Group-level cooperation in chimpanzees is shaped by strong social ties

Humans maintain extensive social ties of varying preferences, providing a range of opportunities for beneficial cooperative exchange that may promote collective action and our unique capacity for large-scale cooperation. Similarly, non-human animals maintain differentiated social relationships that promote dyadic cooperative exchange, but their link to cooperative collective action is little known. Here, we investigate the influence of social relationship properties on male and female chimpanzee participations in a costly form of group action, intergroup encounters. We find that intergroup encounter participation increases with a greater number of other participants as well as when participants are maternal kin or social bond partners, and that these effects are independent from one another and from the likelihood to associate with certain partners. Together, strong social relationships between kin and non-kin facilitate group-level cooperation in one of our closest living relatives, suggesting that social bonds may be integral to the evolution of cooperation in our own species. We conducted our study on adult (>12 years) male and female chimpanzees (Pan troglodytes verus) of three different communities (i.e., North, South and East) at the Taï National Park, Côte d’Ivoire (5°45′N, 7°07′W). Christophe Boesch started habituation efforts with North group in 1979 (61 North: 1979; South: 1993; East: 2000; see Fig. 4 for home range estimations of the three groups), and established systematic observations and the collection of demographic data. Once the chimpanzees were habituated to researcher presence, trained observers began nest to nest focal-follow data collection62 of adult individuals from each community on a daily basis (North: 1990–present; South: 1999–present; East: 2007–present). Focal-follow data included documentation of all changes in activity, social interactions (e.g., grooming) and vocalizations involving the focal individual63. Furthermore, due to the dynamic fission–fusion social system of chimpanzees61,64 we continuously recorded of all changes in party size and composition. Presented are the 95% (dashed lines) and 99% (solid lines) kernel home range utilization areas of North group (blue), South group (yellow), and East group (red). The 95% and 99% kernel home range sizes were 19 and 30 km2 in North, 38 and 49 km2 in South, and 32 and 43 km2 in East, respectively. The home range overlap at 99% Kernel was 1.3 km2 between North and South, and 15 km2 between South and East. There was no home range overlap between North and East. The study was done in compliance with the ethics policy of the Max Planck Society and was approved by the Ministries of Research and Environment of Côte d’Ivoire, and Office Ivoirien des Parcs et Réserves. Chimpanzee intergroup relations include patrols around border areas and into the territories of neighbours, and hostile and aggressive interactions with neighbours upon encounter37,44,45. Both types of territorial activities are associated with potential costs and unpredictability of outcome23. However, here, we focused on intergroup interactions that resulted in vocal or physical encounters with neighbours (hereafter intergroup encounters), as those pose a greater direct risk to participating individuals. To examine what influences males and females’ participation propensity in intergroup encounters we used detailed all occurrence data on intergroup encounters collected between the years 1997–2018. If an intergroup encounter was recorded, observers interrupted the focal follow data collection to document all occurrence data on participation, vocalizations and aggressions observed. In our analyses, we only included intergroup encounters in which at least one within-group member actively approached and engaged in the conflict and marked those that did so as participants. Thus, cases of intergroup encounters in which all party members showed no reaction to vocalizations by neighbours and/or retreated into their core area after detecting neighbours auditory and/or visually (e.g. ambush) were excluded, as no voluntary participation occurred22. Non-participation also meant that we could not evaluate how relationships with participants may contribute to participation, or what causes some community members to participate in the encounter but not others. By limiting our data to intergroup encounters involving an active approach or vocal response towards a neighbouring group, we can reliably investigate influences of inter-individual variation on participation. For a subset of the intergroup encounter data with an active response (collected between the years 2013–2018 and comprising almost half of the entire dataset) we had accompanying information on whether or not intergroup encounters involved a border patrol. Border patrols are often initiated while in core areas and include cohesive movements of in-group members toward and beyond the borders of their territory23, during which participation is considered voluntary22. Overall, more than 92% of active intergroup encounters followed a border patrol behaviour, thus, this approach sufficiently accounts for participation decisions. We differentiated between two types of active intergroup encounters both involving an active approach towards the out-group but that differ in the degree of contact, either (a) only vocal (vocal encounters) or (b) also visual and/or physical (contact encounters). Maternal kinship and paternity for individuals of all ages of the different groups was established by means of pedigree and genetic data extracted from faecal samples65 in the genetic lab of Linda Vigilant at the Max Planck Institute for Evolutionary Anthropology. The current Taï genetic dataset includes 259 individuals with an average of 83% complete genotypes at 19 autosomal loci. In brief, we extracted faecal samples (~100 mg) using either the QIAamp DNA stool kit (Qiagen) or the GeneMATRIX Stool DNA Purification Kit (Roboklon) following manufacturer instructions. We first simultaneously amplified aliquots at all 19 loci and subsequently reamplified using fluorescently labelled primers as detailed in ref. 66. To confirm individual identities and assign paternities we compared the resultant genotypes using the ‘identity analysis’ or the ‘parentage analysis’ functions of CERVUS, respectively, with confidence assessments of 80% and 95%. Potential sires were excluded by two or more mismatches such that each paternity assignment was of high likelihood. We determined within-group dominance relationships and their changes over time by applying a likelihood-based