The flattop mountains (tepuis) of South America are ancient remnants of the Precambrian Guiana Shield plateau. The tepui summits, isolated by their surrounding cliffs that can be up to 1000 m tall, are thought of as “islands in the sky,” harboring relict flora and fauna that underwent vicariant speciation due to plateau fragmentation. High endemicity atop tepui summits support the idea of an ancient “Lost World” biota. However, recent work suggests that dispersal between lowlands and summits has occurred long after tepui formation indicating that tepui summits may not be as isolated from the lowlands as researchers have long suggested. Neither view of the origin of the tepui biota (i.e., ancient vicariance vs. recent dispersal) has strong empirical support owing to a lack of studies. We test diversification hypotheses of the Guiana Shield highlands by estimating divergence times of an endemic group of treefrogs, Tepuihyla. We find that diversification of this group does not support an ancient origin for this taxon; instead, divergence times among the highland species are 2–5 Ma. Our data indicate that most highland speciation occurred during the Pliocene. Thus, this unparalleled landscape known as “The Lost World” is inhabited, in part, not by Early Tertiary relicts but neoendemics.

The tepuis (flattop mountains) of northern South America are arguably the most dramatic example of sky islands on Earth. With relatively flat summit plateaus sitting atop sheer cliffs of up to 1000 m, these tabletop mountains form a discontinuous ecosystem called Pantepui (Huber 1988) and harbor numerous endemic lineages. For example, within the Pantepui, 60% of the vascular plant species and 87% of the frog species are endemics, often to a single tepui summit (Duellman 1999; Berry and Riina 2005; McDiarmid and Donnelly 2005). The tepuis are part of the Guiana Shield, which is situated between the Orinoco and Amazon Basins, accounts for 9% of the land area of South America, and shows biogeographic affinities with the Amazon and Andes.

Early reports of the unique topography and biodiversity of the Pantepui inspired Sir Arthur Conan Doyle to write of long-isolated dinosaur populations surviving to the present day in his novel “The Lost World,” which inspired the namesake Lost World hypothesis (Rull 2004a; or Plateau hypothesis, Chapman 1931). This is the main biogeographic hypothesis put forward to explain the highly endemic fauna and flora of this region (Chapman 1931; Maguire 1970; Hoogmoed 1979; McDiarmid and Donnelly 2005; Heinicke et al. 2009). This hypothesis predicts that highland species, which are often endemic to a single tepui, are relicts of formerly widespread plateau taxa and arose through vicariance following ancient fragmentation of the plateau. Thus, summit taxa are hypothesized to have been isolated for millions of years. In examining the especially high endemism levels of Pantepui frog genera, most researchers support the idea that diversification must be explained, at least partially, by long-term isolation of Plateau paleoendemics (Hoogmoed 1979; MacCulloch and Lathrop 2002; McDiarmid and Donnelly 2005; Heinicke et al. 2009). However, to date, it is unclear what fraction of the highland endemism, in frogs or other groups, is explained by the Lost World hypothesis.

Many global biodiversity hotspots are tropical mountainous areas, suggesting that highland areas promote higher rates of diversification (Orme et al. 2005). In most uplifted mountain ranges such as the Andes, the summits are the most recently exposed surfaces, and these seem to be associated with recent species divergences (Hughes and Eastwood 2006). In contrast, the tepui summits are of Precambrian origin and are thought to harbor ancient lineages (Chapman 1931; Maguire 1970; Hoogmoed 1979; McDiarmid and Donnelly 2005; Heinicke et al. 2009). The tepuis were formed about 70–90 Ma, after the Guiana Shield plateau underwent several periods of erosion and plateau uplift and fragmentation starting around 300 Ma, resulting in isolated sky islands (Briceño et al. 1990; Briceño and Schubert 1990; Gibbs and Barron 1993). Because the tepuis were formed mostly by erosion rather than uplift, the summit surfaces are geologically older than the adjacent slopes.

Many studies have suggested that vicariance alone (i.e., the Lost World hypothesis) cannot explain the current distribution of highland vascular plants (Huber 1988; Givnish et al. 1997), birds (Mayr and Phelps 1967), and ants (Jaffé et al. 1993), nor the available pollen deposits (Rull 2005; Rull and Nogué 2007). A set of alternative hypotheses states that diversification occurred more recently, after the summits were colonized by lowland species through various mechanisms such as habitat shifts, vertical (cool climate) displacement, and island hopping (Huber 1988; Mayr and Phelps 1967; Rull 2004a). The Island-Hopping hypothesis (Chapman 1931) suggests aerial dispersal among the tepuis. This hypothesis seems plausible (though not obligatory) for organisms with higher dispersal abilities (e.g., birds, vascular plants, insects; Mayr and Phelps 1967; Jaffé et al. 1993; Givnish et al. 1997; Rull and Nogué 2007) but unlikely for low-vagility organisms. The Habitat Shift hypothesis (Mayr and Phelps 1967) suggests that lowland species adapted to cooler climates, allowing colonization of the highlands (Mayr and Phelps 1967; Huber 1988; McDiarmid and Donnelly 2005). The Vertical Displacement hypothesis (Rull 2004a,c, 2005), which is not exclusive of the Habitat Shift hypothesis, suggests that cooler climates, especially during Quaternary climate oscillations, promoted downward elevational shifts of habitat and range expansion of highland species into the lowlands, thus connecting previously isolated populations. Subsequent warmer interglacials promoted upward shifts of these cold-adapted populations, isolating them on the summits. Thus, the Vertical Displacement hypothesis specifically invokes historical climate change, whereas the Habitat Shift hypothesis focuses on populations adapting to new habitats.

