Maintenance costs of male dominance and sexually antagonistic selection in the wild

1. Variation in dominance status determines male mating and reproductive success, but natural selection for male dominance can be detrimental or antagonistic for female per -formance, and ultimately their fitness. Attaining and maintaining a high dominance status in a population of competing individuals is physiologically costly for males. But how male dominance status is mediated by maintenance energetics is currently not well under -stood, nor are the corresponding effects of male energetics on his sisters recognized. 2. We conducted laboratory and field experiments on rodent populations to


| INTRODUC TI ON
Intralocus sexual conflict can drive males and females from their sexspecific life-history optima, thereby compromising lifetime fitness (Bonduriansky, Maklakov, Zajitschek, & Brooks, 2008;Mokkonen, Koskela, Mappes, & Mills, 2016). Such conflict occurs when the same alleles have opposing fitness between males and females that, unless there is sex-limited gene expression, impedes adaptive evolution (Bonduriansky & Chenoweth, 2009;Van Doorn & Fe, 2009). Intralocus conflict has the potential to generate sexually antagonistic selection affecting fitness through survival and the reproductive components of fitness, such as mating and reproductive success. However, male success in mating and reproduction is physiologically costly (Vehrencamp, Bradbury, & Gibson, 1989). Currently, the maintenance energetics mediating male fitness are not clear, nor is the presence of sexual antagonism for such maintenance costs.
Behavioural dominance often defines a male's access to mates (Qvarnström & Forsgren, 1998). Testosterone, which mediates the dominance hierarchy in males, can impose differential expression of many physiological pathways between males and females (Peterson et al., 2014). Testosterone can affect mating behaviour Mokkonen, Koskela, Mappes, & Mills, 2012) and fitness: For example, red deer male calves born to first-time mothers were less likely to survive if they had high neonatal testosterone level (Pavitt, Walling, Mcneilly, Pemberton, & Kruuk, 2014). Male dominance is also directly influenced by the tactics of other males in the population, showing negative frequency-dependent selection in the wild, that is, being costly for males when common in the population (Mokkonen et al., 2011). Dominance status and testosterone level can both be heritable and respond to selection (Pavitt, Walling, Mcneilly, et al., 2014;Schroderus et al., 2010), but the evolution of sex-specific testosterone levels can be constrained by cross-sex genetic correlation . The status of a highly dominant male, with a high testosterone level, can reveal physiological costs (Bryant & Newton, 1994;Røskaft, Järvi, Bakken, Bech, & Reinertsen, 1986). However, selection for testosterone level can differentially associate with son versus daughter reproductive success, causing a negative correlation in fitness between siblings Mokkonen et al., 2012). Testosterone may differentially affect overall gene expression between males and females, affecting many metabolic and physiological traits (Peterson et al., 2014). Yet, currently it is unclear how selection for dominance in males affects maintenance costs in both sexes, females in particular, or whether frequency-dependent selection on dominance promotes or constrains adaptations in metabolic traits (Buchanan, Evans, Goldsmith, Bryant, & Rowe, 2001).
