Behaviour (2020) DOI:10.1163/1568539X-bja10034 brill.com/beh

Look at me while having sex! Eye-to-eye contact affects homosexual behaviour in females

G. Annicchiarico a, M. Bertini a, G. Cordoni a and E. Palagi a,b,∗ a Natural History Museum, University of Pisa, Via Roma 79, 56011 Calci (Pisa), Italy b Unit of Ethology, Department of Biology, University of Pisa, Via A. Volta 6, 56126 Pisa, Italy *Corresponding author’s e-mail address: [email protected]; ORCID: https://orcid.org/0000-0002-2038-4596

Received 6 December 2019; initial decision 18 March 2020; revised 11 August 2020; accepted 11 August 2020

Abstract In , eye-to-eye contact (EEC) or mutual gazing is a reflexive predisposition occurring in intimate contexts. We investigated the role of EEC during bonobo socio-sexual contacts. Fe- males engage in homosexual ventro-ventral, genito-genital rubbing (VVGGR) during which they embrace each other while rubbing part of their vulvae and, sometimes, clitoris. VVGGR facili- tates conflict resolution, anxiety reduction and social bonding. We found that EEC was negatively affected by female bonding: the more the eye contact, the weaker the social relationship. This suggests that EEC promotes an intimate contact between the more unfamiliar subjects. More- over, VVGGRs were successfully prolonged in presence of at least one event of EEC compared to VVGGRs during which none of the partners looked towards the other or only one looked at the other’s face. EEC has been probably favoured by natural selection to enhance the cohesion between bonobo females, who can gain social power through socio-sexual contacts.

Keywords Pan paniscus, socio-sexuality, genito-genital rubbing, emotional engagement, mutual gazing.

1. Introduction Mutual gazing or Eye-to-Eye contact (EEC) interactions are widely studied in humans and are considered to be innate and unconscious predispositions that typically occur in affiliative contexts (Farroni et al., 2002). According to previous studies, EEC develops in the first months of life (Johnson et al., 1991; Batki et al., 2000; Farroni et al., 2002) and plays a crucial role in

© Koninklijke Brill NV, Leiden, 2020 DOI 10.1163/1568539X-bja10034 2 Behaviour (2020) DOI:10.1163/1568539X-bja10034 mother-infant attachment (Senju & Johnson, 2009). Furthermore, an interest- ing eye-tracking study (Senju & Csibra, 2008) demonstrated that 6-month- old infants were more likely to follow the adults’ gaze toward an object (joint ) only when this action was preceded by an EEC. These findings suggest that mutual gazing functions as an effective communicative signal from the early phases of life. In adult humans, EEC is an important source of non-verbal communi- cation, which can settle shifts in a conversation, exercise social control, or express intimacy and a feeling of closeness (Kleinke, 1986; Batki et al., 2000). Evidence has also shown that people who spent more time to- gether engage in longer and more frequent EEC during dyadic conversations (Cordell & McGahan, 2004). Data also indicate that as the duration of EEC increases, individuals judge each other more positively (Napieralski et al., 1995). In this sense, EEC seems to be strongly related to the levels of in- timacy/closeness between the interacting subjects (Argyle & Dean, 1965; Rubin, 1970; Kleinke, 1986; Iizuka, 1994), suggesting functions in promot- ing emotional connectivity. EEC could represent an automatic form of emo- tional contagion, since it requires a synchronization of the eye movements between the subjects (Prochazkova & Kret, 2017). EEC can also be related to other types of mimicry such as hand movements (Wang et al., 2011) and parent-infant synchrony (Feldman, 2012). Additional data come from neuro- psychological investigations. People suffering with (Baron-Cohen & Gillberg, 1995), psychopathy (Dadds et al., 2008) and damage to the amyg- dala region (Adolphs et al., 2005) show deficits in processing emotions and a reduced tendency to engage in EEC interactions, thus confirming the strong linkage between mutual gazing and emotional regulation. EEC is also considered a powerful tool to establish a communicative link between people in romantic contexts (Grewen et al., 2005; Schneiderman et al., 2012). In humans, during the first phases of romantic interactions, the occurrence of EEC creates intimacy between the partners compared to those interactions without eye-contact (Croes et al., 2020). Romantic partners who frequently gaze at each other also report high levels of emotional engagement (Kellerman et al., 1989). However, direct naturalistic observation of EEC during intimate and spontaneous social interactions in humans, regardless of whether during sexual or platonic contexts, is difficult due to ethical and conventional constraints. G. Annicchiarico et al. / Behaviour (2020) 3

The current study aims to identify the potential occurrence of EEC during non-conceptive sexual interactions in one of our closest relatives, the bonobo (Pan paniscus) (Savage-Rumbaugh & Wilkerson, 1978). Specifically, in this species socio-sexual behaviour plays an important role to the point of be- ing defined as the key to bonobo social life (de Waal, 1995). Sex not only serves a reproductive function, but also has important social functions. For instance, a study conducted by Paoli and colleagues (2007) demonstrated that in high arousal contexts (i.e. feeding), displayed non-reproductive sexual behaviours, such as mounting and homo-sexual contacts, more fre- quently than reproductive ones. In this sense, socio-sexual contacts can be considered a powerful mechanism to reduce social tension in this species. The tension reduction hypothesis has also been demonstrated by Clay & de Waal (2015) who found that socio-sexual interactions were used to reconcile and console. Therefore, non-reproductive sex in bonobos seems to play a role in promoting social bonds and cooperation (de Waal, 1989; Wrangham, 1993; Hohmann & Fruth, 2000; Furuichi, 2011; Moscovice et al., 2019). Although socio-sexual behaviour occurs in all age and sex combinations, it seems to be particularly fruitful for adolescent and adult females (Kuroda, 1980; Hohmann & Fruth, 2000), who represent the decisional and emo- tional core of bonobo social groups (Furuichi, 2011; Demuru & Palagi, 2012; Tokuyama & Furuichi, 2016). Due to their limited relatedness (female exogamy), female bonobos adopt homosexual behaviours to create strong alliances, based on the development and preservation of close social bonds (Kano, 1992; de Waal, 1995; Parish, 1996; Gruber & Clay, 2016; Moscovice et al., 2019). Ventro-Ventral Genito-Genital Rubbing (VVGGR) is the most frequent type of sexual contact between females (Kano, 1992; Hohmann & Fruth, 2000). It consists in embracing each other while the two subjects are rubbing part of their vulvae and, sometimes, clitoris (Kuroda, 1980; de Waal & Lanting, 1997). This ventral position increases the coordination of body movements to facilitate genital stimulation (Kano, 1992; Moscovice et al., 2019). VVGGR often involves face-to-face interaction, which is a phenomenon not yet reported outside sexual context in this species (Savage- Rumbaugh & Wilkerson, 1978). In bonobos, qualitative observations indi- cate that the success of the sexual interaction, inferred by its duration, can be associated with the maintenance of mutual gazing during sexual contacts (Savage-Rumbaugh & Wilkerson, 1978). 4 Behaviour (2020) DOI:10.1163/1568539X-bja10034