The Vertical Displacement hypothesis is similar to the well-known Forest Refuge hypothesis of Haffer (1969) in that both invoke glacial–interglacial climate fluctuations resulting in alternation between isolated refugia and widespread habitat types. Hypotheses of Pleistocene refugia have been extremely popular, though also highly disputed, for explaining biogeographic patterns both in temperate (Knowles 2000; Johnson and Cicero 2004; Galbreath et al. 2009) and tropical regions (Mayr and O’Hara 1986; Hooghiemstra and Van der Hammen 2004; Carnaval and Moritz 2008). The postulated Pleistocene refugia of the Guiana Shield are not restricted to tepui summits. For example, in low-elevation mountains in the eastern Guiana Shield, the effects of Pleistocene climatic fluctuation have been posited to promote extensive secondary contact among species typically found in mid-elevations (Noonan and Gaucher 2005, 2006).

Debate on the origins of the Pantepui biota thus focuses on the relative importance of recent dispersal versus ancient vicariance. Not unexpectedly, the Pantepui biota is likely to be a mosaic of remnant Guiana Shield lineages and more recent colonizers (Rull 2004b). Divergence times from molecular analyses of tepui species, and between these tepui inhabitants and their closest lowland relatives, can differentiate among these alternatives. To date, however, few studies have tested these hypotheses using phylogenetic approaches (but see Givnish et al. 1997, 2004), largely due to the great difficulty and expense of conducting fieldwork in the Pantepui ecosystem. To our knowledge, we offer the first explicit test of diversification hypotheses of a Pantepui-endemic vertebrate group by inferring the phylogenetic relationships and divergence times for species of Tepuihyla, which comprises seven allopatric treefrog species that occur only at mid- to high-elevations (Fig. 1).

Figure 1.

Map of the distribution of Tepuihyla within the Pantepui of the Guiana Shield: (1) T. aecii on Duida tepui; (2) T. luteolabris on Marahuaka tepui; (3) T. edelcae on Auyan tepui; (4) T. edelcae on Chimantá massif; (5) T. rimarum on Ptari tepui; (6) T. rodriguezi on Sierra de Lema; (7) T. galani on Guadacapiapu tepui; (8) T. talbergae at Kaieteur falls.

Materials and Methods


The flattop mountains that make up the Pantepui ecosystem are remnants of the Guiana Shield Plateau and derived from Roraima Group sandstones (Huber 1987). The Pantepui occupies mainly southern Venezuela but also adjacent regions of northeastern Guyana, southern Suriname, and northern Brazil (Huber 1987). The concept of Pantepui that we follow is the association with the geological formation and not a particular elevation or floristic composition.


We sampled four of the seven recognized species of Tepuihyla. All Tepuihyla species are allopatric and inhabit montane and submontane areas. Of the seven species, four are known only from a single tepui, one (T. edelcae) from several adjacent tepui summits, and two (T. rodriguezi and T. talbergae) from lower elevation localities (Fig. 1). One nominal species, T. celsae, is not considered in our discussions. It is known only from a single locality well outside the Pantepui region. Barrio-Amorós (2004) noted that the specimens of T. celsae, which were not collected by the authors who described the species, have a doubtful locality and he attributed this new species and locality to mislabeling. Although the poor condition of the specimens makes identification challenging, the specimen is likely a mislabeled T. luteolabris from Duida tepui (C. Barrio-Amorós, pers. comm.).

Tissue samples were obtained from the Pontificia Universidad Católica del Ecuador (QCAZ), Museo de Historia Natural La Salle, Venezuela (MHNLS), and United States National Museum (USNM). See Appendix for GenBank accession numbers. The entire ingroup sample includes sequences of 86 individuals of Lophiohylini (Hylidae; Trueb 1970; Faivovich et al. 2005), of which 43 were generated by us and 43 were downloaded from GenBank. The outgroup consists of nine species of the sister clade Hylini.

Extraction and isolation of DNA, and amplification and sequencing of mitochondrial (mtDNA) and nuclear (nDNA) genes were done using standard techniques. Genomic DNA was extracted with the Viogene Blood and Tissue Genomic DNA Extraction Kit. Two different polymerase chain reaction protocols were used, one for the nDNA (the proopiomelanocortin gene POMC, approx. 460 bp) and another for the mtDNA (12S, valine tRNA, and 16S; approx. 2400 bp). The T. aecii tissue was from an old museum specimen, and extraction yielded low DNA concentrations, so only part of the 12S–16S could be amplified.

The mtDNA was amplified in four overlapping segments and using nine primers (MVZ59: 5′ATAGCACTGAAAAYGC
TDAGATG3′[Goebel 1999, #29], tRNA-phe: 5′GCRCTG
TAAG3′[Pauly et al. 2004], tRNA-val: 5′GGTGTAAGC
GAGAGGCTT3′[Goebel 1999, #73], 16SH: 5′GCTAG
ACCATKATGCAAAAGGTA3′[Goebel 1999, #76], 16SC: 5′GTRGGCCTAAAAGCAGCCAC3′[Pauly et al. 2004], 16SA: 5′ATGTTTTTGGTAAACAGGCG3′[Goebel 1999, #87], 16SD: 5′CTCCGGTCTGAACTCAGATCACGTAG3′[Goebel 1999]). The nDNA was amplified in one fragment (POMCR1: GGCRTTYTTGAAWAGAGTCATTAGWGG [Vieites et al. 2007], POMCF1: ATATGTCATGASCCAYTTYCGCTGGAA [Vieites et al. 2007]) .

