Male morphological traits are heritable but do not predict reproductive success in a sexually-dimorphic primate

Sexual selection favours traits that increase reproductive success via increased competitive ability, attractiveness, or both. Male rhesus macaque (Macaca mulatta) morphological traits are likely to reflect the effects of multiple sexual selection pressures. Here, we use a quantitative genetic approach to investigate the production and maintenance of variation in male rhesus macaque morphometric traits which may be subject to sexual selection. We collected measurements of body size, canine length, and fat, from 125 male and 21 female free-ranging rhesus macaques on Cayo Santiago. We also collected testis volumes from males. We used a genetic pedigree to calculate trait heritability, to investigate potential trait trade-offs, and to estimate selection gradients. We found that variation in most male morphometric traits was heritable, but found no evidence of trait trade-offs nor that traits predicted reproductive success. Our results suggest that male rhesus macaque morphometric traits are either not under selection, or are under mechanisms of sexual selection that we could not test (e.g. balancing selection). In species subject to complex interacting mechanisms of selection, measures of body size, weaponry, and testis volume may not increase reproductive success via easily-testable mechanisms such as linear directional selection.


Results
trait heritability. In our models using data from both parents, all measurements except testis volume were heritable (h 2 ≥ 0.1 59 , Table 1a). DIC values for models including the animal (heritability) random effect were lower than DIC values for all models excluding the animal term, indicating that the addition of the heritability term produced a better fitting model (Table 1a,b), even though the testis volume heritability estimate did not meet our threshold to be considered heritable. Testis volume heritability estimates were extremely low (h 2 < 0.1), and about half of the variance in testis volume could be explained by the date the animals were measured (see more details below). Date measured was not a significant contributor to variance in any of the other traits. Confidence intervals for the heritability and maternal effects were wide, so we interpret the HDPI values with caution, given the very low lower limits of these intervals. None of our sex-linked heritability models converged.
Evolutionary and developmental trade-offs. Seasonality. Date measured influenced three of our traits of interest: body mass, testis volume, and crown-rump length ( Table 2). Males measured later in the trapping period had higher body masses, longer crown-rump lengths, and larger testis volumes (Fig. 1). Date measured did not influence measurements of canine length or abdominal skinfold thickness. These results contrast with those from our heritability models, where we treated date measured as a random, rather than a fixed effect. In the heritability models, date measured only contributed to a large proportion of the variation in testis volume.
Correlations between traits. In our dataset, body mass was significantly correlated with crown-rump length, testis volume and abdominal skinfold thickness (Table 3, Supplementary Fig. S1). Testis volume (either absolute or relative) was not correlated with either abdominal skinfold thickness or canine length (Table 3). Because relationships involving testis volume did not change after controlling for body mass, we used absolute testis volume in all subsequent analyses. Age was a significant term in all of the models, and was always negative.
Correlations between dominance rank and trait values. There was no relationship between current ordinal dominance rank (low, medium, high) and any morphometric trait (ANOVA, p > 0.05, Supplementary Table S1).

Selection on traits.
We found no definitive evidence for directional, stabilizing, disruptive, or correlational selection on any of our focal morphometric traits (Tables 4-6a,b). Neither age nor date measured influenced relationships between trait values and reproductive success. Dominance rank (as assumed from dispersal rate) had a minor effect in several of our models, but the term was not statistically significant (0.05 < p < 0.10).

Discussion
We used a quantitative genetic approach to investigate the production and maintenance of variation in male rhesus macaque morphometric traits putatively associated with intrasexual competition. Our results suggest that male morphometric traits are heritable, but that variation in these traits does not predict reproductive success. We also found that male morphometrics were not influenced by dominance rank, and we found no evidence for trade-offs in investment between morphometric traits.
For several traits, either additive genetic variance (heritability) or maternal ID contributed to a moderate proportion of the phenotypic variation in the trait. Canine length, body mass, and abdominal skinfold thickness all had moderate heritability values (h 2 > 0.3), though crown-rump length values were lower (h 2 < 0.2). We found very low additive genetic variance in testis volume. It is possible that intraindividual seasonal increases in testis volume have confounded our results. However, low additive genetic variance may also be the result of strong selection on testis volume, as strong selection may erode additive genetic variation 60 . Additionally, all of our estimates for heritability and maternal effects have wide confidence intervals, ranging from very low (0.02) to quite high (0.75) in some cases. Future studies with larger samples might help to resolve these values with greater confidence. Overall, our results suggest that variation in most male morphometric traits has a genetic basis, and as such, that these traits can evolve under selection.