In bonobos eye-fixation seems to be a common phenomenon that favours affiliative behaviour and social tolerance (Hare & Woods, 2017; Harrod et al., 2020). In this regard, in a recent study, Kano and colleagues (2015) showed some pictures to bonobos and representing conspecific and allospecific faces. Chimpanzees were more interested in looking at the genital areas and mouth, whereas bonobos paid more attention to the faces, by fixing the eyes with a very short latency and for longer periods. Kret and colleagues (2016) evaluated the attentional bias towards conspecifics’ emotions and showed that bonobos were specifically attracted by images showing conspecifics while grooming and engaging in sexual interactions (e.g., positive valence emotions). For all these reasons, VVGGR in female bonobos represents a valuable social tool to test the role of EEC outside the reproductive context. Several studies indicate that in wild bonobos VVGGR generally occurs more frequently among females that are weakly rather than strongly asso- ciated (Hohmann & Fruth, 2000; Tokuyama & Furuichi, 2016; Moscovice et al., 2017). However, it is unknown if VVGGRs occurring between weakly bonded females are also characterized by the presence of EEC. If, as it occurs in our species (Argyle & Dean, 1965; Rubin, 1970; Kleinke, 1986; Iizuka, 1994), EEC in bonobos varies with the level of social closeness, we expect that EEC occurs more frequently between those females which score high level of affiliation (Prediction 1a). Alternatively, if EEC reflects the distri- bution of VVGGR and serves to form short-term social ties, we predict that EEC will be more frequent between females sharing weak rather than strong bonds and, potentially, showing higher scores in rank distance (Prediction 1b). Furthermore, if EEC represents an effective mechanism to promote emo- tional/attentional engagement thus prolonging social interactions (Cordell & McGahan, 2004), we predict that EEC can positively affect the duration of the sexual interaction thus increasing its success (Prediction 2).

2. Methods 2.1. The study group Behavioural data were collected on a colony of bonobos (Pan paniscus) housed at the Wilhelma Zoo (Stuttgart, Germany) during five months of observations (November 2017–March 2018). The colony consisted of 17 individuals belonging to all age classes. G. Annicchiarico et al. / Behaviour (2020) 5

In the current study, we focused on the socio-sexual behaviours of eight fully reproductive unrelated females (showing anogenital swellings and men- struation): six adults (>13 years) and two late-adolescents (10–12 years, Table 1) (Boose & White, 2017). Throughout the observation period, one female was lactating and the others were fully swollen. During our study period, the colony was divided into two subgroups la- belled, “group-left” (Sx) and “group-right” (Dx) in relation to the indoor facility they occupied. The subgroup composition changed almost every month in order to replicate as much as possible the fission-fusion social or- ganization observed in the wild. During the observation period, we followed six different sub-groups (DxA,DxB,DxC,SxA,SxB,SxC; Table A1 in the Appendix). Each subgroup was housed in two adjacent indoor enclosures (approx. total of 350 m2, including the off-exhibit zones) consisting of two connecting rooms for “group-left” and one unique large space for “group- right”. Although the two groups could not physically interact, they were in visual and acoustic contact. Both indoor facilities were provided with several elevated platforms, ropes and trunks to permit the movement of the animals in three dimensions. Bonobos were fed three times per day, at about 8.45 am, 11.30 am and 03.00 pm. The diet included fresh vegetables, fruits, nuts and yogurt. Specif- ically, food was scattered on the ground or concealed into plastic sticks in order to promote the enrichment. Water was available ad libitum. No stereo- typic or aberrant behaviours were ever observed in the two study groups. 2.2. Data collection Data were collected over a 6-h period from early morning to the afternoon, 6 days per week. To collect data on the relationship quality shared by subjects, we applied the scan sampling method at 10-min interval to record the frequency of the grooming bouts. Scan data were collected by speaking into a tape recorder and subsequently computer-transcribed. Via this method we gathered a total of 288 hours of scans for “group-left” and 309 hours for “group-right”. Since it has been demonstrated that ranking position can affect the dis- tribution of VVGGR in female bonobos (Clay & Zuberbühler, 2012), we also collected data on agonistic interactions via the all occurrences sampling method. For each unidirectional conflict, we registered the identity of the winner and the loser. The conflicts without clear winners and losers were 6 Behaviour (2020) DOI:10.1163/1568539X-bja10034 discarded from the data. Furthermore, before starting data collection the ob- servers underwent a training period (30 h) in order to acquire adequate skills regarding animal identification and behaviours (grooming and agonistic con- flicts). Training ended when the observations produced a Cohen’s K>0.80 for the behavioural items considered (Kaufman & Rosenthal, 2009). By using all occurrences sampling technique (Altmann, 1974) we also video-recorded all socio-sexual behaviours occurring during our observa- tion period (Digital Videocamera Panasonic HC V-180EG-K Full-HD optical zoom 90×). Due to the clear sexual invitations, sexual contacts were easily predicted and the observers were able to anticipate the forthcoming sexual interaction. This permitted the registration of each sexual contact well be- fore its beginning. When sexual contacts involving different dyads occurred in a tight sequence, the camera was never switched off to avoid losing in- teractions. This yielded 944 sexual interactions occurring between all the subjects of the colony. 321 out of 944 were ventro-ventral female homosex- ual interactions (Ventro-Ventral Genito-Genital Rubbing, VVGGR). The videotaped sequences of female homo-sexual contacts were analysed and coded using the program VideoLAN Client 2.2.1 (VLC Jump to Time). Before starting the video-analysis, the two observers (G.A., M.B.) underwent a second training that lasted 1 month (three times per week) for a total pe- riod of 60 hours (April 2018). For the following 5 months (May-September 2018), at the beginning of each week, we checked for observer reliability for the behaviours included in the study (Ventro-Ventral Genito-Genital Rub- bing, VVGGR; no-look; look-at Eye-to-Eye). The Cohen’s values were never below 0.80 for each of the behavioural patterns considered. 2.3. Operational definitions and statistics A sexual interaction began when the two females entered into mutual gen- ital contact and ended when one of them moved away. If the two subjects engaged in a second sexual interaction or, during a VVGGR event, paused and then started again after a delay of 10 s, that event was counted as new. For each sexual interaction we recorded (a) the identity of the females, (b) the duration of each VVGGR (seconds) (c) the exact duration of each gazing done and/or received by the two females (with a 2-centisecond accuracy). If the two females did not engage in any visual contact (the faces of the subjects were turned in opposite directions), we defined the condition as no- look (A|B) (Figure 1a). When only one of the females looked at the other’s G. Annicchiarico et al. / Behaviour (2020) 7