The thermocycler protocol for POMC was 2 min at 94˚C followed by 12 cycles of 30 s at 94˚C, 30 s at 65˚C (–5˚C every two cycles), and 60 s at 72˚C. The thermocycler protocol for the mtDNA was 2 min at 94˚C followed by 34 cycles of 30 s at 94˚C, 30 s at 46˚C, and 60 s at 72˚C. Amplified products were purified using the Viogene Gel-M Extraction Kit. Sequencing was performed using an ABI3100 PRISM sequencer.

Contiguous sequences were assembled for 12S–16S in Sequencher 4.8 and alignments were constructed using ClustalX 2.0 with default parameters (Thompson et al. 1997) followed by manual editing in MacClade 4.08 (Maddison and Maddison 2001). Several regions in the mitochondrial dataset totaling 194 bases were judged to be unalignable and were excluded.


The combined dataset of mtDNA and POMC included 3039 bases. All individuals were represented in the mtDNA alignment, but sequences of POMC were not available for all individuals so three analyses were performed. The first included 84 ingroup individuals plus outgroups (95 individuals total) with complete mtDNA data. To minimize missing data, the second included 46 individuals plus seven outgroup species with complete data for mtDNA and POMC (53 individuals). The third included only POMC sequences (53 individuals), so that the nDNA and mtDNA topologies could be compared. Three partitioning schemes were used in MrBayes (Ronquist and Huelsenbeck 2003): a single partition including all data; two partitions, by locus (mtDNA and POMC); and three partitions, by gene and codon position (separating third codon position for POMC). Bayes factors were used to determine the preferred partition (Kass and Raftery 1995), which was by gene and codon position. This partition was used in all subsequent analyses. The best-fitting model of evolution for each of the four partitions, GTR + G + I, was determined using MrModeltest (Nylander 2004). Maximum likelihood analyses were performed using RAxML using the model GTR + G rather than GTR + G + I (Stamatakis 2006). Node support was measured by standard nonparametric bootstrapping (1000 replicates) using RAxML, and by Bayesian posterior probabilities using MrBayes (two independent runs, 10 million generations, sampled every 1000, with burn-in of 1000 out of 10,000 samples). Convergence and stationarity of runs was assessed in Tracer (Drummond and Rambaut 2007).

We compared the results for Tepuihyla to another frog group, Stefania, with similar Pantepui distribution. Stefania is a Pantepui endemic clade, but only two of the species available on GenBank occur on tepui summits. We added all available GenBank data for 12S and 16S (1510 bp total) for five of the 19 described species of Stefania (Hylidae). The sequences included were the following: Flectonotus fitzgeraldi (AY819355, DQ679381), Stefania coxi (DQ679265, DQ679415), Stefania ginesi (DQ679266, DQ679417), Stefania scalae (DQ679267, DQ679418), Stefania schuberti (AY843768), Stefania evansi (AY843767), and Stefania evansi (AY819359). The sequences were aligned and analyzed using the same protocols and programs as for Tepuihyla.


Divergence times were estimated using BEAST (Drummond and Rambaut 2007), which allows for simultaneous inference of topology and divergence times. We used two general methods of calibration to test for consistency among divergence estimates: fossils and paleobiogeography, and published substitution rates. The fossil calibrations included five hylid frogs (Smith et al. 2005). These are problematic for several reasons. First, only one pre-Holocene fossil is assigned to Lophiohylini (Osteopilus septentrionalis from the Pleistocene of the Bahamas; Sanchíz 1998) and none for the group under study (Osteocephalus and Tepuihyla). Second, for many frog fossils, the only recovered elements are ilia; rarely are even partially articulated frogs found, rendering identification less certain. Third, assignment of frog fossils to species or genera is often done from overall similarity rather than evidence of phylogenetic relationship from synapomorphies (Bell et al. 2010). Nonetheless, dates of occurrence of some frog fossils, assigned only on the basis of general similarity (Holman 1998, 2003) have been used to calibrate chronograms of hylid frogs (Smith et al. 2005; Lemmon et al. 2007).

We also used a paleobiogeographic calibration to the common ancestor of Osteopilus and Phyllodytes auratus, using the GAARlandia or Aves Ridge landbridge hypothesis (Iturralde-Vinent and MacPhee 1999). GAARlandia is a hypothesized Caribbean landbridge that connected the Greater Antilles to northern South America between 33 and 35 Ma. The support for this hypothesis is mixed; some authors support it as a major means of dispersal to the Caribbean islands (i.e., mammals, spiders, frogs; Dávalos 2004; Crews and Gillespie 2010; Alonso et al. 2012) whereas others strongly oppose it (Hedges 2006). However, Moen and Wiens (2009) found that the divergence of Osteopilus from other mainland Lophiohylini overlapped with the 33–35 Ma timeframe of this hypothesis. The prior for the GAARlandia node was drawn from a normal distribution with a mean of 34 Ma and standard deviation of 1 Ma, based on the proposed range of 33–35 Ma (Iturralde-Vinent and MacPhee 1999). Although the GAARlandia hypothesis is highly debated, we believe the inclusion of this calibration did not bias our results, since it did not change overall estimates. Furthermore, it substantially improved the computational performance of the analysis, in that good effective sample sizes (ESS > 200) were never reached (>80 M generations) without the GAARlandia calibration. With the calibrations, we analyzed the mtDNA dataset only as well as the combined mtDNA and nDNA datasets. Data partitioning was identical to that in the MrBayes analysis.

