Polyadic Grooming Patterns and Network in a Free-Ranging Group of Japanese Macaques at Awajishima
ABSTRACT
Grooming is the most common affiliative behavior in many primate species. While traditionally viewed as an exclusively dyadic interaction, polyadic grooming involving more than two individuals also occurs. Few studies have explored the characteristics or functions of polyadic grooming in comparison with dyadic grooming. However, some studies suggest that polyadic grooming may have distinct characteristics and could be associated with social tolerance. We analyzed polyadic grooming patterns and its network structure in a free-ranging group of Japanese macaques at Awajishima, Japan. This group exhibited higher social tolerance than typical Japanese macaque groups. We found that polyadic grooming was common in this group, with higher frequency than other Japanese macaque groups as well as other primate species except chimpanzees. We also found that polyadic grooming typically occurred with multiple groomers sharing a single groomee, a pattern commonly observed across most primate species. Social network analyses showed that the polyadic grooming network was centralized around high-ranking males, reflecting the frequent grooming from multiple females to a single adult male. In addition, the network of polyadic grooming was less dense and more cliquish than that of dyadic grooming. This potentially suggested that polyadic grooming was more strongly kin-biased than dyadic grooming. Our results support the prediction that polyadic grooming may be associated with higher levels of social tolerance. Moreover, polyadic grooming may function to access valuable social partners more efficiently than dyadic grooming.
Summary
-
Polyadic grooming is frequently observed in a socially tolerant group of Japanese macaques and has distinct features compared with dyadic grooming.
-
The study highlights the potential importance of polyadic grooming for our understanding of primate society.
1 Introduction
Social grooming is one of the most frequently observed affiliative interactions among primates, which strengthens the social bonds and can be exchanged for various benefits such as agonistic support (Schino and Aureli 2008). While primates primarily rely on grooming for social bonding, humans use vocal communication, particularly conversation (Dunbar 1993). A key distinction is that grooming typically occurs between two individuals, whereas conversation can involve multiple participants simultaneously (Dunbar 1993). From this point, Dunbar (1993) theorized that human language evolved as a more efficient communication method, particularly in the form of gossip, which functions as “grooming at a distance” and enables us to maintain multiple social relationships simultaneously.
However, grooming involving more than two individuals (hereafter, polyadic grooming) occurs among various primates (e.g., Assam macaque (Macaca Assamensis): Cooper and Bernstein 2000; bonobo (Pan paniscus): Sakamaki 2013; Girard-Buttoz et al. 2020; chimpanzee (Pan troglodytes): Nakamura 2003; hanuman langur (Semnopithecus entellus): Sommer et al. 2002; Japanese macaque (Macaca fuscata): Nakamichi et al. 2020; long-tailed macaques (Macaca fascicularis): Mishra et al. 2020; rhesus macaque (Macaca mulatta): Mielke et al. 2021; tufted capuchin (Cebus apella): Parr et al. 1997), albeit at a lower frequency than dyadic grooming. Given that primates often engage in various types of polyadic affiliative interactions, such as huddling clusters (Zhang and Watanabe 2007), this prevalence of dyadic interactions in grooming behavior appears noteworthy. There are only a few studies that have focused on polyadic grooming per se (Girard-Buttoz et al. 2020; Nakamichi et al. 2020; Nakamura 2003). In most studies, when researchers observed polyadic grooming bouts, they divided a single bout into multiple dyadic grooming bouts (Mishra et al. 2020; Sommer et al. 2002) or simply discarded such data (Zhu et al. 2018). This indicates that researchers have paid little attention to this behavior, presumably due to its low frequency as well as the analytical challenges posed by polyadic interactions.
A fundamental unexplored question is what factors influence differences across populations and species in the prominence of polyadic grooming. While polyadic grooming is generally rare, its frequency varies among species. For example, in the genus Pan, polyadic grooming is commonly observed in chimpanzees (Nakamura 2003), while it is rarer in bonobos (Girard-Buttoz et al. 2020). Group differences are also evident: in rhesus macaques, no instances of polyadic grooming were observed in two groups living on Cayo Santiago Island during 688.3 h of observation (Brent et al. 2013), while it occurred frequently in a captive group (Mielke et al. 2021). This variation suggests the existence of factors that either constrain or facilitate polyadic grooming.
Another understudied aspect is whether polyadic grooming has any different characteristics or functions than dyadic grooming. Although few studies have compared polyadic with dyadic grooming, evidence suggests that polyadic grooming has distinct features. Nakamura (2003) reported that adult female chimpanzees spent more of their grooming time in large polyadic clusters than in dyadic sessions. In bonobos, polyadic grooming increased the number of grooming partners accessible per minute, suggesting that it is a more efficient form of social interaction than dyadic grooming for them (Girard-Buttoz et al. 2020).
