Vascularization underlies differences in sexually selected skin coloration in a wild primate

Male reproductive competition can select for condition‐dependent, conspicuous traits that signal some aspect of fighting ability and facilitate assessment of potential rivals. However, the underlying mechanisms that link the signal to a male's current condition are difficult to investigate in wild populations, often requiring invasive experimental manipulation. Here, we use digital photographs and chest skin samples to investigate the mechanisms of a visual signal used in male competition in a wild primate, the red chest patch in geladas (Theropithecus gelada). We analysed photographs collected during natural (n = 144) and anaesthetized conditions (n = 38) to understand variability in male and female chest redness, and we used chest skin biopsies (n = 38) to explore sex differences in gene expression. Male and female geladas showed similar average redness, but males exhibited a wider within‐individual range in redness under natural conditions. These sex differences were also reflected at the molecular level, with 10.5% of genes exhibiting significant sex differences in expression. Subadult males exhibited intermediate gene expression patterns between adult males and females, pointing to mechanisms underlying the development of the red chest patch. We found that genes more highly expressed in males were associated with blood vessel development and maintenance but not with androgen or oestrogen activity. Together, our results suggest male gelada redness variability is driven by increased blood vessel branching in the chest skin, providing a potential link between male chest redness and current condition as increased blood circulation to exposed skin could lead to heat loss in the cold, high‐altitude environment of geladas.


| INTRODUC TI ON
Male reproduction is a zero-sum game that can lead to intense competition (Weir et al., 2011). Because actual fighting is energetically costly and exposes both the winner and loser to injury and infection, rival males can benefit from displaying their strength or condition using signals that allow for competition without costly engagement in fighting (Bradbury & Vehrencamp, 2011;Maynard Smith & Parker, 1976). The conspicuous male traits used for rival assessment (sexually selected signals) allow for conflict resolution at the lowest cost to both males (Maynard Smith, 1982;Maynard Smith & Harper, 1995, 2003, and are often condition-dependent, such that they reliably indicate some aspect of 'quality' that predicts their ability to win a physical fight (i.e. body size, body condition, current health status) (Fisher, 1915;Penn & Számadó, 2020;Trivers, 1972;Zahavi, 1977). Because the expression of the signal is limited by a male's condition, these signals tend to be honest indicators of ability because only high-quality males can produce the strongest signals (Weaver et al., 2017).
While the functional consequences of signals can be observed noninvasively and are thus relatively well-studied, the mechanisms underlying these traits are much less well-understood. For example, although observational data demonstrate that higher ranking male drills (Mandrillus leucophaeus) exhibit redder lip and groin coloration than lower ranking males (Marty et al., 2009), identifying the causal mechanism of this difference would require invasive methods such as surgical implantation of slow-release hormone devices, social manipulation, or genetic manipulation, which are often not possible or ethical to conduct in wild populations (Emlen et al., 2012;Hau et al., 2000;Karubian et al., 2011). However, capture-and-release programs in wild populations allow tissue sample collection for genomic and transcriptomic analyses (Tung et al., 2010), providing a minimally invasive technique to investigate the molecular correlates of signal mechanisms without experimental manipulation.
Here, we investigated the mechanisms underlying a visual signal that may mediate male competition in a wild primate: the brilliant red chest patch in male geladas (Theropithecus gelada), a cercopithecine primate endemic to the highlands of Ethiopia (Figure 1). Geladas have a multi-tiered social system in which multiple 'reproductive units' (one dominant adult male, one or more subordinate adult males, up to a dozen adult females and related offspring) associate to forage, rest and move. The dominant 'leader' male of units have the majority of the mating opportunities and displays the reddest chests as compared to the subordinate 'follower' males (with many fewer reproductive opportunities) and 'bachelor' males living in all-male groups (with no reproductive opportunities) (Bergman et al., 2009;Bergman & Beehner, 2008). Because geladas live in large, fluid societies where males frequently congregate and forage with other males that they do not recognize individually (Bergman, 2010), the chest patch is hypothesized to be a sexually selected signal that mediates male rival assessment and allows males to 'size up' others prior to engaging in conflict (Benítez, 2016). Bachelor males, which tend to be young adult males awaiting their chance to overthrow and replace leader males (Pappano & Beehner, 2014), are particularly attentive to leader male chest colour.
For example, across a 1-year study, leaders with redder chests (after a vigorous ritualized 'vocal display' accompanied by running, throwing rocks, climbing trees or shaking branches) were less likely to be overthrown by bachelors than their less-red counterparts (Benítez, 2016).
Previous research conducted on the same population has shown that chest redness in geladas is mediated by increased blood flow with a concomitant increase in surface temperature. Specifically, redder chests are associated with (1) increased physical activity (Bergman et al., 2009;DeLacey et al., 2022) and (2) higher chest skin surface temperature, whether measured using internal body heat or following the application of an external heat pack (DeLacey et al., 2022). The relationship with physical activity could be similar to skin flushing in humans during exercise, when skin blood flow increases to dissipate heat generated by muscle contractions (Kenney & Johnson, 1992). Further, the production of skin redness has been linked to characteristics of the blood vessels beneath the skin in humans and other primates where higher haemoglobin oxygen saturation increases skin redness in the short term (Changizi et al., 2006). branching in the chest skin, providing a potential link between male chest redness and current condition as increased blood circulation to exposed skin could lead to heat loss in the cold, high-altitude environment of geladas.

