Social stability via management of natal males in captive rhesus macaques (Macaca mulatta)

Abstract

Keystone individuals are expected to disproportionately contribute to group stability. For instance, rhesus macaques (Macaca mulatta) that police conflict contribute towards stability. Not all individuals’ motivations align with mechanisms of group stability. In wild systems, males typically disperse at maturity and attempt to ascend via contest competition. In a captive system, dispersal is not naturally enabled – individuals attempt to ascend in their natal groups, which can be enabled by matrilineal kin potentially destabilizing group dynamics. We relocated select high-ranking natal males from five groups and assessed group stability before and after. We quantified hierarchical metrics at the individual and group level. After removal, we found significantly higher aggression against the established hierarchy (reversals), indicative of opportunistic attempts to change the hierarchy. Mixed-sex social signaling became more hierarchical, but the strength of this effect varied. Stable structure was not uniformly reached across the groups and alpha males did not all benefit. Indiscriminate natal male removal is an unreliable solution to group instability. Careful assessment of how natal males are embedded within their group is necessary to balance individual and group welfare.

INTRODUCTION

Social stability is an emergent group-level phenomenon that can be difficult to assess (Beisner et al., 2015; McCowan et al., 2008; Oates-O’Brien et al., 2010). Stable groups exhibit a robust ability to adapt to daily perturbations through conflict policing [Flack, Krakauer, et al., 2005]), while failure to maintain stability results in group fission or hierarchical collapse (Beisner et al., 2015; Oates-O’Brien et al., 2010). Just as the presence of keystone individuals can support group stability by investing in conflict policing, we might expect that some individuals disproportionately undermine group stability. It is not entirely clear, though, how influential a single individual may be towards group collapse.

Candidate individuals that disproportionately contribute towards group collapse might be those whose self-interests are at odds with mechanisms of group stability. Despite a natural tendency to disperse (Drickamer & Vessey, 1973; Oi, 1990; Van Noordwijk & Van Schaik, 1985), captive male macaque (Macaca mulatta) subjects are constrained in their capacity to emigrate without human intervention. Thus, captive male macaque self-interests might be posited to be at odds with their circumstances.

Wild macaque dominance relationships are expected to be adapted to maintain relatively stable female membership and rank, with turnover in male membership and rank (Berard, 1999; Sade, 1972; Sprague, 1998). Rhesus macaques typically have a discrete birthing season (Thierry, 2017), live in large mixed-sex groups (9–140 individuals, with a M:F sex ratio ranging from 1.43:1 to 0.21:1) (Ménard, 2004), and females attain rank through matrilineal youngest ascendancy (Chauvin & Berman, 2004). Maturing free-ranging male rhesus macaques often retreat to the periphery of their natal group and disperse at puberty (mean = 4 years, range 2.5–6.5 years) (Colvin, 1986; Drickamer & Vessey, 1973; Koford, 1963; Sugiyama, 1976). In captivity, however, natal male presence is not influenced by socioecological conditions – natal males stay until removed via human intervention. This is a challenge to managing captive macaques because, rather than become increasingly peripheral as they mature (Colvin, 1986; Drickamer & Vessey, 1973; Koford, 1963; Sugiyama, 1976), natal males with high-ranking female kin engage in frequent dominance interactions, displace well-connected males, and impact hierarchical stability (Beisner et al., 2011b; Koyama, 1967; Sade, 1967). Younger, 4–6 year-old, natal males can outrank older non-natal adult males (Chapais, 1983; Koford, 1963). Therefore, the captive setting is an ideal setting to quantify how influential particular individuals can be in the social dynamics of a large mixed-sex group, and how individual gains may be at odds with group-wide stability. Indeed, some captive primate centers pre-emptively remove natal males to prevent group disruption; at the California National Primate Research Center (CNPRC), however, males in outdoor housing units are left to mature in their natal groups. This is, in part, because young high-ranking natal males might also contribute to group stability. For instance, high-ranking natal males are themselves unusually central in dominance information networks, and their removal could destabilize group structure. Furthermore, natal males might contribute to social stability via social policing in fights among group members (Beisner & McCowan, 2013; Flack, Krakauer, et al., 2005).

Defining group stability is challenging. Rates of aggression are often unreliable indicators of social stability because aggression can increase in unrelated contexts– e.g., clumped resources (Gore, 1993), increased density (Judge & DeWaal, 1997), with breeding opportunities (Wilson & Boelkins, 1970). Furthermore, high rates of aggressive impartial interventions have been associated with reductions in trauma (Beisner et al., 2019). Instead, dominance behaviors via subordination signals (Beisner & McCowan, 2014) are necessary to describe hierarchical stability: i.e., displacements and peaceful silent-bared-teeth signals (pSBTs).

Displacements are important measures of conflict avoidance associated with social rank status (Judge & de Waal, 1993; Simpson & Simpson, 1982), whereby subordinates avoid high ranking individuals shortly after approaches that are not immediately preceded or followed by aggression or affiliation. pSBTs are formal signals of subordination communicating a pair’s true dominance relationship (Beisner & McCowan, 2014; de Waal & Luttrell, 1985; Thierry, 2007). Rhesus macaque pSBTs’ networks have a directed acyclic (i.e., no circular pathways) structure, occurring as one connected component with multiple hierarchical tiers (Beisner et al., 2016). The frequency and diversity of pSBTs received predicts conflict policing ability (Beisner et al., 2016) and has been linked to social stability (Beisner & McCowan, 2013; Flack, Krakauer, et al., 2005). Importantly, dramatic loss of pSBTs’ network complexity has been linked to social collapse (Fushing et al., 2014).

