is sociality required for the evolution of …eerc.unsw.edu.au/ord/ord_garciaporta2012.pdfresearch...

doi: 10.1098/rstb.2011.0215, 1811-1828367 2012 Phil. Trans. R. Soc. B

Terry J. Ord and Joan Garcia-Porta in diverse taxonomic groups

hypothesescomplexity? Evidence weighed against alternative Is sociality required for the evolution of communicative

References

http://rstb.royalsocietypublishing.org/content/367/1597/1811.full.html#related-urls Article cited in:

http://rstb.royalsocietypublishing.org/content/367/1597/1811.full.html#ref-list-1

This article cites 101 articles, 26 of which can be accessed free

Subject collections

(497 articles)evolution (360 articles)ecology

(357 articles)behaviour Articles on similar topics can be found in the following collections

Email alerting service hereright-hand corner of the article or click Receive free email alerts when new articles cite this article - sign up in the box at the top

http://rstb.royalsocietypublishing.org/subscriptions go to: Phil. Trans. R. Soc. BTo subscribe to

on June 1, 2012rstb.royalsocietypublishing.orgDownloaded from

http://rstb.royalsocietypublishing.org/content/367/1597/1811.full.html#ref-list-1http://rstb.royalsocietypublishing.org/content/367/1597/1811.full.html#related-urlshttp://rstb.royalsocietypublishing.org/cgi/collection/behaviourhttp://rstb.royalsocietypublishing.org/cgi/collection/ecologyhttp://rstb.royalsocietypublishing.org/cgi/collection/evolutionhttp://rstb.royalsocietypublishing.org/cgi/alerts/ctalert?alertType=citedby&addAlert=cited_by&saveAlert=no&cited_by_criteria_resid=royptb;367/1597/1811&return_type=article&return_url=http://rstb.royalsocietypublishing.org/content/367/1597/1811.full.pdf?keytype=ref&ijkey=ang7ZkCzJbAzDh4http://rstb.royalsocietypublishing.org/subscriptionshttp://rstb.royalsocietypublishing.org/


Phil. Trans. R. Soc. B (2012) 367, 1811–1828

doi:10.1098/rstb.2011.0215

Research

* Autho

One concommun

Is sociality required for the evolution ofcommunicative complexity? Evidence

weighed against alternative hypothesesin diverse taxonomic groups

Terry J. Ord1,* and Joan Garcia-Porta2

1Evolution and Ecology Research Centre, and the School of Biological, Earth and Environmental Sciences,The University of New South Wales, Kensington, New South Wales 2052, Australia2Institute of Evolutionary Biology (CSIC-UPF), Passeig Marı́tim de la Barceloneta,

37-49, 08003 Barcelona, Spain

Complex social communication is expected to evolve whenever animals engage in many and variedsocial interactions; that is, sociality should promote communicative complexity. Yet, informal com-parisons among phylogenetically independent taxonomic groups seem to cast doubt on the putativerole of social factors in the evolution of complex communication. Here, we provide a formal test ofthe sociality hypothesis alongside alternative explanations for the evolution of communicative com-plexity. We compiled data documenting variations in signal complexity among closely related speciesfor several case study groups—ants, frogs, lizards and birds—and used new phylogenetic methods toinvestigate the factors underlying communication evolution. Social factors were only implicated inthe evolution of complex visual signals in lizards. Ecology, and to some degree allometry, were mostlikely explanations for complexity in the vocal signals of frogs (ecology) and birds (ecology and allom-etry). There was some evidence for adaptive evolution in the pheromone complexity of ants,although no compelling selection pressure was identified. For most taxa, phylogenetic nullmodels were consistently ranked above adaptive models and, for some taxa, signal complexityseems to have accumulated in species via incremental or random changes over long periods of evo-lutionary time. Becoming social presumably leads to the origin of social communication in animals,but its subsequent influence on the trajectory of signal evolution has been neither clear-cut norgeneral among taxonomic groups.

Keywords: animal communication; phylogenetic comparative methods; adaptation;sexual selection; natural selection

1. INTRODUCTIONComplex communication is classically linked to theevolution of complex animal societies [1]: as thenumber and context of social interactions increase,communication mediating those interactions tends tobecome increasingly elaborate. Indeed, it is difficultto think of any highly social animal that does not pos-sess a complex system of communication. Humans arean obvious example, as are many other primates andlong-lived mammals. Chimpanzees form complexhierarchies and alliances among troop members andrely on an elaborate array of vocal and visual signalsto mediate those relationships [2–5]. Elephants havesimilarly complex social interactions and use an exten-sive repertoire of social signals as an apparentconsequence [6–9]. But is it correct to say that social-ity is a necessary prerequisite for the evolution of

r for correspondence ([email protected]).

tribution of 13 to a Theme Issue ‘The social network andicative complexity in animals’.

1811

communicative complexity? Are less social speciesreally predisposed to basic forms of communicationmore than socially complex species?

These questions lie at the very foundation of howwe believe communication evolved in animals. If com-municative complexity and sociality are tightlycoupled, then both the origin and direction ofcommunication evolution has essentially been a by-product of factors driving the evolution of socialitymore generally. Alternatively, communication mighthave originated to mediate social interactions amongconspecifics, but its ultimate elaboration has beendriven by other factors independent of changes in soci-ality. Freeberg et al. [1] have discussed in detail thepossible non-social pressures that might produce com-municative complexity. These include environmentalfactors that influence signal fidelity, the need forreliable species recognition and neutral or non-adaptive evolutionary processes. Some of these factorshave empirical support (species recognition; [10,11]),whereas others have very little (neutral processes;reviewed by Freeberg et al. [1]). The challenge is to

This journal is q 2012 The Royal Society

mailto:[email protected]://rstb.royalsocietypublishing.org/

1812 T. J. Ord and J. Garcia-Porta Sociality and animal communication


test the sociality hypothesis alongside its alternativesand in a way that allows the weight of evidence foreach hypothesis to be directly compared.

One can try to make broad taxonomic comparisonsto argue that group-living primates, for example, havesignificantly more elaborate systems of communica-tion relative to claw-waving territorial crabs becauseprimates form complex societies, whereas crabs donot (by comparison). Yet there are many othernon-primate groups that do exhibit extraordinary com-plexity in their communication, and some researchershave even gone as far as to say have levels of signal com-plexity comparable to primates [12,13]. Consider theexquisite song and courtship dances of many birds[14], the elaborate headbobbing and colourful dewlapdisplays of anole lizards [11,15], the rippling colourand movement displays of cuttlefish [16,17] or the elab-orate foot thumping signals of ornate wolf spiders[18,19]. These are all examples of complex communi-cation—complex in repertoire, the number and typeof components making up a signal and, in some cases,the number of sensory modalities used for communi-cation (e.g. signals that are both auditory and visual).Yet, these animals do not exhibit the same level ofsociality seen in primate societies.

Such examples seem to weaken the putative linkbetween communicative complexity and sociality.However, broad comparisons among disparate groupslike these (primates versus crabs or birds) actually pro-vide little insight into the evolution of animalcommunication. The vast number of attributes thatvary among species at these broad phylogenetic scalesmake it virtually impossible to determine what factorsmight or might not account for taxonomic differencesin communication. More informative are comparisonsamong closely related species within broad taxonomicgroups. In almost any group that depends on socialcommunication, closely related species differ, to alesser or greater degree, in their social behaviour andcomplexity of communication. Furthermore, at thisfiner phylogenetic scale, the specific selection pressuresthat direct the evolution of communicative complexitybecome more apparent (reviewed by Ord [20]). Ratherthan conducting a literature review of the evidence forand against the sociality hypothesis, we chose to test thehypothesis directly alongside other potential causal fac-tors, using empirical data collected for closely relatedspecies.

To this end, we developed six datasets or ‘casestudies’ to represent a variety of taxonomic groups(frogs, birds, lizards and ants) and signal classes(acoustic, visual and chemical). The data for eachcase study documented variations among closelyrelated species in signal complexity, social behaviourand other factors predicted to influence the evolutionof communicative complexity (see §1a–d). We thenused new phylogenetic comparative approaches andassembled new phylogenies to evaluate alternative evo-lutionary models of how communicative complexitymight have evolved in each case study. The models cor-responded to one of the four hypotheses and were notmutually exclusive. We assembled findings across thecase studies to determine the generality of each hypoth-esis in describing the evolution of communicative

Phil. Trans. R. Soc. B (2012)

complexity over diverse taxa and signal classes. Asecondary goal of our study was to provide communi-cation biologists, who might not be familiar withrecent advances in phylogenetic techniques, with aheuristic example of how multiple hypotheses can beconsidered jointly, using methods that also inform onthe probable mode of evolution.

(a) Social drivers of signal complexityWhen the frequency or importance of social inter-actions increases, signals used to mediate thoseinteractions should become more complex for arange of reasons: to convey information more reliably,to manipulate social partners more effectively or toprovide relevant social cues in different contexts [1].For instance, as sexual selection intensifies, eitherbecause mates become choosy in what they find attract-ive or as competition for mates and territoriesincreases, signal repertoires are expanded or thedesign of signals are elaborated to help the signal-ler outcompete courtship and territorial rivals[11,21–25]. For species living in groups or otherwiseinteracting with a range of different social partners(e.g. mates, rivals, juveniles, adults, dominates andsubordinates), different signals or cues are oftenrequired for different contexts, leading to an increasein repertoire size or the complexity of existing signalsto convey multiple messages [26]. The socialityhypothesis predicts variation among species in thedesign or repertoire of signals whenever species differin the frequency or context of social interactions.

