key: cord-0722615-8msne79m
authors: Balasubramaniam, Krishna N.; Aiempichitkijkarn, Nalina; Kaburu, Stefano S. K.; Marty, Pascal R.; Beisner, Brianne A.; Bliss-Moreau, Eliza; Arlet, Malgorzata E.; Atwill, Edward; McCowan, Brenda
title: A comparative network approach to assess the social-ecological underpinnings of zoonotic outbreaks at human-wildlife interfaces
date: 2021-07-20
journal: bioRxiv
DOI: 10.1101/2021.07.19.452944
sha: f6623238f7b0a461cf2e993f0f1e413bf34772e7
doc_id: 722615
cord_uid: 8msne79m

Pandemics caused by wildlife-origin pathogens, like COVID-19, highlight the importance of understanding the ecology of zoonosis at human-wildlife interfaces. To-date, the relative effects of human-wildlife and wildlife-wildlife interactions on zoonotic outbreaks among wildlife populations remain unclear. In this study, we used social network analysis and epidemiological Susceptible Infected Recovered (SIR) models, to track zoonotic outbreaks through wild animals’ social-ecological co-interactions with humans and their social grooming interactions with conspecifics, for 10 groups of macaques (Macaca spp.) living in (peri)urban environments across Asia. Outbreak sizes predicted by the SIR models were related to structural features of the social networks, and particular properties of individual animals’ connectivity within those networks. Outbreak sizes were larger when the first-infected animal was highly central, in both types of networks. Across host-species, particularly for rhesus and bonnet macaques, the effects of network centrality on outbreak sizes were stronger through macaques’ human co-interaction networks compared to grooming networks. Our findings, independent of pathogen-transmissibility, suggest that for wildlife populations in the Anthropocene, vulnerability to zoonotic outbreaks may outweigh the potential/perceived benefits of interacting with humans to procure anthropogenic food. From One Health perspectives, animals that consistently interact with humans (and their own conspecifics) across time and space are useful targets for disease-control.

do so to different extents across individuals, time, and space, and may form patterns of 93 associations through such interactions that could influence the acquisition and transmitting 94 infectious agents. Social Network Analysis (SNA) offer exciting avenues to captures such 95 patterns of associations [18] [19] [20] [21] [22] . Improving on traditional epidemiological frameworks, SNA offers 96 promising quantitative ways to allow for animals' tendencies to interact differently, and to 97 varying extents, with different socio-ecological aspects of their environment (e.g., their 98

conspecifics, other animals, humans), and assess how such non-random contact patterns and 99 behavior can impact infectious disease transmission [18] [19] [20] [21] [22] . Yet while SNA-based epidemiological 100 assessments have been increasingly implemented to evaluate disease transmission in both group-101

living wild animals and humans 18-23 , HWIs have seldom been the foci of such assessments. 102

To-date, epidemiological studies that have implemented SNA have largely focused on 103

animal-animal interactions, and often on single behavioral features that define such interactions 104 (reviewed below). However, infectious agents at HWIs may spread differently through different 105 types of interactions, for instance through networks of human-wildlife and wildlife-wildlife 106

interactions. Some examples of wildlife-wildlife social networks that have been associated with 107 increased risk of infectious agent transmission include affiliative contact associations (e.g., 108

Tasmanian devils, Sarcophilus harrisii 24 ; skinks, Egernia stokesii 25 ; giraffes, Giraffa 109 camelopardalis 26 ), aggression (e.g., meerkats, Suricata suricatta 27 ), social grooming (e.g., 110

Japanese macaques, Macaca fuscata 28 ; brown spider monkeys, Ateles geoffrey 29 ), and contact-111

huddling (rhesus macaques, M. mulatta 30 ). While interesting, such studies largely do not capture 112 important complexity and variability in animal social systems, such as individuals potentially 113

interacting in different ways with their conspecifics compared to with other, shared features of 114 their environment (e.g., accessing and sharing space and resources, interspecies encounters with 115 predators, prey, and humans) [18] [19] [20] [21] [22] . In particular, disease transmission among wildlife at HWIs 116 may be driven by such multiple, potentially interplaying types of interactions, including inter-117

individual differences in animals' interactions with conspecifics 31,32 , humans 2,7-9 , and 118 anthropogenic features like contaminated water, soil, human foods, livestock, and other feral 119 mammals 20,33 . It is therefore crucial to assess zoonotic transmission through multiple (rather than 120 single or specific) types of interactions and their resultant network connections at HWIs. 121 Another issue is that epidemiological assessments of zoonotic transmission at HWIs 122 continue to be hampered by many ecological (e.g., trapping and sampling of wildlife) and 123

logistical (e.g., tracking human behavior and long-term health indicators) constraints. When data 124 is incomplete or unavailable, mathematical models offer critical insights into the occurrence of 125 real-world epidemiological processes 34, 35 . In this regard, network approaches have been 126 extensively combined with epidemiological models of the 'Susceptible Infected Recovered 127 (SIR)' type 10,31,36-38 , to simulate disease transmission and its associated outcomes through human 128 (reviewed in 31,36,37 ) and (less so) nonhuman animal (individual empirical studies 10,38-41 ) 129

networks. SIR models are bottom-up, compartmental epidemiological models that simulate 130 disease spread by causing entities (individuals) to move across 'susceptible', 'infected', and 131 'recovered' disease states. They do so at dynamic probabilities that, based on user specifications 132 of model complexity, may depend on a combination of one or more pathogen-specific 133 epidemiological variables (e.g., transmissibility, basic reproduction number: defined below), host 134 contact patterns (e.g., spatial or social network connectedness), and host attributes (e.g., age-sex 135 class) or intrinsic states (e.g., physiology, rates of recovery). To date, studies that have 136