We compared chronograms obtained from fossil and paleobiogeographic calibrations to chronograms generated using two estimates of rates of evolution for anuran mitochondrial 12S and 16S genes, one from Xenopus (0.00249 substitutions per site per lineage per Myr, hereafter the “Xenopus rate”; Evans et al. 2004) and one from Pseudacris (0.00277 substitutions per site per lineage per Myr, the “Pseudacris rate”; Lemmon et al. 2007). The Pseudacris rate also relied in part on the hylid frog fossil calibrations (Smith et al. 2005) so it is not independent of our rate estimates. For these analyses, only the mitochondrial dataset was used because the published rates of evolution (Evans et al. 2004; Lemmon et al. 2007) were not estimated using POMC.

The BEAST analyses were run with sufficient generations (35–60 million) to yield ESS of at least 200 for all parameters. The GTR + G model was used with uncorrelated lognormal distributions of branch lengths and with no specification of a prior tree. Every 1000th generation was sampled, and the first 10% of the samples were discarded as burn-in based on examination of all parameter estimates in Tracer. Trees were summarized using TreeAnnotator (in the BEAST package) with the target tree type set as maximum clade credibility.

The analyses of the Stefania GenBank sequences were conducted similarly as in Tepuihyla/Osteocephalus. Two separate analyses were performed using the Xenopus and Pseudacris rates. Given that the Stefania dataset is small (compared to Tepuihyla), we only used BEAST (and exclude RAxML and MrBayes analyses) to estimate tree topology and divergence times.



Maximum likelihood and Bayesian analyses indicate that species of Tepuihyla (four out of seven recognized species; 11 individuals) form a strongly supported clade for both mitochondrial and nuclear data (1.0 Bayesian posterior probability, BPP; Figs. 2 and 3; 100% likelihood bootstrap support, BS; Fig. 3). The samples of Osteocephalus (11 out of 24 species) revealed that Tepuihyla and Osteocephalus exophthalmus form a strongly supported clade (100% BS, 1.0 BPP; Figs. 2 and 3) that is the sister group of all other Osteocephalus. Thus Osteocephalus as currently delimited is paraphyletic, and henceforth, we denote this by referring to “Osteocephalusexophthalmus with quotation marks pending our resolution of this taxonomic issue, which is in progress.

Figure 2.

Divergence time estimates and 95% confidence bars for Tepuihyla (node A), the sister species “O.exophthalmus (node B), and their lowland closest relatives, Osteocephalus (node C) for the mtDNA dataset. Divergence estimates of major clades are shown for outgroups. Estimates were obtained using the Lemmon et al. (2007) rate of substitution. Asterisks indicate Bayesian posterior probabilities greater than 95%. The tree and estimates were obtained from BEAST. Numbers next to terminal taxa refer to individual specimens in Appendix.

Figure 3.

Maximum likelihood topology for: (A) the combined dataset of 12S, 16S, and POMC (present in all terminals), and (B) the POMC only dataset. Bootstrap values are shown above the nodes and Bayesian posterior probabilities are shown below. Support values are omitted for the outgroups for simplicity. Numbers next to terminal taxa refer to individual specimens in Appendix.

We also analyzed all genera within the Lophiohylini, which includes Osteocephalus and Tepuihyla. Although the Lophiohylini was strongly supported as monophyletic, relationships among the deeper nodes of the lophiohyline clades (Fig. 3) are generally poorly supported and our topology is somewhat inconsistent with other studies (Salducci et al. 2002; Faivovich et al. 2005; Wiens et al. 2010). However, the sister-group relationship between Tepuihyla and Osteocephalus holds in all studies.

The divergences among the T. edelcae samples from the summits of Auyan Tepui and the Chimantá Massif (Fig. 1) are similar to the among-species divergences within Tepuihyla. Further, the mtDNA and nDNA phylogenies are incongruent regarding the monophyly of T. edelcae. The mtDNA only (Fig. 2) and the combined mtDNA and POMC (Fig. 3A) topologies both group T. edelcae from Chimantá as more closely related to the mid-elevation T. rodriguezi/T. talbergae than to the T. edelcae from Auyan indicating that T. edelcae is paraphyletic. However, based on nDNA only (Fig. 3B), T. edelcae is monophyletic, though this is poorly supported (44 BS, 0.81 BPP). The low variability in the nuclear locus does not resolve the species-level relationships within Tepuihyla. Thus, the apparent discrepancy between the nuclear and mitochondrial trees for T. edelcae might be due to mitochondrial introgression from recent hybridization between T. edelcae and T. rodriguezi/T. talbergae or to incomplete lineage sorting.