K E Y W O R D S
gene expression, male competition, Theropithecus gelada, visual signal F I G U R E 1 An adult male gelada (Theropithecus gelada) with a brilliant red chest patch. Photo by E. Roberts.
Sexually selected signals in males are often mediated by testosterone, a steroid hormone involved in regulation of reproductive function in male vertebrates (Ketterson & Nolan Jr., 1999;Plant & Zeleznik, 2014) that is associated with reproductive benefits (Enstrom et al., 1997) and physiological costs (Muehlenbein et al., 2006;Muehlenbein & Bribiescas, 2005). High doses of circulating testosterone are known to dilate vascular networks and increase blood flow, providing an avenue for testosterone to work alongside a blood flow mechanism to influence chest redness in geladas (Molinari et al., 2002;Webb et al., 1999). Testosterone can alter gene transcription by (1) binding to the androgen receptor (AR), (2) aromatizing to oestradiol and then binding to oestrogen receptor (ER ) or (ER ) or (3) converting to 5 -reductase which binds to AR but cannot convert to oestradiol (Hau & Wingfield, 2011). Local conversion of testosterone to oestradiol is a particularly strong candidate for chest redness regulation based on the results of a similar study; in the closely related male rhesus macaque (Macaca mulatta), increases in testosterone increased both redness and blood flow in sex skin areas. Moreover, the administration of an aromatase inhibitor (which prevents the conversion of testosterone to oestradiol) decreased skin redness (Rhodes et al., 1997).
Therefore, to better understand the molecular correlates of the chest patch signal mechanism, we collected chest skin biopsies from male and female geladas from a wild population in Ethiopia during non-lethal, immobilizations (hereafter 'capture-and-release') to explore differences in gene expression local to the chest patch.
Digital photographs complement this dataset to assess variability in male and female chest redness both in natural conditions and while under anaesthesia. We tested the hypothesis that male chest redness is a condition-dependent signal used during rival assessment.
We predicted that, when compared to females, males would have: (1) redder chests and a larger within-individual range in chest redness under natural conditions and while under anaesthesia, (2) increased expression of genes associated with vascularization and (3) increased expression of genes associated with androgen and oestrogen regulation. were monitored for temperature, pulse and respiration every 10 min during data collection, and sedation was reversed with atipamezole (0.2 mg/kg). Individuals were monitored through recovery until they returned to their social unit. All data were collected with permis-

