Differential effects of multiplex and uniplex affiliative relationships on biomarkers of inflammation

Introduction

For decades, research has shown that social relationships impact individual health and fitness in a variety of animal species, with estimates of the magnitude of the association between social factors and mortality in humans on par with other well recognized mortality risks (e.g., smoking, alcohol consumption) (Holt-Lunstad, Smith & Layton, 2010). While the importance of social factors for health and fitness are widely recognized, the mechanisms by which social life exerts its influence are still not well understood (McCowan et al., 2016; Ostner & Schülke, 2018). One reason is that social life is complex; it encompasses both agonistic and affiliative interactions that vary across a range of dimensions such as frequency of interaction, symmetry, tenor (friendly or hostile), predictability, and stability (Hinde, 1976a; Hinde, 1976b; Silk, Cheney & Seyfarth, 2013; Fischer et al., 2017). As such, new approaches addressing this complexity are needed to better understand the biological mechanisms by which social relationships influence health.

Two common methods for quantifying relationships and their association with health and fitness outcomes (e.g., reproduction or lifespan) (Silk, Cheney & Seyfarth, 2013) include the use of sociality indices and social network analysis. The dyadic sociality index (DSI) is a commonly used measure of relationship quality in non-human primates. Relationships with high DSI scores (e.g., strength of top three bonds or scores greater than the group mean) are thought to be of high quality and are commonly referred to as strong bonds and tend to be equitable, stable, involve frequent interaction, and are most common between kin and peers (Silk et al., 2010; Silk, 2014b; McFarland et al., 2017; Ostner & Schülke, 2018). In animals, a higher number or quality of these strong bonds has been associated with acute hormonal responses (e.g., oxytocin or cortisol levels), increased reproduction, and longer survival (Crockford et al., 2013; Ostner & Schülke, 2018; Silk, Seyfarth & Cheney, 2018; Ellis et al., 2019). However, these strong affiliative bonds represent only a small fraction of the affiliative relationships an individual has (Silk et al., 2010; Schülke, Wenzel & Ostner, 2013; Fischer et al., 2017; McFarland et al., 2017; Silk, Seyfarth & Cheney, 2018) with weaker affiliative bonds comprising the remainder. The function of these other, weaker bonds has been hypothesized to increase social flexibility (e.g., social connections can shift with environmental demands), allowing general social integration and indirect connections that might provide access to others who may have resources or information (McFarland et al., 2017; Silk, Seyfarth & Cheney, 2018). Yet, evidence for an association between weak bonds and health and fitness is mixed.

Social network analysis is a method well suited to quantifying the structure of social relationships and their potential to impact health and fitness. Typically, animal researchers use social network analysis to understand how one’s direct interactions with others or indirect connections (e.g., friends of friends) impact a variety of health-related metrics and/or fitness outcomes. Direct connections measure the number of unique social partners one has (degree) or the frequency of social interactions (strength) which together have been proposed as measures of social integration (Ostner & Schülke, 2018; Ellis et al., 2019). For example, having more social partners has been associated with reduced likelihood of pathogen infection and fitness benefits (Balasubramaniam et al., 2016; McFarland et al., 2017; Brent, Ruiz-Lambides & Platt, 2017a). Indirect connections can be measured in a variety of ways that describe how one is connected in the broader social network ranging from measures of social embeddedness (closeness), cliquishness (clustering coefficient), and social bridging (betweenness). While the general pattern seen across these measures is that greater connectedness or centrality in specific behavioral networks is associated with lower risk for gastrointestinal pathogens, increased reproduction, and longevity (Brent et al., 2013; Balasubramaniam et al., 2016; Cheney, Silk & Seyfarth, 2016; Ostner & Schülke, 2018), there is little consistency across studies in identifying which specific network metric is important (Ostner & Schülke, 2018; Ellis et al., 2019; Snyder-Mackler et al., 2020). For example, eigenvector centrality, a measure of social capital (i.e., one is well-connected to well-connected others) has been associated with health and fitness benefits in some studies (Brent et al., 2013; Balasubramaniam et al., 2016; Cheney, Silk & Seyfarth, 2016) but has been found to have no association in others (Brent et al., 2013; Balasubramaniam et al., 2016; Ellis et al., 2019).

