Abstract. Introduction

Transcription

1 Molecular Ecology (2013) doi: /mec Genetic and morphometric evidence on a Galapagos Island exposes founder effects and diversification in the first-known (truly) feral western dog population SINI E. M. REPONEN,* SARAH K. BROWN,* BRUCE D. BARNETT and BENJAMIN N. SACKS* *Mammalian Ecology and Conservation Unit, Veterinary Genetics Laboratory, University of California, One Shields Avenue/Old Davis Road, Davis, CA 95616, USA, Department of Anthropology, University of California, One Shields Avenue, Davis, CA 95616, USA, Barnett Environmental, 5214 El Cemonte Ave., Davis, CA 95618, USA, Department of Population Health and Reproduction, University of California, One Shields Avenue/Old Davis Road, Davis, CA 95616, USA Abstract Domesticated animals that revert to a wild state can become invasive and significantly impact native biodiversity. Although dogs can be problematic locally, only the Australasian dingo is known to occur in isolation from humans. Western dogs have experienced more intense artificial selection, which potentially limits their invasiveness. However, feral dogs eradicated from Isabela Island, Galapagos in the 1980s could be the first-known exception. We used DNA and morphometric data from 92 of these dogs to test the hypotheses that (i) these dogs persisted independently of humans for up to a century and a half since descending from a handful of dogs introduced in the early 1800s, vs. (ii) similarly to other western feral dog populations, they reflected continuous recruitment of strays from human settlements on a portion of the Island. We detected one dominant maternal lineage and one dominant paternal lineage shared by the three subpopulations, along with low autosomal genetic diversity, consistent with the hypothesized common origins from a small founder population. Genetic diversity patterns among the three island subpopulations were consistent with stepping-stone founder effects, while morphometric differentiation suggested rapid phenotypic divergence, possibly due to drift and reinforced by selection corresponding to distinct microclimates and habitats on Isabela. Despite the continued presence of free-ranging dogs in the vicinity of settlements on Isabela and other Galapagos Islands, feral populations have not reestablished in remote areas since the 1980s, emphasizing the rarity of conditions necessary for feralization of modern western dogs. Keywords: de-domestication, feral dogs, founder effects, Galapagos, invasive species Received 10 September 2013; revision received 14 November 2013; accepted 15 November 2013 Introduction Domesticated animals that have reverted to a wild state can become invasive and threaten native biodiversity. However, domestic animals tend to lack adaptations of their wild ancestors and therefore often are unable to attain complete independence from humans, limiting their invasiveness (Daniels & Bekoff 1989a,b; Boitani & Ciucci 1995). Understanding the potential for domestic Correspondence: Benjamin N. Sacks, Fax: ; bnsacks@ucdavis.edu animals to become truly feral defined here as self-sustaining and independent of humans (similar to de-domesticated ; Baker & Manwell 1981) is important for predicting risks to depauperate communities, such as on islands, which often contain vulnerable endemic species that lack the antipredator defences of their mainland counterparts. Domestic dogs accompany humans to the most remote corners of the earth and can often threaten wildlife through competition, predation or disease, which can be especially devastating to insular fauna that have evolved in the absence of predators (Barnett & Rudd

2 2 S. E. M. REPONEN ET AL. 1983; Barnett 1987; Taborsky 1988; Butler et al. 2004; Vanak & Gompper 2009; Young et al. 2011; Hughes & MacDonald 2013). Among the impacts dogs can have on wildlife, predation appears to be the most widespread problem (Hughes & MacDonald 2013). The severity of predation depends primarily on anthropogenic food subsidies, which on the one hand reduce per capita predation by dogs on wild prey but on the other hand elevate dog population density (Vanak & Gompper 2009). The degree to which dogs impact wild prey also depends on the size of the prey and on the sympatric wild predator community, which can limit the dog population and, consequently, their impacts (Butler et al. 2004; Campos et al. 2007). Free-ranging dogs, however, usually depend on anthropomorphic food sources and/or recruitment of strays to maintain packs, which can restrict their geographic impacts. It is rare for domestic dogs to achieve sufficient independence from humans to invade more remote wildlands where many of the earth s most vulnerable species presently occur. Where various apparently feral dog populations have been identified, only the dingo (and perhaps some South Asian pariah dog populations) has achieved long-term ecological and demographic independence from humans (Boitani & Ciucci 1995; Vanak & Gompper 2009; Young et al. 2011). Dingoes in Australia and other parts of South-East Asia and New Guinea arose from Neolithic ancestors that were probably not domesticated to the extent of modern western dogs (Corbett 1995; Vanak & Gompper 2009). Even modern village dogs in parts of South-East Asia have largely escaped the hyperdomestication and inbreeding bottleneck associated with the post-victorian West (Irion et al. 2005; Brown et al. 2011; Sacks et al. 2013). Thus, the existence of these populations does not necessarily imply that western breed dogs can reach a similar state. Indeed, modern western dogs have peculiarities that could differentially affect their feralization potential. All western dogs purebred or mongrel reflect especially intensive selection and inbreeding over the last few hundred years that, on the one hand, could have magnified their reliance on humans and, on the other hand, created unprecedented phenotypic diversity (Parker et al. 2004; Lindblad-Toh et al. 2005; Neff & Rine 2006; Gray et al. 2009; Drake & Klingenberg 2010). Many of these distinct forms (i.e. breeds), in turn, are characterized by phenotypes coded for by