Here we observe multiple captive rhesus macaque social groups to examine how individual tendencies (e.g., social rank ascension) might impinge upon group stability (e.g., the status quo of social rank). These social groups are complex systems, and it is challenging to determine prior to management action whether a natal male’s presence is providing beneficial structural features to the social group that are not readily apparent through individual behavior alone. Therefore, experimental removals are necessary to truly assess an individual’s impact on group dynamics. To do so, we observed groups with high-ranking natal males for a baseline period, then examined how the group responded to the removal of this potentially key individual.

The social dynamics of changes in male status and/or role in the social group are critical for making informed management decisions and preventing social collapse in rhesus macaque breeding groups (Beisner et al., 2011a; McCowan et al., 2008; Oates-O’Brien et al., 2010). Consequently, we had the a priori expectation that the dominance certainty of high ranking individuals (in this case, alpha males) strongly contributes to the totality of group stability in despotic rhesus macaques, captive and wild. While changeover of alpha males is a natural process that may impact group stability, our prevalence of natal male occupancy is falsely imbued via human management. Therefore we examined baseline and post-knockout social structure changes at multiple scales: differences in (group-level) hierarchical structure, individual (node-level) network changes for the alpha male, and the inferred role of the natal male based on behaviors expressed in baseline.

Differences might be expected in hierarchical structure after natal male removal because their presence might alter the flow of hierarchical information from low-ranking individuals to the alpha male. As such, when they are physically removed, one might expect that network level metrics of hierarchical structure would increase or remain high. We examined four metrics of hierarchical structure in pSBT and displacement mixed-sex and male-only networks.

We expected that natal males might undermine the alpha male and act as terminal nodes for submissions, not submitting to the alpha male. As such, when they are physically removed, one might expect that alpha males would directly receive submissions from conspecifics – increasing their displacement centrality and increasing acyclicity. Furthermore, as removal of a high-ranking individual might create opportunities for rank ascension, we examined changes in the initiation and success of conflict reversals – insubordination, i.e., conflict that flows against the established hierarchy – before and after knockouts, following Flack et al. (2005).

Finally, we draw upon our findings to reconstruct what the role of each of the natal males was prior to their removal. Was their presence consequential to group stability? Did natal males exhibit homogeneity in their approach to rank ascension? Given these points, we emphasize the implications of our results for the captive management of social groups of macaques.

METHODS

Ethics statement

All research adhered to the American Society of Primatologists (ASP) Principles for the Ethical Treatment of Nonhuman Primates and was approved by the University of California Davis Institutional Animal Care and Use Committee.

Subjects and Housing

The subjects of this study were rhesus macaques from five groups at the California National Primate Research Center (CNPRC) in Davis, California, United States (Table 1). Five natal males (ages 3–6, mean = 4.2 years) were selected for permanent removal from their social groups. Four were sons of the α-female (groups B, C, D, E) and one was the nephew of the α-female (group A). In study groups B and C, the natal male subject had a 2-year-old brother who was also removed to avoid the possibility that the younger brother would attempt to fill his network role. The natal males’ matrilines were of variable size: A = 11, B = 5, C = 22, D = 26, E = 8. Animals were socially housed in 0.2 ha outdoor enclosures. Details of housing conditions are provided in Beisner et al. (2016).

Table 1.

Study group characteristics. Study subjects included all individuals ≥ 3 years of age. Groups had a similar number of infants (< 1 year): 31.2M±4.97sd.

Group	Group Size	Infants & Juveniles (≤ 3 years)	Sub- & Young-Adults (3 – 5 years)	Adults (≥ 6 years)	M:F Sex Ratio (≥ 3 years)	Natal male age at removal
A	168	79	33	56	0.53 : 1	3 years, 6.7 months
B	169	85	37	47	0.42 : 1	4 years, 4.0 months
C	147	46	39	62	0.35 : 1	4 years, 0.2 months
D	125	70	13	42	0.41 : 1	4 years, 5.2 months
E	204	103	43	58	0.51 : 1	6 years, 0.5 months

Behavioral data collection

Groups were observed for 12 weeks, two in spring (group C: Feb-May 2012, group E: Mar-May 2014), one in summer (group B: May-Aug 2012), and two in the autumn-to-winter breeding season (group A: Sep-Dec 2012, group D: Sep-Nov 2013). Group E was conducted in the same social housing unit as group B, two years after our 2012 natal male removal. Only 62% of all group E members (including infants) had been group members in B: 52% of B group’s adult males had been removed and the alpha male had been replaced (as a management decision unrelated to this project’s design). Observation periods were divided into: 6-week baseline and 6-week post-removal, with the natal male removal event occurring the start of week 7 (the first week in which post-removal data collection occurred). Data collection continued even on the day of removal. The exact day the week for natal male removal varied across groups as did the time of removal. If the natal male had already engaged in active conflict on that day, then we pooled that day’s data with baseline. If not, we pooled the data with the post-removal period.

We used an event sampling method to record all instances of aggressive, submissive, and status signaling interactions, and the level of aggressive and submissive behavior by each participant. Threats, lunges, and short chases represent lower levels of aggression whereas long chases, attacks, and bites represent more severe aggression (a more detailed description of the levels of aggression and submission can be found in Beisner et al. [2015]). Two observers collected data for 6 hours on 4 days per week from 0900–1200 h and 1300–1600 h. Inter-observer reliability was established prior to data collection, with a Krippendorf’s alpha threshold of ≥0.85. Observers collected data concurrently, with one observer collecting aggressive data detailed here and the other conducting scan sampling of affiliative data (e.g., within-arm’s length proximity, directional grooming). Events were recorded as a series of dyadic interactions that were linked by temporal proximity (i.e. < 20 seconds between dyadic interactions) and a common thread of participants (e.g., A chases B, C enters the fight and chases B also, and finally C redirects aggression toward D). A total of 50,904 events were recorded during 864 hours of observation.

