Association patterns in a high-elevation chimpanzee community in Rwanda

Investment in social interaction and affiliative behaviour is often related to variation in sex-specific dispersal patterns among species but can also vary within species in response to local environmental conditions and feeding competition. Chimpanzee (Pan troglodytes) association patterns have been studied primarily in low and mid-elevation tropical forests. This study adds to our understanding of association patterns in chimpanzees by presenting the first data on association patterns in a high-elevation population at Nyungwe in equatorial Rwanda. Using data on co-occurrence in social parties (party association) and calculating gregariousness indices, we quantified levels of associations among age-sex classes relative to food (fruit) distribution and the presence of oestrous females. Overall, adult females in Nyungwe are less gregarious than males. Female but not male gregariousness increased when feeding on clumped foods, indicating that resource characteristics impact female association tendencies more intensely than males. Despite high elevation forests representing a habitat with relatively fewer fruit resources for chimpanzees, at an edge of range location, we found that gregariousness patterns were driven by similar factors as those in mid and low elevation chimpanzee populations. HIGHLIGHTS – We quantified association patterns in a high elevation population of chimpanzees at Nyungwe (Rwanda) – Adult females were less gregarious than males – Female but not male gregariousness increased when feeding on clumped foods, indicating that resource characteristics impact female association tendencies more intensely than males


INTRODUCTION
Group living reflects inter-individual attraction to social benefits such as opportunities for reproduction, cooperation, protection and information acquisition. However, costs such as susceptibility to pathogens and, in particular, intra-group competition over food resources moderate the benefits of group living, and can constrain group size or destabilise groups (Clutton-Brock 2016; Grueter et al. 2020). Intragroup scramble feeding competition tends to increase with group size and entails the energetic costs of travelling farther or visiting additional food patches (Chapman & Chapman 2000;Koenig 2002). Among mammals, these foraging costs are particularly applicable to females because access to food/nutrition is a stronger constraint on female than male reproductive performance (Trivers 1972). Male reproductive success is primarily contingent on the ability to secure access to fertile females, and therefore, the spatio-temporal distribution of males tends to closely follow that of females (Emlen & Oring 1977;Altmann 1990).
Association patterns are often the outcome of an individual's decision to join together with other conspecifics or avoid them (Cords 2002;Mourier et al. 2012;Shizuka et al. 2014). However, the mere fact that individuals are associated does not necessarily imply that they are socially bonded (Pepper et al. 1999;Gilby & Wrangham 2008). Associations can be driven by external forces such as resource distribution or access to shelter (Campbell et al. 2006;He et al. 2019). Furthermore, the physical features of the habitat can also impose direct constraints on the pool of individuals that one can associate with, e.g. by shaping movement patterns and encounter probabilities (Titcomb et al. 2015;He et al. 2019).
Associations offer a substrate for individuals to form bonds, i.e. strong, equitable and enduring social relationships (Ostner & Schülke 2014). Social bonds result from repeated affinitive and affiliative interactions (Hinde 1976;Kappeler & van Schaik 2002) including allogrooming (Dunbar 1991), food sharing (Wittig et al. 2014), spatial proximity maintenance (Cords 1997) or behavioural synchrony (Dunbar & Shultz 2010). The strength of bonds can vary with respect to sex; they are typically more fully developed among the philopatric sex (van Hooff & van Schaik 1994), often -but not necessarily -aided by inclusive fitness benefits (Sterck et al. 1997;Silk 2006). Male-male bonds can enable the formation of dyadic or polyadic coalitions against an in-group member (Noë & Sluijter 1995;Berghänel et al. 2011) and group-level cooperative activities such as defence against intrusions from external males (Pope 1990;van Hooff 2000). Ultimately, social bonds can boost an individuals' reproductive success (Schülke et al. 2010). However, male bonds are expected to deteriorate when competition over fertilisations predominates (van Hooff & van Schaik 1994;Qi et al. 2017). Female-female bonds can be beneficial in the context of resource defence against outgroups (Wrangham 1980) and can also facilitate coalitions against male aggression (Newton-Fisher 2006;Tokuyama & Furuichi 2016). Moreover, they also play a role in facilitating allocare and infant handling (Xi et al. 2008;Hrdy 2009;Yu et al. 2013) as well as providing dependent offspring with opportunities for socialisation (Williams et al. 2002). Female investment in strong, enduring social bonds can be rewarded with lifetime fitness benefits (Silk et al. 2010). Lastly, male-female bonds can function to protect females and their infants from infanticide, harassment or predators (Smuts 1985;van Schaik & van Noordwijk 1989;Palombit 2000;Newton-Fisher 2006;Grueter et al. 2012). Social bonds between males and females can also have important fitness consequences for both sexes (Ostner et al. 2013;Archie et al. 2014).
