<<

Downloaded by guest on September 27, 2021 miti- living. universally group of structure impact disease modular the that gates not hypothesis contact does the work local of our support estimates Overall, and necessary. data-limited are size when consequences group epidemic sufficient about be networks may knowledge heterogeneity null basic Otherwise, on subdivided. based highly are within interactions population that neces- the suggests show knowledge are struc- we prior structure when Finally, a only modular outbreaks. sary cause incorporating disease models also of network spread and that the subgroups to few delay structurally a tural to within able are infections subgroups trap cohesive fragmented with Highly networks cohesion. social subgroup net- organization: and modular net- fragmentation of social mechanisms work modular two highly by in driven burden is works disease illustrate of results lowering Our the smaller duration. that disease experience longer threshold that and modular although burden found disease structure, a we modular beyond by work, networks unaffected social prior largely is to influ- risk Contrary networks disease dynamics. social in to disease subdivisions model ence how network and theoretical when a investigate developed and net- modularity with associated work patterns we interaction studies, of features empirical the these identified mod- From societies. of animal consequences in to epidemiological structure species ular the animal of 43 study conflicting. from our been networks motivate social has empirical societies, hypothesis modu- analyzed animal this We Although in toward cost living. this evidence group reduce 2016) empirical 15, to of August thought cost review is for potential organization (received lar a 2017 2, is March approved risk and Disease NJ, Princeton, University, Princeton Levin, A. Simon by Edited 20892 MD Bethesda, www.pnas.org/cgi/doi/10.1073/pnas.1613616114 spread. disease of structure dynamics modular the affects how networks understand social mechanistically in and to studies) past need levels in high the investigated the typically exhibit subdivision networks modular social of animal past modular- all in of not ranges (because discrepancy realistic ity consider This to need 10). the higher implies (9, research to due subgroups size within outbreak connectivity increase have to few reduce structure a modular (6–8), subdivisions found bottlenecks modular structural creating that by Although suggested transmission past. the have in studies results some mixed spread- produced on have behavior, modularity processes of However, social ing impact of factors. the of combination landscape investigations a theoretical and from environmental, emerge species can demography, solitary and social relatively 5), including across forming (4, species, widespread group, wildlife is the in in structure more networks individuals modular other other Such each with subgroups. do with they interact than consistently often conspecifics of 3). sets (2, burden disease alleviate and fre- sug- size interactions been group higher social between has a in association it the to subdivisions Recently, modular due (1). that burden hosts associated gested between disease also contact elevated of is of quency living costs success, group the foraging Conversely, with young, defense. of group care and mat- cooperative predation, opportunities, from suc- protection ing improved competition, from intergroup benefit in cess societies Group-living living. group S 59715; networks a social animal Sah Pratha in structure modular mechanisms of and consequences disease the Unraveling eateto ilg,Gogtw nvriy ahntn C20057; DC Washington, University, Georgetown Biology, of Department oua raiaineegsi nmlgop hnsub- when groups animal in emerges organization Modular neouinr rdc fscaiyi nmlppltosis populations arthropods. animal in and sociality vertebrates of product in evolutionary ubiquitous An is behavior ocial c eateto ilg,Pnslai tt nvriy nvriyPr,P 60;and 16802; PA Park, University University, State Pennsylvania Biology, of Department | sociality a,1 tpa .Leu T. Stephan , | omnt structure community a alC Cross C. Paul , | idiedisease wildlife b ee .Hudson J. Peter , | infection b otenRcyMuti cec etr SGooia uvy oea,MT Bozeman, Survey, Geological US Center, Science Mountain Northern c 1073/pnas.1613616114/-/DCSupplemental at online information supporting contains article This 1 in ihnsbrusi qiaett h est finteractions subdi- of density of the to strength equivalent When is the subgroups (11). within as estimate tions networks serve to in can measure modularity, vision edges used Newman where spread. commonly infection species a of animal routes for 43 realistic networks across social groups Networks. available) Social 69 publicly Animal (and in published Disease obtained and Organization Modular Discussion and Results models null necessary. animal appropriate in are of consequences groups epidemic popu- use of estimates the animal data-limited suggest when in modular further spread which We disease in lations. infectious context the influences on structure insights study our mechanistic Overall, spread. provides disease accu- of build models to in network sufficient) contact organization rate (and necessary modular is networks of social knowledge animal and whether effects; investigate observed under disease behind (iii) conditions of mechanisms the dynamics the explain the and explore spread influences (ii) organization organiza- networks; modular modular social which of their strength in the tion increase that patterns action 1). vari- (Fig. drive size and subgroup individual members), in (i.e., subgroup ation with cohesion interact subgroup num- to the promote preferences (i.e., subgroups), fragmentation ones of network the associated of ber including are groups, the change that animal that in patterns subdivisions interaction modular with of features structural aadpsto:Terpiaindt eotdi hspprhv endpstdi the in deposited been have paper this in Dataverse, reported Harvard data replication The deposition: Data Submission. Direct PNAS a is article This interest. of conflict no declare paper. authors the The wrote P.S. and data; analyzed S.B. and P.J.H., P.C.C., uhrcnrbtos ..adSB eindrsac;PS efre eerh .. S.T.L., P.S., research; performed