Divergence time estimates based on calibrations from either fossils or substitution rates showed similar results (Fig. 4). Taking into account all results from different datasets and calibrations, divergences between the Pantepui-endemic Tepuihyla and its sister taxon were estimated to be 14.7–23.6 Ma, which is far more recent than the formation of the tepuis. Furthermore, the oldest estimates for the divergence between this clade (Tepuihyla+“O.” exophthalmus) and Osteocephalus are still at least 20 Ma more recent than tepui formation. Divergences among Tepuihyla species or populations inhabiting different tepuis are also relatively recent, from 0.7 to 8.1 Ma, with several divergences overlapping the Quaternary.

Figure 4.

Divergence time estimates (top) in millions of years and 95% confidence intervals (bottom) obtained in all BEAST analyses for the main clades of the highland group, Tepuihyla, and its closest relatives, “O.exophthalmus and Osteocephalus. The clades are numbered in the simplified cladogram to the right. Node 1 was not always recovered, and thus we report the deepest divergence within that group and indicate the unrecovered node with *. The differences in number of generations in the left column are due to the fact that different analyses required different numbers of samples to reach stationarity. All divergence time estimates obtained with the calibrations of the outgroup fall within the 95% confidence intervals obtained with the previously estimated rates of evolution.

The divergence times from the Xenopus estimate were the oldest, possibly because Xenopus is very distantly related. The rates found using the hylid fossil calibrations are all in close agreement, the only difference being that the estimates obtained from the fossil calibrations (compared to the Pseudacris estimate) had much higher variance, probably because none of the calibrations lies within the group of interest (Tepuihyla and Osteocephalus). The only calibration within Lophiohylini is the paleobiogeographic calibration of Osteopilus. Given this, we focus the discussion on the analysis using the Pseudacris estimate. Estimated divergence times among Tepuihyla species range from 0.7–5.3 Ma. The oldest node, the split between T. aecii and all other Tepuihyla, is estimated at 5.3 Ma (Fig. 2, node A). Notably, this split corresponds to the geographic division between the western tepui group and the eastern tepui group (Huber 1988). Tepuihyla aecii is from Cerro Duida in the western group, and the remaining species are from the eastern group.

The youngest split (0.7 Ma) is between T. talbergae (Kaieteur National Park, Guyana, 366 m, but known from higher elevations) and T. rodriguezi (Gran Sabana, Venezuela, 800–1200 m); both species inhabit the lowest elevations known for Tepuihyla. Not all analyses recovered this poorly supported node; in the mtDNA tree (Fig. 2) T. rodriguezi was paraphyletic, which suggests either incomplete lineage sorting, or simply that they are a single species. We refer to this complex hereafter as T. talbergae/rodriguezi. The divergence between Tepuihyla and its sister species “O.” exophthalmus is estimated at 14.7 Ma (Fig. 2; node B), and that between Tepuihyla+“O.” exophthalmus and Osteocephalus is estimated at 24.7 Ma (Fig. 2; node C).

The results of the Stefania analyses using the Xenopus and Pseudacris rates were very similar, so we only report the divergence times from the Pseudacris rate. The deepest divergence found is 34.4 Ma, between S. ginesi and all other included species (Fig. 5). The most recent divergence is estimated at 7.1 Ma, between S. scalae and S. evansi.

Figure 5.

Divergence time estimates and 95% confidence intervals for Stefania. Divergence time estimates are shown in millions of years. Time of tepui fragmentation is shown as a reference. Estimates were obtained using the Lemmon et al. (2007) rate of substitution.



The low divergence estimates among Tepuihyla species (<5.3 Ma) indicate that this Pantepui clade did not speciate under a Lost World vicariance scenario. Even when accounting for uncertainty in rates of molecular evolution, credibility intervals, and time calibrations, it is evident that species diversity within this clade is not the product of ancient dissection of the Guiana Shield plateau. Furthermore, the origin of the Pantepui-endemic treefrogs does not predate the formation of the tepuis, which is another prediction of the Lost World hypothesis.

Previous phylogenetic analyses that included lophiohyline taxa (Faivovich et al. 2005; Wiens et al. 2010) sampled only one species of Tepuihyla so the monophyly of this group had never been tested. With our increased sampling (four out of seven recognized species), we found Tepuihyla to be monophyletic. However, Osteocephalus as currently delimited (Trueb 1970; Faivovich et al. 2005) is not monophyletic because “O.exophthalmus was found to be the sister taxon to Tepuihyla. “Osteocephalusexophthalmus is a Pantepui-endemic known from low and middle elevations in Guyana. In contrast to this clade of Guiana shield endemics (i.e., “O.”exophthalmus+Tepuihyla), Osteocephalus is primarily a lowland Amazonian clade; a few species occur in the lowlands of the Guiana Shield. Thus, the key node to understanding the divergence of the Pantepui clade is that between Tepuihyla+“O.exophthalmus and Osteocephalus (node 6, Fig. 4).

Tepuihyla+“O.exophthalmus diverged from Osteocephalus about 24.7 Ma (Figs. 2 and 4, node 6). This timing of separation between Osteocephalus and Tepuihyla overlaps with that of a marine incursion into the present Amazon Basin starting in the Early Miocene, 23–16 Ma (Hoorn 1993). This may explain the largely allopatric distributions of Osteocephalus and Tepuihyla.


