As a note of caution, it must be recalled that the present study makes a compilation of GenBank sequences, sequences that may not be devoid of errors. Problems may arise from taxonomic misidentification (Vilgalys, 2003; Stuart and Fritz, 2008; Fritz et al., 2010), sequencing errors (Harris, 2003), and pseudogene amplification (Fritz et al., 2010). We did not try to remove rogue taxa (Sanderson and Shaffer, 2002) from the analysis, because the maximum likelihood method we employed for phylogenetic inference, and the resulting confidence values do not make use of bootstrapping (Guindon and Gascuel, 2003; Anisimova and Gascuel, 2006). Only highly supported branches (cv > 0.9) will be discussed here.
Many studies assumed a priori the monophyly of Pleurodira and Cryptodira, and rooted the Testudines tree at the branch joining these two groups (Shaffer et al., 1997; Fujita et al., 2004; Thomson and Shaffer, 2010). Studies that used one or two outgroups (Gallus and/or Alligator) found ambiguous results concerning the monophyly of Pleurodira and Cryptodira, depending on the method used for phylogenetic inference (Cervelli et al., 2003; Krenz et al., 2005; Barley et al., 2010). Sterli (2010) recently recovered a sister relationship between Pleurodira and Trionychia, based on morphological characters and 12S, 16S, cytb, RAG1 and R35 intron. This unorthodox result may be due to the inclusion of extinct species in the analysis of morphological characters, resulting in the basal position of Chelonioidea as sister to all other extant Testudines lineages. Using outgroups from four clades (Squamata, Rhynchocephalia, Aves and Crocodylidae), we find here good support for a basic divergence of Pleurodira and Cryptodira. The early appearance of Trionychia within Cryptodira is also better supported than in previous molecular studies with outgroups (Krenz et al., 2005; Barley et al., 2010).
The phylogeny of Cryptodira has been a matter of debate (Shaffer et al., 1997; Fujita et al., 2004; Krenz et al., 2005; Chandler and Janzen, 2009; Thomson and Shaffer, 2010). We find a polytomy composed of four clades: (i) Chelonioidea (cv = 0.999), (ii) Chelydridae (cv = 1), (iii) Kinosternoidea (cv = 1), and (iv) a group formed by Geoemydidae, Testudinidae, Emydidae and Platysternon megacephalum (cv = 1). Various studies have found different topologies, either grouping Chelonioidea with Kinosternoidea (Fujita et al., 2004; using R35 intron), or grouping Chelonioidea with Geoemydidae, Testudinidae, Emydidae and Platysternon megacephalum (Parham et al., 2006; using complete mtDNA). As the number of species from Kinosternoidea included in phylogenies is usually low (here only five out of 26 species), DNA sequencing of more species from this clade could help clarify these relationships. However, the reasons why certain deep nodes are difficult to resolve are probably twofold: (i) a relatively rapid radiation, and (ii) phylogenetic analyses using genes saturated with mutations. Recently, Barley et al. (2010) provided new sequence data for eight nuclear genes, and found good support for a sister relationship of Chelonioidea to a group formed by Chelydridae and Kinosternoidea. We find the same relationships (cv > 0.9) after adding the sequences of Barley et al. (2010) to our alignment (results not shown).
The phylogenetic position of Platysternon megacephalum has long been enigmatic (Parham et al., 2006). Indeed, the first analyses, based on morphological characters, have led authors to relate P. megacephalum with Chelydridae (Chelydra serpentina and Macrochelys temminckii). However, some molecular studies found that P. megacephalum was grouped with Testudinoidea (Cervelli et al., 2003, based on U17 small nucleolar RNA; Krenz et al., 2005, based on 12S, cytB and RAG1). In the phylogeny that Parham et al. (2006) obtained with complete mitochondrial genomes, P. megacephalum was included in Testudinoidea as the sister-species to Emydidae. We here find good support (cv = 0.980) for the same hypothesis, as did Thomson and Shaffer (2010).