P.S. research; designed S.B. and P.S. contributions: Author n heaBansal Shweta and , owo orsodnemyb drse.Eal s7@ereoneuo shweta. or [email protected] Email: addressed. be may correspondence whom To [email protected]. feiei osqecsaenecessary. are consequences of epidemic estimates use of data-limited the when recommend models results network and null our networks appropriate these social validate animal through We real spread burden. using structural that disease induce infections reducing networks, of networks, trapping social and in delay modularity associated high both are with which cohesion, subgroup fragmen- high high and that tation show We subdivided. net- extremely social are when works by except unaffected organization, largely modular is underlying the spread demon- disease networks networks infectious theoretical that with social strates along group-living empirical species in animal of burden 43 analysis of disease our of However, cost hypothe- species. the is alleviate networks to social sized animal in organization Modular Significance dniyfaue faia inter- animal of features identify (i) objectives: three have We the of investigation driven empirically an propose we Thus, d oat nentoa etr ainlIsiue fHealth, of Institutes National Center, International Fogarty https://dataverse.harvard.edu PNAS | a,d,1 pi 8 2017 18, April . Q | 0 = (DOI: o.114 vol. www.pnas.org/lookup/suppl/doi:10. h est finterac- of density the , 10.7910/DVN/WP3BV7). | o 16 no. | 4165–4170 Q We is ,

ECOLOGY AB This is because, in principle, we expect the maximum modu- larity Qmax of all networks to be equal to one. However, for smaller networks, Qmax tends to be a much smaller value (Fig. 2A). We therefore estimated relative modularity (obtained by normalizing Q by Qmax ) to facilitate comparison of modular subdivision strength across different animal groups. Animal social networks in our database ranged from homogeneous (Qrel = 0) to highly subdivided structure (Fig. 2A). Animal interactions are dynamic in nature, and therefore social net- works of the same set of individuals can also fluctuate between being relatively homogeneous to highly modular over time C D (Fig. 2B). To gain intuition about how modular organization in ani- mal social networks influences epidemiological outcomes, we first generated homogeneous null networks for each empirical animal in our database. In homogeneous null networks, higher-order organization in social networks (such as modular structure, clustering coefficient, cliquishness, and degree ) are randomized, weakening relative modu- larity (SI Appendix, Fig. S1), but individual contact heterogene- ity () of the empirical network is preserved. We next performed susceptible–infected–recovered (SIR) dis- ease simulations of a moderately transmissible infectious disease (basic reproduction number, R0 = 1.2) through these empir- ical networks and their homogeneous null network counter- Fig. 1. Schematic representation of a homogeneous social network with parts. Based on previous work suggesting a protective effect no modular subdivision (A) and highly modular social networks with high of modular structure on disease risk (3, 6), we expected a cohesion within subgroups (B), high network fragmentation (C), and high lower disease burden (measured as the proportion of popu- subgroup size variation (D). Each network has 60 nodes. Orange edges lation infected) in empirical networks than in homogeneous indicate interactions within a subgroup, and purple edges are between- null networks. Reduced disease burden, however, was apparent subgroup interactions. Q is Newman modularity (11) and Qrel is relative only in social networks with Qrel > 0.6 for moderately spread- modularity. ing infectious disease (Fig. 2C and SI Appendix, Fig. S2). Addi- tionally, none of the empirical networks demonstrated a major reduction in disease burden for a highly transmissible infec- between subgroups. Higher values of Newman modularity indi- tious disease (R0 = 4.8, SI Appendix, Fig. S3). This implies that cate stronger subdivisions of social networks. However, using the protective effect of modular subdivisions is realized only Newman modularity to compare modular subdivision across net- at high values of relative modularity, but the presence of a works is problematic because the measure is inherently cor- threshold depends on pathogen transmissibility (SI Appendix, related to the size and number of links in the network (12). Fig. S3).

ABC

Fig. 2. Animal social networks with modular subdivisions. (A) Values of Newman modularity, Q, and maximum possible modularity given the network configuration, Qmax , across all animal groups in the database. Point color represents the relative modularity estimated as the ratio of Q over Qmax . The dashed line represents Q = Qmax and the maximum value of Qrel = 1. (B) Relative modularity values of dynamic interactions in ants (Camponotus fellah), raccoons (Procyon lotor), and field voles (Microtus agrestis), where time points represent consecutive days, weeks, and trapping sessions, respectively. (C) Comparisons between real (filled points) and their homogeneous null networks (tips of arrows) with respect to the percentage of infected individuals (outbreak size) due to an outbreak with basic reproduction number, R0 = 1.2. Point color corresponds to the taxonomic class of the animal group. Social networks with nonsignificant modular subdivision (as indicated by t test analysis) have been excluded in C. The generated homogeneous null networks preserve only the local heterogeneity of contacts among individuals; the arrows therefore indicate the change in direction and magnitude of outbreak size under the scenario in which all higher-order structural complexities (including modular subdivisions) are removed from animal social networks. The shaded area represents the

region where empirical networks tend to experience reduced outbreak size (at Qrel > 0.6; SI Appendix, Fig. S2). The networks that experienced lower disease burden included social networks of raccoons (P. lotor), elephant seals (Mirounga angustirostris), ants (Camponotus pennsylvanicus), bottlenose dolphins (Tursiops truncatus), and Australian sleepy lizards (Tiliqua rugosa). In A and C, we used a randomly selected network snapshot for groups with temporal interaction data.