Figure 1. Drawings showing the three possible face interactions: (a) no-look, none of the interacting subjects looks at the other; (b) look-at, only one of the interacting subjects look- at the other; (c) Eye-to-Eye contact, the two interacting partners look-at each other. Credits Fosca Mastrandrea. face in a non-reciprocity condition (the face of one subject was turned away), we categorized the pattern as look-at (A→B) (Figure 1b). If the look-at was reciprocated with the two females looking into each other eyes, the be- haviour was defined as Eye-to-Eye (EEC) (A↔B) (Figure 1c). A look-at event started from the first frame in which one of the two females looked at the other and ended with the first frame at which she interrupted the visual contact. A look-at could become an EEC interaction only if the second part- ner reciprocated the gaze. EEC events started from the first frame in which both females looked into each other eyes and ended with the first partner looking away. We considered these conditions (no-look, look-at, EEC) as mutually exclusive. If the VVGGR included both no-look and look-at, such event fell into the look-at condition. If the VVGGR included both look-at and EEC, such event fell into the EEC condition. This procedure avoided any pseudo-replication of the data. For each of these interactions, the eyes of the subjects always had to be visible to the observers. All doubtful cases were discarded from the analyses (N = 102). Since the dominance hierarchy seems to have an important role in regu- lating bonobo sexual interactions (Hohman & Fruth, 2004), we included the rank distance as a fixed factor in the mixed model analysis (see below for the statistical definitions). We used the data on agonistic contacts to create dom- inance matrices. Normalized David’s scores (NDS) were calculated on the basis of a dyadic dominance index (Dij ) in which the observed proportion of wins (Pij ) is corrected for the chance occurrence of the observed outcome. This value was calculated on the basis of a binomial distribution with each subject having an equal chance of winning or losing in every dominance encounter (de Vries et al., 2006). This correction is necessary when, as in 8 Behaviour (2020) DOI:10.1163/1568539X-bja10034 the case of our study group, interaction frequency greatly differed among dyads. We determined the NDS-based hierarchy by ranking the individuals according to their NDSs. The NDS for each interacting dyad were then calculated. Due to the importance of the audience effect in modulating the commu- nication in bonobos (Clay & Zuberbuhler, 2012), in both GLMM and LMM analyses we included as fixed factor the subgroup composition (DxA,DxB, DxC,SxA,SxB,SxC; Table A1 in the Appendix). The hourly frequency of grooming was used to assess the quality of re- lationship between females and was included as fixed factor in the mixed model analyses due to its importance in establishing social bonds, especially among females (Palagi et al., 2004; Furuichi, 2011). 2.4. Statistical model analysis

To evaluate the frequency of no-look, look-at, EEC per VVGGR contact at an individual level, we applied the Friedman Exact test. The Bonferroni- Dunnett Post-Hoc test was employed for the multiple pairwise comparisons. To verify which variables could affect the occurrence of EEC and the duration of VVGGR we ran Generalized Linear Mixed Models (GLMM) and Linear Mixed Models (LMM), respectively. GLMM is an extension of LMM to analyse non-normal distribution. In the GLMM analysis, the dependent variable was multinomial and included three cases (no-look, A|B; look-at, A→B; EEC, A↔B). In the LMM, the dependent variable was normally distributed (VVGGR duration in seconds; Anderson-Darling test, ns, EasyFit 5.5 Professional Program, www.mathwave.com). The distribution of EEC and duration of VVGGR can be affected by several variables. Thus, in addition to grooming and rank distance (NDS), other variables were included as fixed factors in both analyses (GLMM, LMM). The following fixed factors were included in both models: grooming, subgroup, NDS. For the GLMM, we included as fixed factor the VVGGR duration (seconds). For the LMM, we included as fixed factor the variables “no-look, look-at, EEC” and the presence/absence of the interference by a third subject during an ongoing sexual contact (interference). For both models, the interaction of the identity (ID) of the subjects involved in the sexual contact with their relative position (upper, UP and lower, LOW) was included as random factor (IDUP × IDLOW). G. Annicchiarico et al. / Behaviour (2020) 9