single genes of large effect, or combinations (across multiple loci) of fixed genes, which when crossed, can produce novel genomic combinations with large phenotypic consequences (Gray et al. 2009; Akey et al. 2010; Boyko et al. 2010). Thus, while western dogs may retain fewer ancestral adaptations than their Eastern counterparts, which put them at a relative disadvantage in the wild, they have a genetic architecture to produce unnaturally high phenotypic diversity upon which selection can rapidly act, potentially facilitating their ability to fill otherwise empty niches. Among the small number of apparently feral (freeranging) western dog populations, those on Isabela Island in the Galapagos appear to have maintained long-term human independence (Kruuk & Snell 1981; Barnett & Rudd 1983; Phillips et al. 2012). Ephemeral feral dog populations have occurred on several of the Galapagos Islands since the early 1800s, but most of these were sustained only as long as they continued to recruit strays from the mainland or nearby human settlements (Phillips et al. 2012). Evidence reviewed in detail elsewhere (Barnett & Rudd 1983; Barnett 1986) suggests that the dogs of Isabela Island could have been an exception, reflecting a demographically expanding, phenotypically evolving and self-sustaining population. Dogs were likely first introduced to Isabela Island in 1835 the year of Darwin s visit to the archipelago after being taken to Isabela on a hunting trip and abandoned on the southern slopes of the Sierra Negro volcano by Ecuadorian General Jose Villamil. These first dogs were primarily sighthounds, but also included dogs resembling today s pointers and a couple other European hunting breeds (Barnett 1986). Although no humans permanently inhabited this island until 1897, free-ranging dogs were subsequently observed on the Island near their initial introduction point ( Sierra Negra ) and later in more remote portions of the island, including the south slope of the Cerro Azul volcano ( Cerro Azul ) and the northern coast of the island s southern portion ( North Coast ; Fig. 1). The North Coast dogs preyed extensively on marine iguanas (Amblyrhynchus cristatus; Kruuk & Snell 1981) and juvenile sea lions; the Cerro Azul dogs primarily preyed upon and scavenged feral cattle (also purportedly in this area since early human settlement of the island), and the Sierra Negra dogs primarily scavenged prey caught in snares by hunters from nearby villages (Barnett & Rudd 1983). These dogs, however, could have been merely a series of ephemeral populations deriving from periodic introductions, rather than a self-sustaining population originating over a century and a half earlier (Phillips et al. 2012). Mainland Ecuadoran dogs have been continuously imported to the coastal village of Puerto Villamil and the highland, agricultural community Santo Tomas close to the site of initial introductions since 1897 (Barnett & Rudd 1983; Barnett 1986; Gingerich et al. 2010). It is possible that subsequently imported dogs were essential to the persistence of free-ranging populations on the Island, either through augmentation or continual replacement.

3 FERALIZATION OF DOGS IN THE GAL APAGOS 3 North.. Puerto Villamil Santo Tomás achieved demographic growth without drawing significantly from human-associated domestic dog populations. On the other hand, if these dogs reflected multiple sources (augmented, multiple introductions hypothesis), we predicted that the population would be composed of many genetic variants, possibly with differing frequencies on different parts of the island. Materials and methods Study area Fig. 1 Map of the study area on southern Isabela Island, Galapagos, illustrating locations of two villages, the North Coast, Cerro Azul and Sierra Negra dog populations prior to their eradication, and the expanse of lava thought to restrict dog movement. Black arrows indicate hypothetical sequence of stepping-stone founder populations. Barnett observed that the three subpopulations described above differed phenotypically and ecologically, which, if the result of evolution from a single ancestral introduction, suggests the potential for rapid divergence either through drift alone or potentially reinforced by local adaptation. Alternatively, phenotypic differentiation between these island populations could reflect distinct introductions of different breeds or breed mixtures. The current study tested these hypotheses (single-origin vs. augmented, multiple introductions) and described feral subpopulations in terms of genetic, demographic and phenotypic characteristics. Because of their immense impact on endemic species through predation and disease, the dogs were eradicated in the early 1980s using Compound 1080 (sodium monofluoroacetate) poison bait stations (Barnett 1986, 1987). Ninety-two carcasses of the feral dogs removed from the Island were collected and used in the present study. Based on genetic markers, the single-origin hypothesis predicts that the subpopulations would be dominated by a small number of shared maternal and paternal haplotypes and low nuclear and mitochondrial genetic diversity (e.g. similarly to Australian dingoes, Savolainen et al. 2004; Sacks et al. 2013). Moreover, the haplotypes present in these dogs should be consistent with those found in the putative founder breeds. Given that hundreds of dogs had become established in remote portions of Isabela Island at the time of their eradication in the early 1980s, low genetic diversity and shared alleles would imply also that this population The Sierra Negra subpopulation occurred on the western slope of the Sierra Negra volcano near the island s only two human villages (Fig. 1). This subpopulation was the one most likely to reflect some genetic admixture from physically dissimilar dogs more recently imported to the island communities from the South American mainland. The other two subpopulations occurred in remote parts of the Island geographically close to one another but corresponding to very different habitats, prey communities and microclimate. The rugged terrain separating the Sierra Negra and