4166 | www.pnas.org/cgi/doi/10.1073/pnas.1613616114 Sah et al. Downloaded by guest on September 27, 2021 Downloaded by guest on September 27, 2021 a tal. et Sah o estimated. not VIF 0.66 average size Subgroup variation cohesion Subgroup cohesion* Subgroup fragmentation* Network size Network Intercept variable Explanatory orga- modular of (Q determinants nization on analysis Multivariable 1. network Table of level the change can interactions social season- with driven species or ally fission–fusion individuals, in nomadic with members species However, species, group efforts. com- between sampling consistent interactions unequal for dynamic with database our species in across group parison animal each for tions state reproductive female 16). and 15, (13, hierarchy, resource dominance in as ratio), (such changes and sex factors by demographic dolphins, about aggression, brought intragroup bottlenose modularity. be availability, can network chimpanzees, and in lions) on (e.g., elks, African species effect social hyenas, many negative in spotted common weak subgroup are sizes in a subgroup variation Fluid have found to we fragmen- network cohesion, sizes to subgroup contrast and In [e.g., (15)]. tation competition monkeys intragroup spider brown high den- in or resource distribution, on constraints resource by sity, fragmen- caused network be can High conversely, (14)]. tation, elephants African matrilineal in or [e.g., together, in foraging descent [e.g., spent requirements time nutrient of (13)], similar hyenas result spotted to direct due a cohe- bonds be effects social subgroup can strong positive Cohesiveness factors, strong organization. three had modular on fragmentation the network Of and 1). siveness (Table variation subgroup size and net- fragmentation, social network animal cohesion, in subgroup works: organization modular driving features work correla- networks temporal intragroup these below). avoid analyze one separately we to (and selected metrics analysis network randomly in the we tion for network, network social tem- a snapshot multiple subgroups had of that among snapshots groups individual poral contacts For of in variation). degree number variation (subgroup in and variation (degree); in and contacts variation in average and variation average sizes; subgroups); network; subgroup within the occur across cohesion that subgroup contacts total propor- the of as measured tion cohesion, (subgroup the subgroup in own fea- association subgroups with social preferential of network fragmentation); the number (network these the) in network of social present of (log individuals size); some of (network there- to network number We driv- attention the networks. patterns including our tures, social interaction turned animal of first in features fore This subdivision varying networks. modular to mod- across ing the due inconsistent above be fairly reduction could was size outbreak threshold Addi- of ularity threshold. extent certain a the above tionally, only and strength modularity, the relative on of depending context-dependent, Networks. are structure Social ular Animal in 2C Organization Fig. Modular of Determinants ugopdge aito 0.01 Sociality 0.07 — effects Random variation degree Subgroup variation degree Individual average degree Individual variation* size Subgroup aooi order Taxonomic seik n odtx niaesgicne xlntr aibe with variables Explanatory significance. indicate text bold and Asterisks interac- of snapshot single a only considered have we far, Thus, net- three identified model beta-regression mixed-effects The > eedopdfo h oe,adteeoeterefc ie were sizes effect their therefore and model, the from dropped were 5 ugssta h pdmooia osqecso mod- of consequences epidemiological the that suggests rel cosgop fdfeetspecies different of groups across ) fetsz 5 ofiec intervals confidence 95% size Effect −0.03 −0.07 −0.42 −0.13 1.14 — aineetmt (σ estimate Variance <0.001 0.066 .4t 0.08 to −0.14 0.03 to −0.17 .8t 0.10 to −0.08 .9to −0.59 .6to −0.26 .3t 1.25 to 1.03 0.82 to 0.51 .4t 0.18 to 0.04 — — −0.25 −0.01 2 ) aacsteoealdsaesra,adtu oua structure, modular thus and spread, disease tradeoff overall This the 3C). net- (Fig. balances transmission these disease in global within contacts reduce contacts works intersubgroup of low density high conversely, a homoge- subgroups; to with due transmission Compared local higher 3C). (with subgroups (Fig. the networks within spread on neous transmission disease depends global disease modularity and local and between relation- burden The tradeoff disease networks? social between animal ship in burden disease reduce angustirostris). (. lotor oehs neato ewrs nordtbs,treani- in three occur database, our do In demonstrated rare, networks. networks. species although homogeneous mal interaction networks, were with host modular compared pathogens some