Japanese macaques, one of the most studied primate species, typically engage in dyadic grooming, with a proportion of polyadic grooming reported to be less than 2% (Fujimoto and Takeshita 2007; Nakamichi et al. 2020). However, the Awajishima group, a provisioned group on southern Awajishima Island, seems to show exceptionally frequent polyadic grooming (Nakamichi and Yamada 2010). This group is also notable for their higher levels of social tolerance than typical Japanese macaques, particularly exemplified by peaceful co-feeding behavior (Kaigaishi et al. 2019). Nakamichi et al. (2020) proposed that the typically low frequency of polyadic grooming in Japanese macaques reflects their low social tolerance, as polyadic grooming requires multiple individuals to be in close proximity. In fact, the strength of social bonds or kinship plays a crucial role in the occurrences of polyadic grooming in various primate species (Girard-Buttoz et al. 2020; Mielke et al. 2021; Nakamichi et al. 2020). The high social tolerance of the Awajishima group may facilitate their frequent polyadic grooming, although quantitative data are still needed.
In this study, we quantitatively analyzed polyadic grooming patterns in the Awajishima group of Japanese macaques. We predicted that polyadic grooming would be observed more frequently in the Awajishima group than in the other Japanese macaque groups, although the observational methods differed across the studies (Fujimoto and Takeshita 2007; Nakamichi et al. 2020; see Table 2 for detail). We also analyzed the form of each grooming clique. Polyadic grooming can take two forms: (1) multiple individuals grooming a single partner simultaneously (e.g., A → B ← C, defined as the focused type), or (2) one individual grooming another while being groomed by a third (e.g., A → B → C, defined as the chained type). Although not many studies have paid attention to this classification, there appears to be a species-specific bias about the form of polyadic grooming: chimpanzees and bonobos more frequently engage in the chained type, where individuals form a chain of grooming. In contrast, Japanese macaques are more likely to form the focused type, in which multiple individuals groom one partner (Nakamichi et al. 2020; Nakamura 2003; Sakamaki 2013). We predicted that the macaques in the Awajishima group would follow this pattern, engaging in the focused type more frequently. In addition, we used social network analysis to capture the differences in social interaction patterns between dyadic and polyadic grooming. Since polyadic grooming is likely to face stronger social constraints than dyadic grooming in Japanese macaques, we predicted that these constraints would influence partner selection and social interaction patterns, resulting in distinct characteristics for each grooming network structure. Specifically, we hypothesized that individuals playing central roles in social networks would differ between dyadic and polyadic networks. In dyadic networks, high-ranking individuals, especially males, were expected to exhibit higher centrality, as they typically play central roles in Japanese macaque social network (Sueur et al. 2011). Conversely, in polyadic networks, such individuals would likely show lower centrality, as males tend to participate less frequently in polyadic grooming (Nakamichi et al. 2020). Furthermore, we predicted that polyadic networks would display lower cohesiveness (i.e., lower density and higher modularity) compared to dyadic networks, as polyadic grooming tended to occur more frequently among closely bonded individuals in rhesus macaques (Mielke et al. 2021) and between kin-related females in Japanese macaques (Fujimoto and Takeshita 2007; Nakamichi et al. 2020).
2 Methods
2.1 Study Group, Subjects, and Data Collection
We conducted this study on a free-ranging group of Japanese macaques inhabiting the southern part of Awajishima Island, Hyogo prefecture, Japan (34°14'38.4“N, 134°52'58.9“E). This group has been artificially provisioned by the Awajishima Monkey Center since 1967. The group typically stays at the feeding site from morning until evening (approximately 5 PM), except during early summer and autumn when the group rarely visits the feeding site and ranges in the forest for abundant natural food. The macaques are provisioned three times a day with wheat and soybeans. Behavioral study at this field site began in 1976 (Nakamichi et al. 1983), and information on kin relationships is presently known only partially. Group size was estimated to be 399 in 2017 (47 adult males [> 5 years], 181 adult females, 58 subadults [4-5 years] and 113 juveniles [1-3 years] and infants [0 years]), 381 in 2018 (35 adult males, 186 adult females, 60 subadults and 100 juveniles and infants), 471 in 2019 (55 adult males, 223 adult females, 53 subadults and 140 juveniles and infants), and 454 in 2020 (45 adult males, 227 adult females, 42 subadults and 140 juveniles and infants).
We conducted behavioral observations from June 2017 to July 2020, 289 days in total (73 days in 2017, 126 days in 2018, 55 days in 2019, and 35 days in 2020). Since the study group was provisioned, most macaques remained in the vicinity of the feeding site, which enabled us to easily observe them throughout the day. We primarily collected data during the summer and winter seasons, encompassing both the mating season (approximately September to March) and birth season (approximately April to August). During early summer and autumn, when the group rarely visited the feeding site, we collected data only when the whole group remained for sufficient time to record grooming relationships. We selected adult males and sexually mature females (including all adult females and some subadult females which gave birth at the age of 5), excluding males present only during mating seasons. The grooming cliques involving such males were not recorded. This yielded 216 subjects in 2017 (male: 22; female: 194), 218 in 2018 (male: 22; female: 192), 214 in 2019 (male: 22; female: 192), and 220 in 2020 (male: 20; female: 200). For kinship analyses, we classified dyads with maternal relatedness ≥ 0.125 as kin (mother-daughter, sister-sister, grandmother-granddaughter, and aunt-niece relationships). However, matrilineal relationships were only partially known within the group: among the study subjects, only 95 individuals had known matrilineal information, yielding 449 kin dyads. We excluded individuals with unknown maternal kinship from the analyses examining the effect of kinship on grooming relationships. No paternal kinship data were available in our study group.
