Natural selection and animal personality NielsJ.Dingemanse1,2) &DenisRéale3) (1AnimalEcologyGroup,CentreforEvolutionaryandEcologicalStudies,Universityof Groningen,POBox14,9750AAHaren,TheNetherlands;3CanadianResearchChairein BehaviouralEcologyandGroupedeRechercheenEcologieComportementaleetAnimale, DépartementdesSciencesBiologiques,UniversitéduQuébecàMontréal,CP-8888, SuccursaleCentre-ville,Montréal,Québec,H3C3P,Canada) (Accepted:1February2005) Summary Recentprogresshasbeenmadeonthestudyofpersonalityinanimals,bothfromamecha- nisticandafunctionalperspective.Whilewestartknowingmoreabouttheproximalmech- anismsresponsiblefortheconsistentdifferencesinbehaviourbetweenindividualsinapop- ulation,littleisknownyetabouttherelationshipbetweenthephenotypicdistributionofper- sonality traits, or combinations of traits, and fitness. Here we provide an overview of the available literature on the fitness consequences of personality traits in natural populations. Westartbyadescription oftwocasestudiesthathaveexaminedthe roleofnaturalselec- tiononpersonalitytraitsinthewild(i.e.,thegreattit,Parusmajorandbighornsheep,Ovis canadensis),andreviewotherstudiesthathavereportedsomelinksbetweenpersonalitytraits andfitnessindices,inalargevarietyofanimalspecies.Wecontinuebyoutliningbothdirect approaches(i.e.,measuringcorrelationalselectiononpersonalitytraitcombinations)andin- directapproaches(i.e.,comparingcorrelationsbetweenpersonalitytraitswithinandbetween populations) to study suites of correlated traits from an adaptive perspective. This review, wehope,willbeabletostimulatetheuseofthephenotypicselectionanalysisappliedtothe studyofselectiononpersonalitytraitsinanimals. Keywords:personality,behaviouralsyndromes,reproductivesuccess,sexualselection,corre- lationalselection,geneticconstraints. Introduction Behaviouralflexibilityisoftenregardedtobeunlimited,immediate,andre- versible(Sihetal.,2004a,b),allowingindividualstomaximizetheirfitness 2)Correspondingauthor’se-mailaddress:[email protected] ©KoninklijkeBrillNV,Leiden,2005 Behaviour142,1159-1184 Alsoavailableonline- 1160 Dingemanse&Réale in the many different environments they encounter during life. Contrary to this notion of behavioural plasticity as the major adaptive cause of pheno- typicvariationinbehaviour(Houston&McNamara,1999;Dalletal.,2004; Neff & Sherman, 2004), animals often show very limited behavioural plas- ticity(Sihetal.,2004a,b)andcommonlydifferconsistentlyintheirreaction towardsthesameenvironmentalstimuli(Clark&Ehlinger,1987;Wilsonet al.,1994;Boissy,1995;Wilson,1998;Gosling,2001;Greenberg&Mettke- Hofmann, 2001). These individual differences in behaviour are, moreover, frequently expressed across a wide range of contexts and situations: indi- vidualscommonlydifferconsistentlyinwholesuitesoffunctionally-distinct behavioural traits (Sih et al., 2004a, b). For instance, in birds (Verbeek et al.,1996),rodents(Koolhaasetal.,2001),andfish(Huntingford,1976),an- imals that are relatively aggressive towards conspecifics are also bolder in exploringnovelenvironmentsandpredators.Theseindividualdifferencesin suitesofcorrelatedtraitshavebeennamedbehaviouralsyndromes(Sihetal., 2004a, b), coping strategies (Koolhaas et al., 2001), temperament (Boissy, 1995; Clarke & Boinski, 1995) or animal personality traits (Buss, 1991; Gosling,2001). The evolutionary origin and maintenance of phenotypic variation in an- imal personality is poorly understood (Stamps, 1991; Wilson et al., 1994; Wilson,1998;Dingemanseetal.,2004;Dalletal.,2004;Sihetal.,2004a). Both mechanistic and functional approaches need to be applied to gain full understanding of why this behavioural variation persists (Stamps, 1991, 2003). The mechanistic approach seeks to evaluate how phenotypes result fromthecombinedinfluencesofgeneticandenvironmentalfactors;thefunc- tional approach how the interaction between phenotypes