Pternohyla fodiens, Boulenger, 1882

FAIVOVICH, JULIÁN, HADDAD, CÉLIO F. B., GARCIA, PAULO C. A., FROST, DARREL R., CAMPBELL, JONATHAN A. & WHEELER, WARD C., 2005, Systematic Review Of The Frog Family Hylidae, With Special Reference To Hylinae: Phylogenetic Analysis And Taxonomic Revision, Bulletin of the American Museum of Natural History 2005 (294), pp. 1-240 : 37-48

P. fodiens . In this analysis we include P. fodiens .

Plectrohyla: This genus was reviewed by Duellman (2001) and discussed by McCranie and Wilson (2002). Its phylogenetic relationships were addressed by Duellman and Campbell (1992), Wilson et al. (1994a), and Duellman (2001). The monophyly of Plectrohyla does not appear to be controversial. Duellman and Campbell (1992) listed six synapomorphies: bifurcated alary process of premaxilla; sphenethmoid ossified anteriorly, incorporating the septum nasi and projecting forward to the leading margins of the nasals; frontoparietals abutting broadly anteriorly and posteriorly, exposing a small area of the frontoparietal fontanelle; hypertrophied forearms; and absence of lateral folds in the oral disc. Wilson et al. (1994a) added ‘‘prepollex enlarged, elongated, ossified, flat, terminally blunt.’’ Duellman (2001) interpreted that this definition corresponded to more than one character, and so he divided it into two characters, the derived state of the first one being ‘‘enlarged and ossified prepollex in both sexes’’, and the derived state of the second one being ‘‘enlarged and truncate prepollex.’’ See comments under the Hyla bistincta group.

Plectrohyla currently contains 18 species: P. acanthodes , P. avia , P. chrysopleura , P. dasypus , P. exquisitia , P. glandulosa , P. guatemalensis , P. hartwegi , P. ixil , P. lacertosa , P. matudai , P. pokomchi , P. psiloderma , P. pycnochila , P. quecchi , P. sagorum , P. tecunumani , and P. teuchestes . In this analysis we include P. guatemalensis , P. glandulosa , and P. matudai .

Ptychohyla: This group was reviewed by Duellman (2001). Campbell and Smith (1992) suggested three synapomorphies for Ptychohyla : the presence of two rows of marginal papillae, an increased number of tooth rows in larvae (from 3/5 to 6/9), and a strongly developed lingual flange of the pars palatina of the premaxilla. Duellman (2001) also suggested as synapomorphies the presence of ventrolateral glands in breeding males, and the coalescence of tubercles to form a distinct ridge on the ventrolateral edge of the forearm. The increase in the number of tooth rows could actually be a synapomorphy not of Ptychohyla but for a more inclusive clade containing Ptychohyla plus other species of stream­breeding hylids that also have a tooth row formula larger than 2/3. Similarly, ventrolateral glands are present also in some species of Duellmanohyla (Campbell and Smith, 1992; Duellman, 2001).

For an unstated reason, Savage (2002a) excluded Ptychohyla legleri and P. salvadorensis from Ptychohyla , placing them back in Hyla . These two species were originally in Hyla (former H. salvadorensis group; see Duellman, 1970) until Campbell and Smith (1992) transferred them to Ptychohyla . Because we are not aware of any evidence supporting Savage’s action, we consider them to be members of Ptychohyla .

Ptychohyla is composed of 12 species: P. acrochorda , P. erythromma , P. euthysanota , P. hypomykter , P. legleri , P. leonhardschultzei , P. macrotympanum , P. panchoi , P. salvadorensis , P. sanctaecrucis , P. spinipollex , and P. zophodes . In this analysis we include P. euthysanota , P. hypomykter , P. leonhardschultzei , P. spinipollex , P. zophodes , and Ptychohyla sp. , an undescribed species from Oaxaca, Mexico.

Smilisca: This genus was reviewed by Duellman and Trueb (1966) and Duellman (1970, 2001). Duellman (2001) could advance no evidence for the monophyly of Smilisca . He presented a phylogenetic analysis rooted with a hypothetical ancestor, whose strict consensus showed Pternohyla plus Triprion nested within Smilisca , being more closely related to S. baudinii and S. phaeota . The synapomorphies supporting Pternohyla 1 Triprion 1 ‘‘ Smilisca ’’ are the presence of lateral flanges on the frontoparietals, and the unexposed frontoparietal fontanelle. The species of Smilisca have been divided (Duellman and Trueb, 1966) into the S. sordida group ( S. puma and S. sordida ), the S. baudinii group ( S. baudinii , S. cyanosticta , and S. phaeota ), and S. sila , a form considered intermediate between these two groups. In Duellman’s (2001) phylogenetic analysis, S. sila plus the S. sordida group is monophyletic, with its synapomorphy being the short maxillary process of the nasal. Smilisca contains six species: S. baudinii , S. cyanosticta , S. phaeota , S. puma , S. sila , and S. sordida . In this analysis we include the three species in the S. baudinii group, S. bau­ dinii, S. cyanosticta , S. phaeota , and one species of the S. sordida group, S. puma .

Triprion: This genus was reviewed by Trueb (1969) and Duellman (1970, 2001). In Duellman’s (2001) phylogenetic analysis of Pternohyla , Smilisca , and Triprion , the monophyly of Triprion is supported by three synapomorphies 20: maxilla greatly expanded laterally, prenasal bone present, and presence of parasphenoid odontoids. See comments for Smilisca and Pternohyla . Triprion is composed of two species, T. petasatus and T. spatulatus . In the analysis we include T. petasatus .