implemented SIR models on wildlife spatial and social networks have revealed strong 137 associations between network connectedness of the first-infected individual and simulated 138 disease outcomes such as the times to saturation (i.e. when all individuals have been infected 139 thereby leaving no other susceptible individuals) or extinction (i.e. when all individuals have 140 recovered from the disease and no more individuals can be infected), and outbreak sizes (mean 141 % of infected individuals) of pathogens 38, 40, 41 . To our knowledge, SIR models remain largely 142 unimplemented at HWIs (see 41 for an exceptional study of Barbary macaques, M. sylvanus), and 143 especially in contexts of human-wildlife interactions in (peri)urban ecological settings where 144 contact between people and wildlife maybe highly frequent and vary across time and space. 145

Human-nonhuman primate interfaces are well-suited to address the above gaps. and overlap more with humans and anthropogenic environments 46,66 . Yet they typically show 205 more nepotistic (than bonnet macaques) social systems, with individuals preferring to engage 206 more with just specific subsets of group conspecifics than with others 62 . Given these differences, 207

we tested the following predictions. Across network-type for each host-species, we predicted that 208 the co-interaction network centrality of first-infected macaques would have a stronger effect on 209 outbreak sizes than grooming network centrality for rhesus macaques and long-tailed macaques, 210

but that bonnet macaques would show the opposite effect. Across host-species for each network 211 type, we predicted that the effect of co-interaction network centrality of first-infected macaques 212 on outbreak sizes would be higher for rhesus macaques and long-tailed macaques compared to 213 bonnet macaques, but that the reverse would be true (bonnet macaques > rhesus and long-tailed 214 macaques) for the effects of grooming network centrality on outbreak sizes. 215 We also examined the effects of sociodemographic (sex, dominance rank) characteristics 216 of the first-infected macaque on outbreak sizes. Since females and high-ranking individuals form 217 the core of macaque grooming networks 62,67 , we predicted that outbreak sizes through grooming 218 networks would be higher when the first-infected individuals were females (versus males) and 219

higher-ranking (versus lower-ranking) individuals. On the other hand, given the exploratory and 220

increased risk-taking behavior of males resulting in their being more well-connected in co-221

interaction networks compared to females 60, 64 , we predicted that outbreak sizes through co-222 interaction networks would be higher when the first-infected individuals are males (versus 223 females). Finally, we also explored whether the overall anthropogenic exposure of first-infected 224 macaques, specifically their frequencies of interactions with humans, and time spent foraging on 225 anthropogenic food, influenced zoonotic outbreak sizes. 226 227

Results:

Construction of macaques' co-interaction networks and social grooming networks: 230 231

For each of 10 groups of three macaque species observed for between 11-18 months in 232 (peri)urban environments in India and Malaysia (details in 64 ; Supplementary Table 1;  233 Supplementary Figure 1) , we collected and analyzed demographic and behavioral data on 234 human-macaque interactions, macaque activity budgets, and macaque-macaque social behavior. 235

From this data, we constructed weighted, undirected human co-interaction networks 64 . In these, 236 the nodes were individual, pre-identified macaques, and the weighted edges represented the 237 frequencies with which animals jointly engaged in taking risks in anthropogenic environments, 238

i.e. co-interacted with one or more humans within the same time-frame and anthropogenic 239 space 64 . We also constructed weighted, undirected social grooming networks in which individual 240 monkeys (nodes) were linked based on their frequencies of engaging in social grooming 241

interactions (edges) with their conspecifics 64 . 242 243

Impact of the centrality of first-infected individuals by network-type and host-species on disease 244 outbreak sizes: 245 246

Using the SIR model simulations, we examined the impact of network-type, host-species, 247

the network centrality of randomly selected first-infected macaque, and the interactions between 248 them, on mean outbreak sizes. We also examined the effects of the first-infected individuals' 249 sociodemographic characteristics (sex, dominance rank), and overall exposure to anthropogenic 250 factors (frequency of interactions with humans, time spent foraging on anthropogenic food), on 251 mean outbreak sizes. Outbreak sizes were calculated as the mean % of infected individuals 252 within a macaque group at the end of each epidemiological SIR model simulation run marked by 253 either disease extinction or saturation 40 (see Introduction and Methods below for definitions). For 254 each macaque group and network-type, we ran 5000 SIR model simulations, 500 simulations for 255 each of 10 artificially introduced pathogens that were allowed to infect a randomly-chosen 'first-256

infected' macaque, and whose transmissibility  ranged from 0.05 (lowest) to 0.50 (highest) in 257 increments of 0.05. Transmissibility values and ranges were selected based on the corresponding 258 values for pathogens of average basic reproduction numbers (R0) that ranged from 'low' (=1.6) 259

to 'high' (=14.0) in accordance with the human literature 38,40 (see Methods for details). 260

Simulated outbreak sizes were then averaged for each macaque for each network-type from 261 across all its first-infected simulation runs irrespective of pathogen-specific . As measures of 262 network centrality, we calculated, for both co-interaction networks and grooming networks, 263

individuals' strength centrality (the number and sum of its direct connections or edge-264 weights 69,70 ), but also its betweenness centrality (the tendency for an individual to inter-link or 265 'bridge' different parts of a network 19,71,72 ), and eigenvector centrality (the number and strength 266

of an individuals' direct and secondary network connections 73-75 ) centrality measures (see 267