highly 58% transmissible to Such moderately up of reduced sizes outbreak contagious (Q highly (transmissibility not but pathogens 0.08–0.16) = (transmissibility tagious 3 Fig. in with line solid ( threshold; outbreak homogeneous in in larity with reduction modularity 10% compared relative a of least size at level experience minimum thresh- networks the which epidemic calculated the we beyond old, transmissibility pathogen Sec- pathogens. for contagious low ond, for outbreaks the disease alleviate large networks of in modular risk highly line that implying modu- solid networks, low lar and threshold; with homogeneous with is networks (epidemic compared threshold found there outbreak epidemic We which large 3A ). a below Fig. of min- contagiousness the risk pathogen estimated no we of First, level ways. two imum in spread disease toward S12). Fig. mon- Appendix, spider empiri- (SI brown of (19) parasitic no network keys social gastrointestinal found modular a low we the of in addition, trapping infection structural In of S11). evidence Fig. cal Appendix , (SI of spread (18) the investigate parasites to data protozoan empirical parasite using global inhibit by networks transmission modular low) not (but transmissi- highly 3 that moderately (Fig. for pathogens networks ble modular experi- extremely size in only outbreak (Q of (Q networks homogeneous networks magnitude of modular the low net- by that modular enced found synthetic networks. on we social simulations works, in disease organization disease SIR epidemiolog- modular Performing the influence of isolate can to consequences appropriate ical networks is approach of this spread, features Because (17). the structural networks to degree several homogeneous close in as length) while realized path (such modularity, configuration and features of coefficient, clustering network strength homophily, higher-order tunable gen- model other a The networks keeping 17. with modular ref. networks in synthetic proposed erates used model networks? we our con- social using question, disease generated animal this the in answer determine subdivisions social To also modular animal factors sub- of in two sequences structure within the modular cohesion Do with networks. and associated fragmentation are groups network Spread. Disease that of Dynamics shown on Modularity of Impact The within modularity relative on correlation groups. negative animal weak subgroups among a contacts in had variation and size vari- subgroup addition, in In or ation S1 ). Table cohesion Appendix, subgroup (SI fragmentation higher network with previous modular groups our animals increasingly with that become found Consistent we interactions, available. aggregated are of groups analysis animal data few such a which for anal- for interactions the dynamic repeated considering therefore by We ysis 2B). (Fig. time over modularity h usinte swyd o l eeso modularity of levels all not do why is then question The enx unie h outeso oua oilnetworks social modular of robustness the quantified next We rel ,adnrhr lpatseals elephant northern and agrestis), (M. voles field ), Q > rel .)hdasrkn mato ies transmission— disease on impact striking a had 0.8) ≥ 0. 5 iiae ies pedfrmdrtl con- moderately for spread disease mitigated PNAS Babesia A Q rel | and pi 8 2017 18, April 0 = > Q ntesca ewrso edvoles field of networks social the in rel rel .6.Hgl oua networks modular Highly 0.16). .W orbrtdorfinding our corroborated We B). ,mdlrntok experience networks modular ), 0 = ausof values .W on htnetworks that found We B). .Otra iewsreduced was size Outbreak ). rel Q Q rel < rel | a iia othat to similar was 0.6) > 0 = o.114 vol. 0. .—acos(P. >0.8—raccoons ewrs(modu- networks ) 6 ohv higher a have to | o 16 no. ehave We | 4167

ECOLOGY infection in modular social networks. The structural trapping effect (albeit weak) has been observed in a field study of pneu- monia transmission in highly subdivided networks of bighorn lambs (SI Appendix, Fig. S13) (21) and others (SI Appendix, Table S2). We also found empirical evidence of structural delay in the spread of mycoplasma in the low-modular network, but highly cohesive subgroups, of house finches described in ref. 22 (SI Appendix, Fig. S14). A high probability of extinction associated A B with the structural trapping effect has been reported in theoret- ical studies of disease spread in spatially structured metapopu- lation models (7, 8). We note that the magnitude of structural trapping and delay depends on the contagiousness of the infection—moderately transmissible pathogens spread slowly through populations and are therefore able to perceive modular structures present in networks. For rapidly spreading pathogens, highly fragmented and cohesive networks do not contain, but delay, the spread of disease to other subgroups, creating only a structural delay effect.