Cerro Azul subpopulations consisted of a 16-km-wide expanse of lightly vegetated Pahoehoe lava called El Quemado and a narrow band of older, more vegetated, less daunting lava along the island s southwestern coast. Dogs must have either traversed this terrain at least once for initial population establishment or were independently introduced to those areas. The Cerro Azul and North Coast subpopulations were also separated by an ~10-km-wide lava field, except for a narrow, vegetated passage along the northwestern coast near Punta Cristobal. And the Sierra Negra and North Coast populations were separated by a 15- to 20-km lava flow on the volcano s northern slope. These subpopulations also occupied very different habitats, exploited very different prey and were exposed to distinct microclimates. The North Coast population ranged over open lava fields with high daily temperatures, arguably favouring the predominantly short-haired and light-coloured individuals with large erect ears (Barnett 1986). In contrast, the climate on the south-facing slopes of Cerro Azul and Sierra Negra was cooler and wetter and supported large numbers of feral hoof-stock. Samples and approach Intensive invasive predator control programmes initiated in the late 1970s led to the eradication of the feral dog populations on Isabela by the early 1980s, limiting research findings to a few localized behavioural observations, morphometric measurements and research into disease prevalence and transmission (Barnett & Rudd

4 4 S. E. M. REPONEN ET AL. 1983; Barnett 1987). The genetic material used for the present analysis was extracted from the skulls of 92 dogs collected from the three island subpopulations during the eradication of : 41 dogs from North Coast, 16 from Cerro Azul and 35 from Sierra Negra. Estimation of age (see below), morphometric measurements and coat-colour descriptions were recorded at the time of collection in 1980 and Only the skulls of these dogs, maintained in a collection subjected to ambient climate changes for 30 years, provided analysable DNA material. The storage site in Davis, California (N38.55, W ), experienced hot, dry summers and cool, wet winters, fluctuations which were expected to reduce DNA quality and yield. Therefore, we treated the samples with the same stringency as we would treat ancient DNA samples (Brown et al. 2013). Laboratory procedures To avoid and detect contamination and errors due to postmortem DNA degradation, we applied strict ancient DNA protocols, including use of a dedicated ancient DNA facility, negative controls at all steps and replication of extractions and polymerase chain reactions (PCR). We attempted to extract DNA from 250 to 500 mg of tooth material in sets of nine samples and one extraction blank using a phenol chloroform procedure (Sambrook & Russell 2001). We attempted to amplify via PCR a 582-bp fragment of the hypervariable region I of the mitochondrial genome in 1 3 overlapping fragments using published primers (Savolainen et al. 2002; Brown et al. 2011). Due to limitations on the amount of DNA available, we opted not to attempt smaller fragments (e.g. as in Brown et al. 2013). We conducted PCR and sequencing as described previously (Brown et al. 2011, 2013). We edited and aligned sequences manually in SEQUENCHER â 5.0 DNA software (Gene Codes Corporation, Ann Arbor, MI, USA). We compared sequences to published dog haplotypes using DNACOLLAPSER v1.2 (Villesen 2007) and to those accessioned in GenBank via Basic Local Alignment Search Tool (BLAST). To quantify nuclear DNA, we used quantitative realtime PCR (qpcr) to amplify a canine-specific nuclear gene (Mc1r) and included internal positive controls in each sample to test for PCR inhibitors (Evans et al. 2007). We attempted to genotype five dinucleotiderepeat microsatellites from the nonrecombining region of the Y chromosome: MS34A, , , MS41B, and as described previously (Sacks et al. 2013). We also genotyped one sample corresponding to each potentially distinct Y microsatellite haplotype at each of 29 NRY SNPs identified by Ding et al. (2011). We used iplex Sequenom MassARRAY system (Sequenom Inc., San Diego, CA, USA) as described by Sacks et al. (2013), except that we pre-amplified the samples and used a 50-fold dilution of the pre-amplification product as template for the reaction. We PCR-amplified 12 autosomal microsatellite loci in two multiplexes as described previously (Moore et al. 2009). One multiplex included loci REN54P11, AHTh171, CPH18, CXX468 and CXX602(dog), and another included loci AHT137, AHT142, CXX279, CXX374, INU055, REN162C04, REN169O18. We performed PCR reactions in 20 ll volumes using Qiagen Multiplex PCR reagents (Qiagen Sciences Inc., Germantown, MD, USA) according to the manufacturer s guidelines with Q-solution, along with 2 ll of DNA template and 14 ll of Chill-out (Bio-Rad Laboratories Inc., Hercules, CA, USA) per reaction. We attempted 3 10 independent PCR amplifications for each sample, depending on the amount of DNA available. Genetic data analysis We used mitochondrial and Y chromosome haplotypes to determine minimum numbers of maternal and paternal founders within and among the Isabela Island subpopulations and compared them to worldwide databases (Savolainen et al. 2002; Pang et al. 2009; Brown et al. 2011; S. K. Brown & B. N. Sacks, unpublished) to assess consistency with the particular putative founding breeds. For autosomal microsatellites, we estimated allelicdropout and false-allele error rates through comparisons between replicate genotypes of the same DNA samples (Creel et al. 2003). This approach was conservative because estimates applied to a single genotyping attempt, whereas the final genotypes used in our analyses were based on a minimum of three replicate genotypes. We used Genepop to assess Hardy Weinberg equilibrium and estimate F IS (Raymond & Rousset 1995) and tested significance based on permutation tests in FSTAT (Goudet 1995). We investigated population genetic structure based on allele-frequency differences between populations using an estimator of F ST (Weir & Cockerham 1984) and evaluated statistical significance using permutation tests in FSTAT followed by Bonferroni adjustments for multiple comparisons (Goudet 1995). To account for uneven samples sizes among subpopulations, we used HP-Rare to estimate allelic richness and private allelic richness rarified to the smallest sample size (Kalinowsi 2005). To detect substructure within subpopulations, we used a Bayesian-model-based clustering method implemented in program STRUCTURE v (Pritchard et al. 2000). We used the correlated allele frequencies model (Falush et al. 2003) and