Having rejected the Lost World hypothesis for Tepuihyla, it is now clear that the divergence times among Tepuihyla species require dispersal between the lowlands and the summits. All three dispersal hypotheses (Island-Hopping, Habitat Shift, and Vertical Displacement) generate predictions related to dispersal frequencies and times of divergence. The Island-Hopping hypothesis is unlikely because island hopping should be a rare event for Tepuihyla given its low vagility. In addition, our finding of multiple relatively recent dispersal events is inconsistent with the predictions of Island Hopping. However, given our estimates of divergence times and phylogenetic relationships, the biogeographic history of Tepuihyla does not definitively reject either of the remaining dispersal hypotheses.

Given the occurrence of Tepuihyla in mostly middle and high elevations, as well as the recent divergence times, the establishment of the current distribution probably involved elements of both the Habitat Shift and Vertical Displacement hypotheses. Because all known Tepuihyla species are associated with the Pantepui, we infer that the common ancestor of Tepuihyla was also a Pantepui inhabitant that adapted to these conditions (as the Habitat Shift hypothesis suggests). However, the sympatry of “O.” exophthalmus with T. rodriguezi/T. talbergae throughout their limited distributions in lower Pantepui elevations of the eastern Guiana Shield suggests that the association of this clade (“O.exophthalmus+Tepuihyla) with the tepuis and possible adaptation to higher elevations occurred as far back as 14.7 Ma (Fig. 2).

It is unlikely that Pleistocene climatic shifts were the only processes involved in divergences within Tepuihyla. The three deepest divergences within Tepuihyla predate the Quaternary, supporting the overlooked importance of pre-Quaternary diversification (Rull 2008). On the other hand, large confidence intervals among Tepuihyla divergences that overlap part of the Pleistocene, plus the small divergences between T. edelcae and T. rodriguezi/T. talbergae suggest that Pleistocene glacial cycles may have shaped the current distributions of these species. However, to determine the general significance of Pleistocene climate fluctuations, extensive sampling within recently diverged Tepuihyla species is needed to estimate recent gene flow using coalescent analyses.


This analysis is the first to estimate DNA-based divergence times of a Pantepui-endemic animal taxon. Our analyses of Tepuihyla recover relatively recently diverged species (Plio-Pleistocene) with good species sampling (4/7 species). Although small samples of two other Pantepui frog groups have been analyzed phylogenetically (Stefania and Ceuthomantis, see below), the only other studies (to our knowledge) that explicitly treat the historical biogeography of a Pantepui clade include members of the Bromeliaceae and Rapateaceae (Givnish et al. 1997, 2000, 2004, 2011).

Bromeliaceae and Rapateaceae are families of flowering plants within the order Poales. Within the Rapataceae, the tepui-endemic crown clade Stegolepidieae (Stegolepis, Amphiphyllum, and Epidryos) diversified 10 Ma (Givnish et al. 2004). Dispersal, and not vicariance, is argued to be the principal correlate of divergence among these lineages (Givnish et al. 2004). Furthermore, the presence of some Stegolepis in lowland as well as intermediate and summit habitats corroborates recent dispersal.

Similarly, divergence times for many clades within Bromeliaceae have been estimated. The most recent common ancestor of crown-group Bromeliaceae likely inhabited the Pantepui at 19.1 Ma (Givnish et al. 2011: Fig. 7). Brocchinioideae and Lindmanioideae, the two earliest branching lineages within Bromeliaceae, are tepui endemics. The range of divergences between these clades is 8.9–19.0 Ma. The Bromeliaceae, then, does not owe its diversification (as a crown group) to the vicariant dissection of Pantepui, but more likely to late Cenozoic dispersal across a dissected landscape, a scenario similar to that observed for Tepuihyla.

The Pantepui endemic bromeliads in the genus Brocchinia are incredibly diverse in morphology and ecology (Givnish et al. 1997). This diversity could suggest that this clade is an “ancient” lineage. However, species within Brocchinia diverged relatively recently; the age of the crown group is at least 13.1 Ma (Givnish et al. 2011). Interestingly, Tepuihyla species seem to depend on water accumulation in phytotelmata of some species of Brocchinia (B. hectiodes and B. acuminata; Ayarzagüena et al. 1992), but no hypothesis of co-divergence of Tepuihyla and Brocchinia has been put forward.

We also examined divergence times of another Pantepui-endemic group, Stefania, for comparison to the Tepuihyla results. Stefania, like Tepuihyla, are highly dependent on the phytotelmata of pitcher plants. However, the reproductive biology of Stefania is quite distinct in that females are obligate dorsal egg brooders and the embryos undergo direct development (Salerno and Pauly 2012). In contrast, Tepuihyla have the more common reproductive mode of depositing eggs in bodies of water with the eggs hatching into a larval (tadpole) stage that undergoes metamorphosis (Ayarzagüena et al. 1992). Several authors have suggested that distributions and phylogenetic relationships in Stefania support a vicariant speciation scenario following the Lost World hypothesis (Hoogmoed 1979; MacCulloch and Lathrop 2002; McDiarmid and Donnelly 2005). However, as is the case for most Pantepui groups, these assessments were based on limited sampling and distributional data with relationships among individuals solely based on morphological comparisons. Our analysis of the available sequences (five out of 19 species) shows that the deepest divergence within Stefania is 34.4 Ma, and the youngest is 7.1 (Fig. 5). Interestingly, the youngest divergence is found between two lowland/midland species (S. scalae and S. evansi), which is the same pattern, though different timing, observed for the two Tepuihyla lowland/midland clades, T. rodriguezi/T. talbergae. Furthermore, these two species pairs have similar geographic distributions, which may indicate shared biogeographic histories. Even though the divergence estimates for Stefania clades are on average more deeply diverged than Tepuihyla, Stegolepidae, Brocchinia, and Lindmania, the divergences within Stefania indicate that this lineage radiated more recently than the dissection of the tepui summits (70–90 Ma). Thus, Stefania follows the general diversification pattern in Tepuihyla and Bromeliaceae.

