Sex differences in audience effects on anogenital scent marking in the red-fronted lemur

How the presence of conspecifics affects scent mark deposition remains an understudied aspect of olfactory communication, even though scent marking occurs in different social contexts. Sex differences in scent-marking behaviour are common, and sex-specific effects of the audience could therefore be expected. We investigated sex differences in intra-group audience effects on anogenital scent marking in four groups of wild red-fronted lemurs (Eulemur rufifrons) by performing focal scent-marking observations. We observed a total of 327 events divided into 223 anogenital scent-marking events and 104 pass-by events (i.e. passage without scent marking). Using a combination of generalised linear mixed models and exponential random graph models, we found that scent marking in red-fronted lemurs is associated with some behavioural flexibility linked to the composition of the audience at the time of scent deposition. In particular, our study revealed sex differences in the audience effects, with males being overall more sensitive to their audience than females. Moreover, we show that these audience effects were dependent on the relative degree of social integration of the focal individual compared to that of individuals in the audience (difference in Composite Sociality Index) as well as the strength of the dyadic affiliative relationship (rank of Dyadic Composite Sociality Index within the group). The audience effects also varied as a function of the audience radius considered. Hence, we showed that scent marking in red-fronted lemurs is associated with some behavioural flexibility linked to the composition of the audience, ascribing red-fronted lemurs’ social competence in this context.

The traditional approach of considering communication as information transfer between a sender-receiver dyad connected by a transmission channel 1 has been extended by the concept of communication networks. Indeed, in many social groups, individuals are closely spaced, and signals reach multiple individuals, including both intended and unintended receivers [2][3][4] . Unintended receivers, i.e. eavesdroppers, can exploit information to their benefit, sometimes at a cost to the sender 3,5 . Accordingly, senders may be sensitive to the presence and characteristics of receivers and may, thus, exhibit behavioural flexibility by initiating, inhibiting, or varying the rate or nature of signal deposition 6 . Such effects are defined as 'audience effects' 6,7 .
Although olfactory signals represent a main modality of communication in most mammals 5,8 , audience effects have mainly been studied for vocal and visual signals 7 . This inbalance can be explained by different reasons. First, olfactory signals are long-lasting, remaining in the environment long after the sender left the location. Hence, these signals may be perceived even in the absence of an audience at the time of their deposition. Second, historically, research on olfactory communication focused mainly on solitary species, where audience effects on the deposition of olfactory signals appeared to be less relevant 7 . However, scent signals have now been shown to be deposited in many different social contexts and are recognised as important in social species both for withinand between-group communication [9][10][11][12][13][14] . Moreover, recent frameworks highlighted the importance of selective pressures arising from the social domain on the evolution of communicative systems across all modalities 15,16 .
When considering the anogenital-marking network (exponential random graph model), overall, sex and/or sociality (i.e. DSI rank of the dyad focal-audience, difference in the CSI values of the individuals within a given dyad, CSI of the individual in the audience) had an effect on the probability of an individual to scent mark in front of another individual (full-null model comparison: 3 m chi 2 = − 697.5, df = 580, p < 0.001, p adjusted < 0.001; 5 m chi 2 = − 1263.1, df = 580, p < 0.001, p adjusted < 0.001; 10 m chi 2 = − 1982.8, df = 580, p < 0.001, p adjusted < 0.001).
In particular, there was a significant effect of the interaction between the combination of sexes and the DSI rank of the respective dyad on the probability of scent marking within the 5 m and 10 m radius (full-reduced model comparison: 5 m chi 2 = − 1412.5, df = 568, p-value = 0.003, p adjusted = 0.010; 10 m chi 2 = − 2213.5, df = 568, p-value < 0.001, p adjusted < 0.001; Table 3) but not within the 3 m radius (chi 2 = − 749.5, df = 568, p-value = 0.184, p adjusted = 0.550). Males scent marked more often in front of females with whom they had a stronger relationship (smaller DSI rank). In contrast, females scent-marked more often in front of females with whom they had a weaker relationship (higher DSI rank) (Fig. 2).