While these approaches have proved fruitful, previous studies have either analyzed affiliative behaviors separately (e.g., grooming or proximity) or lumped them together into a single index giving them roughly equal weight (e.g., dyadic sociality index) (Silk, Cheney & Seyfarth, 2013; Ostner & Schülke, 2018). Unexplored is whether the diversity or breadth of affiliative behaviors exhibited in a relationship and the extent to which different types of affiliative behaviors are integrated (Hinde, 1976b, p. 197; Silk, Cheney & Seyfarth, 2013) which may provide key information as to the function or quality of a relationship. For example, a relationship in which a specific dyad engages in multiple types of affiliation (e.g., contact sitting, grooming, and proximity) may differ from one in which a dyad engages in only one single behavioral domain (e.g., contact sitting only, grooming only, or proximity only), even if the rate of interaction is the same. We may expect relationships identified as strong bonds (e.g.,  via DSI metrics) would likely also be those in which animals interact using multiple affiliative behaviors. It is somewhat unclear, however, what might be expected of relationships in which only one behavior is present, as the function of these relationships may be more closely tied to the specific behavior being measured (e.g., thermoregulation might explain a relationship characterized by contact sitting only (Campbell et al., 2018), but this is unlikely to be a good explanation for proximity only). The choices animals make regarding how to interact with social partners (i.e., the breadth of behaviors social partners engage in) may provide information about what function those relationships serve. In fact, Hinde (Hinde, 1976b) states that “the most important clues as to the significance and meaning of an interaction to the participants may be the context of other interactions in which it lies (p 4)”.

Recent advances in social network analysis and theory, and specifically multilayer or multiplex networks, (Beisner et al., 2020; e.g., multiple types of interactions among the same set of individuals) may eventually provide a solution with tools to address such issues. However, currently, these analytical methods (e.g., metrics like versatility and multiplex PageRank) are not sufficiently sophisticated to account for the many structural differences that exist between behavioral networks of different types (e.g., proximity networks are often dense and undirected, contact networks may be more sparse and undirected, whereas grooming networks are directed) (Beisner et al., 2020). Yet, given the mounting evidence for the importance of multidimensionality in social relationships, the integration of diverse affiliative behaviors at this dyadic level may provide important additional information as to the quality or function of those relationships and their potential downstream impacts on health and fitness. Indeed, in human studies, multidimensional measures are often the strongest predictors of health and fitness (Berkman et al., 2000; Holt-Lunstad, Smith & Layton, 2010; Snyder-Mackler et al., 2020). Yet few animal studies to date (except see Fischer et al., 2017) have examined whether the integration of that diversity of interactions in a relationship provide clues as to the nature of those social relationships. One exception, conducted by Balasubramaniam et al. (2019), found that highly connected rhesus macaques in a grooming network were more likely to share E. coli species (a commensal gut bacteria), but only if they were also well connected in a huddling network, suggesting that the presence of multiple affiliative behaviors across an individual’s dyadic connections (e.g., integration) may influence the potential transmission of gastrointestinal microbes.

