INTRODUCTION

The Bataguridae is the largest and phenotypically most divergent family in the order Testudines, and is distributed mainly from the temperate to tropical Asia but also in southwestern Europe, northern Africa and Latin America (Iverson, 1992). This family consists of two subfamilies: Batagurinae and Geoemydinae (Gaffney and Meylan, 1988). The former accommodates 22 extant species of 12 genera (David, 1994; McCord and Iverson, 1994), whereas the latter 43 species of 13 genera (David 1994; McCord et al., 1995, 2000; Fritz and Wischuf, 1997; Fritz et al., 1997; McCord, 1997; Wink et al., 2000; Parham et al., 2001; Yasukawa et al., 2001).

Within the Bataguridae, kinetic plastron with a hinge between the pectoral and the abdominal is observed in five geoemydine genera: Cistoclemmys Gray, 1863 (type species: Ci. flavomarginata), Cuora Gray, 1855 (type species: Cu. amboinensis), Cyclemys Bell, 1834 (type species: Cy. dentata), Notochelys Gray, 1863 (monotypic with N. platynota), and Pyxidea Gray, 1863 (monotypic with P. mouhotii). These turtles are well known as Asian “box turtles”. Bramble (1974) pointed out the absence of substantial differences in the plastral hinge structure between Cuora and Cistoclemmys, and Sites et al. (1984), on the basis of allozymic analyses, suggested a closest affinity between these two genera. Based on these pieces of information, some authors regard Cistoclemmys as a junior synonym of Cuora (Ernst and Barbour, 1989; McCord and Iverson, 1991; Iverson, 1992; David, 1994; Fritz and Obst, 1997: Wu et al., 1999). Fritz and Obst (1997) further doubted the separation Pyxidea from Cuora sensu lato because of states of a few morphological characters in Ci. serrata that are intermediate between states of corresponding characters in Cistoclemmys and Pyxidea. However, this taxonomic account lost its base with the elucidation of actual origin of Ci. serrata as an intergeneric hybrid (Parham et al., 2001).

On the other hand, cladistic analyses of morphological characters suggested a distant location between Cistoclemmys and Cuora sensu stricto, and a closest affinity of the former with Geoemyda and Pyxidea (Fig. 1A: Hirayama 1984; Gaffney and Meylan, 1988), or solely with Pyxidea (Fig. 1B: Yasukawa et al., 2001). Furthermore, Hirayama (1984) went so far as to suspect that the Testudinidae had derived from an ancestor exclusively common to some members of the Geoemydinae (i.e., some Rhinoclemmys species, Geoemyda, Pyxidea and Cistoclemmys: Fig. 1A). This view lead to serious controversy concerning the extent of application of the generic name Cuora, and several authors regarded Cistoclemmys as valid (Bour, 1980; Hirayama, 1984; Gaffney and Meylan, 1988; King and Bruke, 1989; Ota, 1991; Zhao and Adler, 1993; Yasukawa and Ota, 1999; Yasukawa et al., 2001).

Fig. 1

Phylogenetic hypotheses of the Geoemydinae based on morphological characters. Circles denote positions of evolution of the plastral hinge. (A) Hirayama (1984: this view was also favored by Gaffney and Meylan [1988]). (B) Yasukawa et al. (2001). (C) Bramble (1974).

In these two decades, the genus Cuora sensu lato have also been receiving great attentions from a viewpoint of the species taxonomy, leading to descriptions, synonymizations and revalidations involving a large proportion of the species included (see McCord and Iverson [1991] for example). Moreover, the genus seems to be an ideal material for bio-geographical studies, because members of Cuora sensu lato occur not only in the continental part but also over almost whole of the insular region of Southeast and East Asia (Iverson, 1992).

Recently molecular studies yielded advanced phylogenetic hypotheses for some badgered turtles (e.g., Shaffer et al., 1997; Wu et al., 1999; McCord et al., 2000). However, none of these studies could appropriately address the phylogenetic, taxonomic and biogeographical problems of Cuora sensu lato, mainly due to the insufficiency in samples analyzed therein.