In the GLMM analysis, we tested the combinations of the fixed variables that helped understand the role of grooming in the occurrence of the EEC phenomenon (no-look, look-at, EEC) during VVGGR interactions. In the LMM analysis, we tested the combinations of the fixed variables that helped understand whether the duration (in seconds) of a VVGGR event was affected by the presence of EEC. To select the best model, we used the Akaike’s Corrected Information cri- terion (AICc, for small sample sizes). We calculated the difference (AICc) between the AICC value of the best model and the AICC value for each of the other models. Models with AICc values less than two are considered to be essentially as good as the best model and models with AICc up to seven should not be discounted (Burnham et al., 2011). Moreover, to assess the relative strength of each candidate model, we employed AICc to cal- culate the evidence ratio and the Akaike weight (wi). The wi (ranging from 0 to 1) is the weight of evidence or probability that a given model is the best model, taking into account the data and set of candidate models (Symonds & Moussalli, 2011). The evidence ratio of the model weights is calculated by dividing the wi of the top-ranked model by the wj of the other models considered (wi/wj). For the non-parametric analyses and GLMM/LMM models we used the SPSS Statistic 20.0. When some variables were significant, pairwise comparisons were run using randomization procedures to avoid pseudo-replication due to the non- independence of data. The randomization tests were employed with 10,000 permutations using the software Resampling Procedures 1.3 (David C. How- ell, freeware; www.uvm.edu/~dhowell/StatPages/Resampling/Resampling Package.zip).

3. Results 3.1. Descriptive results Analysis at the individual level showed that the frequency of EEC, look- at, and no-look significantly differed during VVGGR (Friedman Exact test; χ 2 = 13.07, N = 8, df = 2, p = 0.001). The post-hoc tests revealed that EEC (number of EEC per VVGGR event: mean ± SD 0.65 ± 0.17) was significantly more frequent than no-look (Bonferroni-Dunnett Post-Hoc test: q =−1.63, p = 0.003; number of no-look per VVGGR event: mean ± 10 Behaviour (2020) DOI:10.1163/1568539X-bja10034

Table 1. Identity of individuals and sex (M = male; F = female), kinship, age classes (Ad = adult; Adol = adolescent; Imm = immature) and date of birth.

Individuals Kinship Age Birth No. No. No. EEC and sex class (year) no-look look-at per per per VVGGR VVGGR VVGGR event event event ∗ Banbo (F) Yanola mother Ad 2002 0.21 0.05 0.74 Bobali (M) Imm 2013 – – – ∗ Chimba (F) Koju mother Ad 1995 0 0 1 ∗ Chipita (F) Kasai mother Ad 1991 0.20 0.20 0.60 ∗ Fimi (F) Adol 2008 0 0.44 0.56 ∗ Haiba (F) Ad 2001 0.08 0.46 0.47 ∗ Huenda (F) Makasi mother Adol 2006 0.28 0.08 0.64 Kasai (M) Chipita son Ad 2004 – – – Kolela (F) Liboso Imm 2016 – – – daughter ∗ Kombote (F) Ad 1966 0.03 0.23 0.74 Koju (M) Chimba son Imm 2017 – – – ∗ Liboso (F) Kolela, Lubao Ad 1997 0.08 0.46 0.47 mother Lubao (M) Liboso son Imm 2013 – – – Makasi (M) Huenda son Imm 2015 – – – Mobikisi (M) Ad 1980 – – – Yanola (F) Banbo Imm 2016 – – – daughter

All the adult and adolescent females were unrelated. For the definition of age classes see the Methods. An asterisk (∗) indicates the subjects on whom the analysis has been performed and that occupied the upper position in the dyad during the VVGGR.

SD 0.11 ±0.11) and look-at (q =−1.38, p = 0.018; number of look-at per VVGGR event: mean ± SD 0.24 ± 0.19). No difference was found between the number of no-look and look-at per VVGGR event (q =−0.25, p = 1.00). See Table 1 for individual descriptive statistics. At a dyadic level (a → b + b →a = one dyad) we counted: three dyads always engaging in EEC during their VVGGR contacts, two dyads never engaging in EEC, three dyads engaging in both EEC and no-look, three dyads engaging in EEC and look-at, and four dyads engaging in no-look, look-at and EEC. G. Annicchiarico et al. / Behaviour (2020) 11

Table 2. The values of AICc, wi and evident ratio of each model tested in the GLMM analysis (occurrence of Eye-to-Eye contact = dependent variable, binomial distribution).

Models AICc wi Evidence ratio gr 1758.973 0.876 ∗ intercept (null model) 1763.183 0.107 8.207 gr, NDS 1767.137 0.015 59.264 NDS 1770.237 0.003 279.220 gr, duration 1803.985 1.473E-10 5 946 091 798 gr, duration, NDS 1811.788 2.976E-12 2.942E+11 duration 1815.001 5.970E-13 1.467E+12 duration, NDS 1821.754 2.040E-14 4.293E+13 subgroup 2400.671 3.975E-140 2.203E+139 NDS, subgroup 2408.583 7.609E-142 1.151E+141 gr, subgroup 2409.776 4.190E-142 2.090E+141 gr, NDS, subgroup 2421.139 1.428E-144 6.131E+143 duration, subgroup 2482.171 7.978E-158 1.098E+157 gr, subgroup, duration 2483.912 3.341E-158 2.621E+157 duration, NDS, subgroup 2489.554 1.990E-159 4.402E+158 gr, duration, NDS, subgroup (full model) 2495.242 1.158E-160 7.564E+159

Legend: gr = grooming, NDS = delta Normalized David Score.

3.2. Variables affecting eye-to-eye interaction Testing which variables affected the occurrence of EEC interactions during sexual contacts, we found one best model that included only the variable “grooming” (AICc = 1758.97, wi = 0.876) with the probability of 87.60% to be the best model (Table 2). Since the null model had a AICc = 4.21 (<7, Burnham et al., 2011), it cannot be discounted. This model (AICc = 1763.18, wi = 0.107) had the 10.70% of probability to be the second best model (Table 2). The evidence ratio for the first model versus the second model is 8.21, the empirical support for the first model is 8.21 times that of the second model. In the first model, the variable grooming was significant (p = 0.010; effect size = 0.339; Table 3 and Table A2 in the Appendix). Females who groomed more were less likely to engage in EEC during VVGGR interactions (Figure 2) (Prediction 1b supported). 3.3. Variables affecting the duration of sexual contacts Testing which variables affected the duration of VVGGR interactions, we found that the full model was the best model (AICc = 1408.257, wi = 12 Behaviour (2020) DOI:10.1163/1568539X-bja10034

Table 3. Results of the best two models explaining the occurrence of Eye-to-Eye contact (no- look/look-at/Eye-to-Eye) (dependent variable = Eye-to-Eye).

