MolecularEcology(2012)21,3879–3892 doi:10.1111/j.1365-294X.2012.05644.x Mate choice and the genetic basis for colour variation in a polymorphic dart frog: inferences from a wild pedigree CORINNE L. RICHARDS-ZAWACKI,*†IAN J. WANG‡ andKYLE SUMMERS§ *Department ofEcology andEvolutionary Biology, 400LindyBoggs Bldg.,Tulane University, New Orleans,LA 70118,USA, †Smithsonian TropicalResearch Institute, Apartado Postal 0843-03092,Balboa, Ancon, Panama,Republica de Panama, ‡DepartmentofOrganismic and Evolutionary Biology, Harvard University, Cambridge, MA02138, USA,§Department of Biology, East CarolinaUniversity, EastFifth St., Greenville,NC 27858, USA Abstract Understandinghowreproductivebarriersevolveduringspeciationremainsanimportant question in evolution. Divergence in mating preferences may be a common first step in this process. The striking colour pattern diversity of strawberry dart frog (Dendrobates pumilio) populations has likely been shaped by sexual selection. Previous laboratory studieshaveshownthatfemalesattendtomalecolorationandprefertocourtwithmales oftheirowncolour,suggestingthatdivergentmorphsmaybereproductivelyisolated.To test this hypothesis, we used molecular data to estimate pedigree relationships from a polymorphic population. Whereas in the laboratory both red and yellow females preferred to court with males of their own phenotype, our pedigree shows a pattern of assortative mating only for red females. In the wild, yellow females appear to be less choosy about their mates, perhaps because they incur higher costs associated with searchingthanfemalesofthemorecommonredphenotype.Wealsousedourpedigreeto investigatethegeneticbasisforcolour-patternvariation.Thephenotypefrequencieswe observed were consistent with those expected if dorsal background coloration is controlledbyasinglelocus,withcompletedominanceofredoveryellow.Ourresultsnot onlyhelpclarifytheroleofsexualselectioninreducinggeneflow,butalsoshedlighton the mechanisms underlying colour-pattern variation among sympatric colour morphs. The difference we observed between mating preferences measured under laboratory conditionsandthepatternofmatechoiceobservedinthewildhighlighttheimportance of field studies for understanding behavioural reproductive isolation. Keywords: heritability, mate choice,pedigree, polymorphism,sexualselection, speciation Received16September 2011;revision received18April2012; accepted 24April 2012 reproductive barriers can continue to accumulate and Introduction strengthen after new species have formed, the barriers The question of how reproductive isolating barriers that cause speciation are those that reduce gene flow evolve among incipient species is one of the central before reproductive isolation is complete. For this problems of speciation. In many cases, divergent popu- reason, studies of recently or nearly diverged popula- lations show strong assortative mating in the absence of tions, where the action of these barriers can be more intrinsic post-zygotic incompatibilities, suggesting that directly observed, provide the most accurate picture of divergence in mating signals and associated female the process (Coyne & Orr 2004; Rundle & Nosil 2005; preferences may be a common first step in the specia- Sobel et al. 2010). tion process (McMillan et al. 1997; Seehausen et al. The strawberry dart frog (Dendrobates pumilio, 1999;Jiggins&Mallet2000;Jigginset al.2004).Whereas formerly Oophaga pumilio, see Santos et al. 2009) is amazingly variable in coloration on the islands of the Correspondence:CorinneRichards-Zawacki, Bocas del Toro archipelago and adjacent mainland of Fax:+1 504 862 8706;E-mail:[email protected] Panama (Daly &Myers 1967; Summers et al. 2003). This (cid:2)2012BlackwellPublishingLtd 3880 C. L. RICHARDS-ZAWACKI ET AL. variation apparently arose with the formation of the about potential mates. In natural populations, con- archipelago over the past 6000years (Summers et al. straints on time, memory and mobility prevent females 1997;Anderson&Handley2002).Althoughmost colour from visiting all possible mates (Janetos 1980). There- and pattern variants (hereafter, colour morphs) are allo- fore, the degree to which mate choice can be predicted patric, there are a few areas where two or more distinc- from studies of mating preferences will depend upon tive colour morphs occur sympatrically. Despite their the sampling strategies females use to select their amazingdiversity, allofthe Panamaniancolour morphs mates. are considered to be conspecific based on the similarity If female D. pumilio mate with the first male they of male mating calls (Daly & Myers 1967; Prohl et al. encounter (random mating), preferences observed in the 2007) and the fact that all populations (with the excep- laboratory would not be a good predictor of the pattern tion of frogs from the much more distant island of of mate choice that females exhibit in the wild. Even if Escudo de Veraguas) fall into one clade with minimal females sample multiple males, the attractiveness of a genetic divergence in rapidly evolving molecular mark- given mate may depend upon the others with whom he ers (Summers et al. 1997; Hagemann & Prohl 2007; is being compared (Bateson & Healy 2005). Hence, mat- Wang& Shaffer 2008;Hauswaldt et al. 2010). ing preference results from laboratory experiments may Laboratory experiments, in which females are simul- or may not predict patterns of mate choice, depending taneously exposed to males of two or more phenotypes, upon how many males are compared and how the suggest that female D. pumilio prefer to court with experiments were designed. In either case, the previous males of their own colour over males of other colours experiments could have either underestimated the com- (Summers et al. 1999; Reynolds & Fitzpatrick 