Here we employed a social network approach to explicitly explore whether the integration of diverse affiliative behaviors (i.e., grooming and contact sitting) within a relationship can provide new information regarding the quality and function of relationships and their impact on health-related outcomes in a commonly studied non-human primate model system, the rhesus macaque (Macaca mulatta). We use rhesus macaques as a group-living, nonhuman primate (NHP) model because their social relationships are highly differentiated, exhibit a high degree of complexity and individual variability, and have been linked to a variety of health and fitness outcomes (Vandeleest et al., 2016; Snyder-Mackler et al., 2016b; Brent, Ruiz-Lambides & Platt, 2017a; Capitanio, Cacioppo & Cole, 2019). Affiliation in primates takes many forms, including grooming, contact sitting, proximity, embracing, and coalitionary support (Sussman, Garber & Cheverud, 2005; Aureli & Schaffner, 2007; Pallante et al., 2019). In this study, we selected two of the most well studied affiliative behaviors in rhesus macaques, grooming and contact sitting, and explored whether relationships in which both were observed differed from those in which only a single behavior occurred. Grooming is the most widely studied affiliative behavior in rhesus monkeys and is used as a key indicator for the presence of affiliative relationships. Grooming has been proposed to serve multiple social functions including: to maintain social bonds (Dunbar, 1991) and social cohesion (Lehmann, Korstjens & Dunbar, 2007), and in exchange for tolerance from dominants, for agonistic support, or for access to resources (Noë & Hammerstein, 1995; Schino & Aureli, 2008; Borgeaud & Bshary, 2015; Snyder-Mackler et al., 2016a; Barrett & Henzi, 2006). Other affiliative behaviors in rhesus monkeys include spatial proximity, huddling, contact sitting, and alliances. Here we chose to use contact sitting behavior which is a combination of sitting in physical contact and huddling (contact sitting with trunk-to-trunk contact or while sleeping (Ueno & Nakamichi, 2018) as a second indicator of affiliative relationships (Campbell et al., 2018). While less frequently studied than grooming, contact sitting and huddling are behaviors frequently used to aid in thermoregulation in cold climates and preferentially used among animals with close relationships (Bernstein & Mason, 1963; Ueno & Nakamichi, 2018; Campbell et al., 2018).

To assess the integration of behaviors within a relationship we took a traditional single behavioral network (e.g., grooming) and filtered it into “multiplex” and “uniplex” networks (Table 1). We use the term “multiplex” to refer to networks in which edges reflect the co-occurrence of multiple affiliative behaviors and the term “uniplex” to refer to networks in which edges reflect only one specific type of interaction (e.g., grooming or contact sitting, not both) (Hinde, 1976b; Berkman et al., 2000). We compared network structure between three pairs of networks. First, we compared uniplex grooming (dyads only groom; edge-weights reflect grooming frequency) to multiplex grooming (dyads groom and contact sit; edge-weights reflect grooming frequency). Our rationale for constructing multiplex and uniplex networks in this manner was inspired by Balasubramaniam et al. (2019) which identified the combination of grooming and huddling behavior to be especially relevant for sharing of gastrointestinal microbes. Further, this distinction could also be hypothesized to relate to the different functions of grooming (social bonding/cohesion vs. exchange for tolerance and resources may potentially reflect strong vs. weak social bonds). Then to ensure that observed structural differences reflected biological differences and were not artifacts of our novel network filtering procedure, we also compared uniplex and multiplex contact sitting as well as networks generated from all the grooming vs. all contact sitting interactions. These relationships may promote health through multiple pathways, including: by buffering individuals from the negative consequences of stressors (Cohen & Wills, 1985; Balasubramaniam et al., 2019) or by impacting general physiological processes related to anabolic repair and maintenance (Cacioppo & Hawkley, 2003). Therefore, we further explored whether an individual’s position in these paired networks (i.e., uniplex grooming vs. multiplex grooming, uniplex contact sitting vs. multiplex contact sitting, and all grooming vs. all contact sitting) was differentially associated with biomarkers of inflammation (i.e., serum pro-inflammatory cytokines), which are common, well-established indicators of individual health status (Uchino et al., 2012; Capitanio, Cacioppo & Cole, 2019). Our prediction was that individuals exhibiting more central roles in multiplex networks would show lower levels of inflammatory cytokines than those exhibiting more central roles in uniplex networks.

Materials and Methods

Results

Relationship dimensionality and biomarkers of inflammation

IL-6. There was a single top model identified which included uniplex and multiplex grooming closeness (Table 5). Specifically, individuals that were more central (i.e., higher closeness) in the uniplex grooming network exhibited higher inflammation while the opposite was true for multiplex grooming (Figs. 1A–1B).

Table 5:

Candidate model outputs.

