Many species have been shown to associate non-randomly (e.g., honeybees (Apis mellifera
], guppies (Poecilia reticulata
], Columbian ground squirrels (Urocitellus columbianus
], African elephants (Loxodonta africana
], bottlenose dolphins (Tursiops truncatus
], Japanese macaques (Macaca fuscata
], feral goats (Capra hircus
], and yellow baboons (Papio cynocephalus
]), leading to distinct patterns of association within larger collectives at the group or population level [9
]. Additional layers of social complexity arise due to differentiated social roles within groups based on sex, breeding status, morphological caste, and individual differences to name a few [10
]. The emergent social structure of these groups is defined by the nature, quality, and patterning of constituent relationships over time, which has been suggested to influence the individual fitness of group members [12
]. Additionally, social structure (as a product of behavioural interactions over time) has been shown to be linked to individual welfare in many species [13
]. Consequently, changes in overall social structure or individual social role can be indicative of changes in individual health and welfare status [14
Because of this link, many modern zoo-based animal management systems have recognized the importance of maintaining species-specific social structures when seeking to safeguard individual welfare in captivity [15
]. To meet this aim, new tools for improving the subjective experience and affective states of captive animals continue to be developed [16
]. Social network analysis is one such tool that has been suggested to be able to provide husbandry teams with insight into how changes in the social or physical environment, as well as changes in routine, affect the overall structure of a group and the bonds within that group [17
]. Changes in social stability and individual social role can indicate fluctuations in complex affective states in an easily quantifiable way [16
Social network theory allows for the analysis of the connection between individual constituent behaviour and the functionality of a group from an evolutionary and ecological perspective [12
]. When applied to animal systems, social network analysis can produce a graphical representation of dyadic interactions within a group, based on adjacency matrices, which is used to identify individuals or classes of individuals (referred to as nodes) crucial to the structural integrity of the group [14
]. This information can then inform species-specific husbandry decisions, including future population planning, institutional translocations, current enclosure changes, and group structure manipulation to maximize social stability and individual welfare [17
]. However, the application of social network theory to zoo animal management is currently in its infancy.
Here, we apply social network analysis to improve the understanding of the species-specific social organisation of the Livingstone’s fruit bat (Pteropus livingstonii
), which will aid in the evidence-based welfare assessment and captive management of this species and other closely related fruit bat species. As many fruit bat species are endemic to islands subject to extreme effects of climate change, the role of zoos and captive breeding in conservation efforts has only increased [18
]. Evidence-based management of fruit bat species is currently impeded by a lack of research on their captive welfare [19
]. The welfare of bats in captivity may be acutely important to the success of breeding programs, as poor welfare has been directly linked to decreased reproductive success in other social species [20
]. However, there remains little research on how to best assess or improve the welfare of bat populations housed in captive environments.
Previous studies of bat sociality in situ have employed a wide variety of network-based methodologies and have increasingly emphasized the importance of social interaction to bat population health and integrity [23
]. Several species have been shown to associate non-randomly, have preferred partners, and form distinct communities that sometimes exist across several different roost sites [23
]. We demonstrate here that the quantification of the social environment can be similarly applied in captive settings, through social network analysis, to maximize the evidence-based welfare management and ex situ conservation of fruit bat species through an improved understanding of captive sociality and social roles. We believe that this is the first study to implement these techniques on a zoo-housed fruit bat species, the Livingstone’s fruit bat.
The Livingstone’s fruit bat (Pteropus livingstonii
) is native to the islands of Anjouan and Moheli of the Comoros West-Indian archipelago, off the coast of northern Madagascar [24
]. The total wild population is thought to consist of approximately 1120 individuals [24
]. The IUCN (International Union for Conservation of Nature) first listed P. livingstonii
as endangered in 1988, and then as critically endangered in 1996 [25
]. To safeguard the genetic diversity of the species, a captive breeding program was established in 1989 when the Durrell Wildlife Conservation Trust signed an agreement with the Comorian government to capture several wild individuals [24
]. Four expeditions brought 17 individuals to Jersey Zoo, Channel Islands, by 1995 [26
In 2020, the captive population of P. livingstonii
consists of 67 individuals housed across three institutions (including one surviving member of the wild-caught group of individuals). Jersey Zoo still houses the majority of individuals, with a total population of 60 bats. The action plan formulated by the IUCN Species Survival Commission Chiroptera Specialist Group specifically recommends further research on the feeding ecology, population biology, and the social organisation of P. livingstonii
to ensure the success of future conservation interventions [27
]. Therefore, we implement social network analysis techniques on this critically endangered species to address these recommendations, as well as provide a framework as to how they might be utilized in other captive settings.