(b) Ecological influences on signal complexityThe range of factors that make up the ecology ofspecies is diverse and there are a number of possibleways in which ecology can direct the evolution ofcommunicative complexity. The influence of theenvironment on the transmission of animal signals iswell documented [27–30]. Background acoustic orvisual noise masks calls [31,32] and displays [33,34]and obstructions in the environment deflect and scat-ter sound waves [28] and obscure visual signals[35,36], as does the reduced visibility imposed bypoor light [37–39]. The strategies animals adopt toenhance signal fidelity in noisy, cluttered or dim habi-tats are varied. Background noise can limit the range offrequencies heard by birds [40] or the types of move-ments that can be seen by lizards [34], and thisrestricts the range of signal designs that can be readilydetected by receivers. Namely, difficult environmentalconditions limit signals to those that are simple indesign. There are, however, some instances wherenoisy environments can facilitate signal complexityby promoting the evolution of alert components.These are new components added to signals to attractthe attention of receivers, before the more informationrich portion of the signal is delivered [39,41]. Here,poor signal environments promote the evolution ofmore complex signals (i.e. increases in componentnumber through the addition of alerts and othercomponents to enhance signal detection).

Microhabitat or the site within the environmentfrom which animals communicate with one another

http://rstb.royalsocietypublishing.org/

Sociality and animal communication T. J. Ord and J. Garcia-Porta 1813


can increase or decrease the severity of environmentalconditions that affect signal efficiency [42]. Acoustic-ally communicating species living near streams aresusceptible to noise generated by flowing water inwhat otherwise might be a quiet macrohabitat (e.g. asheltered temperate forest). In the case of somefrogs, species near streams facilitate call detection byproducing calls at ultrasonic frequencies above thebroadband frequency of running water [43]. Callsgiven near the ground are also more susceptible todegradation through muffling and deflection [28].Likewise, the visibility of displays is enhanced fromperches above undergrowth and other low-lyingvisual obstructions [44]. On the one hand, the con-straints acting on signal efficiency in microhabitatsnear noise sources or physical obstructions limit therange of signal designs that can be detected by receiv-ers. On the other hand, animals can minimize maskingby adding alert or amplifier components to signals and,in the process, increase the complexity of their signals[39,41]. Whether environmental variables constrain orpromote signal complexity is unclear (but see [35]).Yet it is reasonable to expect environmental condi-tions to play some part in directing the evolution ofcommunicative complexity.

Finally, when animals frequently encounter hetero-specific congeners in the environment and are not indirect competition with those congeners for resources(e.g. mates), the need for accurate species recognitionbecomes important (reviewed by Ord et al. [45]). Thedesign of social signals often provides the best cues forascertaining species identity. The number of sympatricspecies an animal encounters should prompt increasesin signal complexity to facilitate recognition [10,11].That is, signal elaboration results in a unique,species-typical signal that can be easily distinguishedfrom the signals of sympatric congeners.

(c) The allometry of signal complexityBody size influences numerous features of an animal.Of special relevance to communication is the allometryof physical structures and the physiological mechan-isms that govern signal production. For example, thelength of the vocal tract is heavily dependent onbody size (larger animals have longer vocal tracts[46]). The vocal tract in turn determines the types ofsounds that can be produced by an animal [47,48].The allometry of the vocal tract should therefore leadto disparity in vocal signals among species that differin size. In addition, size-specific metabolic rate seemsto explain a large portion of the variance amongspecies in the rate, frequency and duration of acousticsignals in groups as diverse as insects and mammals:large species typically produce low-frequency calls oflong duration and at low rates [49]. A similar physio-logical mechanism has been implicated in theevolution of movement-based displays in lizards [50].Larger lizards have higher energetic costs associatedwith movement compared with smaller lizards, andthis has been suggested to constrain the number, dur-ation or type of movements that large lizards caninclude in displays [50]. However, in the case ofstatic visual signals, possessing a large body might


instead provide more surface area for the expressionof big or elaborate ornaments.

Taken together, the allometry of communicative com-plexity will depend on the modality and type of signalcharacteristic examined. Larger bodied acousticallycommunicating species will have vocalizations of longerduration than smaller bodied species [49]. Whethersize-specific metabolic rate, or body size more generally,influences other indices of vocal complexity such asrepertoire size or note number is unknown. In visuallycommunicating lizards, size-specific energetic costsshould limit the evolution of movement-based displaycomplexity [50], but larger body sizes should allow theevolution of large or more numerous ornaments becauseof increased surface area.

(d) Non-adaptive signal complexityDivergence in song complexity can occur ‘passively’among populations in wide-ranging species throughcultural or genetic drift (e.g. birds: [51,52]; mammals:[53]). This leads to the unusual hypothesis that sto-chastic processes can generate signal complexity inthe absence of selection (see also [54]). Genetic driftand neutral mutation (mutations that are neither dele-terious nor beneficial) may incidentally increase signalcomplexity over evolutionary time or following bouts ofrapid evolution (e.g. during speciation). If true, vari-ations in communicative complexity will tend to trackphylogeny (closely related species will tend to sharesimilar levels of signal complexity, whereas distantlyrelated species will not) and, in particular, match a pat-tern of stochastic evolution. Recent developments inphylogenetic comparative analyses allow the joint esti-mation of both phylogenetic inertia and stochasticityin evolutionary diversification [55]. This neutralityhypothesis will be novel to most communication biol-ogists, but it is an important null model missing frommost studies of signal adaptation.

2. MATERIAL AND METHODSWe identified recent studies in which the design orrepertoire of communication had been surveyed for alarge number of closely related species. Our criterionfor selecting case studies was dependent on several fac-tors. First, we needed to be fairly certain that we couldobtain adequate social, ecological or morphologicaldata for the species in question. Some of this infor-mation was reported in the original sources, or theauthors of those sources were willing to share unpub-lished data with us when contacted directly (seeAcknowledgements). In other cases, we used elec-tronic databases and other published sources tosupplement datasets. Second, to construct phylogeniesfor each case study, there had to be adequate and com-prehensive genetic markers for species in GenBank.Third, we wished to cover several diverse taxonomicgroups and obtain representatives of both the mostcommonly studied and least-studied forms of com-munication (in decreasing order of researchattention: acoustic signals, static visual ornaments,movement-based displays and chemical signals). Weexcluded primates and other mammals because thesesystems were either well represented in earlier tests of




the sociality hypothesis (e.g. [56–58]; reviewed inFreeberg et al. [1]) or were the focus of other studiesincluded in this theme issue [59–61]. Finally, insome cases, we selected two case studies for thesame taxonomic group (specifically birds and lizards),one that represented phylogenetically diverse speciesfrom multiple genera and families, while the secondconsisted of closely related species from the samegenus or phylogenetically adjacent genera. We referto these case studies as ‘distantly related’ and ‘closelyrelated’ examples, respectively. Our motivation herewas to assess the sensitivity of our estimates of phylo-genetic inertia and stochasticity to taxon sampling.

Details on the data and analyses used are describedin the following sections. All data and GenBank acces-sion numbers for DNA sequences used to assemblephylogenies have been deposited in files assigned tothis article in the public electronic database DryadDigital Repository (see Acknowledgements).

(a) Communication data and social indicatorsWe compiled data on signal and social characteristicsthat we believed were consistent with current definitionsof complexity [1]. Indices of signal complexity were: callamplitude modulation (frogs); call, song or display dur-ation (frogs, birds and lizards); song or syllablerepertoire size (birds); number of ornaments (lizards);number of separate components making up signals(lizards and ants; this includes colour dichromatism inlizards, measured as the number of colourful bodypatches exhibited by males but not seen in females[62]). Indices of sociality were: sexual size dimorphism(frogs and lizards; in these taxa, size dimorphism isbelieved to reflect the intensity of competition amongmales for mates and territories [24]); levels of extra-pair paternity (birds); mating system (in birds this wascoded as ‘monogamy’, ‘irregular polygyny’ and ‘regularpolygyny’; in ants, ‘no polygyny or polyandry’ versus‘polygyny or polyandry present’); and colony size(ants). All data on signal complexity were compileddirectly from published studies [25,62–67] except forfrogs (see below). Indices of sociality were obtainedfrom the same sources used for communication data(frogs [68]; lizards, distantly related [25]; birds [66]),unpublished data from the authors of these studies(ants—E. van Wilgenburg, M. R. E. Symonds & M. A.Elgar 2011, unpublished data) or compiled separatelyfrom other literature (lizards, closely related [69]).

To obtain data on the signals of frogs, we used oscil-lograms of species-typical frog calls to estimate callduration and amplitude modulation from a com-prehensive field guide on the anuran fauna of theKaiteur National Park in Guyana [68]. Oscillo-grams were digitally scanned and IMAGEJ v. 1.42q(W. Rasband 1997–2009, NIH) used to measure theduration of calls in seconds from the start of the firstnote to the end of the last note. Call amplitude modu-lation was estimated as the coefficient of variation(CV) of the peak sound pressure computed acrosspulses making up a call. That is, a call with manypulses varying in peak sound pressure had a higherestimated CV than a call with pulses that peaked atconsistent sound pressure levels.


