Social proximities of developing gorilla males (Gorilla gorilla gorilla) in European zoos: The consequences of castration and social composition

In the European captive population of western lowland gorillas (Gorilla gorilla gorilla), the harem social structure and an even sex ratio at birth result in a surplus of males and consequent management difficulties. This study seeks to assess the socialization differences between captive juvenile and adolescent male gorillas according to their fertility status (intact vs castrated) in different social compositions (familial vs bachelor groups), and to evaluate the suitability of castration as a management tool for the EEP gorilla population. We carried out social network analyses (SNA) to assess the “positive” proximity pattern of behaviour in 93 western lowland gorillas aged from 0 to 45 years old and housed in 11 social units (seven familial and four bachelor groups). We compared the data recorded for the 27 juvenile and adolescent (i.e. subadult and blackback) males included in our sample size. Although no differences were revealed between the intact juveniles and the castrated juveniles living in familial groups, our results showed that castrated adolescent individuals showed more cohesiveness within their familial group than their intact conspecifics in terms of their activity budget. They also displayed a “positive” proximity pattern of behaviour with all group members, including adults (silverback and females). Despite being significantly more isolated, the intact adolescent males living in bachelor groups do not differ from their castrated and intact counterparts of the same age class living in familial groups in terms of their strength of “positive” behaviour when close to group conspecifics. This effect highlights the social benefits of male-male interactions within gorilla species. Our results may be evidence that both of the management strategies compared here, i.e. bachelor groups and castration, could be appropriate socio-behavioural enrichments during juvenility and adolescence. These findings also highlight the need to continue investigations until the study subjects reach adulthood to validate and/or improve these tools for the welfare-compliant management of gorilla male surplus in captivity.