In the present study, we partially sequenced mitochondrial 12S and 16S ribosomal RNA (rRNA) genes for samples representing various testudinoid taxa including all but one species of Cuora sensu lato, and analyzed resultant data phylogenetically. Our purpose is to establish robust phylogenetic hypotheses for the Bataguridae with a particular emphasis on the interrelationships of Cuora and the presumptive relatives (see above). Also, we discuss implications of the resultant phylogeny on the taxonomy and biogeography of the Asian box turtles.

MATERIALS AND METHODS

Samples representing five of the six extant of Cuora sensu stricto, and the two species and one subspeices of Cistoclemmys (see McCord and Iverson, 1991; Yasukawa and Ota, 1999; Hilton-Taylor, 2000) were examined. We failed to examine specimens of Cu. mccordi due to the rarity of this recently described continental species (Ernst, 1988). Representatives of other genera of the Bataguridae, as well as those of the Emydidae, Testudinidae and other families, were also subjected to the analyses on the basis of newly obtained (Table 1, see Appendix for further details) and previously published data (Honda et al., 2002: Table 1). Of these, Staurotypus triporcatus of the Kinosternidae and Pelodiscus sinensis of the Trionychidae were regarded as outgroups, because these families are considered to be basal to the superfamily Testudinoidea (Gaffney and Meylan, 1988).

Table 1

Batagurid and other species used for the present analyses. Data source are (a) Honda et al. (2002); (b) this study. See Appendix for further details.

Extraction, amplification and sequencing procedures of DNA are described in detail elsewhere (Honda et al., 1999a,b). A part of mitochondrial 12S and 16S rRNA genes were amplified using the polymerase chain reaction (PCR) with primers L1091, H1478, L2606 and H3056 (Kocher et al., 1989; Hedges et al., 1993). Alignments for DNA sequences were determined based on maximum nucleotide similarity following Honda et al. (2000). The sequence data have been submitted to the DDBJ database, and their accession numbers are given in Table 1. Aligned data will be provided by the corresponding author on request.

The neighbor-joining (NJ) method (Saitou and Nei, 1987) was applied to infer relationships among taxa on the basis of a pairwise matrix of the distance from Kimura's (1980) two-parameter model. The maximum-likelihood (ML, with empirical base frequencies and equal rate substitution model) and maximum-parsimony analyses (MP, no bias between transition and transversion) were also conducted using heuristic search option. These analyses were performed by use of PAUP* 4.0b (Swofford, 1998) excluding gap sites, and confidences of branches were assessed by bootstrap resamplings (Felsenstein, 1985).

RESULTS

The 12S rRNA fragment consisted of 410 total sites, 168 of which were variable. For the 16S rRNA fragment, there were 472 total aligned sites, 178 of which were variable. Intergeneric nucleotide replacements within the Bataguridae varied from 16 bp (Chinemys reevesii vs. Mauremys caspica) to 95 bp (Callagur borneoensis vs. Notochelys platynota). Interspecific nucleotide replacements involved 34–35 bp, 2–47 bp, 20–26 bp and 2–47 bp in Cistoclemmys, Cuora, Mauremys and the Cuora group (assemblage of Cuora, Cistoclemmys and Pyxidea: see below), respectively. In continental Cuora (Cu. aurocapitata, Cu. pani, Cu. trifasciata and Cu. zhoui), nucleotide substitutions were observed in 2–11 bp, whereas three base-pairs were replaced between representatives of the two subspecies of Ci. flavomarginata.

The NJ dendrogram derived from the Kimura's (1980) distance matrix from aligned sequences is shown in Fig. 2A. The monophyly of the Testudinoidea was strongly supported (node 1: bootstrap proportion [BP] = 98%). The ingroup portion of this dendrogram was divided into two major clusters: the Emydidae (node 2: 100%), and an assemblage of the Testudinidae and the Bataguridae (node 3: 87%). The latter further split into two clusters, of which node 4 (99%) consisted of testudinid genera, whereas node 5 (93%) accommodated all the batagurid species examined. Within this last node, the Callagur—Geoemyda—Hardella—Orlitia assemblage (node 6: 94%) diverged first, followed by the Cyclemys—Heosemys—Notochelys—Sacalia (node 8: 84%) and the Chinemys—Mauremys assemblage (node 10: 93%) in order, leaving the Cistoclemmys—Cuora—Pyxidea assemblage as a monophyletic group (node 11: 100%).

