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

Social relationships profoundly impact health in social species. Much of what we know regarding the impact of affiliative social relationships on health in nonhuman primates (NHPs) has focused on the structure of connections or the quality of relationships. These relationships are often quantified by comparing different types of affiliative behaviors (e.g., contact sitting, grooming, proximity) or pooling affiliative behaviors into an overall measure of affiliation. However, it is unclear how the breadth of affiliative behaviors (e.g., how many different types or which ones) a dyad engages in impact health and fitness outcomes. We used a novel social network approach to quantify the breadth of affiliative relationships based on two behaviors: grooming and sitting in contact. Dyadic relationships were filtered into separate networks depending on whether the pair engaged in multiple affiliative behaviors (multiplex networks) or just one (uniplex networks). Typically, in social network analysis, the edges in the network represent the presence of a single behavior (e.g., grooming) regardless of the presence or absence of other behaviors (e.g., contact sitting, proximity). Therefore, to validate this method, we first compared the overall structure of the standard network for each affiliative behavior: all grooming interactions regardless of contact sitting, and all contact sitting interactions regardless of grooming. We then similarly compared the structure of our filtered multiplex vs. uniplex networks. Results indicated that multiplex networks were more modular, reciprocal, and kin-based while connections in uniplex networks were more strongly associated with social status. These differences were not replicated when comparing networks based on a single behavior alone (i.e., all grooming networks vs. all contact sitting networks). Next, we evaluated whether individual network position in multiplex vs. uniplex (novel approach) or grooming vs. contact sitting (traditional approach) networks differentially impact inflammatory biomarkers in a commonly studied non-human primate model system, the rhesus macaque (Macaca mulatta). Being well connected in multiplex networks (networks where individuals both contact sat and groomed) was associated with lower inflammation (IL-6, TNF-alpha). In contrast, being well connected in uniplex grooming networks (dyad engaged only in grooming and not in contact sitting) was associated with greater inflammation. Altogether, these results suggest that multiplex relationships may function as supportive relationships (e.g., those between kin or strong bonds) that promote health. In contrast, the function of uniplex grooming relationships may be more transactional (e.g., based on social tolerance or social status) and may incur physiological costs. This complexity is important to consider for understanding the mechanisms underlying the association of social relationships on human and animal health.

Methods
Data collection:

Data were collected on a four groups of Rhesus macaques. Behavioural observations were conducted all adult individuals (3+ years) in the group and serum samples were collected. Affiliative and agonistic interactions were recorded.  Serum samples were assay for inflammatory cytokines.

Data processing:

Behavioural observations were used to construct a weighted, directed behavioural network for each of the following behaviours: (1) all grooming, (2) all contact sitting, (3) multiplex grooming (dyads that both groomed and contact sat, edge-wieghts reflect grooming frequency), (4) uniplex grooming (dyads that only groomed and never contact sat, edge-wieghts reflect grooming frequency), (5) multiplex contact sitting (dyads that both groomed and contact sat, edge-wieghts reflect contact sitting frequency), and (6) uniplex grooming (dyads that only contact sat and never groomed, edge-wieghts reflect contact sitting frequency).  Whole network metrics including density, modularity, eigenvector centralization, average edge weight, clustering coefficient were calcualted.  For calculation of individual level network metrics all networks were treated as undirected, these metrics included degree, strength, eigenvector, betweenness, closeness centralities and local clustering coefficient.  Dominance ranks were calculated from agonistic interactions for all adult individuals using the Perc package.

Files include the code used to execute these processing steps as well as the processed datasets.