Hierarchical Structure

Displacements and pSBTs

We constructed directed, weighted, networks of pSBTs as well as displacement approach–move away interactions. To avoid incorporating pSBTs in displacement networks, we excluded displacements accompanied by pSBTs (2.93% and 3.68% of displacements in baseline and post-removal, respectively). We focused on unweighted global reach centrality as an indicator of hierarchical flow in network structure (Mones et al., 2012). Global reach centrality (GRC) is the average of the difference between local reach for each node and the maximum local reach, where local reach is proportion of nodes directly connected to a node via out-edges (Mones et al., 2012). GRC ranges from 0–1, with 1 indicative of a high structural capacity for hierarchical flow. We expected that overall global reach centrality would increase after natal male removal, as a proxy for increased hierarchical structural flow.

During post hoc data analyses we focused more on displacement networks, as pSBT network densities were lower than found in previous work (Beisner et al., 2016). We attribute this lower density to changes in data collection methods, relative to prior work. We also included three other measures of hierarchical structure: 1) the proportion of nodes in any cycle up to 10 path lengths (Prop; a less computationally intensive modification to number of edges in cycles [Zafeiris & Vicsek, 2017]), which ranges from 0–1 with low scores indicative of strong hierarchical structure; 2) group-wide linearity estimates (h’ index) (De Vries, 1995) which imputes dyads to assess the number of dominant individuals (adjusted for unknown rank relationships), with high scores indicative that every individual in the group has clear dominance relationships with conspecifics; and, 3) triangle transitivity (ttri) (McDonald & Shizuka, 2013), which is the proportion of triadic interactions that are transitive, relative to the combined total of transitive and cyclic triads minus the expectation of transitive triangles by chance (0.75). We computed these metrics for mixed-sex and male-only displacement networks.

Node-Level Changes

Displacement network position

Natal males have been shown to exhibit atypically high centrality in displacement networks (Beisner et al., 2011b). Thus, we also focused on the ‘Authority’ metric, which is similar to eigenvector centrality but distinguishes authorities from hubs (Supplementary Figure 1). Here, authorities would be nodes that approached individuals and ‘received’ displacements but were not frequently displaced by others, as distinct from hubs that would be well-connected nodes that receive and express displacements. This is useful in lieu of eigenvector centrality as hierarchically central nodes could be expected to be hubs, while terminal nodes would be identified as authorities. Authority scores range from 0–1. High ranking individuals generally have higher authority scores than low ranking individuals, but authority is not linearly representative of rank (Supplementary Figure 2). Each network had a single node with a maximal authority score (1.0) with very few individuals having high authority scores; in our baseline networks, only 6.67% of subjects had authority scores > 0.25. We utilized topological knockouts using baseline data to determine how natal male removal might be expected to alter the authority scores as a function of node removal (following: Flack et al., 2006).

Rank reversals in aggression

Males might exhibit heightened competition for social status after removal of high-ranking males (Flack, Krakauer, et al., 2005). To examine whether our data supported this expectation, we extracted direct and redirected conflict events where lower ranked challengers initiated conflict against higher rank individuals. We examined whether subordinate males might initiate more conflict to opportunistically ascend in rank post-removal. We also examined whether rank structure might be destabilized by subsetting reversals that were successful – whereby lower ranked individuals received, but did not express, submission from higher ranked individuals.

Role of the Natal Males

Natal male rank

Ranks were determined using the Perc package in R, a network-based method for calculating dyadic dominance probabilities and dominance ranks using network pathways (Fujii et al., 2016; Fushing et al., 2011). We used decisive wins and sparring interactions to calculate rank. Ranks were determined for the entire group (i.e. including both sexes) as well as for males only.

Dominance probability of the natal male

We derived two dominance probability measures using Perc: the probability that the alpha male outranked the natal male subject based upon the network pathways between them; and, the mean dominance probability of the natal male’s relationship with all other group members. Dominance probability values range from 0–1: 0 signifies i is completely submissive to j, 1 signifies i is completely dominant to j, and 0.5 signifies ambiguity (Vandeleest et al., 2016).

Subordination signaling network position

We calculated natal males’ in-degree centrality (i.e. pSBTs received) from directed, weighted pSBT networks, z-scored to permit comparison across groups, to determine if subjects held positions in the network similar to alpha males.

Fights initiated by natal males

We counted the total number of agonistic events per study group and then calculated what proportion of those fights were started by natal male subjects in each group (i.e., a natal male subject was the initiator of the first dyadic transaction of the event). We also counted the total number of instances of severe aggression per study group (i.e. long chase, pin to the ground, and/or bite; these could be any dyadic transaction within an event) and calculated the proportion of severe aggression performed by natal male subjects in each group.

Conflict interventions by natal males

We tallied all conflict interventions per study group, which are defined as a third-party entering an on-going fight by directing aggression at, directing submission at, affiliating with, or approaching one or both combatants. We then calculated the proportion of all interventions performed by natal male subjects and what proportion of each natal male subject’s agonistic interventions had a duration of > 5 seconds to estimate each subject’s tendency to prolong agonistic events. Finally, to assess whether natal males’ presence might be beneficial to group stability, we calculated how often each subject performed conflict policing interventions. We distinguish between two types of conflict policing established as group stabilizing mechanisms: (1) impartial policing: the intervener enters and on-going fight and treats all conflict participants equally (approaching or directing aggression at all participants [Flack, Waal, et al., 2005]), and (2) support policing: the intervener defends a non-kin subordinate animal (Beisner & McCowan, 2013).

Similarity between alpha and natal males’ connections

Finally, we calculated Jaccard similarity coefficients of natal and alpha males’ connections during baseline (Fuxman Bass et al., 2013; Jaccard, 1912). To do so, we extracted the number of partners that both the alpha and natal male interacted with and divided it by the total number of partners that either the alpha and natal male interacted with (that is, both unique and shared partners). This index is bound between 0 and 1, with higher scores signifying similarity via redundant connections. We calculated Jaccard coefficients on their direct conflict, displacement, and grooming connections.