CDIs Knowledge of Modular Structure Necessary (and Sufficient) to Predict Disease Outcomes? Our investigations of disease conse- quences on synthetic networks suggest that only modular sub- division beyond a certain threshold can reduce disease burden of moderately transmissible pathogens. Other topological features Fig. 3. Overall disease implications of modular subdivisions in synthetic of the network—and, in particular, heterogeneity in host con- modular networks. (A) Average outbreak size, measured as the percent- age of infected individuals, over increasing subdivided social networks and tact (23)—are, however, bound to confound the effect of mod- pathogen transmissibility. Outbreak size values have been normalized to the ularity on infectious disease spread. Our work thus far assumes maximum observed outbreak. The solid line indicates epidemic threshold— high variation in contacts among individuals (24), although fac- namely, the threshold value of pathogen contagiousness below which there tors including sociality (25) and nonpersistent space use (26) are is no risk of a large outbreak (> 10% outbreak size; Materials and Methods). known to reduce variation in individual contact rate. Does indi- (B) Epidemic robustness of networks with increasing value of relative mod- vidual contact heterogeneity then affect the way modular organi- ularity, measured as the percentage reduction in outbreak size compared zation influences disease spread? We found that for social net-

with outbreak size experienced by homogeneous (Qrel = 0) networks. The works with low contact heterogeneity, high modularity brings solid line indicates the modularity threshold, where networks experience at about a greater reduction in disease burden compared with net- least a 10% reduction in outbreak size. (C) Infection transmission events, works with high contact heterogeneity (SI Appendix, Fig. S8). expressed as the percentage of total outbreak size, within subgroups (local) These results imply that the local contact patterns should also and between subgroups (global); pathogen transmissibility = 0.18. (D) Dis- be considered when estimating the reduction of disease burden ease implications of modular subdivisions as a function of subgroup cohe- in extremely subdivided social networks. sion and network fragmentation (measured as the log-number of subgroups Although modular subdivision is a common feature observed in present in the network). many animal groups, it is often correlated with other higher-order network structures, such as clustering coefficient and degree homophily (27). In empirical networks, it is therefore impor- under most conditions, does not lower disease burden. The con- tant to ask whether modular organization has any epidemio- ditions under which modularity does reduce disease burden of logical consequence above and beyond these network proper- social networks is largely determined by the two factors underly- ties, after accounting for contact heterogeneity. To answer this ing network modularity—network fragmentation and subgroup cohesion (Table 1). Synthetic networks that possess both high subgroup cohesion and high network fragmentation experience lower outbreak size compared with homogeneous and low mod- 70 A B ular social networks (Fig. 3C and SI Appendix, Fig. S6). In addi- Cohesion Percent outbreaks 60 60 tion, high variation in subgroup sizes along with high subgroup 1.00 5 0.75 cohesion and fragmentation further reduce outbreak size and 0.50 10 0.25 15 50 0.00 increase outbreak duration in social networks (SI Appendix, Fig. 40 S7). Disease consequences in real animal social networks, how- ever, tend to be driven by subgroup cohesion because they typi- 40 cally exhibit a much lower range of fragmentation and subgroup 20 size variation compared with subgroup cohesion (SI Appendix, 30 Figs. S4 and S5).

Time to disease invasion 20 Uninfected subgroups (%) Uninfected 0 Mechanisms That Drive the Impact of High Modular Structure on Disease Spread: Structural Delay and Trapping. We next examined 12345678910 0.00 0.25 0.50 0.75 1.00 the mechanisms by which highly fragmented social networks Ordered subgroups Subgroup cohesion with cohesive subgroups experience reduced disease burden. The Fig. 4. Mechanisms behind the effect of modular organization on disease presence of only a few interactions between highly cohesive sub- spread of a moderately transmissible pathogen (= 0.1). (A) Structural delay groups increased the amount of time it took for an infection effect in a moderately fragmented network (fragmentation = 2.30): We con- to spread from one subgroup to another (the structural delay sidered the time to disease invasion for a subgroup as the number of time effect; Fig. 4A), causing longer disease outbreaks. When net- steps it takes for at least 2% of individuals to acquire infection. Subgroups works were fragmented into highly cohesive subgroups, the struc- at the x axis are sorted according to the increasing disease invasion time. tural delay imposed by cohesion caused infection to be local- (B) Structural trapping effect in a highly fragmented network (fragmenta- ized to a small proportion of subgroups before dying out (Fig. tion = 4.83): High fragmentation and subgroup cohesion localize infection to 4B). Popularly known as structural trapping (20), this effect also a small proportion of subgroups in the social network. Structural trapping reduces the overall likelihood of a major outbreak by a novel also increases the likelihood of stochastic extinction of disease outbreaks.