Chimpanzees feature all sex combinations of associations, albeit at varying frequencies. Chimpanzee males tend to be highly gregarious and exhibit stronger affiliative and cooperative behaviour than females (Goodall 1986;Newton-Fisher 1999;Boesch & Boesch-Achermann 2000;Wrangham 2000;Mitani 2009;Bray & Gilby 2020; for adolescent males see Sandel et al. 2020). The more gregarious nature of males may stem from them being less restricted by scramble feeding competition than females (Wrangham 2000;Wakefield 2008), and from the fitness benefits provided by defence of territorial space and coalitionary support for rank acquisition and maintenance (Langergraber et al. 2007;Gilby & Wrangham 2008;Gilby et al. 2013). Male party size is generally positively correlated with the presence of females with sexual swellings (Pepper et al. 1999;Boesch & Boesch-Achermann 2000;Hashimoto et al. 2001). However, at Ngogo, Uganda, male gregariousness (measured using levels of association among different age-sex classes in a party) decreased in parties with oestrous females, which may reflect high intrasexual mating competition and competitive exclusion in this large community (Mitani et al. 2002).
Chimpanzee females are typically less gregarious than their male conspecifics but there is considerable variation around this, which may be related to ecological conditions such as predation pressure, food distribution/abundance and the subsequent degree of intrasexual competition (Lehmann & Boesch 2008). In East Africa, chimpanzee females exhibit relatively weak associations with other females and low levels of gregariousness (Gombe: Goodall 1986; Kanyawara: Wrangham et al. 1992;Wrangham 2000; but see Ngogo : Wakefield 2008;Langergraber et al. 2009). In contrast, at Taï, Côte d'Ivoire, in West Africa, higher female gregariousness and stronger dominance hierarchies exist (Boesch & Boesch-Achermann 2000;Wittig & Boesch 2003;Lehmann & Boesch 2008). Wittiger and Boesch (2013) reported weaker dyadic female associations compared to earlier research on the same population, attributing this to an overall decline in community size in Taï. This finding emphasises that patterns of association and gregariousness may change over time and in response to changes in site-specific demographic and ecological circumstances (Wittiger & Boesch 2013).
Male-female affiliative social bonds have been shown to be less important than male-male bonds, and male-female associations are usually the result of co-utilisation of space and not necessarily indicative of a strong bond (Machanda et al. 2013).
Moreover, male-female associations fluctuate in line with the female oestrous cycle and thus more resemble transient reproductive relationships than lasting bonds (ibid.). Male chimpanzees at Ngogo, however, exhibited more stable associations with particular females (Langergraber et al. 2013). Adolescent and young adult males formed affiliative, yet coercive, relationships with females and these extended outside the immediate context of mating (Reddy & Mitani 2020).
Assessing gregariousness patterns in differing communities and habitats can further our understanding of how ecological pressures (in concert with reproductive drivers) shape chimpanzee sociality and grouping patterns. Here we focus on a chimpanzee community inhabiting an ecologically challenging montane rainforest characterised by relatively low fruit diversity and availability (Gross-Camp & Kaplin 2005;Matthews et al. 2019) and a relatively high importance of terrestrial herbaceous vegetation (THV) as a food source (Matthews et al. 2019). A previous study on correlates of party size in the same community has shown that party size was directly influenced by the proportion of fruit in the diet, the size of fruit patches and the presence of oestrous females (Matthews et al. 2021). There was also a direct effect of fruit distribution on the presence of oestrous females, thereby indirectly influencing party size (ibid.).