adaptation of the Elo rating approach67–69. We used submissive uni-directional pant grunt vocalizations to estimate the dominance hierarchy within each of the sexes and groups separately (males: 9064, 10,011 and 4704 pant grunt vocalizations in North, South and East, respectively; females: 966, 1302 and 207 pant grunt vocalizations in North, South and East, respectively), standardized to a range from 0 to 1. We used focal follow data collection of grooming behaviour (grooming bouts: North—14,599, South—18,948, East—9,040) to assess dyadic relationship quality following Samuni et al. 16. In brief, we implemented the dynamic dyadic sociality index method16,69,70, which provided continuous directed daily measures for grooming for each combination of within-group adult individuals. Using this method, dyads achieve high values by investing in grooming with one another more than others but maintain stable high values only if the investment provided was regular and consistent. We then used the directed grooming scores to evaluate preferences of grooming partners separately for same and opposite sex dyads. If both individuals of a dyad were each other’s top preferred grooming partners, this dyad received a score of 1. Conversely, when the top grooming preference was one sided or entirely absent the dyad received a score of 0. As such dyads who scored 1 were those with mutual and stable grooming relationships. Using this method in the same population, we have previously shown that social bond partners are more likely to engage in the cooperative behaviour of food sharing than non-social bond partners16. Chimpanzees live in a fission–fusion social system with transient and dynamic associations61,64. Association tendencies of males and females in Taï show intraindividual repeatability across days and years71, indicating temporally stable phenotypes of gregariousness in this population which may influence participation likelihood. Thus, to evaluate spatial proximity, we used party association data collected during the year prior to the intergroup encounter date to construct matrices of dyadic association values of all adult individuals in a community. The time frame of one year prior to the encounter provides sufficient data to construct reliable association matrices, but as well likely the best representation of current social conditions (e.g., demography, group size, or female reproductive status) as the encounter which might influence association patterns. The dyadic association values were based on the duration two individuals were observed in the same party as a function of the total duration each of them was observed in total. For each intergroup encounter event we calculated the average dyadic association values with all other adult individuals that participated in the encounter. Higher average association values represent individuals that more frequently associated with encounter participants in comparison to lower association values. Thus, we could account for association degree with encounter participants in shaping participation decisions (see “Statistical analysis”). Chimpanzees use their home range according to the distribution of food sources. Thus, to account for the potential effect of food availability on participation likelihood, we calculated a monthly fruit availability index following a standard index for Taï chimpanzees72. We compiled the index based on the absence/presence of mature fruits, and the density and mean basal area of tree species. Phenology data was collected within the home-range of each chimpanzee group, thus, reflecting local variation in fruit productivity. We fitted generalized linear mixed models (GLMM73) with binomial error structure and logit link function, to investigate variables affecting the likelihood to participate in intergroup encounters for adults. We evaluated ‘participation decisions’ by determining for each intergroup encounter all living adult community members and whether they participated or not (0/1). By including all adult individuals in the community as potential participants, we attempt to capture the patterns that predict the likelihood of some group members to engage in intergroup encounters but not others. This definition follows a previous study in chimpanzees22. As different factors may affect participation likelihood of the different sexes (e.g., reproductive state of females) we fitted separate models for the two sexes (‘male participation model’ and ‘female participation model’). To test whether more cohesive engagement of group members (i.e., number of participants) affected participation in intergroup encounters via ‘strength in numbers’37,39,41, we included the number of other adult male and other adult female participants (not including the individual itself) as two separated test predictors. Furthermore, to test whether predictability of coalitionary support and interaction exchange is a mechanism influencing intergroup encounter participation, we investigated the effects of social preference and relatedness on participation likelihood. Hence, we included the presence or absence of adult maternal kin and social bond partners (preferred grooming partner) in the encounter as additional test predictors. By testing the two hypotheses (i.e., number or preference of participants) in a single analysis we can evaluate their independent contribution to the response. To assure that participation patterns are not an artefact of association patterns with other participants and in order to account for geographic decay (i.e. decrease in social ties with increased spatial distance between individuals), we included the average of the dyadic association indices with all participants (i.e., spatial proximity) as a control predictor in both models. We controlled for an additional set of predictors that may affect participation likelihood. Individuals may gain from intergroup encounter participation, through inclusive fitness, if they have many relatives in the community. Thus, in both models we included the number of living maternal kin group members of all ages (for males—mother, maternal sibling, and maternal sister’s offspring; for females—offspring and daughter’s offspring) as a control predictor. We also included males’ paternity success assessed by the number of living offspring in the community at the time of the encounter, as a control predictor for males. These potential benefits follow previous studies22,25,49. Males are the philopatric sex in chimpanzees with high male–male competition over mates, and over time successful intergroup encounters may lead to territory expansion35 which may attract mating partners. If intergroup encounter participation