| Photo collection, measurement and analyses
To assess patterns in male and female chest redness in geladas, we used objective colour measurement methods for digital photo- Females do not exhibit these displays, so we did not exclude any photos from females. We only included photos taken in March and April of each year because: (1) this was the only dataset available for females, and (2) previous analyses have demonstrated that males exhibit seasonal trends in chest redness that could skew male results if selected over multiple seasons (Benítez, 2016). While the animals were under anaesthesia, chest redness was measured from photos taken at the start of anaesthetization for both males (n = 20) and females (n = 18). For a subset of individuals (n = 13 males and n = 3 females) we paired the photo taken at the start of anaesthetization with a second chest photo after a heat pack was applied to one side of the chest. We include a visualization of the within individual change in redness after temperature treatment in the supplementary material ( Figure S1), but we did not run statistical tests due to the small sample size.
For all photos, we also photographed a colour standard, the X-Rite ColorChecker Ⓡ Classic chart (hereafter, 'ColorChecker chart'), to correct for variable light conditions by adjusting the colour in the photograph to the known colour levels in the chart squares.
Although the digital camera brand and model were not consistent across all photographs, f-stop, shutter speed and white balance settings remained consistent between chart and chest photos.
JPEG format photos were analysed in Adobe Photoshop (Adobe Inc. 2022) using colour profiles in the RGB colour space created in ColorChecker Camera Calibration (v2.2.0; X-Rite Inc.) software designed for use with the ColorChecker chart. We measured redness as the Red to Green Ratio (hereafter, 'Red/Green') because the value in each RGB channel is only informative relative to values in the other channels (Bergman & Beehner, 2008). Detailed methodology and instructions for photo measurement can be found elsewhere (DeLacey et al., 2022).
To determine whether males are redder than females during natural conditions, we constructed a linear mixed-effect model (LMM) with chest redness as the outcome variable and sex as the predictor variable while including ID and camera brand as random effects (R packages lme4 (Bates et al., 2015) and lmerTest (Kunzetsova et al., 2017)). Next, to assess whether males have a larger range in chest redness during natural conditions, we ran a linear regression model with the range in chest redness within an individual (maximum R/G -minimum R/G for each individual) as the outcome variable and the interaction between sex and camera brand as the predictor variable. Lastly, to determine whether males are redder than females under anaesthesia at baseline, we ran a linear regression model with chest redness as the outcome variable and camera brand and sex as the predictor variables.

| Skin biopsy collection
We collected chest skin biopsies from 15 adult males, 15 adult females, six subadult males and two subadult females during the annual SMGRP capture-and-release campaigns. Adulthood was de- were mapped to the Macaca mulatta reference assembly Mmul_10 (Warren et al., 2020) using kallisto (v0.43.1) (Bray et al., 2016). We chose the genome of Macaca mulatta, a closely related cercopithecoid, over the Theropithecus gelada genome (Tgel_1) that our team assembled  because of the richer annotation of Mmul_10. Mmul_10 has 20% more annotated genes than Tgel_1, which is likely due to the fact that Mmul_10 was annotated using RNA-seq data from 13 tissues, while Tgel_1 was annotated using only two RNA-seq libraries derived from cultured fibroblasts. Given that our current study focuses on a more diverse population of cells from skin patches, we were concerned that relevant genes would be poorly annotated or missing in the gelada genome. Thus, to ensure that we were capturing the most relevant and complete set of chestpatch-expressed genes, we chose to map to the macaque genome.
Indeed, we found that more of our reads were assigned to an annotated gene when we used Mmul_10 compared to Tgel_1. Further, given that all of our samples were collected from the same population, and the macaque represents a shared outgroup, all samples are all equally evolutionarily distant from the macaque reference, and thus any reference genome alignment bias, if present, will be the same across all samples (i.e., both males and females).

| Read count normalization
First, we removed reads mapping to seven genes encoding ribosomal RNA subunits and haemoglobin genes to remove the influence of blood contamination on the skin biopsies. While investigating RNA integrity, we removed two samples with low RNA quality (retaining samples with RQN > mean -2 standard deviations), leaving a final sample size of 36 geladas (n = 20 males, n = 16 females). We then removed genes with low expression (median TPM < 10) for either males or females and 14 Y-chromosome genes (removing sex differences that are a product of their location on the male-specific Ychromosome), which resulted in 10,212 detectably expressed genes for our downstream analysis. We normalized read counts using the voom function in the R package limma (Ritchie et al., 2015).