(b) Ecological and morphological dataMost ecology data were obtained from sources report-ing communication data and included: macrohabitatfor the ‘distantly related’ lizards [25]; microhabitatfor both lizard case studies [63,64] as well as thefrogs [68]; environmental noise for the frogs [68];whether species were migratory for the ‘closely related’bird case study [65]; species geographical range for the‘distantly related’ lizard case study [62]; the number ofsympatric species for frogs [68], and two climatic vari-ables for ants [67]. Macrohabitat was generally codedas ‘open’ (e.g. grasslands, deserts and plains) or‘closed’ habitats (e.g. forests and woodlands). This isa biologically relevant categorization of habitat forcommunication because closed environments gener-ally reduce the range of detection for acoustic andvisual signals compared with open environments (see§1). Microhabitat in frogs was the average heightabove ground of call sites for each species, while inlizards it was whether species were arboreal or terres-trial. In frogs, species that were reported preferringenvironments near streams were assumed to experi-ence environmental noise from running water andcategorized as occupying ‘high noise’ environments,whereas all other species were categorized as occupy-ing ‘low noise’ environments. Whether species weremigratory or occupying large geographic ranges wasused as an index of the likely habitat heterogeneityexperienced by species, as well as the likelihood ofinteracting with heterospecifics. Habitat heterogeneitywas expected to influence the evolution of communi-cative complexity because heterogeneity will eithertruncate the set of signals that are detectable acrossdifferent habitats or because heterogeneity facilitatesthe evolution of new signal components or largerrepertoire sizes (see §1). Widely distributed speciescan also be expected to interact with more sympatricspecies compared to localized species, and sympatryis in turn predicted to promote the evolution ofsignal complexity (see §1). In frogs, the number ofsympatric species was estimated as the number ofco-occurring frog species reported at sites where aspecies was stated to occur. Climatic variablesexpected to influence the composition of cuticularhydrocarbons in ants are rainfall and temperature[67], which may limit the range of chemical com-ponents included in an olfactory signal.

We used a GIS approach to obtain information onthe ecology of species for two case studies for whichdata were not reported in original sources. For onelizard case study (communication data from Martins[63]), we used GIS distribution data to compute thegeographic range of species, the percentage range over-lap with congeneric heterospecifics (species of thesame genus; this index is hereafter referred to as ‘sym-patry: overlap’), the number of sympatric species(hereafter ‘sympatry: number’) and the level of habitatheterogeneity likely experienced by species. Specific-ally, geographical distribution data for 79 speciesof Sceloporus (88% of described species within thegenus) were obtained from the IUCN red list data-base (http://www.iucnredlist.org/technical-documents/spatial-data) and mapped using a cylindrical equalarea projection. The range for the species of interest

http://www.iucnredlist.org/technical-documents/spatial-datahttp://www.iucnredlist.org/technical-documents/spatial-datahttp://www.iucnredlist.org/technical-documents/spatial-datahttp://rstb.royalsocietypublishing.org/



was measured in square kilometres and estimated asthe area enclosed by the polygon corresponding toeach species distribution. We then calculated the per-centage area of a species range that was overlappedby at least one Sceloporus congener. We also estimatedthe number of sympatric species by tallying thenumber of overlapping species. Habitat heterogeneitywas computed by intersecting the ecoregion GISlayer described in Olson et al. ([70]; availableelectronically at http://www.worldwildlife.org/science/data/item1875.html) with the projected distributionof species. Initially, the percentage overlap of ecore-gions for a given species was grouped into four broadcategories—forest, shrubland, grassland and desert—but these were later merged into ‘forest’ and ‘non-forest’ environments (forest versus all others) becausethis dichotomous categorization was more biologicallyappropriate for testing our hypotheses (see openingparagraph in this section).

The same methods were used to obtain ecologicaldata for one bird case study (communication datafrom Soma & Garamszegi [66]). In this case, speciesdistributions were obtained from the BirdLife Inter-national and NatureServe database ([71]; http://www.birdlife.org/datazone/info/spcdownload3, accessed onAugust 19, 2011). Our projected distributions werebased on the known ‘breeding range’ if species weremigratory (given that song charactersitics were usedin mate choice and territory defence [66], activitiesspecific to the breeding season) or the ‘resident range’if species were non-migratory. Ecoregion data and ana-lyses were the same as those of the lizard exampledescribed earlier. However, we were unable to obtaindata on sympatric species for this bird case study, sosympatry was not evaluated. All GIS analyses wereperformed using ARCGIS v. 9.3 (ESRI, Redlands CA).

Data on body size were largely obtained fromstudies reporting communication data (the only excep-tion being the Soma & Garamszegi [66] case study; seeabove). Measures of size included snout-to-vent lengthfor frogs and lizards, and tarsus length in birds (tarsuslength is a common metric of size in birds—e.g.[72,73]—and is a useful measure because it reflectsboth the overall size and mass of the bird).

(c) PhylogeniesWe developed phylogenies for each case study usingthe two most comprehensive mitochondrial DNAmarkers deposited in GenBank [74]. The longestsequences available for a given species were selectedand aligned using the program MAFFT v. 6 ([75];http://www.ebi.ac.uk/Tools/msa/mafft/) with the fol-lowing parameter settings: gap penalty ¼ 1.53; gapextension ¼ 0.123; tree rebuilding number ¼ 100;maximum number of iterations ¼ 100; and fast Four-ier transforms (FFTS) ¼ local pair. When sequencedata for a given species were not available in GenBank,we used surrogate sequences from two species of thesame genus (when possible; in rare instances onlyone species of the same genus was represented inGenBank). Surrogate sequences were required onlyfor five of the 199 species included in our study.


For every case study, we generated two ultrametricphylogenies using BEAST v. 1.6.1 [76]. We used twotrees rather than a single phylogeny for each casestudy to incorporate alternative phylogenetic hypoth-eses into our analyses (results from comparativeanalyses are partly dependent on the phylogenyused). The first tree was based on species relationshipsinformed exclusively by the two mitochondrial DNAgenes downloaded from GenBank. The second treeused additional information from the most recentlypublished phylogenies for a given group to ‘constrain’the topology of the tree such that only branch lengthswere estimated in our phylogeny using the mitochon-drial DNA genes from GenBank. This lead to anoverall topology in the second tree that was congruentwith existing phylogenies (frogs: [77]; birds: [78–83];lizards: [84–88]; ants: [67]). We refer to these treesas ‘unconstrained’ and ‘constrained’ phylogenies,respectively.

Topology, node supports (in the unconstrainedtrees) and branch lengths (in both the unconstrainedand constrained trees) were inferred using theBayesian algorithms implemented in BEAST. A Yulebranching process with a uniform prior was used andan uncorrelated branch rate variation was modelledusing a lognormal distribution. For both genes, themodel of evolution was set to GTRGAMMAI. Themean global substitution rate was set to unity and pro-duced ultrametric trees with branch lengths expressedin units of substitutions per site. The analysis consistedof two to four independent Markov Chain MonteCarlo (MCMC) chains that ran for 20 000 000 gener-ations with parameters and trees sampled every 3000generations. Independent runs in all cases convergedon very similar posterior estimates and were combinedusing LOGCOMBINER v. 1.5.4 (included in the packageBEAST). In all runs, the first 10 per cent of gener-ations were considered to belong to the burn-inphase of the analysis and were excluded. The programTRACER v. 1.2 [89] was then used to confirm conver-gence and good mixing of the combined MCMCchains. Finally, summary trees were obtained withmean node heights computed using TREEANNOTATORv. 1.5.4 (in the package BEAST), with a posteriorprobability limit set to 0.5.

(d) AnalysisWe used model fitting within a phylogenetic frame-work to assess the extent to which different predictorvariables explained variation among species in com-municative complexity. We used the second-orderAkaike’s Information Criterion (AICc) to determinethe level of support for each model. AICc is a modifi-cation of AIC that corrects for sample size; as samplesize increases, AICc values converge on those of AIC[90]. As applied here, an AICc value reflects the like-lihood that a given model fits the observed variationin signal complexity among species, given the phylogen-etic relationships among those species. The modelwith the lowest AICc value is the model that best fitsthe data, although any model within two AICc unitsof this lowest value is by convention considered to fitthe data just as well [90]. We also computed model

http://www.worldwildlife.org/science/data/item1875.htmlhttp://www.worldwildlife.org/science/data/item1875.htmlhttp://www.worldwildlife.org/science/data/item1875.htmlhttp://www.birdlife.org/datazone/info/spcdownload3http://www.birdlife.org/datazone/info/spcdownload3http://www.birdlife.org/datazone/info/spcdownload3http://www.ebi.ac.uk/Tools/msa/mafft/http://www.ebi.ac.uk/Tools/msa/mafft/http://rstb.royalsocietypublishing.org/



weights, AICw, to provide a more intuitive indicationof the support of a given model relative to all modelsconsidered. See Burnham & Anderson [90] for detailson the calculation of AICc and AICw.