Methods for more detailed definitions), all of which may influence disease transmission and 268 outbreak sizes 19 . 269

To test our predictions, we used Generalized Linear Mixed Models (GLMMs) with 270 macaque group ID within host-species entered a random effect and implementing a corrected 271

Akaike Information selection Criterion (AICc) 76,77 to identify a single best-fit model from each 272 set. To ensure our findings were not impacted by inter-dependencies in network measures across 273 individuals, we calculated permuted p values for each predictor in our best-fit models, 274

implementing a permutation-based 'null-model' approach that used a post-network node-275 swapping procedure 78-80 . In support of our prediction, we found that across network-types and 276 host-species, the strength centrality of the first-infected macaque, which better predicted 277 outbreak sizes than betweenness centrality or eigenvector centrality (model 1 in Supplementary  278  Tables 2A-C, 3A-B) , was significantly, positively correlated to mean outbreak size (Tables 1 and  279 2; Figures 1 and 2) . In other words, disease-causing agents generally infected more individuals if 280 they entered into a population by first infecting central or more well-connected individuals. 281

Moreover, the magnitude of these effects of first-infected macaque centrality on outbreak 282 sizes varied across network types and species, although not always in the predicted directions. 283

For a given host-species but across the two different types of networks, we found a significant 284 interaction between network-type and strength centrality for rhesus macaques and bonnet 285 macaques, but not for long-tailed macaques (Table 1; Figure 1 ). As predicted, rhesus macaques 286

showed a significantly stronger effect of the mean centrality of first-infected individuals on 287 outbreak sizes through their co-interaction networks compared to their grooming networks 288 (Table 1 ; Figure 1 ). In other words, disease-causing agents were likely to infect more individuals 289

if they entered into a population by first infecting monkeys that were more central in human co-290 interaction networks, compared to by first infecting monkeys that were more central in grooming 291

networks. Contrary to our predictions, bonnet macaques also showed the same (rather than the 292 opposite) effect as rhesus macaques, although the magnitude of difference was somewhat lesser 293 than for rhesus (Table 1; Figure 1 ). Finally, although the centrality of first-infected macaques 294

within their co-interaction networks once again showed an overall greater effect on outbreak 295

sizes than the centrality of macaques within their grooming networks for long-tailed macaques, 296 this difference was not significant (Table 1 ; Figure 1 ). Moreover, long-tailed macaques also 297 seemed to show separate groupings within each network-type ( Figure 1 ). In other words, they 298 seem to show intra-specific differences in the effects of the network centrality of macaques 299

within each network-type on outbreak sizes (Discussion). 300

For a given network-type but across host-species, we found a significant interaction 301 between species and strength centrality for both co-interaction networks and grooming networks 302 (Table 2) . For co-interaction networks, rhesus macaques showed the strongest effect of strength 303 centrality on outbreak sizes as predicted. Contrary to our predictions, bonnet macaques fell 304 within the range of rhesus macaques, and long-tailed macaques showed a significantly lower 305 effect than both rhesus and bonnet macaques (Table 2; Figure2) . For grooming networks, the 306 differences were in the directions we predictedbonnet macaques showed the strongest effects 307 of strength centrality on outbreak sizes, followed by long-tailed macaques, and finally rhesus 308 macaques that showed a significantly lower effect compared to bonnet macaques (Table 2;  309  Figure2 ). For all three species, the magnitude of the effects of strength centrality on outbreak 310 sizes was markedly greater for co-interaction networks compared to grooming networks ( Figure  311 2). In other words, across host-species, the infection of macaques that were central in their co-312

interaction networks led to consistently higher disease outbreaks (more individuals infected) than 313 the infection of macaques that were central in their grooming networks. 314

For grooming networks, but not for co-interaction networks, we also found a significant 315 effect of sex and dominance rank of the first-infected individual on mean outbreak sizes -316 disease outbreak sizes were higher when first-infected macaques within grooming networks were 317 females compared to males, and higher-ranking compared to lower-ranking individuals (Table  318 2). However, the magnitude of these effects were much lower than those of the strength 319 centrality of first-infected macaques (Table 2) . Finally, the overall anthropogenic exposure of 320 first-infected macaques, i.e. their frequencies of interactions with humans and times spent 321

foraging on human foods, had no impact on disease outbreak sizes ( macaques' co-interactions with people make them especially highly vulnerable to disease 397 outbreaks has implications for our understanding of contemporary evolution, and specifically of 398 the behavioral flexibility of wild animals living in dynamic, varying (peri)urban environments. In 399 a previous study, we showed that macaques that engaged in specific forms of social affiliation 400 that were shorter in duration were also more likely to jointly take risks in anthropogenic 401 environments by co-interacting with humans within the same time and space 64 . We speculated 402 that such joint risk-taking would better enable wild primates to procure high-energy 403 anthropogenic foods 64,68 . Here, our findings suggest that the benefits of procuring such foods 404 may be offset by high zoonotic risk. Evaluating these costs-benefits tradeoffs on the fitness and 405

survival of wildlife in human-impacted environments is important but may be complicated by the 406 frequent trapping and re-location of animals in (peri)urban environments, that may hinder long-407

term studies on these populations. For all three species, we found that the centrality of macaques within their co-interaction 414 networks consistently led to higher disease outbreak sizes compared to their centrality within 415 grooming networks. In a previous study, we revealed that macaques' grooming relationships 416