Here we quantify levels of associations among age-sex classes relative to food distribution and presence of oestrous females. The gregariousness of males is expected to be less contingent on food resources than that of females. We therefore predict, in line with previous studies, that male gregariousness in general is higher than female gregariousness. If food resources limit female gregariousness, then we expect an increase in gregariousness when foods are more clumped (Wittiger & Boesch 2013). The role of clumped resources in inducing associations should be particularly strong in mountainous environments where traversing steep slopes is energetically costly (Green et al. 2020a), food resources are more scarce (Yamagiwa et al. 1996), and energy expenditure from travel can be reduced when feeding on fruit trees with a clumped distribution (Chapman et al. 1995).
The Mayebe chimpanzee community near Uwinka has been subject to nest-to-nest follows and habituation efforts since 1997 through the Projet Conservation de la Fort de Nyungwe individually identifiable chimpanzees, including 14 adult males, 18 adult females, four subadult males, seven subadult females, 12 juveniles (eight males, three females, one unknown), and 12 infants (Green et al. 2020b). The chimpanzees were habituated to observation at an average distance of 10-20 m.

Data collection
Observations. From September 2016 to August 2017, we conducted observations for 10 days each month, with an average of 47.9 contact hours per month (SD = 16 hr; see Table S1 in Supplemental Data). When the coordinates of a nest site from the previous day were known, we located the chimpanzees before dawn and attempted to follow them throughout the day to an evening nest site. If we lost a party due to fast travel, poor visibility, or rugged terrain, we attempted to relocate them through tracking. Otherwise, we listened for vocalisations, inspected recently used feeding sites, and systematically scoured the study area for signs of chimpanzees or fresh tracks. Fleeting encounters were not included; the shortest encounters included were 45 min (see Supplemental Data).
Vegetation sampling. Using randomised sampling stratified by elevation, we established 64 vegetation plots throughout the focal community's home range. Following Potts (2008), we used a nested sampling design with 5 × 50 m plots to measure the diameter at breast height (DBH) of all trees with a DBH > 10 cm, 25 × 50 m plots to measure all trees with a DBH > 30 cm, and 50 × 50 m plots to measure all trees with a DBH > 80 cm (Matthews et al. 2019). We measured the spatial distribution of fruit tree species with the Index of Dispersion (ID; variance/mean number of stems/plot; Greig-Smith 1983). An ID higher than one represents a clumped distribution, an ID equal to one represents a random distribution, and an ID less than one represents a more uniform distribution.
Party size and composition. We aimed to follow chimpanzee parties of all sizes indiscriminately; however, larger parties were typically louder, more conspicuous and easier to follow, which may have created biased results. We defined party size as all independent individuals within 50 m or within visual range of at least one other party member (Pepper et al. 1999;Mitani et al. 2002;Wakefield 2008). Party size and composition were recorded using instantaneous scan samples taken at 15-min intervals. During scans, we counted all individuals in view (excluding infants) and recorded their age-class (juvenile, subadult or adult; Goodall 1968), sex, and, in the case of females, reproductive state (oestrous; anestrous and non-lactating; anestrous and lactating). We scored the sexual receptivity of cycling females based on the size of their sexual swelling, and only females with maximal swelling tumescence (estimated using descriptions in Dahl et al. 1991) were designated as "in oestrous" (Tutin 1979;Wakefield 2008).
A new record of party size comprised the initial count following a change in party size or composition between scans (Boesch 1996). Consecutive scans of the same party were discounted. A fission event was deemed to have occurred when we failed to observe one or more individuals from the focal party in two sequential scans. In the event of a fission, we tossed a coin to determine which party (or individual) to follow.

Data analysis
We performed all statistical analyses using R version 3.4.3 (R Core Team 2017) and SPSS 13.0 (SPSS Inc., Chicago, USA). In order to assess the overall tendency to associate with conspecifics, we used a general gregariousness index (following Underwood 1981;Pepper et al. 1999): where a i is the number of members of age-sex class a in party i, s i is the size of party i, and (s i − 1) is the total number of associates for each member of party i. We used a group randomisation technique (Smolker et al. 1992;Pepper et al. 1999) to generate expected levels of association and assess statistically whether observed gregariousness deviates from chance expectations. We used the vegan package in R version 3.4.3 (Oksanen et al. 2018) to create a matrix of party compositions where rows represent an observed party, columns represent age-sex classes and cells represent the number of individuals of a given class in a given party. The cells within the matrix representing the membership of observed parties were then repeatedly reshuffled while retaining the row and column totals of the original data.