| Modelling the effect of sex on gene expression
We used a mixed modelling approach with the R package EMMREML to quantify the effect of sex on gene expression while controlling for sample collection year, RNA extraction date, RNA concentration and RNA quality (Akdemir & Godfrey, 2015). We used an identity matrix as the known covariance structure which is required for the EMMREML mixed modelling approach. To focus on the gene expression of this putatively sexually selected trait, we only included adults in this model (n = 14 adult males, n = 14 adult females). We calculated the false discovery rate (FDR) for each gene using the R package qvalue (Storey et al., 2022). Genes that passed a threshold of a FDR of 20% were considered differentially expressed between the sexes: 'male-biased' if they were more highly expressed in males and 'female-biased' if they were more highly expressed in females. We then Z-transformed expression values for this set of male-biased and female-biased genes across all 36 individuals (n = 14 adult males, n = 14 adult females, n = 6 subadult males, n = 2 subadult females) and averaged the Z-transformed expression levels of these genes per individual to obtain a composite sex-biased gene expression score for each individual. Finally, we ran a linear regression model with expression score as the outcome variable and age category (adult male, adult female, subadult male, subadult female) as the predictor variable to investigate whether subadults differed from adults in sexbiased gene expression.

| Enrichment analyses
To test our prediction that males would exhibit increased expression of genes associated with vascularization, we conducted gene ontology (GO) enrichment analyses using the R package topGO to identify biological processes that were enriched in genes differentially expressed between males and females (Alexa & Rahnenfuhrer, 2020).
We searched for terms related to angiogenesis, the formation of new blood vessels, including the phrase 'angio' and removed those not involved in angiogenesis (i.e. 'lymphangiogenesis'). We compared the standardized effect of sex for these angiogenesis genes with the standardized effect of sex for all other detectably expressed genes with a Kolmogorov-Smirnov test.
To test our prediction that males would have higher expression of genes associated with oestrogen and androgen regulation, we used the Online Predicted Human Interaction Database (OPHID) to assess protein-protein interaction (PPI) networks associated with hormone receptor proteins of interest: ER , ER and AR (Brown & Jurisica, 2005). We used ENSEMBL orthology information queried through the R package biomaRt (Durinck et al., 2009) to identify oneto-one orthologs in the human genome, then filtered and reindexed our expression matrix to detectably expressed genes in the human genome. We retained a total of 8255 human genes for analysis.
Next we ran a query in OPHID to identify the set of human genes involved in PPI networks with each hormone receptor protein and used biomaRt to convert genes names from UniProt to Ensembl (The UniProt Consortium, 2021). We found the genes associated with PPI networks for ER , ER and AR and compared the standardized effect of sex for these genes with the standardized effect of sex for all other detectably expressed genes with human orthologs with a Kolmogorov-Smirnov test.

| Chest redness in male and female geladas
Under natural conditions (i.e. not anaesthetized), chest redness for males and females overlapped substantially, with males displaying only marginally redder chests than females ( = 0.11, p = .06,

| Sex differences in gene expression
We then quantified how sex was associated with chest skin gene expression while controlling for sample collection year (year was significantly associated with both the first and second principal component of gene expression; PC1: year = 0.39, p = .02; PC2: year = −0.22, p = .003, Table S1, Figure S2). Males and females differed along the first principal component of gene expression while controlling for year (PC1: sex = − 0.39, p = .01, Figure S3; PC2: sex = −0.13, p = .049). The results of the principal component analysis did not change when we removed genes located on sex chromosomes and analysed only those found on autosomes ( Figure S4).
At the level of the individual genes, we found that 10.5% of the 10,212 detectably expressed genes exhibited significant differential expression across males and females (n genes = 1068, FDR < 20%,

| Sex-biased genes involved in vascularization
In line with our prediction that males would exhibit increased expression of genes associated with vascularization, we found that genes more highly expressed in males were enriched for biological processes associated with angiogenesis (K-S Test: D = 0.19, p = 7.25 × 10 −7 ; Figure 4). Additionally, these genes in males were also enriched for biological processes associated with blood pressure regulation and blood vessel maintenance (K-S Test: D = 0.19, p = 2.59 × 10 −5 ; Figure S6). Contrary to our predictions, we did not find evidence for increased expression of genes associated with oestrogen or androgen regulation in males-including no evidence for indirect effects through protein-protein networks: there was no enrichment for genes involved in PPI networks for ER (K-S Test: D = 0.04, p = .07), ER (K-S Test: D = 0.05, p = .17), or AR (K-S Test: D = 0.04, p = .46). For subset of individuals for which we had matched skin biopsies and chest redness at baseline while under anaesthesia (n = 10 males, n = 8 females), we did not find a correlation between the average standardized expression level of genes involved in PPI networks for ER , ER or AR and chest redness within males or within females ( Figure S7).