were not related to their tendencies to co-interact with people, which led us to speculate that the 417 patterning and distribution of social contact with humans versus may offer potentially different 418 'pathways' for disease transmission 64 . Testing this speculation, here we reveal that co-419

interactions with humans may generally pose an even greater risk of diseases reaching higher 420 outbreak sizes, than macaques' social interactions with conspecifics. This finding has major 421 implications the consideration of how human-environmental interactions impact both human and 422 environmental welfare and health outcomes in equal measurea public health approach referred 423 to as "One Health" 87,88 . and their social networks as targets for disease prevention and control within these human-441 natural systems (e.g., vaccination, antimicrobial treatment) 40 . 442 We found cross-species differences in the extent to which co-interaction networks more 443 strongly predicted the sizes of disease outbreaks compared to grooming networks. As predicted, 444 rhesus macaques were the most vulnerable to disease outbreaks through co-interaction networks, 445 and the least vulnerable through their grooming networks. This highlights the importance of 446 evaluating the relative effects of multiple (rather than single, as is often the case) single aspects 447 of animal ecology on disease transmission 38,83 . Rhesus macaques, more so than the other two 448 macaque species, may preferentially engage in affiliative behaviors such as grooming with close 449 kin or allies 62,67 ; the resultant sub-grouping of individuals within their social networks may 450 potentially function as 'social bottlenecks' to disease transmission in this species 10,92 . Yet 451 animals that show sub-divided social networks may nevertheless be vulnerable to outbreaks 452 through other types of associations, and often in specific social-ecological contexts around 453

human-provisioned food that may cause wild animals to aggregate together 86 and co-interact 454

with people (as we have shown 64 ). 455

Contrary to our predictions the effects of co-interaction networks on outbreak sizes in 456 bonnet macaques were marginally greater (rather than lesser) than the effects of grooming 457 networks, and were in fact within the range of rhesus macaques. One reason for this may be the 458 spatial distribution of human-wildlife interactions in this population. Bonnet macaques are less 459 geographically widespread and ecologically flexible compared to rhesus macaques 84 . Although 460 the bonnet macaques in our study experienced markedly lower frequencies of interactions with 461 humans compared to rhesus macaques and long-tailed macaques 12 , these interactions were highly 462 geospatially restricted to within specific areas or 'blocks' within their home-range. It is likely 463 that such spatially dense social-ecological associations with people, through increasing the 464 connectivity of macaques within their co-interaction networks, leads to a considerable increase in 465 the risk of zoonotic outbreaks despite their relatively lower overall frequencies of interactions 466 with humans. More generally, this finding suggests that zoonotic agents may enter into and 467 rapidly spread even through populations of less ecologically flexible or socially gregarious 468 wildlife that, despite interacting less frequently with humans or their conspecifics, may 469 congregate within specific parts of their home-range around anthropogenic factors (e.g., contexts 470 of food provisioning, crop-foraging, ecotourism activity) 86 . Our network approach, through 471 capturing the spatiotemporal aspects of these human-wildlife interactions, provided a more 472 accurate estimation of zoonotic risk in these populations than just their overall frequencies of 473 interactions with humans. Aside from being the least ecologically flexible of the three species in 474 this study, bonnet macaques are also the most vulnerable to human-impact 82 , with many 475 populations facing the imminent threat of local extinction. Thus, the identification and treatment 476 of 'super spreader' individuals may be especially important in this population. 477

Contrary to our prediction, long-tailed macaques showed no differences in disease 478 outbreak sizes across network-types. At least one explanation for this may be intra-specific 479 variation, specifically between-group differences in their overall exposure to humans. We 480 observed two groups of long-tailed macaques at a Hindu temple and popular tourist location 481

within Kuala Lumpur, where the monkeys were exposed to dense human populations with whom 482 they interacted highly frequently 58 . On the other hand, we observed two other groups in at a 483 recreational park at the edge of the city bordering a fragmented forest area, where interactions 484 with humans were comparatively less frequent 58 . Moreover, long-tailed macaques also showed 485 marked differences in their grooming behavior across these locations as a response to 486 interactions with humans 58 . This explanation seems to be supported by the separate groupings for 487 the relationships between network centrality and outbreak sizes for long-tailed macaques, even 488

for the same network-type (Figure 1) . A more comprehensive assessment of the disease 489 vulnerability of these populations would require within-species, cross-group comparisons 490 (analysis on-going for a future study). 491

Disease outbreak sizes through macaques' grooming networks were generally higher 492 when the first-infected individuals were females compared to males, or when they were higher-493 ranking compared to lower-ranking individuals. Nevertheless, the effect sizes of sex and 494 dominance rank on disease outbreaks were a lot weaker than the effects of individuals' network 495

centrality. In many wildlife species, animals' sociodemographic attributes like their age, sex and 496

dominance rank may influence their behavioral ecology, specifically their life-history strategies, 497

social interactions, and their adaptive responses to changing (anthropogenic) environments 12,68 . It 498 may therefore be important to evaluate the potentially interactive effects of such factors with 499 animals' network connectedness on disease outbreaks. 500