facilitates future access to mates, increased reproductive opportunities could benefit males, but as well their mothers (through inclusive fitness). Thus, mothers may participate in intergroup encounters to foster future reproductive opportunities for their sons. Standard physical and behavioural criteria suggests that late juvenility/early adolescent life stages of male chimpanzees start at the age of 864,74, confirmed by physiological measures of the onset of puberty75. Thus, to investigate whether mothers are more likely to participate in intergroup encounters if they have sons approaching reproductive age, we included the presence of sons above the age of 8 in the group as a control predictor (independent son yes/no) for females. Energetic constraints also influence intergroup encounter participation across primate species, with the prediction that the costs are higher in older and lower ranking individuals (reduced physical condition)22,24–26. Therefore, in both models, we included the squared term of age and dominance rank of individuals as potential costs indicating individuals’ physical condition. In addition, we accounted for other potential energetic costs that may affect participation likelihood of females. Since pregnancy and early lactation are energetically demanding (considerably more in the first two years of lactation76) we controlled for the number of days into pregnancy and the presence of an offspring under 2 year of age as potential costs25,26. We estimated the onset of gestation by subtracting 228 days (average gestation length77) from parturition date. The year and month of birth was known for all offspring, and we assigned the 15th day of the month in cases where the day of birth was unknown. We as well accounted for females’ reproductive status (i.e., maximal tumescence yes/no) Finally, we accounted for fluctuations in fruit availability driving participation of males and females25. In both models we controlled for community membership (North, South or East) and included the type of the encounter, whether it was strictly vocal or involved visual or physical contact, the latter we assume to be riskier. We also added the log-transformed number of potential participants (i.e., adult community size) as an offset term. Adding the number of potential participants as an offset term allowed us to account for differences in participation likelihood due to overall community size (e.g., general participation might be higher/lower in smaller/larger communities22,27). Due to high correlation between the number of living maternal kin of all ages in the community and the presence of adult maternal kin in the encounter for males (Pearson’s R = 0.67) we could not include both in the ‘male participation’ model. Thus, for the males we fitted two models that were identical in all terms except of including either the number of maternal kin or the presence of adult kin as predictors. In the “Results” section we present the results of the model with the lower AIC score, as this model performs better at explaining the response78. Nonetheless, we provide full information on the results of the other model in Table 3. This was not a problem in the female model as the number of living maternal kin in the community was mostly influenced by the presence of non-adult offspring. Second ‘Male participation’ GLMM testing the effect of strength in numbers and predictability of coalitionary support on participation likelihood (Here, the number of living maternal kin is included instead of adult maternal kin presence in the encounter. N = 342 encounters, 36 male subjects). In bold appear the CIs that do not overlap 0. The coded levels are indicated in parenthesis. a–fz-transformed, mean ± SD of the original variables: a0.59 ± 1.05, b5.28 ± 3.53, c3.04 ± 1.29, d0.51 ± 0.18, e2.35 ± 2.38, f0.69 ± 0.26 (range 0-1 with 1 being the highest social rank), g21.31 ± 8.91, h1.81 ± 1.37. To avoid pseudo-replication and account for non-independent sampling of certain individuals or intergroup encounters disproportionally affecting participation likelihood, we included the identities of encounters and potential participants as random effects. To keep type I error rate at the nominal 5%, we included the maximal random slope structure79,80. We thus included random slopes for age (linear and squared), dominance rank, number of maternal kin, number of offspring for males, and average dyadic associations within both random effects. In addition, we included the number of male and female participants, fruit availability and number of days into gestation within the random effect of potential participant. See Supplementary Code 1 for the complete specification of the full and null models. Our datasets included 343 intergroup encounter events with 75 potential female and 36 potential male participants, of three communities, resulting in 3,799 participation decisions for females and 1,377 participation decisions for males. We processed the data and fitted all models in R (version 4.0.281), using the function lmer of the R package ‘lme4’82. Before fitting the models, we z-transformed all of the covariates to a mean of zero and a standard deviation of one83 and presented the original distribution of covariates in the table legends. We applied the function vif of the R package ‘car’84, to a standard linear model (lacking the random effects and slopes) to derive variance inflation factors (VIF), which revealed no collinearity issues (largest VIF: ‘male participation model’ = 2.11; ‘female participation model’ = 1.9385). Using a likelihood ratio test we evaluated model significance by comparing the fit of the full models with those of a respective null model lacking only the four test predictors86. To obtain individual p-values for all fixed effects we compared the full model with a series of models in which each fixed effect was systematically dropped one at a time, using the drop1 function in R79. Model stability, assessed by comparing the estimates of the full model with those derived from a series of models excluding the different levels of the two random effects (identities of potential participants and encounters) one after the other, revealed no influential identities. We report the confidence intervals which were derived by means of parametric bootstraps with the function bootMer of the package ‘lme4’. We calculated models‘ effect sizes (R²) using the function r.squaredGLMM from the R package’MuMIn’, and report the variance explained by the fixed (marginal-R²) and fixed and random (conditional-R²) effects87. Further information on research design is available in the Nature Research Reporting Summary linked to this article.