Model fitting and subsequent selection using infor-mation and likelihood theory is becoming increasinglycommon in evolutionary ecology studies (see [91] forreview). This is because it offers an enticing alternativeapproach to traditional p-value-driven statistical ana-lyses, with several statistical advantages that areparticularly amenable for biological study. First,there is often a range of biologically plausible variablesthat might account for the observed patterns we see inthe natural world. These variables might be alternativeindices associated with a specific hypothesis or corres-pond to different hypotheses in their own right. Modelfitting incorporates this notion of multiple potentialcausal factors as an explicit part of its analytic philos-ophy by allowing simultaneous evaluation of a numberof alternative models. Second, such multiple compari-sons are a menace for the interpretation of p-valuesbecause of the increased chance of falsely concludinga ‘significant’ relationship when many statistical com-parisons are performed (i.e. type II error ratesincrease with the number of statistical analyses per-formed on a dataset). While there are variouscorrections that can be applied to p-value thresholdsfor judging significance (Bonferroni, false discoveryrate), the problem is circumvented entirely usingmodel fitting methods such as AIC.

This ability to consider a range of different modelsof how evolution might have occurred in a group isespecially useful for phylogenetic comparative analysesin which there are typically several potential causal fac-tors. In our study, we had a number of predictorvariables associated with several hypotheses. Thesehypotheses were not mutually exclusive. While themodel fitting framework allows all possible multi-variate combinations of variables to be considered,we choose to focus on a relatively simple set ofmodels to facilitate interpretation and avoid generatinglarge unwieldy AICc tables. We therefore relied onmodel weights to identify which hypothesis, or combin-ation of hypotheses, most likely accounted for theevolution of signal complexity in a given case study.

Models were fitted individually to the communi-cation data using the phylogenetic comparativesoftware SLOUCH v. 1.1 [55] run in R v. 2.8.1 (RDevelopment Core Team). Thomas Hansen providesan introduction to his program and its analyses inaccessible non-technical language in the accompany-ing user manual. A more detailed description of hisapproach is given in Hansen et al. [55] (see also[92]). We elaborate here on some of the key aspectsof the program that are relevant for the interpretationour results.

SLOUCH has several features that make itespecially attractive for comparative study among thedaunting array of methods currently available. In par-ticular, rather than assume a particular mode ofevolution at the outset (a critical flaw of the popularindependent contrasts method ([93]; see discussionin Ord & Martins [94]), SLOUCH uses likelihoodto estimate the level of phylogenetic inertia and


stochasticity in the evolution process based on the dis-tribution of species data across the tips of the phylogenyand the nature of the phylogeny itself (e.g. its topology,length of branches). In the case of phylogenetic inertia,SLOUCH computes the phylogenetic half-life, t1/2, ofthe phenotype (here signal complexity), which is thetime the phenotype would likely take to evolve halfwaytowards an adaptive optimum. The value of t1/2 is adirect function of phylogenetic inertia: strong phyloge-netic inertia is reflected in large values of t1/2 and isconsistent with incremental phenotypic change overlong periods of evolutionary time, whereas weak phylo-genetic inertia is reflected in small values of t1/2 andbursts of phenotypic change over short periods of evol-utionary time. Physiological or morphologicalconstraints, genetic correlations or low mutation rateare some of the factors that might contribute to phylo-genetic inertia and large values of t1/2.

SLOUCH parametrizes t1/2 from zero to infinity, arange that includes a rapid and directed phenotypicchange in response to selection, to gradual phenotypicchange occurring via a process of Brownian motion(which may or may not be directed by selection). Inthe instance of adaptive evolution, the phenotypemight track a continuously shifting selection regime orbe maintained at an optimum phenotype through stabil-izing selection. This provides considerable versatilitybecause it allows for the possibility of non-adaptiveevolutionary change, adaptive evolutionary changethrough directional selection and adaptive evolutionarychange through stabilizing selection.

In the context of signal evolution, values of t1/2approaching infinity imply that signals have evolvedvia an incremental change culminated over longperiods of evolutionary time. In this scenario, speciestend to share similar levels of signal complexity as afunction of phylogenetic relatedness. Adaptive evo-lution may have occurred, but the process has beenslowed by strong phylogenetic inertia. Intermediatevalues of t1/2 imply that signal evolution has tended toproceed towards an optimal phenotype via Brownianmotion and, if reached, stabilizing selection has keptsignal designs at or near this optimum. Values of t1/2approaching zero suggest that signal designs amongspecies are effectively independent and have littlerelationship to phylogeny. In this situation, animal sig-nals have been free to vary adaptively and retain nosignature of evolutionary history.

There are other phylogenetic comparative methodsthat use likelihood to estimate parameters reflectingthe extent phenotypic evolution has been influencedby factors associated with phylogeny. For example,Martin & Hansen’s [95] phylogenetic generalizedleast squares model (PGLS) estimates a, or the rateof adaptation, which can be used to compute t1/2[55]. Other methods compute parameters such as K[96] or l [97] that estimate the extent phenotypic evo-lution has followed a Brownian motion process,commonly interpreted as the level of phylogeneticsignal in phenotypic traits. The utility of SLOUCHlies in its foundation on the Ornstein–Uhlenbeck pro-cess, which includes both Brownian motion and thepossibility of adaptive evolution towards an optimum.Furthermore, unlike other methods, SLOUCH




computes a separate estimate of stochasticity. Stochas-ticity can generate non-adaptive phenotypic variationsamong living species despite the presence of strongdirectional or stabilizing selection and contributes to arandom phenotypic change under Brownian motion.

SLOUCH estimates stochasticity, vy, over a rangefrom zero to infinity. Values approaching infinityimply that stochastic processes have been highly in-fluential in the course of phenotypic evolution.Stochasticity can be generated by factors such as gen-etic drift or secondary selection pressures acting onother traits that are genetically correlated or duringthe process of evolutionary change more generally.Values approaching zero imply that there have beenfew perturbations from an adaptive optimum orduring the process of evolutionary change generally.

There is always the danger of over interpreting t1/2and vy parameters and others like them in phylogeneticcomparative methods (K or l). It must be kept in mindthat these parameters are not direct measures of theevolutionary process. Rather they are statistical par-ameters measuring patterns in the data that areconsistent with phylogenetic inertia and stochasticityin phenotypic evolution. But these parameter valuescan reflect other non-biological factors, such as thelevel of measurement error in data or patchiness intaxon sampling. Practitioners of phylogenetic com-parative methods should consider carefully thewarnings of Revell et al. [98] and Freckleton [99].With a reasonable degree of caution, parameterspotentially reflecting phylogenetic inertia and sto-chasticity can still provide some insight into theevolutionary processes that might have lead to speciesvariation in signal complexity. To facilitate interpret-ation, we provide a measure of confidence in t1/2 andvy estimates by reporting the range of values withintwo support units of the best estimate [55]. As withAICc, values within this two-unit range are those thatare essentially equally well supported.

Finally, SLOUCH offers two methods for examin-ing phenotypic evolution. The first assumes that thephenotype has evolved towards discrete stationaryoptima (either a common optimum or several alterna-tive optima; e.g. a specific phenotype selected for andmaintained in a given environment via stabilizingselection). The method relies on ancestor state recon-structions of a categorical predictor variable onto thephylogeny (e.g. habitat type), which is then used toassess whether different phenotypes have beenfavoured in different regions of the phylogeny specifiedby that predictor variable. We used this approach to fitmodels with the following variables: open versusclosed macrohabitats, arboreal versus terrestrial micro-habitats, high-noise versus low-noise environments,migration and mating system. These categories werereconstructed onto the phylogeny using parsimony inthe program MESQUITE v. 2.74 [100]. Ideally, likeli-hood reconstructions should be used here, butlikelihood assigns a probability that a categorical vari-able is present in a given ancestor. To implement theoptimality analysis in SLOUCH, these probabilitieswould have to be manually assigned using some arbi-trary cut-off (e.g. an ancestor with a probabilitygreater than 50 per cent coded as having lived in an


open rather than closed habitat). To avoid this, we fol-lowed current convention and assigned ancestor statesas present or absent using parsimony [101,102]. Themethod of reconstruction does influence ancestorassignments. However, our preliminary analysesshowed that changes to the phylogeny had a greatereffect on ancestor reconstructions than the methodof assignment (parsimony versus likelihood). All ofour analyses were repeated on alternative phylogeniesfor each case study.

Reconstructions were imported into SLOUCH andused to fit models that assumed that reconstructedvariables (e.g. open versus closed habitats) haveselected for different adaptive optima in signal com-plexity (e.g. complex signals in open habitats, simplesignals in closed habitats).

The second method in SLOUCH is similar in therespect that it also assumes that the phenotype hasevolved towards an adaptive optimum, but differs inthe sense that this optimum is not stationary andvaries as a function of fluctuations in a continuous pre-dictor variable. The model fitted is essentially aregression of signal complexity on the predictor vari-able and was used for all continuous predictors.SLOUCH also provides output for an optimal andevolutionary regression. The optimal regressiondepicts the relationship between the predictor variableand signal complexity assuming phylogenetic inertiawas absent, whereas the evolutionary regression pro-vides the ‘observed’ relationship between thepredictor and signal complexity as a function of bothadaptation and the constraining force of phylogeneticinertia. We report the results from evolutionaryregressions only.

For all best supported models, we report phylogen-etic effect sizes (r-values) to provide an indication ofthe direction and magnitude of relationships inthe data.

3. RESULTSSupport for models applied to each case study arereported in tables 1–3. We provide a brief summaryof key findings below and in figures 1 and 2, andelaborate on the combined findings of the analyses inthe discussion (§4).