Introduction
At the beginning of 2016, the EAZA (European Association of Zoos and Aquaria) ex-situ programme for gorillas (GEEP) comprised a total population of over 475 individuals in 74 institutions. Zoos have become successful in breeding western lowland gorillas, and they have considerably improved the captive management of this species over the years by sharing experience and knowledge that can help to improve daily husbandry (Ogden and Wharton, 1997;Stoinski et al., 2001;Strong et al., 2017).
However, the balanced sex ratio that naturally occurs at birth for this species and the long life of breeding males in captivity result in a problem of surplus males (Stoinski et al., 2013;Vermeer and Devreese, 2015;Vermeer et al., 2014). This male overpopulation has been managed by the creation of bachelor gorilla groups in zoos over the last three decades, namely through the application of previously published knowledge of wild mountain gorillas (Johnstone-Scott and Johnstone-Scott, 1988;Pullen, 2005;Watts and Meder, 1996). Several factors including age, early familiarity of individuals, group composition, number of males, exhibit design, rearing experience and personality have already been identified as possible elements that play roles in the successful formation and maintenance of captive multi-male gorilla groups (Coe et al., 2009;Gold and Maple, 1994;Kuhar et al., 2006;Pullen, 2005;Stoinski et al., 2004a,b). In the European region, twenty gorilla bachelor groups were established in 17 institutions between 1995 and 2015. A backward analysis of the last two decades of bachelor group management in European institutions highlights social instabilities that could be dangerous over time, such as aggressions and injuries. However, recent studies on the North American captive gorilla population reported similar rates of contact aggression shown by males in both familial and bachelor housing conditions. The influence of young silverbacks was a strong confounding factor that increased both agonistic behaviour and the wounding rate in the bachelor groups (Leeds et al., 2015;Stoinski et al., 2013). These authors concluded that all-male groups could therefore be a long-term housing strategy offering male gorillas the benefits of socialization if appropriate social grouping prevents the risk of escalated aggression and the maintenance of affiliative relationships. As early as 2001, the zoo community emphasized the importance of describing and studying how adolescent blackback males or young silverback males change their social relationships with the dominant silverback male as they grow older (Nakamichi and Kato, 2001). Indeed, key predictors of social instability must be identified to prevent severe aggressive outbreaks in bachelor groups when bachelor males become silverbacks. Likewise, knowledge of key social stability predictors is also necessary to ensure that the affiliative relationships of young individuals are maintained throughout maturation until adulthood. While our knowledge of instability indicators for post-adolescent males and young silverbacks is increasing, very few studies to date have investigated the stability indicators emerging from data collected on younger individuals (juveniles and adolescents). Yet this information could be of crucial importance to guide management decisions attempting to control captive male overpopulation.
However, this bachelor group management strategy may not be sufficient to ensure the long-term relocation of future surplus males. The solutions that have been proposed and discussed include the use of sex ratio selection, a promising approach that requires further improvement. The Gorilla EEP (GEEP) has already reduced the number of females per breeding male to current levels of 2 or 3 at the most, thus providing a normal social upbringing for the offspring. Another proposal is to extend the time that young males remain in their natal group, or house them alone. This also appears to be impossible: blackback subadult transition is driven by intra-group competition, generally through unidirectional aggression exhibited by the silverback (Robbins, 2007), so remaining longer in their familial group will directly impact their well-being. It also appears unethical to house males alone in zoos, because this solution could impact individual welfare with more self-harming or abnormal behaviours such as hair plucking, regurgitation and reingestion, as reported for gorillas housed alone (Less et al., 2013). Although the use of euthanasia as a population management strategy would free up hosting capacities that could be used for housing other gorillas, this strategy is not recommended by the Great Ape TAG (Taxon Advisor Group), who limit its use to exceptional circumstances: "The Great Ape TAG accepts use of euthanasia for Great Apes in preference to keeping them long term under conditions which significantly reduce the individuals' welfare, and if there is no other alternative" (Abelló et al., 2017 page 102). Castration meets this criteria (see Moreno Rivas and colleagues (2018) for additional explanations of the benefits of castration, such as improving genetic variability or reducing the potential drastic negative demographic effects due to an overuse of females contraception), and also reduces the cost of accommodation structures for this species. Given the expense of creating new facilities for big mammals like gorillas, keeping the castrated individuals in their familial group would also have financial advantages. For all these reasons, the Gorilla EEP agreed to test the use of castration among males that were genetically over-represented in the population as a supplementary tool to complement other, insufficient management strategies. Castration at an early age is a possible management strategy insofar that mature castrated males can remain in the presence of the silverback male in their native familial group without the risk of conflicts, or can form part of a bachelor group without competing for dominance. This study is a first step in this new EEP management strategy, and focuses on young gorillas in a familial housing situation to observe their social development. Until 2011, there was only one castrated gorilla in the GEEP and he is the only recorded castrated individual to have reached adulthood. The reports collected from caretakers reveal that this male had a good quality of life (both in the nursery and in the social group). Given this positive feedback, the GEEP committee recommended the castration of ten immature males living in familial groups between 2007 and 2016, followed by a regular monitoring of their development and careful observation to identify any possible impact on the physical, psychological and behavioural development of these males and of the other members of their familial group.
This study investigates the effect of such castration on immature male gorillas. Social behaviour changes dramatically during primate adolescence, and it is unknown how far adolescent social behavioural development is reliant on testosterone and other gonadal hormones (Richards et al., 2009). In western lowland gorillas, these behavioural and hormonal changes mainly occur during juvenescence (from 3 to 4 years to 6-7.5 years) and adolescence (from 6 to 7.5 years to 12-14 years), and continue until they reach adulthood as young silverbacks (from 12 to 14 years to 18-20 years) (Breuer et al., 2009;Stoinski et al., 2002). It is therefore important to improve our knowledge of how castration effects individuals during these life stages.
Zoos do not base their gorilla husbandry decisions in zoos on the physiological and nutritional needs of gorillas alone; they also take the fundamental social needs of this species into account in order to ensure acceptable levels of welfare. We define the socialization process as the acquisition of a social repertoire and the construction of a social network during growth. It is essential to understand the important role that social environment plays in an individual's health and how the management of different social housing situations can improve individual health and welfare (Rose and Croft, 2015;Sueur and Pelé, 2015). These criteria are vital welfare clues to ensure the provision of optimal care and govern the management of surplus males in the captive Gorilla EEP population.
More recently, social network analysis (SNA) has been proved to be an efficient tool for the management of zoo animals (McCowan et al., 2008;Sueur et al., 2011), as reviewed by Rose and Croft in 2015, and is increasingly used in this domain. In 2011, a social network analysis of social preferences in a captive chimpanzee community (Clark, 2011) supported the hypothesis that a large and modern exhibit facilitates the expression of social preferences among individuals. The same year, captive chimpanzees with a high social status were shown to occupy central positions in the grooming network (Kanngiesser et al., 2011). This highlighting of the pivotal role played by specific individuals in maintaining cohesiveness could influence how zoos manage the restructuring of groups. The use of SNA may provide valuable information about the different social life stages in intact or castrated immature gorilla males living in familial and bachelor captive social groups.
Our study seeks to understand the social differences experienced by castrated and intact individuals living in different social situationsnamely familial and bachelor -from the juvenescent to the adolescent maturity phases. Our study hypothesises that these immature captive gorilla males should all show different behavioural types or strategies, depending on their social housing and their age. Publications describing captive and wild populations report that intact males emigrated from the familial groups following their gradual peripheralization and a decrease in the number and quality of relationships they had with other group members, particularly the silverback (Abelló et al., 2017;Harcourt, 1978;Harcourt andStewart, 2007, 1977). In contrast, the weaker expected physical development of castrated males on reaching adulthood could facilitate their relationships with younger group members. Maturing intact gorillas that live in bachelor groups instead of living alone could also benefit from the experience of silverbacks (Levréro et al., 2006). We therefore make the following hypotheses: 1) For intact males living in familial groups, positive social interactions will decrease with age; 2) For castrated males living in familial groups, positive social interactions will remain stable or increase; therefore 3) An increase will probably be observed in the number and quality of relationships between castrated immature males and adult conspecifics in familial groups; and finally 4) Significantly more numerous positive social interactions will be observed for intact males living in bachelor groups than for intact males living in familial groups.