There was a significant interaction effect between the sexes and the CSI difference between individuals of the respective dyad within the 3 m and 10 m radius (full-reduced model comparison: 3 m chi 2 = − 740.8, df = 568, p-value = 0.004, p adjusted = 0.011; 10 m chi 2 = − 2225.5, df = 568, p-value < 0.001, p adjusted < 0.001; Table 3) but not within the 5 m radius (chi 2 = − 1393.5, df = 568, p-value = 0.152, p adjusted = 0.455). Females scent marked more often in front of females that were more social than themselves (i.e. when the difference in CSI was negative; Fig. 3). Females scent-marked more often when males that were as social as themselves (i.e. when the difference in CSI is small) were present in the 10 m range, but their probability to scent-mark also increased when in close proximity (< 3 m) with males that were less social than themselves (i.e. when the difference in CSI is positive; Fig. 3).
The probability for an individual to be observed marking in front of a given conspecific increased as a function of the number of times the individual was observed in the presence of this individual (3 m: b = 0.783, p < 0.001, p adjusted < 0.001; 5 m: b = 0.645, p < 0.001, p adjusted < 0.001; 10 m: b = 0.494, p < 0.001, p adjusted < 0.001; Table 3). Hence, individuals were more likely to scent mark in front of individuals that were more often in proximity to them. Moreover, there was a significant effect of the group on the probability to observe scent marking (full-null Table 2. Results of the models of the effects of audience composition within 3. 5 and 10 m radius, age, context and season on the probability that a female scent− marked when passing a marking spot.  Table 2). There was no significant effect of the mutual term after correcting for multiple testing ( Table 3). There was also no significant effect of the cyclical term after correcting for multiple testing ( Table 3).

Discussion
In this study, we investigated intra-group audience effects on anogenital scent marking in wild red-fronted lemurs. Our results indicated that scent marking in red-fronted lemurs is associated with some behavioural flexibility linked to the composition of the audience at the time of scent deposition. Moreover, our findings also showed that the nature of the audience effects differed between males and females, with males being more sensitive to their audience than females. On the intrasexual level, males were observed to scent mark significantly less often when a greater proportion of males of their group were within the 3 m radius. This observation is reinforced by the lowest anogenital scent-marking probabilities associated with the male-male category in the outputs of the exponential random graph analyses (Figs. 2 and 3). However, the effect of the proportion of males present in the audience on the probability of a male to anogenital scent-mark was detected only by trend in the 5 m radius and absent in the 10 m radius). In principle, it is possible that the individuals present in the 3 to 10 m range of the scent-marking spot were too far away to be attentive to scent mark deposition of other individuals. However, scent-marking rates were predicted by the strength of the social relationship with the individuals in these larger distance categories, suggesting that individuals even when they were farther away might still be attentive to scent-mark depositions. Indeed, the probability that a male would scent mark in front of another male decreased when these two males had a weaker social relationship (greater DSI rank). Males also tended to scent mark less often in front of males that were more social than themselves (negative values of the difference in CSI). Hence, males seem to avoid scent marking in close proximity with an increased number of males, especially if the latter are more social than themselves, and in the presence of males with whom they have weak affiliation.
Therefore, it is possible that even if there is no linear hierarchy and low aggression levels in male red-fronted lemurs 65 , the risk of physical aggression might be elevated when scent marking in close proximity. It might also be that males inhibit their scent-marking behaviour to avoid their scent mark being quickly overmarked by other males 11 . In addition, the probability of the sender receiving an aggression and/or overmarking might be higher when the male in the audience is not a close affiliate and is more central than the scent marker. Investigations on the probability of exhibiting aggression and overmarking at different distances and depending on the social value of the relationship between two individuals might help to test this prediction. Alternatively, males may give priority to other males to scent mark the spot when they are in proximity and prefer to pass by the scentmarking spot without depositing a scent mark. Hence, competition among males may result in having priority of access to these specific scent-marking spots. If the male to whom the priority would be given is at a distance of 5 or 10 m, the focal male might still have time to scent mark before its arrival. Our results indicate that priority of scent marking seems to be given to the most social males, which may also contribute towards explaining why the central males have been observed to be the ones scent marking more frequently 65 . Hence, in red-fronted  www.nature.com/scientificreports/ Females were observed to scent-mark more often in the presence of females that were more social than themselves (negative values of the difference in CSI) in the 10 m radius. This effect was not significant when looking at smaller radii, suggesting that females give more importance to the overall audience than to proximity in this context. Females were also observed to scent-mark more often in the presence, at any distance range, of females with whom they had weaker social relationships (greater DSI rank). Hence, despite the absence of hierarchical dominance instigated through overtly aggressive behaviour, females may signal their social status via scent marks. www.nature.com/scientificreports/ However, the highest scent-marking probabilities are associated with the female-female category in the outputs of the exponential random graph analyses (Figs. 2, 3), suggesting that overall, females seem to be less sensitive to the presence of females in their audience than to the presence of males.