5 FERALIZATION OF DOGS IN THE GAL APAGOS 5 discarded the first iterations as burn-in, basing estimates of ancestry on the subsequent iterations. We performed this procedure 10 times each for a range of K (number of clusters assumed). To assess relatedness among individuals within clusters, we employed a maximum-likelihood estimator of the proportion of alleles shared by descent, implemented in ML Relate (Kalinowski et al. 2006). Demographic and morphometric analysis We aged tooth samples to the nearest year by sectioning, staining and enumerating cementum annuli from a lower canine root (Linhart & Knowlton 1967), except for those from juvenile dogs with deciduous teeth or incomplete root closure. We estimated survival rate by regressing the natural logarithm of the number of individuals sampled on year class and then exponentiating the slope of the regression (Krebs 1999). The 95% confidence interval was calculated by exponentiating the confidence limits, which were calculated according to the z-distribution and the standard error of the estimate of the slope (Zar 1999). To qualitatively assess whether populations differed morphologically, we first conducted a principal components (PC) analysis to reduce the 22 skull measurements to orthogonal multivariate shape parameters (i.e. PCs) and then conducted a multivariate analysis of variance (MANOVA) to assess whether PCs differed significantly among populations based on the Wilk s Lambda test (Tabachnick & Fidell 2001). We then performed discriminant analyses (DA) to characterize interpopulation differences in morphology based, first, on the 22 skull measurements and, second (for the subset of individuals for which we had body measurements), all 29 skull and body measurements (Table 1). Only adults were used in morphometric analyses. All variables were inspected for deviations from normality based on a normal probability plot and determined to conform well to normal expectations. Therefore, multivariate normality was assumed; all variables were log-transformed to minimize heteroscedasticity and standardized prior to entering into the initial PCAs and DAs. Next, to derive models based only on the most informative measurements, we performed a stepwise discriminant analysis, where variables were added one at a time in order of highest to lowest F-ratios (Tabachnick & Fidell 2001). Variables were removed if their F-ratios fell <3.9 (F-to-remove) and included if their F-ratio exceeded 4.0 (F-to-enter). Pelage characteristics The coat colours of dogs were recorded and fell into one of seven categories: brindle, tan, white, black, white with black spots, white with tan spots and white with brindle spots. We report all data, but due to low sample size in some categories, for statistical comparisons among populations, we collapsed white coat colours (i.e. regardless of spots) to a single category. Differences in frequencies were assessed with a chi-square test of independence. Coat-colour diversity was estimated for each population as 1 minus the sum of squared frequencies (of the four main coat-colour types). Except as noted above for genetic data, statistical analyses were conducted in SYSTAT (v. 9.0). Results Mitochondrial DNA None of the extraction blanks contained amplified DNA, indicating no detectable contamination. We successfully sequenced mtdna fragments from 54 of the 92 dogs, of which 50 shared identical sequences along fragments ranging bp (Table 2). Based on the full 582-bp sequences (n = 23), these were previously identified as A22 (Savolainen et al. 2002). The partial fragments were missing portions that distinguished A22 from only two other haplotypes, A23 and A163, each of which was described previously in a single dog (<0.001 frequency). All individuals in the North Coast (28 dogs) and Cerro Azul (six dogs) populations carried the A22 (or presumptive A22) haplotype, as did 16 of the 20 Sierra Negra dogs. In addition, the Sierra Negra population contained one individual carrying an A17 haplotype and three individuals carrying a haplotype matching B1, B14 or B21 (Table 2). The dominant haplotype (A22) was consistent with sighthound ancestry. In a worldwide survey of mtdna in >2000 dogs (Savolainen et al. 2002; Pang et al. 2009; Brown et al. 2011), the A22 haplotype was found in only 14 of 234 breeds (6%) surveyed overall, yet this 6% included half (n = 6) of 12 sighthound breeds in that survey, which is a highly significant association (Yates corrected v 2 1 = 35.7, P << ). Moreover, the particular sighthound breeds carrying the A22 haplotype (Sloughi, Greyhound, Saluki, Sica, Irish Wolfhound, Scottish deerhound) were geographically widespread in origin, implying ancient incorporation of this haplotype into the group. Y chromosome DNA The qpcr assay indicated the presence of quantifiable nuclear DNA in 38 of the extracts, including 24 males (10 from the North Coast, five from Cerro Azul and nine from Sierra Negra). Of these, 14 individuals provided at least partial Y microsatellite haplotypes