and their environ- mentaffectsfitness.Variationinpersonalityhasreceivedconsiderableatten- tionfromthemechanisticviewpoint;theemergingpatternisthatindividual differencesinsinglecomponentsofanimalpersonality(e.g.,aggressiveness) are moderately heritable and relatively stable over the entire life of the in- dividual (Boissy, 1995; Koolhaas et al., 1999; Bouchard & Loehlin, 2001) and that phenotypic correlations between components of personality (e.g., between aggressiveness and boldness) often originate from strong underly- inggeneticcorrelations(Bakker,1994;Bult&Lynch,2000;vanOersetal., 2004a; Bell, 2005). In contrast, functional approaches towards understand- ing variation in personality have received far less attention (Réale & Festa- Bianchet, 2003; Dingemanse et al., 2004; Dall et al., 2004), despite the fact Naturalselectionandanimalpersonality 1161 thatonlythecombinationofbothapproaches(inthesamestudysystem)will allowaninformedevaluationofhowbehaviouraltraitsmight(co)evolveun- derdifferentenvironmentalconditions(Fisher,1930;Endler,1986). This paper has a three-fold aim. First, we aim to provide an overview of theavailableliteratureonthefitnessconsequencesofpersonalitytraitsinnat- ural populations. In doing so, we largely concentrate our discussion on few studysystemswherebothmechanisticandfunctionalapproacheshavebeen appliedtounderstandnaturalvariationinpersonality,andwhereselectionon behavioural phenotypes has been measured in all major life-history stages andoveranumberofyears;thusprovidingfirstdetaileddescriptionsofhow andwhennaturalselectionmayoperateonanimalpersonalitytraits.Wefur- therdiscussexamplesofarangeofstudysystemswhereselectiononanimal personality has been measured incompletely using short-term proxies for fitness only. We emphasize the importance of applying a holistic approach when studying animal personality from an adaptive perspective. Notably, various studies have addressed some (supposed) functional aspect of hu- manpersonality,e.g.,susceptibilitytoillness(Grossarthmaticek&Eysenck, 1990; Schmitz, 1992) or levels of stress hormones (Bruce et al., 2002), but direct links between personality and fitness have rarely been addressed (but see Eaves et al., 1990; Mealey & Segal, 1993; Wilson, 1994; Nettle et al., 2005).Ourreviewthusfocussesprimarilyonnonhumananimals. To date, the few available naturalistic studies have addressed the func- tional consequences of animal personality by describing how selection op- erates on single components of animal personality (e.g., exploratory behav- iour of novel environments as a measure of ‘avian personality’ in great tits; Dingemanse et al., 2002). Our second aim is to point out that, ultimately, suchstudiescannotprovidefunctionalexplanationsfortheexistenceofani- malpersonalityperse,asthiswouldrequireinsightinwhyindividualsshow consistency in their behaviour, either across time, generations, contexts or situations (Dall et al., 2004; Sih et al., 2004a). We outline both direct and indirectapproachestostudysuitesofcorrelatedtraitsfromanadaptiveper- spective. Our third and last aim is to stimulate the use of the phenotypic selection approach(Lande&Arnold,1983)whenquantifyingfitnessconsequencesof animal personality, as this approach produces standardized estimates of the strength of selection that can directly be compared with those derived from otherstudiesonthesameorothertypesoftraits(Kingsolveretal.,2001). 