2007; plexity or misrepresented the nature of sexual selection Maan & Cummings 2008, 2009; Richards-Zawacki & among D. pumilio morphs. Whereas studies of mating Cummings 2011). Whereas most of these studies have preference are important for understanding the sensory involved only females from monomorphic populations, underpinnings of differences in mating behaviour assortative mating preferences were also observed for among morphs, a clearer understanding of the pattern females from a polymorphic population on the island of mate choice exhibited in natural populations is of Bastimentos in Panama (Richards-Zawacki & needed to assess the extent to which mating behaviour Cummings 2011). Both red and yellow females from leads to reproductive isolation. One way to obtain such this population preferred to court with males of their information is by testing for a pattern of assortative own phenotype when given a choice between red or mating using pedigree information from a polymorphic yellow. If the female preferences observed in these population. choice experiments are representative of the mate choice decisions that female D. pumilio make in the The power of pedigrees wild, some degree of behavioural reproductive isolation may already be present, even among sympatric colour Divergence in mating preferences among populations, morphs in polymorphic populations. This assumption, like any evolutionary change, requires a combination of however, deserves explicit testing as in several frog two elements: (i) heritable genetic variation that under- species, mating preferences observed in the laboratory lies the distribution of phenotypes and (ii) a mechanism are not reflected in the pattern of mating in the field (such as selection or genetic drift) that alters that distri- (Gerhardt 1992). bution from generation to generation. Pedigrees can be used to study both of these elements of evolutionary change: they can be used to quantify the phenotypic Preference vs. choice covariance of relatives, providing insight into the Differences between patterns observed in laboratory genetic basis of traits, and breeding patterns from choice experiments and field studies may be common pedigrees can be linked to these traits to estimate because these two approaches are in fact measuring dif- selection(Kruuk&Hill2008).Althoughgeneticistshave ferent aspects of female mating behaviour. Whereas recognized the power of pedigrees for more than a choiceexperimentsareappropriateforstudying‘mating century, only recently have biologists begun to realize preferences’ (sensory and behavioural properties that their potential in providing insight into the forces influence the propensity of females to mate with certain driving evolutionary dynamics in wild animal phenotypes, sensu Heisler et al. 1987), field studies pro- populations (Pemberton 2008). For example pedigrees vide estimates of ‘mate choice’ (the pattern of mating have recently been used to study breeding patterns that arises in part as a result of mating preferences). (e.g., Blackmore & Heinsohn 2008; Grant & Grant 2008; Mate choice depends not only on mating preferences, Ursprung et al. 2011a), heritability and the quantitative but also on how females acquire and use information genetic architecture of traits (e.g., Charmantier et al. (cid:2)2012BlackwellPublishingLtd MATE CHOICE ON A HERITABLE COLOUR POLYMORPHISM 3881 2006) and gene flow (e.g., Zeyl et al. 2009) in wild ani- in the cleared areas. Frogs in this population differ in mal populations. As direct observations of reproduction dorsal coloration and each colour morph appears more are usually unavailable, pedigrees for wild populations or less evenly distributed across the study area (Fig. S1, are usually estimated using molecular marker data. In Supporting information). The frogs were captured these cases, DNA derived from tissue, scat, fur or feath- between July 2007 and January 2009 and transported to ers are used to infer relationships statistically. The the Smithsonian Tropical Research Institute’s Bocas del resulting pedigree estimates are particularly useful in Toro Research Station. There, digital photographs of the the study of wild animal mating systems, especially for dorsal and ventral surfaces, as well as body size (snout- taxa where direct observation of mating is impossible, vent length, SVL) measurements of each frog were prohibitively time-consuming or potentially misleading taken against a grey card with a flat spectral response (e.g., Carpenter et al. 2005; Gottelli et al. 2007; reviewed across the visible spectrum (Digital Gray Card(cid:4), Robin inJones &Wang2010). Myers Imaging). Photographs were taken, within 6 h of capture, from a distance of 10cm under uniform light- ing, and after the frogs had acclimated for at least Study objectives 30min to room temperature (23(cid:3)C). Frogs were then Using molecular information gathered in conjunction weighed using a digital scale (Gempro-500, accurate to with a mark recapture study, we estimated pedigree 0.002g), classified by sex and the tip of the fourth toe relationships among D. pumilio individuals from a of each hindlimb was removed. Like some other poison polymorphic population on the northwest tip of the frogs (e.g., Allobates femoralis, Ursprung et al. 2011b), island of Bastimentos. Using simulations, we tested the D. pumilio regenerates its digits after toe-clipping and effect of our mating system assumptions on the confi- no adverse effects of this tissue sampling method were dence of our pedigree assignments. The phenotypes of seen over the course of our study. Sex was determined mated pairs and their offspring were then used to test based on SVL and the presence (male) or absence for a pattern of assortative mating by colour, consistent (female)ofadark throatpatch. Malesinthis population with results of a laboratory study (Richards-Zawacki & usually had dark throat patches by 18.0mm SVL; there- Cummings 2011). We also investigated the mode of fore, any frog 18.0mm or larger with no dark throat inheritance of colour–pattern variation by asking patch was treated as female. Frogs smaller than whether the patterns we observed could be explained 18.0mm SVL were treated as juveniles. Because frogs by a single locus of large effect showing simple Mende- in this population breed year-round, other than lian dominance. Our results not only contribute impor- distinguishing juveniles from adults, we were unable to tant new insight into the degree to which colour distinguish cohorts of similar age within our sample. Of morphs are reproductively isolated, but also shed light the 169 adult frogs recaptured during this study, six on the mechanisms underlying colour–pattern variation (3.6%) had been classified as the wrong sex upon first and mating preferences among morphs. These findings capture. All of these were males whose dark throat help to improve our understanding of the role of sexual patch was missed or was not present at first capture. selection in the rapid evolution of colour–pattern poly- Frogs were returned to their point of capture within morphism among Panamanian D. pumilio, and empha- 48h. size the importance of field studies for studies of mate choice. Colour analysis To quantify colour variation, colour and contrast values Methods among a subset of digital photographs were standard- ized by adjusting midtones to a target value using the Field sampling and measurements of phenotype ‘auto colour’ command in Adobe Photoshop CS 8.0. A total of 677 Dendrobates pumilio (312 male, 274 female, Colours were characterized by averaging the red, green, 91 juvenile) were captured by hand from a 0.75ha area blue (RGB) colour values from three random points on of secondary-growth forest on the northwest tip of the each of the dorsal and ventral surfaces (excluding the island of Bastimentos (9.3468(cid:3)N, 82.2064(cid:3)W). The study dark throat patch of males). Variation among individu- site was bounded on its northern and western sides by als was summarized using a principal components cleared land (a cemetery and residential areas) and on analysis. A two-step cluster analysis using the Bayesian its eastern and southern sides by a stream. We have Information Criterion was used to estimate the number never encountered D. pumilio in the marshyhabitat that and composition of natural colour groups in the lies on the other side of the stream; however, occasion- population. All statistical analyses were performed in ally individuals can be found in and around large trees SPSS 18.0. Results of the cluster analysis were compared (cid:2)2012BlackwellPublishingLtd 3882 C. L. RICHARDS-ZAWACKI ET AL. to assignments of the same frogs to colour groupings Pedigree inference madebyahumanobserver(CRZ)todeterminewhether visual inspection of photographs was a reliable way of As we could not distinguish cohorts (other than juve- assigningfrogs tocolour groups. nilesvs.adults)bybodysizeorothermetric,wepooled all adult males as candidate fathers, all adult females as candidate mothers and all frogs (male, female and juve- Molecular genetic methods nile) as candidate offspring for pedigree analysis. Real- Toe clip tissues were preserved in a salt-saturated izing the challenges this poses for pedigree estimation dimethyl sulfoxide (DMSO) and ethylenediaminetetraa- (reviewed in Koch et al. 2008), we used a multifaceted ceticacid(EDTA)solutionatroomtemperatureforupto approach involving three pedigree estimation programs 20months prior to extraction. Genomic DNA was iso- to improve confidence in our inferred relationships and latedusingaQIAGENDNeasykitaccordingtothestan- the conclusions that were drawn from them. Pedigree dard protocol for animal tissue. Extracted samples were analyses were performed using CERVUS 3.0 (Kalinowski dilutedto2.5ng⁄lLpriortoamplificationbypolymerase et al. 2007), COLONY 2.0 (Jones & Wang 2009) and MASTER- chainreaction.Tissuesamplesfrom677individualswere BAYES2.47 (Hadfield et al. 2006),asdescribed below. amplified at 15 microsatellite loci using primers previ- CERVUS is a likelihood based program that uses simu- ously developed by Wang & Summers (2009; Dpum12, lations to determine thresholds for parentage assign- Dpum13, Dpum14, Dpum24, Dpum44, Dpum92 and ment. For our analysis, we used simulations of 10000 Dpum110) and Hauswaldt et al. (2009; OopB8, OopB9, offspring with a genotyping error rate of 1%. For simu- OopC3, OopC11, OopD4, Oop E3, OopF1 and OopH5). lation, CERVUS requires an estimate of the average num- Forward primers were labelled with a 5¢-fluorescent tag ber of candidate mothers and fathers per offspring (6-FAM, NED, VIC or PET). polymerase chain reactions (which should include all adults known or thought to were carried out in 10lL volumes containing 1· buffer be present at the time of breeding). To estimate these (Applied Biosystems), 0.2mM each dNTP, 0.5lM for- values, we used the POPAN open-population model in ward and reverse primers, 0.25U AmpliTaq Gold Taq MARK 5.1 (White & Burnham 1999) on mark–recapture DNA polymerase (Applied Biosystems) and 2.5ng tem- data recorded during five sampling sessions: (i) July plateDNA.ForalloftheHauswaldtet al.(2009)primers, 13–14, 2007; (ii) March 24–April 1, 2008; (iii) June 16–17, Dpum12 and Dpum13, the MgCl concentration was 2008; (iv) October 20–22, 2008 and (v) January 9–14, 2 2.5mM.FortherestoftheWang&Summers(2009)prim- 2009. Photographs of the dorsal colour-pattern were ers, the MgCl2 concentration was 3.75mM. For Dpum12 used to identify individuals. This was possible because andDpum13,5%DMSOwasalsoaddedtothepolymer- in this population, markings are both unique and fixed asechainreactionmastermix.Polymerasechainreaction soon after metamorphosis.The reliability ofthismethod cyclingconditionswereasinHauswaldtet al.