Mirror replication of sexual facial expressions increases the success of sexual contacts in bonobos

Rapid Facial Mimicry (RFM), one of the possible predictors of emotional contagion, is defined as the rapid, involuntary and automatic replication of a facial expression. Up to now, RFM has been demonstrated in nonhuman animals exclusively during play. Since in bonobos, as in humans, socio-sexuality is a powerful tool for assessing/strengthening inter-individual relationships, we investigated RFM in this domain. Bonobos displayed silent bared-teeth (sbt, the most common facial expression during sexual contacts) more frequently after the detection of an sbt emitted by the trigger than in the no-detection condition. This is the first demonstration of the presence of RFM during sex. The occurrence of RFM was positively affected by the sex of the partners with female homo-sexual contacts being punctuated by a higher presence of RFM. At an immediate level, RFM increased the duration of homo- and hetero-sexual contacts. This finding suggests that RFM can increase individuals’ potential fitness benefits. By prolonging their sexual contacts, females can strengthen their social relationships thus increasing the probability to obtain priority over resources (RFM indirect fitness benefits). Via longer copulations, males can increase the probability to make females pregnant (RFM direct fitness benefits). In conclusion, in bonobos the access to the partner’s face during sexual contacts (face-to-face, proximate factor) and the role of socio-sexuality in increasing the individual direct and indirect fitness (ultimate factor) could have favoured the evolution of specific sexual facial expressions and their rapid mirror replication. Our findings on bonobos expand the role of RFM well beyond the animal play domain thus opening new scenarios for future comparative studies exploring the evolution of socio-sexuality in humans.

Testing the occurrence of Rapid Facial Mimicry (RFM). The aim of this analysis is to verify if the detection of the facial expression sbt is able to evoke a mirror response in the receiver (RFM). We found that the rates of sbt performed in detection condition (direct visual contact between the trigger and the observer, frontal view in Fig. 2) were significantly higher than the rates of sbt performed in no-detection condition (no direct visual contact between the trigger and the observer, blind zone in Fig. 2) only in the first second after the detection of the stimulus (Randomization paired sample t-test, Bonferroni correction p = 0.01; N receivers = 9; t 1sec = − 5.446, p 1sec = 0.003; t 2sec = − 2.472, p 2sec = 0.043; t 3sec = − 0.754, p 3sec = 0.464; t 4sec = − 0.058, p 4sec = 1.00; t 5sec = − 1.547, p 5sec = 0.186; Fig. 3). Only subjects with at least 5 detected and no-detected stimuli were included in this analysis. These results confirm the presence of RFM (defined as a rapid, automatic replication of others' facial expressions) during bonobo sexual interactions within 1 s after the detection of sbt performed by the trigger. The mean latency time of the response was 42.0 csec ± 4.0 SE. Only the sbt events occurring within 1 s from the triggering sbt stimulus were included in the following analyses.
Variables affecting the occurrence of RFM. The model built to analyze the variables (for the definition see Table 1 and the Method section) possibly affecting the absence/presence of RFM (absence = sbt perceived but not mimicked by the observer; presence = sbt perceived and mimicked by the observer within 1 s) significantly differed from the null model (χ 2 = 21.222, df = 9, p = 0.002). We found that the only factor with a significant effect on the occurrence of RFM was the type of sexual contact ( Table 2). The pairwise comparisons via Randomization two independent test revealed that RFM occurred more frequently during ventro-ventral genito-genital rubbing (VVGGR) compared to the other two types of sexual contacts, ventro-ventral mounting copulations  Scheme illustrating the criterion used to evaluate the attentional state of the observer in relation to the detection of sexual bared-teeth. Only when the observer was in front of the trigger (direct visual condition) we considered the stimulus as detected. All cases of the lateral views were discarded from the analysis. www.nature.com/scientificreports/   Variables affecting the duration of sexual contacts. When analyzing the duration of the sexual contacts (LMM), the full model significantly differed from the null model (χ 2 = 1999.837, df = 13, p = 0.0005). For the definition of the fixed and random variables see Table 3 and the Method section. We found that the only factor with a significant effect on duration of sexual contacts was the variable SBT_not perceived_perceived_mimicked ( Table 4). The category not perceived included the sexual contacts with at least one silent bared teeth (sbt) not perceived by the observer, the category perceived involved the sexual contacts with at least one sbt perceived but not mimicked by the observer, and the category mimicked included sexual contacts with at least one sbt perceived and mimicked by the observer within 1 s (for details see Methods). The pairwise comparisons via Randomization two independent test revealed that the duration of the sexual contacts significantly varied across the three categories (perceived > not perceived; t = -2.113; p = 0.036; mimicked > perceived; t = 2.225; p = 0.027; mimicked > not perceived; t = -4.907; p = 0.0001) (Fig. 5).