BMR and testosterone can be involved in the expression of life-history traits linked to fitness (Moore & Hopkins, 2009). For example, females with higher metabolism can transfer more testosterone to eggs (Tschirren et al., 2016), while the close physiological association between testosterone level and metabolic rate can result in testosterone-mediated honest sexual signalling in males (Buchanan et al., 2001). But it is still unclear how BMR affects the evolution of sexually antagonistically selected life-history traits and whether it links to frequency-dependent selection on male tactics (Mills et al., 2009Mokkonen et al., 2011). Bateman's principle predicts that male fitness is primarily shaped by mating success, whereas female fitness is primarily shaped by longevity to optimize their lifetime reproductive effort (Rolff, 2002). As mating and reproductive success can be affected by energetic physiology (Boratyński & Koteja, 2010), selection for increased male fitness would theoretically affect physiological performance of both sexes as well as their level of BMR (i.e., maintenance costs). In particular, according to the parental care concept of the evolution of endothermy, which provides an adaptive explanation for the evolution of a high level of BMR, selection for increased reproductive output should result in a correlative response in increase of the level of maintenance metabolism, at least in females (Bacigalupe, Moore, Nespolo, Rezende, & Bozinovic, 2017;Koteja, 2004). However, it is unclear whether the cross-sex genetic correlation of metabolism can constrain responses to antagonistic selection between males and females.
We would predict that (1) if the cross-sex genetic correlation for BMR is relatively low in our study species (Boratyński et al., 2013), males and females can theoretically approach their optimal levels of maintenance metabolism. Being dominant is energetically costly, and thus according to the "increased intake" hypothesis, we would predict that (2) metabolic performance, rather than size alone, along with the level of expressed testosterone, can constrain expression of male dominance (Buchanan et al., 2001). Alternatively, and according to the "allocation" hypothesis, low BMR might release resources to invest in behavioural performance such as dominance.
We would therefore predict that (3) energetic costs for individuals expressing high dominance status might be different between males whose fathers were dominant versus those whose fathers were subordinate. Males not genetically predisposed to high dominance might manifest higher energetic costs (sensu "increased intake" hypothesis) of expression of such status than those males which inherited genes of dominance (sensu "allocation" hypothesis).
Theories predicting the evolution of a high level of BMR (Angilletta & Sears, 2003;Farmer, 2000) postulate that individuals with a higher metabolic capacity can sustain elevated energetic demands of reproduction (e.g., males maintaining larger territories, females providing their pups with more food). However, information is lacking on the fitness costs of high versus low maintenance physiology, and BMR, for males and females selected for high versus low reproductive output. Theoretically, we predict that (4) a high BMR could result from correlative selection for high reproductive output and physiological capacity. However, previous studies have shown that not only can the direction of selection on BMR differ between males and females, and among seasons (Boratyński & Koteja, 2009, but a low BMR can generally be beneficial for fitness (Boratyński et al., 2013).
We experimentally tested these hypotheses in a field study on bank voles subjected to artificial selection in the laboratory that resulted in the selection lines of behaviourally dominant males with sisters of low fecundity and subordinate males with sisters of high fecundity (Mokkonen et al., 2011). We tested whether selective breeding of male bank voles with high and low dominance status influenced their BMR and whether any associated sexually antagonistic effects can be observed in female metabolic performance.
We tested whether BMR is inflated in males due to inherited correlational line-specific selection effects or whether they are related to behavioural (dominance) and physiological (testosterone level) phenotypic status. We also estimated whether male dominance status is primarily determined by testosterone level or whether it is constrained by BMR. Ultimately, we tested whether BMR influenced the main reproductive fitness components, mating and reproductive success, and whether frequency-dependent selection operated on male dominant vs. subordinate breeding lines in the field.

