Sexual coercion in a natural mandrill population

Increasing evidence indicates that sexual coercion is widespread. While some coercive strategies are conspicuous, such as forced copulation or sexual harassment, less is known about the ecology and evolution of intimidation, where repeated male aggression promotes future rather than immediate mating success with targeted females. Although known in humans, intimidation was recently reported in chimpanzees (Pan troglodytes) and chacma baboons (Papio ursinus), where males are regularly violent against females. Here, we investigate the nature of male coercive strategies in wild mandrills (Mandrillus sphinx), a primate living in large polygynandrous groups where severe male aggression towards females is rare and females can form coalitions against males. Yet, we found support for all three predictions of the sexual coercion hypothesis, namely that male aggression (1) specifically targets sexually receptive females, (2) inflicts costs to these females, and (3) increases male mating success in the long-term. These results hold true when considering only non-physical threats, or only severe aggression. Finally, we show that high-ranking females are most targeted by males, probably because of their higher reproductive performances, while high-ranking males are most coercive. These results indicate that sexual intimidation is widespread in sexually dimorphic and group-living mammals, and that males and females vary in their propensities to use, and to be exposed to sexual coercion, respectively.

tematically recorded copulations of males with females (n=275). Male aggressive events towards 135 females included grasping/hitting (n=401), biting (n=18), chasing (n=65), lunging (n=383), slap-136 ping the ground (n=138) and head bobbing (n=567). For the analyses below, we ran the models 137 including all these behaviours and we also replicated the analyses using only severe aggression 138 (grasping/hitting, biting and chasing) or only threats (lunging, slapping the ground and head 139 bobbing) because both categories produce different female behavioural reactions (see discussion). 140 Dominance ranks were established separately for each sex (on a yearly basis for females and on a 141 monthly basis for males) based on avoidance and displacements and calculated using normalized 142 David's score ([46]; as per [47]). Female rank is maternally inherited and generally stable during 143 a female's life [48]. Here, females were divided into three classes of equal size (high-, medium-and 144 low-ranking) while male rank was considered as a binary variable (alpha versus non-alpha) because 145 of the distinct behavioural characteristics of the alpha male, who monopolizes most swollen females 146 and is relentlessly challenged by other males [49]. In the test for intimidation, in case the swollen 147 period spanned over two consecutive months, a male was considered as alpha if he achieved the 148 highest position for at least one of these two months.  Census data allowed to reconstitute patterns of male residency in the group. Here, we considered a male as resident in a given mating season when censused in the group late during the preceding 158 birth season, between January and March. When censused for the first time during the mating 159 season (which takes place once per year between April and September) we considered the male 160 as immigrant. For immigrant males, the first census date was the "arrival date". Each year, the 161 day of arrival of the first immigrant male in the group was considered as the onset of the mating 162 season (figure S1). 163 2.4 Female reproductive state and sex ratio 164 During each female estrous cycle, the perineal swelling inflates for several days until reaching a 165 maximal swelling size around ovulation. Swelling size remains maximal for a few days before 166 deflating within a few days. We used a scale from 0 to 3 (by increments of 0.5) to evaluate the 167 swelling size of each female on a near-daily basis. The reproductive state of each adult female was 168 also recorded on a near-daily basis. Each female was classified as: "non-swollen" (i.e. non-fertile 169 phase of the cycle that does not fall within the following three categories), "swollen" (i.e. with a 170 perineal sexual swelling), "pregnant" (i.e. with a characteristic pregnancy swelling and/or if she 171 gave birth 163-190 days afterwards (average gestation length: mean±SD: 175.0±4.7 days; [32]) or 172 "lactating" (i.e. nursing a ≤6 month-old infant without having resumed cycling). Finally, females 173 were considered as nulliparous until their first parturition, and parous afterwards. We calculated 174 monthly adult group sex ratio (SR) or group operational sex ratio (OSR) as the number of females 175 (for SR) or females with inflating sexual swelling or swelling of maximal size (for OSR) divided by 176 the number of males aged 9 yrs and above that were censused in the group that month.

Injuries
We recorded the occurrence, type of wound, freshness and body location of any injury on a near-179 daily basis on all subjects [51]. A total of 90 injuries (limping n=15, puncture of the skin n=11, 180 bleeding or swollen skin n=48, other n=16) were recorded on 43 females over the study period. For 181 most injuries, we did not witness the interaction and the cause but in the three cases with a known 182 context the injury was inflicted by a male. We never observed violent female-female aggression 183 resulting in an injury.