We recorded grooming cliques (a configuration of individuals directly connected through grooming: Nakamura 2003) via line census along a predetermined route, as the macaques were dispersed across the feeding site. We chose the route to ensure a complete circumnavigation of the feeding site, allowing us to observe all areas where the macaques were dispersed. We began the census when the whole group was resting around the feeding site. We did not conduct observations 30 min before and after provisioning by the park manager or when tourists intensively fed the macaques, because most macaques gathered around limited area for feeding during that time. The interval between the censuses was at least 1 h to maximize the independence of each census. During each census, we recorded all visible grooming cliques involving only adults, noting the clique type, the identity of participants, and their roles (groomer or groomee) at first encounter. We did not record subsequent observations of the same clique, and when individuals participated in multiple cliques during a census, we recorded only their first participation. We conducted a total of 871 censuses (3.1 censuses per day on average). The mean duration of each census was 25.3 (SD ± 7.8) min. Due to the nature of direct field observations, observers were aware of the research questions during data collection. Regarding the type of grooming clique, we defined four distinct types (see Figure S1 for the photo of each type): dyadic (a clique involving two individuals), focused (multiple groomers directed their grooming toward one groomee), chained (one individual grooms another, who, in turn, grooms a third individual), and compound (a polyadic grooming clique where both focused and chained types coexisted).
2.2 Social Network Analysis
From the recorded grooming interactions, we constructed asymmetric matrices representing groomer-groomee relationships, where matrix elements indicated grooming frequencies per dyad. We analyzed directed and weighted networks of dyadic and focused types only, excluding chain and compound types because of their extremely low frequency (see Results). However, we also conducted a complementary analysis in which we analyzed polyadic grooming networks combining all polyadic grooming interactions of any kind (see supplemental material S1 and Table S1 for details). Separate analyses were conducted for each observation year (2017-2019) due to slight changes in group composition, resulting in six social networks.
For individual-level metrics, we calculated in-strength, out-strength, and eigenvector centrality. Out-strength represents the number of grooming an individual performed, while in-strength indicates the number of grooming interactions an individual received. Thus, these centrality measures suggest how actively an individual was involved in grooming interactions, either as a groomer (out-strength) or as a recipient (in-strength). Eigenvector centrality evaluates an individual's social position by weighing not only their own direct connections but also the connectivity of their social partners throughout the network. Individuals with higher eigenvector centrality can be interpreted to be more socially influential, possessing higher social capital (Brent et al. 2011). We predicted that in dyadic grooming, high-ranking males would primarily serve as grooming recipients (higher in-strength) while females would act as groomers (higher out-strength). In contrast, in polyadic grooming, males would be less involved overall (lower in-/out-strength), with females taking more active roles (higher in-/out-strength). Additionally, we expected higher-ranking males to occupy core network positions in dyadic grooming networks, whereas females would take central roles in polyadic grooming networks (higher eigenvector centrality). At the group level, we analyzed density and modularity. Density represents the ratio of observed to possible relationships (range: 0-1); a density value close to 0 indicates that only a limited number of potential relationships were realized, while values approaching 1 indicate that most possible relationships occured in the network. Modularity quantifies the degree to which the network is divided into distinct communities (range: 0–1), with higher values indicating a more pronounced community structure where individuals interact more frequently within than between communities. Such a community structure may reflect, for example, preferential associations among kin-related individuals.
2.3 Hierarchical Rank Estimation
We calculated the hierarchical rank of each individual based on the results of food-dominance tests and dyadic aggression in non-feeding contexts observed from 2017 to 2020. In the food-dominance tests, we placed a piece of food between two individuals sitting 1–2 m apart and recorded which individual took it. We conducted the test when both individuals were attentive to the food, and we placed the food equidistant from the two individuals. If one individual took the food, with or without aggression, while the other made no attempt to do so, the former was recorded as dominant and the latter as subordinate. The tests were conducted once or multiple times per dyad (1.65 times for females and 3.01 times for males on average). For aggressive interactions in non‑feeding contexts, we used ad libitum sampling to record the identity of the aggressor (defined as the individual displaying any aggressive behavior such as threatening and biting) and the victim (the individual receiving the aggression). We excluded interactions involving any form of counter-aggression by the victim or those involving more than two individuals. Because male Japanese macaques are generally dominant over females (Johnson et al. 1982), and because both our direct observations of win-loss interactions and dominance rank calculations showed that all males were dominant over all females (including the highest ranking ones) in the Awajishima group during the study period (Kaigaishi, personal observation), we analyzed male and female dominance ranks separately. Consequently, we recorded 590 interactions for males (26 aggressive interactions and 564 food-dominance tests: accounting for 38.7% of all possible combinations) and 14270 interactions for females (657 aggressive interactions and 13613 food-dominance tests: accounting for 41.5% of all possible combinations). Of these, both food-dominance tests and aggressive interactions were recorded in 403 pairs. The two indicators yielded consistent results in 95% (382/403) of these cases, demonstrating concordance in dyadic dominance relationships. In addition, the outcomes of all win-loss interactions (food-dominance test and aggressive interactions) were highly unidirectional (98% of interactions occurred in the same direction within pairs). Consequently, we decided to integrate the results from both indicators into the hierarchical rank calculation.