Many endemic representatives of the “Lost World” are touted as “living fossils,” that is, a species or group of species that was formerly speciose or widespread in time and/or space, but has suffered extinction (Brown and Lomolino 1998). However, highly endemic clades are not necessarily “living fossils.” There is no evidence that the clades Tepuihyla, Stefania, Lindmania, Brocchinia, and Stegolepidae are remnants of ancient widespread lineages. These taxa are more appropriately termed neoendemics, that is, endemic taxa of relatively recent autochthonous origin (Brown and Lomolino 1998).

The frog Ceuthomantis smaragdinus from Mt. Kopinang, Guyana also is claimed to be a “living fossil” (Heinicke et al. 2009). In contrast to the previous examples, this species is deeply diverged from its extremely speciose sister taxon at about 60 Ma (Heinicke et al. 2009). Two other species are tentatively referred to this genus, but no sequences are available. Although this divergence time is much greater than that of the other taxa considered here, the divergence of only one species makes it difficult to determine the minimal age at which that lineage (one species) was present within the Pantepui. Thus, further work is needed to assess whether this lineage is a paleoendemic.


Many tepuis have sheer cliffs that have been proposed to be physical barriers to dispersal between low and high elevations, thus isolating summit taxa (Chapman 1931; Maguire 1970; Hoogmoed 1979). However, divergences among Tepuihyla species took place relatively recently, at least 60 Ma after the tepuis were fully formed, indicating that tepui walls are not a complete physical barrier between the summits and lowland forest even for organisms with low vagility.

Tepuihyla has an extremely fragmented distribution, but it is widespread in the Guiana Shield, occurring in three of the four Tepui provinces (Huber 1988). All species are endemic either to the summits or to mid-elevation regions of tepui remnants and are generally found in Rapateaceae- or Bromeliaceae-dominated meadows (Ayarzagüena et al. 1992). That Tepuihyla is absent from the lowlands between the tepuis suggests the habitat is currently unsuitable. Physiological constraints associated with increasing elevation (and thus decreasing temperature and increasing daily and yearly temperature variation) have been shown to enhance isolation among populations at different elevations, especially in ectothermic vertebrates (Navas 2006). The drastic elevational differences between tepui summits and the surrounding lowlands result in extreme differences in annual mean temperature, precipitation, soil composition, and phytogeographic regions (Steyermark 1979; Huber 2006). Drastic microclimatic and habitat differences between summits and lowlands may create a challenge for dispersal, thus restricting Tepuihyla to suitable mid- and high-elevation conditions. However, as stated in the Vertical Displacement hypothesis, these climatic conditions may have been different in the past, allowing for a more widespread lowland existence.

Our results indicate that the sheer escarpments of tepuis have not prevented dispersal of Tepuihyla species across the lowlands between the summits during the last 5.3 Ma. Furthermore, the first association of the ancestor of Tepuihyla+“O.”exophthalmus with the Pantepui likely occurred around 14.7 Ma, which deeply postdates tepui fragmentation. Because most Tepuihyla divergences are during the Pliocene (5.3–2.6 Ma), our results also highlight the importance of pre-Quaternary speciation (Rull 2008). However, we cannot completely reject the effects of Pleistocene glaciation, because the timing of the most recent divergences is consistent with climatic fluctuations. Our analysis clearly demonstrates dispersal of Tepuihyla to and/or from the tepui summits long after their formation. Furthermore, comparisons to other taxa such as Stefania and Bromeliaceae seem to indicate that dispersal has occurred across widely different tepui taxa, from pollen-dispersing plants to low-vagility organisms such as frogs. Thus, even though the Lost World hypothesis is attractive in nature and has been largely popular in the literature, so far there is no empirical dataset that shows unambiguous support for it.

Associate Editor: B. Fitzpatrick


We thank A. Wynn (United States National Museum) for tissue sample loans, the Ministerio del Poder Popular para el Ambiente (no. 2373, no. 0076, no. 41–0088) and Inparques (no. 0271) for permits, and C. Barrio-Amorós, G. Pauly, and members of the Cannatella laboratory for comments on the manuscript. This work was supported by funds from the American Philosophical Society, Explorers Club Exploration Grant, and the Ecology, Evolution and Behavior Graduate Program of the University of Texas to PES, and by U.S. National Foundation Grant EF0334952 to DCC and D. M. Hillis.


List of specimens. Specimens that were sequenced are bold and marked with an asterisk. All other specimens were obtained from GenBank. Field/museum codes are only shown for specimens sequenced herein. Field numbers correspond to Brice P. Noonan (BPN) and Patricia Salerno (PS). [Corrections added after online publication May 4, 2012: the numbers in column one of the Appendix were updated].