4168 | www.pnas.org/cgi/doi/10.1073/pnas.1613616114 Sah et al. Downloaded by guest on September 27, 2021 Downloaded by guest on September 27, 2021 a tal. et Sah and Q ubr eoesprt ruso h aespecies. same the of groups separate denote Numbers PC, tonkeana; Macaca MA, canus; DR, cattle; dairy DC, sylvanicus; CC, values error percentage secondary curve, modularity solid relative of (red value increasing the social to The according network. ordered empirical are to networks the of due networks null homogeneous networks or modular social animal 19 S for with outbreak networks an null homogeneous and 5. Fig. sup- evidence past species, group-living allevi- of to burden disease postulated the been ate has organization social modular While Conclusions burden. disease true the estimating in per- better networks null formed modular conversely, strain, moder- the transmissible For ately strain. low-transmissibility the outbreak for accurate predictions and size identical produced networks modular homogeneous and predictions, theoretical the to Congruent S15). Fig. outbreak predict lizards, to two networks of null size homogenous the relative performance and the confirm modular of compared empirically we of levels simulations, and To disease all our S10). of homogeneous Fig. across results Appendix, of similar (SI are networks, estimates modularity networks the size null of outbreak modular complexities therefore higher-order and by unaffected infec- spreading (R are moderately slowly to the diseases, contrast overestimate tious In therefore, burden. and, disease networks true struc- empirical modular effect. the the trapping of preserve structural ture not the do to networks null due Homogeneous structure this at modular size of outbreak low level experience networks social animal tions, (Q modular- high relative was when networks empirical except of networks, outcome ity social disease animal predicting most in for well equally performed however, and percent- networks zero) 5 null not modular (Fig. from (but predictions low size outbreak found in the We error age 2C). to Fig. (similar in randomized used was networks networks distribution, struc- degree null network except all modular where ture, networks, of null homogeneous performance of that the if with compared networks subdivisions. null also modular error of modular We effect high the from a mask features prediction expected structural size other ran- We outbreak but structure. in counterparts, network rate other network all empirical domize their degree and of level modularity distribution the networks. preserve null modular networks animal their null and 19 Modular database through our in simulations networks disease social performed we question, emp rel ecpteu campbelli Cercopithecus × IAppendix SI au ssmlrt h hehl dnie bv Fg 2C (Fig. above identified threshold the to similar is value 0,where 100, IAppendix, (SI 28 ref. in described data using rugosa, T. ecnaeerri ubeksz rdcin sn modular using predictions size outbreak in error Percentage MF, angustirostris; M. .Hmgnosnl networks, null Homogeneous S9). Fig. Appendix, SI amnlaenterica Salmonella R S i.S2 Fig. , 0 ubeksize, outbreak = = TR, cynocephalus; Papio .Pretg ro scluae s(S as calculated is error Percentage 1.2. BA, <15%. y CCR, ; xs.Tesae einidctsterneof range the indicates region shaded The axis). 0 .Frmdrtl rnmsil infec- transmissible moderately For ). < HM, rotundus; Desmodus MM, fuscata; Macaca )adrpdysraiginfections spreading rapidly and 1) CF, crocuta ; Crocuta BB, arachnoides ; Brachyteles emp tan nteAsrla sleepy Australian the in strains miia ewr,and network, empirical = rel > TT, rugosa; T. .) ent htthis that note We 0.6). CP, fellah; C. MT, mulatta; Macaca amrosmexi- Haemorhous .truncates. T. emp io bison; Bison .penn- C. − null S null are )/ ihysniiet h tt fntokmdlrt ttetm of time the at modularity be network of outbreaks. will disease state however, the periods, period. to infectious sensitive infectious net- highly short average of with the scale infections than time Acute shorter the is if relatively fluctuations modularity be work network will high period by infectious addition, long unaffected consequently, In a connected. with and, more infections contact, appear chronic might temporary subgroups of social role the the windows time net- amplify large social may these over of interactions ignore modularity pooling may Conversely, the works. window overestimate aggre- time and small contacts subgroups, fleeting a infre- between over be particularly interactions can gating interactions intermittent, animal or many disease quent Because of 29). models network (7, how- accurate pathogen, spread the developing the to to of crucial relative is period ever, interactions infectious animal and of of mode Consideration transmission interval simulations. time relevant disease relevant scale our the temporal in the spread with infection coincide aggre- net- to database dynamic the our of that “snapshots” in assume temporal works and we networks cri- Second, static weighting networks. gated spread appropriate contact the disease of identify historical terion to or available, populations unclear where captive data, transmis- generally in leveraging is studies advise potential sion We transmission context-dependent. inten- usually on or and contact) duration, of frequency, of represent to sity impact they the individuals because the (whether between was filtered weights This interactions we networks. First, num- social unweighted system. construct a to specific made assigned a we before to weights considered results groups, be the should taxonomic that extending assumptions of simplifying range of ber broad a across communica- through information ecological of networks. spread and tion as metabolic well transportation in as and spread networks, networks perturbation dis- interaction and infectious human networks, S2), in Table at spread Appendix , processes ease (SI spreading scales including systems, ecological applicable of potentially higher range are study wide this a mod- of across synthetic results in the spread networks, contagion ular on of based investigation is analysis systematic our transmis- a Because local of modularity. network reduce spread high but at global modularity, sion network minimize An low to modularity. for network aim infection control of should of levels intervention design different the effective for modularity, burden vary of disease will levels overall overall strategy most the although at reduce that, unaffected therefore note is and We subgroups burden. few disease a frag- to and transmissible moderately infections cohesion trap high structurally subgroup networks have except in High and mentation groups, subgroups. fragmented within extremely animal largely are cohesion is