The evolution of South American endemic canids: a history of rapid diversification and morphological parallelism

Authors


Carlos G. Schrago, Departamento de Genética, A2-097, Universidade Federal do Rio de Janeiro, CCS, IB, Rua Prof. Rodolpho Paulo Rocco, S/N, Cidade Universitária, PO Box 68011, Rio de Janeiro, RJ, CEP 21.941-617, Brazil. Tel.: +55 21 2562 6397; fax: +55 21 2562 6333; e-mail: guerra@biologia.ufrj.br

Abstract

The origin of endemic South American canid fauna has been traditionally linked with the rise of the Isthmus of Panama, suggesting that diversification of the dog fauna on this continent occurred very rapidly. Nevertheless, despite its obvious biogeographic appeal, the tempo of Canid evolution in South America has never been studied thoroughly. This issue can be suitably tackled with the inference of a molecular timescale. In this study, using a relaxed molecular clock method, we estimated that the most recent common ancestor of South American canids lived around 4 Ma, whereas all other splits within the clade occurred after the rise of the Panamanian land bridge. We suggest that the early diversification of the ancestors of the two main lineages of South American canids may have occurred in North America, before the Great American Interchange. Moreover, a concatenated morphological and molecular analysis put some extinct canid species well within the South American radiation, and shows that the dental adaptations to hypercarnivory evolved only once in the South American clade.

Introduction

The evolution of the Canidae began in North America, as documented in Eocene sediments (40 Ma) with fossils of Hesperocyon (Wang et al., 2008). The vast North American fossil record strongly suggests that not only the origin but also the early diversification of canids took place on this continent. Successive radiation events soon gave rise to the three recognized subfamilies, Hesperocyoninae, Borophaginae and Caninae, by the late Oligocene (Wang et al., 2004). The first two subfamilies were endemic to North America and have been extinct since the middle Miocene and early Pleistocene respectively (Wang & Tedford, 2007; Wang et al., 2008). The sole surviving subfamily, Caninae, is now distributed worldwide and includes all 36 extant species of the family.

After being confined to North America for a period of about 30 Myr, canid lineages spread across the world. Members of the genus Eucyon reached the Asian continent through the Beringian land bridge around 6 Ma and continued on to colonize Europe (Wang et al., 2008). In South America, canid records are found much later, in late Pliocene sediments (ca. 3 Ma) (Wozencraft, 2005; Lucherini & Vidal, 2008). Despite the North American origin and the recent invasion, the largest extant canid diversity is actually found in the South American continent, with six genera and 11 species recognized (Wozencraft, 2005).

Extant canid diversity in South America certainly originated from more than one incursion from the north, given the presence of different lineages. Obvious representatives of North American species, such as the grey fox (Urocyon cinereoargentus) and the extinct Canis dirus, are found in South America (Berta, 1987; Prevosti & Rincon, 2007). The remaining assemblage of South American canids includes the so-called endemic fauna: the maned wolf (Chrysocyon brachyurus), the bush dog (Speothos venaticus), the crab-eating fox (Cerdocyon thous), the small-eared dog (Atelocynus microtis) and the South American foxes (Lycalopex spp.) (Wozencraft, 2005). The monophyly of the South American endemic fauna is well supported by molecular studies (e.g. Wayne et al., 1997; Lindblad-Toh et al., 2005), although some morphology-based phylogenies have included the Asian Nyctereutes procyonoides, the raccoon dog, within the group (Berta, 1987; Tedford et al., 1995).

Knowledge of the timescale of the evolution of the South American endemic fauna is critical to comprehend the history of the lineage. Nevertheless, most canid molecular studies focused solely on the phylogenetics of the group, neglecting divergence time issues (Bardeleben et al., 2005; Lindblad-Toh et al., 2005). The time of entry in the continent must be considered as important as a robust phylogenetic proposal if the scenario of canid diversification in South America is to be unveiled.

In this paper, we have attempted to reconstruct the evolutionary history of South American canids, establishing a chronological line for the cladogenetic events within the Canidae. In addition, in order to investigate the phylogenetic position of the canid fossils endemic to South America, we conducted a phylogenetic analysis that combines molecules and morphology for extinct and extant canid species.

Materials and methods

Sequences and alignment

We composed a data set combining the mitochondrial genes COI, COII and CYTB used by Wayne et al. (1997) and the 22 nuclear loci sequenced by Lindblad-Toh et al. (2005) and Bardeleben et al. (2005). A total of 27 canid species were studied. In addition, two outgroups were included in our data set, the giant panda, Ailuropoda melanoleuca, and the American black bear, Ursus americanus (Table 1). The GenBank accession numbers are included in Appendix S1 in the Supporting Information.