Methods
This research project was designed in accordance with the best practices and highest ethical standards, and meets national and international legislation requirements. Moreover, all methods comply with the EAZA Code of Ethics (www.eaza.net) and the WAZA Code of Ethics and Animal Welfare, namely the 'Ethical Guidelines for the Conduct of Research on Animals by Zoos and Aquaria' (www.waza.org).

Subjects and housing
Small sample size is a common cause of difficulty in many studies seeking to carry out long-term monitoring of socialization processes in captive animals (Watters et al., 2009). Our study addresses this problem through the observation of 93 western lowland gorillas (51 males and 42 females) living in eleven different social units (seven familial groups and four bachelor groups) located in nine different zoological institutions; each group was studied for approximately one month (see Table 1 for full details). Individuals between 0-3 years of age were classified as infants (INF), juveniles (JUV) of both sexes were aged between 3− 6 years, and adolescents (ADO) were 6− 10 years old for females and 6− 12 years old for males. Females were considered adults (A) at 10 years of age, and males were considered adults at the age of 18 after a transitory young silverback (YSB) age class from the age of 12-18. All the groups had daily access to naturalistic outdoor enclosures or islands measuring between 400 and 10 000 square meters in addition to their indoor areas. All the different environments were equipped with appropriate permanent and temporary enrichment for these animals. The housing in all the institutions concerned is accredited under the same Gorilla EEP standards (Abelló et al., 2017, Section 2, pages 32-51). All of the 10 castrated immature gorillas present in our study were castrated by surgery before the age of three, and a minimum of 6 months before the observations began.

Data collection
BL collected all data for each group over approximately one month between April 2016 and May 2017 (see Table 1). Observations were carried out from 9:00 a.m. to 5:00 p.m, when some of the groups observed solely had access to their outside enclosure and the others were free to move between the inside and outside enclosures according to group management and weather conditions. Individuals in the infant age class (0-3 years old) were not included in observation data because they were still dependent on their mothers.
We defined "social preference proximity" as a spatial association within one meter of the focal individual, with or without contact, and the "social tolerance proximity" as distances of one to five meters between the focal individual and a conspecific. The sum of social preference and tolerance proximities provides the "general social proximity" spatial associations (within five meters of the focal individual).
Depending on the number of individuals per group, observations were made using two to four 10-minute continuous focal animal samplings per day for each group member. We recorded proximities between individuals but excluded those relating to aggressions, thus retaining proximities that occurred during affiliative or neutral behaviours (see Table S1) in order to solely interpret data that are collected in Table 1 Composition details of the familial and bachelor groups observed. FSC = familial social context; BSC = bachelor social context. The two grey columns highlight the subjects for whom data are retained for analysis and interpretation in this study (detailed in Table 2).  a context of "positive" welfare and comparable for all groups. When a focal individual was in the proximity of more than one conspecific, the observer recorded the number of individuals and the exact duration whenever the perimeters of each individual changed (from social preference proximity to social tolerance proximity, within a 5 m radius of the focal individual). Samplings were randomized throughout the day and thirty 10-min focal samplings (i.e. a total of five hours) had been obtained per individual by the end of the observations. In parallel, instantaneous samplings were collected every 15 min and provided information on "positive" proximity patterns (using the same ethogram detailed in Supplementary Material, Table S1). The scan was valid if, and only if, all the group members were seen within one minute of beginning the scan sampling. Approximately 20 scans were carried out per day to obtain 200 scans per group.
Proximity data were discarded from the dataset when aggressive behaviours were recorded, to ensure that we only measured proximities as a proxy of affiliative relationships (stability indicator). We decided to concentrate our analyses solely on a positive pattern of proximities as a "positive index of welfare" because aggressive behaviours are rare in both wild and captive environment (Nakamichi and Kato, 2001;Stokes, 2004). Moreover, additional data collection for aggressive behaviours would have required all occurrences samplings, which was not achievable with a single observer. The data collected from focal animal samplings (in terms of durations), and from the instantaneous samplings (in terms of occurrences) were used to build proximity matrices of the "positive" patterns performed at social preference, social tolerance, then finally general social proximities for each group (see next paragraph).