| DISCUSS ION
We sampled chest skin biopsies from wild geladas to directly measure putative mechanisms underlying a uniquely evolved sexual signal. We found that male and female geladas showed a substantial overlap in chest redness under natural and unmanipulated conditions, but males exhibited a wider within-individual range in baseline redness under natural conditions (Figure 2a,b). Further, subadults displayed an intermediate gene-expression pattern from adult males and adult females (Figure 3). We also found sex differences in gene expression, where higher expression in males was associated with angiogenesis, blood pressure and blood vessel maintenance, suggesting that blood flow and vascularization may underlie sex differences in this sexually selected signal (Figure 4). Contrary to our predictions, genes encoding proteins that interact with androgen or oestrogen were not more highly expressed in males. Together, these results suggest that males may have more variable chest redness due to increased blood flow and blood vessel branching in the chest patch skin.
Chest photograph measurements revealed an overlap in redness between the sexes at baseline in both natural and anaesthetized conditions, but males overall exhibited a wider withinindividual range in redness. Selection on an ornamentation trait in one sex can create a correlated response in the opposite sex within a species (e.g. male and female colouration are highly correlated in passerines), suggesting that changes in one sex can be constrained by changes in the other sex (Dale et al., 2015;Poissant et al., 2010;Potti & Canal, 2011). In geladas, this overlap in chest redness between the sexes could simply be the result of a positive genetic correlation. Alternatively, as male and female colour traits function differently, female chest redness in geladas could have continued to evolve under another selective pressure (Dale et al., 2015;Tobias et al., 2012), and the overlap in redness could be caused by each sex using the chest patch to communicate different signals. In males, chest redness varies among males by status (Bergman et al., 2009)  to communicate reproductive status through hormonal and blood signalling as they have the reddest chests late in gestation when oestrogen levels and blood volume are the highest (Hytten, 1985;Roberts et al., 2017). In addition to chest colour variation, gelada females exhibit sexual swellings consisting of cutaneous vesicles surrounding the chest region where vesicle turgidity varies across the ovarian cycle, suggesting sexual swellings work in tandem with chest redness to signal a different aspect of reproductive state (Roberts et al., 2017).
The sex difference in gelada chest skin gene expression aligns with findings in humans where a wide variety of tissues exhibit small effects of sex on gene expression (Lopes-Ramos et al., 2020;Oliva et al., 2020). However, small expression changes have been shown to have large phenotypic effects, particularly in the manifestation of disease (Khramtsova et al., 2019). Within primates, sex-biased gene expression has also been detected in rank-related genes, immune regulation and ageing in blood samples collected from wild baboons (Anderson et al., 2021;Lea et al., 2018). The magnitude of sex-biased gene expression has been shown to increase across development with the greatest differences in adult tissue (Mank et al., 2010;Perry et al., 2014), and differences are particularly exaggerated in sexually dimorphic tissues such as elaborated weaponry (Zinna et al., 2018).
Subadult geladas showed an intermediate pattern of gene expression between that of adult males and females suggesting that gene expression differences increase at sexual maturity when sexually selected signals develop for mate acquisition. Further, an analysis of an avian clade that found the degree of sexual selection predicts the proportion of male-biased gene expression (Harrison et al., 2015).
This finding is consistent with gelada chest skin, as we found more genes that were more highly expressed in males compared to genes that were more highly expressed in females in this species with a high male reproductive skew.