6 6 S. E. M. REPONEN ET AL. Table 1 Abbreviations, descriptions and references for the 29 morphological measurements taken for analysis of interpopulation differences Abbreviation Description References Cranial L1 Condylobasal length Jolicouer (1959) L2 Palatal length Jolicouer (1959) L3 Postpalatal length Jolicouer (1959) L4 Cranial length Stockard (1941) L5 Nasal length Stockard (1941) J1 Minimum height of jugal at right angles to axis of bone Lawrence & Bossert C1 Least width of the braincase Jolicouer (1959) C2 Width between the auditory bullae Jolicouer (1959) C3 Maximum breadth of braincase at parieto-temporal suture Lawrence & Bossert O1 Minimum distance from alveolus of M2 to depression anterior to bulla at base of styloid process Lawrence & Bossert O2 Minimum vertical distance from alveolar margin of M1 to base of orbit Lawrence & Bossert T1 Alveolar length of the upper carnassial Jolicouer (1959) T2 Crown length of first upper molar Jolicouer (1959) T3 Crown length of upper cheekteeth from canine to M2 Lawrence & Bossert T4 Crown width across upper incisors Lawrence & Bossert T5 Maximum antero-posterior width of upper canine taken at base of enamel Lawrence & Bossert T6 Minimum width of crown of P4 taken between roots Lawrence & Bossert W1 Zygomatic width Jolicouer (1959) W2 Palatal width outside first upper molars Jolicouer (1959) W3 Palatal width outside second upper premolars Jolicouer (1959) W4 Width between postglenoid foramina Jolicouer (1959) W5 Interorbital width Jolicouer (1959) Full body BL1 Body length H1 Front height H2 Rear height BL2 Tail length BL3 Ear length BL4 Hind foot length WT Body mass (Table 3). All but one Y microsatellite haplotype matched the 9l haplotype in haplogroup H23 (Brown et al. 2011; Sacks et al. 2013). However, one individual was typed for an allele at locus MS41B that differed by two repeats from the others. Because the other three loci that were successfully genotyped in this sample matched the 9l haplotype and because MS41B was one of the two most mutable loci (the other being 79.3), it is likely that the novel mutation(s) at this locus was (were) recently derived from the same patriline as the other dogs, rather than reflecting an independent ancestral lineage. The primary Y chromosome haplotype, H23-9l, was consistent with sighthound ancestry. Based on a previous survey of 102 purebred dogs from 35 breeds and 300 South-East Asian and Middle Eastern village dogs, this haplotype was found only in two (of 136) Middle Eastern village dogs (Brown et al. 2011; Sacks et al. 2013). The only breeds of the 35 surveyed with haplotypes clustering close to 9l, and with Middle Eastern village dogs, generally, were the Saluki and Afghan hound, which were also the only sighthound breeds in the survey (Brown et al. 2011). Autosomal microsatellites We obtained autosomal microsatellite genotypes consisting of 7 12 loci for 30 individuals, based on 3 7

7 FERALIZATION OF DOGS IN THE GAL APAGOS 7 replicates each, depending on available DNA (Appendix S1, Supporting information). The observed genotyping error rate was 0.02 allelic dropouts per locus and no false alleles. Based on a binomial approximation (e.g. McKelvey & Schwartz 2004), this error rate implied a probability >99% that two multilocus genotypes of the same sample would agree at >85% of their allele calls. Conversely, pairwise comparisons between multilocus genotypes of different individuals revealed <5% of comparisons to match at 85% of loci. Thus, despite the low quantity of DNA present generally in these samples, the accuracy of resulting genotypes was reasonably high. Table 2 Frequencies of three mtdna haplotypes observed in 54 dogs from three Isabela Island populations Population Mitochondrial DNA haplotypes A22 A17 B1 North Coast 28 * 0 0 Cerro Azul Sierra Negra *23 Individuals had incomplete sequences, matching haplotypes A22, A23 and A Individuals had incomplete sequences, matching haplotypes A22 and A Individuals had incomplete sequences, matching haplotypes A22, A23 and A163. Sequences were incomplete, matching B1, B14 and B21. Allelic diversity was low in all subpopulations, but especially in the two remote subpopulations (Table 4), with the greatest number of alleles and the highest heterozygosity found in the Sierra Negra subpopulation. Importantly, no locus contained more than three alleles in the two remote subpopulations (pooled or separately), consistent with origin from as few founders as a single pregnant female. The only locus to show statistically significant heterozygote deficiency was CXX279 in the Sierra Negra subpopulation. The F IS value also was highest in this subpopulation, indicating heterozygote deficiency apparently associated with substructure or admixture. All three subpopulations shared 1 3 alleles at each locus and the allelic diversities of both remote subpopulations were almost completely included within the allelic diversity observed in the Sierra Negra subpopulation, which had nearly as many private alleles as were shared with the two remote subpopulations (Fig. 2). Notably, the sample size of the Sierra Negra subpopulation was little more than half that for the other two combined, despite its having approximately twice the number of alleles, suggesting the likely presence of additional, undiscovered alleles in the Sierra Negra subpopulation. The indexes rarified to the lowest sample size confirmed these patterns: allelic richness indicated higher allelic diversity in the Sierra Negra subpopulation, and private allelic richness was many times greater for the Sierra Negra population than the other two, consistent with the latter being subsets of the former (Table 4). Table 3 Putative Y chromosome (Y STR and Y SNP) haplotypes of 14 males among 38 dogs with detectable amounts of nuclear DNA ID Population MS34A MS41B Y STR* Y SNP* S North Coast l H23 S North Coast l (H23) S North Coast l H23 S Cerro Azul l (H23) S Cerro Azul (9l + 2) H23 S Cerro Azul l (H23) S Sierra Negra l H23 S Sierra Negra (9l) (H23) S Sierra Negra (9l) (H23) S Sierra Negra (9l) (H23) S Sierra Negra (9l) (H23) S Sierra Negra l (H23) S Sierra Negra 126 (9l) (H23) S Sierra Negra (9l) (H23) *The STR haplotypes in parentheses were presumed based on partial haplotypes that completely matched the dominant haplotype. The SNP haplotypes were only tested on the two samples that differed in at least one of the STR loci and two samples that had full Y STR profiles. Others (in parentheses) were presumed based on their matching STR alleles. The STR haplotype (9l + 2) indicates a haplotype matching 9l at three loci and differing by two steps at locus MS41B. The discrepant allele could not be replicated in two subsequent attempts.