(a) Complexity in vocal communicationThere was little support for the role of sociality inthe evolution of complexity in vocal signals (mean+s.e. AICw ¼ 0.07+0.01). Instead, phylogenetic nullmodels and a range of different ecological models—the probability of encountering heterospecifics,migration/geographic range and habitat preference—were among the best-supported models (table 1).For example, the phylogeny of frogs, their probableencounter rate with heterospecifics and call site pos-ition above ground fit variations in call complexitywell. Plots of sympatry and call modulation revealedseveral prominent outliers (figure 1a), but the overalltrend was consistent with the species recognitionhypothesis. Call site was positively correlated to callduration and, to some extent, levels of call modulation,implying that calling high above the ground has


Tab

le1.

Fac

tors

infl

uen

cin

gth

eev

olu

tion

of

com

ple

xit

yin

voca

lco

mm

un

icat

ion

.S

SD

,se

xu

al

size

dim

orp

his

m.

case

-stu

dy,

sign

al

vari

able

mod

el,

ran

k

con

stra

ined

tree

un

con

stra

ined

tree

AIC

cA

ICw

rt 1

/2(s

up

port

regio

n)

vy

(su

pp

ort

regio

n)

AIC

cA

ICw

rt 1

/2(s

upport

regio

n)

vy

(su

pport

regio

n)

frog

s,n¼

32

spec

ies

call

am

plitu

de

mod

ula

tion

1.

phylo

gen

y:

nu

ll292.1

0.3

7n

.a.

0(0

–10)

440

(260

–760)

290.5

0.3

7n

.a.

20

(0–

140)

440

(300

–1)

2.

ecolo

gy:

sym

pat

ry294.4

0.1

2�

0.1

6a

1(0

–10)

420

(260

–720)

292.9

0.1

13.

ecolo

gy:

mic

rohabit

atb

294.2

0.1

3292.5

0.1

42

0.1

420

(0–

160)

430

(280

–1)

4.

ecolo

gy:

nois

ec

294.2

0.1

3292.8

0.1

25.

soci

al:

SS

D294.5

0.1

1293.1

0.1

06.

allom

etry

:bod

ysi

ze294.7

0.1

0292.6

0.1

37.

ecolo

gy:

mac

rohabit

at296.7

0.0

4294.8

0.0

4

call

du

rati

on

1.

ecolo

gy:

mic

rohabit

atb

211.8

20.8

10.4

5900

(170

–1)

30

(0–

70)

57.1

0.1

2

2.

phylo

gen

y:

nu

ll2

7.2

00.0

854.6

0.4

2n

.a.

460

(870

–1)

10

(0–

20)

3.

ecolo

gy:

sym

pat

ry2

5.1

60.0

357.3

0.1

14.

allom

etry

:bod

ysi

ze2

5.2

00.0

357.2

0.1

15.

ecolo

gy:

nois

ec

24.6

30.0

257.3

0.1

16.

soci

al:

SS

D2

4.5

90.0

257.3

0.1

1

7.

ecolo

gy:

mac

rohabit

at2

3.5

30.0

160.0

0.0

3

bird

s(d

ista

ntly

rela

ted),

n¼

23

spec

ies

syllable

reper

toir

esi

ze1.

allom

etry

:bod

ysi

ze46.2

0.3

50.3

575

(25

–1)

25

(0–

700)

47.4

0.3

40.3

4500

(25

–1)

175

(0–

700)

2.

phylo

gen

y:

nu

ll46.3

0.3

3n

.a.

300

(25

–1)

125

(0–

825)

47.4

0.3

4n

.a.

175

(25

–1)

75

(0–

825)

3.

ecolo

gy:

bre

edin

gra

nge

48.6

0.1

149.7

0.1

1

4.

ecolo

gy:

mac

rohabit

at48.7

0.1

050.0

0.0

95.

soci

al:

extr

a-p

air

pat

ern

ity

48.9

0.0

949.9

0.1

06.

soci

al:

mat

ing

syst

em52.1

0.0

253.0

0.0

2

son

gre

per

toir

esi

ze1.

phylo

gen

y:

nu

ll55.0

0.4

6n

.a.

975

(0–1)

600

(0–1)

56.3

0.4

6n

.a.

975

(0–1)

575

(0–1)

2.

ecolo

gy:

bre

edin

gra

nge

57.0

0.1

70.2

1950

(0–1)

550

(0–1)

58.3

0.1

70.2

0975

(0–1)

575

(0–1)

3.

allom

etry

:bod

ysi

ze57.9

0.1

159.2

0.1

14.

soci

al:

extr

a-p

air

pat

ern

ity

58.0

0.1

059.2

0.1

15.

ecolo

gy:

mac

rohabit

at58.0

0.1

059.1

0.1

16.

soci

al:

mat

ing

syst

em59.2

0.0

661.3

0.0

4

bird

s(c

lose

lyre

late

d),

n¼

23

spec

ies

syllable

reper

toir

e1.

ecolo

gy:

mig

rati

on

241.4

0.4

2�

0.5

2d

1(6

30

–1)

10

(0–

20)

237.5

0.1

32.

phylo

gen

y:

nu

ll2

41.1

0.3

6n

.a.

940

(500

–1)

10

(0–

30)

240.5

0.5

7n

.a.

910

(480

–1)

10

(0–

30)

3.

ecolo

gy:

mac

rohabit

at2

38.7

0.1

12

37.8

0.1

54.

allom

etry

:bod

ysi

ze2

38.7

0.1

12

38.0

0.1

6

syllable

du

rati

on

1.

phylo

gen

y:

nu

ll2

64.1

0.6

6n

.a.

1(8

30

–1)

10

(0–

20)

264.2

0.6

6n

.a.

1(8

30

–1)

10

(0–

20)

2.

ecolo

gy:

mac

rohabit

at2

61.2

0.1

52

61.3

0.1

63.

allom

etry

:bod

ysi

ze2

61.2

0.1

52

61.2

0.1

54.

ecolo

gy:

mig

rati

on

257.9

0.0

32

58.1

0.0

3

(Con

tinued

.)





Tab

le1.

(Con

tinued

.)

case

-stu

dy,

sign

al

vari

able

mod

el,

ran

k

con

stra

ined

tree

un

con

stra

ined

tree

AIC

cA

ICw

rt 1

/2(s

up

port

regio

n)

vy

(su

pp

ort

regio

n)

AIC

cA

ICw

rt 1

/2(s

upport

regio

n)

vy

(su

pport

regio

n)

son

gd

ura

tion

1.

phylo

gen

y:

nu

ll2

57.8

0.6

4n

.a.

1(7

80

–1)

10

(0–

20)

252.2

0.1

12.

allom

etry

:bod

ysi

ze2

55.2

0.2

72

55.0

0.4

30.2

81

(780

–1)

10

(0–

20)

3.

ecolo

gy:

mac

rohabit

at2

54.9

0.1

52

54.6

0.3

50.0

51

(780

–1)

10

(0–

20)

4.

ecolo

gy:

mig

rati

on

252.1

0.0

62

52.2

0.1

1

aT

her

eare

seve

ral

pro

min

ent

ou

tlie

rsm

akin

gd

irec

tin

terp

reta

tion

of

this

coef

fici

ent

mis

lead

ing

(figu

re1a).

bM

icro

habit

atw

as

defi

ned

as

the

hei

ght

(m)

of

call

site

sabove

gro

un

d.

cT

he

leve

lof

bac

kgro

un

dn

ois

epote

nti

ally

exper

ien

ced

by

asp

ecie

sw

as

det

erm

ined

by

whet

her

spec

ies

wer

ere

port

edto

call

nea

rst

ream

s.dT

he

stan

dard

erro

rsof

para

met

erva

lues

for

mig

rati

on

wer

eve

ryla

rge

makin

gbio

logic

al

inte

rpre

tati

on

dif

ficu

lt.




facilitated the evolution of more complex calls. Therewas some support for the allometry hypothesis inbirds (e.g. body size correlated positively with syllablerepertoire size; figure 1b).

There were high levels of possible phylogenetic iner-tia in most vocal signals. It was particularly strikingthat the estimates of phylogenetic inertia (implyingevolution via Brownian motion) were consistentacross measures of signal complexity and for data-sets in which species were sampled quite differently(distantly related versus closely related). Figure 2a,billustrates repertoire size in birds as it relates to phyl-ogeny. Both examples recorded high phylogeneticinertia, but differed in estimates of stochasticity: sto-chasticity in repertoire evolution was low for theclosely related species (figure 2a), but high among dis-tantly related species (figure 2b). This difference instochasticity probably reflects differences in taxonsampling between the two case studies.

(b) Complexity in visual communicationVisual communication in lizards provided the onlycompelling support for models of sociality (mean+s.e. AICw ¼ 0.44+0.13). Effect sizes showed thatsexual size dimorphism, a proxy for the intensity ofmale–male competition within species, was positivelycorrelated with the number of ornaments (e.g. horns,spines, tail crests; figure 1c), colour dichromatismand headbob display duration (table 2). In the caseof colour dichromatism, our analyses were focusedon colour signals occurring on two separate bodyregions because previous studies have shown that theselection pressures acting on dorsal coloration(‘exposed’; these signals are visible to aerial predators)have been different from those acting on ventral color-ation (‘concealed’; these signals are only exposedduring headbob displays to territorial rivals [25]). Phyl-ogeny and whether species were arboreal or terrestrialwere other models that received good support. On thelatter, consistent with call evolution in frogs, headbobdisplay evolution in lizards seems to have beenfacilitated by an arboreal lifestyle (figure 1d).