Table 1.   Species included in this study.
SpeciesCommon name
  1. *South American endemic canids.

Atelocynus microtis*Short-eared dog
Ailuropoda melanoleucaGiant panda
Canis adustusSide-striped jackal
Canis aureusGolden jackal
Canis familiarisDomestic dog
Canis latransCoyote
Canis mesomelasBlack-backed jackal
Canis lupusGrey wolf
Cerdocyon thous*Crab-eating fox
Chrysocyon brachyurus*Maned wolf
Cuon alpinusDhole
Lycalopex culpaeus*Culpeo
Lycalopex fulvipes*Darwin’s fox
Lycalopex griseus*Chilla
Lycalopex gymnocercus*Pampas fox
Lycalopex sechurae*Sechuran fox
Lycalopex vetulus*Hoary fox
Lycaon pictusAfrican wild dog
Nyctereutes procyonoidesRaccoon dog
Otocyon megalotisBat-eared fox
Speothos venaticus*Bush dog
Urocyon cinereoargenteusGrey fox
Urocyon litorallisIsland fox
Ursus americanusAmerican black bear
Vulpes corsacCorsac
Vulpes lagopusArctic fox
Vulpes macrotisKit fox
Vulpes vulpesRed fox
Vulpes zerdaFennec fox

Alignments were conducted independently for each gene at the T-Coffee online server (Poirot et al., 2003). Gap-rich flanking regions were excluded in each individual gene alignment after visual inspection. All molecular evolutionary analyses were based on the 25-loci concatenated sequence, comprising 15 570 bp in total. The final alignment is available from the authors upon request.

Molecular phylogenetic analysis

Phylogenetic reconstructions were performed with the maximum likelihood (ML) and the Bayesian inference (BI) methods. The choice of evolutionary model was performed in the HyPhy program (Pond et al., 2005), which implements a likelihood ratio test (LRT) using the same hierarchical model arrangement of modeltest (Posada & Crandall, 1998). The significance level of the LRT was set at 1%. The GTR + G4 + I was the model that maximized the likelihood of the data and, thus, selected for both analyses.

The ML tree was built with PhyML 3.0 (Guindon & Gascuel, 2003). Branch support values for the topology were estimated with the bootstrap test. Our test values were obtained with 1000 replicates in the PhyML program. A Bayesian phylogenetic analysis was conducted also using the GTR + G4 + I model, with the MPI version of MrBayes 3.1 (Ronquist & Huelsenbeck, 2003). To obtain posterior distributions, via the Markov chain Monte Carlo (MCMC) algorithm, four independent runs were used with four chains each. MCMC runs were visited 10 000 times every 100th cycle. Convergence of the chains was checked by visual inspection of the log-likelihood plot using the program Tacer 1.4.1 (Drummond & Rambaut, 2007). For each run, 25% of the collected samples were removed (burn-in). Thus, the final joint sample was composed of 30 000 trees (7500 × 4), which were used for topological inference. Statistical support for the clades was measured using the Bayesian posterior probability.

In order to account for heterogeneity of evolutionary process over sites, a second Bayesian tree was inferred using the mixture model GTR–CAT available in PhyloBayes 2.3 (Lartillot & Philippe, 2004). Mixture models are known to reduce artefacts such as long-branch attraction (Lartillot et al., 2007). Two independent Markov chains were run until 10 000 sample trees were collected. To summarize the posterior distribution, 25% of the samples were discarded as burn-in, yielding a total of 15 000 trees (7500 × 2). Statistical support for the clades was also measured by the Bayesian posterior probability.

Combined morphological and molecular analysis

In order to further explore the scenario of the evolution of South American canids, the phylogenetic position of the extinct fauna was accessed by assembling morphological data from fossil (Speothos pacivorous, Theriodictis platensis, Protocyon scagliarum, Protocyon troglodytes and Dusicyon australis) and from extant canid species. Data on extant species were based on Tedford et al. (1995). As their matrix coded only canid genera, individual canid species and D. australis were included in our study by also considering the modified matrix of Zrzavy & Ricankova (2004). Extinct species were included into the matrix using literature descriptions (Berta, 1984, 1988; Cartelle & Langguth, 1999; Prevosti et al., 2004, 2005). Only cranial characters were coded, as most of the fossil descriptions were based on skulls. Noninformative characters were excluded, yielding a morphology-based matrix composed of 32 cranial characters (see Appendix 1).