in networks burden subdivisions social disease when modular that to by shows Contrary 3). study unaffected (2, our living group hypothesis, of this costs disease the reduce works animal in organization societies. modular bio- of and mechanisms disease of model into consequences insights network derive to theoretical simulation disease this realistic logically with species across networks animal network social 43 published higher-order combined of we net- Third, presence social properties. the animal minimizing empirical while (of works), networks mod- null synthetic of and model generation networks mechanistic a ular the using epidemio- for by allows the that organization (17) modu- studied modular to of systematically lead consequences we logical can Second, that structure. mechanisms distinct lar two identified and modular- relative nor- called (Q a modularity introduced ity Newman we of First, version ways. three malized consequences in disease modularity on network research we of past study, of this In ambiguity before. the identified resolved been not have networks subdivided social be animal to cause that patterns interaction the of importantly, More features equivocal. been has hypothesis this porting eas hssuyivle eaaayi fsca networks social of meta-analysis involved study this Because net- social animal in subgroups that hypothesized been has It rel oalwcmaio cosntok fdfeetsizes different of networks across comparison allow to ) PNAS | pi 8 2017 18, April | o.114 vol. | o 16 no. | 4169

ECOLOGY This study suggests the presence of a high modularity thresh- size, etc., toward network modularity, we ran a mixed-effects beta regres- old, above which social networks experience reduced outbreak sion model using the glmmADMB package (Version 0.8.3.3) in R (Version size. However, we caution against the use of a single mod- 3.2.3). We treated the species nested within the taxonomic order of the ani- ularity threshold that is applicable to all systems. Multiple mal group as a random effect in the model to account for any correlations behavioral, ecological, and epidemiological factors—including in network metrics. However, the variance accounted for by the nesting of 2 pathogen infectiousness (Fig. 3B), infectious period, local con- species was negligible (σ ≈ 0), and we therefore considered only taxo- tact heterogeneity, seasonality of contacts, and transmission, nomic order as the random effect in the analysis. We also accounted for coupled with recruitment of susceptible through births—can differences between the social organization of different species by defining three broad levels of sociality—relative solitary, social, and fission-fusion— influence the Qrel threshold and the extent to which network fragmentation and subgroup cohesiveness structurally trap and and included sociality as a random effect in the model. delay infection spread. Therefore, to assess the disease implica- Generation of Synthetic Modular and Null Networks. We used the network tions of modular structure observed in specific social groups, we generator described in ref. 17 to generate networks of a specified modu- recommend the use of null networks, as presented in this study. larity and degree sequence, while keeping other network properties close Comparison of disease consequences on empirical network data to homogeneous null networks. For Figs. 3 and 4, we generated synthetic with those on a hierarchy of null networks clarifies whether mod- modular networks with 10,000 nodes, with an exponential degree distribu- ular structure is important. If natural history and limited obser- tion with a mean degree of 10. Two types of null networks were generated vations suggest a species’ network will be highly modular, then for each animal social network. Modular null networks were created by ran- models that incorporate modularity should be used; otherwise, domizing within- and between-subgroup connections. The modular null net- homogeneous null models (based on basic knowledge about net- works therefore had identical modular configuration and degree sequence as work size and local heterogeneity) may be sufficient when data- empirical networks, but were random with respect to other higher-order net- limited estimates of epidemic consequences are necessary. work properties. We generated homogeneous null networks by performing simultaneous edge swaps over within- and between-subgroup connections, Materials and Methods which preserves the local contact heterogeneity, but randomizes all higher- order features of the network, including network modularity. Extended methods are provided in SI Appendix, SI Text. Disease Simulations. We performed Monte Carlo simulations of a discrete- Measuring Modularity. We used modularity (Q) proposed by Newman (11) time SIR model of disease spread. Because we were interested only in major to measure the strength of modular organization in networks. Modularity outbreaks, we considered only those simulations in our calculations where  Lw 2 PK k  Lk  can be defined as Q = k=1 L − L ), where Lk is the total number at least 10% of the population acquired infection. Transmission occurred according to a pathogen transmissibility, defined as the probability of trans- of edges in a subgroup k, of which Lw are the edges within the subgroup, k mission from an infected to susceptible host during the period when the and L is the number of total edges in the network. , or the number and composition of subgroups, for each animal social network infected host is infectious. The recovery probability was assumed to be con- was estimated by using the Louvain method (30). The highest possible mod- stant across hosts. ularity in a network (Q ) is achieved when all individuals in a subgroup k max ACKNOWLEDGMENTS. We thank Phil O’Neill for feedback on a previous only interact with each other and no edges are present between subgroups version of the manuscript. We thank all the researchers who have made their (i.e., subgroups are disjointed). In other words, Qmax of a network is when animal social network and disease data openly accessible, without which w PK Lk  Lk  Lk = Lk, and can be written as Qmax = k=1 L 1 − L . We measured this study would not have been possible. This work was supported by the National Science Foundation Ecology and Evolution of Infectious Diseases the relative modularity of networks as Q = Q . rel Qmax Grant 1216054. S.T.L. was supported by a postdoctoral Endeavour Research Fellowship from the Australian Department of Education and Training. Any Mechanisms of Modular Organization. To examine the relative contribution mention of trade, firm, or product names is for descriptive purposes only of factors such as network size, network fragmentation, average subgroup and does not imply endorsement by the U.S. government.