(2009)and is supported by the observation that of 679 frogs Wang&Summers(2009).Eachlocuswasamplifiedindi- diagnosed by dorsal pattern as being distinct individu- vidually and run on an ABI 3730 Genetic Analyser als, in only two cases (0.3%) did we later find identical (Applied Biosystems). Eleven reference samples with genotypes. Photographs allowed us to diagnose both of known genotype for each locus were included in each these as recaptures where the colour-pattern match had runtoensurereliablescoringofgenotypesacrossdiffer- initially been missed, rather than identical genotypes ent gels. Fragments were sized with LIZ-500 size stan- occurring in two individuals. The best-supported dard (Applied Biosystems) and collected with GeneScan POPAN model resulted in estimates of N=965 females version3.1(AppliedBiosystems).Scoringwasperformed and N=1044 males, where N is an estimate of the total withGeneMarker(SoftGenetics,LLC). number of frogs that came into the population during the study period (the super population size). This resulted in estimates of 0.308 and 0.302 respectively, for Population genetic analysis the proportion of candidate mothers and fathers We tested for deviations from Hardy–Weinberg sampled. equilibrium (HWE) and linkage disequilibrium (LD) Unlike CERVUS, which uses pairwise comparisons to using GenePop 4.1 (Raymond & Rousset 1995). Allele assign parentage, COLONY considers the likelihood of the frequencies, observed (H ) and expected (H ) entire pedigree structure in assigning pedigree relation- O E heterozygosity, number of alleles and frequency of null ships. Our COLONY analysis was attempted in multi-core alleles for each locus were estimated using CERVUS 3.0 processor mode, using the full likelihood method and a (Kalinowski et al. 2007). Genetic structure among colour medium run length. We assumed both male and female groups was estimated using the infinite alleles model polygamy and a 1% error rate. This analysis proved to (FST)inARLEQUIN3.0(Schneider et al. 2000). be prohibitively long for our full data set, likely as a (cid:2)2012BlackwellPublishingLtd MATE CHOICE ON A HERITABLE COLOUR POLYMORPHISM 3883 result of the large number of candidate offspring, and Mode of inheritance of dorsal coloration was terminated prior to completion after several months. However, we were able to use COLONY to pro- Usingthemother–father-offspringtriosidentifiedbyour vide a check of the parent–offspring trios identified by pedigree analysis, and the frequency of the yellow phe- CERVUS. We ran COLONY ten times, each time with all notype in this population, we tested the hypothesis that individuals as potential parents and a different subset dorsal coloration could be largely controlled by a single of individuals as potential offspring. Each offspring locus with two alleles and complete dominance of red subset contained all offspring identified by CERVUS plus over yellow (i.e. simple Mendelian dominance: X=red, an equal number of randomly chosen individuals. Par- x=yellow;Fig. 1).Thisisthescenariothatbestexplains ent–offspring relationships supported at 95% confi- limited observations from captive breeding of these two dence in at least one COLONY run were used in morphs: yellow pairs have only ever produced yellow subsequent analyses. offspring,whereaspairswhereatleastoneparentisred Pedigree analyses in MASTERBAYES can be run using (red+red and red+yellow) produce a mix of red and either maximum likelihood or Bayesian approaches. If yellowoffspring(CRZ,unpublisheddata).Thefrequency only genotypic data are being used, confidence levels ofyellowdorsalcolorationintheBastimentospopulation for assignments can be calculated analytically using (q2) was estimated from photographs of all individuals maximum likelihood, removing the need for a Bayesian captured during the 18-month mark–recapture study. approach (Hadfield et al. 2006). Nevertheless, we ran From this value, the frequency of the recessive (q) and both maximum likelihood and Bayesian parentage anal- dominant(p)allelesaswellastheexpectedfrequencyof yses in MASTERBAYES. The offspring pool for these analy- homozygous(p2)andheterozygous(2pq)redfrogswere ses was limited to all juvenile individuals plus adult derived, assuming HWE. The observed and expected individuals identified as offspring by CERVUS because frequencies of yellow offspring for different combina- preliminary trials with MASTERBAYES indicated poor per- tions of parent phenotypes were compared to test their formancewhenthereisoverlapbetweenparentandoff- fittoHardy–Weinbergexpectations. spring pools (individuals were often assigned as their ownparent).Theparentpoolconsistedofalladultindi- Results viduals except those identified as trio offspring by CER- VUS. For the maximum likelihood analysis, results from Colour variation a run where MASTERBAYES was allowed to estimate the number of unsampled males and females were com- The principal components analysis of averaged R, G pared with those of a run where estimates from the and B values extracted a single significant component mark–recapture analysis were provided. For the Bayes- (PC) for each of dorsal (n=217) and ventral (n=68) ian analysis, we ran the Markov chains for 250000 iter- coloration. For dorsal coloration, the first PC (eigen- ations with a burn-in of 50000 iterations and a thinning value=1.905) explained 63.5% of the variance in color- interval of100. ation and was positively correlated with variation in G and B (coefficient scores: R=)0.89, G=0.505, B= 0.512). The