8 8 S. E. M. REPONEN ET AL. Table 4 Genetic diversity statistics, average number of alleles per locus (N A ), allelic richness (AR) and private allelic richness (PAR) rarified to three genotypes per locus, expected heterozygosity (H e ), observed heterozygosity (H o ) and inbreeding coefficient (F IS ) estimated from 30 dogs with 7 12 loci Population n N A AR Range N A PAR H e H o F IS North Coast (0.24) 1.77 (0.21) (0.09) 0.41 (0.06) 0.32 (0.04) 0.20* (0.09) Cerro Azul (0.22) 1.92 (0.15) (0.05) 0.30 (0.07) 0.20 (0.05) 0.25* (0.12) Sierra Negra (0.36) 2.84 (0.19) (0.17) 0.61 (0.05) 0.37 (0.05) 0.37* (0.08) Standard errors are in parentheses. *P < 0.05, based on Bonferroni-corrected permutation test in FSTAT. 1 1 Table 5 Pairwise genetic differentiation (F ST ) and 95% confidence intervals between populations Population North Coast Cerro Azul North Coast Cerro Azul Sierra Negra ( 0.04 to 0.13) ( ) 0.13 ( ) 11 Sierra Negra 22 Samples sizes are given in the diagonal. No pairwise differences were statistically significant after Bonferroni correction. Sierra Negra Fig. 2 Venn diagram illustrating sharing of 50 autosomal microsatellite alleles among three dog populations, North Coast (n = 13), Cerro Azul (n = 6), and Sierra Negra (n = 11), on Isabela Island, Galapagos. Tests of pairwise genetic differentiation among the three subpopulations indicated greater genetic differentiation between the Sierra Negra and both remote subpopulations than between the two remote subpopulations (Table 5). Although none of these pairwise comparisons was statistically significant after Bonferroni correction, allele frequencies differed significantly (i.e., after Bonferroni correction) between Sierra Negra and the two remote subpopulations pooled (F ST = 0.12; 95% CI = ; P = 0.05). Correspondingly, the number of genetic clusters to best fit the data was K = 2, which grouped North Coast and Cerro Azul subpopulations together along with three of the 11 Sierra Negra dogs (Figs 3 and S1, Supporting information). The increase to K = 3 was not accompanied by a substantial increase in likelihood and, more importantly, resulted only in fractionating individuals rather than explaining substructure among them (Fig. 3). The remainder of the Sierra Negra subpopulation formed a distinct cluster at K = 2. The eight individuals forming the distinct cluster in the Sierra Negra subpopulation exhibited an average pairwise relatedness of 0.08, indicating that the substructure within this subpopulation was not due to sampling of close relatives. These findings suggest that the three Sierra Negra individuals clustering with the remote subpopulations were first-generation immigrants from one of the remote subpopulations. Demographic characteristics Eighty-five of the 92 carcasses recovered were successfully aged to year class (Appendix S2, Supporting information). There were no significant differences (v 2 2 = 0.13, P = 0.93) in the sex ratios among the three subpopulations, which collectively contained 34 F and 51 M. The overall sex ratio, 0.67 F:1 M, was significantly male biased (Binomial test, P = 0.016). There were no significant differences in the age-class ratios (<1 vs. >1 year) of the three subpopulations (v 2 2 = 1.12, P = 0.57). Pooling data across subpopulations indicated a range in age to 10 years, with modes at 0 and 3 years old (Fig. 4). Other than the deficiency of 1- and 2-yearolds, the age distribution suggested that survival and fecundity rates had been relatively stable at least for 7 of the 10 years reflected in this cross-sectional sample. The deficiency of 1- and 2-year-olds reflected poor

9 FERALIZATION OF DOGS IN THE GAL APAGOS 9 mtdna A22 B-1 Y chromosome 9l 9l+2 Ln Pr(X K) K North Coast Cerro Azul Sierra Negra Fig. 3 Individual-level ancestry assignments based on autosomal microsatellite Bayesian cluster analysis and corresponding mtdna and Y STR haplotypes. Upper right inset shows the average and SD among 10 runs of the logarithm of the probability of the data as a function of the number of clusters (K). K = 2 K = 3 S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S recruitment just prior to and during the 2(+)-year collection period, possibly related to disproportionate vulnerability of juveniles to the poison baits (e.g. Sacks et al. 1999), which were initially distributed 6 months prior to beginning collections for this study. An estimate of adult survival based only on individuals 3 years old was 0.79 (95% CI = ). Morphometric analysis The PCA using 22 skull measurements from all 74 adults resulted in four PCs that explained 75% of the variance in the measurements (e.g. 48% PC1, 13% PC2; Appendix S2, Supporting information). Based on the MANOVA, phenotypic differences distinguished all three subpopulations (multivariate F 8,136 = 22.4, P < ). Specifically, PC2 distinguished North Coast (F 2,71 = 69.66, P < 0.001), whereas PC1 scores distinguished Sierra Negra (although not significantly; F 2,71 = 1.67, P = 0.21). The discriminant analysis using 22 skull measurements from all 74 adults was highly significant (F 44,100 = 7.16, P < ; Fig. 5A), with jackknife crossvalidation correctly classifying 100%, 79% and 74% of dogs from North Coast, Cerro Azul and Sierra Negra, respectively. The stepwise analysis, which identified five variables (Table 6), was highly significant (F 10,134 = 28.8, P < ; Fig. 5B). Although this analysis revealed less separation than the analysis using all variables, classification accuracy was higher, with jackknife cross-validation indicating 97%, 86% and 85% No. individuals F Age (Years, No. cemetum annuli) Fig. 4 Age and sex distribution for 85 feral dogs collected , Isabela Island, Galapagos. correctly classified in North Coast, Cerro Azul and Sierra Negra subpopulations, respectively. The discriminant analysis including all 29 variables in the subset of 56 adult dogs with complete measurement data sets was also highly significant (F 58,48 = 4.120, P < ), as was the stepwise analysis (F 16,90 = 17.4, P < ). The classification accuracy based on jackknife cross-validation in the stepwise analysis was similar to that using only skull measurements above, with 97%, 93% and 78% correct classifications in North Coast, Cerro Azul and Sierra Negra, respectively. This M