The final morphological matrix was combined with the molecular data and analysed in a Bayesian framework using the MrBayes. For extinct taxa, molecular characters were coded as missing data. The Lewis (2001) model was used for the morphological part of the matrix. Morphological characters were mapped on the resulting phylogeny using Mesquite 2.6 (Maddison & Maddison, 2009).

Inferring divergence times

Divergence time estimates for all nodes were computed with relaxed molecular clock methods assuming uncorrelated and correlated models of rate evolution. The uncorrelated model was implemented in beast 1.4.8 (Drummond & Rambaut, 2007) using the log-normal distribution to model the evolution of evolutionary rates (Drummond et al., 2006). The tree prior was obtained using the Yule process. The Markov chain was sampled 30 000 times and 10% of the states were discarded as burn-in. For the sake of comparison, node ages were also estimated with the correlated model of Thorne & Kishino (2002) available in the Multidistribute package (see also Thorne et al., 1998). Branch lengths and the variance–covariance matrix for each gene were estimated in ESTBRANCHES using the F84 + G4 model. These estimates were then used by multidivtime to approximate the likelihood function and to run the MCMC algorithm. To obtain the posterior distribution of the ages, a burn-in period of 1 000 000 generations was used. The chain was then sampled 30 000 times every 1000 cycles. multidivtime also requires assignment of priors for several parameters. The mean (and SD) of the prior distribution for the rate of the root node was set as 0.04 (±0.04), whereas the mean (and SD) of the prior for the Brownian motion constant (ν) was set at 0.8 (±0.8).

Fossil calibrations

Reliable calibrations are crucial to a consistent timescale (Marjanovic & Laurin, 2007). One advantage of the methods detailed above is that calibration times are entered as distributions rather than fixed points. Hence, minimum and maximum ages, for each calibrated node, may be associated with a probabilistic distribution. In our case, we used three fossil constraints as time priors. In beast, the age of the root, namely the Canidae/Ursidae split, was constrained by a Gamma prior with shape parameter 1 and scale parameter 4 (Fig. 1a). Values younger than 40 Ma for this node were discarded, as this is the age of the first Hesperocyoninae fossil records (Wang et al., 2008).

Figure 1.

 Probability distributions (priors) assigned to the age of the nodes in the uncorrelated log-normal model adopted in beast. See text for details.

The second prior used was the Ailuropoda/Ursus split. This divergence was also modelled by a Gamma prior (shape = 1 and scale = 8; Fig. 1b) with values younger than 7.5 Ma assigned zero probability, which is justified by the early Ailuropodini record of Ailurarctos (Qiu & Qi, 1989). The third time constraint is related to the split of Canini/Vulpini tribes (sensuMcKenna & Bell, 1997). In this case, a Gamma prior (shape = 1 and scale = 5; Fig. 1c) was assumed with an offset of 8 Ma, which corresponds to the earliest record of Eocyon (an early member of Caninae). The gamma-scale parameter was set in order to include values near 30 Ma in the tail of the distribution because Archaeocyon, the oldest recorded Borophaginae (a sister group of Caninae), is approximately 32 Ma (Wang et al., 2004).

In multidivtime, the outgroup is ignored; thus, the priors for Canidae/Ursidae and Ailuropoda/Ursus splits were not used in this analysis. In addition, this method uses uniform time intervals; hence, the only time constraint applied was the age of the Caninae split, the ingroup root, which was forced to lie between 8 and 32 Ma.

Results

Phylogenetic analyses

Our results show no topological differences between Bayesian and ML trees inferred from the molecular data (Fig. 2). The topology obtained was very close to the maximum parsimony tree of Lindblad-Toh et al. (2005), who used a similar set of genes. The monophyletic condition of the South American endemic fauna has significant statistical support, with 100% Bayesian posterior probability in MrBayes (BPP), 100% Bayesian posterior probability under the mixture model (MM-BPP) and 99% bootstrap support in ML analysis (BS). Within this clade, two groups were strongly supported: the (Chrysocyon + Speothos) lineage (100% BPP, 100% MM-BPP and 100% BS) and the South American fox-like lineage, which comprises Lycalopex + Cerdocyon + Atelocynus (100% BPP, 100% MM-BPP and 99% BS). Phylogenetic relationships of the South American fox-like canids, however, remain unresolved, although the genus Lycalopex was significantly supported and a Lycalopex + Cerdocyon association is suggested (79% BPP, 68% MM-BPP and 62% BS).

Figure 2.