Visual signal evolution in lizards was associated withestimates of phylogenetic inertia close to zero. Ouranalyses also suggested very little stochasticity in evo-lutionary diversification. Taken together, visual signalshave potentially been free to respond quite rapidlyto selection.

(c) Complexity in chemical communicationWe found little support for the role of sociality in theevolution of signal complexity in ants (mean+ s.e.AICw ¼ 0.14+0.003). The ecological models fairedno better (table 3). Phylogeny appeared to be the onlymodel that fit the data at all and even then the level ofsupport was only marginally better than the othermodels considered (table 3). Almost no phylogeneticinertia or stochasticity was detected in the data. Thatis, evolutionary changes in the number of cuticularhydrocarbons appear to have occurred extremelyrapidly and not stochastically. This implies that thesesignals could have been targets for selection. Whatselection pressure(s) this might have been is unknown.


Tab

le2.

Fac

tors

infl

uen

cin

gth

eev

olu

tion

of

com

ple

xit

yin

vis

ual

com

mu

nic

atio

n.

case

-stu

dy,

sign

al

vari

able

mod

el,

ran

k

con

stra

ined

tree

un

con

stra

ined

tree

AIC

cA

ICw

rt 1

/2(s

up

port

regio

n)

vy

(su

pp

ort

regio

n)

AIC

cA

ICw

rt 1

/2(s

upport

regio

n)

vy

(su

pport

regio

n)

lizard

s(d

ista

ntly

rela

ted),

n¼

55

–59

spec

ies

no.

orn

am

ents

1.

soci

al:

SS

D210.7

1.0

00.5

840

(0–

450)

100

(20

–1)

208.8

1.0

00.5

860

(0–

400)

140

(20

–1)

2.

allom

etry

:bod

ysi

ze230.7

0228.6

03.

phylo

gen

y:

nu

ll234.4

0231.9

04.

ecolo

gy:

mac

rohabit

at236.5

0233.8

0

5.

ecolo

gy:

ran

ge

236.6

0221.1

0

colo

ur

dic

hro

mat

ism

,ex

pose

d1.

phylo

gen

y:

nu

ll161.8

0.3

7n

.a.

125

(0–

800)

225

(25

–1)

150.6

0.3

9n

.a.

50

(0–1)

75

(25

–1)

2.

soci

al:

SS

D162.6

0.2

40.1

5100

(0–

825)

175

(25

–1)

152.2

0.1

70.1

1175

(0–1)

250

(25

–1)

3.

allom

etry

:bod

ysi

ze163.8

0.1

30.0

7277

(0–

800)

225

(25

–1)

152.6

0.1

40.0

750

(0–1)

75

(25

–1)

4.

ecolo

gy:

mac

rohabit

at163.9

0.1

3152.7

0.1

4

5.

ecolo

gy:

ran

ge

163.9

0.1

3152.4

0.1

650

(0–1)

75

(25

–1)

colo

ur

dic

hro

mat

ism

,co

nce

ale

d1.

soci

al:

SS

D172.2

0.7

90.3

125

(0–

800)

50

(25

–1)

169.7

0.7

60.3

125

(0–

800)

50

(25

–1)

2.

phylo

gen

y:

nu

ll176.7

0.0

8174.0

0.0

93.

allom

etry

:bod

ysi

ze177.1

0.0

7174.1

0.0

84.

ecolo

gy:

ran

ge

178.8

0.0

3175.8

0.0

4

5.

ecolo

gy:

mac

rohabit

at178.9

0.0

3176.2

0.0

3

lizard

s(c

lose

lyre

late

d),

n¼

22

spec

ies

bob

no.

1.

ecolo

gy:

mic

rohabit

ata

125.3

0.6

70.4

90

(0–

10)

10

(0–

30)

125.3

0.6

90.4

90

(0–

10)

10

(0–

30)

2.

ecolo

gy:

sym

pat

ry,

ove

rlap

128.5

0.1

4128.5

0.1

4

3.

phylo

gen

y:

nu

ll129.3

0.0

9129.3

0.0

9

4.

soci

al:

SS

D131.2

0.0

4132.3

0.0

25.

ecolo

gy:

sym

pat

ry,

no.

131.2

0.0

4132.4

0.0

26.

allom

etry

:bod

ysi

ze132.4

0.0

2132.4

0.0

27.

ecolo

gy:

mac

rohabit

at134.1

0.0

1132.3

0.0

2

dis

pla

yd

ura

tion

1.

phylo

gen

y:

nu

ll132.7

0.3

5n

.a.

0(0

–10)

20

(5–

35)

132.7

0.3

5n

.a.

0(0

–50)

20

(0–1)

2.

soci

al:

SS

D134.1

0.1

80.2

91

(0–

50)

20

(0–1)

134.1

0.1

70.2

90

(0–

70)

20

(0–1)

3.

allom

etry

:bod

ysi

ze134.8

0.1

2134.8

0.1

24.

ecolo

gy:

sym

pat

ry,

ove

rlap

135.0

0.1

1135.0

0.1

1

5.

ecolo

gy:

mic

rohabit

ata

135.2

0.1

0135.1

0.1

0

6.

ecolo

gy:

sym

pat

ry,

no.

135.6

0.0

8135.4

0.0

97.

ecolo

gy:

mac

rohabit

at136.7

0.0

5136.4

0.0

5

aM

icro

habit

atw

as

defi

ned

as

arb

ore

al

or

terr

estr

ial

(that

is,

typic

ally

dis

pla

yin

gfr

om

tree

sor

from

the

gro

un

d).





0

10

20

30

40

50

60

70

80

90

–1 5 11 17 23 29 35

call

ampl

itude

mod

ulat

ion

(CV

of

call

ampl

itude

)

no. sympatric species

(a) (b)

(c) (d)

0

2

4

6

8

10

12

arboreal terrestrial

1

10

100

1000

10 15 20 25 30 35tarsus length (mm)

sylla

ble

repe

rtoi

re s

ize

–1

0

1

2

3

4

5

6

0.9 1.0 1.1 1.2 1.3 1.4sexual size dimorphism (SVLm/SVLf)

no. o

rnam

ents

microhabitat

Figure 1. Factors predicting variations in signal complexity in (a) frogs, (b) distantly related birds, (c) distantly related agamidlizards and (d) closely related Sceloporus lizards. The arrows in (a) highlight outliers not included in the computation of thetrend line depicted in the plot.

Table 3. Factors influencing the evolution of complexity in chemical communication. CHCs, cuticular hydrocarbons.

case-study,signalvariable model, rank

constrained tree unconstrained tree

AICc AICw r

t1/2(supportregion)

vy(supportregion) AICc AICw r

t1/2(supportregion)

vy(supportregion)

ants, n ¼ 40 speciesno. different

CHCs

1. phylogeny:

null

298.9 0.46 n.a. 0 (0–10) 90 (50–

150)

298.9 0.46 n.a. 0 (0–10) 90 (50–

150)2. social:

colony size301.2 0.14 301.2 0.14

3. ecology:

rainfall

301.3 0.14 301.3 0.14

4. social:matingsystem

301.4 0.13 301.4 0.13

5. ecology:

temperature

301.4 0.13 301.4 0.13





(a) (c)

(b)

1 10syllable repertoire size

0 1 10 100 1000song repertoire size

0 20 40 60no. cuticular hydrocarbons

Figure 2. The phylogeny of signal complexity in (a) closely related birds, (b) distantly related birds and (c) ants. Strong phy-logenetic inertia was estimated for the repertoires of both bird groups, while virtually no phylogenetic inertia was found for the

pheromones of ants.



Variations in signal complexity across the ant phylogenyis depicted in figure 2c.

4. DISCUSSIONThe goal of our study was twofold: (i) to evaluate therelative support of several hypotheses for the evolutionof communicative complexity, and (ii) to illustrate hownew phylogenetic approaches can be incorporated intothe study of animal communication more generally.With this second goal in mind, we briefly discusssome practical points first to provide readers who arenot familiar with the techniques used with a better per-spective on how to interpret our findings, beforeelaborating on the broader implications of our study.

An important point to remember for any statisticalanalysis is that the validity of results is contingent on


the extent to which the data accurately reflect the bio-logical variables being investigated. For example, theparticular index of signal complexity we used for agiven taxonomic group may or may not be functionallyrelevant for the species in question. Communicationsystems are also often complex in ways that are noteasily quantified by a single metric. For example,males might be the predominant signallers in onespecies, whereas both sexes might rely heavily onsocial communication in another species. In thisinstance, it is not immediately obvious whether reper-toire size (for example) should be summed across thesexes to provide a common metric for both species,or whether each should be sex-evaluated separately.Similar difficulties can exist for metrics of sociality.As an example, the evolution of sexual size dimorphismcan reflect other factors aside from sexual selection.


Table 4. A summary of the relative support for factors expected to influence the evolution of communicative complexity.