1162 Dingemanse&Réale Naturalselectionandpersonality Natural selection is measured by the covariance between traits and fitness (Endler,1986),allowingonetoestimateboththeshape(Brodieetal.,1995) (i.e.,directional,stabilizing,disruptive)andthestrengthofselection(King- solver et al., 2001), including patterns of selection on correlated characters (Lande, 1979; Lande & Arnold, 1983; Kingsolver et al., 2001) as well as correlational selection (selection for optimal trait combinations; Barton & Turelli, 1991). These estimation models thus provide a suitable method for studyingselectiononsuitesofcorrelatedtraits,likelife-historysyndromesor personalitytraits(Réale&Festa-Bianchet,2003).Whenquantitativegenetic parametershavealsobeenquantified(i.e.,heritabilityofandgeneticcorrela- tions between components of personality traits), one can make an informed evaluationoftheevolutionaryconsequencesoftheimposedselectiveregime (Falconer & Mackay, 1996; Roff, 1997; Lynch & Walsh, 1998), including theevolutionofgeneticcorrelationsbetweenbehaviouraltraits(Roff,1996), i.e., animal personality. Here we discuss the few yet available field studies on the fitness consequences of animal personality, with special reference to twostudyspecies inthe wild,greattit (Parusmajor)andthe bighornsheep (Ovis canadensis), where both quantitative genetics parameters and fitness landscapesofanimalpersonalityhavebeenquantifiedsimultaneouslyandin natural populations, providing a first insight in the evolutionary potential of personalitytraits. Fitnessconsequencesofpersonalityingreattits Individualdifferencesinsuitesofcorrelatedtraits Laboratory studies on hand-reared great tits showed that individuals dif- fered in their reaction toward novel or challenging stimuli, comparable to howrodentsdifferinreactivity(Verbeeketal.,1994;Koolhaasetal.,2001; Groothuis&Carere,2005).Inthesebirds,speedofexplorationofnovel(lab- oratory) environments is positively correlated with aggressiveness towards conspecifics (Verbeek et al.,1996; Carere et al., in press),boldness towards novel objects (Verbeek et al., 1994), risk-taking (van Oers et al., 2004b, 2005a)andscrounging(Marchetti&Drent,2000)duringforaging,andstress responsiveness (Carere et al., 2001, 2003; Carere & van Oers, 2004). Two bi-directionalselectionexperiments,thefirstonacombinedscoreforexplo- ration and boldness (‘early exploratory behaviour’; Drent et al., 2003) and Naturalselectionandanimalpersonality 1163 the second on risk-taking behaviour (van Oers et al., 2004b), gave realized heritabilitiesof0.54and0.19respectivelyandevidenceforastronggenetic correlation(0.84)betweenearlyexploratorybehaviourandrisk-takingunder laboratory conditions (van Oers et al., 2004a). Repeatabilities (range: 0.27- 0.66; Dingemanse et al., 2002) and narrow-sense heritabilities (0.34±0.13; Dingemanse et al., 2004) were considerably lower for wild great tits, sug- gestingthatenvironmentalfactors(i.e.,thosecontrolledforinthelaboratory) alsoinfluencedexploratorybehaviourinthewild(Riskaetal.,1989;seealso Carereetal.,2005foradiscussiononenvironmentalsourcesofvariationin great tits). Alltogether, these quantitative genetics studies showed that great titsdifferinsuitesof(genetically)correlatedtraits,withtheextremesofthe trait distribution (ranging from ‘slow’ to ‘fast’ exploratory behaviour) re- flecting alternative behavioural strategies to cope with novel or challenging stimuli(Verbeeketal.,1994;Drentetal.,2003;Carereetal.,inpress).Rel- ativelyhighlevelsofbothadditiveandnonadditivegeneticvarianceinearly exploratory behaviour of laboratory-bred great tits (van Oers et al., 2004c) suggesteda history offluctuating selection pressuresin this population (see vanOersetal.,2005b). Measuringpersonalityofwildanimals To quantify selection on avian personality in the wild, large numbers of wildgreattits(1342individualsbetween1998-2002;N.J.Dingemanse,pers. comm.) were captured from a nest-box population in the Netherlands and transported to the laboratory where they were housed individually (Dinge- manseetal.,2002).Thefollowingmorning,exploratorybehaviourwasmea- sured for each bird individually in a sealed room containing five artificial trees (following Verbeek et al., 1994), before the birds were released back in the wild at their individual place of capture. The