 Phylogeny of extant Canidae based on 22 nuclear and three mitochondrial loci. Numbers on nodes refer to Bayesian posterior probability/mixed model Bayesian posterior probability/bootstrap support. Asterisk indicates polytomy.

South American canids are recovered as being strongly related (100% BPP, 100% MM-BPP and 100% BS) to a clade composed of the (Canis + Lycaon + Cuon) genera. Although the (Canis + Lycaon + Cuon) clade is also strongly supported (100% BPP, 100% MM-BPP and 96% BS), the genus Canis was recovered as a nonmonophyletic assemblage, with one lineage (lupus/familiaris + latrans + aureus) closely related to the dhole (Cuon alpinus) (100% BPP, 100% MM-BPP and 96% BS), whereas the remaining Canis species (C. adustus and C. mesomelas) are strongly associated (100% BPP, 100% MM-BPP and 90% BS) and related to the (Canis + Cuon + Lycaon) clade.

The other major branch of the molecular tree includes a monophyletic genus Vulpes (100% BPP, MM-BPP and BS). A weak statistical support was encountered for the (Otocyon + Nyctereutes) grouping, related to Vulpes, in both ML and Bayesian trees. The phylogenetic position of Otocyon, Nyctereutes and Urocyon could not be elucidated with our data.

The combined molecular and morphological phylogeny placed all extinct endemic South American canids within the South American clade (Fig. 3). Protocyon, Theriodictis and Speothos presented evolutionary affinities with the (Chrysocyon + Speothos) clade. They were associated to the extant Speothos with statistical value of 87% BPP. Neither Speothos nor Protocyon were monophyletic, because T. platensis was inferred as a sister taxon of P. troglodytes (95% BPP), with the exclusion of P. scagliarum. Speothos pacivorus was positioned as a sister group of the (T. platensis + P. troglodytes + P. scagliarum) clade. The recently extinct D. australis was placed deep within the Lycalopex clade as a sister taxon of Lycalopex culpaeus (87% BPP). Such phylogenetic arrangement clearly points to a close relatedness between Dusicyon and Lycalopex.

Figure 3.

 Bayesian phylogeny of a combined data set of molecular and morphological data, including fossils, with the evolution of some morphological characters mapped onto the tree.

Our combined analysis also resolved the position of Urocyon as the sister group of Vulpes (93% BPP). Both taxa are associated with Otocyon (96% BPP), forming a Vulpini clade. The position of Nyctereutes remains unresolved, positioned in a trichotomy at the base of the tree with the Canini and Vulpini radiations.

Timescale

The time of the most recent common ancestor (TMRCA) of South American canids was estimated at 4.3 Ma using the uncorrelated model, with a 95% Bayesian highest probability density (HPD) from 3.4 to 5.5 Ma (Table 2). The time estimate for this split was essentially the same for the correlated model, at 4.2 Ma (3.3–5.2). This estimate places extant South American radiation in the Pliocene. The Chrysocyon/Speothos divergence occurred around 3.0 Ma, whereas the TMRCA of the (Atelocynus + Lycalopex + Cerdocyon) clade was inferred slightly earlier, at 2.4 and 2.7 Ma by the correlated and uncorrelated models respectively. This suggests that both lineages radiated relatively recently. The diversification of the Lycalopex species took place within approximately 1.3 Myr in the Pleistocene.

Table 2.   Times (Ma) of South American Canidae divergences.
DivergenceTime (95% HPD)
CorrelatedUncorrelated
Diversification of modern Caninae8.7 (8.0–9.9)8.8 (8.0–10.5)
Basal SA canid split4.2 (3.3–5.2)4.3 (3.4–5.5)
Basal (Atelocynos + Cerdocyon + Lycalopex)2.4 (1.7–3.2)2.7 (2.0–3.6)
Chrysocyon/Speothos3.1 (2.2–4.0)3.0 (2.1–3.9)
Diversification of Lycalopex spp.1.3 (0.8–1.8)1.3 (0.9–1.8)
Diversification of (Canis + Lycaon + Cuon)3.8 (2.9–4.7)4.3 (3.4–5.5)

Our results indicate that rapid diversification patterns are found throughout Canidae. Extant canids are descendents of a lineage from the end of the Miocene, around 8.7 (correlated) and 8.8 Ma (uncorrelated). The diversification of the golden jackal, the coyote and the grey wolf occurred within 1.7 Myr. The domestication of the grey wolf is estimated to have begun from 136 000 to 280 000 years ago. Apart from the divergence between the Arctic fox and the kit fox, all Vulpini splits are older than 2 Ma. Curiously, the basal diversification of the South American canids, the Canis-like Canini and the Vulpini took place relatively synchronously, at around 4.5 Ma.