second PC explained 33.2% of the variance, Effects of multiple matings on confidence in pedigree was marginally non-significant (eigenvalue =0.997) and assignments was strongly positively correlated with variation in R Our CERVUS analysis assumed no inbreeding and that no (coefficient scores: R=0.988, G=0.164, B=0.011). For close relatives were present among the candidate ventral coloration, the first PC (eigenvalue =2.481) parents.Thecriticaldeltascoresfor90%and95%confi- explained 82.7% of the variance in coloration and was dence obtained from this analysis were used to identify approximately equally influenced by R, G and B (coeffi- mother–father–offspring trios. To examine how the cient scores: R=0.330, G=0.396, B=0.371). Frogs clas- presence of relatives with a coefficient of relationship sified by eye as belonging to red and yellow dorsal (r, Wright 1922) of 0.25 or more would affect confidence colour groups are clearly distinguishable in a plot of in our parent–offspring assignments, we ran a series of the first two dorsal PCs (Fig. 2a). However, these two simulations in CERVUS with different numbers (1–10) and groups are not differentiated in terms of ventral colora- typesofrelatives.Wethenusedfieldestimatesofclutch tion (Fig. 2b). A two-step clustering analysis suggested size and survival to adulthood to estimate the true that dorsal coloration falls into two natural clusters, confidence, and the number of full siblings likely to be which almost perfectly match the assignments made by present in our data set, as a function of the number of eye (2 of 217, or <1% of assignments differed). This repeatmatingsbetweenparents.Detailsofthesecalcula- supports the idea that colour groupings made by visual tionscanbefoundinthesupportinginformation. inspection of photographs accurately represent natural (cid:2)2012BlackwellPublishingLtd 3884 C. L. RICHARDS-ZAWACKI ET AL. Fig. 1 Expectations for offspring phenotypes if red is completely dominant over yellow. The frequency of yellow dorsal coloration in thepopulation(q2) wasfoundtobe 0.35.Fromthisvalue,assumingHardy–Weinberg equilibrium, thefrequency ofthe recessive (q=0.592) and dominant (p=0.408) alleles, as well as the expected frequency of homozygous (p2=0.166) and heterozygous (2pq=0.484)redfrogswerederived.Theproportionofredfrogscarryingarecessiveallelewascalculatedas2pq⁄(2pq+p2)=0.735. Fromthisvalue,theexpectedfrequenciesofeachgenotypecombinationforred+yellowandred+matingswerecalculated. groupings in this population. Through visual inspection an average of 18.27 alleles per locus and an average H O of photographs, the frequency of yellow dorsal colora- of 0.785 (Table 1). Frogs from the two dorsal colour tion was found to be 0.35 for both adult males (110 of groups (red and yellow) were not differentiated geneti- 312) andfemales(96 of274). cally(F =0.0003,P=0.17). ST Population genetic analysis Pedigree inference Linkage disequilibrium was detected between OopD4 From among the 677 genotyped frogs, our CERVUS pedi- and Dpum110. Because linkage between these two loci gree analysis identified 32 mother–father–offspring trios was not found in other Dendrobates pumilio populations with 90% confidence, of which 24 were also identified (CRZ unpublished data) and because linkage between with 95% confidence. For each of these trios, a second one pair of loci is unlikely to bias confidence in parent- round of screening was performed by comparing age assignments (Marshall et al. 1998), these two loci mother, father and offspring alleles at two additional were used in the pedigree analyses. Four loci (Dpum12, loci (Dpum14 and Dpum24). As both these loci are Dpum13, Dpum14 and Dpum24) were found to be out known to harbour null alleles, incidences where a mis- of HWE after Bonferroni correction. Comparisons match could be attributed to the presence of a null among individuals that were genotyped multiple times allele were ignored. However, there were two cases at indicate that two of these loci were unreliable because the 90% confidence level where alleles at these two loci they produce non-repeatable results (Dpum12 and were incompatible with the proposed parent–offspring Dpum13). The other two loci (Dpum 14 and Dpum 24) relationship. These trios were considered false positives produced repeatable results but harboured null alleles and not used in subsequent analyses, bringing the total at high frequencies. Only the 11 loci in HWE were used to 30 trios with 90% confidence and 24 with 95% confi- in the pedigree analyses to assign mother–father–off- dence (Fig. S3, Supporting information). Eleven (36.7%) spring relationships. However, alleles at Dpum14 and of the 90% confidence trios had juvenile frogs at the Dpum24 were later used to check the accuracy of those offspring position. The proportion of trios with juvenile assignments (see pedigree analysis, below). Across the offspring was similar (8 of 24, or 33.3%) for the 95% 11 microsatellite loci in HWE and 677 frogs, we found confidence trios. Considering that we only sampled 91 (cid:2)2012BlackwellPublishingLtd MATE CHOICE ON A HERITABLE COLOUR POLYMORPHISM 3885 Table 1 Genetic diversity and results of test for departure 4 fromHWEfor677inindividualsofDendrobatespumilio 3 Numberof Nullallele Uncorrected 2 2 Locus alleles HO HE frequency P-value C 1 P al 0 Usedinpedigreeestimation s OopH5 19 0.874 0.878 0.002 0.1912 or D –1 OopB9 14 0.869 0.876 0.004 0.6465 OopC3 14 0.871 0.897 0.014 0.2806 –2 OopF1 16 0.870 0.875 0.002 0.3985 –3 OopE3 25 0.920 0.923 0.001 0.7778 –2 –1 0 1 2 3 4 OopC11 25 0.867 0.888 0.013 0.1002 Dorsal PC 1 OopD4 18 0.706 0.718 0.009 0.0675 OopB8 21 0.924 0.919 )0.003 0.1820 3 Dpum110 17 0.850 0.867 0.010 0.0087 Dpum92 2 0.009 0.009 )0.001 0.9999 2 Dpum44 30 0.878 0.884 0.003 0.1558 2 1 