Analyses

All analyses were performed in R (R Core Team, 2013, 2022), with the exception of global reach centrality calculations which were calculated in Python (v 3.11; van Rossum & de Boer, 1991) using NetworkX 3.1 (Hagberg et al., 2008). To examine whether post-removal GRC metrics differed from what we might expect from chance, we permuted (N=1000) removals from baseline data (excluding the natal male) and plotted actual post-removal GRC values relative to these simulated values. We removed a random individual for each permutation in groups A, D, and E; we removed two random individuals in groups B and C to account for the removal of the natal males’ kin. Where data permitted, we conducted Wilcoxon signed rank tests to test whether natal male removals altered alpha male authority scores or increased the number of successful conflict reversal events. We also tested whether the probability and number of initiated reversals shifted after the natal male removal. These latter data were overdispersed and zero-inflated, therefore we implemented a negative binomial model with a Bayesian Regression Model using Stan (brm) (Bürkner, 2017). We used weakly informative priors with a warm-up of 1000 on 4 chains, running for 5000 iterations and a thin of 2, resulting in 8,000 post-warmup draws in the final models. We used expected log pointwise predictive densities (loo_compare()) and graphical posterior predictive checks (pp_check() and pairs()) to assess model fit.

RESULTS

Hierarchical Structure

Our metrics of hierarchical flow did not exhibit a uniform change across groups and displacement networks from baseline to post-knockout. Our mixed-sex pSBT networks had low GRC overall, indicating a reduced capacity for hierarchical flow. In the pSBT networks, C & D become more hierarchical (C = 0.09 to 0.12; D = 0.09 to 0.19, respectively); while A, B, and E become less hierarchical (A = 0.14 to 0.11; B = 0.18 to 0.07; E = 0.12 to 0.08). While prior work has emphasized the importance of pSBT networks, we found sparse pSBT networks (mean graph density = 0.02); thus, we elected to proceed with analyses focused on displacement networks.

In the mixed-sex displacement networks, GRC and the proportion of nodes in cycles (Prop) were poorly correlated (r₁₀ = 0.13, p = 0.71). The removal of the natal male resulted in heightened hierarchical structure across all of our groups, as characterized by increase GRC and decreased Prop (Figure 1). Permutations based on 1000 simulated removals from baseline data, however, suggested that only groups A, B, and D had higher than expected post-removal GRC based on 95% confidence intervals (A: removal GRC = 0.26, lower CI = 0.05, upper CI = 0.12; B: removal GRC = 0.46, lower CI = 0.20, upper CI = 0.30; D: removal GRC = 0.22, lower CI = 0.08, upper CI = 0.16); group E was 0.00008 away from its upper CI (GRC = 0.11, upper CI = 0.11). Two of our groups, B and C, had more pronounced hierarchical structure relative to the other groups. We present the dynamics of the female-only displacement networks in Supplementary Figure 2.

Figure 1.

Global reach centrality (top row of panels; GRC) and proportion of nodes with cyclicity (bottom panels, Prop) in the mixed-sex and male-only (columns) displacement networks across the five groups (x-axis). Bars are colored by group, with the hue indicating before and after the natal male knockout.

The male-only networks had higher GRC in group D and E, relative to the mixed-sex GRC. The proportion of nodes with cyclic interactions were low in male-only networks; we found a moderate association between GRC and Prop (r₁₀ = 0.31, p = 0.39) in male-only networks. Contrary to our findings in the mixed-sex networks, groups D and E had some of the highest GRC values, indicative of hierarchical structure – while these groups had low GRC in the mixed-sex networks. Furthermore, groups C, D, and E had no nodes with cyclic paths after the removal of the natal males. Despite this finding, GRC did not uniformly increase, rather A and C decreased, while B, D, and E increased in GRC. Permutations based on 1000 simulated removals from baseline data, however, suggested that group C had lower, while group B, D, and E had higher than expected post-removal GRC based on 95% confidence intervals (B: removal GRC = 0.43, lower CI = 0.24, upper CI = 0.38; C: removal GRC = 0.21, lower CI = 0.34, upper CI = 0.50; D: removal GRC = 0.50, lower CI = 0.26, upper CI = 0.45; E: removal GRC = 0.42, lower CI = 0.30, upper CI = 0.41).

Unlike our findings with GRC and Prop, linearity and ttri exhibited marginal changes after the manipulation across both networks (male-only and mixed-sex) and groups (A-E) (Supplementary). Excluding group D, linearity was generally low across the mixed-sex networks (range = 0.06–0.09), but was higher in male displacement networks (range = 0.15–0.21). Group D had higher linearity than the other groups in the mixed-sex (baseline = 0.19; post-knockout = 0.20) and male-only networks (baseline = 0.35; post-knockout = 0.42). Across groups in the mixed-sex networks, ttri was generally high (>= 0.91); but, in the male networks, group C, D, and E increased in ttri (from, respectively, 0.88, 0.93, and 0.95 in baseline to 1.00 post-knockout), parallel to our findings for Prop.

Node-Level Changes

Displacement network position

In the male-only displacement networks, topological knockouts indicated that alpha males were expected to increase in authority or remain 1.00 (Table 2). Contrary to this topological expectation, physical knockouts did not result in a uniform change in alpha male authority scores, relative to baseline (Figure 1; Table 2) (mean alpha males’ authority in baseline = 0.65±0.35sd; topological = 0.89±0.23; physical = 0.78±0.25). Due to several ties, we did not use non-parametric significance tests in the alpha male scores. Authority scores across all males present in baseline and physical knockout, however, were significantly different between the two periods (Paired Wilcoxon signed rank exact test: V = 2025, p-value = <0.001). Despite the lack of uniform differences in the alpha males, two alphas (groups C and D) had a quantitative increase in authority for male displacement networks (Table 2), two alphas (groups A and B) decreased or remained low, and the fifth alpha remained high but unchanged (group E). Supplementary Figure 3 shows example displacement networks for groups C and D.