We applied the percolation-conductance method for rank calculation, which is conservative for data containing many unknown dominance relationships (Fushing et al. 2011). This method determines individual ranks based on both direct (A > B) and indirect (A > C > D > B) win-loss interactions. By incorporating all these relationships, it calculates dominance certainty for each dyad, the probability that one individual is dominant over (or subordinate to) another, ranging from 0.5 to 1, with higher values indicating clearer dominance relationships (Fushing et al. 2011). Through 1000 random walk permutations, the method assigns unique ranks to individuals (Fujii et al. 2021). In our data set, the average dominance certainty was high for both females (N = 19110; 0.96 ± 0.09) and males (N = 253; 0.95 ± 0.10), while the linearity being 0.99 both for females and males. These results indicate that individuals in this group possessed clear dominance ranks while the rank structure exhibited extremely high linearity, as observed in typical despotic Japanese macaque groups (Nakamichi et al. 1995). It is also notable that when analyzing silent bared-teeth displays in another research project, we found they were consistently directed from the lower- to higher-ranking individuals, further validating our calculated dominance hierarchy (Kaigaishi, unpublished data). Since ranks were calculated separately for each sex, we rescaled them to range from 0 (lowest) to 1 (highest) for use in statistical models.
2.4 Statistical Analysis
To investigate whether kinship had a stronger effect on the probability of polyadic grooming compared to dyadic grooming, we ran a multinomial logistic regression with random effects. The response variable was the type of grooming (‘dyadic’, ‘focused’, ‘others’ [combining ‘chained’ and ‘compound’ types due to small sample sizes]), with ‘dyadic’ as the reference category. The fixed effect was whether each groomer-groomee dyad was kin (1) or non-kin (0). We included the ID of grooming clique, line census session, grooming pair, groomer, and groomer as random effects to account for multiple dyads in the polyadic grooming cliques. We excluded dyads with unknown kinship and those involving males, as maternal kinship data were available only for females.
For the social centrality measures, we aimed to examine how sex and rank could be associated with social centralities based on the different types of grooming (namely, dyadic type and focused type). Accordingly, we ran GLMMs with the value of each social centrality as the response variable (with Poisson distributions for in-strength (total number of records of received grooming) and out-strength (total number of records of grooming given) and a beta distribution for eigenvector centrality (0–1 normalized score)). We constructed models for each of the 3 social centrality measures of two types of grooming networks, resulting in 6 models. The models included sex, rank score, and the interaction between them as fixed effects, and year and individual ID as random effects. We interpreted the coefficients of the main effects (sex and rank score) only when the interaction term was not significant.
To address the nonindependence of social network metrics, we ran the GLMMs with a node-label permutation procedure (Farine and Whitehead 2015). We reshuffled the sex and rank attributes of each individual while maintaining the whole network structure, repeating this process 10000 times. Then, we ran the GLMMs on the 10,000 reshuffled datasets and compared the model coefficients from the observed data with those from the permuted data. The effects of the model parameters were considered significant if the permuted model coefficients were larger or smaller than the observed model appearing less than 500 times (i.e., p < 0.05). Moreover, since we repeatedly applied GLMMs using the same set of predictor variables, we needed to control for multiple comparisons across the 6 GLMMs (2 networks × 3 centrality measures). We applied the Benjamini-Hochberg procedure to adjust the permutation-based p-values, while maintaining false discovery rate at 0.05.
3 Results
3.1 Frequency, Variation, and Characteristics of Polyadic Grooming Patterns
Polyadic grooming was widespread in the study group across the study period: 184 individuals (85.1%) in 2017, 208 individuals (93.7%) in 2018, and 181 individuals (84.6%) in 2019 were observed to engage in such interactions. Only five individuals were never found to be involved in any polyadic grooming cliques, and these were all lowest-ranking males. Of the 13,438 grooming cliques recorded, 10.0% (1,348) were polyadic, comprising 11 distinct forms (Figure 1). The focused type was predominant (95.6%), whereas the chained and compound types were much rarer (3.7% and 0.7%, respectively). The largest was the focused type involving six individuals ([12] in Figure 1). Quadratic chains of grooming were extremely rare (two cliques; [5] and [10] in Figure 1).
Regarding sex composition within cliques, in dyadic grooming, female-to-female cliques were the most common (85.8%, N = 12090), followed by female-to-male (10.4%), male-to-female (2.9%) and male-to-male interactions (0.8%). Similarly, in polyadic grooming, females were the primary participants across all types. In the focused type, the most common clique was the one involving only females (i.e., multiple females grooming a female recipient: 73.7%, N = 1288), followed by multiple females grooming a male (21.7%). Males rarely participated in focused type grooming as groomers (4.3%), and cliques involving only males were particularly scarce (0.3%). The chained type had a similar female-biased pattern, with female-only chains accounting for 39 of 49 cliques (79.6%). The remaining cliques involved only a single male, and chains involving multiple males were never observed. Compound type consisted almost exclusively of female participants (9 of 10 cliques). There was one clique that involved a single male and multiple females.