Mean 18.27 0.785 0.794 C Notusedinpedigreeestimation P al 0 Dpum14* 8 0.063 0.079 0.107 <0.0001 Ventr –1 DDppuumm2142*** 156 00..684191 00..970281 )00..106767 <<00..00000011 –2 Dpum13** 9 0.767 0.774 0.020 <0.0001 –3 Locidenotedby(*)wereusedonlyasasecondarycheckof –3 –2 –1 0 1 2 3 4 parent–offspringassignments.Locidenotedby(**)werenot Ventral PC 1 usedatall.Abbreviations:H =observedheterozygosity, O H =Expectedheterozygosity.CriticalP-valueforHWEtest E Fig. 2 Plots of dorsal (a) and ventral (b) colour variation (a=0.05),afterBonferronicorrection=0.0045. among individual Dendrobates pumilio. Symbols indicate dorsal colourcluster:blackcircles=frogsclassifiedasredbyeyeand by cluster analysis; open circles=frogs classified as yellow by femalefrogsofeithercolorshouldshareparenthoodwith eye and by cluster analysis; grey triangles=frogs classified as yellowmales35%ofthetime.Yellowfemalesappearto redbyeyebutyellowbyclusteranalysis. followthispattern.InthesetoftriossupportedbyCERVUS at 90% confidence, 38% of yellow females mated with juveniles out of a total of 677 individuals, the probabil- yellow males (Table 2). Alternative analyses including ity of parents being assigned was higher for juveniles more conservative subsets of these trios (i.e. those sup- (12.1%) than for adults (3.2%). This pattern is expected ported by CERVUS at 95% confidence, those containing as juvenile offspring are more likely to have parents juveniles as offspring and those also supported by COL- still alive and available for sampling within the popula- ONY and⁄or MASTERBAYES) produced similar results tion than offspring that have already reached sexual (Table S1, Supporting information). The frequency of maturity and may be anywhere between 10months and yellow–yellowpairingsdoesnotdifferfromtheexpecta- several yearsold themselves. tions for random mating in any of these analyses (Bino- Parent–offspring relationships for 25 of the 30 trios mial Exact Test, P‡0.437), although for some analyses supported by CERVUS were also supported by COLONY the power to detect such a difference would be low and⁄or MASTERBAYES with 95% confidence (Fig. S3, Sup- becauseofthesmallsamplesizeofyellowfemales. porting information). For MASTERBAYES, the same relation- Unlike yellow females, red females show a statisti- ships were supported regardless ofwhether the number cally significant pattern of assortative mating (Binomial of unsampled parents was estimated directly from the Exact Test P£0.049). This is true whether we consider data or whether estimates of these numbers from the the CERVUS trios (Table 2), where 91% of red females POPAN mark–recapture analysis were used. The same mated with red males, or any of the more conservative relationships were also supported regardless of whether alternative analyses (Table S1, Supporting information), the Bayesian or maximum likelihood module of MASTER- where the frequency of red–red pairings was ‡88%. For BAYESwasused. the CERVUS analyses with all trios considered, our results Thefrequencyofyellowdorsalcolorationamongadult are robust to potential error in our estimation of the male frogs in this population is estimated to be 0.35. frequency of the yellow phenotype (q2=0.35±0.05 for Therefore, under the null hypothesis of random mating, 95%confidence and0.35±0.1for 90%confidence). (cid:2)2012BlackwellPublishingLtd 3886 C. L. RICHARDS-ZAWACKI ET AL. Table 2 Frequencies of assortative⁄disassortative mating for (a) 100 matedpairssupportedbyCERVUSwith90%confidence %) 90 Malephenotype: Binomial e ( observed(expected) exacttest nc 80 e d P-value onfi 70 c Own Other (1-tailed) e u Tr 60 Redfemale 0.91(0.65) 0.09(0.35) 0.006 (N=22) 50 Yellowfemale 0.38(0.35) 0.63(0.65) 0.572 0 2 4 6 8 10 (N=8) Opposite sex relatives of candidate parents (b) 100 The error rate for parentage assignments made by CERVUS tends to be higher than that in COLONY and MAST- %) 90 ERBAYES (Walling et al. 2010). Although we screened our nce ( 80 initial CERVUS trios for matching genotypes at an addi- de tional two loci, it is still possible that some of the trios onfi 70 c supported by CERVUS but not by other programs reflect ue this larger error rate. However, in our study, the pres- Tr 60 ence of erroneous parentage assignments among trios 50 would tend to obscure a pattern of assortative mating 0 2 4 6 8 10 rather than strengthen it. For this reason, and because Same-sex relatives of candidate parents our results were unchanged when only the subset of relationships supported by two or more pedigree (c) 100 analyses (CERVUS and COLONY or MASTERBAYES) was considered, wefeelconfident thatthe significant pattern %) 90 e ( of assortative mating seen among red females in CERVUS nc 80 e is an accurate reflection of the mating behaviour of this d population. onfi 70 c e u Tr 60 Effects of multiple matings on confidence in pedigree assignments 50 0 2 4 6 8 10 Our CERVUS analysis was conducted using the assump- Relatives of candidate offspring tion that no close relatives of the candidate parents and offspring are present in the sample. However, six indi- Fig. 3 Effect ofcloserelativesofparentsandoffspringontrue viduals that were captured as adults during our first confidence in pedigree analysis. If relatives are present in the sampling session were recaptured 18months later, data set, using the critical values from the CERVUS simulation whichindicates thatfrogsinthispopulation arecapable assuming no relatives will lead to differences between intended and true confidence. True confidence is plotted as a of living at least 2 years, and perhaps much longer. In function of the number and type of relative present. Line col- this case, our sample likely comprises more than two ourindicateswhichconfidencelevelfromtheno-relativessim- generations of