Social network analysis
Social network analysis (SNA) required the collection of all dyadic proximity patterns in a studied group. Observation matrices (at social preference, tolerance, and general proximity) were built by considering the quotient recorded during the observations for each of the two individuals, thus obtaining an average dyadic social proximity value. The sociality focal and scan matrices for each group were compared using Dietz' R matrix correlation tests (1 000 permutations in SOCPROG 2.8, see Supplementary Material, Table S2). The average R-correlation coefficients of the 11 groups were R=0.89, R=0.82, and R=0.77 for general social proximity, social preference proximity and social tolerance proximity, respectively, and were highly correlated (combined P-value<0.0001; R-package CombinePValue). These correlations confirm the validity of all data collected and the use of both matrices (built after focal and scan observations) in the same way for the statistical analysis.
The indices of the matrices of associations were obtained with SOCPROG 2.8 software (Whitehead, 2009). The weighted networks of familial and bachelor groups are represented as graphs via Gephi software (https://gephi.github.io) (see Supplementary Material, Figures S1 to S10). We focused on the strength index for an individual, which is equal to the sum of the total duration/occurrence of proximities with all group peers in our study. Individuals with a high strength index have strong associations with others and/or have many associates (Carter et al., 2013). By dividing the strength index of a chosen individual by the number of group partners, we obtained what we called a personal sociality index (PSI). This index thus determines both the activity budget time and social affinities by revealing "what the individual wants to do near which conspecifics". It is therefore an individual indicator of cohesion.

Statistical analyses
Analyses were performed to compare 27 JUV and ADO males. Amongst these males, 9 intact ADO males lived in bachelor groups and the remaining eighteen males lived in familial groups: nine were castrated (5 JUV and 4 ADO) and nine were intact (5 JUV and 4 ADO) -see details of immature males in Table 2. The data for Mapenzi (castrated ADO in the Beauval group) and Nakouh (intact ADO in the Burgers zoo group) were excluded from all tests after being identified as outlier values on the boxplot distribution for a number of variables (see Figure S11 in Supplementary Material, and further details in the Discussion section). We used tests to look for differences as a function of age classes, fertility status (castrated vs. intact), and housing situation (familial or bachelor groups) for the following variables: (1) PSI, (2) percentage of time spent in isolation (no conspecifics within a 5 m radius), (3) percentage of time spent in social play behaviours, (4) percentage of time spent in social proximity with the silverback, (5) percentage of time spent in social proximity with the mother, and (6) percentage of time spent in social proximity with adult females. Both social preference and Table 2 Details concerning the immature males studied in FSC (familial groups) and BSC (bachelor groups). Among the 29 subjects, the two grey individuals were excluded as outliers. tolerance proximities were taken into account when testing the variables for PSI and the percentage of time spent in proximity with the adult peers. We ran our analyses in three steps: • Comparing variables with age.
First, Spearman rank correlation tests with permutations (Spearman test function in the R coin package) were conducted to establish the "positive" social patterns of the captive immature males through determined functions (fertility status and housing situation).
• Comparing variables specifically for adolescents.
In a second step, one-way analysis with permutation tests (perm. oneway.anova in the R-package wPerm) and pairwise permutation posthoc tests were performed to compare the mean as a percentage of time for the three categories of ADO males for variables 1-3 (intact males living in bachelor groups, intact and castrated males living in familial groups).
• Comparing variables specifically for immature males living in familial groups.
In a last step, permutation tests for independent samples (permTS in the R package perm) were performed to look for any significant differences among the castrated and intact immature males living in familial groups (variables (1) to (6) for the JUV age class, and variables (4) to (6) for the ADO age class). Despite our small sample size, we decided to exclude the data of the two outliers given that the high number of permutations strengthened the results of our tests.
All the analyses run in R (version 3.4.2) were two-tailed and the probability of a type I error was maintained at 0.05. All tests were performed with 9999 permutations.

Results
Matrices of scan associations were analysed to obtain the sociograms of the 11 groups ( Fig. 1: Artis group example, see Supplementary Material for the other groups, Figures S1 to S10).