total number of flights and hops within the first 2 mins were used as an index of their exploratory behaviour(Dingemanseetal.,2002).Exploratorybehaviourwasbothrepeat- able and heritable (see above), and unrelated to age, body condition, or sex (Dingemanseet al., 2002). Subsequent field studies revealed that the fitness consequencesofavianpersonalitywerecomplex(Figures1,2). Fitnessconsequences:adultannualsurvival Selection as measured by adult annual survival acted on exploratory behav- iour(basedononetestperindividual),buttheeffectswerealwaysopposite 1164 Dingemanse&Réale Figure1. Overviewoftheconsequencesofexploratorybehaviourofwildadultgreattits (Parusmajor)fortwomajorfitnesscomponents(survivalandproductionofrecruits).Arrows representmeasured(solidlines)orpresumed(brokenlines)directorindirectrelationshipsas basedonthecorrelationalstudiesdiscussedinthetext.Symbolsindicatetheshapeoflinear (+: positive; −: negative) and non-linear (s: stabilising selection; d: disruptive selection) relationships. Notably, the relation between exploratory behaviour and offspring mass was variable (denoted ‘v’), as offspring body mass was a function of the interaction between the personality type of the individual and its mate. In cases where the consequences of exploratorybehaviourdifferedbetweenyearsorclassesofindividuals,therelationshipshave beengivenforeachgroupseparately(formoredetailsseeFigure2).Thesedescriptivestudies suggestedthatfast-exploringadultssurvivedrelativelywellinenvironmentswithintensified intra-sexualcompetition(ISC),butthattheysurvivedrelativelypoorlywhenISCwasrelaxed, andthattheoverallshapeofnaturalselectionwasstabilizing.Formoredetailsseethetext. formalesandfemales,andreversedbetweenyears(Figure2;Dingemanseet al.,2004).InayearwithmastingofbeechesFagussylvaticus(2000),when great tits experience relaxed competition for winter food (Perdeck et al., 2000),andsubsequenthighrecruitmentratesinspring(i.e.,intensifiedcom- petition for territorial space), fast-exploring adult males and slow-exploring adult females had highest survival. This pattern was reversed in two years (1999, 2001) with little winter food and subsequent low recruitment rates, whenslow-exploringadultmalesandfast-exploringadultfemaleshadhigh- estsurvivalrates. Naturalselectionandanimalpersonality 1165 Figure 2. Schematic overview of the fitness consequences of exploratory behaviour of novelenvironments(rangingfrom‘slow’to‘fast’)inwildgreattits(Parusmajor)fortwo typesofyears(poor[1999/2001]=nobeechmasting;rich[2000]=beechmasting)(Dinge- manse et al., 2004). The arrows indicate the shape of selection (→ directional selection favouring fast; ← directional selection favouring slow; →← stabilising; ←→ disruptive) fortwomainfitnesscomponents,adultannualsurvivalandoffspringrecruitment,Hatched barsindicatenonsignificanttrends.Formoredetailsseetext. Temporalvariabilityinenvironmentalconditions Dingemanse et al. (2004) hypothesized that beech masting affected the strength of intra-sexual competition but that these effects were always op- positeforterritorialmalesandfemales(Figures1,2):becausefemaleswere subordinatetomales(Dingemanse&deGoede,2004),theywerelikelytobe mostaffectedbycompetitionforwinterfood.Beechmastingthereforemay resultinrelaxedintra-sexualcompetitionamongfemales,whilecompetition is intensified in years without beech mast. As only males defend territories, they are likely to be most affected by competition for territorial space. In years with beech masting, recruitment rates are high (Perdeck et al., 2000), resulting in intensified intra-sexual competition among males (Both et al., 1999), while in non-beech mast years competition is relaxed. The complex patterns in adult survival (Figures 1, 2) may thus reflect that fast-exploring adultssurvivedrelativelywellinyearswithintensifiedintra-sexualcompeti- tion,andthattheysurvivedrelativelypoorlywhencompetitionwasrelaxed. Notably, the