Shown are median ranks of the highest ranked model for a given hypothesis for each case study based on the results from theconstrained trees only or both the constrained and unconstrained trees. Calculations used the actual ranks of models whenthose models were ranked within two AICc units of the best-supported model or an assigned rank of 100 if models felloutside this region. Subsequent median ranks of 100 were then classified as ‘unranked’.

hypothesis

signal modality

acoustic visual chemical all modalities

constrained trees only

phylogeny 1 1 1 1ecology 1 1-unranked unranked 1–2social unranked 1–2 unranked unrankedallometry unranked 3-unranked not tested unranked

constrained or unconstrained trees

phylogeny 1 1 1 1ecology 1 2 unranked 1–2allometry 1 3-unranked not tested 3social unranked 1–2 unranked unranked



The sexes might appear monomorphic despite malesize being the target of sexual selection because natur-al selection on females has lead to the evolution of largefemale body size for enhanced fecundity [103]. Alter-natively, the presence of size dimorphism mightreflect sex-specific divergence in the ecologicalresources exploited (e.g. differences in the size of preyeaten by the sexes [104]). In comparative analyses,errors in the phylogeny can influence findings as well,as can the assumed mode by which evolution has pro-ceeded. To circumvent these latter issues, we relied onalternative phylogenies and a comparative method thatassessed the fit of a range of parameter values designedto assay varying levels of phylogenetic inertia andstochasticity in the evolution process.

The consequence of error in either the phenotypicdata or the phylogeny is the generation of noise in ana-lyses. Data noise makes it difficult to detectrelationships of small effect that might otherwiseexist in nature (i.e. enhanced type I statistical error).For this reason, it is hard to conclusively reject ahypothesis if it does not gain compelling support ina comparative analysis. This is exacerbated whendata are complied from a range of different sources(noise in the data can increase because sourcesdefine variables differently or use different methodsfor measuring variables). By the same token, broadtrends that are in fact revealed in comparative analyseswill probably reflect major evolutionary phenomenabecause only factors leading to strong biological effectswill tend to be detected. In this respect, phylogeneticcomparative analyses can offer conservative supportfor a hypothesis. This support can then be used to jus-tify a more refined comparative analysis in which directmeasures or more accurate data are collected by thecomparative biologists themselves (e.g. see exploratoryliterature-based analysis of Ord & Martins [11] andsubsequent follow-up field study by Ord et al. [105])or focused experimental research that confirmscausal links between a putative selective force and itsadaptive outcome [39,106].

Our study is therefore an exploratory analysis thatidentifies potentially productive avenues for future


research. Our findings are not meant to provide defini-tive conclusions on the specific evolutionary causes ofsignal complexity in the groups studied (although sev-eral strong candidates are highlighted (figure 1a–d).Rather the purpose of our study was to use thesediverse groups as case studies to offer broader insightinto the evolution of communicative complexity gener-ally. On this front, we found that sociality – based onthe metrics we were able to compile from the literatureand electronic databases – was not as influential in theevolution of signal complexity as we had anticipated.Indeed, it appears to have only been an importantfactor in the evolution of signal complexity in lizards(table 2 and figure 1c; this is consistent with earliercomparative analyses [24,25]).

Table 4 provides an overview of the median modelrank associated with each of the four hypotheses. Itshould be noted that the table weighs evidence fromeach of the case studies equally, and the communi-cation systems and ecology of these groups do differin a number of potentially important ways (e.g. duetingsongbirds versus chorusing male frogs; colonially livingants versus territorial lizards). However, if a hypothesisis a general explanation for the evolution of communi-cative complexity, then it should be largelyindependent of the social or ecological peculiarities ofa given species or taxonomic group. That is, supportfor a given hypothesis should be apparent acrossbroad taxonomic groups rather than exclusive toselect species. Table 4 shows that the phylogeneticnull model, not social factors, was most often thebest-supported model, followed closely by modelsreflecting ecology. In some instances, phylogeneticinertia and stochastic processes can explain variationsin signal complexity among species. Signal complexitytherefore has the potential to accumulate in lineages vianeutral or non-adaptive processes (e.g. syllable dur-ation in birds; table 1). When evidence of adaptationwas found, it was more often associated with variationsin ecological factors and (to a lesser extent) allometrythan it was with social pressures. For example, signalcomplexity increased in frogs as a possible function ofthe number of sympatric species encountered (table 1


Table 5. A summary of phylogenetic patterns associated with estimates of communicative complexity. CHCs, cuticular

hydrocarbons.

case-study, signal variable Nspecies phylogenetic inertia (t1/2) stochasticity (vy)

vocal signalsfrogs

call amplitude modulation 32 low highcall duration 32 high low

birds (distantly related)syllable repertoire size 23 moderate moderate

song repertoire size 23 high highbirds (closely related)

syllable repertoire 23 high lowsyllable duration 23 high low

song duration 23 high low

visual signalslizards (distantly related)

number of ornaments 59 low low–moderatecolour dichromatism, exposed 55 low–moderate low–moderatecolour dichromatism, concealed 55 low low

lizards (closely related)bob number 22 virtually zero lowdisplay duration 22 virtually zero low

chemical signals

antsno. different CHCs 40 virtually zero low



and figure 1a) or the height of call sites above theground (table 1). In lizards, the switch from a terrestriallifestyle to being arboreal appears to have facilitated theevolution of more elaborate territorial displays (table 2and figure 1d; see also [63]). The difficulty in interpret-ing the role of ecology here is whether variables such ascall or display site drive or constrain signal complexity.On the one hand, the range over which signals mustremain effective increases with perch elevation, leadingto potential directional selection on signals for longerduration or more components to facilitate detectionby distant receivers. On the other hand, the environ-mental constraint leading to simple signals whencommunication is conducted close to the ground isno longer present or reduced for species signallingfrom elevated perches, opening the door for more com-plex signals to evolve via other factors. Semantically thedistinction is subtle, but biologically it is importantfor inferring causality. But again, causality can onlyreally be confirmed by means of empirical study andexperimental manipulation [106].

That said, empirical and experimental studieswithin species alone do not adequately identify theselection pressures that have directed signal evolution.Experimental studies can demonstrate current utility,but they cannot reveal the evolutionary history of com-munication, which in itself can have importantconsequences on how species adapt to contemporaryselection pressures [107]. Consider the phylogeny ofsignal complexity in the case studies we examined.Table 5 illustrates remarkable consistency withinsignal classes in possible levels of phylogenetic inertia.If these estimates are accurate, they imply that theevolution of complexity in vocal signals exhibits fargreater phylogenetic inertia, and subsequently lowerrates of adaptation, than other signal modalities.


This could reflect major physiological or metabolicconstraints on the production of auditory signals[49,108,109]. We also found evidence for possibleselection on the pheromone signals of ants, eventhough none of our selection models obtained anycompelling support. The evolution of pheromonecomplexity appears to have been non-random andnot the product of the gradual accumulation of cuticu-lar hydrocarbons over long periods of evolutionarytime (indicated by low values of stochasticity in theevolution process and low values of phylogenetichalf-life, respectively). We reiterate that the interpret-ation of statistical parameters reflecting phylogeneticpatterns need to be made with caution. With this inmind, our results are consistent with the notion thatan unknown selection pressure has promoted rapid,predictable changes in pheromone complexity inants. However, these results may also reflect that thehydrocarbons assayed have little functional relevancefor communication or fitness generally.

Future research will be needed to clarify the signifi-cance of our findings in relation to the specific casestudies examined. But the general outcome of ourinvestigation is that sociality is not always requiredfor the evolution of communicative complexity. Or atleast, communicative complexity is the product of anintricate evolutionary process that cannot be distilledto a single factor. The complexity of form in the wayanimals communicate with one another fascinatesbiologists and amateur naturalists alike. On an infor-mal level, communicative complexity implies richnessin the social lives of animals. This sophistication insocial behaviour probably enticed many of us into acareer of studying animal communication in the firstplace. Indeed, in some regards, the study of communi-cation in non-avian and non-primate species might be




under-represented in the literature because of the per-ceived notion that such systems lack a degree ofsociality and are therefore less attractive systems forthe study of communication. Yet the generality ofany hypothesis of why and how animal communicationevolved is reliant on testing hypotheses on diverse taxa.We conducted such a study here and found compellingevidence—ironically, it seems—only for social factorsin the evolution of visual signals in lizards. Whethersociality is a prerequisite for the evolution of commu-nicative complexity in other systems awaits furtherinvestigation. Our results suggest that it may not be(see [56–58] for positive tests in mammals).

We thank Matthew Symonds, Carlos Botero and LászlóGaramszegi for generously providing us access to theiroriginal data, Victor Soria-Carrasco for his helpful advicein GIS matters, and Todd Freeberg and two anonymousreviewers for comments on a previous version of this paper.J.G.P. was supported by a pre-doctoral research fellowship(JAE) from the CSIC. All data used in this study havebeen archived in the Dryad Digital Repository(doi:10.5061/dryad.nf7697j6).