Our multinomial logistic regression model showed that, among females, kinship significantly and positively influenced the probability of both focused type (β = 0.830, z = 7.195, p < 0.001) and the ‘other’ category of grooming (comprising chained and compound types; β = 0.799, z = 2.189, p = 0.029) relative to dyadic grooming. However, as a substantial proportion of dyads had unknown kinship relationships across all types, these patterns should be interpreted with caution. The full model results are shown in the supporting material (Table S2).
3.2 Topological Features of Dyadic and Focused Grooming Networks
The dyadic networks consistently showed higher density values (2017: 0.05; 2018: 0.10; 2019: 0.06) than the focused ones (2017: 0.02; 2018: 0.03; 2019: 0.02), indicating that grooming spanned a broader range of individuals in the dyadic networks. Conversely, the modularity score was higher for the focused type (2017: 0.64; 2018: 0.55; 2019: 0.64) than for the dyadic networks (2017: 0.26; 2018: 0.22; 2019: 0.27). To address whether the higher modularity in focused networks might be an artifact of their lower interaction frequency (approximately 10 times fewer interactions than dyadic networks), we conducted a subsampling analysis. We repeatedly sampled the dyadic networks down to the same size as the focused networks (1000 iterations) and compared the resulting modularity values. The analysis revealed that the focused networks exhibited significantly higher modularity than would be expected from random subsamples of the dyadic networks of equivalent size (p < 0.001 for all networks; see Figure S2). These results suggest that focused type grooming tended to occur more frequently among individuals within specific subgroups, with less common interactions between different subgroups (see Figure S3 for the graphical representation of each network). Results were the same for the network where all types of polyadic cliques were involved (Supplemental Material S1).
3.3 Effects of Sex and Rank on Social Centrality Measures in Dyadic and Focused Type Grooming
After controlling the p-values for multiple comparisons, we found different patterns in the two grooming networks (Table 1). In the dyadic network, a significant interaction effect of sex and rank emerged only in the in-strength model, indicating that higher-ranking males were particularly likely to be the target of grooming (Figure 2). Hierarchical rank showed consistently positive main effects on the other centrality measures, indicating that higher-ranking individuals played important roles in the dyadic network. Males exhibited lower out-strength centrality than females, suggesting that they were less likely to perform groomer roles in dyadic grooming. The main effect of sex was not significant in the eigenvector centrality model (Figure 3).
| Dyadic network | Focused type network | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Centrality | Effect | Coefficient | Standard error | Z value | p value | Coefficient | Standard error | Z value | p value |
| In-strength | Intercept | 2.387 | 0.235 | 10.174 | 1.000 | 0.631 | 0.196 | 3.218 | 1.000 |
| Rank | 0.422 | 0.098 | 4.306 | < 0.001 | 0.638 | 0.211 | 3.025 | 0.053 | |
| Sex | −0.499 | 0.212 | −2.358 | < 0.001 | 0.139 | 0.411 | 0.339 | 0.468 | |
| Rank × Sex | 1.068 | 0.332 | 3.218 | < 0.001 | 1.210 | 0.649 | 1.863 | 0.032 | |
| Out-strength | Intercept | 2.346 | 0.237 | 9.896 | 1.000 | 0.380 | 0.208 | 1.823 | 1.000 |
| Rank | 0.594 | 0.120 | 4.931 | < 0.001 | 1.379 | 0.204 | 6.754 | < 0.001 | |
| Sex | −1.056 | 0.283 | −3.725 | < 0.001 | −0.457 | 0.504 | −0.906 | 0.006 | |
| Rank × Sex | 0.035 | 0.449 | 0.078 | .915 | −1.771 | 0.821 | −2.157 | 0.003 | |
| Eigenvector centrality | Intercept | −3.552 | 0.166 | −21.436 | < 0.001 | −5.017 | 0.202 | −24.857 | < 0.001 |
| Rank | 3.043 | 0.225 | 13.504 | < 0.001 | 2.271 | 0.278 | 8.159 | 0.006 | |
| Sex | −0.037 | 0.481 | −0.077 | .915 | −0.529 | 0.691 | −0.766 | 0.033 | |
| Rank × Sex | −0.006 | 0.745 | −0.008 | .990 | 2.188 | 1.032 | 2.121 | 0.043 | |
- Note: p values are adjusted for multiple testing using false discovery rate correction. Positive coefficients for Sex indicate higher values in males than females. Values in bold indicate p < 0.05.
In the focus between sex and rank was significant for all types of social centralities (Figure 2). While the interaction term had a negative effect in out-strength models, it showed a positive effect in in-strength models. These results suggest that, in focused type grooming, higher-ranking females actively took groomer roles while higher-ranking males were particularly likely to be chosen as grooming partners. Moreover, a significant and positive interaction between sex and rank was observed in the eigenvector centrality model, suggesting that high-ranking males particularly occupied central positions in the overall grooming network, forming social connections with other socially central individuals. For the main effects of sex and rank in the models with significant interaction effects, see Figures S4–S7. The GLMMs for the compound polyadic network (combining all polyadic grooming types) yielded identical trends (see Supplemental Material S1).