Table 2.

Displacement network raw authority metric scores in the male-only and mixed sex displacement networks for alpha and natal males (with the ordinal rank of the authority score in parentheses), and mean authority scores ±sd for the remaining animals. Note that the topological knockout only removed the natal male from the baseline networks.

Network	Subject	Period	Group
			A	B	C	D	E
Male-only	Alpha Male	Baseline	0.47 (3)	1.00 (1)	0.18 (2)	0.60 (2)	1.00 (1)
		Topological KO	0.48 (3)	1.00 (1)	1.00 (1)	1.00 (1)	1.00 (1)
		Physical KO	0.45 (2)	0.86 (4)	0.59 (2)	1.00 (1)	1.00 (1)
	Natal Male	Baseline	0.04 (9)	0.09 (3)	1.00 (1)	1.00 (1)	0.69 (2)
	Others (M±sd)	Baseline	0.09 ±0.22	0.02 ±0.04	0.03 ±0.05	0.04 ±0.05	0.05 ±0.10
		Physical KO	0.07 ±0.21	0.28 ±0.37	0.11 ±0.23	0.05 ±0.05	0.09 ±0.20
Mixed-Sex	Alpha Male	Baseline	0.34 (6)	0.65 (2)	0.32 (6)	0.63 (2)	1.00 (1)
		Topological KO	0.34 (6)	0.86 (2)	0.35 (5)	0.68 (2)	1.00 (1)
		Physical KO	0.28 (4)	0.15 (4)	0.19 (5)	0.36 (2)	1.00 (1)
	Natal Male	Baseline	0.17 (13)	0.59 (3)	1.00 (1)	0.59 (3)	0.38 (2)
	Others (M±sd)	Baseline	0.09 ±0.16	0.08 ±0.13	0.08 ±0.13	0.07 ±0.15	0.03 ±0.05
		Physical KO	0.07 ±0.13	0.04 ±0.12	0.05 ±0.12	0.05 ±0.14	0.06 ±0.10

In the mixed-sex displacement networks, topological knockouts indicate that alpha males were expected to increase or remain the same in authority (Table 2). Contrary to these expectations and relative to baseline, physical knockouts resulted in a uniform decrease in alpha male authority scores in all but group E (mean authority scores for alpha males: baseline = 0.59±0.28sd; topological = 0.65±0.30; physical = 0.40±0.35). Authority scores across all subjects present in baseline and physical knockout were significantly different between the two periods (Paired Wilcoxon signed rank exact test: V = 53,932, p-value = <0.001).

When divided by removal period and group displacement network, only one subject has a score of 1.00 in authority per network. Besides alpha or natal males, we identified six other individuals with an authority of 1.00. Two males (both with 1.00 in baseline and post-knockout) and two females (one with a score of 1.00 in both the baseline and post-knockout periods, the other female only in post-knockout) in the mixed-sex network; one of these males had a score of 1.00 in baseline and post-knockout for the male-only displacement network, and two other males had scores of 1.00 in the post-knockout period.

Rank reversals in aggression

To determine whether the initiation of rank reversals changed among males after natal male removal as an individual level frequency tabulated per project period, we built a negative binomial brm with a hurdle distribution. In the hurdle, we included rank, age, a random effect for group, and an offset for days of group observation; in the negative binomial formula, we included project period (baseline, post-knockout), a random effect for group, and an offset for days of group observation. Including period in the hurdle did not meaningfully contribute, and reduced model fit; similarly, including rank and age in the negative binomial did not improve model fit. We used pp_check() and loo_compare() to assess and compare model fits (Supplementary Figures 4 & 5). We observed slight under-prediction at tail values. There was 1 run divergence, but all Pareto k estimates were good. Rhat values, bulk_ESS, and tail_ESS values were acceptable (Table 3).

Table 3.

Output from Bayesian regression model using Stan testing whether period predicts initiation of reversals.

Family: Hurdle_Negative Binomial
Link function: mu = log; shape=identity; hu = logit
Draws: 4 chains, each with iter = 5000; warmup = 1000; thin = 2; total post-warmup draws = 8000
Formula:
Initiations ~ Period + offset(log(Days)) + (1 | Group)     [negative binomial]  hu ~ offset(log(Days)) + Rank * Age + (1 | Group)          [hurdle]
Group-Level Effects:
	Estimate	Est.Error	l-95% CI	u-95% CI	Rhat	Bulk_ESS	Tail_ESS
sd(Intercept)	0.41	0.31	0.04	1.21	1	3788	4396
sd(hu_Intercept)	1.09	0.70	0.33	2.93	1	3541	4149
Population-Level Effects:
	Estimate	Est.Error	l-95% CI	u-95% CI	Rhat	Bulk_ESS	Tail_ESS
Intercept	−2.21	0.29	−2.81	−1.68	1	5622	6109
hu_Intercept	−3.91	0.64	−4.95	−2.31	1	4395	3638
Period	0.55	0.20	0.17	0.95	1	7109	6939
hu_Rank	−1.40	0.38	−2.16	−0.68	1	6937	6843
hu_Age	−1.70	0.58	−2.85	−0.58	1	6703	6546
hu_Rank:Age	−5.11	0.91	−6.92	−3.38	1	6797	6724
Family Specific Parameters:
	Estimate	Est.Error	l-95% CI	u-95% CI	Rhat	Bulk_ESS	Tail_ESS
Sigma	0.85	0.27	0.38	1.42	1	7022	5612
Bayesian R-squared:
	Estimate	Est.Error	Q2.5	Q97.5
R2	0.130	0.043	0.065	0.234

Model Notes: There was 1 divergent transitions after warmup. Draws were sampled using sampling (NUTS). For each parameter, Bulk_ESS and Tail_ESS are effective sample size measures, and Rhat is the potential scale reduction factor on split chains (at convergence, Rhat = 1).