Fig. 2

(A) Neighbor-joining (NJ) dendrogram derived from distance matrix from 12S and 16S rRNA sequence data. Numbers beneath branches are BPs at least 50% of the 2,000 bootstrap replications. Nodes with bold numbers indicate relationships referred to in the text. Bar equals 0.1 unit of Kimura's two-parameter distance. (B) Maximum-likelihood (ML) dendrogram (ln likelihood=–6152.8). Branches without BP values were not supported in ≥50% of the 100 bootstrap replicates. Bar equals 0.1 unit. Bold numbers above branches are identical with those in NJ and MP dendrograms. (C) Maximum parsimony (MP) dendrogram using heuristic using heuristic option (strict consensus of four parsimonious trees, 894 steps, 208 bp informative under the condition of parsimony, consistency index=0.50, homoplasy index=0.50, retention index=0.54). Branches without BP values were not supported in ≥50% of the 2,000 bootstrap replicates.

Both the two subspecies of Ci. flavomarginata and members of Chinese Cuora (Cu. aurocapitata, Cu. pani, Cu. trifasciata and Cu. zhoui) constituted monophyletic groups with complete BP (100%) within the node 11. In contrast, monophyly was not supported for Cistoclemmys and Cuora sensu stricto due to the initial divergence of Ci. galbinifrons from the remaining node 11 components and an apparently distant affinity of Cu. amboinensis with the Chinese congeners, respectively.

Relationships resulting from ML (Fig. 2B) and MP analyses (Fig. 2C) were consistent with those expressed in the NJ dendrogram in terms of topology of the nodes 1–11, although they failed to support three of them (nodes 5, 7 and 8) with substantial BP values (i.e., >50%). Also, both analyses, while supporting monophyly of the two Ci. flavomarginata subspecies and of the Chinese Cuora with complete or nearly complete BPs, did not offer any support to the monophyly of Cistoclemmys and Cuora sensu stricto.

DISCUSSION

Bramble (1974), on the basis of the shell-closing mechanism, and morphology of the pelvic girdle and Musclus testoscapularis, assumed that Cuora sensu lato (as consisting of Cuora sensu stricto and Cistoclemmys), Cyclemys and Pyxidea diverged from a Heosemys-like ancestor (Fig. 1C). He also noted that there were no substantial differences in the plastral hinge mechanism between Cu. amboinensis and Ci. flavomarginata. This has been one of the major reasons to regard Cistoclemmys as a junior synonym of Cuora (see above). However, phylogenetic analyses of morphological data by Hirayama (1984) and Yasukawa et al. (2001) strongly suggested that Cistoclemmys constitutes a monophyletic group with Geoemyda and Pyxidea, and thus is distinct from Cuora (Fig. 1A, B).

Based on sequence data of mitochondrial cytochrome b gene, McCord et al. (2000) provided a phylogram showing the monophyly of Cuora, Pyxidea and three other genera exclusive of Geoemyda. This relationship contradicts with the hypothesis by Hirayama (1984) and Yasukawa et al. (2001) (see above). In the results of McCord et al. (2000), however, phylogenetic positions of the batagurid box turtles remained equivocal because McCord et al. combined those genera, including Cuora sensu lato, into one operational taxonomic unit, and did not illustrate any relationships among them. Our results thus provide for the first time a substantial support to the monophyly of Cistoclemmys, Cuora sensu stricto and Pyxidea (henceforth referred to as the Cuora group) (contra Bramble 1974; Hirayama, 1984; Yasukawa et al., 2001), and detailed relationships among their species.

Bramble (1974) postulated that in the Bataguridae the plastral hinge evolved in the common ancestor of Cistoclemmys, Cyclemys, Cuora and Pyxidea (Fig. 1C), and also in the lineage leading to Notochelys (not shown). However, results of the morphological analyses by Hirayama (1984) and Yasukawa et al. (2001), respectively, suggest that the hinge independently evolved five and three times in this family (Fig. 1A, B). Present results support none of these hypotheses, because they suggest that the plastral hinge emerged in the common ancestor of the Cuora group and supposedly in the common ancestor of Cyclemys and Notochelys (Fig. 2).