The consistently stronger effects of strength centrality compared to betweenness 501 centrality or eigenvector centrality on outbreak sizes suggests that animals' direct connections 502 played a greater role in disease transmission than their secondary connections. This finding is 503 consistent with many previous studies (reviewed in 19,82 ), but not so with others (e.g., 504

betweenness as a stronger predictor of outbreaks across communities of humans 93 and 505 chimpanzees 40 ). Such differences in the role of direct versus indirect connections in disease 506 transmission may depend on the host population, network-type, or more global aspects of 507 networks 38,39,82 . These may be community modularity or the tendency of animals to form sub-508 groups 39 , inter-individual differences in network connectedness or centrality 80 , and the efficiency 509 of the flow or transfer of information through networks 94 . Examining how these global aspects of 510 macaques' co-interaction networks and grooming networks may impact infectious agent 511 transmission and consequential disease outbreaks in these populations would be a critical next 512 step. 513

Our results were independent of pathogen-specific transmissibility which, through 514 influencing basic reproduction numbers (R0 values), may strongly impact disease outbreaks. We 515 chose to account for, rather than quantitatively evaluate, the effects of a suite of zoonotic 516 respiratory pathogens of different transmissibility (e.g., influenza virus, measles virus, 517

Mycobacterium spp., SARs-CoV-2) 38 inter-pathogen differences would need to be considered while constructing more sophisticated 529 but system-specific epidemiological models of disease transmission for these and other HWIs. 530

In conclusion, our findings suggest that in the Anthropocene, wild animals remain highly 531 vulnerable to zoonotic outbreaks through their social-ecological interactions with humans, in 532 addition to their social interactions with conspecifics. Even in ecologically flexible, (peri)urban 533 wildlife, disease-related costs may likely outweigh the potential or perceived benefits of 534 increased assess to anthropogenic food. From One Health perspectives, our network approaches 535 and findings demonstrate the importance of considering animals that consistently co-interact 536 with humans across time and space (rather than just those that frequently interact with humans or 537 their conspecifics), as targets for disease control. This is critical for preventing disease outbreaks 538 in wildlife, but also for preventing cross-species wildlife-to-human disease spill-over events 539

which have the potential to trigger future global pandemics. 540 541

Methods: 542 543

Study Locations and Subjects: 544 545

We observed 10 macaque groups representing three different species at human-primate 546

interfaces across three locations in Asiafour groups of rhesus macaques in Shimla in Northern 547

India ( Figure 1) . All macaque groups were observed 551 in (peri)urban environments, and their home-ranges overlapped with humans and anthropogenic 552

settlements -e.g., Hindu temples (Shimla and Kuala Lumpur), recreational parks (outskirts of 553

Kuala Lumpur, Thenmala), roadside areas (Thenmala, Shimla)to varying extents 12,64,68 . 554

Subjects were adult male and female macaques which were pre-identified during a two-month 555 preliminary phase prior to data collection at each location. More details regarding the study 556 locations, macaque group compositions and subjects, and observation efforts, may be found in 557 our previous publications 12,6864 and in Supplementary Table 1.  558  559  Data Collection:  560  561 We collected behavioral and demographic data in a non-invasive manner using 562 observation protocols that were standardized across observers within and across locations 563 (details in 12,68,95 ). All data were collected for five days a week, between 9:00 am and 5:00 pm. 564

To record and spatiotemporally capture variation in human-macaque social-ecological 565

interactions for the construction of co-interaction networks, we used an event sampling 566 procedure 96, 97 . For this we divided pre-identified parts of the home range of each macaque group 567 in which human-macaque interactions were most likely to occur, into blocks of roughly equal 568 area and observability 64 . We visited these blocks in a pre-identified, randomized order each day. 569

Within a 10-minute sampling period, we recorded interactions between any pre-identified subject 570 macaque and one or more humans that occurred within that block, in a sequential manner. 571

Human-macaque interactions included all contact and non-contact behaviors initiated by 572 macaques towards humans (e.g., approach, aggression, begging for food), or vice-versa (e.g. 573

approach, aggression, provisioning with food) (more details in 12,68,95 ). We undertook this 574 sampling approach of visiting blocks at random in order to avoid over-sampling of human-575 macaque interactions in more (versus less) densely populated areas of macaques' home-ranges. 576

To record macaques' social behavior, and their overall anthropogenic exposure 577 independent of spatiotemporal context, we used focal animal sampling 96 . For this we followed 578 individual subjects in a pre-determined, randomized sequence for 10-minute durations. In a 579 continuous manner, we recorded, within each focal session, instances of social grooming, and 580 dyadic agonistic interactions that involved aggression (threat, lunge, chase, attack) that was 581 followed by submission (avoidance, silent bared teeth, flee), between the focal animal and its 582 group conspecifics. We also recorded interactions between the focal animal and one or more 583 humans in a continuous manner (see above for definitions). Once every two minutes, we ceased 584 recording continuous data to conduct a point-time scan 96 of the focal animal's main activity, i.e. 585 one of resting, locomotion, socializing, interacting with a human, foraging on natural food, or 586

foraging on anthropogenic food. More details regarding the data collection protocols and 587

definitions of behaviors may be found in our previous publications 12,68,95 . 588

We entered all data into Samsung Galaxy Tablets using customized data forms created in 589