On the intersexual level, we found no effect of the proportion of individuals of one sex present in the audience on the probability of an individual of the opposite sex to anogenital scent mark. However, females were observed to increase their scent-marking probability when in close proximity (< 3 m) with males that were less social than themselves (positive values of the difference in CSI). When considering a 10 m radius females were observed to scent mark more often when males that were as social as themselves (i.e. when the difference in CSI is small) than when males that were less social than themselves were present, suggesting that the proximity of the males has an effect on a female's decision to scent-mark. Females may generally prefer to scent mark in front of the most socially integrated males of the group but may also address their scent mark to the less integrated males when in close proximity with them (personal observations).
Males were observed to scent mark more in the presence of a female when they had a stronger relationship with that female (lower DSI rank). This effect was highly significant at 5 and 10 m, showing that the presence of such females was more important than her proximity to the focal male. Hence, males may particularly address scent-mark signals to females with whom they maintain a close relationship. This outcome is in line with earlier research suggesting that males that are involved in more social interactions with females than all other males (i.e. central males) are the ones scent marking the most 65 .
At the intersexual level, scent-marking behaviours may serve to maintain the pair-bonding, as shown in both pair-living (red-bellied lemurs Eulemur rubriventer 81 ) and group-living species (Coquerel's sifakas Propithecus coquereli 82 ). Scent-marking signals have also been suggested to be directed towards the opposite sex as a form of mate attraction (ring-tailed lemurs 83 and grey mouse lemurs 78 ). Both functions are not contradicted by our results but further research on the function of scent-marks is required.
Overall, our results indicate that males seem to be more sensitive than females to their audience when scent marking. Whereas both males and females seem to be sensitive to the audience, social facilitation of scent marking may occur in females, whereas in males social inhibition of scent marking may occur. Social facilitation and inhibition are defined respectively as an increase or decrease of the initiation, frequency or intensity of a response in the presence of other individuals [84][85][86][87] . Hence, males seem to be more constrained in the expression of scent signals and appear to adjust their scent-marking behaviour in a more fine-tuned manner to the composition of the audience than females. Less social males, which scent mark less frequently in the presence of other males, may rely primarily on the long-lasting component of the signal to advertise their social status to a future audience, thereby avoiding potential aggression from other males.
Interestingly, male genital secretions have been shown to be chemically richer than the genital secretions of females in true lemur species without overt dominance relationships 14,16 . Social constraints on signal deposition may be balanced by a more elaborate signal design in these species. Studying the flexibility of multicomponent signal usage across social contexts (audience compositions) contributes to uncovering the social features eliciting or constraining complex signal expression 15,88 . These social characteristics may, in turn, constitute social  www.nature.com/scientificreports/ pressures acting for or against the evolution of complex signalling behaviour 6,15,22,89 . Moreover, in true lemurs, diversification of means of olfactory communication covaried with the diversification of social systems, making them excellent models for comparative studies in this context 16,53 . Hence, further research combining chemical analyses with observations of scent-marking behaviour and audience effects across true lemur species are now indicated to further understand the social function of scent-marking behaviours. The term social facilitation is used both in the case when the other individuals are engaged in a similar task or behaviour (co-action) or when they are passive observers (restrictive use of the term "audience effect") 9,84-87 . In the present study, the individual scent marking did neither always observe another individual scent marking nor pass in proximity to a recently deposited scent mark. Indeed, this is the case for the first event of almost all video recordings as we started recording the individuals before observing a scent-marking event. Moreover, the 15 min duration of the focal observations, allowed for individuals or sub-groups isolated from the rest of the group to also perform behaviours without having been part of the audience of an individual recorded earlier. For these reasons, social facilitation via co-action is unlikely in the context of this study.