We found no evidence for trade-offs between investment in different morphometric traits. Body mass was positively correlated with crown-rump length, testis volume, and abdominal skinfold thickness. These correlations likely reflect allometric relationships, not investment in multiple reproductive strategies, and are consistent with previous work on male morphometric traits in rhesus macaques 61 . We found no correlation between canine length and testis volume, and we also found that jointly, these traits did not explain variation in reproductive success, suggesting that inter-individual variation in male morphology does not reflect relatively higher investment in either direct or sperm competition. These findings do not support our prediction that testis volume would be inversely correlated with variation in body size (mass and length) and canine size. This finding is consistent with inter-specific analyses that have shown that species with lower levels of direct male-male competition do not exhibit strong trade-offs between investment in weaponry and investment in testis volume 10 but contrasts with intra-specific studies of other taxa, which have clearly demonstrated trade-offs between traits that are involved in pre-and post-copulatory competition 31,32,62 . This finding may also reflect different patterns of investment in  canines and testes: once canines are formed, no additional energetic investment is required -in contrast, testes need to be maintained throughout adult life 12 . We would need many data points collected across the lifespan of individual males in order to test this idea. Furthermore, our prediction that current dominance rank was not correlated with morphometrics was supported, providing further evidence that competitive ability is not important for dominance acquisition in this species. Lastly, these analyses revealed an effect of age on variation in morphology. Age was always a negative term in the models, suggesting that older animals are smaller than younger ones. We cannot determine whether this reflects the aging process, cohort effects, or selective mortality of particular male phenotypes with our data. Additional work is necessary to address whether smaller males are more likely to survive to older ages. Our prediction that body mass, testis volume, and fat mass would be higher in males captured later during the capture-release period was partially supported. We found that males captured later had higher testis volumes than those captured earlier, which confirms prior work demonstrating that rhesus males undergo dramatic increases in testis volume prior to the mating season, indicating strong investment in sperm competition 56 . We also found that males trapped later in the trapping season had higher body masses and longer body lengths, but not higher fat mass. This finding may either reflect seasonal increases in body size or an effect of body size on a male's ability to be captured. In order to test for seasonal increases in body size, we would need to collect data on body size in the months leading up to the mating season. Since it is not possible to trap animals on Cayo Santiago during this period, one possible way of doing this would be to collect multiple body length measurements per male (e.g., one  www.nature.com/scientificreports www.nature.com/scientificreports/ measurement per week in the three months leading up to the mating season) using photogrammetric methods (e.g., Breuer et al. 63 ; Wright et al. 64 ). Alternatively, this result may reflect the fact that larger males happen to be trapped later than smaller ones because they are harder to capture.
We did not find evidence of selection on any of our morphometric traits. Our results echo those from Atlantic cod 23 , bighorn sheep 24 , and sifakas 25 but contrast with those from mandrills 22 , Soay sheep 21 , red deer 20 , kangaroos 19 , minnows 18 , and field crickets 16 , among others. Our findings provide additional evidence that even in sexually dimorphic species, larger or more highly-weaponized males do not always enjoy the highest reproductive success. Rather, among rhesus macaques, female preference (such as that based on facial coloration 41,43 ) and male behavioural strategies 51,52 are likely stronger predictors of male reproductive success.
Our results could be interpreted in multiple different ways. One interpretation is that the measured male morphometric traits are not under selection. Alternatively, the results are also largely consistent with previous evidence of flat fitness landscapes and multiple routes to male reproductive success in male rhesus macaques 51,52 . Under these scenarios, we are unlikely to find clear linear or quadratic relationships between specific traits  www.nature.com/scientificreports www.nature.com/scientificreports/ and reproductive success, because there are multiple routes to equal levels of success. We found weak, but not statistically-significant, evidence that male dominance rank influences reproductive success independent of variation in male morphometric traits, consistent with previous analyses of reproductive skew 47 . Because rhesus males do not contest dominance, this result provides further evidence that direct male-male competition is not a strong selection pressure in this species. In general, our results confirm prior research indicating that variation in morphological traits associated with competition is not a strong predictor of reproductive success in this species 54,61 . Additional studies are necessary to determine whether male reproductive strategies or aggression levels are correlated with morphology.
Our findings illustrate that in species that are evolving under multiple sexual selection pressures, such as rhesus macaques, male traits like large body size, enhanced weaponry, and large testis volumes may not increase reproductive success through linear, quadratic, or correlational selection. Our results highlight the importance of understanding how interactions between sexual selection pressures, and between behavior and morphology, function to influence male reproductive success.