185
To test whether male aggression targets swollen females preferentially (first prediction), we ran 186 a binomial generalized linear mixed models (GLMMs) with a logit link function to study the re-187 lationship between the probability that a female received aggression by any (adult or subadult) 188 male during that female focal observation (0/1; response variable) and her reproductive state at 189 the time of observation (non-swollen, swollen, pregnant and lactating; for sample sizes, see table 190 S1). We further controlled for the following fixed effects: female dominance rank (high-, medium-191 or low-ranking) to test if higher-ranking females are preferentially targeted by males, parity (nul- we ran a similar model (same structure of fixed and random effects) with the response variable 197 corresponding to the probability that a female received aggression by groupmates other than adult 198 or subadult males. By doing so, we tested if swollen females were generally more targeted than 199 any other female, regardless of the age-sex group of the aggressor. To test whether swollen females were more injured than females in other states (second predic-202 tion), we ran a binomial GLMM with a logit link function to study the relationship between the 203 probability that a female got injured (observed injured for first time) on a given day (0/1; response 204 variable) and her reproductive state that same day. As above, we further controlled for the follow-  We then tested whether males who were more aggressive also had a higher mating probability with 210 their victim (third prediction). To study intimidation, we performed a binomial GLMM with a 211 logit link function to test whether the rate of aggression received by a female from a given male 212 (continuous fixed effect) before the next estrous cycle of the female increased the probability of 213 copulation of that heterosexual dyad during the female's swollen period (0/1; response variable).

214
The "aggression window" before the swollen period was defined as the time elapsed between the 215 onset of the mating season (for resident males) or a male's arrival in the group a given year (for 216 immigrant males) and until the beginning of the swollen period of the female (spanning from the 217 first day of a female's sexual swelling to the last day where swelling size was maximal: mean±SD: 218 10.6±5.1 days; figure S1). We pooled focal observations from females and males (table S1). We 219 controlled for the following fixed effects in our model: female dominance rank and parity, OSR 220 (since we focused only on swollen females for that prediction) in the month corresponding to the 221 first day of maximal swelling, male dominance rank (alpha vs. non-alpha) that same month in 222 interaction with the rate of male aggression (to test whether the aggression of alpha males had 223 a greater impact on their mating success than the aggression of subordinate males) and the total 224 focal observation time of the studied heterosexual dyad (during the swollen period of the female) 225 to control for the time of observation. Female identity, male identity and year of observation 226 were fitted as random factors. We restricted our analyses to those heterosexual dyads that were 227 observed for at least 30 minutes of focal time during the female swollen period to avoid biases 228 due to under-sampling that would prevent us from estimating reliably mating probability. How-229 ever. we validated that our results remained similar when we used slightly different thresholds 230 (25 or 35 minutes) or no threshold at all. We further ran the same model but restricting the 231 swollen period to the few days of the cycle during which the female was maximally swollen (i.e.

232
where the probability of conception is the highest; mean±SD: 2.9±2.9 days). Finally, to test for 233 immediate effects of male aggression, we ran the same model as above considering the rate of ag-234 gression received by a female from a given male during her swollen period only (figure S1, top line).

236
To test for sexual harassment, we assessed for each female and male focal observation during which 237 an aggressive event was recorded from a male to a swollen female, whether a copulation occurred or 238 not between that same heterosexual dyad in the 150 seconds following the aggression (see electronic 239 supplementary material; figure S2). To test for male punishment, we assessed for each female and 240 male focal observation during which a copulation event was recorded between a male and a swollen 241 female, whether an aggression from a different male occurred towards the copulating female in the 242 150 following seconds (figure S2; table S1).

244
We further ran GLMM with a negative binomial distribution to test whether alpha males were 245 more aggressive than subordinates during the mating season. We used as a response variable the 246 number of aggression events a male directed towards all adult females during each month of the 247 mating season (April to September). We considered only aggression towards females that were potential mating partners for males: late lactating females (during the 5th and 6th month of 249 lactation when some females have already resumed cycling; MJEC personal observation), "non-250 swollen", "swollen" and early pregnant females (during the first two months of pregnancy, since 251 males may not be able to distinguish early pregnant from "non-swollen" females). We pooled focal 252 observations from females and each given male (table S1). We included the following explanatory  We explored an alternative scenario to sexual coercion, the "aggressive male phenotype" hypothe-        We did not find support for sexual harassment and punishment. Following aggression, females  ; table 4). In addition, males were more aggressive 330 (marginally significant effect; table 4) when there were more swollen females in the group in rela-331 tion to males but male aggression did not depend on its age (table 4). Lastly, we did not find evidence for a female preference for aggressive male phenotypes, as fe- We found support for all three core predictions of the sexual coercion hypothesis in mandrills. 338 First, swollen females received significantly more male aggression than other females. Elevated    Here we report new evidence for sexual intimidation in a species where males, despite being much 458 larger than females, are not conspicuously aggressive towards them (at least from a human observer 459 perspective). The temporal uncoupling between male aggression and copulation explains why 460 sexual intimidation may have long been overlooked, while it increasingly appears influential at 461 shaping the social structure and mating system of polygynandrous mammals [20].

462
Ethics: All applicable international, national, and/or institutional guidelines for the care and use    (for sample sizes, see table S1). Significant p-values and confidence intervals that did not cross zero appear in bold. The significance of each variable was assessed using chi-square tests (Chisq), while the significance of each level of a categorical variable was evaluated against a reference level (noted 'Ref') according to whether their confidence intervals (CI) overlap or not.

Fixed Factor
Level Estimate CI 95% Chisq P-value