	IL-6	TNF-α
Model parameters	Model 1	Model 1	Model 2
Intercept	−0.102	6.53**	4.83**
Uni GR closeness	7.92**	–	–
Uni GR strength	–	23.81**	25.36**
Multi GR closeness	−3.68*	−6.26**	–
Multi GR degree	–	–	−0.12**
ΔAIC	0	0	1.59
Model Weight	1	0.689	0.311

DOI: 10.7717/peerj.19113/table-5

Notes:

*p < 0.05.

**p < 0.01.

Uni

Uniplex

Multi

Multiplex

GR

Grooming network

All models were run using a negative binomial distribution and included a random effect of group.

TNF-α. There were two candidate models identified; both of which included comparisons of multiplex and uniplex grooming centrality (Table 5). As with IL-6, lower centrality in the multiplex grooming network (degree or closeness) but higher centrality in the uniplex grooming network (strength) were consistently associated with higher levels of TNF-α (Figs. 1C–1D).

Discussion

Social primates have a complex web of differentiated social relationships, which vary in their quality, structure, and function. While strong affiliative social relationships are usually associated with better health, less is known on how the multidimensionality or integration of different affiliative behaviors within a social relationship might impact health. We identified affiliative relationships that were multiplex (dyads affiliated using both grooming and contact sitting behavior) versus uniplex (dyads only groomed or only contact sat). Examination of these networks revealed that they differed in topology, kinship, and associations with rank. Multiplex networks were more modular, clustered, reciprocal, had higher average edge weights, and were more strongly associated with kinship. In contrast, dyads in the uniplex networks tended to be of more disparate ranks. Notably, these differences in kinship and rank between multiplex and uniplex networks were not apparent when looking at all grooming or contact sitting interactions. The health impacts of these two networks differed as well, with females that were less socially connected in multiplex grooming networks exhibiting higher levels of inflammatory cytokines (IL-6 and TNF-α), whereas females more socially connected in uniplex grooming networks exhibited higher levels of inflammatory cytokines. Notably, these effects were primarily seen in multiplex and uniplex grooming, not in the other networks tested. These results suggest that grooming that occurs in the context of multiplex affiliative relationships may result in health benefits (i.e., reduced inflammation). In contrast, grooming occurring in uniplex affiliative relationships may result in health costs (i.e., higher inflammation).

Networks consisting of dyads with multiplex relationships showed differences from uniplex relationships in network topology, kinship, and associations with dominance. Multiplex networks had structural characteristics consistent with strong bonds or supportive affiliative relationships (Silk et al., 2009; Silk et al., 2010; Silk, 2014a). Specifically, interactions in the multiplex networks were more likely to be reciprocal, frequent (i.e., higher edge-weight), clustered, and associated with kinship, suggesting they are relationships that are regularly maintained and potentially more stable across time (Silk et al., 2010; Silk, 2014a). Previous methods demonstrating that strong bonds enhance fitness, particularly those using sociality indices, have also used multiple behaviors to assess relationship strength (e.g., grooming and proximity (Fischer et al., 2017; McFarland et al., 2017; Silk, Seyfarth & Cheney, 2018)). However, these methods rely on total duration or frequency of affiliation to describe relationships rather than characterizing the breadth or dimensionality of the relationships (e.g., dyads can have high DSI through grooming, proximity, or both). Similar to strong bonds, multiplex affiliative relationships may improve health and fitness by buffering individuals from the negative impacts of stress, improving predator detection, promoting offspring survival, and improving social stability (Beisner et al., 2011; Micheletta et al., 2012; Balasubramaniam et al., 2016; Cheney, Silk & Seyfarth, 2016).