HanDBase® application (DDH software). From these we exported and tabulated all the data into 590 MS Excel and MS Access databases daily. All observers within and across locations passed 591

inter-observer reliably tests using Cohen's kappa (> 0.85) 98 . 592 593

Construction of Co-interaction Networks and Social Grooming Networks: 594 595

From the human-macaque interactions collected using event sampling data, we 596 constructed social-ecological co-interaction networks ( Figure 3A ). In these, nodes were 597 individual macaques. Edges were based on the frequency with which pairs of macaques jointly 598 engaged in interactions with humans at the same block and within the same event sampling 599 session, per unit of event sampling observation time during which both members of the pair were 600 present in the group and (thereby) observable 64 . Since our data were limited by lacking the 601 identities of individual people, we refrained from constructing and analyzing bi-modal social-602 ecological networks with both in which both macaques and humans were considered as nodes (as 603 in 99 ). Instead, we "projected" interactions between humans and macaques into co-interaction 604 networks, in which nodes were macaques and edges defined by their joint tendencies to interact 605 with humans within the same time-frame and anthropogenic space. We thus subject these social-606 ecological interactions to SNA (similar to the conversion of primate-parasite bimodal networks 607 to social networks in 65 ). 608

We constructed macaque-macaque social grooming networks using the focal sampling 609 data ( Figure 3B ). In these, we linked individual macaques (nodes) based on the frequency which From the data on dyadic agonistic interactions with clear winners and losers, we 644 calculated macaques' dominance ranks for each group, separately for male-male and female-645 female interactions, using the network-based Percolation and flow-conductance method (Package 646

Perc in R 101 ). Perc is a network-based ranking method that combines information from direct 647 dominance interactions with information from multiple indirect dominance pathways (via 648 common third parties) to quantify dyadic dominance relationships, and yield ordinal ranks from 649 such relationships 101 . Aside from being successfully implemented in our previous studies (e.g. 650 30, 58 ), this method has been shown to yield animal rank orders that are highly consistent with 651 those yielded by other, popularly used ranking methods in behavioral ecology, such as David's 652 score, I&SI ranks, and Elorating 102 . As with network centrality, we converted ordinal ranks of 653 macaques within each group into percentile values that ranged between 0 (lowest-ranked 654 individual) and 1 (highest-ranked individual). From the continuously collected focal sampling 655 data, we calculated frequencies of human-macaque interactions per unit focal observation time. 656 We also calculated, for each macaque, its time spent foraging on anthropogenic food as the ratio 657 of the number of point-time scans in which it was foraging on anthropogenic food (Fa) to the 658 total number of scans in which it was foraging on either anthropogenic food (Fa) or natural food 659 (Fn), i.e. Fa/ (Fa + Fn). 660 661

Zoonotic Disease Simulations: 662 663

To simulate the spread of zoonotic agents of varying transmissibility () on macaques' co-664 interaction networks and grooming networks, we ran a series of Susceptible Infected Recovered 665 (SIR) epidemiological models (using the Epimdr R package: 103 ) ( Figure 4A, B) . We define '' as 666 a pathogen-specific characteristic, i.e. its probability of infecting a susceptible host within its 667 infectious period 38,41 .  is a function of the probability of pathogenic infection ( ) and recovery 668 rate (), and is calculated as /( + ) 38 . For each network-type (human co-interaction, social 669 grooming) and macaque group, we ran 5000 model simulations, 500 for each of 10 different 670 values of  ranging from 0.05 -0.50 in increments of 0.05. These selections were based on the 671 human literature that indicates that these values of  correspond to zoonotic agents that range 672 from low (e.g., influenza virus 104,105 ), to moderate (e.g., respiratory pathogens like 673

Mycobacterium spp. and SARS-CoV-2 38,106 ), to high (e.g., measles 17 ) contagiousness, and 674 average basic reproduction numbers (R0) of between 1.6 -14.0 38,40 . We thus ran a total of 675 100,000 simulations (5000 per macaque group times 10 groups times two network-types). In 676 each simulation, we deemed all macaques within a group to be initially 'susceptible', and then 677 infected one individual (node) at random with an artificial zoonotic agent of a given . A 678 simulation proceeded using a discrete time, chain binomial method 38,107 that dynamically and 679 temporally tracked the spread of infection through a weighted, undirected network through time 680

(example in Figure 4B ). In each simulation, animals were allowed to transition from 681 'susceptible' to 'infected' states, as a function of their network connections to individuals already 682 in 'infected' states and the pathogen  value. 'Infected' individuals were then allowed to 683 transition into 'recovered' states at a fixed recovery rate () of 0.2 that corresponds to an average 684 infectious period of five days 38 . Each simulation was allowed to proceed until the disease 685 proceeded to extinction when there were no remaining infected individuals in the network. At the 686 end of each simulation, we calculated the disease outcome of 'mean outbreak size', as the 687 average % of infected macaques (the number of 'infected' individuals divided by the total 688 number of individuals) across all time-units of the simulation. We also extracted, for each 689 simulation, the identity of the first-infected macaque 'k' (Figure 4A ), and calculated an average 690 of disease outbreak sizes from across all its first-infected simulation runs. We then matched this 691

individual-level mean outbreak size with the sociodemographic characteristics, network 692 centrality, and overall anthropogenic exposure of this (first-infected) individual. of the centrality of the first-infected macaque by network-type (co-interaction versus grooming) 714

for a given host-species (bonnet or long-tailed or rhesus) on mean outbreak sizes, we ran three 715 sets of three GLMMs each, one for each macaque species (details in Table 3A ). In all models, we 716 set the number of macaque subjects within the group (or 'effective group size') to be an offset 717

variable, since group size can impact our outcome variable of mean outbreak sizes 38,39 . In all 718 models, we also included 'animal ID' (a repeated measure for co-interaction networks and 719 grooming networks) nested within macaque 'group ID' as a random effect to control for 720 intraspecific variation. For each species, we ran three models, in each of which we included just 721 one of the three different network measures of the centrality of the first-infected macaque, i.e. the 722 strength, betweenness, or eigenvector, as a main effect. We favored this approach in order to 723 avoid the confounding effects of potential inter-dependencies of network centrality measures 72 . 724