4 Discussion
We compared the observed pattern of polyadic grooming in the Awajishima group with quantitative data from various primate species reported in previous studies (Table 2). Direct comparisons and statistical analyses were difficult due to the variation in observational methods across the studies. Nevertheless, Table 2 indicates that the proportion of polyadic grooming in the Awajishima group was relatively higher than those in previous studies. In particular, the Awajishima macaques engaged in polyadic grooming more frequently than two other Japanese macaque groups (1.6% in Arashiyama: Fujimoto and Takeshita 2007; 1.7% in Katsuyama: Nakamichi et al. 2020), supporting previous observations of higher polyadic grooming frequency in this group (Nakamichi and Yamada 2010). The Awajishima macaques were also exceptional in their variation of clique types and sizes, showing the second largest variation after chimpanzees (Nakamura 2003) and frequently forming cliques of four or more individuals. In contrast, other species showed only 3–7 types, and cliques larger than three individuals were absent or extremely rare. These results suggest that polyadic grooming may be as common a daily social interaction for the Awajishima group as it is for chimpanzees (Nakamura 2003).
| Species | Proportion of polyadic grooming in all grooming | The number of observed clique types | The largest clique size | Proportion of focused type in all polyadic grooming | Proportion of chained type in all polyadic grooming | Analytical unit | Observational 0method | Study population | Citation |
|---|---|---|---|---|---|---|---|---|---|
| Macaca fuscata | 10.0% (1348/13433 cliques) | 11 | 6 | 95.6% | 3.7% | Frequency | Line census | Awajishima | This study |
| Macaca fuscata | 1.6% (19/1179 episodes) | Not described | 3 | Not described | Not described | Frequency | Focal sampling | Arashiyama E | Fujimoto and Takeshita (2007) |
| Macaca fuscata | 1.7% (193/11100 bouts) | 4 | 4 | 72.9% | 26.1% | Frequency | Line census and Ad libtum samplinga | Katsuyama | Nakamichi et al. (2020) |
| Macaca assamensis | 7.2% (390/5397 episodes) | 3 | 4 | 100.0% | 0.0% | Frequency | Ad libitum sampling | Tukeswari temple | Cooper and Bernstein (2000) |
| Macaca fascicularis | 1.1% (12/1047 bouts) | Not described | Not described | Not described | Not described | Frequency | Focal sampling | Campbell Bay TR | Mishra et al. (2020) |
| Macaca mulatta | 0.0% (total duration of grooming is not described) | 0 | 2 | — | — | Duration | Focal sampling | Cayo Santiago V | Brent et al. (2013) |
| Macaca mulatta | 0.0% (Total duration of grooming is not described) | 0 | 2 | — | — | Duration | Focal sampling | Cayo Santiago F | Brent et al. (2013) |
| Macaca mulatta | 5.2% (77/1488 episodes) | 5 | 4 | 76.6% | 22.1% | Frequency | Scan sampling | Yerkes Regional Primate Research Center | Maestripieri (2000) |
| Colobus guereza | 0.0% (0/13 bouts) | 0 | 2 | — | — | Frequency | Focal sampling | Bole Valley 1a | Dunbar (2022) |
| Colobus guereza | 0.0% (0/7 bouts) | 0 | 2 | — | — | Frequency | Focal sampling | Bole Valley 1b | Dunbar (2022) |
| Colobus guereza | 0.0% (0/10 bouts) | 0 | 2 | — | — | Frequency | Focal sampling | Bole Valley 3a | Dunbar (2022) |
| Colobus guereza | 0.0% (0/9 bouts) | 0 | 2 | — | — | Frequency | Focal sampling | Bole Valley 4 | Dunbar (2022) |
| Colobus guereza | 0.0% (0/18 bouts) | 0 | 2 | — | — | Frequency | Focal sampling | Bole Valley 5 | Dunbar (2022) |
| Colobus guereza | 0.0% (0/24 bouts) | 0 | 2 | — | — | Frequency | Focal sampling | Bole Valley 8 | Dunbar (2022) |
| Pan paniscus | 3.5% (Total duration of all cliques was 6923.5 min) | 7b | 4 | 37.4% | 60.9% | Duration | Focal sampling | Wamba E1 | Sakamaki (2013) |
| Pan paniscus | 7.4% (Total duration of all cliques was 67.5 h) | Not described | Not described | Not described | Not described | Duration | Focal sampling | LuiKotale Bompusa | Girard-Buttoz et al. (2020) |
| Pan troglodytes | 15.8% (Total duration of all cliques was 137 h) | 27b | 7 | Not described | Not describedc | Duration | Focal sampling | Mahale M | Nakamura (2003) |
| Pan troglodytes | 15.1% (Total duration of all cliques was 37.7 h) | Not described | Not described | Not described | Not described | Duration | Focal sampling | Taï South | Girard-Buttoz et al. (2020) |
| Pan troglodytes | 15.4% (Total duration of all cliques was 35.6 h) | Not described | Not described | Not described | Not described | Duration | Focal sampling | Taï East | Girard-Buttoz et al. (2020) |
| Rhinopithecus bieti | 0.3% (Total number of bouts is not described) | Not described | 3 | Not described | Not described | Duration | Focal sampling | Xiangguqing | Zhu et al. (2018) |
| Theropithecus gelada | 1.6% (3/186 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber N39 | Dunbar (2022) |
| Theropithecus gelada | 1.6% (12/745 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber N16 | Dunbar (2022) |
| Theropithecus gelada | 0.9% (7/739 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber N2 | Dunbar (2022) |
| Theropithecus gelada | 1.6% (5/320 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber N19 | Dunbar (2022) |
| Theropithecus gelada | 2.5% (2/81 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber N20 | Dunbar (2022) |