Discussion

Phylogeny of South American canids and the evolution of hypercarnivory

Although questioned in some earlier studies (Wayne et al., 1997), the monophyly of South American endemic canid fauna is well supported in most molecular phylogenies (Bardeleben et al., 2005; Lindblad-Toh et al., 2005). Our molecular data set, comprising more than 15 000 bp, provides strong support for the monophyletic assemblage of South American endemic canids. Morphological studies also support a monophyletic South American Canidae but with the inclusion of the raccoon dog, Nyctereutes (Tedford et al., 1995). Berta (1987) also included this species as part of the South American radiation. In our study, the position of Nyctereutes is unresolved in both the molecular and combined analyses.

At first glance, it seems biogeographically unlikely that an Eastern Asian species would be deeply nested into the South American Canidae radiation. Nevertheless, it has been argued that the disappearance of the Bering land bridge (ca. 4 Ma) was the geological event that isolated Cerdocyon and Nyctereutes ancestors (Berta, 1987). In this scenario, Cerdocyon would have invaded South America independently of other South American canids (ca. 3 Ma). Part of this argument is based on the fossil record of Cerdocyon-like canids in North America (see below).

Among the morphological characters uniting Nyctereutes to South American canids is, for example, the presence of a subangular lobe in the mandible (Fig. 3). This feature also occurs outside South American diversity, in Urocyon and, in a more extreme form, in Otocyon, and it may have functional significance. This feature is also absent in many South American canids, being present only in Cerdocyon and Atelocynus. Other characters, like the extension of the palatine bones beyond the tooth row, characteristic of South American canids, are absent in some South American taxa and present in the highly specialized Otocyon (Fig. 3).

There is consensus that South American canids are most closely related to dog-like canids (Canini) among the Canidae. Some previous phylogenies suggested an affinity of Chrysocyon with Canis-like taxa or, together with Speothos, in an unresolved position within Canini (Berta, 1987; Wayne et al., 1997). Our phylogenies corroborate those of Bardeleben et al. (2005) and Lindblad-Toh et al. (2005), which groups Chrysocyon with the morphologically divergent taxon Speothos as a sister clade of the South American foxes and the small-eared dog.

The bush dog has already been associated with Cuon and Lycaon in the subfamily Simocyoninae (Simpson, 1945). This association was based on the absence of the entoconid on the inferior carnassial molar (m1), a feature that creates a trenchant heel, an adaptation to hypercarnivory. This feature, however, has long been considered homoplastic in nature, having evolved many times in different carnivore lineages (Van Valkenburgh, 2007).

In South America, apart from Speothos, a trenchant heel is also found in the extinct Theriodictis and Protocyon. Interestingly, our combined analysis unites Speothos, Theriodictis and Protocyon in a monophyletic group, implying that hypercarnivory evolved only once in South American endemic canids (Fig. 3). This result conflicts with the previous phylogenetic hypothesis of Berta (1987, 1988) that associated these extinct taxa with the South American foxes (Lycalopex spp.) and with those studies that indicate a relationship with Old World hypercarnivorous taxa such as Lycaon (Prevosti, 2005).

Our data also suggest that the genus Protocyon may be paraphyletic, as P. troglodytes is the sister taxon of Theriodictis. Prevosti (2007) comments that the genus Theriodictis may indeed be paraphyletic, with some members more closely related to Protocyon. One possibility is that the older P. scagliarum, recorded from the Lower-Middle Pleistocene of Argentina (Prevosti et al., 2005), might belong to the stock from which the Late Pleistocene species of both Protocyon and Theriodictis evolved, with the characters uniting P. scagliarum and P. troglodytes being plesiomorphic in nature. However, our approach to these fossil taxa was superficial, and conclusions are pending a more comprehensive analysis.

We also found that the genus Speothos is paraphyletic. Such results, however, might be a consequence of the molecular bias in the combined analysis. As the phylogenetic signal on the molecular matrix is strong, it may cause an attraction of extant S. venaticus to Chrysocyon. If we consider a purely morphological analysis, it shows S. venaticus and S. pacivorous in a basal polytomy at the Protocyon + Theriodictis radiation (data not shown).