Statistic df1 df2 p (effect size) Fixed variables (AICc = 1758.973) ∗∗ Grooming F = 4.716 2 215 0.010 (0.34) Random variables IDup × IDlow Z = 1.729 0.084 Fixed variables (AICc = 1763.183) Null F = 0.215 1 135 0.644 Random variables IDup × IDlow Z = 1.706 0.088

AICc = Akaike’s Corrected Information Criterion.

Figure 2. Boxplots showing the relation between the VVGGR events characterized by the three possible face interactions (no-look — look-at — Eye-to-Eye) and the hourly grooming frequency exchanged by the interacting subjects. Thick horizontal lines indicate medians; height of the boxes corresponds to interquartile range; thin horizontal lines indicate the range of observed values. The boxplot shows both ‘extreme’ and ‘mild’ outliers. Extreme outliers areanyscoremorethan3× IQR from the rest of the scores, and are indicated by stars. IQR stands for ‘interquartile range’, and is the middle 50% of the scores. Mild outliers are any score more than 1.5 × IQR from the rest of the scores and are indicated by open dots. IQR stands for ‘interquartile range’ and is the middle 50% of the scores. G. Annicchiarico et al. / Behaviour (2020) 13

Table 4. The values of AICc, wi and evident ratio of each model tested in the LMM analysis (duration of the Ventro-Ventro-Genito-Genital-Rubbing = dependent variable, normal distribution).

Models AICc wi Evidence ratio subgroup, EEC, gr, NDS, interference (full) 1408.257 0.999 ∗ subgroup, gr, NDS, interference 1422.936 0.001 1539.940 EEC, subgroup 1424.671 0.000 3666.523 subgroup, gr, NDS 1427.508 6.596E-05 15 146.110 subgroup, interference 1435.176 1.426E-06 700 464.524 subgroup 1439.465 1.670E-07 5 980 405.901 EEC, NDS, gr 1440.498 9.966E-08 10 024 062.39 gr, EEC 1443.626 2.086E-08 47 893 914.1 EEC, NDS 1449.426 1.148E-09 870 430 957.1 EEC 1452.577 2.375E-10 4 206 929 990 gr 1456.417 3.481E-11 28 695 294 742 interference 1460.822 3.848E-12 2.596E+11 NDS 1461.824 2.332E-12 4.285E+11 intercept 1465.107 4.516E-13 2.212E+12 EEC, interference 1448.371 1.945E-09 513 622 439.5 EEC, gr, NDS, interference 1435.993 9.479E-07 1 053 890.381

Legend: gr = grooming, NDS = delta Normalized David Score, EEC = Eye-to-Eye contact.

0.999) with the probability of 99.99% to be the best model. The AICc between the best and the second model was 14.679 (Table 4). In the best model, the variables no-look/look-at/EEC (mean VVGGR duration ± SD: no-look27vvggr = 10.91 ± 3.11; look-at53vvggr = 13.42 ± 2.62; EEC139vvggr = 15.15 ± 2.25; Figure 3) and subgroup (Figure 4) were significant (Table 5 and Table A3 in the Appendix). The presence of EEC significantly increased the duration of VVGGR contacts (Two independent sample randomization t-test — Bonferroni correction; no-look vs look-at: t =−2.016, p = 0.053, effect size = 0.229; no-look vs EEC: t =−4.295, p = 0.0001, effect size = 0.448; look-at vs EEC: t =−2.395, p = 0.015, effect size = 0.184) (Predic- tion 2 supported).

4. Discussion Our study provides the first quantitative analysis of the presence and poten- tial functions of EEC during genito-genital rubbing in female bonobos. As predicted, the occurrence of EEC was primarily affected by the degree of 14 Behaviour (2020) DOI:10.1163/1568539X-bja10034

Figure 3. Duration of the Ventro-Ventral Genito-Genital Rubbing (VVGGR) in seconds as a function of the three possible face interactions (no-look — look-at — Eye-to-Eye).

Figure 4. Duration of the Ventro-Ventral Genito-Genital Rubbing (VVGGR) in seconds as a function of the six different subgroup composition. The subgroups were labelled as “group- left” (Sx) and “group-right” (Dx) in relation to the indoor facility the animals occupied. We observed three group-right (DxA,DxB,DxC) and three group-left (SxA,SxB,SxC). See also Table A1 in the Appendix. G. Annicchiarico et al. / Behaviour (2020) 15

Table 5. Results of the best model explaining the duration of VVGGR (dependent variable = duration of VVGGR (s)).

F df1 df2 p Fixed Variables (AICc = 1408.257) ∗∗ Intercept 2.462 10 208 0.008 ∗ Subgroup 2.569 5 208 0.028 ∗∗ Eye-to-Eye 5.121 2 208 0.007 Grooming 0.417 1 208 0.519 NDS 0.872 1 208 0.351 Interference 2.294 1 208 0.131 Random variables IDup × IDlow Z = 7.467 0.173