Early (unpublished) data collected by our group on the social experience of female P. livingstonii
in captivity suggests that preferred affiliative associations exist and that these are based on kinship and age class homogeneity [28
]; however, the overall social structure of a mixed sex and age group of P. livingstonii
has not previously been elucidated. This study implements novel social network techniques to quantify the social environment experienced by individuals of this species through the analysis of observational data on proximity-based association, and affiliative and aggressive interactions, over two discrete seasons. We utilize this information to explore social structure through the quantification of levels of complexity, the relationship between spatial association and more complex types of interaction, and by the identification of trait-based assortment (individuals choosing to associate based on similarity). Additionally, we implement node metrics to determine whether individual social roles exist, and if so, whether they are predicted by the sex, age, or dominance level of individual P. livingstonii
. The ultimate aim of this research is to demonstrate the implementation of social network analysis in a captive environment, while simultaneously aiding zoo management in making evidence-based decisions to improve individual welfare, based on an increased knowledge of social structure in this species.
This study has demonstrated how novel social network analysis-based techniques can be implemented to study the sociality of captive populations. Our results can now inform the management of a critically endangered species to safeguard individual welfare through the preservation of crucial social relationships and underpinning structure. Captive P. livingstonii
display a variety of conspecific relationships and substantial social complexity. Different types of social relationships in other species (as categorized by strength and frequency of interaction) have been shown to provide varied ecological and emotional benefits within social systems [71
]. The composition of three or four interaction types (K) (based on mean strength of association) present within this population at increasing frequency within each network may present P. livingstonii
individuals with a gradation of relationship strengths, each with potential fitness benefits.
Analysis of the level of crossover between types of social interaction was carried out through the implementation of an MRQAP test [47
] to quantify the predictive power of affiliative and aggressive networks on proximity-based association. The results of this test showed that the association network was significantly predicted by both the affiliation and aggression networks from both data collection periods. This suggests that P. livingstonii
interact more frequently, both affiliatively and aggressively, with conspecifics with whom they also spend more time in close proximity. However, both association networks were more highly predicted by their corresponding affiliation networks rather than their aggression networks. From a management perspective, this indicates that frequent spatial association can be interpreted as evidence of social bonds between individuals.
To further explore the underpinning variables of the observed social complexity, the dominance hierarchy of this population was assessed through the calculation of individual David’s scores. A basic examination of predictors of these scores through linear modelling revealed that age alone was found to affect dominance rank in both data collection periods. One possible explanation for the exclusion of sex as an influential factor, as would be initially assumed from previous research on harem structure in this and other related species in the wild [73
], is that size has been shown to positively correlate with age in other fruit bat species [74
]. Older individuals may be larger, and therefore more likely to win agonistic interactions regardless of sex [74
]. As David’s score is based on the outcome of aggressive encounters, it would follow that older, larger individuals may be more dominant in this population.
To further quantify the social organisation of P. livingstonii
, the network-based assortment of variables describing individual characteristics (i.e., sex, age, and dominance rank) was calculated. No network from either data collection period was found to be significantly assorted by sex, indicating that individuals interacted with conspecifics of the same and opposite sex at relatively similar rates. This highlights the need for future research on preferential inter- and intra-sexual relationships in captivity, as previous research on the social organisation of P. livingstonii
suggested that, because of harem-based reproduction and male resource guarding, male to male interactions were mostly aggressive in nature, and male to female interactions were mostly affiliative [73
]. The lack of obvious harem structuring in this population may be due to a number of factors presented by a captive environment, including but not limited to the abundant nature of resources or the limited spatial choices available to individuals. This points to a need for the exploration of territoriality and captive space use in this species.
Another notable social pattern present in this population identified during assortment analysis was that all statistically significant assortment was positive, suggesting that individuals have a higher degree of preference for interaction with members of like classes [50
]. The aggression network from the Summer 2019 data collection period and the association network from the Spring 2020 data collection were both positively assorted by age. As well as having a similar age in years, these individuals often share a life-stage. This means that, perhaps more than age alone, P. livingstonii
chooses in captivity to associate with conspecifics who share similar ecological goals such as social requirements, reproductive or energetic status [76
]. Group cohesion in other social species (e.g., yellow-bellied marmots (Marmota flaviventris
], Merino sheep (Ovis ares
], Columbian ground squirrels (Spermophilus columbianus
], etc.) has also been shown to improve when individuals have fewer conflicting interests, so individuals of similar biological and social requirements often associate at a higher frequency [79
In addition to age, two out of three networks (association and aggression for Summer 2019; affiliation and aggression for Spring 2020) from each data collection period were also positively assorted by dominance, as represented by David’s scores. As the calculation of David’s score is based on the outcome of agonistic interactions, it follows logically that aggression networks would be positively assorted by dominance. The positive assortment of the association network from Summer 2019 and the affiliation network from Spring 2020 suggest that different seasons may present social environments where interaction preference changes. One possible factor influencing this change may be the reproductive cycle of P. livingstonii
. In captivity, the mating and birthing seasons are somewhat longer than what has been observed in the wild [80
]. It has been estimated that parturition takes place in captivity during the warmer months from March to September, meaning that mating (although witnessed infrequently) takes place from November to March.