| Study species and artificial selection for male dominance
The bank vole is a Palearctic species with a polygynandrous mating system. Bank voles have been shown to experience sexually antagonistic selection for testosterone: Artificial selection for testosterone is associated differentially with son versus daughter reproductive success, causing a negative correlation in fitness between full-siblings Mokkonen et al., 2012). To further investigate phenotypic traits related to sexually antagonistic relationship between son and daughter fitness, we artificially selected bank voles in the laboratory based on male dominance, the details of which are described elsewhere (Mokkonen et al., 2011). Briefly, two selection lines were created for this experiment using artificial selection; high-dominance males, M, were mated with females of low fertility, f (Mf line), and low-dominance

| Field selection experiment
To measure strength and form of natural selection on the phenotype, we used a total of 91 females and 91 males in a field experiment. In total, 20 populations were created in two replicates (11 and 9 populations) using 11 field enclosures measuring 40 × 50 m each (Mokkonen et al., 2011). Each population consisted of an equal number of males to females and contained 1 male and 1 female from a given selection line with 4 (1st) or 3 (2nd replicate) males and females from the other selection group (e.g., population A: 1 Mf male + 1 Mf female + 4 mF males + 4 mF females; population B: 4 Mf males + 4 Mf females + 1 mF male + 1 mF female) to test for the frequency-dependent selection on breeding lines (Mokkonen et al., 2011). Individuals were randomly assigned to enclosures; however, sibling assignments to the same enclosure were avoided. Each enclosure contained 20 Ugglan live traps in a 4 × 5 grid pattern, and they were spaced 10 m apart. Sheet metal fencing (1.0 m above, 0.5 m below ground) surrounded each enclosure, preventing escape of study individuals while at the same time allowing possible predators to enter the enclosures. Study individuals relied on natural field conditions and resources to survive. Initially, females were released to enclosures. After 4 days, males were then released and all individuals were left to survive and breed in the field enclosures. Eighteen days after males were introduced, all study individuals were trapped out of the enclosures and brought to the laboratory for females to give birth. In the laboratory, pregnant females were monitored every 24 hr. After a birth, tissue samples were taken from each offspring individual and common litter characteristics were recorded. Each offspring was genotyped by extracting DNA using the Qiagen DNeasy Tissue kit and KingFisher magnetic particle processor, and then, Cervus 3.0 software was used to assign paternity (Kalinowski, Taper, & Marshall, 2007). The use of study animals and all above protocols adhered to ethical guidelines for animal research in

| Correlative responses in BMR
We used a generalized mixed model (GLMM) procedure to test for female and male correlative responses (maternal and heritable effects) in BMR to artificial selection for male dominance status. Logtransformed BMR was included as the dependent variable (Gaussian family function), male selection line affiliation (dominant vs. subordinate line) and sex as fixed cofactors, and (log-transformed) age and body mass at metabolic trials as covariates. Due to a significant effect of sex on dependent variables, models were also run separately for males and females.

| BMR, dominance phenotype and genotype
To test for energetic costs between dominance phenotypes and genotypes of male bank voles we ran a GLMM including log-transformed BMR as the dependent variable (Gaussian family function), dominance status (dominant vs. subordinate status) and selection line affiliation (dominant vs. subordinate line) as fixed cofactors, and (log-transformed) testosterone level, age and body mass at metabolic trials as covariates.
All models included mother ID and respirometry chamber ID as random factors (to control for common early environment/relatedness among individuals and variation in respirometry machine) and tested factorial interactions (those insignificant were sequentially excluded). Effect sizes are presented as percentages of differences between back-transformed least-squares means predicted from above models. Statistical tests from models with residual BMR (from linear regression of BMR on age and body mass) as the dependent variable (or covariate) are presented in the main text for simplicity.

| Selection analyses
The strength and form of selection on phenotype in the wild was and its frequency as fixed cofactors, and (log-transformed) BMR, body mass and age at the onset of experiment as covariates. We tested factorial interactions between all characters while controlling for maternal, common environmental and replication effects (random factors of mother ID, population ID and replication of experiment). Linear and quadratic effects and interactions between BMR, age and body mass were tested to account for directional, stabilizing/disruptive and correlational selections, respectively (due to limited power, interactions were first tested in separate models per interaction type, and finally, significant terms were included in one model; Artacho, Saravia, Ferrandière, Perret, & Le Galliard, 2015;Rønning et al., 2016). Age did not affect fitness components and it was excluded from final tests. To remove the correlation between linear and quadratic terms, continuous predictors were standardized within datasets. Statistical modelling with GLMM was performed with "glmmADMB" (http://glmmadmb.r-forge.r-project.org/) and "lme4" packages in R (www.r-project.org). For comparison of strength and form of selection with other studies standardized selection gradients were estimated (Artacho & Nespolo, 2009;Lande & Arnold, 1983;Pettersen, White, & Marshall, 2016;Schluter, 1988).
Linear (β) and nonlinear (γ; quadratic, correlational) selection gradients were estimated by means of coefficients from multiple regression analyses, separately for females and males, and within selection lines, with relative fitness as dependent variable and standardized quantitative traits (BMR and body mass) as predictors. Linear effects (β) were estimated in multiple regression models including only main effects, while nonlinear effects (γ) were estimated in models including also quadratic or correlational terms along the linear effects.
Coefficients for quadratic terms were multiplied by two (Fairbairn & Reeve, 2001). The 95% confidence intervals of the coefficients were estimated with 1,000 bootstrapping.