REFERENCES1 Freeberg, T. M., Dunbar, R. I. M. & Ord, T. J. 2012

Social complexity as a proximate and ultimate factorin communicative complexity. Phil. Trans. R. Soc. B367, 1785–1801. (doi:10.1098/rstb.2011.0213)

2 Pollick, A. S. & de Waal, F. B. M. 2007 Ape gesturesand language evolution. Proc. Natl Acad. Sci. USA104, 8184–8189. (doi:10.1073/pnas.0702624104)

3 Vick, S.-J, Waller, B. M., Parr, L. A., Smith Pasqualini,

M. C. & Bard, K. A. 2007 A cross-species comparisonof facial morphology and movement in humans andchimpanzees using the facial action coding system(FACS). J. Nonverb. Behav. 31, 1–10. (doi:10.1007/s10919-006-0017-z)

4 Arbib, M. A., Liebal, K. & & Pika, S. 2008 Primatevocalization, gesture, and the evolution of humanlanguage. Curr. Anthropol. 49, 1053–1076. (doi:10.1086/593015)

5 Dunbar, R. I. M. 2012 Bridging the bonding gap: the

transition from primates to humans. Phil.Trans. R. Soc. B 367, 1837–1846. (doi:10.1098/rstb.2011.0217)

6 McComb, K., Reby, D., Baker, L., Moss, C. J. &

Sayialel, S. 2003 Long-distance communication ofacoustic cues to social identity in African elephants.Anim. Behav. 65, 317–329. (doi:10.1006/anbe.2003.2047)

7 Soltis, J., Leong, K. & Savage, A. 2005 African elephant

vocal communication. I. Antiphonal calling behaviouramong affiliated females. Anim. Behav. 70, 579–587.(doi:10.1016/j.anbehav.2004.11.015)

8 Soltis, J., Leong, K. & Savage, A. 2005 African elephantvocal communication. II. Rumble variation reflects the

individual identity and emotional state of callers.Anim. Behav. 70, 589–599. (doi:10.1016/j.anbehav.2004.11.016)

9 Meyer, J. M., Goodwin, T. E. & Schulte, B. A. 2008Intrasexual chemical communication and social

responses of captive female African elephants, Loxodontaafricana. Anim. Behav. 76, 163–174. (doi:10.1016/j.anbehav.2007.12.019)

10 Figuerola, J. & Green, A. J. 2000 The evolution of

sexual dimorphism in relation to mating patterns,cavity nesting, insularity and sympatry in the


Anseriformes. Funct. Ecol. 14, 701–710. (doi:10.1046/j.1365-2435.2000.00474.x)

11 Ord, T. J. & Martins, E. P. 2006 Tracing the origins of

signal diversity in anole lizards: phylogeneticapproaches to inferring the evolution of complex behav-iour. Anim. Behav. 71, 1411–1429. (doi:10.1016/j.anbehav.2005.12.003)

12 Hailman, J. & Ficken, M. 1986 Combinatorial animal

communication with computable syntax: chick-a-deecalling qualifies as ‘language’ by structural linguistics.Anim. Behav. 34, 1899–1901. (doi:10.1016/S0003-3472(86)80279-2)

13 Martins, E. P. 1994 Structural complexity in a lizardcommunication system: the Sceloporus graciosus ‘push-up’ display. Copeia 194, 944–955. (doi:10.2307/1446717)

14 Catchpole, C. K. & Slater, P. J. B. 2008 Bird song:biological themes and variations, 2nd edn. Cambridge,UK: Cambridge University Press.

15 Nicholson, K. E., Harmon, L. J. & Losos, J. B. 2007Evolution of Anolis lizard dewlap diversity. PLoS ONE2, e274. (doi:10.1371/journal.pone.0000274)

16 Adamo, S. A. & Hanlon, R. T. 1996 Do cuttlefish(Cephalopoda) signal their intentions to conspecificduring agonistic encounters? Anim. Behav. 52, 73–81.(doi:10.1006/anbe.1996.0153)

17 Mäthger, L. M., Shashar, N. & Hanlon, R. T. 2009 Do

cephalopods communicate using polarized light reflec-tions from their skin? J. Exp. Biol. 212, 2133–2140.(doi:10.1242/jeb.020800)

18 Hebets, E. A. & Uetz, G. W. 2000 Leg ornamentation

and the efficacy of courtship display in four species ofwolf spider (Araneae: Lycosidae). Behav. Ecol. Sociobiol.47, 280–286. (doi:10.1007/s002650050667)

19 Gordon, S. D. & Uetz, G. W. 2011 Multimodal com-munication of wolf spiders on different substrates:

evidence for behavioural plasticity. Anim. Behav. 81,367–375. (doi:10.1016/j.anbehav.2010.11.003)

20 Ord, T. J. 2010 Phylogeny and the evolution of com-munication. In Encyclopedia of animal behavior (edsM. D. Breed & J. Moore), pp. 652–660. Oxford, UK:

Academic Press.21 Krebs, J. R., Ashcroft, R. & Webber, M. 1978 Song

repertoires and territory defence in the great tit.Nature 271, 539–543. (doi:10.1038/271539a0)

22 Read, A. F. & Weary, D. M. 1992 The evolution of bird

song: comparative analyses. Phil. Trans. R. Soc. Lond. B338, 165–187. (doi:10.1098/rstb.1992.0137)

23 MacDougall-Shackleton, S. A. 1997 Sexual selectionand the evolution of song repertoires. Curr. Ornithol.14, 81–124.

24 Ord, T. J., Blumstein, D. T. & Evans, C. S. 2001 Intrasex-ual selection predicts the evolution of signal complexityin lizards. Proc. R. Soc. Lond. B 268, 737–744. (doi:10.1098/rspb.2000.1417)

25 Stuart-Fox, D. & Ord, T. J. 2004 Sexual selection, natur-al selection and the evolution of dimorphic colorationand ornamentation in agamid lizards. Proc. R. Soc.Lond. B 271, 2249–2255. (doi:10.1098/rspb.2004.2802)

26 Martins, E. P., Ord, T. J. & Davenport, S. W. 2005 Com-bining motions into complex displays: playbacks with arobotic lizard. Behav. Ecol. Sociobiol. 58, 351–360.(doi:10.1007/s00265-005-0954-2)

27 Wiley, R. H. & Richards, D. G. 1982 Adaptations for

acoustic communication in birds: sound transmissionand signal detection. In Acoustic communication inbirds, vol. 1 (eds D. E. Kroodsma, E. H. Miller & H.Ouellet), pp. 131–181. New York, NY: AcademicPress.

http://dx.doi.org/10.1098/rstb.2011.0213http://dx.doi.org/10.1073/pnas.0702624104http://dx.doi.org/10.1007/s10919-006-0017-zhttp://dx.doi.org/10.1007/s10919-006-0017-zhttp://dx.doi.org/10.1086/593015http://dx.doi.org/10.1086/593015http://dx.doi.org/10.1098/rstb.2011.0217http://dx.doi.org/10.1098/rstb.2011.0217http://dx.doi.org/10.1006/anbe.2003.2047http://dx.doi.org/10.1006/anbe.2003.2047http://dx.doi.org/10.1016/j.anbehav.2004.11.015http://dx.doi.org/10.1016/j.anbehav.2004.11.016http://dx.doi.org/10.1016/j.anbehav.2004.11.016http://dx.doi.org/10.1016/j.anbehav.2007.12.019http://dx.doi.org/10.1016/j.anbehav.2007.12.019http://dx.doi.org/10.1046/j.1365-2435.2000.00474.xhttp://dx.doi.org/10.1046/j.1365-2435.2000.00474.xhttp://dx.doi.org/10.1016/j.anbehav.2005.12.003http://dx.doi.org/10.1016/j.anbehav.2005.12.003http://dx.doi.org/10.1016/S0003-3472(86)80279-2http://dx.doi.org/10.1016/S0003-3472(86)80279-2http://dx.doi.org/10.2307/1446717http://dx.doi.org/10.2307/1446717http://dx.doi.org/10.1371/journal.pone.0000274http://dx.doi.org/10.1006/anbe.1996.0153http://dx.doi.org/10.1242/jeb.020800http://dx.doi.org/10.1007/s002650050667http://dx.doi.org/10.1016/j.anbehav.2010.11.003http://dx.doi.org/10.1038/271539a0http://dx.doi.org/10.1098/rstb.1992.0137http://dx.doi.org/10.1098/rspb.2000.1417http://dx.doi.org/10.1098/rspb.2000.1417http://dx.doi.org/10.1098/rspb.2004.2802http://dx.doi.org/10.1098/rspb.2004.2802http://dx.doi.org/10.1007/s00265-005-0954-2http://rstb.royalsocietypublishing.org/



28 Brenowitz, E. A. 1986 Environmental influences onacoustic and electric animal communication. BrainBehav. Evol. 28, 32–42. (doi:10.1159/000118690)

29 Romer, H. 1993 Environmental and biological con-straints for the evolution of long-range signalling andhearing in acoustic insects. Phil. Tran. R. Soc. Lond. B340, 179–185. (doi:10.1098/rstb.1993.0056)

30 Forrest, T. G. 1994 From sender to receiver: propa-

gation and environmental effects on acoustic signals.Am. Zool. 34, 644–654.

31 Potvin, D. A., Parris, K. M. & Mulder, R. A. 2011 Geo-graphically pervasive effects of urban noise on frequency

and syllable rate of songs and calls in silvereyes (Zosteropslateralis). Proc. R. Soc. B 278, 2464–2469. (doi:10.1098/rspb.2010.2296)

32 Francis, C. D., Ortega, C. P. & Cruz, A. 2011 Differentbehavioural responses to anthropogenic noise by two

closely related passerine birds. Biol. Lett

is sociality required for the evolution of …eerc.unsw.edu.au/ord/ord_garciaporta2012.pdfresearch...

Documents