10 10 S. E. M. REPONEN ET AL. Table 6 Discriminant function coefficients associated with eight variables used to differentiate three populations of feral dogs on Isabela Island, Galapagos Variable 22 Skull measurements All 29 measurements DF1 DF2 DF1 DF2 L O L L C BL H BL Table 7 Coat color frequencies of 90 feral dogs on Isabela Island, Galapagos Population* Coat color NC CA SN Brown brindle Tan Black White No spots Black spots Brindle spots Tan spots *North Coast (NC), Cerro Azul (CA), and Sierra Negra (SN). (A) Discriminant function 2 All 22 skull measurements CA NC SN but smaller in terms of postpalatal length, cranial length, body length and braincase width, whereas discriminant function 2, which differentiated Sierra Negra from the other two subpopulations, indicated that the Sierra Negra dogs were larger in terms of condylobasal length and shoulder height and smaller in terms of maxillary depth, ear length, postpalatal length. Coat-colour comparison (B) Discriminant function 2 Discriminant function 1 The 5 significant skull measurements Discriminant function 1 Fig. 5 Morphometric separation of 74 adult feral dogs among three subpopulations on Isabela Island, Galapagos along two discriminant functions based on (A) all 22 skull measurements and (B) five statistically significant measurements selected in the final discriminant analysis. Samples sizes were 16, 33 and 25 dogs from Cerro Azul (CA), North Coast (NC) and Sierra Negra (SN), respectively. Filled triangles represent three SN dogs with genetic assignments consistent with the other two subpopulations. analysis identified eight variables as most influential, including the same five skull measurements in addition to three body length measurements (Table 6). Discriminant function 1, which differentiated North Coast from the other two subpopulations, indicated that North Coast dogs were larger in terms of condylobasal length, CA NC SN Coat-colour descriptions were obtained for 90 dogs from the three subpopulations (Table 7; Appendix S2, Supporting information). Frequencies of coat colours differed between subpopulations (v 2 6 = 35.85, P < ). Diversity of the four basic coat colours (brindle, tan, black, white) was highest in Sierra Negra (0.74), followed by Cerro Azul (0.29), and North Coast (0.19). Discussion Were the Isabela Island dogs self-sustaining and independent of humans? The findings in this study provide strong evidence that the dogs eradicated from Isabela Island in the early 1980s were the product of a single founding population and up to a century and a half of persistence and evolution in the absence of human influence. Maternally, paternally and biparentally inherited genetic markers were consistent with the founding of the remote subpopulations by as few as a single breeding pair (or pregnant female). Given the size of these subpopulations at the time of eradication (several hundred individuals), the initial establishment must have been many generations in the past. Even if the population grew immediately from a founding pair by a doubling every generation an unrealistically high population growth rate a minimum of nine generations (~30 years) would

11 FERALIZATION OF DOGS IN THE GAL APAGOS 11 have been required to produce several hundred dogs. Other studies of free-ranging dog populations indicated very low juvenile survival and little to no reproduction from individuals born in the wild (Boitani & Ciucci 1995; Vanak & Gompper 2009). Therefore, the Isabela population more likely required many more than nine generations to reach its ultimate size. Additionally, both the maternal and paternal haplotypes were consistent with the first breeds reportedly introduced to Isabela over 150 years earlier (although we had no information on the breeds subsequently brought to the Island). No previous study has demonstrated the existence of a truly feral (i.e. wild) dog population derived from western breeds. In-depth studies investigating canid behavioural ecology and demographics throughout Europe and the Americas have invariably found that domestic dog populations were unable to sustain themselves through reproduction due to high mortality of pups born in the wild, often attributable to the lack of subterranean dens, seasonal synchrony of oestrus and parturition to seasonal conditions, biparental care or other adaptations characteristic of wild canids (Daniels & Bekoff 1989a,b; Boitani & Ciucci 1995). In most cases, dog populations also appear to be poor hunters of large prey relative to their wolf ancestors, due to lacking the proportionally large jaws, carnassials and temporalis muscles necessary to efficiently bring down large ungulates (Boitani & Ciucci 1995). In most cases, free-ranging dogs depend on humans through scavenging, although a few examples have described dogs capable of hunting and surviving off small mammalian prey (Campos et al. 2007). Founder effects, genetic structure and phenotypic divergence Our findings also corroborate previous hypotheses about the spread of dogs on the Island (Barnett & Rudd 1983; Barnett 1986). In particular, the subpopulations were thought to have stemmed from the initial establishment of a small number of dogs introduced to the southeast portion of the Island, ultimately seeding the Cerro Azul and North Coast subpopulations as later subsets of this initial introduction. Consistent with an initial founding from a small number of dogs, the genetic diversity of all three subpopulations at all three marker types was generally low compared with that seen in other free-ranging or purebred dog populations. For example, compared with the 19 dogs of the two remote subpopulations in this study (which exhibited 1 3 alleles per locus), village dogs from each of six Islands in South-East Asia exhibited alleles per locus in the same microsatellite loci (Pedersen et al. 2013). Although samples sizes were larger in most of those village dog populations, expected heterozygosity, which is not biased by sample size, also was much higher in the South-East Asian village dog populations (H e = ) than in the present study (H e = ). The numbers of uniparentally inherited haplotypes in the six South-East Asian Island village dog populations also were much higher than in the present study (where a single haplotype of each paternal lineage dominated), ranging from 8 to 16 Y chromosome haplotypes and 7 to 18 mitochondrial haplotypes, and Australian dingoes had 20 Y chromosome and 11 mitochondrial haplotypes (Brown et al. 2011; Sacks et al. 2013). Similarly, the number of alleles and heterozygosity estimates among five dog breeds on the Canary Islands ranged and , respectively, albeit based on different loci, and each breed had an average of 5.4 mitochondrial haplotypes and 16 in total (Suarez et al. 2013). Purebred dogs tend to have low genetic diversity due to inbreeding, but nevertheless tend to exhibit higher diversity than we observed in our study. For example, eight breeds genotyped with the same loci used in the present study, with sample sizes as low as 11 dogs, exhibited a range of alleles per locus, H e estimates ranging , and 4 7 mitochondrial haplotypes (Pedersen et al. 2013). The Y chromosome diversity, however, was especially low in the purebred dogs (1 or 2 haplotypes per breed) due to controlled breeding using a small pool of males (Sundqvist et al. 2001; Pedersen et al. 2013). The low genetic diversity of the Sierra Negra subpopulation was especially noteworthy given