In the hurdle output, the 95% credible intervals gave high certainty that age and rank predicted whether males would initiate a reversal (Figure 2), and age interacted with rank (Figure 3). Higher ranked males were more likely to initiate reversals relative to lower ranked males. Younger males were more likely to initiate reversals relative to older males. Importantly, examination of the interaction effect indicated that younger males were more likely to initiate reversals if they were ranked higher. While older males were less likely to initiate reversals if they were ranked higher, instead older males were more likely to initiate reversals if they were mid- or low-ranking.

Figure 2.

Posterior prediction density plots for whether the fixed effects of period, rank, and age predict reversal initiations. Note that Age and Rank are only in the hurdle formula (initiation of reversals), while Period is only in the negative binomial formula (counts of reversal initiations). Estimated mean (Table 4) is emphasized for each density. Uncertainty intervals are included via shading at the 80% interval, and density outlines cease at the 99% interval.

Figure 3.

Posterior prediction density plots for the brm interaction between rank (y-axis groupings) and age (fill). Probability (x-axis) is the probability individuals will attain zero and not cross the hurdle – that is, initiate no reversals. High and low predicted values for age and rank were estimated using the maximum and minimum values in the dataset; we used the project mean for the offset (days).

Period predicted the rate of initiations after the hurdle, such that initiations occurred at higher rates in the knockout period, relative to baseline (Figure 2). Bayes_R2 for the full model was 0.13±0.04. Males that did initiate conflict did so at a higher rate after the natal male knockout (Table 3). The average rate across groups, per day was: Baseline 0.10M±0.14sd; Post Knockout 0.14±0.22sd (Supplementary Figure 6).

We did not observe more successful challenges post-knockout. Due to the low number of subjects that successfully challenged, we analyzed these data lumped by group, rather than across individuals. The mean rate across groups, per day was: baseline 0.59 ±0.35sd; post removal 0.76 ±0.44sd. Though all but one group (C) increased in the rate of successful male challenges, paired Wilcoxon signed rank exact test suggests that groups did not significantly differ in successful challenges between the two periods (V = 4, p = 0.44).

Role of the Natal Males

We assessed numerous metrics of natal male behavior and social status position (Table 4). The natal males from groups A and B shared several features, relative to natal males from groups C, D, and E. Natal males from groups A and B had lower relative rank in the male and group hierarchy with dominance probability scores that were relatively high with the alpha male, compared to their probability relative to the group. To clarify, natal males in remaining groups (groups C, D, and E), uniformly held the beta male position, had ambiguous relationships with the alpha male (evident via dominance probability), had high overall (mean) dominance probability, performed a large percentage of the group’s interventions, and exhibited greater use of severe aggression. Furthermore, natal males in C, D, and E had greater redundancy in the identity of their social conflict and displacement partners, as evident by higher Jaccard similarity indices, relative to groups A and B. There were no discernable patterns in grooming similarity across the groups.

Table 4.

Behavior and network position metrics of natal male subjects

		Groups
		A	B	C	D	E
Natal male network position	rank in male hierarchy	7	3	2	2	2
	rank in group hierarchy	10	6	3	3	4
	dominance probability with alpha male	0.97	0.99	0.72	0.68	0.87
	mean dominance probability	0.80	0.75	0.94	0.90	0.93
Natal male agonistic behavior	group fights initiated	3.03%	2.90%	5.90%	6.74%	4.73%
	group severe aggression	3.33%	2.72%	6.72%	6.34%	5.98%
	group interventions	1.88%	2.53%	5.65%	4.98%	3.72%
	interventions duration > 5 s	32.2% (N=10)	22.2% (N=4)	45.6% (N=36)	49.0% (N=24)	30.5% (N=18)
Natal male policing interventions	support policing	19.3% (N=6)	11.7% (N=2)	21.5% (N=17)	15.6% (N=7)	31.6% (N=19)
	impartial policing	12.9% (N=4)	5.9% (N=1)	26.6% (N=21)	6.7% (N=3)	15.0% (N=9)
Jaccard similarity coefficient for natal & alpha males’ partners in baseline	conflict similarity	0.31	0.24	0.50	0.51	0.55
	displacement similarity	0.16	0.14	0.30	0.55	0.47
	grooming similarity	0.22	0.27	0.24	0.20	0.33

We emphasize, however, that all natal male behaviors were not uniform among groups C, D, and E. For instance, the natal males from groups C and D initiated a higher percentage of group fights and had more long duration interventions, relative to all the other groups. As counts and percentages of general (non-policing) interventions might be expected to scale with the number of kin present, we examined the proportion of group animals that were in the natal males’ matriline relative to total group members in the housing unit. Groups C and D also had the highest proportion of natal male matriline membership to total group population (0.15 and 0.21, respectively) relative to the remaining groups (A = 0.07, B = 0.03, E = 0.04). We also examined policing interventions, whereby natal males from group E and C performed more support policing interventions (as a count, but not as a percentage), relative to all the other natal males. Furthermore, the natal male from group C performed more impartial interventions than the combined total of natal males from all other groups.