Personal social index (PSI)
The evolution of the personal social index PSI was compared to age for the three categories of males using Spearman rank correlation tests with permutations. A negative correlation was noted for intact males living in familial groups at social preference proximity ( Fig. 2A) (rho=-0.95, P=0.007) and at social tolerance proximity (Fig. 2B) (rho=-0.88, P=0.015). No correlation was found for castrated males living in familial groups at social preference proximity (rho=-0.37, P=0.3), whereas a positive correlation was observed at social tolerance proximity (rho=0.72, P=0.043). No correlations were observed for the intact males living in bachelor groups at both social preference (rho=0.13, P=0.704) and social tolerance proximity (rho=0.19, P=0.586). Permutation tests in the two JUV categories (intact and castrated) living in familial groups did not reveal any significant differences at social preference (P=0.373) or social tolerance proximity (P=0.984). One-way analysis of the three ADO categories of immature males did not reveal any significant differences at social preference proximity, with a value of P=0.207. However, a trend of significant difference (P=0.085) was observed at social tolerance proximity. Pairwise permutation tests revealed that the PSI at social tolerance proximity for castrated ADO individuals living in familial groups was significantly higher than that of intact ADO males living in the same context (P adjust =0.032). The other pairwise comparisons reveal no differences (intact in familial groupsintact in bachelor groups: P adjust =0.636; castrated in familial groupsintact in bachelor groups: P adjust =0.128).

Isolation time
Spearman rank correlation testing between the percentage of time budget spent in isolation and the age of individuals indicated a positive correlation (rho=0.90, P=0.011) for intact males living in familial groups (see Fig. 3A). There were no correlations for castrated males living in familial groups (rho=0.13, P=0.706) and for intact males living in bachelor groups (rho=0.10, P=0.776). Permutation tests for the two JUV categories living in familial groups did not reveal any significant differences in mean values for isolation (P=0.249). One-way analysis for the three ADO categories of immatures males for the percentage of time spent alone revealed significant differences (P < 0.001), with significantly less isolation in ADO castrated males (mean=15.15%±2.68) than in both intact ADO males living in familial groups (mean=30.61%±3.8) (P adjust =0.033) and intact ADO males living in bachelor groups (mean=50.95%±6.08) (P adjust =0.009). Intact ADO males living in bachelor groups were also more isolated than intact males living in familial groups (P adjust =0.029).

Social play
Spearman rank correlation testing between the percentage of time budget spent in social play and the age of individuals (see Fig. 3B) revealed a trend of negative correlation for intact males living in bachelor groups (rho=-0.64, P=0.072) and for intact males living in familial groups (rho=-0.60, P=0.08). However, no correlation was found for castrated males living in familial groups (rho=-0.18, P=0.604). Permutation tests for the two JUV categories living in familial groups did not reveal any significant differences in mean values for social play (P=0.786). One-way analysis of the percentage of time budget spent in social play highlighted significant differences between the three ADO categories of immatures males (P=0.016). Post-hoc pairwise comparisons showed that castrated ADO males living in familial groups tended to be significantly more involved in social play than intact ADO males living in bachelor groups (Padjust=0.056), whereas no differences were found for the other pairwise comparisons (intact ADO males in familial groups -intact ADO males in bachelor groups: P adjust =0.965; intact ADO males in familial groups -castrated ADO males in familial groups: P adjust =0.118). However, complementary permutation tests comparing the amount of social play by males during juvenescence and adolescence highlighted that intact adolescents were significantly less involved in play than intact juveniles (P=0.015), contrary to observations in castrated immature males (P=0.73).

"Positive" pattern of proximity with the silverback
The following results concern social variables that are only present in the familial housing situation, and highlight the differences between the two fertility statuses (intact vs. castrated). Fig. 4   percentage of budget time used to perform positive behaviour in the proximity of the silverback. The Spearman rank tests revealed a negative correlation between age and the percentage of proximity with the silverback at social preference proximity (Fig. 4A, rho=-0.85, P=0.017) and at social tolerance proximity (Fig. 4B, rho=-0.95, P=0.008) for the intact males. However, no correlations were found for the castrated males at social preference proximity (Fig. 4A, rho=-0.15, P=0.666) or at social tolerance proximity (Fig. 4B, rho=-0.32, P=0.370). Permutation tests comparing associated means revealed that castrated ADO males performed significantly more positive behaviours at social tolerance proximity with the silverback than the intact ADO males did (P=0.027). No differences were found for the JUV and ADO males at social preference proximity (for JUV P=0.421; for ADO P=0.543), or for the JUV males at social tolerance proximity (P=0.857).