The clade composed of the South American foxes (Cerdocyon and Lycalopex) and Atelocynus was previously suggested by molecular data (Wayne et al., 1997; Bardeleben et al., 2005; Lindblad-Toh et al., 2005), but morphological analyses have shown more controversial results (Berta, 1987; Tedford et al., 1995). In our analyses, the monophyly of the Lycalopex foxes is significantly supported, but the phylogenetic position of Cerdocyon is not evident with the molecular data alone. However, the inclusion of morphological characters strongly associates Cerdocyon with Atelocynus. A close relationship is also suggested between L. culpaeus with the recently extinct Falkland-wolf (D. australis) by our combined matrix. This result renders a paraphyletic status for Lycalopex. If this relationship holds, the generic name Dusicyon Hamilton Smith (1839) would have priority over both Lycalopex Burmeister (1839) and Pseudalopex Burmeister (1856) (other commonly used denomination). The short diversification time shown by the clade composed by (Lycalopex + Dusicyon) contributes to the idea that all the included species may be congeneric.

Divergence time and evolutionary scenarios

The basal split of this South American endemic clade was inferred at 4.2 and 4.3 Ma by the correlated and uncorrelated models respectively (Fig. 4). This time is slightly older than the estimated age of the complete rise of the Isthmus of Panama, from 2.7 to 3.6 Ma (Haug & Tiedemann, 1998). If the 95% HPD interval for the age of the node is considered (3.5–5.8), the hypothesis that the early clade divergence took place in South America cannot be strictly discarded. Subsequently, a single colonization event by the ancestors of the endemic fauna could be assumed. The fossil record, however, holds evidence that challenges such a straightforward scenario. The contradiction exists because it has been claimed that members of the SA endemic fauna were actually found in North America before the rise of the Panamanian land bridge (Munthe, 1998). This argument is specifically supported by North American fossils assigned to the genera Cerdocyon and Chrysocyon.

Figure 4.

 Timescale of Canidae diversification. Bars on nodes indicate the 95% highest probability density interval for the time estimates. Thick bars represent the time span of the fossil record of the species studied. Dashed line indicates the rise of the Panamanian land bridge. Canid illustrations adapted from St George Mivart. 1890. Dogs, Jackals, Wolves, and Foxes: A Monograph of Canidae. Illustrated by J. O. Keulemans. Dulau and Co., London.

Fossils currently assigned to the genus Cerdocyon are found in North America in late Miocene to early Pliocene deposits, ca. 5.3 Ma (Torres-Roldán & Ferrusquía-Villafranca, 1981; Berta, 1987; Miller & Carranza-Castañeda, 1998). Therefore, the earliest North American Cerdocyon record pre-dates the divergence time of the two major South American lineages (≈4 Ma). The age of these fossils is actually closer to the divergence time between the South American clade and the ‘Canis’ clade (≈5 Ma).

The simplest explanation for this disparity is that these fossils are not actually related to the extant crab-eating fox. North American Cerdocyon might correspond to the ancestral morphotype of South American Canidae, which is still observed in living Cerdocyon. It is interesting to note that the generalized nature of the Cerdocyon morphology has been previously noted by Torres-Roldán & Ferrusquía-Villafranca (1981), who postulated that this genus was ancestral to most other South American Canidae.

Fossils attributed to Chrysocyon are known from fragmentary remains from early to mid-Blancan (ca. 3.5 Ma) of Arizona and New Mexico (Berta, 1987; Munthe, 1998). These findings are incongruent with the timescale proposed in this paper, which estimates the Chrysocyon/Speothos divergence around 3 Ma. However, the extreme divergent post-cranial adaptations of extant Chrysocyon (Dietz, 1985), which are mainly on post-cranial characteristics, presumably could not be evaluated on North American Chrysocyon.

The existence of these North American fossils closely related to the two main lineages of South American canids, i.e. North American Cerdocyon to the (Atelocynus + Cerdocyon + Lycalopex) lineage and North American Chrysocyon to the (Chrysocyon + Speothos) lineage, suggests that the early South American canid diversification may have occurred in North America before the invasion during the Great American Interchange. Unfortunately, however, these records are known only from fragmentary remains. Furthermore, it is clear that both Cerdocyon and Chrysocyon show a rather generalized dentition. Thus, it is plausible that they represent basal members of each lineage and may not be directly related to the extant forms assigned to these genera. Curiously, early splits of these South American lineages occurred near the rise of the Panamanian land bridge, which further corroborates a two-lineage colonization scenario (Fig. 5).

Figure 5.

 Hypothesis of the evolution of South American canids.