Within Pleurodira, a problematic taxon is Elseya, which appears polyphyletic, with the inclusion of Rheodytes leukops as sister to Elseya dentata (cv = 1), and of Elusor macrurus as sister to Elseya purvisi (cv = 0.929). The same position, albeit less supported, was already found for Rheodytes leukops by Seddon et al. (1997), based on 12S sequences. Thomson and Georges (2009) recently proposed the erection of a new genus, Myuchelys, including species previously named Elseya latisternum (type-species of the new genus), Elseya georgesi, Elseya purvisi and Elseya novaeguineae. More sequencing work would be useful to assess the monophyly of the resulting taxon, because it is not monophyletic in our phylogeny. Finally, Emydura is problematic, because Emydura macquarii is grouped with Elseya georgesi and Elseya latisternum (cv = 0.934), and not with Emydura subglobosa. However, it must be noted that Emydura subglobosa is here represented by only one sequence. The topology we find for Podocnemididae is in agreement with the one recently reported by Vargas-Ramirez et al. (2008), based on 12 genes (six mitochondrial and six nuclear).
Within Trionychia, our results are comparable with those of Engstrom et al. (2004), based on ND4, cytB and R35, except within Nilssonia. We find high support for a sister relationship between Nilssonia hurum and Nilssonia nigricans (cv = 0.988), and between Nilssonia gangeticus and Nilssonia leithii (cv = 0.933). This is in agreement with the results reported by Praschag et al. (2007a), based on cytB sequences. Within Kinosternoidea, Thomson and Shaffer (2010) found good support for a monophyletic Kinosternon. In contrast, we find a sister relationship between Kinosternon baurii and Sternotherus odoratus (cv = 0.923). However, only one DNA sequence is available for Kinosternon baurii, so this result is only preliminary and warrants further investigation. The phylogeny we find within Chelonioidea is identical to that already described (Naro-Maciel et al., 2008). Within Testudinidae, we find a sister relationship between Kinixys belliana and Kinixys spekii (cv = 0.985), whereas Thomson and Shaffer (2010) found good support for a close relationship between Kinixys belliana and Kinixys natalensis. The reason for this difference is unclear because gene sampling for Kinixys was presumably very similar in both studies. There is also significant disagreement between our phylogeny of Testudininae and the one by Lourenço et al. (2012), who found Indotestudo and Malacochersus nested within Testudo. This may result from different species sampling because Lourenço et al. (2012) did not include Testudo hermanni and Indotestudo travancorica in their analysis. This is not the first time that Homopus is found to be polyphyletic. Thomson and Shaffer (2010) found the same relationships for Homopus, showing Homopus signatus and Homopus boulengeri grouped with Chersina angulata with good support, and Homopus aerolatus and Homopus femoralis grouped with Psammobates with lower support. We suggest that a taxonomic revision could be useful here.
A taxon that has been plagued by nomenclatural problems is Geoemydidae. Based on molecular phylogenies, Spinks et al. (2004) made three nomenclatural suggestions. They proposed (i) including all Chinemys and Ocadia species in Mauremys, (ii) re-including Chelopus annulata and Chelopus rubida in the genus Rhinoclemmys, and (iii) classifying K. tecta, K. tentoria and K. smithii as members of a new genus, Pangshura. All these propositions are supported by our phylogeny and result in monophyletic genera. However, the taxonomic position of the remaining species of Kachuga (K. dhongoka, K. kachuga, K. sylhetensis and K. trivittata) is still problematic. As Le et al. (2007) and Praschag et al. (2007a) already recommended, we propose the inclusion of Kachuga sylhetensis in Pangshura, with which it is clearly grouped (cv = 0.999), and the inclusion of Callagur borneoensis, Kachuga kachuga, Kachuga dhongoka, and Kachuga trivittata in Batagur. Within Geoemydinae, the taxonomy within Cuora and Cyclemys has long been uncertain. Here we find Cuora trifasciata grouped with Cuora aurocapitata (cv = 1) and Cuora galbinifrons grouped with Cuora mccordi (cv = 