The gregariousness index was calculated after each randomisation, and the mean of 10,000 iterations was considered the expected value. This process controls for the distribution of party sizes and frequency of each class within the dataset when generating a null hypothesis of unbiased association. The ratio of the observed to expected values (O/E − 1) is used to present the results graphically, where positive values indicate higher gregariousness than expected by chance and negative values indicate lower gregariousness than expected. We statistically tested whether observed gregariousness could result from unbiased interactions in each age-sex class by using the proportion of randomised indices that were at least as extreme as the observed index. This proportion represented a one-tailed P-value which we doubled for use in a two-tailed statistical test. To account for multiple comparisons between age-sex classes, we used a sequential Bonferroni correction: where a is the overall experiment wise error rate (0.05), k is the total number of tests, and i is the ith sequential test (Holm 1979).
To assess whether the gregariousness of age-sex classes changed in response to the distribution of fruit and/or presence of oestrous females, we calculated three separate gregariousness indices (modified from Mitani et al. 2002). The first index conveys gregariousness for control parties, i.e., parties that were feeding on foods with an Index of Dispersion (ID) within one standard deviation above or below the average, and with no sexually receptive females present. The second index was calculated for parties feeding on clumped food resources, i.e., an ID one standard deviation above the average and with no sexually receptive females present. We compared periods of food dispersion rather than food abundance since the abundance index was not a good indicator of food consumption (Matthews et al. 2019) or party size (Matthews et al. 2021). We calculated the third index for parties with oestrous females.

RESULTS
Adult females (anestrous non-lactating and lactating) were significantly less gregarious than expected by chance during the "control" period, i.e., parties without oestrous females feeding on fruits with dispersion indices within one standard deviation of the average (Fig. 2a). When food was clumped, and when oestrous females were present, adult female gregariousness was not significantly higher or lower than expected (Fig. 2b-c). Subadult females demonstrated a slightly different trend than adult females and displayed significantly higher gregariousness than expected when food was clumped, but gregariousness did not deviate from expectation during the "control" period (parties without oestrous females feeding on fruits with dispersion indices within one standard deviation of the average) or when oestrous females were present (Fig. 2) Fig. 2. -Gregariousness indices for different chimpanzee age-sex classes. (a) Control period: parties without estrous females feeding on fruits with dispersion indices that did not deviate by more than one standard deviation from the average. (b) Clumped food distribution: parties without estrous females feeding on fruits with dispersion indices that exceeded one standard deviation above the average. (c) Parties with estrous females. Depicted are the ratios of observed to expected associates minus one; ratios above zero indicate higher than expected levels of gregariousness and values below zero indicate lower than expected levels of gregariousness. Age-sex classes with significantly more or fewer associates than expected by chance are indicated by grey bars.
significantly different from expected rates of gregariousness during any of the time periods (Fig. 2), but subadult males were significantly more gregarious than expected when oestrous females were present (Fig. 2c).

DISCUSSION
This study has provided details on association patterns in a high-elevation chimpanzee population that only recently has become the target of research efforts. Relatively little is known about the ecology of high-elevation chimpanzee populations, and findings from such populations, which may be considered to be at the edge of the habitat range of chimpanzee, are important to comprehend the scope of within-species socioecological adaptability. The level of associations we observed among chimpanzee age-sex classes relative to food distribution and oestrous females indicates that, overall, adult females in Nyungwe are less gregarious than males. This finding fits with the conventional principle that male primates living in territorial multi-male/multi-female groups with female-biased dispersal and male philopatry (Wrangham 1979;Di Fiore et al. 2009) have stronger associations than females (Wrangham & Smuts 1980;Chapman 1990;Shimooka 2005). Male-male bonds can be beneficial for a variety of purposes such as coalitionary support (Silk 1994;Gilby et al. 2013) and territorial defence (Pope 1990;Samuni et al. 2021). It is as yet unknown whether associating males in this population are closely related and accrue inclusive fitness benefits from these associations (Pope 1990;Krützen et al. 2003;Mitani 2009).