frogs, and potentially several types of ulation was used for comparison: Black=95%, grey=90%. close relatives. If close relatives are present in the data Linetypeindicateswhetherthecoefficientofrelationship(r)is set, using the critical values from the CERVUS simulation 0.5 (solid) or 0.25 (dashed). The presence of half siblings assuming no relatives will lead to differences between and⁄or grandparent⁄grandchild relationships (r=0.25) has lit- assumed (90% or 95%) and true confidence in pedigree tle effect on the confidence of pedigree assignments (a–c). However, the presence of same-sex full siblings of parents (b) relationships. To estimate the impact that close relatives and full siblings of offspring (c) can lead to confidence levels would have on confidence levels for our analysis, we thataremuchlowerthanintended. conducted a series of simulations with different num- bers and types of close relatives of candidate parents and offspring. The results suggest that our confidence among the candidate parents and offspring (Fig. 3a–c) estimates are robust to the presence of relatives with and to the presence of full siblings (r=0.5) among the r=0.25 (half siblings, grandparents or grandchildren) candidate parents so long as the candidate parent and (cid:2)2012BlackwellPublishingLtd MATE CHOICE ON A HERITABLE COLOUR POLYMORPHISM 3887 its close relatives are of the opposite sex (e.g. father’s tionships and that the patterns of mate choice and sister or mother’s brother) (Fig. 3a). However, if several inheritance seen among these trios can reproduce the same-sex full siblings of the candidate parents are pres- morph frequencies we observed in a larger sample of ent (e.g. mother’s sister or father’s brother), the true the population. confidence in our pedigree assignments may be much lower than 90% or 95% (Fig. 3b). The presence of full Discussion siblings among the candidate offspring has a similarly dramatic effecton confidence (Fig. 3c). One of the central problems of speciation is the origin of isolating barriers that actually or potentially prevent gene flow in sympatry (Mayr 1942; Coyne & Orr 2004). Mode of inheritance of dorsal coloration Reproductive barriers can take many forms (e.g. pre- Observations from captive breeding of these two mor- zygotic, post-zygotic), each of which can prevent gene phs (CRZ, unpublished data) suggest that red dorsal flow in different ways. Because speciation is usually a coloration shows complete dominance of red over yel- long process during which many barriers can arise, sev- low. However, that experiment has not yet yielded suf- eral will likely be present by the time reproductive iso- ficient data to merit statistical testing. Using the lation is nearly complete (reviewed in Coyne & Orr mother–father–offspring trios identified by our pedi- 2004). The most important are those that reduce gene gree analysis, we tested the null hypothesis that dorsal flow to the greatest extent prior to complete speciation. coloration could be controlled by a single locus with A number of species show strong assortative mating in two alleles and complete dominance of red over yellow the absence of post-zygotic isolation, suggesting that (Fig. 1). Table 3 shows the observed and expected fre- divergence in mating preferences may be a common quencies of yellow offspring from different combina- first step inthe speciation process (McMillan et al. 1997; tions of parent phenotypes using the results from Seehausen et al. 1999; Jiggins & Mallet 2000; Jiggins CERVUS at 90% confidence. The observed frequencies of et al. 2004). offspring phenotypes were not significantly different The amazing colour variation found among straw- from expected, regardless of which confidence level berry dart frog populations in Panama appears to have (90% or 95%) was used, or whether trios were limited evolved recently and rapidly, suggesting the action of to those with juvenile offspring or those also supported strong selection (Brown et al. 2010). Studies of recently by COLONY and⁄or MASTERBAYES (Table S2, Supporting diverged populations, like the colour morphs of Dendro- information). This suggests that dorsal coloration in bates pumilio, provide direct insight into which barriers this population could be controlled in large part by a play a major role in reducing gene flow, leading to single locus with complete dominance of red over yel- reproductive isolation. In this study, we used a wild low. However, for some parent combinations (e.g. pedigree, estimated using molecular markers, to investi- red+yellow) and data sets (e.g. juvenile offspring), gate whether females of this species mate assortatively our sample size is small and thus the power to detect by colour to clarify the extent to which sympatric mor- a departure from Hardy–Weinberg expectations would phs are reproductively isolated from one another as a be quite low. The observed frequencies of offspring result of mating behaviour. Using simulations, we phenotypes were also not significantly different from investigated the impact of our assumption of no close expectations based on our mark–recapture study (i.e. relatives on true confidence in pedigree assignments. 