0.943), in contrast with Honda et al. (2002), using 12S and 16S sequences, or Stuart and Parham (2004) and He et al. (2007), using COI and ND4. The phylogeny we obtain for Cuora is consistent with the one obtained by Spinks and Shaffer (2007), using COI and ND4 (but not what they obtained with nuclear DNA). Noticeably, we here used complete mtDNA for Cuora aurocapitata, Cuora flavomarginata and Cuora mouhotii. Uncertainty of specimen identification or hybridization between species, as is known to occur in Cuora, may also explain these discrepancies (Spinks and Shaffer, 2007). Within Cyclemys, we find close relationships between a group formed by Cyclemys atripons and Cyclemys pulchristriata, and a group formed by Cyclemys bellii, Cyclemys enigmatica, Cyclemys dentata and Cyclemys ovata. This contrasts with the topologies found by Praschag et al. (2009) and Fritz et al. (2008), using cytB, c-mos, RAG2 and R35 intron. However, these studies obtained different topologies when analyzing separately mitochondrial and nuclear sequences, indicating that mitochondrial introgression may have occurred through hybridization. Finally, we suggest that the Batagurinae and Geoemydinae should be re-delimited, since Geoemydinae is currently not monophyletic (Table S2, Fig. 2d).
Within Emydidae, Clemmys guttata has been consistently recovered in two different positions: (i) as sister to Terrapene (Feldman and Parham, 2002, based on cytB and ND4; Stephens and Wiens, 2003, based on 16S, ND4 and cytB; Spinks and Shaffer, 2009, based on cytB, R35, RAG1 plus five more nuclear loci) or (ii) as sister to the Emys + Emydoidea clade (Stephens and Wiens, 2009, based on morphological data, 16S, cytB, ND4, control region, and R35; Wiens et al., 2010, based on cytB, ND4, R35 plus five more nuclear loci). Our results support the first hypothesis (cv = 0.930). However, the second hypothesis for Clemmys guttata was favored by data sets that were not totally included in ours, so we cannot reject it. Our results on Deirochelyinae are in good agreement with previous topologies (Stephens and Wiens, 2003, 2009; Wiens et al., 2010), but differ from those reported by Spinks et al. (2009), using seven nuclear loci including R35 and RAG1. We find Trachemys monophyletic (cv = 1), grouped with Graptemys and Malaclemmys terrapin (cv = 0.999), and Pseudemys grouped with Chrysemys picta (cv = 0.995). Within Graptemys and Pseudemys, relationships are poorly resolved because mitochondrial DNA seems to exhibit low divergence between species (Wiens et al., 2010), and we have excluded fast evolving DNA regions from our 230-species alignment.
A parsimony analysis of the DNA matrix was also performed, and yielded a less resolved tree, with low bootstrap support for some clades that were well resolved in maximum likelihood analysis (results not shown). However, the monophyly of Pleurodira and Cryptodira is supported (bootstrap support = 0.81 and 0.76, respectively), as is the sister relationship of Trionychia (bootstrap support = 0.85) to a group formed by all other Cryptodira (bootstrap support = 0.87). When only bootstrap support > 0.7 is taken into account, the parsimony tree is fully compatible with the one obtained with maximum likelihood.
Our study has provided the largest phylogeny of turtles to date. By using both mtDNA and nuDNA data, we find that most genera are now monophyletic, with strong support, but we suggest some nomenclatural revisions and point at specific taxa that warrant further sequencing work. Polytomies still observed in our phylogeny (cv < 0.5) are related with the species with the lowest number of sequences. On the 21 species with only one gene sequenced, one third is directly involved in a polytomy (this proportion is 0.11 when considering all species). New sequences from these seven species (Cyclemys ovata, Emydura subglosa, Graptemys oculifera, Graptemys versa, Pseudemys alabamensis, Pseudemys gorzugi, Pseudemys suwanniensis) should be obtained in priority to better resolve the phylogeny. Because all but two turtle genera are represented in our phylogeny, our work provides a solid basis to help in further studies of the evolution of some characters in turtles or the ancient biogeographical distribution of turtles.