F I G U R E 4
Genes more highly expressed in males were enriched for biological processes associated with angiogenesis D = 0.19,. Beeswarm dots represent individual genes within each subcategory of the angiogenesis gene ontology category, and the solid vertical lines reflect the average standardized effect of sex for genes for each subcategory. The solid red line represents the median standardized effect size of sex for all genes not in angiogenesis-associated categories.
Male geladas expressed genes associated with angiogenesis, blood pressure regulation and blood vessel maintenance more highly than females. The mechanism of increased blood vessel branching in the chest skin may indicate chest redness is a condition-dependent signal where the differential costs of signalling based on current body condition inhibit low-quality males from investing in the signal (Grafen, 1990;Penn & Számadó, 2020). We propose energy balance and heat loss as possible costs associated with producing a red chest. Male geladas may develop more extensive blood vessel branching in the skin compared to females through engaging in vocal displays (a behaviour females do not exhibit). Post-display chest redness increases with display rate per hour in gelada males (Benítez, 2016) which suggests that, after frequent activity has built up vascular networks, an instance of increased blood flow will prompt a larger increase in chest redness. Among males, leaders spend less time resting, more time engaging in low-intensity aggression and produce more calls per vocal display bout compared to bachelors Perlman, 2021). The physical effort required to engage in aggression and vocal displays may contribute to ensuring only high-quality males in good body condition have red chests (if it is difficult to break the link between exertion and vascularization). Further, redder chests have higher surface skin temperatures which indicate that the increased blood flow to this area may also result in heat loss in the cold, high-altitude environment of the Simien Mountains (DeLacey et al., 2022). These potential constraints could provide an avenue for chest redness to communicate current body condition to potential rivals.
Contrary to our predictions, males did not have increased expression of genes associated with androgen and oestrogen regulation in the chest skin. Although this result could simply indicate that androgen and oestrogen regulation are not important to sex differences in chest redness, it could also (1) indicate both males and females use the same androgen and oestrogen regulation pathways in the chest skin or (2) be a product of sequencing skin biopsies in particular as sex-biased genes have tissue-specific expression profiles (Lopes-Ramos et al., 2020;Yang et al., 2006). Perhaps, we would detect more sex differences in expression in brain regions involved in the regulation of hormone secretion rather than the target tissue (Becker et al., 2007). Additionally, we measured the expression of genes that interact with oestrogen and androgen receptors, but circulating hormones such as testosterone or changes in androgen receptor density may play a larger role in regulating redness in primates (Dixson, 1983;Rhodes et al., 1997;Setchell & Dixson, 2001).
As yet, no relationship has been identified between testosterone and chest redness in adult male geladas (DeLacey and Beehner, unpublished data). This may be because this putative signal is not testosterone dependent, or because we are only able to measure faecal androgen metabolite levels (capturing an averaged level of the hormone over the past day) rather than actual circulating testosterone levels. Further, we may not detect a relationship between testosterone and chest redness because oestrogens directly regulate chest redness and testosterone only indirectly influences redness through aromatization to oestrogens.

| Limitations
We only collected skin biopsies from the chest to limit the number of biopsies collected per individual during the capture-and-release.
However, comparing chest skin gene expression to gene expression in a non-sexual skin area would allow us to assess (1) whether higher male expression associated with blood flow and vascularization is unique to chest skin biopsies and (2) whether both male and female chests are hotspots for expression of androgen and oestrogen-related genes. While we were able to identify interesting sex differences, we were not able to assess gene expression differences between male status categories because anaesthetizing a leader male to collect a biopsy sample would disrupt the reproductive unit and potentially put animals at risk. Lastly, we restricted our male chest photo dataset to only photos taken in March and April to match our more limited female chest photo dataset. This facilitated a more direct comparison between the sexes, but may have reduced our ability to identify more within-individual variation, particularly in males.

| CON CLUS ION
Here, we provide mechanistic evidence that gelada chest redness is linked to increased blood flow near the surface of the skin, specifically through increased blood vessel branching in the chest skin in males. Although we did not detect a sex difference in chest skin gene expression related to androgen and oestrogen regulation, future research into the correlation between fluctuations in faecal testosterone metabolites and chest redness within males will help determine whether another aspect of hormonal regulation is involved in chest redness signalling. Going forward, we hope to better understand whether increased angiogenesis creates a current-condition signal for males in the cold, high-altitude environment of the Ethiopian highlands by leveraging biomarkers of energy balance or heat loss.