AICc = Akaike’s Corrected Information Criterion. NDS = Normalized David’s Scores. social bonding between sexual partners. Other variables such as rank differ- ence (NDS values), and the duration of the sexual interactions had a very limited effect on the presence of EEC. We found that EEC interaction was more likely between those females who shared weak social bonds (measured via grooming scores) (Prediction 1b supported). Therefore, EEC in bonobos does not seem to vary with the level of intimacy and closeness as it does in humans (Argyle & Dean, 1965; Rubin, 1970; Kleinke, 1986; Iizuka, 1994). Instead, EEC seems to reflect the distribution of VVGGR reported in the wild studies showing that this sexual contact occurs more frequently among females who groom less or are less preferred associates in other contexts (Hohmann & Fruth 2000; Tokuyama et al. 2016; Moscovice et al. 2017). Our finding suggests that EEC could be an additional strategy that is enacted to overcome the low level of familiarity between the subjects, who rarely engage in reciprocal grooming exchange. An interesting study focusing on 30 species of primates has demonstrated that in large groups, when grooming involves only a limited number of indi- viduals, the exchange of gazing can represent a substitute means in creating and reinforcing social relationships, especially between those subjects that cannot groom each other (Kobayashi & Hashiya, 2011). This is in agreement with the Cooperative Eye Hypothesis (CEH), which predicts that gaz- ing has evolved to maintain cooperative behaviours (Tomasello et al., 2007). According to the CEH, in humans there is a strict linkage between gazing behaviour and the production of oxytocin, a neuropeptide strongly related 16 Behaviour (2020) DOI:10.1163/1568539X-bja10034 in promoting emotional recognition, interpersonal communication and so- cial approach (Guastella et al., 2008; Auyeung et al., 2015). Interestingly, homosexual interactions in bonobo females increase urinary oxytocin lev- els, as well (Moscovice et al., 2019). The next step could be the evaluation of the possible linkage between EEC during sexual contacts and oxytocin production in this species. The presence of EEC in dyads sharing weak bonds could be indicative of the need to improve relationship quality through sexual contacts when so- cial bonding (via grooming) has not yet developed. In this regard, it would be interesting to verify whether VVGGR, not characterized by EEC, occurs in different types of social contexts (e.g., bargaining food, coalitionary sup- port, reconciliation, consolation) compared to VVGGR in which the EEC is present. We can also hypothesize that subjects sharing lower levels of social bonding require a finer communicative exchange to interact successfully. A VVGGR can produce a certain level of arousal between the females, espe- cially when the sexual contact occurs during tense social situations (e.g., feeding context) (Paoli et al., 2007). During a competitive feeding task, Hohmann et al. (2009) observed that the socio-sexual interactions between bonobo females increased. Those females who engaged in higher levels of sexual contacts also shared a higher amount of food and showed a decrease in their salivary cortisol levels. In this view, an EEC interaction can be an ef- fective mechanism for conveying essential information about the motivation of the subjects to engage in a relaxed interaction. In this view, the presence of EEC could reduce the social tension arising from the close physical as- sociation between subjects (Wrangham, 1993). Our finding about the role of EEC in prolonging the duration of the sexual contacts seems to corroborate this hypothesis. Those sessions characterized by the presence of EEC lasted longer than those in which none of the females looked at each other or only one female looked at her partner (Figure 3, Prediction 2 supported). It seems that the degree of mutual attention is important to maintain high motivation to prolong the contact (Prochazkova & Kret, 2017). The duration of the sexual interactions also varied in relation to the other variables considered in the analysis although most of them were not statis- tically significant. In addition to the EEC, the only variable, which reached statistical significance, was the different composition of the subgroups. It is difficult to evaluate which aspects of subgroup composition may elicit longer G. Annicchiarico et al. / Behaviour (2020) 17

VVGGR interactions. Previous research has analysed the importance of “au- dience effect” during sexual interactions in bonobo females, showing that the presence of specific individuals within the group can have an effect on the communicative tactics used by animals (Clay & Zuberbuhler, 2012). We can only hypothesize that a general audience effect could also have a role in determining the duration of sexual contacts, thus suggesting that the at- tentiveness of the subject is not only focused on the sexual partner but also on the rest of the group. Socio-sexual behaviour is a competitive strategy among bonobo females who engage in such kind of interactions to create alliances and obtain support during competitive circumstances. Therefore, each female represents an important resource for the other females that com- pete to gain the maximum number of homosexual interactions (Vervaecke & van Elsacker, 2000). Quantitative analyses on the influence of subgroup composition on socio-sexual communication in bonobos are needed to draw reliable conclusions on the role of audience effect phenomenon. To sum up, EEC in female bonobos seems to be an important communica- tive tool to engage in prolonged VVGGR interactions, which can be particu- larly successful in strengthening affiliation and cooperation between females. In this species, mutual gazing behaviour, favoured by the ventro-ventral po- sition during homosexual behaviour, acquires even more importance when the interactions occur between females who do not share close relationships. As a whole, EEC could be an evolutionary trait, which has been favoured by natural selection to enhance the cohesion between the bonobo females, who gain social power and alliances through the exchange of successful socio- sexual contacts.

Acknowledgements

We wish to thank the Wilhelma Zoo (Germany) and specifically the mam- mal curator Marianne Holtkötter and the bonobo keepers for allowing and facilitating this work. We also thank Chiara Bresciani for the unconditional support; Fosca Mastrandrea for drawings and Candace Burke and Gordon Burghardt for the accurate language revision and comments. This research did not receive any grant from funding agencies in the public, commercial, or not-for-profit sectors. This research complies with the current laws of Ger- many and Italy. 18 Behaviour (2020) DOI:10.1163/1568539X-bja10034