| Theropithecus gelada | 0.0% (0/397 bouts) | 0 | 2 | — | — | Frequency | Scan sampling | Sankaber N14 | Dunbar (2022) |
| Theropithecus gelada | 0.7% (4/536 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber N21 | Dunbar (2022) |
| Theropithecus gelada | 0.0% (0/131 bouts) | 0 | 2 | — | — | Frequency | Scan sampling | Sankaber N1 | Dunbar (2022) |
| Theropithecus gelada | 0.5% (1/188 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber N17 | Dunbar (2022) |
| Theropithecus gelada | 0.4 (4/912 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber N12 | Dunbar (2022) |
| Theropithecus gelada | 0.7% (2/292 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber N5 | Dunbar (2022) |
| Theropithecus gelada | 0.4% (1/267 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber N28 | Dunbar (2022) |
| Theropithecus gelada | 2.6% (15/580 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber N27 | Dunbar (2022) |
| Theropithecus gelada | 1.3% (8/613 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber N13 | Dunbar (2022) |
| Theropithecus gelada | 0.5% (1/218 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber H39 | Dunbar (2022) |
| Theropithecus gelada | 0.0% (0/48 bouts) | 0 | 2 | — | — | Frequency | Scan sampling | Sankaber H31 | Dunbar (2022) |
| Theropithecus gelada | 0.6% (1/174 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber H22 | Dunbar (2022) |
| Theropithecus gelada | 0.0% (0/207 bouts) | 0 | 2 | — | — | Frequency | Scan sampling | Sankaber H50 | Dunbar (2022) |
| Theropithecus gelada | 0.6% (1/160 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber H17 | Dunbar (2022) |
| Theropithecus gelada | 3.9% (9/233 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber H74 | Dunbar (2022) |
| Theropithecus gelada | 0.0% (0/104 bouts) | 0 | 2 | — | — | Frequency | Scan sampling | Sankaber H60 | Dunbar (2022) |
| Theropithecus gelada | 0.8% (1/131 bouts) | Not described | 3 | Not described | Not described | Frequency | Scan sampling | Sankaber H51 | Dunbar (2022) |
- a The authors recorded dyadic grooming employing line census and polyadic grooming employing Ad lib tum sampling.
- b The author included mutual grooming clieques in which two individiuals simultaneoulsly groomed each other.
- c While the proportion of the focused or the chained type in clieques larger than three is not described, the author mentioned that chained type occurred longer than the focused one in triadic grooming cliques.
It is important to discuss the possibility that our data collection method affected the results. Firstly, we recorded the grooming cliques during line census, but could not record which individuals were absent during each census. Since our group was provisioned and the whole group usually remained around the feeding site during daytime, we are confident that our observations captured the majority of grooming interactions. However, this methodology may have introduced unintentional bias into our results, such as the potential underrepresentation of peripheral individuals ranging outside the main area. This methodological limitation may have led to underestimation of social centralities for such individuals. Secondly, we conducted observations primarily during the season when the group heavily relied on artificially provisioned food. Some studies suggest that food provisioning alters activity budgets in primates, potentially increasing time for social interactions (Wong 2019). Consequently, this modified activity budget may have enhanced the likelihood of observing polyadic grooming in our study group. Nevertheless, it is noteworthy that polyadic grooming remained extremely rare in two other provisioned Japanese macaque groups (Fujimoto and Takeshita 2007; Nakamichi et al. 2020).
Although quantitative information on the form of polyadic grooming is still limited, Table 2 suggests that the focused type is generally more common than the chained type except in Pan species, which exhibit the opposite pattern. This predominance of focused type grooming may reflect cognitive constraints, as individuals in the chained type must simultaneously take both groomer and groomee roles (Nakamichi et al. 2020; Nakamura 2003). An observational study suggested that taking both roles within one grooming bout may be cognitively challenging even for chimpanzees (Zamma and Nakamura 2015). Interestingly, the overlap of groomer and groomee roles is evident even in mutual grooming (two individuals sitting face-to-face simultaneously groom each other), which is common in chimpanzees (Nakamura 2003) but rare in most other primate species including our study group (Kaigaishi, unpublished data). However, a study reported that it is frequent in small-brained red-fronted lemurs (Eulemur fulvus rufus) (Port et al. 2009), which suggests that mutual grooming may not be so cognitively demanding for primates. Thus, the potentially higher cognitive load of chained grooming in comparison with focused grooming might stem not from the mere role overlaps but rather from the requirements to simultaneously interact with two partners. Consequently, for most primates, focused type grooming may function as a means of social interaction that is more efficient than dyadic grooming and imposes lower cognitive loads than the chained type.