"Positive" pattern of proximity with the mother
As some individuals did not have their mother in their group, it is important to note that the sample size is smaller than the initial one for the tests on this variable (N=8 for intact males (5 JUV + 3 ADO); N=7 for castrated males (4 JUV + 3 ADO)). For both categories (intact and castrated), Spearman rank correlation tests revealed negative correlations at social preference proximity when juveniles males were compared to older adolescents (Fig. 5A, intact: rho=− 0.98, P=0.009; castrated: rho=− 0.89, P=0.029). No correlations were found for the same test at social tolerance proximity (Fig. 5B, intact: rho=− 0.57, P=0.120; castrated: rho=0.29, P=0.484). The comparisons of associated means did not reveal any differences between the two categories of males for JUV and ADO age classes (permutation tests at social preference proximity for JUV P=0.413 and for ADO P=0.5; and at tolerance proximity for JUV P=0.111and for ADO P=0.3).

"Positive" pattern of proximity with the adult females
A ratio was used to allow comparisons between different numbers of females in each group. Spearman rank correlation tests did not reveal any correlation at social preference proximity when intact juveniles males were compared to older intact adolescents, and a trend of significant positive correlation was observed for castrated males (Fig. 6A, intact: rho=0.10, P=0.813; castrated: rho=0.68, P=0.053). A positive correlation was observed for castrated males at social tolerance proximity, whereas positive behaviours in the proximity of adult females decreased with age for intact males (Fig. 6B, intact: rho=− 0.53, P=0.136; castrated: rho=0.82, P=0.019). The permutation tests did not reveal any significant differences at social preference proximity (for JUV P=0.532 and ADO P=0.257). Likewise, no difference was observed for the JUV age-class (P=0.571) at social tolerance proximity. However, castrated ADO males performed significantly more positive behaviours  towards adult females than the intact ADO males did (P=0.029).