DISCUSSION

Maintenance of social stability is important for the welfare management of captive social groups, where reductions in stability have a variety of negative consequences for animals’ physical, social, and psychological well-being (McCowan et al., 2008; Oates-O’Brien et al., 2010). The presence of high-ranking natal males past the normal age of dispersal, which can occur in both wild and captive settings, has a unique potential to impact group stability when they threaten the alpha male’s position (Beisner et al., 2011b). Despite this assertion, we found inconsistency in how groups responded to the removal of natal males. Our study found variation across five high-ranking natal male subjects in the nature of their dominance relationships, aggressive behavior, and position within group social networks. Three of the five natal male subjects showed evidence for disrupting the status of the alpha male. Experimental removal of these three subjects resulted in higher, or stable but high, authority scores in the male displacement network. The removal of the remaining two natal males, however, decreased alpha male authority scores, or retention of low scores. Via analyzing hierarchical structure metrics, node-level changes, and the role of each natal male within their group we find a complex portrait: high ranking natal males are not uniformly a threat to alpha male status and group stability, rather their influence is diverse. Group hierarchical flow and, potentially, individual ascension show considerable variation even in a managed environment.

Hierarchical Structure

Removal of the natal males, identified a priori as potentially challenging to group stability, did not uniformly increase hierarchical flow in our displacement and pSBT networks. Our displacement networks showed variation between groups, across periods, and between mixed- and male-only networks. Mixed-sex displacement networks generally increased in GRC and decreased in Prop after natal male removal, relative to baseline, indicative of more hierarchical structure – two groups (C & E), however, did not have a pronounced increase after removal. Rhesus macaque hierarchies are often interspersed, such that females and males can occupy high ranking positions. The increase in hierarchical structure could be the result of at least two concurrent processes: individuals might be asserting their dominance relationships more frequently as a high ranking individual has been removed; or, the natal male’s kin may support the status quo alpha male.

Male-only networks had very low Prop scores across groups, while GRC was moderate. Both of these characteristics are evident of hierarchical structure. Despite higher scores of male metrics relative to mixed-sex networks in a subset of groups, we did not find a uniform response to natal male removal. Three of the groups markedly increased in GRC, but our biggest change exhibited a decrease in GRC indicative of less hierarchical structure via a reduced disparity in reach across the network. Our general findings in male displacement networks were equivocal.

Node-Level Changes

Alpha male authority scores did not uniformly increase after natal male removal. Compared to the remaining groups, natal males in A and B were lower ranked and had lower authority scores in their network (Tables 2 & 4). Alpha males had higher male-only authority scores in three groups (A, B, E), with a similar pattern in the mixed-sex networks – though the alpha and natal male had similar authority in D. Thus, the removal of the natal male did not uniformly benefit the alpha male in their position as a terminal hierarchical node.

Interestingly, we found strong supportive evidence that the number of rank reversals increased after the natal male removal. Broadly, these analyses indicate that males were exhibiting heightened competition for social status after natal male removal (Flack, Krakauer, et al., 2005). Our results provide a nuanced picture, however, whereby individuals were not more likely to initiate reversals after the knockout, but those that did initiate were more likely to do so at a higher rate. We found a strong age-rank interaction, with independent effects, that constrained whether individuals would initiate a reversal. These outcomes may indicate that viable individuals are attempting to ascend in rank, competing for the natal male position. Alternatively, because our measures of rank were attained from baseline data, the higher number of reversals may signify alterations to the hierarchy that have already occurred.

Role of the Natal Male

Three of the natal males (groups C, D, and E) showed similarities, relative to the remaining groups (A, B): high rank, ambiguous dominance relationships with the alpha males, less ambiguous dominance relationships with other group members, high authority scores in the displacement networks, more pSBTs, higher Jaccard coefficients between alpha and natal males’ conflict and displacement partners, and higher percent contribution to severe aggression and interventions. Subadult male macaques are typically socially peripheral (Koford, 1963; Sugiyama, 1976) and as such, are expected to hold more peripheral network positions. Therefore, these three natal male subjects held atypical (i.e. highly central) network positions. As beta males with ambiguous relationships with their alpha males, these three natal males likely represented a threat to the alpha male’s position. Indeed, two days prior to the scheduled experimental removal (outside observation hours), we saw one natal male (group C) displace and threaten the alpha male. Thus, this subset of high ranking natal males exhibited (a) behaviors that may exacerbate group-level aggression (e.g. greater use of severe aggression), (b) uncertainty in his dominance relationship with the alpha, and (c) an uncharacteristically central social position in the group (e.g., high SBT in-degree, high displacement authority scores).

Natal males’ policing behavior was not uniform across groups. Conflict policing is a social mechanism that improves the stability of a social group by preventing or curtailing cascades of aggression and other forms of deleterious aggression (Beisner & McCowan, 2013; Flack, Krakauer, et al., 2005; Von Rohr et al., 2012). These data suggest that natal males have the capacity to contribute to the stabilizing mechanism of conflict policing and, in some instances, contributed almost a third of the baseline policing events (group E). We acknowledge, however, that these natal males can contribute to group stability via policing, while also remaining a threat to the alpha male’s status.

Captive Management

Our study’s outcomes highlight the different influence that natal males can have on group stability and that their rank ascension is not intrinsically detrimental to group stability. Animal care staff should closely monitor natal males for rank competition, which may indirectly occur through pSBT or displacement networks. Likely because monitoring groups is logistically and temporally consumptive, colony managers might be inclined to remove all natal males to preempt their perceived negative impact on social stability. We assert, however, that the consequences of this strategy may be fiscally and socially costly.

First, from an individual welfare perspective: captive rhesus macaques raised in their natal social groups for at least the first three years of life have the best behavioral and physiological health (Gottlieb et al., 2013; Lutz et al., 2007; Novak et al., 2006). When reared in a context other than a large outdoor social group (i.e., small peer-group outdoors, indoor mother-reared, or nursery reared), rhesus macaques are 4.44–11.02 times more likely to develop abnormal behaviors (Gottlieb et al., 2013) and 2.86–4.95 times (Gottlieb et al., 2013) – even up to 30 times (Rommeck et al., 2009) – more likely to develop self-injurious behaviors. Furthermore, males are more susceptible to developing abnormal and self-injurious behaviors than females (Gottlieb et al., 2013; Lutz et al., 2007; Novak et al., 2006; Rommeck et al., 2009) and these behaviors are difficult to impossible to treat (Baker et al., 2012; Rommeck et al., 2009). Immature high-ranking natal males removed from natal groups would be at high risk for developing self-injurious behaviors if not successfully reintroduced to another social group, compromising the welfare of the animals and their scientific utility, while potentially incurring chronic medical costs to manage these behaviors.