35% yellow), regardless of which confidence level or We also used our pedigree to investigate the genetic subset of trios was used (Tables 3 and S2, Supporting basis for colour-pattern variation in this population, information). This suggests that our pedigree contains providing the first glimpse into the heritability of a trait a representative subset of mother–father–offspring rela- on which matechoice appears tobe based. Table 3 Comparison of offspring phe- Parents Offspring Frequencyofyellow Binomial notypes to expectations if dorsal phenotypes offspring exacttest colorationiscontrolledbyasinglelocus with the red allele being dominant over P-value yellow Phenotypes Genotypes Red Yellow Observed Expected (2-tailed) Yellow·yellow xx+xx 0 3 1.00 1.00 Red·yellow xx+X_ 4 3 0.43 0.37 0.99 Red·red X_+X_ 16 4 0.20 0.14 0.61 Total 20 10 0.33 0.35 0.99 (cid:2)2012BlackwellPublishingLtd 3888 C. L. RICHARDS-ZAWACKI ET AL. Effect of multiple matings on confidence in pedigree Preference vs. choice assignments Laboratorystudies,whichcomparethecourtshipbehav- The effect of the ‘no-relatives’ assumption on confi- iour of females towards males of different phenotypes, denceappearstodependverymuchonthetypeofrela- haveshownthatfemaleD. pumilioattendtomalecolor- tionship. Our simulations suggest that the presence of ationandprefer tocourtwithmalesoftheir owncolour half siblings, grandparents or grandchildren (r=0.25) (Summers et al. 1999; Reynolds & Fitzpatrick 2007; of candidate parents or offspring would have very little Maan & Cummings 2008, 2009; Richards-Zawacki & effect on confidence (Fig. 3). Other potential relation- Cummings 2011). This suggests the potential for ships within our study population (e.g. cousins, great- behavioural pre-zygotic isolation, caused by divergent grand parents or great-grand children) would likely female preferences, to limit gene flow among morphs. have even less effect as their coefficient of relationship However, for some species (including several frogs, see is lower (r£0.125). The presence of full siblings Gerhardt 1992), mating preferences observed in the (r=0.5), however, could cause true confidence to be laboratory are not reflected in the pattern of mating in much lower than the assumed cut-off levels for assign- wild populations. This appears also to be the case for at ment of pedigree relationships. This would be the case leastonecolourmorphofD. pumilio. if full siblings of the offspring or same-sex full siblings A previous laboratory experiment found evidence of of the candidate parents were present in the data set. assortative mating preferences for both red and yellow However, the presence of opposite-sex full siblings of females from the polymorphic population on the north- the candidate parents (e.g. sister of father or brother of west corner of the island of Bastimentos (Richards- mother) had little effect on confidence. Taken together, Zawacki & Cummings 2011). This suggests that females our simulations suggest that the presence of any full prefer males of their own colour over males of other siblings of the candidate offspring, and approximately phenotypes they would come in contact with in the one half of any full siblings of candidate parents (same- wild. The patterns of mate choice we observed in this sex siblings) present could cause true confidence in study, however, tell a slightly different story. Whereas pedigree assignments to be much lower than assumed. red females showed a significant tendency to mate But how likely is it that these types of relatives are assortativelyinthewild,yellowfemalesdidnot.Infact, presentinour dataset? the pattern of mate choice for yellow females did not Because D. pumilio has a small clutch size (avg. 4.2 differ fromexpectations for randommating. eggs, Prohl 2005) and both males and females are There are a number of potential explanations for the known to mate with multiple partners (Prohl & Hodl discrepancy between the laboratory study and this 1999), it is unlikely that large numbers of full siblings field study. For example it is possible that yellow of offspring or parents would be present within our females do mate assortatively in the field and that this sample. The high mortality of this species in early study’s small sample size (n=8 yellow females) is not development (only one out of six eggs are likely to accurately representing the behaviour of yellow survive to metamorphosis, Prohl & Hodl 1999; Maple females. It is also possible that the contrasting results 2002) and the fact that we only sampled about 30% for yellow females are attributable to differences in of the population mean that a female would have to lighting between the two studies and its effects on mate with the same male many times before we females’ perceptions of colour. We investigated this would likely sample two of their full sibling offspring possibility by plotting red and yellow dorsal colour (Fig. S2, Supporting information). Under this scenario, phenotypes and calculating colour contrasts under both even if a female were to mate with the same male 10 forest lighting and the lighting conditions used in the times, we would only have sampled, on average, two laboratory study (filtered incandescent light) using a of the resulting offspring. Our simulations suggest D. pumilio-specific visual model (details in Supporting that this would lead to a true confidence of 89% Information). Whereas laboratory lighting did generate instead of the intended 95% for assignment of a par- a slightly greater colour contrast than forest light, this ent–offspring relationship, which is still higher than shift is small in comparison to the natural variation in the assignment cut-off of 80% used in many parent- colour found among individuals of each colour morph age studies (e.g., Krutzen et al. 2004; Delgado et al. (Fig. S4, Supporting information). It therefore seems 2008). Given that this high number of matings unlikely that lighting differences can explain the between the same male and female is likely to be difference in yellow female behaviour between the two rare for D. pumilio, and the fact that it only slightly studies. lowered our confidence, we feel justified in trusting Another possible explanation for the apparent incon- our pedigree assignments. sistency could stem from the fundamental difference (cid:2)2012BlackwellPublishingLtd