Discussion
In bonobos, as well as in humans, socio-sexuality is a powerful tool for assessing and strengthening social bonds between group members 28,35 . The habitual use of sex in the social domain is accompanied by a corresponding complexity in communicative behaviour 47,48 . For example, a recent study has evidenced that people, despite the presence of some cultural variants, produce specific facial expressions of sexual pleasure and arousal 30 . In bonobos, as it occurs in humans 27 , we found that sexual contacts can be punctuated by different facial displays even  www.nature.com/scientificreports/ though the silent bared-teeth (sbt) is the most representative being performed in more than half of the contacts recorded ( Fig. 1). The frequent presence of sbt during sexual contacts is in agreement with the data reported by de Waal 44 for the San Diego bonobo colony. Certainly, independently from the expression performed, the facial expressions of sexual arousal are favoured by face-to-face interactions between partners, a characteristic shared by Pan paniscus and Homo sapiens 49,50 . Face-to-face engagement may serve to maintain eye contact between partners also when the sexual interaction is performed dorso-ventrally. Indeed, in various monkey species, females frequently look back at males during dorso-ventral copulations actively searching for their gazing 47 . The face-toface interaction is the prerequisite for the perception of others' facial expressions and their potential replication. Accordingly, we found that bonobos, after engaging in a face-to-face interaction, can mimic the sbt performed by the sexual partner within the first second after the detection of the stimulus (Prediction 1 supported; Fig. 3). This is the first demonstration of Rapid Facial Mimicry (RFM) during socio-sexual contacts in primates. This finding supports the hypothesis that sbt can have an arousal communicative valence (i.e., enjoyment or pleasure), as this facial expression promptly resonates in the face of the sexual partner. Indeed, the involuntary replication of others' facial expression can evoke in the observer the same emotional state related to that facial expression 11 . The relationship quality measured by grooming did not significantly affect the occurrence of RFM in bonobos (Prediction 2 not supported). The variable which had a significant effect on the occurrence of RFM was the type of sexual contact. In particular, female homo-sexual ventro-ventral genito-genital rubbing (VVGGR) had a stronger positive effect compared to hetero-sexual ventro-ventral (VVMoCo) and dorso-ventral mounting/ copulation (DVMoCo) (Fig. 4). Our finding suggests that it is not the ventro-ventral position which affects the presence of the RFM phenomenon, but rather the involvement of two females in the interaction. Probably, during their VVGGR events, females experience a higher level of emotional state matching (e.g., arousal) compared to hetero-sexual contacts. This is in line with the data on the importance of sex in unrelated bonobo females (female exogamy), who engage in homo-sexual behaviour to form alliances, based on the development and maintenance of high quality social relationships 39,51-54 . Table 4. Influence of the fixed factors on the duration of the sexual contacts (in seconds, dependent variable). VVMoCo = Ventro-Ventral Mounting Copulations; DVMoCo = Dorso-Ventral Mounting Copulations. VVGGR = Ventro-Ventral Genito-Genital Rubbing. The significant P values are in bold. a Not shown as not having a meaningful interpretation. b Estimate ± SE refer to the difference of the response between the reported level of this categorical predictor and the reference category of the same predictor. c These predictors were dummy coded, with the "VVGGR", "Not perceived", "pre-feeding" being the reference categories.  www.nature.com/scientificreports/ The results from the first model (GLMM) suggest that the occurrence of RFM is not a byproduct of the duration of the sexual contact. However, the presence of the mimicking response of sbt prolonged the duration of the sexual interactions compared to the presence of sbt not perceived or perceived but not mimicked by the observer (Prediction 3 supported; Fig. 5). It is likely that through the automatic replication of sbt, the sexual arousal experienced by the trigger can be shared with the partner. In this view, it seems that sbt can fairly communicate an emotional engagement necessary to prolong the sexual interaction. A similar immediate effect of RFM on the duration of the interaction has been found in the play domain. In chimpanzees, gorillas 55 , geladas 56,57 , Tonkean macaques 58 , domestic dogs 18 and meerkats 59 , the rapid and unconscious replication of the play faces significantly prolonged the playful sessions. In bonobos, during sex RFM could play the immediate role of synchronizing/ coordinating the motor actions between partners and, thereby, facilitating their reciprocal involvement. The arousal and motor reciprocal engagement can translate into prolonged and successful homo-and hetero-sexual interactions.
This is the first evidence of a connection between RFM and potential direct and indirect fitness benefits. On one hand, through prolonged VVGGR, females can strengthen their social relationships thus increasing the probability to obtain priority over resources (indirect fitness benefits of RFM). On the other hand, via longer copulations, males can increase the probability to make females pregnant (direct fitness benefits of RFM), also considering that prolonged copulations are good indicators of ejaculation in primates 60 .