Second, these natal males are not uniformly deleterious to group stability. If able to mature, adult males are important for maintaining stability via reducing group aggression as policers and by maintaining a group sex ratio similar to wild ratios (e.g. ≥1 male:5 females; [Beisner et al., 2012]). All of our natal male subjects performed policing interventions. Removal of all immature high-ranking natal males would reduce overall policing effort which has been linked to a reduction in social stability (Flack, Krakauer, et al., 2005). In our dataset, authority scores and Jaccard similarity coefficients provided post hoc insight to differentiate natal males potentially disruptive to alpha male dominance. Further data on the implementation of these, or similar metrics, might provide intuitive and easily computed ways to distinguish hierarchical threats to alpha male status, prior to direct turnovers.

Group Dynamics

Shortly after observations, two groups (A & E) were disbanded despite our interventions. Similarly, in group C we observed displacement and threats from the natal male to the alpha, outside of observation hours. Interestingly, these three groups all had a decrease in male-only displacement network GRC after natal male removal; but this difference was only pronounced in group C. The wider implications of this finding from only three events are unclear, especially given the low magnitude of GRC change in two groups. Even so, the expectation of reduced hierarchical structure would be a coupled decrease in group stability. Stable groups (B & D) did not exhibit uniform patterns – though both groups increased in male-only GRC. In group B, baseline data suggests the natal male was unlikely to be a direct threat to the alpha at removal; also, B had high GRC in the mixed-sex displacement network. In group D, the natal male was the authority in the displacement network; the group had low mixed-sex GRC with a high proportion of nodes in cyclic interactions, with high hierarchical structure in the male-only displacement network. Group D was the only group with strong linearity in any of the displacement networks.

We identified six individuals, besides the alpha and natal males, who had high authority scores. In group B, we identified two individuals with the highest authority scores; subject Tank had high authority scores in both baseline and post-removal for the mixed-sex network. Tank was present in B during 2012, but later became alpha. He remained alpha during our 2014 knockout, in group E. The other subject in group B attained a high authority score after our natal male knockout in the male-only network. This subject was later our natal male removal subject in 2014, for group E. In group A and C, males with the high authority scores during this project became alpha males after our study; though group A collapsed with this individual in alpha status. Of the two females with high authority scores, one in group C overthrew her mother to become alpha. These patterns are challenging to interpret, in part because authority scores are assigned to high ranking individuals – through this quality, these individuals are more likely to inherit or ascend to high-rank. Even so, authority scores elucidate two putative pathways to high rank success in captive macaques: via high action in the male-only or mixed-sex displacement networks. Furthermore, they may provide insight into key individuals that ordinal rank, alone, cannot.

Environment and Broader Context

Based on data from 11 group collapses from 1996 to 2007 at CNPRC, there was no consistent bias in whether cage collapses occur in the breeding or birthing season (Oates-O’Brien et al., 2010). This finding was in spite of a known increase in conflict during breeding season for captive (Oates-O’Brien et al., 2010, Theil et al., 2017) and free-ranging populations (Wilson & Boelkins, 1970). Similarly, we did not observe a consistent pattern in whether natal male relocation was more stabilizing for group hierarchical structure based on season. Ruling out the impact of season is challenging, however, given the large differences between the groups’ dynamics and their eventual outcomes. Thus, future research could consider the influence of varying seasonal timing of natal male removals and/or introductions to minimize individual trauma and improve group stability prior to periods of increased trauma. Finally, though the age of relocation for our natal male subjects was similar to those observed dispersing in the wild (Colvin, 1986; Drickamer & Vessey, 1973; Koford, 1963; Sugiyama, 1976), large-scale relocations of natal males also introduces the challenge of introductions. Male introductions can be fiscally and temporally costly, while raising welfare concerns for the risk of injury for the introduced animal (Beisner et al., 2021). Carefully managed retention of natal males in large groups provides welfare benefits at the level of the individual and group, as maintaining sex ratios similar to those observed in the wild has been shown to be of importance to reduce conflict through the action of conflict interventions (Beisner et al., 2012, Bloomsmith et al., 2021).

Conclusion

Importantly, not all natal males are a challenge to group stability. Instead, the individual habits of the male synergizes with their group structure. Retention of natal males in the group saves person hours, reduces costs, and is not intrinsically deleterious to individual or group welfare. If the goal for breeding and housing management is to keep as many unassigned animals as possible in social groups and maintain a sex ratio conducive to effective conflict policing, then keeping males in their natal groups as long as possible may be the easiest and most successful strategy. Not all natal males are problematic, and the blanket removal of any individual holds the risk of decreasing social stability. Indeed, we found evidence for an increase in hierarchical reversals following natal male removal and, in at least one instance, our removal resulted in a new challenging natal male two years later. Thus, our removal may create power vacuums that will opportunistically fill hierarchical slots and may later destabilize group structure. We suggest that the social relationships of natal males be assessed at the individual level to determine if removal from natal groups would be constructive.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are openly available in DataDryad at https://doi.org/10.5061/dryad.g79cnp5wk. [NOTE anonymized data during peer review are available for editorial staff here: https://datadryad.org/stash/share/PhzpsktQK0mFmEY90P6QU3Wm_6qR9iV1eqlguaPjzdo]