Mothers were the least sociable class. Maternal asociality may have social or ecological roots. It may be a means to ensure infant safety by limiting exposure to potentially infanticidal males or females (Otali & Gilchrist 2006;Townsend et al. 2007;Smith et al. 2008;Lowe et al. 2019). It may also be attributable to the stronger constraints of feeding competition on mothers (Wrangham 2000). Overall, adult females were more gregarious when feeding on clumped foods, and when oestrous females were present. Only during the control period were females significantly less gregarious than expected. Similar results were reported for females in Ngogo (mid elevation, Mitani et al. 2002) andTaï (low elevation, Wittiger &Boesch 2013). Increased female gregariousness when oestrous females were present may indicate the benefits of oestrous females to other females e.g., through stimulation of their oestrous swelling (Wallis 1992(Wallis , 1994. The increase in female gregariousness when feeding on clumped foods conforms to the ecological constraints model, since interpatch travel distances become shorter and the energetic costs of joining parties are reduced when food trees are clumped (Chapman et al. 1995). However, as mentioned above, the increased risk to infanticide acts as a counterbalancing force that may curtail gregariousness. The fact that clumped dispersion of fruit resources was associated with greater female gregariousness at both a low-elevation (Tai) and highelevation site (Nyungwe) suggests that this pattern is not a unique characteristic of montane forests.
Interestingly, the trend of increased gregariousness when feeding on clumped foods was strongest for subadult females, who were significantly more gregarious than expected. This result is difficult to explain in light of the fact that few studies have focused explicitly on sociality among subadult chimpanzees (Pusey 1990). However, subadult females receive more male aggression than adult females (Wakefield 2008), and by increasing gregariousness when the associated costs of group feeding are reduced, subadult females may be able to develop social relationships while avoiding male aggression. Alternatively, this trend may simply reflect the reduced energetic costs of both subadult and adult females joining parties when feeding on clumped resources.
Adult male gregariousness did not deviate from expectation, but subadult males showed increased gregariousness when oestrous females were present (contra Mitani et al. 2002). We suggest that this pattern of subadult male gregariousness may reflect a strategy to develop social relationships and status. Male chimpanzees typically exhibit a clear dominance hierarchy with more dominant males experiencing higher rates of reproductive success (Wroblewski et al. 2009;Newton-Fisher et al. 2010). However, Wroblewski et al. (2009) reported that younger and lower-ranking males experienced better than predicted reproductive success, likely from alternative mating strategies such as consortships and success in sperm competition. This reproductive benefit may explain the high rates of subadult gregariousness in the parties with sexually receptive females reported here, but data on male behaviour, paternity patterns, male age and dominance rank are needed to explore this hypothesis further.
It should be noted that habituation levels varied among individuals observed in this study. Generally, females tended to be less habituated than males. While this is not considered to have impacted the validity of the findings, it may have in part contributed the large differences in sex-specific association patterns.
In summary, using data on a montane-dwelling population, this study adds to our understanding of the intricacies of association patterns in chimpanzees. It confirms species-general hallmarks such as overall greater male gregariousness than females and demonstrates covariation between female gregariousness and food distribution. Future research should attempt to identify the mechanisms sustaining associations (e.g., kinship) and establish the reproductive benefits and fitness consequences of bond formation. To further resolve the drivers of chimpanzee gregariousness patterns at Nyungwe, additional research is needed. For instance, a contrast between association patterns in Nyungwe proper (a large, continuous forest) with a nearby forest fragment that has undergone massive anthropogenic change (Moore et al. 2018) could reveal variation in grouping patterns linked to altered or missing ecological elements (e.g. fruit availability and distribution) in the fragment. Moreover, to capture the role of inter-annual variation in food resource characteristics and levels of associations among age-sex classes, data across multiple years are required. How variation in association patterns are linked to site-specific resource characteristics will also require additional comparative work across locales or ecological gradients.

ETHICAL STANDARD
All applicable international, national, and/or institutional guidelines for the care and use of animals were followed. All research protocols were reviewed and approved by the University of Western Australia's Animal Ethics Committee (RA/5/15/1070) as well as the governing body of Nyungwe NP, the Rwanda Development Board/Tourism and Conservation Department. The data are available from the last author (J.K. Matthews) upon request.