In each of these three models, we also included an interaction term of network centrality by 725 network-type (co-interaction versus grooming), to determine whether the magnitude of these 726 effects were different for different types of interactions. In all models, we also included, as main 727 effects, the sociodemographic attributes (sex, dominance rank) and the overall anthropogenic 728 exposure (frequencies of interactions with humans, proportions of time spent foraging on 729 anthropogenic food) of the first-infected macaque. From each model-set of three models, we 730 identified a single best-fit model with a difference in AICc of at least 2 points or lower than the 731 next best-fit model 77 . 732

Second, to examine the effect of the centrality of the first-infected macaque by species 733

(bonnet versus long-tailed versus rhesus) for a given network-type (co-interaction or grooming) 734 on mean outbreak sizes, we ran two sets of three GLMMs each, one for each network-type 735 (details in Table 3B ). Once again, we set the number of macaque subjects to be an offset 736 variable, and included 'group ID' as a random effect, in all the models. For each network-type, 737

we once again ran three models, in each of which we included just one of the three different 738 measures of the centrality of the first-infected macaque as a main effect. In each of these three 739 models, we also included an interaction term of network centrality by host-species (bonnet 740 versus long-tailed versus rhesus), to determine whether the magnitude of these effects were 741 different for different species. Once again, we also included, as main effects, the 742 sociodemographic attributes (sex, dominance rank) and the overall anthropogenic exposure 743

(frequencies of interactions with humans, proportions of time spent foraging on anthropogenic 744 food) of the first-infected macaque. From each model-set of three models, we identified a single 745 best-fit model with a difference in AICc of at least 2 points or lower than the next best-fit 746 model 77 . 747

All GLMMs met the necessary assumptions of model validity (i.e., distribution of 748 residuals, residuals plotted against fitted values 110 ). All statistical tests were two-tailed, and we 749 set the p values to attain statistical significance to be < 0.05. 750 751 van Berkel for their involvement in data collection, processing, and storage in the field. The data 771

for this study was collected as part of a human-primate Coupled Natural and Human 

Competing Interests

Risk of human-to-792 wildlife transmission of SARS-CoV-2

Emerging 795 infectious disease and the challenges of social distancing in human and non-human 796 animals: EIDs and sociality

From Cheetahs to Chimpanzees: A comparative review of the drivers of 798 human-carnivore conflict and human-primate conflict

Human-Wildlife Conflict and Coexistence

The hidden dimensions of human -wildlife 803 conflict: Health impacts, opportunity and transaction costs

Co-managing human-806 wildlife conflicts: A review

Primate-parasitic zoonoses and anthropozoonoses: a literature review. 808 in Primate Parasite Ecology: the Dynamics and Study of Host-Parasite Relationships 809

Infectious disease risk across 811 the growing human-non human primate interface: A review of the evidence

Emerging Infectious Diseases of wildlife -814 threats to biodiversity and human health

Infectious disease 816 and group size: more than just a numbers game

Infectious disease, behavioural flexibility 819 and the evolution of culture in primates

Impact of individual demographic and social factors on 821 human-wildlife interactions: a

A conceptual model for the integration of social and ecological 823 information to understand human-wildlife interactions

The influence of life history milestones and 825 association networks on crop-raiding behavior in male african elephants

A common monkey of Southeast Asia: longtailed macaque populations

Monkeys on the edge: 829 ecology and management of longtailed macaques and their interface with humans

Primate 832 infectious disease ecology: insights and future directions at the human-macaque interface. 833 in The Behavioral Ecology of the Tibetan Macaque

Infectious Diseases of Humans: Dynamics and Control

Heterogeneity in pathogen transmission: 838 mechanisms and methodology

Disease transmission in animal social networks

Infectious disease transmission and contact networks in wildlife and 843 livestock

Network Models: An Underutilized Tool in Wildlife 845 Epidemiology?

Networks and the ecology of parasite transmission: A framework for 847 wildlife parasitology

Social network-based distancing strategies to flatten the COVID-19 curve 849 in a post-lockdown world

Contact networks in a wild 851 Tasmanian devil (Sarcophilus harrisii) population: Using social network analysis to reveal 852 seasonal variability in social behaviour and its implications for transmission of devil facial 853 tumour disease

Network structure and parasite 855 transmission in a group living lizard, the gidgee skink, Egernia stokesii

Linking social and 858 pathogen transmission networks using microbial genetics in giraffe (Giraffa 859 camelopardalis)

Who infects whom? Social networks and tuberculosis transmission in wild 861 meerkats

Monkeys in the Middle: Parasite Transmission through the Social 863 Network of a Wild Primate

Brown spider monkeys (Ateles hybridus): a model for differentiating 865 the role of social networks and physical contact on parasite transmission dynamics

Social 868 buffering and contact transmission: network connections have beneficial and detrimental 869 effects on Shigella infection risk among captive rhesus macaques

When individual behaviour matters: 871 Homogeneous and network models in epidemiology

The evolution of social behavior

Quantifying microbe 876 transmission networks for wild and domestic ungulates in Kenya

Growing artificial societies: social science from the bottom up

Semantic Information Processing

Compartmental models in epidemiology, chapter 2

Modelling the influence of human behaviour on 885 the spread of infectious diseases: A review

Disease implications of animal social network structure: A 887 synthesis across social systems

Community structure and the spread of infectious disease in 889 primate social networks

Network-based vaccination improves prospects for disease control in 891 wild chimpanzees

Implications of 893 Tourist-Macaque Interactions for Disease Transmission

Dating of the human-ape splitting by a 895 molecular clock of mitochondrial DNA

The International Encyclopedia of 897

The spatial ecology of chacma baboons (Papio ursinus) 899 in a human-modified environment

Twenty-year changes in rhesus 901 monkey populations in agricultural areas of Northern India

A common monkey of Southeast Asia: longtailed macaque populations, 904 ethnophoresy, and their occurrence in human environments

Managing humans, managing macaques: Human-908 macaque conflict in Asia and Africa

Female dispersal, social organization, and infanticide in langur: Are they 911 linked to human disturbance?