| Correlative responses in BMR
Residual values were calculated from a multiple linear regression with BMR as the dependent variable and body mass and age as independent variables to account for significant correlations of BMR with age (t = 2.86, p = 0.005) and body mass (t = 8.60, p < 0.001).

| Dominance phenotype and BMR
In male-male competition trials, males were 81.4% more likely dominant in the line selected for dominance status. When analysed over both selection lines, phenotypically dominant males have 8.8% higher BMR than subordinate males (line-by-BMR interaction was insignificant, z = 0.14, p = 0.89; Table 2 and Supporting Information   Table S3). However, phenotypically dominant males from the subordinate selection line have the highest BMR ( Figure 1d). BMR was 11.6% higher than in phenotypically subordinate males from the same line (z = 2.49, p = 0.013) and 8.4% higher than in phenotypically dominant males from the dominant line ( Figure 1d). There was no significant difference in BMR between phenotypically dominant and subordinate males from the dominant selection line (7.1%, z = 0.58, p = 0.30; Figure 1d).

| BMR, dominance phenotype and genotype
We found that variation in male BMR was explained by the interactions between selection line and testosterone level, and between dominance status and testosterone level (but line-bytestosterone-by-dominance status interaction was insignificant: z = 0.71, p = 0.48; Table 3 and Supporting Information Table S4).
TA B L E 2 Generalized mixed models for males only to quantify effects of residual BMR (accounted for variation in age and body mass), selection procedure ( Residual BMR values were calculated from mixed models (accounting for variation in body mass and age, as covariates, and mother and respirometric chamber IDs, as random factors) after excluding significant interactions between testosterone level and dominance, and line and testosterone (Table 3) The BMR of males that behaved phenotypically dominant tended to decrease with increasing testosterone levels, whereas the BMR of males that behaved phenotypically subordinate tended to increase with increasing testosterone levels (Figure 2a). On the other hand, the BMR of males from the male dominant line tended to increase with increasing testosterone, whereas the BMR of males from the male subordinate line tended to decrease with increasing testosterone (Table 3 and Supporting Information   Table S4, Figure 2b).

| Sired litters and offspring
In our initial analysis, we found a significant interaction between frequency (rare or common) and line ( Figure 3a), but a non-significant opposite trend was found between males from the subordinate selection line (average difference in number of litters was 0.6; Table 4; see Supporting Information Table S5 for selection gradients). Furthermore, we found that the BMR-by-line interaction significantly explained variation in the number of sired offspring (Table 4) Table 4).
Frequency-dependent selection on body mass was significant in our overall tests for the number of sired litters and offspring, and in the test within the dominant line for the number of sired offspring (Table 4). Males from the dominant line sired 3.6 more offspring if they were rare than if they were common in the population (5.8 vs. 2.2 offspring sired; Table 4).

| Female mating success and litter size
Females that reproduced were 13.0% heavier than females that did not reproduce, but the trend was not significant (p = 0.059; Table 5).
Females from around the median of the BMR distribution tended to reproduce less often than both females with the lowest (16.7%) and highest (55.1%) BMR (Table 5). The number of pups raised by females from the subordinate line increased as their BMR values increased (2.8 more pups: 6.4 vs. 3.6 pups from females with the highest vs. lowest BMR; Figure 3b), whereas the number of pups raised by females from the dominant line decreased as their BMR values increased (1.9 fewer pups: 3.75 vs. 5.6 pups from females with the highest vs. lowest BMR). However, this contrasting effect of BMR between lines (BMR-by-line interaction on litter sizes: p = 0.015) did not meet significance in our within-line tests (p ≥ 0.13; Table 5; see also results for selection gradients in Supporting Information Table S6).