The earliest South American endemic canid fossils date from the late Pliocene (ca. 2.5 Ma) and were originally assigned to the genus Dusicyon (Prevosti et al., 2005; Prevosti & Rincon, 2007; Lucherini & Vidal, 2008). This genus is related to the South American fox radiation that, in our estimates, initiated at ca. 2.6 Ma. Later genera such as Chrysocyon, Theriodictis and Protocyon appear from the early to mid-Pleistocene (Berta, 1987; Prevosti et al., 2005).

The diversification of Protocyon and Theriodictis has not been tackled in detail. Fossils of these genera occur throughout South America, with Protocyon in Venezuela, Ecuador, Bolivia, Brazil and Argentina (Prevosti et al., 2005; Prevosti & Rincon, 2007) and Theriodictis in Argentina, Bolivia, and possibly Ecuador and southern Brazil (Prevosti et al., 2004). Prevosti & Rincon (2007) speculated, based on the fossil record of Protocyon in northern South America (Venezuela) and the subsequent proximity with the Panamanian land bridge, that this genus could have invaded Central America during the Pleistocene, presumably implying that they believed that Protocyon evolved in situ in South America. Our study assigns both genera to the (Chrysocyon + Speothos) lineage, which diversified around 3 Ma. Therefore, Protocyon and Theriodictis should be interpreted as early members of the radiation.

The speciation of the Lycalopex foxes occurred very rapidly, within approximately 1.3 Ma. This estimate explains why the taxonomy of South American foxes has been regarded as particularly problematic (Lucherini & Vidal, 2008). The radiation of SA foxes may be related to the complex and dynamic process of retraction and expansion of glaciers and climate changes in the Andes and southern South America during the Pleistocene and early Holocene (Markgraf, 1989, 1993). The time span in which the diversification took place is so short that a phylogeographic investigation might be useful to understand Lycalopex spp. systematics.

The timing of the entrance and radiation of the Canidae in South America poses a challenge to the understanding of the recent diversification of the group, as well as to the larger picture of mammalian evolution in South America during the Great American Interchange. The data available so far present a considerable amount of information to be used for chronological inference. For instance, in the correlated model of rate evolution, the influence of the data (alignment) on the prior distribution is significant (Fig. 6). Moreover, such estimates were obtained by applying a considerably wide time range as calibration information for the ingroup root (8–32 Ma). Consequently, our estimates for node ages were robust to the model of rate evolution used (correlated or uncorrelated) and to the prior distribution assigned to the Canini/Vulpini split.

Figure 6.

 Impact of data on the prior distribution of the TMRCA of South American Caninae under the correlated model of rate evolution, showing its effect on narrowing the posterior distribution.

Our findings depict a more complex scenario for the evolution of South American endemic Caninae than a simple invasion followed by diversification. Much of the climatic and ecological history of South America during the last 3 Myr must be assessed to compose a consistent hypothesis regarding this diversification. In particular, it is imperative to compare our results with other late Tertiary invasions that occurred during the Great American Interchange.

Acknowledgments

The authors are greatly indebted to Dr Richard H. Tedford for reviewing an earlier version of this manuscript. This work fulfils part of the requirements for the PhD degree in Genetics for FAP at the Federal University of Rio de Janeiro. FAP is supported by a scholarship from CAPES (Brazilian Ministry of Education). The CGS and CAMR laboratories are supported by grants from FAPERJ (Rio de Janeiro State’s Research Foundation) and CNPq (National Research Council). This study used the computer facilities of NACAD, COPPE/UFRJ. This paper is dedicated to Lund and Luna.

Appendix

Appendix 1: Morphological matrix with character descriptions

Alopex lagopus01110000000000000100000000000001
Atelocynus microtis11011111112001001100010000000001
Canis adustus01001211000010101100001000010011
Canis aureus01001211000010101100001000010011
Canis latrans 01001211000010101100001000010011
Canis lupus 01001211000010101100001000010011
Canis mesomelas01001211000010101100001000010011
Cerdocyon thous11001111112000001100010000000001
Chrysocyon brachyurus01001111111000001100010000000001
Cuon alpinus01001211000110111101001011111110
Dusicyon australis01001111111001001100000100000001
Vulpes zerda01110000000000000100000000000001
Lycalopex vetulus01001111111001001100010000000001
Lycaon pictus01001211000110111101001110111010
Nyctereutes procyonoides11001111112001001100010000000001
Otocyon megalotis21100000000001000211000000000001
Lycalopex culpaeus01001111111001001100100100000001
Lycalopex fulvipes0100111111?001001100??0?00000001
Lycalopex griseus01001111110001001100000000000001
Lycalopex gymnocercus01001111110001001100000000000001
Lycalopex sechurae01001111110001001100000000000001
Speothos venaticus00012111012101011100010011111101
Urocyon cinereoargenteus11100000000000001210110000000001
Vulpes corsac01110000000000000100000000000001
Vulpes macrotis01110000000000000100000000000001
Vulpes vulpes01110000000000000100000000000001
Speothos pacivorous000121111??1?1011100011010110101
Protocyon troglodytes00002211111101001100011110111111
Protocyon scagliarum00002111111101011101011110111111
Theriodictis platensis00002211111101001?00011110111111