In conclusion, in bonobos the possibility to have access to the partner's face during sexual contacts (face-toface, proximate factor) and the important role of socio-sexuality in increasing the individual direct and indirect fitness (ultimate factor) could have favoured the evolution of a specific sexual facial expression, the silent bared teeth, and its automatic mirror replication, one of the processes possibly at the basis of emotional contagion. Our findings on bonobo sexuality expand the role of RFM well beyond the animal play domain thus opening new scenarios for future comparative studies exploring the evolution of socio-sexuality in humans.

Methods
Ethic statement. The Wilhelma Zoo (Stuttgart, Germany) gave the permission for the data gathering. The research complied with current laws of Germany, Italy, and the European Community. The University of Pisa (OPBA committee) waived the need for a permit since the study was purely observational with no manipulation of animals whatsoever. All the observational methods were carried out in accordance with relevant guidelines and regulations.
The study group. The bonobo group was hosted at the Wilhelma Zoo (Stuttgart, Germany) and was composed by 16 individuals (Table S1). The colony was separated into two sub-groups labelled as "Right Group" (RGr) and "Left Group" (LGr) in relation to the indoor facility they occupied. The composition of both subgroups changed once per month in order to maintain group fluidity. During the observation period, we followed six different sub-groups (RGr 1 , RGr 2 , RGr 3 , LGr 1 , LGr 2 , LGr 3 ; Table S2). Each sub-group occupied an enclosure of about 350 m 2 and the two sub-groups were in both visual and auditory contact through the glass parts of the dividing wall. Although platforms, ropes, trunk, straw and other environmental enrichments were present in each enclosure, they did not preclude good visibility to the observers. The bonobos were fed three times per day (08:45 a.m., 11:30 a.m., 03:00 p.m.) and received fruit, vegetables and yogurt. The water was available ad libitum. Data collection. We collected data from November 2017 to March 2018. The observations took place daily over a 6-h period, spanning morning and afternoon and including feeding times. Before commencing systematic data collection, the two observers (G.A., M.B.) underwent 30 h of training to become skilled in individual and behavioural pattern identification (grooming and sexual contacts). During the training period, the two observers collected data simultaneously and the behavioural patterns were compared. Training ended when the observations produced a Cohen's K = 0.85 61 .
Via scan animal sampling 62 we collected grooming every 10 min for all animals in each sub-group. We registered the interacting subjects and the directionality of grooming. This yielded 288 h of observation for LGr 1 , LGr 2 , LGr 3 and 309 h for RGr 1 , RGr 2 , RGr 3 .
Data on aggression were collected via all occurrences sampling method 62 . For each agonistic event, we recorded the identity of the winner (the subject that never displayed submissive and/or fear patterns) and the loser (the subject that displayed submissive and/or fear patterns). We discarded from the analysis all the conflicts without a clear outcome. All occurrences sampling method was also used to obtain detailed data on socio-sexual behaviours, thus all instances of sexual invitations and contacts occurring during the observation period were recorded. Sexual invitations are different in males and females. Males generally sit and slap the feet on the ground while maintaining their legs opened. They can swing their torsos while showing their erect penis. Females usually walk slowly in front of males while watching back towards them. Then, they can lay down or crouch in front of males 50 . Due to the presence of sexual invitations, sexual contacts in bonobos were easily predicted and the observers could easily anticipate the forthcoming bout. This permitted observers to turn on the camera well before the beginning of the sexual contact. Moreover, when sexual contacts occurred repeatedly and involved different subjects, the camera worked in continuum to avoid losing interactions. The good visibility due to the absence of any visual obstacles permitted yielding 944 sexual contacts (from the first to the last contact of the genitals of the two partners) including all the subjects of the colony, including immatures.
The videotaped sequences including only the 10 adults were then analysed and coded using the program VideoLAN Client 2. Demonstrating the presence of RFM phenomenon. To verify the occurrence of Rapid Facial Mimicry (RFM) during sexual interactions, we distinguished two different conditions: detection and no-detection (Fig. 2). In the detection condition, we recorded the number of silent bared-teeth (sbt) performed by the subject when the face of the trigger (defined as the first individual who emitted the sbt stimulus) was directed towards the face of the receiver (direct visual contact condition, detection of the sbt stimulus). The subjects could engage in direct visual contact during both the ventro-ventral and dorso-ventral sexual contacts. In the last case, the subject in the lower position (a female) turned her head to engage in a face-to-face contact with the subject occupying the upper position.