potential cause for the poor survival of fast-exploring adults insuchyearshasnotyetbeenidentified.Thisnotionofdifferentialcompeti- tiveabilitywassupportedbythefindingthatfast-exploringadultsdominated slow-exploring adults when competing for winter food (Dingemanse & de Goede, 2004) and that fast-exploring adults bred on the best breeding terri- 1166 Dingemanse&Réale tories(Bothetal.,2005).Factorsaffectingadultsurvivalaresummarizedin Figure1. Between-yearfluctuationinselectiononpersonalitytraitsinthegreattits is similar to results from other studies on other types of traits (Merilä et al., 2001).Thisresultindicatesthatselectionstudiesshouldbeperformedonthe long-termifwewanttounderstandboththeimmediateconsequences(within a year) and longer term effects (across several generations) of selection on populations. Fitnessconsequences:offspringproductionandrecruitment Slow-exploring females had higher nest success (were more likely to pro- duceatleastonefledgedoffspring),andproducedlargeroffspringthanfast- exploring females (Both et al., 2005). Pairs of assortative phenotypes, con- sisting of two slow partners or two fast partners, produced offspring with highest body mass in all years of the study (Both et al., 2005). Selection as measured by the number of these offspring that survived and bred in the studyarea(‘offspringrecruitment’)actedonfemale,andtoalesserextenton male exploratory behaviour and fluctuated between years (Figure 2; Dinge- manseetal.,2004).Selectiononexploratorybehaviourwasstabilisinginthe two years without beech masting, but was disruptive in the year with beech masting.Thepersonalityofboththemaleandthefemaleparentcontributed to this pattern of disruptive selection, as pairs consisting of assorted part- ners (i.e., fast-fast or slow-slow pairs) produced most recruits in the beech cropyear(Dingemanseetal.,2004).Theseassortedpairsalsoproducedoff- springofhighestbodymass(seeabove),andasbodymassaffectscompeti- tiveabilityandjuvenilewintersurvivalinyearswithintensecompetitionfor resources among juveniles (Both et al., 1999), this pattern of disruptive se- lection probably acted via offspring body massin the yearwith beech crop. Interestingly, pairs of medium-exploring adults nevertheless produced most recruits in years without beech crop, suggesting that the higher than aver- ageoffspringbodymassofassortativepairsonlyincreasedfitnessincertain years and that other characteristics of the offspring phentoype (e.g., their exploratorybehaviour,seeFigure1)affectedoffspringrecruitment. Fitnessconsequences:explainingvariablepatternsinoffspringrecruitment While these variable patterns in adult survival have now resulted in testable hypotheses, (i.e., fluctuating and sex-specific survival (Figure 2) reflected Naturalselectionandanimalpersonality 1167 variable selection for competitive ability with sexes; Figure 1), sources of variation in offspring recruitment are not well understood (see question marks in Figure 1). As outlined above, the variance in offspring recruit- ment partly resulted from variation in parental breeding performance, but primarily in years with beech crop. Offspring recruitment patterns may also have been partly mediated directly via exploratory behaviour inherited fromparentstooffspring(e.g.,byaffectingoffspringforagingsuccess;Fig- ure1).Fieldstudiesshowedthatexploratorybehaviouraffectsbothcompet- itive ability and settlement decisions of juveniles: fast-exploring juveniles hadlowestdominancerankswhennonterritorial (Dingemanse&deGoede, 2004)andcametobreedfurtherfromhome(Dingemanseetal.,2003). Evolutionaryconsequences Natural selection acted on avian personality, but the direction of selection variedbetweensexes,age-classesandyearswithdifferentselectiveregimes. Becauseexploratorybehaviourofwildgreattitsisheritable(seeabove)and