References Adolphs, R., Gosselin, F., Buchanan, T.W., Tranel, D., Schyns, P. & Damasio, A.R. (2005). A mechanism for impaired fear recognition after amygdala damage. — Nature 433: 68. Altmann, J. (1974). Observational study of behavior: sampling methods. — Behaviour 49: 227-266. DOI:110.1163/156853974X00534. Argyle, M. & Dean, J. (1965). Eye-contact, distance and affiliation. — Sociometry 28: 289- 304. Auyeung, B., Lombardo, M.V., Heinrichs, M., Chakrabarti, B., Sule, A., Deakin, J.B., Beth- lehem, R.A.I., Dickens, L., Mooney, N., Sipple, J.A.N., Thiemann, P. & Baron-Cohen, S. (2015). Oxytocin increases eye contact during a real-time, naturalistic social interaction in males with and without autism. — Transl. Psychiatr. 5: e507. Baron-Cohen, S. & Gillberg, C. (1995). Mind blindness: an essay on autism and theory of mind. — Dev. Med. Child Neurol. 37: 1124-1124. Batki, A., Baron-Cohen, S., Wheelwright, S., Connellan, J. & Ahluwalia, J. (2000). Is there an innate gaze module? Evidence from human neonates. — Inf. Behav. Dev. 23: 223-229. Boose, K. & White, F. (2017). Harassment of adults by immatures in bonobos (Pan paniscus): testing the exploratory aggression and rank improvement hypotheses. — Primates 58: 493-504. Burnham, K.P., Anderson, D.R. & Huyvaert, K. (2011). AIC model selection and multimodel inference in behavioral ecology: some background, observations, and comparisons. — Behav. Ecol. Sociobiol. 65: 23-35. Clay, Z. & Zuberbühler, K. (2012). Communication during sex among female bonobos: effects of dominance, solicitation and audience. — Scientific Reports 2: 291. Cordell, D.M. & McGahan, J.R. (2004). Mutual gaze duration as a function of length of conversation in male-female dyads. — Psychol. Rep. 94: 109-114. Croes, E.A., Antheunis, M.L., Schouten, A.P. & Krahmer, E.J. (2020). The role of eye-contact in the development of romantic attraction: studying interactive uncertainty reduction strategies during speed-dating. — Comput. Hum. Behav. 105: 106218. Dadds, M.R., El Masry, Y., Wimalaweera, S. & Guastella, A.J. (2008). Reduced eye gaze explains “fear blindness” in childhood psychopathic traits. — J. Am. Acad. Child Adolesc. Psychiat. 47: 455-463. vJOURNAL0004G05 de Vries, H., Stevens, J.M. & Vervaecke, H. (2006). Measuring and testing the steepness of dominance hierarchies. — Anim. Behav. 71(3): 585-592. de Waal, F.B.M. (1995). Bonobo sex and society. — Sci. Am. 272: 82-88. de Waal, F.B.M. (1989). Peacemaking among primates. — Harvard University Press, Cam- bridge, MA. de Waal, F.B.M. & Lanting, F. (1997). Bonobo: the forgotten ape. — Universiry of California Press, Berkeley, CA. Demuru, E. & Palagi, E. (2012). In bonobos yawn contagion is higher among kin and friends. — PLoS ONE 7(11): e49613. Farroni, T., Csibra, G., Simion, F. & Johnson, M.H. (2002). Eye contact detection in humans from birth. — Proc. Natl. Acad. Sci. USA 99: 9602-9605. G. Annicchiarico et al. / Behaviour (2020) 19

Feldman, R. (2012). Parent–infant synchrony: a biobehavioral model of mutual influences in the formation of affiliative bonds. — Monogr. Soc. Res. Child Dev. 77: 42-51. Furuichi, T. (2011). Female contributions to the peaceful nature of bonobo society. — Evol. Anthropol. Iss. News Rev. 20: 131-142. Grewen, K.M., Girdler, S.S., Amico, J. & Light, K.C. (2005). Effects of partner support on resting oxytocin, cortisol, norepinephrine, and blood pressure before and after warm partner contact. — Psychosom. Med. 67: 531-538. Gruber, T. & Clay, Z. (2016). A comparison between bonobos and chimpanzees: a review and update. — Evol. Anthropol. Iss. News Rev. 25: 239-252. Guastella, A.J., Mitchell, P.B. & Dadds, M.R. (2008). Oxytocin increases gaze to the eye region of human faces. — Biol. Psychiat. 63: 3-5. Hare, B. & Woods, V. (2017). Cognitive comparisons of genus Pan support bonobo self- domestication. — In: Bonobos — unique in mind, brain and behavior (Hare, B. & Ya- mamoto, S., eds). Oxford University Press, Oxford, p. 214-232. Harrod, E.G., Coe, C.L. & Niedenthal, M.P. (2020). Social structure predicts eye contact tolerance in nonhuman primates: evidence from a crowd-sourcing approach. — Sci. Rep. 10: 6971. Hohmann, G. & Fruth, B. (2000). Use and function of genital contacts among female bono- bos. — Anim. Behav. 60: 107-120. Hohmann, G., Mundry, R. & Deschner, T. (2009). The relationship between socio-sexual behavior and salivary cortisol in bonobos: tests of the tension regulation hypothesis. — Am. J. Primatol. 71: 223-232. Iizuka, Y. (1994). Gaze in cooperative and competitive games. — Jpn. J. Exp. Soc. Psychol. 33: 237-242. Johnson, M.H., Dziurawiec, S., Ellis, H. & Morton, J. (1991). Newborns’ preferential track- ing of face-like stimuli and its subsequent decline. — Cognition 40: 1-19. Kano, F., Hirata, S. & Call, J. (2015). Social attention in the two species of Pan: bonobos make more eye contact than chimpanzees. — PLoS ONE 10: e0129684. Kano, T. (1992). The last ape: Pygmy behavior and ecology (Vol. 155). — Stanford University Press, Stanford, CA. Kaufman, A.B. & Rosenthal, R. (2009). Can you believe my eyes? The importance of in- terobserver reliability statistics in observations of animal behaviour. — Anim. Behav. 78: 1487-1491. Kellerman, J., Lewis, J. & Laird, J.D. (1989). Looking and loving: the effects of mutual gaze on feelings of romantic love. — J. Res. Personality 23: 145-161. Kleinke, C.L. (1986). Gaze and eye contact: a research review. — Psychol. Bull. 100: 78. Kobayashi, H. & Hashiya, K. (2011). The gaze that grooms: contribution of social factors to the evolution of primate eye morphology. — Evol. Hum. Behav. 32: 157-165. Kret, M.E., Jaasma, L., Bionda, T. & Wijnen, J.G. (2016). Bonobos (Pan paniscus)show an attentional bias toward conspecifics’ emotions. — Proc. Natl. Acad. Sci. USA 113: 3761-3766. Kuroda, S. (1980). of the pygmy chimpanzees. — Primates 21: 181-197. 20 Behaviour (2020) DOI:10.1163/1568539X-bja10034