Hence, it could be that Summer 2019 dominance rank-based assortment within the association network is an artifact of the individual’s choice to interact with groups that support the ecological and social needs presented by parturition and caring for young. Individuals may be more likely to thrive during this time of year if they more frequently interact with conspecifics of a similar dominance level [79
]. The Spring 2020 data collection took place during the end of the P. livingstonii
mating season. For this reason, the positive dominance-based assortment of the affiliation network from this period could be indicative of mate choice. Perhaps P. livingstonii
preferentially choose mates with a similar, or higher dominance ranking as an indicator of positive fitness. The maximum enclosure temperature during this period was also notably lower than in the summer data collection period (Appendix B Figure A2
and Figure A3
). Perhaps the lack of dominance-based assortment in the association network from this period is a result of individuals roosting in closer proximity to heating units along the northern wall of the enclosure. When space is limited in captivity, social choice is also constrained [81
]. However, the interpretation of these results should be regarded tentatively, as these results may simply be an artifact of reduced data collection in the Spring 2020, due to the Covid-19 pandemic. For this reason, future exploration of variable-based assortment through further data collection of other seasons should be prioritized.
Once the overall complexity and assortment of the social structure of captive P. livingstonii
were described, the social roles of classes of P. livingstonii
were approximated through the estimation of the effects of predictors describing individual characteristics (i.e., age, sex, and dominance rank) on node metrics. Not every best fitting model identified a statistically significant effect, suggesting that something other than the selected characteristics alone may influence social roles. We suggest that individual differences and personality type should be investigated as a potentially significant factor in social role emergence [82
]. Only models applied to node metrics characterising the aggression networks from both data collection periods were found to include statistically significant predictors. During the Summer 2019 period, more dominant individuals had a higher aggression-based closeness centrality and males had a higher weighted degree. This indicates that dominant individuals held higher strength aggression-based connections [9
] and males generally had a higher number of aggression-based connections with conspecifics than did females. This is not to suggest that dominant males were the most aggressive during this period, but rather that dominant individuals (of either sex) had more intense aggressive interactions and that males were part of more aggressive interactions regardless of directionality.
During the Spring 2020 data collection period, males had a higher aggression network-based closeness centrality than females. In contrast with the Summer 2019 period, this indicates that males held higher-intensity agonistic connections with conspecifics. This suggests that social roles based on aggressive interactions could be predicted, to different extents, by the characteristics identified here (i.e., sex and dominance), but also that the social niche of subgroups could change depending on the ecological needs of the group during different seasons.
Though the relationship between descriptive factors (i.e., age, sex, and dominance rank) and social role await further exploration, this study has also demonstrated that individual social roles can remain temporally stable. Through the implementation of a QAP test [68
], individual normalized node metrics from the Summer 2019 data collection were found to be significantly predictive of node metrics from the Spring 2020 data collection in all three network types. Association network-based roles (as characterized by node metrics) showed the highest degree of temporal fidelity. This suggests that, though the variables most predictive of social role change over different seasons, individuals retain socially important standing within the group.
Through the implementation of social network analysis on the behaviour of the population of P. livingstonii housed at Jersey Zoo, we have demonstrated the efficacy of novel methods to quantify factors affecting captive social structure in a highly social mammal. We have quantified observed social complexity through the modelling of categories of interaction based on strength and frequency, as well as through analysis of variable assortment. Additionally, we have started to unravel the ecological basis for social roles, as defined by node metrics, in this species.
Relationship strength appears to be negatively correlated with the frequency at which that type is displayed within the population. P. livingstonii also appear to positively assort based on age and dominance level, with dominance being highly predicted by the age of the individual regardless of sex. Social roles may show consistency over time (shown here over a period of ten months), with variables describing the individual characteristics influencing social role assignment changing over the seasons.
Based on the findings of this analysis, we suggest that captive management of P. livingstonii
should allow for relationships between individuals of similar ages and dominance levels to persist where possible, and separating individuals that interact frequently in an affiliative context should be avoided. Maintaining social bonds could be extremely important for the efficiency of this captive breeding program, as social bonds have been found to play a key role in the reproductive success of other social species such as Assamese macaques (Macaca assamensis
], feral horses (Equus caballus
], and chacma baboons (Papio ursinus
] to name a few. Further, the influence of individual differences and personality on social roles, in addition to the temporal influences on social roles in this species, could be further examined as a direction for future research.
Our results have added to the growing body of literature highlighting the key role of social network analysis in evidence-based management, in particular due to the method’s power to quantify individual social experiences and therefore to safeguard individual welfare.