 TaxonField/museum codeGenBank accession
1Acris crepitansEF566969AY819109
2Aparasphenodon brunoi AY843567 
3Argenteohyla siemersiAY843570
4Corythomantis greeningi AY843578 
5Hyla arboreaAY843601DQ57787
6Hyla arenicolor EF566960 
7Hyla cinereaAY680271AY819116
8Hyla gratiosa AY843630GQ374915
9Hyla meridionalisEF566953GQ374915
10Hyla squirella AY843678AY819120
11Hyla versicolorEF566953DQ55805
12Itapotihyla langsdorffii AY843706AY843706
13Itapotihyla langsdorffii*USNM303287JQ686500JQ868470
14Nyctimantis rugiceps AY843781 
15Osteocephalus alboguttatusDQ380347
16Osteocephalus buckleyi DQ380378EUO34116
17Osteocephalus cabreraiAY843705
18Osteocephalus deridens*QCAZ20868JQ868501JQ868484
19Osteocephalus exophthalmus*BPN166JQ868523
20Osteocephalus exophthalmus*MHNLS19584JQ868525JQ868483
21Osteocephalus exophthalmus*MHNLS19583JQ868524
22Osteocephalus fuscifacies*QCAZ20790JQ868502
23Osteocephalus fuscifacies*QCAZ20788JQ868503JQ868499
24Osteocephalus leprieuriiAY843707
25Osteocephalus leprieurii AY549361 
26Osteocephalus leprieurii*MHNLS18689JQ868505JQ868497
27Osteocephalus leprieurii*MHNLS18619JQ868504JQ868498
28Osteocephalus mutaborDQ380379EUO34117
29Osteocephalus oophagus AF467267 
30Osteocephalus oophagusAY843708
31Osteocephalus planiceps DQ380380EUO43118
32Osteocephalus planiceps*QCAZ19195JQ868521JQ868495
33Osteocephalus planiceps*QCAZ20797JQ868522JQ868494
34Osteocephalus planiceps*QCAZ18844JQ868520JQ868496
35Osteocephalus alboguttatus*QCAZ18186JQ868516JQ868493
36Osteocephalus sp.*QCAZ38420JQ868526
37Osteocephalus cf. taurinus*USNM302469JQ868514 
38Osteocephalus cf. taurinus*USNMFS008803JQ868515
39Osteocephalus cf. taurinus*MHNLS18325JQ868506 
40Osteocephalus taurinusAY843709
41Osteocephalus taurinus*MHNLS18663JQ868509JQ868490
42Osteocephalus taurinus*MHNLS15622JQ868507JQ868492
43Osteocephalus taurinus*PS004JQ868512JQ868487
44Osteocephalus taurinus*MHNLS19633JQ868511
45Osteocephalus taurinus*QCAZ18839JQ868513JQ868488
46Osteocephalus taurinus*MHNLS17336JQ868508JQ868491
47Osteocephalus taurinus*MHNLS18715JQ868510JQ868489
48Osteocephalus taurinusAY326041AY819130
49Osteocephalus verruciger*QCAZ13225JQ868517JQ868486
50Osteocephalus verruciger*QCAZ17283JQ868518
51Osteocephalus verruciger DQ380381 
52Osteocephalus yasuni*QCAZ19245JQ868519JQ868485
53Osteopilus crucialis AY843710EUO34121
54Osteopilus dominicensisAY843711
55Osteopilus dominicensis AY819443EUO34122
56Osteopilus marianaeDQ380383EUO34123
57Osteopilus pulchrilineatus AY819436EUO34124
58Osteopilus septentrionalisAY843712AY819131
59Osteopilus vastus AY843713EUO34128
60Osteopilus wilderiDQ380385EUO34129
61Phyllodytes auratus AY819383AY819133
62Phyllodytes luteolusAY843721
63Phyllodytes sp. AY843722 
64Pseudacris cruciferAY291103EF988269
65Tepuihyla aecii*MHNLS12013JQ868533JQ868478
66Tepuihyla edelcaeAY843770
67Tepuihyla edelcae*PS002JQ868537JQ868475
68Tepuihyla edelcae*MHNLS16090JQ868534JQ868477
69Tepuihyla edelcae*MHNLS05824JQ868535 
70Tepuihyla edelcae*PS001JQ868536JQ868476
71Tepuihyla edelcae*PS268JQ868538 
72Tepuihyla rodriguezi*PS003JQ868540JQ868474
73Tepuihyla rodriguezi*MHNLS19575JQ868539 
74Tepuihyla sp.DQ380389EUO34131
75Tepuihyla talbergae*BPN1101JQ868541 
76Tepuihyla talbergae*BPN1219JQ868542JQ868473
77Trachycephalus coriaceus DQ380386EUO34130
78Trachycephalus hadrocepsAY843717
79Trachycephalus jordani*QCAZ17509JQ868527JQ868471
80Trachycephalus jordaniAY819395AY819145
81Trachycephalus jordani AY326042 
82Trachycephalus jordaniAY843771
83Trachycephalus mesophaeus AY843718 
84Trachycephalus nigromaculatusAY843772
85Trachycephalus resinifictrix*QCAZ20808JQ868528JQ868481
86Trachycephalus resinifictrixAY843719
87Trachycephalus resinifictrix*QCAZ19304JQ868529JQ868482
88Trachycephalus sp.*QCAZ21282JQ868530JQ868479
89Trachycephalus venulosus*QCAZ21283JQ868531JQ868480
90Trachycephalus venulosus*PS013JQ868532JQ868472
91Trachycephalus venulosus AY549362 
92Trachycephalus venulosusAY326048
93Trachycephalus venulosus AY819382 
94Trachycephalus venulosusDQ347027
95Trachycephalus venulosus AY364350