingly demonstrate IaSC with a sex-limited selection experiment, one must show generic improvements in fitness of the selected sex: improvements not derived from adaptation to specific features of the selection protocol. However, when we removed the potential influence of the genetic aberrations and other artefacts—testing the ML males with complete haploid genotypes in a wildtype, heterozygous state—our CRF assays turned up only small and nonsignificant advantages over controls (MC populations). We called this genetic configuration ML_SD (for “single dose”). Males carrying a single-dose of ML autosomes in the absence of the ostensible “hotspot” X (ML_SD(a)) were suggestively lower than ML_SD males in fitness in two of three replicates, but overall this difference did not stand up to statistical scrutiny. Indeed, the only difference that did show significance was that ML males carrying a “double dose” of ML autosomes (ML_DD) performed more poorly than ML males with a single dose (ML_SD) or MC controls. This result is intriguing because the selection protocol would have only allowed autosomal variants to be favored when they had penetrance in the heterozygous state, paired with the autosomal translocation used during selection. Expressed in two copies, ML-evolved autosomes actually depressed male fitness. This result potentially suggests specific adaptation to the translocation, with an “overdose” of compensatory factors occurring in the ML_DD males.

The lack of measurable differentiation in the male fitness experiments is all the more perplexing given the clarity and consistency of the results for females. In keeping with predictions for IaSC, the CRF of females expressing ML hemiclonal genomes was found to be markedly lower than MC female fitness (minus 15% to 20% in all experiments and replicates). In previous studies, the decline in female fitness of ML selected haplotypes, combined with the masculinization of phenotype was treated as clinching evidence of IaSC. While this decline could also be due to a relaxation of selection acting upon the female-limited loci of the genome, we consider this possibility unlikely considering the magnitude and rapidity of decline and the consistency across replicate populations and experiments. More likely, such a consistent reduction in female fitness came about from sexually antagonistic costs. Fitness costs experienced by females carrying ML-selected haplotypes seems to be consistent across assay and evolved conditions, while male benefit variation might be specific to the local conditions of selection.

A few further observations underscore our concerns about specificity of the response to local conditions. Firstly, in the hemiclonal analysis of the ML₁/MC₁ population pair, we found a strong and significant difference in male fitness in the predicted direction (ML > MC by about 20%). Genetic variance for fitness was lower for ML₁ compared to MC₁, as predicted, with the data suggesting that selection had removed genotypes with low male fitness (Fig.  4). However, an important caveat of these specific results is that the Y chromosomes were derived from the ML selection treatment (Fig. S3), which were potentially altered by exposure to generations of maternal transmission. It is possible that the apparent improvement in fitness and reduction in genetic variance for male fitness in MLs actually results from inflated variance in the MCs that were exposed to these novel Y chromosomes. In the same experiment, mean female productivity and genetic variance for productivity was not different between ML₁ and MC₁. We note that for logistical ease we chose to assay female productivity here, rather than fitness under competitive conditions, which might explain the absence of a difference.

Overall, we find that more than 80 generations of ML selection resulted in dramatic improvements in male fitness under evolved conditions, but those increases are not consistently apparent when the evolved genomic haplotypes were tested in matched control conditions. This might reflect the exclusive recruitment of specialized compensatory and local adaptations, but this cannot explain the consistent reduction of fitness in females carrying ML hemiclonal genomes. Alternatively, specialized adaptations to the ML selection regime may have actually generated a cost under control (CRF) conditions, moving them off of their fitness peak with male fitness being locally “neutralized” and female fitness generally measurably reduced. In this case, sexually antagonistic variation was selected upon, but the fitness gains of males were cancelled out, leaving a signature only in reduced female fitness. We note that previous male limited selection experiments have not resolved potential contributions from the array of artefacts used in ML evolution: Rice (1996) (fitness tested under selection conditions), Rice (1998) (CG derived Y chromosomes, double dose haplotypes), Prasad et al. (2007) (translocated autosomes, CG derived Y chromosomes and CG females) and Abbott et al. (2013) (CG derived cytotypes, Y chromosomes and reversed sex chromosome inheritance).

As previously noted, there is a high concentration of sexual conflict research in D. melanogaster upon a single population (LH_m), which may or may not be representative of the species as a whole. Moreover, the relentless operation of selection and drift in finite populations under domestication is likely to lead to changes in genetic variance–covariance matrices, pools of genetic variation, and thus potentially different results from even the same “base population”. LH_m itself has been shown to yield different results with regard to traits related to sexual conflict over both time and space despite defined rearing protocols (Jiang et al. 2011; Collet et al. 2016). In general, one might expect genetic polymorphisms, such as those associated with IaSC, to winnow over time, with increasingly positive intersexual correlations governed by mutation-selection balance. Our results also point to a need to further investigate the impact of specialized local and compensatory adaptation in the application of such selection protocols in order to understand their insights into sexual conflict. We echo Abbott et al.’s (2023) lead off question: “Why is measuring and predicting fitness under genomic conflict so hard?”. In research currently in prep, we aim to answer many of the questions raised here by systematically varying all known artefacts involved in ML evolution.

Author contributions

HT, TD and AKC developed the overall theme of the study; HT and AKC designed the experiments; HT, AKC, IS, MGB, & JAK carried out the experiments; HT analysed the data; HT wrote the first draft of the paper; HT, TD and AKC contributed to the final draft and revisions. Harshavardhan Thyagarajan (Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review & editing), Imran Sayyed (Methodology, Writing—review & editing), Mindy Baroody (Methodology, Writing—review & editing), Joshua Kowal (Methodology, Writing—review & editing), Troy Day (Conceptualization, Writing—review & editing), and Adam K Chippindale (Conceptualization, Writing—original draft, Writing—review & editing)

Data availability

Data archived here: osf.io/jgp59/5.