Also consistent with the literature on strong affiliative bonds, being well connected to others was associated with biomarkers of better health. Specifically, the negative association between multiplex grooming centrality (e.g., degree, closeness centrality) and biomarkers of inflammation indicated that individuals that were generally well connected in the network (e.g., at the core of the group) may be at lower risk for inflammation related diseases (Bruunsgaard et al., 2003). Specifically, having more strong social partners (i.e., degree) or being more embedded in a strongly bonded network (i.e., closeness) could be particularly important. This possibility, however, requires replication due to the influence of network structure (e.g., modularity) on the intercorrelations among individual level centrality metrics. Our results add to the literature suggesting that strong bonds may exert their effects by altering endocrine and immune function (Boccia et al., 1997; Shutt et al., 2007; Crockford et al., 2013; Snyder-Mackler et al., 2020) which could in turn lead to effects on fitness. Consistent with this idea, Yang et al., (2016) found in humans that socially integrated individuals (i.e., those with more social connections across multiple domains) exhibited lower inflammation, whereas social strain (e.g., higher levels of family criticism or demands) was associated with greater inflammation. While our study only examines a single time-point, given that familial and friend relationships tend to endure through extended periods, often persisting over decades (in both humans and NHPs), these relationships may have an important and long-lasting impact on health.

Uniplex grooming relationships may reflect relationships that are more transactional in nature (Silk, 2002). The fact that uniplex grooming relationships are less kin biased but likely to occur between dyads of more disparate ranks suggests that these relationships may be more related to grooming being used as a commodity in exchange for tolerance or support from higher ranking animals. These transactional relationships may reflect a desire to maintain peace/tolerance or used in a biological market exchange (Noë & Hammerstein, 1995; Schino & Aureli, 2008; Borgeaud & Bshary, 2015; Barrett & Henzi, 2006), rather than reflecting a strong affiliative relationship. The positive association between females’ connectedness in uniplex grooming networks and biomarkers of inflammation suggests that uniplex grooming relationships may not be supportive on their own and instead are associated with increased physiological costs, at least in the short term. Specifically, predictors of inflammation in the uniplex grooming networks included strength or closeness centrality. However, the various network centrality metrics from the uniplex grooming network were more highly correlated with each other than the other networks, and therefore it is difficult to identify which specific aspect of uniplex grooming centrality might be driving these effects. However, collectively this group of candidate predictors indicates that greater general connectedness (direct and indirect) in uniplex grooming was associated with increased inflammation. Uniplex grooming relationships were maintained through generally less frequent interactions that were more likely to occur between animals of disparate ranks which may result in greater uncertainty regarding the outcome of any given interaction. This uncertainty may be stressful, and therefore have at least short-term physiological costs (Schino & Alessandrini, 2015; Christensen et al., 2024). If these relationships are more transactional in nature, then maintaining more of these transactional relationships may result in increased stress, which if sustained can result in long-term physiological costs (Snyder-Mackler et al., 2020). It is possible that these short-term costs are actually investments that may manifest in future benefits (e.g., tolerance, alliance support) that would offset these costs, yet this is difficult to test as the “commodities” exchanged may be heterogeneous and the time-scale for market exchanges is often unclear (Dunayer & Berman, 2016). However, other work points to benefits of weak or economically based bonds to survival and reproduction (McFarland et al., 2017; Ellis et al., 2019) (although see Silk, Seyfarth & Cheney, 2018). While these types of connections may have ultimate fitness benefits (e.g., alliance support, increased access to food), this research suggests they may also be associated with proximate costs.

Conclusion

Both humans and many species of NHPs engage in a complex interconnected system of social interactions. Understanding the mechanisms by which social relationships impact health and fitness remains a challenge. Decades of research has established that affiliative social relationships can benefit health, however, the complexity and multidimensionality of relationships has yet to be explored. By utilizing a network approach, we were able to characterize two types of affiliative social relationships that differed in their network topology, kin bias, associations with rank, and importantly their associations with biomarkers of inflammation. Our research has indicated that features of multiplex affiliative relationships are consistent with the concept of a strong supportive relationships and may support health and fitness. In contrast, more transactional affiliative relationships (e.g., uniplex affiliation) may incur short-term health costs yet may result in ultimate benefits through commodity exchange. Still unclear is whether these effects are specific to the combination of behaviors used here (i.e., contact sitting and grooming), or if other affiliative behaviors (e.g., proximity) might provide similar information. Further research into the dimensionality of relationships might reflect different qualities or functions of relationships is needed. However, this complexity is important to consider for understanding the mechanisms underlying the impact of social relationships on human and NHP health.

Supplemental Information