1. Altizer S, et al. (2003) Social organization and parasite risk in mammals: Integrating 16. Lynch Alfaro JW (2007) Subgrouping patterns in a group of wild Cebus apella nigritus. theory and empirical studies. Annu Rev Ecol Evol Syst 34(1):517–547. Int J Primatol 28(2):271–289. 2. Griffin RH, Nunn CL (2011) Community structure and the spread of infectious disease 17. Sah P, Singh LO, Clauset A, Bansal S (2014) Exploring community structure in biological in primate social networks. Evol Ecol 26(4):779–800. networks with random graphs. BMC Bioinformatics 15:220. 3. Nunn CL, Jordan F, McCabe CM, Verdolin JL, Fewell JH (2015) Infectious disease and 18. Davis S, Abbasi B, Shah S, Telfer S, Begon M (2015) Spatial analyses of wildlife contact group size: More than just a numbers game. Philos Trans R Soc Lond B Biol Sci networks. J R Soc Interface 12(102):20141004. 370(1669):20140111. 19. Rimbach R, et al. (2015) Brown spider monkeys (Ateles hybridus): A model for differ- 4. Mourier J, Vercelloni J, Planes S (2012) Evidence of social communities in a spatially entiating the role of social networks and physical contact on parasite transmission structured network of a free-ranging shark species. Anim Behav 83(2):389–401. dynamics. Philos Trans R Soc Lond B Biol Sci 370:20140110. 5. Sah P, et al. (2016) Inferring social structure and its drivers from refuge use in the 20. Weng L, Menczer F, Ahn YY (2013) Virality prediction and community structure in desert tortoise, a relatively solitary species. Behav Ecol Sociobiol 70(8):1277–1289. social networks. Sci Rep 3:2522. 6. Salathe M, Jones JH (2010) Dynamics and control of diseases in networks with com- 21. Manlove KR, Cassirer EF, Cross PC, Plowright RK, Hudson PJ (2014) Costs and bene- munity structure. PLoS Comput Biol 6(4):1–11. fits of group living with disease: A case study of pneumonia in bighorn lambs (Ovis 7. Cross PC, Lloyd-Smith JO, Johnson PLF, Getz WM (2005) Duelling timescales of host canadensis). Proc Biol Sci 281(1797):20142331. movement and disease recovery determine invasion of disease in structured popula- 22. Adelman JS, Moyers SC, Farine DR, Hawley DM (2015) Feeder use predicts both acqui- tions. Ecol Lett 8(6):587–595. sition and transmission of a contagious pathogen in a North American songbird. Proc 8. Cross PC, Johnson PLF, Lloyd-Smith JO, Getz WM (2007) Utility of R0 as a predictor of Biol Sci 282(1815):20151429. disease invasion in structured populations. J R Soc Interface R Soc 4(13):315–324. 23. Bansal S, Grenfell BT, Meyers LA (2007) When individual behaviour matters: Homoge- 9. Lentz HHK, Selhorst T, Sokolov IM (2012) Spread of infectious diseases in directed and neous and network models in epidemiology. J R Soc Interface R Soc 4(16):879–91. modular metapopulation networks. Phys Rev E 85(6):1–9. 24. Krause J, James R, Franks DW, Croft DP (2014) Animal Social Networks (Oxford Univ 10. Nematzadeh A, Ferrara E, Flammini A, Ahn YY (2014) Optimal network modularity Press, Oxford). for information diffusion. Phys Rev Lett 113(8):088701. 25. Hamede RK, Bashford J, McCallum H, Jones M (2009) Contact networks in a wild Tas- 11. Newman MEJ (2006) Modularity and community structure in networks. Proc Natl Acad manian devil (Sarcophilus harrisii) population: Using social network analysis to reveal Sci USA 103(23):8577–8582. seasonal variability in social behaviour and its implications for transmission of devil 12. Good BH, de Montjoye YAA, Clauset A (2010) Performance of modularity maximiza- facial tumour disease. Ecol Lett 12(11):1147–57. tion in practical contexts. Phys Rev E 81(4):1–19. 26. Pinter-Wollman N (2015) Persistent variation in spatial behavior affects the structure 13. Smith JE, Kolowski JM, Graham KE, Dawes SE, Holekamp KE (2008) Social and eco- and function of interaction networks. Curr Zool 61(1):98–106. logical determinants of fission-fusion dynamics in the spotted hyaena. Anim Behav 27. Newman M (2003) Properties of highly clustered networks. Phys Rev E 68(2):26121. 76(3):619–636. 28. Bull CM, Godfrey SS, Gordon DM (2012) Social networks and the spread of Salmonella 14. Archie Ea, Moss CJ, Alberts SC (2006) The ties that bind: Genetic relatedness pre- in a sleepy lizard population. Mol Ecol 21(17):4386–4392. dicts the fission and fusion of social groups in wild African elephants. Proc Biol Sci 29. Holme P (2016) Temporal network structures controlling disease spreading. Phys Rev 273(1586):513–522. E 94(2):1–8. 15. Rimbach R, et al. (2014) Behavioral and physiological responses to fruit availability of 30. Blondel VD, Guillaume JL, Lambiotte R, Lefebvre E (2008) Fast unfolding of commu- spider monkeys ranging in a small forest fragment. Am J Primatol 76(11):1049–1061. nities in large networks. J Stat Mech Theor Exp 2008(10):P10008.

4170 | www.pnas.org/cgi/doi/10.1073/pnas.1613616114 Sah et al. Downloaded by guest on September 27, 2021