(1) Subangular lobe of the mandible: (0) absent, (1) present but undeveloped, (2) present and large;

(2) Horizontal ramus of mandible: (0) deep and thick, (1) shallow and thin;

(3) Paroccipital process width: (0) medio-laterally narrow, (1) broad, closely appressed to bulla, short free tip turned laterally, rarely extends below body of process;

(4) Nasal length: (0) long, usually extending posteriorly beyond the maxillary frontal suture, (1) short, rarely extending to the level of the most posterior position of the maxillary frontal suture;

(5) Paroccipital process posterior expansion: (0) no or little expansion, (1) expands posteriorly from bulla, usually with a prominent free tip, (2) large, greater posterolateral expansion;

(6) Frontal sinus: (0) absence of a depression on the dorsal surface of the post-orbital process, (1) lacks a depression on the dorsal surface of the post-orbital process, (2) large, penetrating post-orbital process and expanding posteriorly toward the frontal–parietal suture;

(7) Mastoid process: (0) small, crest-like, (1) enlarged knob- or ridge-like;

(8) Zygoma orbital partition: (0) presence of a lateral flare and eversion of dorsal border, (1) lack of lateral flare and thickened dorsal border;

(9) Scars of medial masseteric muscle: (0) narrow and uniform width on zygomatic arch and on lateral surface of angular process, (1) wide on zygomatic arch and enlarged on mandible;

(10) Coronoid process: (0) short at base relative to dorsoventral height, (1) long at base relative to dorsoventral height;

(11) Angular process: (0) slender, attenuated with dorsal hook, inferior pterygoid fossa not expanded, (1) large, usually blunt without dorsal hook; fossa for inferior branch of medial pterygoid muscle expanded, (2) deep, short process fossae for the pterygoid muscle are expanded;

(12) Palate: (0) not widened, (1) widened;

(13) Angular process of superior fossa: (0) not expanded, (1) expanded with large fossa for the superior branch of the medial pterygoid muscle;

(14) Palatine length: (0) extends posteriorly to or just anterior to the end of the tooth row, (1) extends beyond the end of tooth row;

(15) Supraoccipital shield: (0) rectangular- or fan-shaped when viewed posteriorly, inion may not overhang condyles, (1) triangular in shape, inion usually pointed and overhangs condyles;

(16) M3: (0) absent, (1) present;

(17) m1 entoconid: (0) conical and enlarged, may coalesce with base of hypoconid to block talonid basin, (1) joining with hypoconid by cristids to form transverse crest;

(18) I1-3 medial cusplets: (0) present, (1) cusplet in I3 absent, (2) cusplet in I1-2 weak or absent;

(19) Crown height of premolars: (0) low-crowned, (1) high-crowned;

(20) p4 anterior cusplet: (0) weak or absent, (1) strong;

(21) m1 protostylid: (0) absent, (1) present;

(22) M1-2 shape: (0) transversely wide for their labial length, (1) narrow for their labial length;

(23) p4 second posterior cusplet: (0) absent, (1) present;

(24) P4 protocone: (0) salient, located medial to anterior border of the paracone, (1) reduced, (2) further reduced or absent, located posterior to the anterior border of the paracone;

(25) m1 metaconid and entoconid: (0) not reduced, (1) greatly reduced or absent;

(26) m2 metaconid and entoconid: (0) not reduced, (1) greatly reduced or absent;

(27) M1-2 hypocones: (0) not reduced, (1) greatly reduced or absent;

(28) M1-2 paracones: (0) not enlarged, (1) enlarged relative to metacone;

(29) MI-2 buccal cingulum: (0) not reduced, (1) reduced or lost;

(30) M2: (0) triple-rooted, (1) double-rooted or M2 absent;

(31) I3: (0) small crown extending to or just below the level of the I1-2, posteromedial cingulum weak or absent, (1) large crown extending markedly below the level of the I1-2, cingulum enlarged, medial crest of I1-2 present merges with cingulum;

(32) P3 and p2-3 posterior cusplets: (0) present, (1) weak or absent;

Ancillary