Social facilitation via co-action historically implies arousal-mediated mechanisms, while the 'audience effect' sometimes refers to the specific effect that an individual is being watched or thinks it is being watched 86,90 . Audience effects may indeed reveal a potential intentional communication, primarily when this variation is based on subtle social and behavioural variations, such as the quality of relationships 6,90-93 . Here we show that red-fronted lemurs do not only scent mark flexibly as a function of the proportion of males and females present in the audience but also based on the strength of the social relationship they maintain with specific individuals present in the audience. Such social competence was described as one indicator of potential intentionality in signalling behaviour 90,94 .
Finally, some caveats and limitations of our study need to be mentioned. First, some individuals may also choose not to pass a specific scent-marking spot in the presence of a particular audience. Hence, we cannot exclude and control for a potential audience effect on the probability to pass this spot or not. Second, the effect of who may have marked beforehand on a specific spot might also be relevant in an individual's choice to mark or not when passing a spot. This aspect is difficult to control in the field because we do not have information on the possible passage on this spot before the observations and video recordings started. Further studies on the patterns of scent-marking behaviour succession occurring on a given scent-marking spot may clarify these questions 8 . Moreover, considering the orientation (e.g. facing or facing away the scent-marking spot) of the individuals in the audience may also be an interesting perspective in this regard. While at 3 m, the individuals may be relatively homogeneously attentive to the scent marking of an individual, when further apart they may notice the scent-marking behaviour only when they are facing the scent-marking-spot. As a consequence, individuals approaching the scent-marking spot may indeed be more attentive than individuals that already overpassed this spot. This may also contribute to the lack of some audience effects observed at larger distances.
Besides intra-group functions, scent marking may also be a form of inter-group communication in resource or territorial defence through individual or group odour deposition 49,[95][96][97] . Female red-fronted lemurs are philopatric and remain in the territory of their mother, so they might be motivated to defend their territory and/or its associated resources. As some of the events reported here occurred in the context of post or pre-inter group encounters (with no extra-group individuals in the audience), they could have impacted our results. However, the effect of context did not influence scent mark deposition. Still, exploring in more detail inter-group audience effects may reveal interesting complementary information to understand further how red-fronted lemur flexibly adapt their behaviour to the social context.
In conclusion, we showed that scent marking in red-fronted lemurs is associated with some behavioural flexibility linked to the composition of the audience (i.e. proportion and social value of the individuals present), ascribing red-fronted lemurs social competence in this context. Moreover, our approach broadens our understanding of signal delivery and its associated sex differences in red-fronted lemurs, providing an avenue for future research addressing the question of the effect of social variation on scent-marking behaviour.

Material and methods
Study site and subjects. We conducted this study in Kirindy Forest, a dry deciduous forest located ca. 60 km north of Morondava, western Madagascar, managed within a forestry concession operated by the Centre National de Formation, d'Etudes et de Recherche en Environnement et Foresterie (CNFEREF) 98 . Since 1996, all members of a local population of red-fronted lemurs inhabiting a 80-ha study area within the forest have been regularly captured, marked with individual nylon or radio collars, and subjected to regular censuses and behavioural observations as part of a long-term study 98 . The data presented in this study were collected from May to November 2018 on 28 individuals belonging to four groups (11 females and 17 males; Table 4). Among males, 14 were adults and 3 sub-adults (1.5-2 years). Sub-adults were included in the study as they were observed to perform scent-marking behaviour as often as adult individuals. Reproduction of the species is seasonal, with a 4-week mating season in May-June and a birth season in September-October 65,99 . All applicable international, national, and/or institutional guidelines for the care and use of animals were followed. The authors complied with the ARRIVE guidelines 100  Data collection. Between May to July (later referred to as mating season) and September to November (later referred to as birth season), data were collected by focal scent mark observations 102 . Scent-marking behaviour were observed ad libitum 102 during 27 to 34 half-days in each group. During these sessions, a total of 120 www.nature.com/scientificreports/ scent-marking behaviour (26 to 34 per group) served as foci for 15 min observations that were video recorded. During these 15 min observation periods, we annotated each individual passing the focal scent-marking spot, its identity, whether it performed scent marking or not, the date, the time, the context and the identity of all the other individuals present in the radius of 3, 5 and 10 m. If an individual scent marked not directly on the original scent-marking spot but on one in close proximity, we also considered it in our analysis and took previous pass-by events on this spot into account. The context was classified using four categories: resting, feeding, travelling and disturbance defining the group activity. The context ' disturbance' referred to situations in which individuals of the group are vigilant, and none of the other three context categories could be attributed to the situation. Cases when individuals of another group were visible were excluded. Additionally, from May to November 2018, we also carried out 30 min individual focal observations in the morning between ca. 07:00-10:00 h and afternoon between14:00-17:00 h. A given individual was never observed for more than one 30  Data analyses. All analyses were carried out using R (version 3.6.0) 103 and RStudio (version 1.2-1335) 104 .