affectscomponentsoffitness,selectiononavianpersonalitycanleadtoevo- lutionary change (Fisher, 1930; Endler, 1986). While considering that the response to selection depends both on the frequency with which individ- uals experience different selective environments as well as the strength of selection in these environments (Figure 1), the overall pattern of selection turned out to be stabilising (Dingemanse et al., 2004): Adults of interme- diate phenotype had highest offspring recruitment rates in most years, as mastingofbeechesoccursonlyaboutonceeverythreeyears(Perdecketal., 2000). Furthermore, the variance in adult survival was lowest for interme- diate phenotypes, resulting in highest overall life expectancy. Taking these long-term fitness consequences into consideration, adult males may have maximized their fitness by means of adaptive mate choice: adult males of extreme phenotype were mated disassortatively with respect to personality type(Dingemanseetal.,2004),allowingthemtoproduceoffspringofinter- mediatephenotypeandincreasetheirlifetimefitness.Notably,disassortative matingseemedmaladaptivewhenonlyconsideringthatassortativepairshad highest reproductive success (Both et al., 2005). Temporal variability in se- lection as observed for this study system can slow down the loss of genetic variation in avian personality (Sasaki & Ellner, 1997; Burger & Gimelfarb, 2002),butitcannot,however,provideanultimateexplanationforthemain- tenance of genetic variation in avian personality. Either a balance between 1168 Dingemanse&Réale mutation,selectionandmigrationinaspatiallyvariableenvironment(Nevo, 1988; Frank & Slatkin, 1990) or frequency-dependent selection (Maynard Smith, 1982) probably need to be invoked to explain this behavioural di- versity from an adaptive perspective (Dingemanse et al., 2004; Both et al., 2005). Similarly, we do yet need to reveal why individual great tits showed suchlimitedbehaviouralplasticity,asbehaviouralflexibilityseemsadaptive insuchatemporallyvariableenvironment(Dall,2004). Fitnessconsequencesofboldnessanddocilityinbighornsheep Individualdifferencesincorrelatedbehaviours In a wild Canadian population of bighorn sheep, individuals differed con- sistently in their willingness to enter corral traps baited with salt (Réale et al.,2000).Thisbehaviouralvariabilitywasassumedtoreflectindividualdif- ferences in boldness (i.e., willingness to take the risk involved in licking salt), where boldness was measured as the yearly number of times a ewe wascapturedinthetrap.Repeatability(betweenyears)andheritabilityesti- mates were 0.36 and 0.21 respectively. Ewes captured in the trap were also compared for their docility during handling: a docility score (based on a 7- point scale) was used to measure how much individuals struggled during handling. Docility was highly repeatable both within (r = 0.65-0.66) and between years (r = 0.86); while some ewes were relatively docile, others struggledtoescape.Therewasanegative—thoughweak—phenotypiccor- relationbetweenboldnessanddocility:shyeweswerealsorelativelydocile. This negative pattern appeared to be caused by the absence of shy, non- docile ewes. Estimation of quantitative genetics parameters using the ‘ani- malmodel’(Lynch&Walsh,1998)revealedsignificantheritabilitiesofboth behavioursaswellasamoderatenegativegeneticcorrelationbetweenthese behaviours(D.Reale&D.Coltman,unpubl.data). Fitnessconsequences:reproductivesuccess Using standard multiple regression techniques to evaluate selection on cor- relatedcharacters(Lande&Arnold,1983),selectiononeachbehaviourwas measured independently of selection on the other (Réale et al., 2000; Réale & Festa-Bianchet, 2003). Selection measured with age at first reproduction as a fitness index acted both on boldness and docility (Réale et al., 2000). Boldewesreproducedatanearlieragethanshyewes.Similarly,docileewes