Patterns of gastrointestinal bacterial exchange between chimpanzees 913 and humans involved in research and tourism in western Uganda

Gastrointestinal 916 bacterial transmission among humans, mountain gorillas, and livestock in Bwindi 917 Impenetrable National Park

Human-nonhuman primate disease transmission

Threatened Hosts, Threatened Parasites? -Parasite 922 Diversity and Distribution in Red-Listed Primates

The human-924 primate interface in the new normal: challenges and opportunities for primatologists in the 925 COVID-19 era and beyond

Comparative ACE2 927 variation and primate COVID-19 risk

Status of the long-tailed macaque macaca fascicularis in Singapore and 930 implications for management

Current situation and status of long-tailed macaques 932 (Macaca fascicularis) in Thailand

The International Encyclopedia of Primatology

Time constraints imposed by anthropogenic environments alter social 936 behaviour in long-tailed macaques

Monkeys on the Edge: Ecology and Management of Long-tailed Macaques 940 and their Interface with Humans

Evolution of Primate Social Systems

The Evolution of Female Social 945 Relationships in Nonhuman Primates

Unity in diversity: Lessons from macaque societies

The role of Macaca fascicularis in infectious disease 949 transmission

Macaques and their Interface with Humans

Implementing network approaches to understand the socio-953 ecology of human-wildlife interactions

Centrality in primate-parasite networks reveals 956 the potential for the transmission of emerging infectious diseases to humans

India's rhesus population: Protection versus 959 conservation management

The influence of phylogeny, social style, and 963 sociodemographic factors on macaque social network structure

Individuals in urban dwelling primate species face unequal benefits 966 associated with living in an anthropogenic environment

Mixing patterns in networks

Social network analysis of animal 973 behaviour: a promising tool for the study of sociality

Constructing, conducting and interpreting animal social 975 network analysis

Some unique properties of eigenvector centrality

Identifying sets of key players in a social network

Friends of friends: are indirect connections in social networks important to 981 animal behaviour?

Model Selection and Multimodel Inference

AIC model selection and multimodel 985 inference in behavioral ecology: some background, observations, and comparisons

Null models for animal social network analysis and 988 data collected via focal sampling : Pre-network or node network permutation ?

Permutation tests for hypothesis testing with animal social 991 data: problems and potential solutions

A guide to null models for animal social network analysis

Integrating contact network 995 structure into tuberculosis epidemiology in meerkats in South Africa: Implications for 996 control

Social structure contains epidemics and regulates individual roles in 998 disease transmission in a group-living mammal

Integrating social behaviour, demography and disease dynamics in 1000 network models: Applications to disease management in eclining wildlife populations

Less than wild? Commensal primates and wildlife 1003 conservation

Quantifying microbe 1005 transmission networks for wild and domestic ungulates in Kenya

Urbanization and the ecology of wildlife diseases

One health, emerging infectious diseases and wildlife

From 'one medicine' to 'one 1012 health' and systemic approaches to health and well-being

Superspreading and the 1015 effect of individual variation on disease emergence

Unifying spatial and social 1017 network analysis in disease ecology

Network theory in the assessment of the sustainability of 1019 social-ecological systems

Social network community structure and the contact-mediated sharing 1021 of commensal E. coli among captive rhesus macaques (Macaca mulatta)

Dynamics and control of diseases in networks with community 1024 structure

Social networks in primates: smart and tolerant species have more 1026 efficient networks

Rates of human-macaque interactions affect grooming behavior 1028 among urban-dwelling rhesus macaques (Macaca mulatta)

Observational Study of Behavior: Sampling Methods

Network stability is a balancing act of personality, power, and conflict 1033 dynamics in rhesus macaque societies

Humans Dominate the Social Interaction Networks of 1036 Urban Free-Ranging Dogs in India

Network measures in animal social network 1038 analysis: Their strengths, limits, interpretations and uses

Perc: Using percolation and conductance to find information flow certainty 1041 in a direct network

Comparative 1043 Investigations of Social Context-Dependent Dominance in Captive Chimpanzees (Pan 1044 troglodytes) and Wild Tibetan Macaques (Macaca thibetana)

Estimated epidemiologic parameters and morbidity associated with 1047 pandemic H1N1 influenza

Estimating the 1049 reproduction number from the initial phase of the spanish flu pandemic waves in Geneva

Assessment of the SARS-CoV-2 basic reproduction 1052 number, R0, based on the early phase of COVID-19 outbreak in Italy

The Mathematical Theory of Epidemics

Fitting linear mixed effects models 1056 using lme4

A Language and Environment for Statistical Computing (R Foundation for Statistical 1058

Experimental Designs and Data Analysis for Biologists