Phylogeny within the Cuora group resulting from our analyses is a great surprise in that it supports the monophyly for none of the genera Cistoclemmys, and Cuora in broad and strict senses so far proposed. Cistoclemmys flavomarginata is known to share a unique shape of the iliac blade exclusively with Ci. galbinifrons and this character state has thus been considered as a synapomorph of the genus (Yasukawa et al., 2001). However, relationships depicted in Fig. 2 suggest that this derived character state actually evolved independently in lineages leading to the two species.

Yasukawa and Ota (1999), on the basis of slight but somewhat consistent morphological differences, argued the recognition of the southern Ryukyu populations of Ci. flavomarginata as a valid subspecies, Ci. f. evelynae. Small, but yet substantial sequence divergence between the southern Ryukyu and Taiwanese samples of the species examined here seems to be congruent with this taxonomic account. It is probable that such morphological and genetic divergences between Ci. flavomarginata populations from southern Ryukyus and Taiwan correspond to the vicariance of these two areas through the insularization of the former, which supposedly took place during the middle-late Pleistocene (Ota, 1998).

With respect to the continental members of Cuora examined here, our analyses yielded robust supports not only for their monophyly, but also for their remarkably low sequence divergence which is almost comparable to the divergence between the two subspecies of Ci. flavomarginata. Species of Cuora distributed in the continent (China and southern Vietnam) are morphologically quite similar to each other and are diagnosed chiefly on the basis of distinct coloration (McCord and Iverson, 1991), whereas Cu. amboinensis, the only species of the genus occurring broadly in the Sunda Islands and the Philippines (Iverson, 1992), exhibits a greater number of morphological characters in unique states (e.g., McCord and Iverson, 1991; Yasukawa et al., 2001). Our results suggest that the ancestral lineage of Cu. amboinensis split relatively early in the Cuora group, and that speciations in the continental Cuora took place much later than that event, or speciations in other related genera, such as Mauremys (Fig. 2).

Our data, while strongly supporting the monophyly of the Cuora group as a whole, did not support any of the previous generic arrangements within the group. Unequivocal inclusion of Pyxidea mouhotii within the Cuora group and its closest affinity to Cuora sensu lato exclusive of Ci. galbinifrons are of particular interests, because the monotypic Pyxidea has never been synonymized with Cuora or Cistoclemmys, and its species was once assigned to Cyclemys (Wermuth and Mertens, 1961; Prichard, 1979; Xiangkui, 1994) or Geoemyda (McDowell, 1964). Considering this, as well as the non-monophyly of Cistoclemmys and Cuora sensu stricto, we suggest the application of the generic name Cuora to all species of the Cuora group by synonymizing Cistoclemmys and Pyxidea with this genus. This view definitely needs further verifications though.

McDowell (1964) suspected that the Bataguridae (as the subfamily Batagurinae of the family Testudinidae) was derived from a common ancestor with the Testudinidae as exclusive of the Emydidae (Emydinae of the Testudinidae). While most subsequent authors explicitly or implicitly assumed a closer affinity of the batagurid turtles with the emydids (e.g., Pritchard, 1979; Ernst and Barbour, 1989; Iverson, 1992), Hirayama (1984) progressed McDowell's (1964) view by arguing the sister relationship between an assemblage of some genera of the batagurid subfamily Geoemydinae and the Testudinidae on the basis of cladistic analyses of morphological characters (Fig. 1A). This view was approved by Gaffney and Meylan (1988).

Based on both molecular and morphological data, Shaffer et al. (1997) depicted the sister relationship between the Testudinidae and the Bataguridae. However, because their material included only two non-geoemydine batagurid genera, they failed to test Hirayama's (1984) hypothesis in a strict manner. Results of our analyses by incorporating data for most taxa crucial to those and related problems, while supporting McDowell's (1964) view for the sister relationships of the Bataguridae and the Testudinidae, negate Hirayama's (1984) view by supporting the monophyly of the Bataguridae. Also, our results do not support any of the sub-familial classifications and other generic grouping within the Bataguridae so far proposed on the morphological ground (Table 1). On the other hand, our results are largely concordant with those of more preliminary molecular studies dealing with turtles (Wu et al., 1999; McCord et al., 2000; Honda et al., 2002). We thus consider that in the testudinoid turtles morphological evolution often leads to an extensive homoplasy presumably under the operation of environmental selective forces.