Social values of individuals. We calculated the CSI (Composite Sociality Index 70 ; Eq. (1)) for each individual based on three mutually exclusive affiliative behaviours: body contact, grooming and huddling. For each individual, we first calculated the proportion of time spent in body contact, huddling and grooming with an individual of its group (except juveniles). The resulting hourly rates for each of the three behaviours ( r.bc i , r.hu i , r.gr i ) were next divided by the respective mean rate for the group of the given individual before being summed up. To obtain the CSI, the summed value was divided by three, corresponding to the number of behaviours considered. We further attributed to each individual a CSI rank within each group and age-sex category, with individuals of rank 1 being the ones interacting the most often. To obtain the difference in CSI between two individuals we subtracted the CSI value of the individual in the audience to the CSI value of the focal. These CSI difference values were scaled using the R function 'scale' within each group.
(1) CSI i = r.bc i mean(r.bc) group + r.hu i mean(r.hu) group + r.gr i mean(r.gr) group 3  (2)) of each dyad of individuals in a given group (excluding juveniles) following the same principle as for the CSI. Because two individuals were never observed simultaneously in a given group, interaction rates for a given dyad A-B could be calculated by summing up the rates associated with A being focal and interacting with B and B being focal and interacting with A. For each individual, we first calculated the time spent in body contact, huddling and grooming with each of its adult group members and divided it by the total observation duration of this individual while its partner was present in the group. We further attributed to each dyad a DSI rank within each group and agesex category, with dyads of rank 1 being the most social dyads of their group. We used rank instead of raw DSI as we were not interested in group differences. In this way, the most social dyad of each group is attributed with the same social value.
Estimation of the audience effect on anogenital scent marking. For a given individual, we only considered anogenital marking events that occurred with a time-lapse of at least 5 min between each other. We selected passing events (without scent marking) on the same criteria. We included only individuals for whom we had at least two observations of each passing and marking. Three males that emigrated during the period had to be excluded because we had only one observation of either passing or marking. The final male dataset included 14 individuals (3 sub-adults and 11 adults) observed for 60 pass events and 105 anogenital marking events. The female dataset included 11 adult females observed for 44 pass-by and 118 scent-marking events. We first fitted two independent Generalized Linear Mixed Models (GLMM) for both sexes, estimating the influence of the audience composition on the probability of anogenital-marking behaviour to occur at a given time. These models had a binomial error structure and logit link function 105 and were run for each audience radius. These models were fitted using the function glmer of the R package lme4 (version 1.1-21) 106 with the optimiser' 'bobyqa' . As fixed effect, we included the proportions of males and adult females present in the given distance radius. To control for age (for males only as we only had one age class for females), context and season we also included these terms in the model as control predictors. Individual identity and date were included as random factors to account for individual variations and the possible effect of particular events.
To reduce the risk of type I errors 107 , we included all possible random slopes components (the proportion of males, the proportion of adult females, context and season within individual identity). We manually dummycoded and centred context, season and age, and z-transformed the proportion of males and the proportion of females before including them as random slopes. Initially, we also included all correlations among random intercepts and slopes for all models. However, for females, these were all estimated to have absolute values being essentially one indicating that they were not identifiable 108 . Hence, we removed these correlations from the female model.