In the no-detection condition, we recorded the number of sbt performed by the subject when he/she was facing away from the trigger who previously emitted the sbt stimulus (without direct visual contact condition, no-detection of the sbt stimulus). If the partner, who was initially looking away, turned his/her face towards the trigger that was still emitting an sbt, this case was categorized as detection condition, because the partner actually perceived the stimulus. All the doubtful cases (n = 163) linked to lateral views were discarded from the analyses (Fig. 2).
Then, we divided the number of facial expressions performed in the detection and no-detection condition by the number of stimuli detected and not detected, respectively. The rates obtained for each condition were compared via randomization paired samples t-tests. The RFM latencies were measured frame-by-frame starting from the onset of the trigger sbt stimulus and ending with the onset of the observer's sbt response both in detection and no-detection condition (with 2-csec accuracy). The comparison was done sec-by-sec within a 5-s time-window. We employed Bonferroni correction to adjust for multiple comparisons.
Variables affecting the occurrence of the Rapid Facial Mimicry. To verify which variable affected the occurrence of RFM (presence/absence), we ran a General Linear Mixed Model (GLMM) by using the lme4 package 63 in R (R Core Team, 2019; version 3.6.1). In this model, the response variable was binomial (absence = each sbt event perceived but not mimicked by the observer, N = 91; presence = each sbt event perceived and mimicked by the observer within 1 s, N = 78). The prerequisite for this analysis was that the stimulus emitted by the trigger (sbt) had to be perceived by the partner. Therefore, all sbt s performed but not perceived were excluded from this analysis. The dataset used for this analysis was N = 169 (all the sbt perceived).
The following variables have been used as fixed factors ( Table 1). www.nature.com/scientificreports/ Context. We defined as "Prefeeding" the last 30-min block before the food provisioning and as "Feeding" the 30-min block starting from food provisioning. The parameter for delimiting the two periods linked to feeding activity was the usual time span necessary for complete food consumption 65 . "No-feeding" context included all the activities performed by the animals outside the "Prefeeding" and "Feeding" contexts.
Relationship quality. The quality of the relationship between the subjects forming each dyad (A-B) was determined by calculating the ratio between the number of the grooming sessions and the total number of scans in which at least one of the subjects of the dyad was present.
Duration of sexual contacts. The duration of each sexual contact was measured in seconds.
To exclude the occurrence of collinearity among the fixed factors, we examined the variance inflation factors (VIF package 66 ). Since collinearity has been found between the relationship quality and rank distance variables, this second factor was excluded from the analysis. We confirmed that the GLMM was not overdispersed (chisq = 170.298, p = 0.274, dispersion parameter = 1.06).
Variables affecting the duration of sexual contacts. To verify which variable affected the duration of the sexual contacts, we ran a Linear Mixed Model (LMM) by using the lme4 package 63 in R (R Core Team, 2019; version 3.6.1). To analyze the duration of the sexual contacts (dependent variable), we used the dataset including 305 sexual events. For this model, we verified the distribution and homogeneity of the residuals by looking at the qqplot and plotting the residuals against the fitted values (a function written by R. Mundry). The distribution of the duration of each sexual contact was normal. To exclude the occurrence of collinearity among predictors, we examined the variance inflation factors (VIF package; 66 ). No collinearity has been found between the fixed factors (Max VIF = 1.19).
The variables Type of sexual contact, Rank distance (ΔNDS), Context, Relationship quality have been defined above.
SBT_not perceived_perceived_mimicked. The category not perceived included the sexual contacts with at least one silent bared teeth (sbt), but none of the sbt s were perceived by the observer (N = 191). The category perceived involved the sexual contacts with at least one sbt perceived but not mimicked by the observer (N = 59). The category mimicked included sexual contacts with at least one sbt perceived and mimicked by the observer within 1 s (N = 55) ( Table 3). Since each sexual contact could include more than one of these categories, we treated them as mutually exclusive. If a sexual contact included both sbt not perceived and perceived but not mimicked, such event fell in the perceived category. If a sexual contact included both sbt perceived and sbt perceived/mimicked, such event fell in the mimicked category. This procedure avoided the pseudo-replication of the data.
The random factors included in both GLMM and LMM were each dyad involved in the sexual contact and the subgroups (Tables 1 and 3).
For both models, we tested the significance of the full model 67 by comparing it against a null model comprising only the random factor, by using a likelihood ratio test (Anova with argument test "Chisq" 68 ). Then, we calculated the p-values for the individual predictors based on likelihood ratio tests between the full and the respective null model by using the R-function "drop1" 69 . We performed all pairwise comparisons via the randomization procedures (two-independent sample test; Resampling Procedures 1.3, David C. Howell, Freeware; 10,000 permutations).

Data availability
All data generated or analysed during this study are included in this published article (and its Supplementary  Information files).