Methods
Field site and subjects. Cayo Santiago is a 15.2 hectare island located off the southeast coast of Puerto Rico. The Caribbean Primate Research Center (CPRC) manages the island and the population of free-ranging rhesus macaques that live there 65 . At the time this study was conducted, the island was inhabited by ~1,500 rhesus macaques divided into seven naturally-formed social groups, all of which descend from a founding population of 409 animals brought to the island from India in 1938 65 . Even though no outside animals have been introduced into the colony, the population is not inbred 66 . The CPRC monitors the population daily and maintains long-term (>75 years) behavioural and demographic databases including data on social group membership for all animals, plus a genetic parentage database for animals born after 1985 65,67,68 . Each year, before the onset of the mating season, a subset of the animals ranging on the island are captured for collection of blood samples and morphometric data, and then released. During the capture-release period, all one-year-old animals are captured, sampled for blood, assigned a unique ID, and tattoed, enabling researchers to easily identify individual animals.
Morphometric data collection. We collected morphometric data from male and female rhesus macaques (n = 146) captured during the annual capture-release period (October 15, 2015 to December 15, 2015). Our sample is composed of all adult males (ages 6 and above) ranging on the island who were able to be captured (n = 125), but we also collected data on females closely related to the males we sampled (n = 21) for use in our heritability analyses. We collected a set of measurements on focal traits that could act as proxies for different types of male-male competition: crown-rump length (direct contest competition), body mass (direct contest competition  Table 4. Linear selection gradients (GLMs) for morphometric traits. Selection gradients are shown as the estimate +/− the standard error. Statistically significant terms (p < 0.05) are shown in bold. β refers to the linear selection gradient, t to the t-value, and p to the p-value.  www.nature.com/scientificreports www.nature.com/scientificreports/ and endurance rivalry), canine length (direct contest competition), testis dimensions (sperm competition), and upper abdominal skinfold thickness (endurance rivalry). We chose these traits based on prior studies of sexual selection in rhesus macaque males 54,61 . Animals were only captured and measured once during the capture-release period. All measurements were collected by one trained observer. Body weight was measured using a hanging scale and all other measurements were collected using either a tape measure or digital calipers (accurate to 0.01 mm). Weight (lbs) was converted to mass (kg) for all analyses. Testis volume was calculated from three dimensions: height (h), width (w), and depth (d) and modeled as an ellipsoid: π = V hwd 4 3 69 . We calculated relative testis volume by dividing testis volume by body mass. We excluded one measurement of an extremely worn or broken canines, one crown-rump length value that was three standard deviations above the mean and likely an error, and measurements affected by pathological conditions (n = 12) from our analyses.

Crown-Rump Length Body
Genetic parentage information. The CPRC maintains a pedigree database containing information on behavioural dams (available for all animals) as well as genetic parentage assignments for dams and sires (available for animals born after 1985). Genetic parentage assignments are made based on a panel of microsatellites 48 . We used the R package MasterBayes to prune the full Cayo Santiago pedigree so that it only included phenotyped animals and those individuals that provided connections between them 70 . Our pruned pedigree spanned ten generations and included 902 animals, with 885 maternities and 567 paternities (pedigree statistics were generated using the pedantics R package 71 ).
We used average number of offspring produced per year as a proxy for male reproductive success. We then calculated relative annual reproductive success (each animal's reproductive success divided by the average value across our entire sample) and used this measure in our selection gradient models 72 . We could not use lifetime reproductive success because the majority of our study animals had not yet reached reproductive senescence (>17 years of age 48 ). Dominance rank. We quantified dominance rank two different ways: first, over the course of one year and second, as an average measure over the animal's life to date (up to when they were captured for morphometric measurements). We used current dominance rank (available for a subset of 55 adult males) to determine whether current rank and male morphology were correlated. We determined the dominance rank of all subjects using pairwise win-loss information from agonistic encounters that were recorded during focal animal samples or during ad libitum observations collected as part of an on-going, unrelated, study. We calculated dominance rank amongst males living within the same social group group. In order to account for variable group sizes, we then calculated dominance rank as the percentage of male groupmates that a subject outranked. We then classed males as either high, mid or low ranking based on this scale, with high ranking animals being those that outranked between 80-100% of males in their group, and low-ranking animals being those that outranked fewer than 49% of males in their group. We chose to bin the ranks this way because the behavioral data used to calculate these ranks are fairly coarse, so these categories are likely to be more accurate than continuous ranks, which may contain errors in exact rank order. Furthermore, the correlation between average dispersal rate and continuous  www.nature.com/scientificreports www.nature.com/scientificreports/ dominance ranks, while strong, is not perfect, so binning the ranks makes the measures more comparable. Finally, this method has been used in previous studies on this population 73,74 so conducting our analyses this way makes our study more comparable with prior work.