Moscovice, L.R., Surbeck, M., Fruth, B., Hohmann, G., Jaeggi, A.V. & Deschner, T. (2019). The cooperative sex: sexual interactions among female bonobos are linked to increases in oxytocin, proximity and coalitions. — Horm. Behav. 116: 104581. Moscovice, L.R., Douglas, P.H., Martinez-Iñigo, L., Surbeck, M., Vigilant, L. & Hohmann, G. (2017). Stable and fluctuating social preferences and implications for cooperation among female bonobos at Lui Kotale, Salonga National Park, DRC. — Am. J. Phys. Anthropol. 163: 158-172. Napieralski, L.P., Brooks, C.I. & Droney, J.M. (1995). The effect of duration of eye contact on American college students’ attributions of state, trait, and test anxiety. — J. Soc. Psychol. 135(3): 273-280. Palagi, E., Paoli, T. & Tarli, S.B. (2006). Short-term benefits of play behavior and conflict prevention in Pan paniscus. — Int. J. Primatol. 27: 1257-1270. Paoli, T., Tacconi, G., Tarli, S.M.B. & Palagi, E. (2007). Influence of feeding and short-term crowding on the sexual repertoire of captive bonobos (Pan paniscus). — Ann. Zool. Fenn. 44: 81-88. Parish, A.R. (1996). Female relationships in bonobos (Pan paniscus). — Hum. Nat. 7: 61-96. Prochazkova, E. & Kret, M.E. (2017). Connecting minds and sharing emotions through mimicry: a neurocognitive model of emotional contagion. — Neurosci. Biobehav. Rev. 80: 99-114. Rubin, Z. (1970). Measurement of romantic love. — J. Personality Soc. Psychol. 16: 265. Savage-Rumbaugh, E.S. & Wilkerson, B.J. (1978). Socio-sexual behavior in Pan paniscus and Pan troglodytes: a comparative study. — J. Hum. Evol. 7: 327-344. Schneiderman, I., Zagoory-Sharon, O., Leckman, J.F. & Feldman, R. (2012). Oxytocin during the initial stages of romantic attachment: relations to couples’ interactive reciprocity. — Psychoneuroendocrinology 37: 1277-1285. Senju, A. & Csibra, G. (2008). Gaze following in human infants depends on communicative signals. — Curr. Biol. 18: 668-671. Senju, A. & Johnson, M.H. (2009). The eye contact effect: mechanisms and development. — Trends Cogn. Sci. 13: 127-134. Symonds, M.R. & Moussalli, A. (2011). A brief guide to model selection, multimodel infer- ence and model averaging in behavioural ecology using Akaike’s information criterion. — Behav. Ecol. Sociobiol. 65: 13-21. Tokuyama, N. & Furuichi, T. (2016). Do friends help each other? Patterns of female coalition formation in wild bonobos at Wamba. — Anim. Behav. 119: 27-35. Tomasello, M., Hare, B., Lehmann, H. & Call, J. (2007). Reliance on head versus eyes in the gaze following of great apes and human infants: the cooperative eye hypothesis. — J. Hum. Evol. 52: 314-320. Vervaecke, H. & Van Elsacker, L. (2000). Sexual competition in a group of captive bonobos (Pan paniscus). — Primates 41: 109-115. Wang, Y., Newport, R. & Hamilton, A.F.D.C. (2011). Eye contact enhances mimicry of intransitive hand movements. — Biol. Lett. 7: 7-10. Wrangham, R.W. (1993). The evolution of sexuality in chimpanzees and bonobos. — Hum. Nat. 4: 47-79. G. Annicchiarico et al. / Behaviour (2020) 21

Table A1. Composition of the six different sub-groups during the observation period.

DxA DxB DxC SxA SxB SxC Kasai Kasai Kasai Mobikisi Mobikisi Mobikisi Haiba Haiba Liboso Kombote Kombote Kombote Liboso Liboso Kolela Huenda Banbo Banbo Chipita Chipita Lubao Banbo Fimi Yanola Chimba Kolela Bobali Fimi Yanola Fimi Kolela Lubao Chipita Yanola Chimba Haiba Lubao Bobali Huenda Makasi Koju Bobali Huenda Makasi Makasi Chimba Koju

Dx = group-right; Sx = group-left.

Table A2. Best GLMM explaining the occurrence of Eye-to-Eye (No-look/Look-at Eye-to-Eye) (depen- dent variable = Eye-to-Eye).

Co SE tp 95% CI

Fixed Variables (AICc = 1758.973) Grooming 23.170 7.695 3.011 0.003 8.001/38.338 Intercept −1.842 0.352 −5.230 0.000 −2.536/−1.148 Fixed Variable (AICc = 1763.183) −1.198 0.299 −4.007 0.0001 −1.788/−0.609 Intercept

AICc = Akaike’s Corrected Information Criterion. 22 Behaviour (2020) DOI:10.1163/1568539X-bja10034

Table A3. Best LMM explaining the duration of VVGGR (dependent variable = duration of VVGGR (s)).

Co SE tp 95% CI

Fixed Variables (AICc = 1408.257) ∗∗∗ Intercept 16.796 1.681 9.992 0.001 13.482/20.110 Subgroup DxA 3.248 2.267 1.433 0.153 −1.221/7.711 Subgroup DxB 1.435 2.129 0.674 0.501 −2.763/5.632 Subgroup DxC −4.289 2.963 −1.447 0.149 −10.131/1.554 Subgroup SxA −2.074 1.523 −1.361 0.175 −5.076/0.929 Subgroup SxB −2.781 1.424 −1.952 0.052 −5.589/0.027 a Subgroup SxC 0 –– – – ∗∗ no-look −4.359 1.467 −2.972 0.003 −7.252/−1.467 look-at −1.851 1.148 −1.613 0.108 −4.115/0.412 EEC 0a –– – – grooming 18.378 28.467 0.646 0.519 −37.742/74.499 NDS 1.322 1.416 0.934 0.351 −1.469/4.113 Interference 0 −1.699 1.122 −1.514 0.131 −3.910/0.513 Interference 1 0a –– – –

AICc = Akaike’s Corrected Information Criterion. NDS = Normalized David’s Scores. a = redundant.