Discussion
This is the first study to focus on the socialization process of immature male gorillas in captivity, according to both the fertility status (castrated or intact) and the social composition (familial or bachelor grouping). These comparisons between the three categories of immature males, i.e. intact and castrated living in a familial group; and intact living in a bachelor group, revealed significant differences according to the personal sociality index, the percentage of time budget spent in isolation, and the percentage of time budget spent on social play.
The first finding is that the differences between the male categories from different age classes only occurred within social tolerance proximity (between one to five meters). It was not surprising that no differences were observed at social preference proximity (within one metre of the focal contact) because unlike chimpanzees and bonobos, gorillas generally perform few positive behaviours in the close proximity of a conspecific, as mentioned in a number of studies (Chelluri et al., 2013;Furuichi and Ihobe, 1994). Secondly, in accordance with the history life trait already described for gorilla species, intact males living in familial groups became less and less socially cohesive as they grew up (i.e. the personal sociality index decreased significantly at both social preference and tolerance proximity). Our results converged with those stating that spatial associations, proximity and affiliation decrease as the frequency and intensity of aggression increases in both wild (Harcourt and Stewart, 1981;Robbins et al., 2004;Watts and Pusey, 1993) and captive environments (Abelló et al., 2017;Stoinski et al., 2004a,b;Vermeer et al., 2014). This generates a progressive peripheralization until the emigration (in the wild) or the transfer (human management in captivity) of the male from the natal group. Additionally, maturation during adolescence is accompanied by intense morphological changes in stature: maturing males increasingly differ from their younger conspecifics in terms of body size and social skills. This leads to unbalanced power relations which explain the increased spatial distance with the other immature members of the group, and, in turn, a lower intragroup cohesiveness (i.e. a decrease in the sociality index for the adolescence age class). Immature males housed in bachelor groups contrasted with those housed in familial groups because their sociality index did not evolve significantly during adolescence. Wild male mountain gorillas have been reported to live together for years in unisex groups if they are not joined by females (Fossey, 1983;Yamagiwa, 1987aYamagiwa, , 1987b. Multimale groups are rare in western lowland gorillas (Magliocca et al., 1999;Parnell, 2002), but an all-male structure was observed for long periods of between 13 months and 3 years in observed groups that contained more than one immature or adult male (Gatti et al., 2004;Robbins et al., 2004). Levrero and colleagues also reported wild western lowland gorilla bachelor groups in 2006, but none of these groups were composed of several silverbacks. Furthermore, strong affiliative relationships among males in a captive bachelor group are most likely to occur when males are immature (Stoinski et al., 2004a,b). It is therefore preferable to transfer immature males at an early age if they are destined for life in a bachelor group, because socialization with other males is easier with younger animals (Abelló et al., 2017). These different aspects, as well as the absence of competition for mating in captive bachelor groups, could explain the stability of the sociality index for the adolescent bachelor males studied. With the exception of the differences highlighted between the two housing situations (familial and bachelor grouping), our results underline contrary types of social changes according to the fertility status of individuals. Comparison of juvenescent and adolescent life stages showed that while intact males are excluded from the familial group at adolescence, castrated males significantly increased their intragroup cohesiveness at tolerance. Moreover, analyses in the familial groups revealed significantly more cohesion in social bonds for castrated adolescents compared to intact males of the same age class.
The results for isolation time are in concordance with our findings on the personal sociality index. The intact adolescent males living in a familial group were significantly more isolated than the castrated adolescent males, and, the intact adolescent males living in a bachelor group spent significantly more time in isolation than both intact and castrated adolescent males living in a familial group. Like in the wild, the "positive" or "negative" reactions between males in a captive bachelor group are expected to be directly related to the perceived affinity with a conspecific, but also to a possible threat of a specific male, reflecting both the degree of vulnerability of the latter (his age and physical condition) and his relative familiarity with the other males. Consequently, zoos that are designing an enclosure for a bachelor group are advised to follow the recommendations made by Stoinski et al., in 2004(b) and Coe et al. in 2009 to ensure that gorilla males can control how much space they are able to put between themselves and conspecifics. Young silverback individuals were present in two of the bachelor groups studied (Amneville group 2 and Loro Parque), and this factor could explain the higher isolation rates observed for intact adolescent males in a bachelor group. Also, time spent in isolation may simply be the expression of the individual's personal desire to be alone. Similarly, the captive housing conditions of a familial group should be carefully considered when anticipating the removal of intact males that have reached the end of adolescence, due to the possible increase in conflicts with the silverback (personal observations). Maturing intact adolescents tend to challenge their fathers (and conversely), and this can cause social stress for all group members. The housing should provide enough space for gorillas to avoid these conflicts, and be complex enough to provide safe escape routes and isolation areas. In this respect, the different management strategies evaluated in this study (familial and bachelor grouping) can permit individuals to avoid conflict, and thus become a critical a reliable tool to improve the well-being of social species maintained in captivity (Morgan and Tromborg, 2007). Given that the significantly higher share of isolation in the time budget of intact males living in bachelor groups does not impact the personal sociality index (i.e. there are no differences between intact males living in the different housing situations), it is likely that male-male interactions are promoted in the unisex social context. These findings suggest that gorillas have the possibility to selectively modify the distribution of some typical patterns of their behavioural repertoire according to the housing situation and the social composition of the group. This high level of behavioural flexibility has been also highlighted in captivity by testing the crowding effect in the Apenheul gorilla group (Cordoni and Palagi, 2007). The dynamic of the social networks emerging from social relationships between members and the pattern of activity time budget are constantly evolving, and therefore require a detailed follow-up to ensure optimal captivity welfare. Our study protocol highlights that social network analysis is a useful tool to improve the monitoring of the different management strategies used for animals in captivity.
Results for the involvement of the immature males in social play are in concordance with the known life history traits for gorilla species. Whatever the housing situation, all intact males tended to play significantly less from juvenility to adolescence, and even during adolescence. This evolution is due to changes in time budgets during maturation (increased foraging time as well as a growing interest in other social activities such as agonism and sex), leading to an overall decline in play at the end of the juvenile period. Moreover, as for other primate species such as the chimpanzee (Shimada and Sueur, 2014) and Japanese macaques (Shimada and Sueur, 2018), it appears that social play mainly fulfils the function of strengthening social bonds during juvenility, with a significant decrease during adolescence. The results of Maestripieri and Ross' study in 2004 were consistent with both motor training and social skills hypotheses for captive gorillas aged from 0 to 5 years. Our results and those of Maestripieri and Ross (2004) highlight the importance of considering "social compatibilities" during the formation of bachelor groups. Like other primate species, immature gorillas are expected to play and interact mostly with individuals that have equivalent body size and social skills (Fagen, 1981;Fairbanks, 1993). This morphological compatibility factor is a major concern for the formation and maintaining of all-male groups in captivity. The absence of appropriate social partners is the major factor that could explain the outlier data recorded for the castrated individual Mapenzi, housed in a familial group at Beauval zoo (see Supplementary Material, Fig. S11). The same may apply to the outlier data for Nakouh, an adolescent intact male in Burgers' zoo (see Supplementary Material, Fig. S11). An adult female from this institution gave birth on the 13th June 2013 to twins, N'kato and N'hasa. This event is very rare in gorillas and may explain the data recorded for Nakouh: the presence of a new-born in a group induces a significant reduction in inter-individual distance not only for characterized specific dyadic interactions (Nakamichi and Kato, 2001;Stoinski et al., 2003) but also for general group cohesion (Kurtycz et al., 2014). Moreover, there were no older immature conspecifics with whom Nakouh could interact. He did live with three younger males during his socialization process. Nakouh's socialization process may have been delayed by this social imbalance. However, complementary tests comparing the share of social play between juvenile and adolescent age classes underline that castrated individuals do not decrease their involvement in social play during these two life stages, unlike the intact males. The combined results on the three variables described (personal sociality index, isolation and share of social play in time budget) highlight that captive western lowland gorillas are socially flexible, and confirm that maturing males adapt well to the bachelor management strategy that has been implemented for over three decades. Indeed, the "positive" interactions prioritised by males maintain social cohesion despite a greater share of isolation in the activity budget time, suggesting that maturing males are adapting well to bachelor group housing. Finally, the different behavioural patterns presented in our study suggest that castration has a positive impact on the intra-group cohesiveness of the individuals during juvenescence and adolescence.
While gorilla familial groups are mainly based on adult male-female bonds (Harcourt and Stewart, 2007), relationships between the silverback and its offspring play a significant role in group structure. As already shown in different studies, the close bonds between the adult male and the immature males (Rosenbaum et al., 2016;Stewart, 2001;Yamagiwa, 1983) are best explained as paternal care (protection and tolerance of proximity) rather than a significant investment in their offspring (resting in contact, grooming, play) (Rosenbaum et al., 2011). The "positive" behaviours that intact immature males performed at preference proximity of the silverback and the mother decreased significantly when they reached maturity. This was also the case at tolerance proximity with the silverback. Conversely, in accordance with our results on the personal sociality index, castrated males keep strong social bonds with the silverback and the adult females when they reach adolescence life stage (i.e. a significant increase at tolerance proximity during maturation). This increase in proximity and interactions might be explained by the fact that conspecifics perceive changes induced by castration that could be visual, auditory, olfactory, and/or behavioural (Caillaud et al., 2008;Masi and Bouret, 2015). Our study therefore highlights not only behavioural changes for castrated individuals, but also a different perception of the latter by their group peers. Although our study did not address this issue, our results suggest an effect of castration on the development of social bonds. The castration induces a cancellation of the growth associated with the first stage of androgen increase during adolescence (Stoinski et al., 2002) (see Supplementary Material, Fig. S12), and could for example increase "the social compatibilities" between the developmentally staggered individuals that compose the group. The behavioural and morphological modifications observed may improve group cohesiveness, in accordance with the social and motor skills hypotheses (Baldwin and Baldwin, 1974;Byers and John, 1998).
As physical, social or psychological alterations are often associated with subtle changes in behaviour, our behavioural monitoring data can be beneficial to many areas of gorilla husbandry. The collaboration of European zoos in this study has ensured a low number of biases due to care routines, diet recommendations and environmental standards that all follow the same set of guidelines recommended by the EEP (for details see Abelló et al., 2017). We acknowledge that caution must be taken when making our conclusions, as the individuals involved were observed for just one month of their development. For this reason, future research should be carried out over the long-term to gain a more detailed knowledge of the social and physiological development of castrated males in familial groups and intact males in bachelor groups.