As an overall test of the effect of audience composition on the probability to anogenital scent mark, we compared the full model with the null model lacking the fixed effects characterising the audience (proportion of males and proportion of females) but comprising the control fixed effects and the same random effect structure as the full model 107 . This comparison was performed using a likelihood ratio test 109 .
Model stability was assessed by comparing the estimates of the model run on the full dataset with the ones run on datasets, excluding each level of the random effects one after the other 110  Confidence intervals were derived using the function bootMer of the package lme4, using 1,000 parametric bootstraps and bootstrapping over the random effects, too (argument' use.u' set to TRUE). Tests of the individual fixed effects were derived using likelihood ratio tests 112 (R function drop1 with argument' test' set to" Chisq"). We determined the proportion of the total variance explained by the fixed effects (R 2 m ; marginal coefficient of determination), and the proportion of the variance explained by both fixed and random effects (R 2 c ; conditional coefficient of determination) following the method recommended by Nakagawa et al. 113 and using the function r.squaredGLMM of the package MuMIn (version 1.43.6) 114 . Because our models seem to suffer singularity issues, we further applied a Bayesian method as recommended by the authors of the "lme4" package 106 . This approach should allow both regularising the model via informative priors and giving estimates and credible intervals for all parameters that average over the uncertainty in the random-effects parameters. Details on the methods and outputs of these models are provided in supplementary material (Supplementary File 2).
To account for the nonindependence of individuals within a group and the network structure of their interactions, we first used valued exponential random graph models (ERGM) 115 to understand how the nature of the audience may influence the probability of anogenital marking. We implemented an ERGM based on a directional weighted matrix corresponding to the number of observed anogenital marking events of a focal individual (tail) when a given individual of its group was in the audience (head).
Models were implemented with a Poisson reference distribution, and the term "sum" corresponding to the sum of the edge weights (equivalent to an intercept in a linear modelling scenario) was added to the model. In addition, a "nonzero" term was added to control for zero inflation in the distribution of edge weights. Moreover, because structural terms are essential for correct model specification 116 www.nature.com/scientificreports/ of the minimum edge weights for each potential edge), and a cyclical weights term allowing for exploring hierarchical structure 118 . Two terms were included as control predictors: an edge covariate term to account for the amount of time an individual was observed in the presence of a given individual in the audience 119 and a nodelevel covariate term to control for the effect of the group. Moreover, an offset term was added to acknowledge the fact that we only consider intra-group interactions. The terms described so far were the terms remaining in the null-model. As node level predictors, we included the interaction between sex (only for adults) and the CSI rank of the individual in the audience (in-edges). As edge covariates, we included the interaction between the sexes and the difference between the CSI of the focal individual and the individual in the audience and the interaction between the sexes and the DSI rank corresponding to the dyad in question. All the terms corresponding to the main effects and the dummy variables (with the exception of the reference male-male) were also included in the model.
ERGMs were implemented in R using the statnet suite of packages 115,[120][121][122][123] . The code to implement this model is provided in the ESM, Supplementary File 3. We manually dummy coded and centred the sex interacting and z-transformed all the explanatory variables before including them into the model. The goodness of fit was assessed for each model by simulating 1000 networks and comparing the distribution of their coefficients to the observed coefficients 124,125 (Supplementary Figs. 1, 2 and 3). MCMC diagnostics were used to assess ERGM convergence ("mcmc.diagnostics" function in the ergm package) (Supplementary Figs. 4, 5 and 6). To assess the overall test of the significance of the interaction between sex and sociality we compared the deviance of the full model to the deviance of the null model described above. This comparison was based on a likelihood ratio-test 107,109 , R function anova with the argument test set to "chisq". To test the significance of the individual interactions between sex and the three social variables, we compared the full model's deviance with that of a corresponding reduced model not comprising this interaction. To control for multiple testing, we corrected the p-values using the p.adjust function with a Bonferroni method. Confidence intervals for the interaction effects were obtained by bootstrapping the response matrix (adding or subtracting 1 to an intra-group edge weight).
Ethics declarations. All applicable international, national, and/or institutional guidelines for the care and use of animals were followed. The authors complied with the ARRIVE guidelines 100

Data availability
The datasets generated and analysed during the current study are available from the corresponding author on reasonable request.