| D ISCUSS I ON
Similarly, the sisters of these males from the dominant line tended to suffer fitness disadvantages of high BMR, characterized by smaller litter sizes (Figure 3b; Supporting Information Table S6). However, a positive trend of BMR on fitness components did not meet significance for individuals from the subordinate selection line (Tables 4 and 5).
Increased maintenance costs, characterized by elevated BMR, in females compared to males in the line selected for male dominance (Figure 1), suggests linkage between male fitness and female energetics (Hayward & Gillooly, 2011;Watson, Arnqvist, & Stallmann, 1998), or more specifically, genetic mito-nuclear incompatibility between the sexes (Hill, 2018;Rand et al., 2001). As the inheritance of mitochondria, the "energetic factories" of the cells, in mammals is mostly via females, selection on a male's energetic performance might be ineffective. However, such selection may promote nuclear genomic variation towards male (and not female) fitness optima due to counteradaptation in males and an imperfect match between (male) nuclear and (female) mitochondrial co-expressed genes (Boratyński, Ketola, et al., 2016;Hill & Johnson, 2013). As low maintenance costs can be, in general, considered selectively beneficial (Boratyński et al., 2013), the increased BMR in sisters of males selected for high dominance may suggest unresolved conflict and gender load mediated by mito-nuclear incompatibilities (Bonduriansky & Chenoweth, 2009;Mills et al., 2012;Petersen et al., 2013). Alternatively, the parental care (and related) models for the evolution of endothermy argued that females benefited from an increased metabolic rate and BMR (Lovegrove, 2017;Wone, Sears, Labocha, Donovan, & Hayes, 2009). Accordingly, our results suggest that selection for male fitness can also benefit females via its physiological performance. However, in our natural selection experiment in the field we had conflicting results. While there was an indication of disruptive selection on female BMR in terms of their probability to reproduce (Table 5), BMR had also opposing effects on a female's litter size between male dominance selection lines (Figure 3b; Table 5 and Supporting Information  Figure 3b). These results suggest that selection for male fitness has rather negative outcomes on female reproductive performance.
In general, males with a high BMR level were more likely to express dominant behaviour (Table 2). But, elevated BMR in phenotypically, but not genetically, dominant males entails high physiological costs of behavioural dominance in males not genetically suited for dominance ( Figure 1d). The proposed models explaining intraspecific variation in BMR distinguish between compensatory and predisposition functions of BMR in relation to other life-history traits (Pettersen et al., 2018;Ricklefs & Wikelski, 2002;Šíchová et al., 2014). The results for male behavioural dominance status (Table 2), and the variation in association between BMR and testosterone levels (Table 3), suggest that the two models may perform on different, phenotypic and genetic, levels (Figure 2). At the phenotypic level, we found that a high BMR, and high maintenance costs, may limit the expression of testosterone in phenotypically dominant males F I G U R E 3 Relative fitness components (individual fitness/population mean fitness; from wild conditions experiment) and its 95% confidence intervals (shadings) for ( shown that negative frequency-dependent selection on bank vole male dominance maintains variation in sexually antagonistic alleles (Mokkonen et al., 2011). Here we showed that the selection on maintenance metabolism in the wild can also constrain the fitness of both sexes. While artificial selection for male dominance resulted in increased male dominance, greater male reproductive success and higher female BMR, the field selection experiment pointed to selective disadvantages of high BMR in both sexes in the dominant line.
However, while high BMR was selected against in males and female from the dominant line, at the same time females from the subordinate line tended (not statistically significant trend) to be selected for higher BMR (Figure 3). As those are the most reproductively successful individuals, this result suggests an opportunity for sexual antagonism over maintenance metabolism in bank voles. Thus, along with the findings of energetic costs of selection for dominance, our results suggest conflict between males and females in energy metabolism. It is possible that energetic capacity, rather than maintenance costs per se, more closely determines fitness, with BMR only being correlated with it. Thus, future work in this and other systems should investigate complete animal energetic budgets to assess the extent mito-nuclear conflicts shape fitness in the field.

ACK N OWLED G EM ENTS
We

DATA ACCE SS I B I LIT Y
Data deposited in the Dryad Digital Repository: https://doi. org/10.5061/dryad.vt4m939 .