In our selection gradient analyses (n = 108), we chose to use an average measure of dominance rank -average annual dispersal rate -because our measure of reproductive success was also averaged over the animal's life (up until they were captured) 41 and we did not have the behavioral data necessary to calculate dominance ranks for many of the males in our sample. Dispersal rate is a good proxy for dominance rank because rhesus males acquire dominance through queuing instead of contest, such that a male's dominance rank can be predicted by group tenure length 46 . In our dataset, dominance rank and tenure length were strongly correlated -males with longer tenure lengths were higher-ranked (Pearson's product moment correlation = −0.579, n = 81, p < 0.001).
Statistical analyses. All statistical analyses were run in R version 3.5.2 75 . We considered p-values to be significant if alpha levels were below 0.05. P-values for generalized linear models (GLMs) were calculated based on a Student t distribution, p-values for ANOVA models were calculated using the F distribution.
Trait heritability. We used animal models to estimate narrow-sense heritability values (h 2 ) for our morphometric traits. Animal models are univariate generalized mixed models that combine phenotypic and pedigree data to parse out the contributions of additive genetic and environmental factors to variation in a trait 76,77 . We implemented our models in the R package MCMCglmm 78,79 . We ran models on a combined sample of both males and females for body mass, crown-rump length, canine length, and upper abdominal skinfold thickness. We included six-year old males in our canine heritability analyses, as canine formation and eruption is generally complete by age six 80,81 , while all other analyses were run on males age seven and above because body growth is generally not complete until age seven. We ran models on mean-scaled measurements. In the pooled sex models, we controlled for age and sex (fixed effects), plus maternal ID, animal ID, and date measured (random effects). We also ran models for testis volume; these included maternal ID, animal ID, and date measured (random effects) and age (fixed effect). Lastly, we ran models for males using only paternal pedigree data to test for sex-linked inheritance, using the same model structure as listed above. We included maternal ID to account for non-genetic differences in maternal care and date measured to account for any changes in morphology over the course of the capture-release period. Animal ID is used to calculate the additive genetic variance.
We ran models on each trait for 2,550,000 iterations with a burn-in period of 50,000 iterations and a thinning interval of 1,000. Although we ran our analyses using a range of prior types and structures, we report values from models with inverse Wishart priors (V = 1, nu = 0.2). Priors with lower values of nu (e.g., nu = 0.002) did not mix well -the chains were autocorrelated -and confidence intervals for random effect terms were very wide. In order to verify that models met assumptions regarding autocorrelation and convergence, we inspected plots of the MCMC chain, ran Heidelberg stationarity tests, and ensured that autocorrelation between estimates was less than 0.1 82,83 . We calculated narrow-sense heritability values (h 2 ) by dividing the proportion of variation due to additive genetic variance (V A ; the posterior distribution of the animal effect) by the total phenotypic variance (V P ; the summed posterior distribution of the maternal effect, date effect, and residual variance). We also ran a set of models without the animal ID term. We then compared DIC values from models with animal ID and those without -models with the lower DIC value were considered to be the best fit models.
Evolutionary and developmental trade-offs. First, we explored how seasonality may influence our morphometric traits of interest, as this has direct implications for our ability to detect trade-offs between traits. We investigated whether the date an animal was measured was related to variation in morphometrics using generalized linear models (GLMs). We set the trait as the response variable and age and date measured as fixed effects.
We then investigated whether males exhibited trade-off between traits associated with different mechanisms of competition using GLMs. We examined relationships between body mass and testis volume, body mass and abdominal skinfold thickness, abdominal skinfold thickness and testis volume (both relative and absolute), canine length and testis volume (both relative and absolute), and canine length and body mass, controlling for age and date measured.
We also explored whether current dominance rank (categorical: low, medium, high) was related to variation in male morphology (crown-rump length, body mass, canine length, testis volume, relative testis volume, and abdominal skinfold thickness) using ANOVA tests. We controlled for age, date measured and social group in our analyses.
Selection on traits. We assessed whether male trait variation predicted variation in reproductive success (measured as average annual offspring production) using selection gradient models 84,85 . We calculated linear selection gradients to estimate directional selection on single traits, quadratic selection gradients to estimate disruptive or stabilizing selection on single traits, and correlational selection gradients to determine if trait values in combination with average annual dominance rank influenced reproductive success. We ran linear selection gradients using mean-standardized trait values, quadratic selection gradients on squared mean-standardized trait values 86 , and correlational selection gradients using mean-standardized values of one trait multiplied by mean-standardized values of another trait 87 . We ran correlational models for trait values and average annual dominance rank, and for pairs of morphometric traits. We used the cube-root of body mass and testis volume in these correlational gradients so that both of our morphometric variables of interest were on the same scale. We controlled for age, dominance rank, and date measured (fixed effects) in our models. We square-root transformed age and average annual dominance rank so that models met assumptions (normally-distributed residuals), but