Conclusion and perspectives
Our study suggests that the housing of castrated surplus males in familial groups is a feasible alternative strategy for positive sociobehavioural management during the maturation of gorilla males. Indeed, our results highlight that castrated adolescents are significantly more cohesive in their native group than their intact counterparts. Furthermore, our results comparing intact immature males living in familial and bachelor contexts are in concordance with those of all the previous studies focusing on captive gorilla bachelor groups. These results strengthen the efforts made by the Gorilla Eaza Ex-situ Programme (GEEP) to improve the different housing management strategies designed to solve the growing problem of surplus male gorillas in captivity. We recommend avoiding the castration of immature males that have no other immature male peers in their group (as seen in the case of Mapenzi, a castrated adolescent male excluded as outlier data from our analyses). Immature males that are transferred into bachelor groups should be pair-generational individuals that share the same developmentally staggered stages, as described by the motor-training hypothesis and the social-relationship hypothesis of play (Maestripieri and Ross, 2004) and social development in general (this study). Other factors such as enclosure size, differences in facility design features, effects of spatial crowding, group size and daily care routine could all affect the proximity between gorillas group members and their behaviour (Coe et al., 2009;Kuhar, 2008;Ross et al., 2011). It is unlikely that all of these factors explain all of the variation in social dynamics, it will nevertheless be essential to include these covariates in the next studies. Moreover, future studies are necessary to investigate whether castrated males have behavioural similarities with females, and understanding how castration influences physical and hormonal development can help us to better understand the long-term effects of this contraceptive method on the social dynamics of the group. Our results should be complemented with a further study comparing the networks of "negative" proximity pattern in the same way as we did here for "positive" proximity. Finally, the long-term monitoring of castrated males is necessary, especially from the young silverback age-class until adulthood, to ascertain whether the use of castration as an additional socio-behavioural management strategy is consistent over time and meets the welfare criteria for maintaining this species in captivity.