its proximity to dogs associated with the island s villages (which themselves received many distinct importations from the mainland; Gingerich et al. 2010). Nevertheless, the genetic diversity of the two remote subpopulations was even lower (substantially so) than the Sierra Negra subpopulation. Finally, the alleles in the remote subpopulations comprised a subset of those in the Sierra Negra population, consistent with the former having been derived from the latter. Although we observed no significant difference in genetic diversity or allele frequencies between the two remote subpopulations, the clear phenotypic difference between these subpopulations suggested they were at least partially reproductively isolated. Due to our small samples size, we had insufficient statistical power to detect even moderate genetic differentiation between the two remote populations. For example, although our F ST estimate did not differ significantly from zero, the upper end of the 95% confidence interval was 0.13, indicating that we could have missed detection of a substantial genetic subdivision. The landscape separating the two remote subpopulations was likely to present a significant, albeit permeable, barrier to movement. Except for a narrow vegetated corridor, along the coast,

12 12 S. E. M. REPONEN ET AL. the regions occupied by these subpopulations were separated by rugged lava. Assuming there was some degree of genetic isolation between the two remote populations, the observed phenotypic differentiation could have evolved primarily through genetic drift. Phenotypic differentiation could also have been reinforced by local selection, especially within the North Coast subpopulation. This subpopulation inhabited the most extreme environment and seemed especially consistent in their physical features and well suited to their harsh environment (Barnett & Rudd 1983; Barnett 1986). In particular, these dogs had large hairless ears and bellies, light coat-colour, long legs and thick footpads (this study; B. Barnett, unpublished), which likely conferred thermal advantages in the extreme heat intensely reflected from the dark volcanic substrate along this narrow equatorial coastline. Regardless of whether due to drift, selection or a combination of the two, the relatively short time frame for the evolution of the phenotypic differentiation of the two remote subpopulations was remarkable. Rapid evolution of these subpopulations could have been facilitated by the unique domestication history of western breed dogs. Although we observed low allelic diversity within selectively neutral loci, the extreme inbreeding that created western breeds during the post-victorian era artificially inflated frequencies of several genes and gene combinations (across loci) of large effect that are associated with the phenotypic diversity among breeds (Gray et al. 2009; Akey et al. 2010; Boyko et al. 2010). Although locally adapted gene complexes require long periods of isolation to develop in wild populations, these can be rapidly constructed through artificial selection. The creation of mongrel populations from differentially purebred dogs entails transfer of large linkage blocks and gametic disequilibrium during initial generations (Gray et al. 2009), which can mimic patterns associated with novel forms arising from hybridizing species (Arnold & Martin 2010). Specifically, admixture between genomes that have been selected (naturally or artificially) to the point where functional combinations of genes are homogenous within lines/species, but distinct between them can provide considerable phenotypic diversity as grist for either chance differentiation or selection, even when the allelic diversity measured within loci is low. The Sierra Negra dogs appeared to be ecologically intermediate between the free-ranging dogs of the villages and feral dogs of the more remote subpopulations. On the one hand, this subpopulation, which pre-dated the Villamil and Santo Tomas communities, became at least commensal with the human populations. In contrast to the remote dogs, which acted more as true predators, the Sierra Negra dogs obtained their food primarily by scavenging from village hunters traps (Barnett & Rudd 1983). Although the opportunistic use of human food sources does not necessarily imply dependence on them, our detection of immigration into this subpopulation from the more remote subpopulation, but not the reverse, suggests that Sierra Negra dogs were less inclined to leave this humandominated region where energy expenditure on food acquisition was limited. On the other hand, interbreeding between the Sierra Negra and free-ranging village dogs appeared limited, suggesting the possibility of some form of behavioural or ecological reproductive barrier (e.g. Sacks et al. 2011). Observations of smaller or female village dogs being killed by feral dogs of Cerro Azul indicated that social barriers could be significant (B. D. Barnett, unpublished). The higher morphological variability of Sierra Negra dogs compared with those in the more remote subpopulations was initially thought to reflect augmentation from dogs more recently imported from the South American mainland (Barnett & Rudd 1983; Barnett 1986). However, as discussed above, the higher phenotypic as well as genetic diversity in the Sierra Negra subpopulation was equally consistent with its being the first population from which the remote subpopulations were founded. The high frequency (80%) of the dominant mitochondrial and Y chromosome haplotypes shared with the remote subpopulations, as well as the lack of visible substructure in the admixture analysis (i.e. excluding the three individuals assigned to the remote subpopulations), suggests little to no admixture with village dogs. Thus, despite the proximity of the Sierra Negra dogs to humans and village dogs, the former seemed to share more, if not all, their ancestry with the remote subpopulations. Conclusions The demonstration of a truly feral population of dogs, as defined here, implying as a self-sustaining and humanindependent population, has important conservation implications. The rarity of truly feral dog populations generally appears to stem from low juvenile survival, low recruitment and the loss of adaptations associated with acquisition of wild prey and competition with wild predators (Daniels & Bekoff 1989a,b; Boitani & Ciucci 1995). Because we lacked direct information on the reproductive rates of the dogs in our study, it was unclear whether juvenile survival and recruitment were any higher than has been typically observed in other free-ranging dog studies. The high adult survival inferred from the age distribution suggests, however, that it need not necessarily have been. For example, application of Lotka Euler equation (Kot 2001) indicates that, for a stable population (not growing or declining),