The high frequency of polyadic grooming in the Awajishima group, characterized by notably high levels of social tolerance (Kaigaishi et al. 2019), supports the prediction that higher social tolerance may enhance polyadic grooming (Nakamichi et al. 2020). However, this relationship between social tolerance and grooming patterns may vary across phylogenetic groups or the type of grooming. For example, in the two Pan species, chimpanzees, the less tolerant species, engage in more frequent polyadic grooming than the more tolerant bonobos (Nakamura 2003; Sakamaki 2013). This appears to go against the prediction about social tolerance and polyadic grooming. However, it is notable that both species show preferences for the chained over the focused type grooming, an exceptional pattern amongst primates. While this preference for chained type in Pan species may reflect their unique cognitive abilities, in most other primate species where focused type grooming is predominant, tolerance levels may be likely to have a greater impact on the frequency of the focused type rather than the chained type. Since grooming can be exchanged for various benefits such as agonistic aid (Schino and Aureli 2008), primates can compete for valuable grooming partners. Therefore, “sharing” grooming partners in focused type grooming may lead to competition for valuable social resources among the groomers, thus higher frequency of this type of grooming in more tolerant species. This competition for social partners may also explain why grooming shows a stronger bias toward dyadic interactions compared to other affiliative behaviors such as huddling, where all participants can simultaneously gain benefits (e.g. thermoregulation) without competing for access to a limited resource.
Another factor potentially influencing the high frequency of polyadic grooming in the Awajishima group is its extremely large group size. Dunbar (2022) suggested that polyadic grooming might help meet increased socialization demands in large groups. Given that polyadic grooming enhanced grooming efficiency in bonobos, measured as the number of grooming partners accessed per minute (Girard-Buttoz et al. 2020), future studies should examine whether (1) polyadic grooming occurs more frequently in larger groups, and (2) polyadic grooming increases the number of accessible partners.
Kinship may be more strongly associated with polyadic than dyadic grooming. Although the interpretation needs caution because most groomer-groomee dyads had unknown kin relationships in this group, this pattern was also observed in other Japanese macaque groups: all triadic grooming occurred between kin-related females in Arashiyama (Fujimoto and Takeshita 2007), and focused type grooming was likely to occur between kin-related dyads in Katsuyama (Nakamichi et al. 2020). Moreover, the strong kin bias in polyadic grooming appears to be reflected in the topological features of its social network. The focused network exhibited a lower density and higher modularity than the dyadic network, suggesting that polyadic grooming was more likely to occur within restricted sets of individuals than dyadic grooming. Therefore, kinship may affect polyadic grooming even more strongly than dyadic grooming in Japanese macaques, although more detailed kin estimation is needed.
Contrary to the prediction, focused type grooming involving adult males was common in the Awajishima group, with the most common mixed-sex clique consisting of multiple females simultaneously grooming a single male. GLMMs showed that higher-raking males were likely to be the target of grooming both in dyadic and focused type grooming (i.e., higher in-strength), and played the most central roles particularly in the focused type network (higher eigenvector centrality). Supporting our prediction, females were active groomers both in dyadic and focused type grooming, with high-ranking females performing more grooming in the focused type network (higher out-strength). These results suggest that, in the Awajishima group, focused type grooming may serve as a specific social strategy for high-ranking females. Given that females benefit from social bonds with males through social support and tolerance (Maestripieri 2000), this simultaneous grooming may allow multiple females to efficiently invest in these valuable social relationships.
In future studies, it will be important to focus on how polyadic grooming occurs. While some studies have analyzed third-party grooming interventions, few have studied the occurrence of polyadic grooming (Mielke et al. 2021; Mielke et al. 2017). They found polyadic grooming (1) frequently occurred in tolerant chimpanzees but not in intolerant mangabeys (Cercocebus atys) (Mielke et al. 2017), and (2) was more likely to occur between familiar dyads in rhesus macaques (Mielke et al. 2021). Extending this approach to other species will elucidate the factors involved in polyadic grooming. In tolerant species, third-party intervention may result in more polyadic grooming than disturbing the dyadic grooming. Although often overlooked in studies of social affiliation, polyadic grooming may hold significant importance in understanding primate societies.
Acknowledgments
We express our sincere appreciation for the managers of the Awajishima Monkey Center for their great support of our study. We would also like to thank the members of the Department of Ethology, Graduate School of Human Sciences, Osaka University. Y.K. personally thanks Prof. Yasunobu Yasoshima and Prof. Tomohide Atsumi in Osaka University for their insightful comments.
Ethics Statement
This study was approved by the animal experimentation committee of Graduate School of Human Sciences, Osaka University (Doujinka29-1-0). We declare that this study was conducted in full compliance with the American Society of Primatologists Principles for the Ethical Treatment of Non-Human Primates and Code for Best Practices in Field Primatology (Riley et al. 2014).
Conflicts of Interest
The authors declare no conflicts of interest.
Open Research
Data Availability Statement
The data set supporting the current analyses is available in Zenodo (https://doi.org/10.5281/zenodo.15293877).