Casque­Headed Frogs and Related Genera

Duellman’s (2001) suggestion of Middle American/Holarctic frogs being monophyletic clearly separates the Middle American casque­headed frogs ( Triprion , Pternohyla ) from the South American and West Indian casque­headed frogs. This is not surprising considering that traditionally the group known as the casque­headed frogs was considered to be nonmonophyletic (Trueb, 1970a, 1970b). However, the position of the South American and West Indian casqueheaded frogs remains controversial, and no author has presented evidence indicating whether they form a monophyletic group.

Aparasphenodon: This genus of casqueheaded frogs was reviewed and characterized by Trueb (1970a) and Pombal (1993). The presence of a prenasal bone is a likely synapomorphy of Aparasphenodon (with a known homoplastic occurrence in Triprion , as reported by Trueb, 1970a). This genus currently comprises three species, A. bokermanni , A. brunoi , and A. venezolanus . We include A. brunoi in the analysis.

Argenteohyla: This monotypic genus was described and reviewed by Trueb (1970b), who segregated it from Trachycephalus , where it had been placed by Klappenbach (1961). Motives for this segregation were the absence in Argenteohyla of several character states of Trachycephalus as redefined by Trueb (1970a), such as the dermal spheneth­

20 On his preferred tree (his fig. 410), one of the character transformations is numbered 18; this is a typographical error for 12, the only other character that supports this clade but not shown in the tree.

moid, the poorer development of ossification and cranial sculpturing, and vocal sacs that when inflated protrude posteroventrally to the angles of the jaw. Possible autapomorphies of this taxon include the fusion of the zygomatic ramus of the squamosal with the pars facialis of the maxilla. The genus comprises a single species, A. siemersi , for which a northern subspecies, A. s. pederseni, was described by Williams and Bosso (1994). In this analysis we included a specimen that corresponds to the northern form.

Corythomantis: This monotypic genus was reviewed by Trueb (1970a). Autapomorphies of this genus include the absence of palatines, and nasals that conceal the allary processes of premaxillaries (Trueb, 1970a). We include the single species Corythomantis greeningi in this analysis.

Osteocephalus: This genus was diagnosed by Goin (1961) and Trueb (1970a) and studied in detail by Trueb and Duellman (1971). These authors recognized five species: Osteocephalus verruciger , O. taurinus , O. buckleyi , O. leprieurii , and O. pearsoni . In the last 20 years, several new species were described, adding to a total of 18 currently recognized species (see Jungfer and Hödl, 2002; Lynch, 2002). Trueb and Duellman (1971) employed 20 character states to characterize Osteocephalus . Jungfer and Hödl (2002) modified some of these characters to take into account subsequently discovered species. As stated by Ron and Pramuk (1999), referring to the diagnostic states employed earlier by Trueb and Duellman (1971), it is unclear which, if any, of the character states are synapomorphic for the genus. Trueb (1970a) and Trueb and Duellman (1971) suggested, based on the presence of paired lateral vocal sacs in the five species then recognized, that Osteocephalus was related to a group composed of Argenteohyla , Trachycephalus , and Phrynohyas .

Martins and Cardoso (1987) described Ostecephalus subtilis that, unlike the other species known at that time, is characterized by a single, subgular vocal sac that expands laterally; a similar morphology was described by Smith and Noonan (2001) in O. exophthalmus . Jungfer and Schiesari (1995) described O. oophagus , a species with a single, median vocal sac, a reproductive mode in­ volving oviposition in bromeliads, and phytotelmous oophagous larvae. Jungfer et al. (2000), Jungfer and Lehr (2001), and Lynch (2002) described four species, O. deridens , O. fuscifacies , O. leoniae , and O. heyeri , which also have a single, median vocal sac. According to Lynch (2002), O. cabrerai also shares this characteristic. Reproductive modes are unknown for O. cabrerai , O. exophthalmus , O. heyeri , O. leoniae , and O. subtilis ; spawning in bromeliads is suspected for O. deridens and O. fuscifacie s (Jungfer et al., 2000). Note that Lynch (2002) doubted a possible relationship between O. heyeri and what he called the ‘‘presumed clade of oophagous species’’ (where he included O. deridens , O. fuscifacies , and O. oophagus ), suggesting instead that it could be related to what he called O. rodriguezi (at that time already transferred to the new genus Tepuihyla by Ayarzagüena et al., ‘‘1992’’ [1993b]). While the species known or suspected to spawn in bromeliads could be monophyletic, we are not aware of any synapomorphy supporting the monophyly of all remaining species of Osteocephalus .

The species currently included in Osteocephalus are O. buckleyi , O. cabrerai , O. deridens , O. elkejungingerae , O. exophthalmus , O. fuscifacies , O. heyeri , O. langsdorffii , O. leoniae , O. leprieurii , O. mutabor , O. oophagus , O. pearsoni , O. planiceps , O. subtilis , O. taurinus , O. verruciger , and O. yasuni . Considering the uncertainties regarding Osteocephalus , we attempted to include representatives of the morphological and reproductive diversity within the genus: O. cabrerai , O. langsdorffii , O. leprieurii , O. oophagus , and O. taurinus .

Osteopilus: The genus Osteopilus was resurrected by Trueb and Tyler (1974) for three apparently related species that were often referred to collectively as the Hyla septentrionalis group (see Dunn, 1926; Trueb, 1970a). Trueb and Tyler (1974) provided a diagnostic definition of the genus; a possible synapomorphy is the differentiation of the m. intermandibularis to form supplementary apical elements. Trueb and Tyler (1974) also maintained, due to the impressive morphological divergence, that Osteopilus , the other Antillean groups then considered to be in Hyla ( H. heilprini , H. marianae , H. pulchrilineata , H. vasta , H. wilderi ), and the new genus they erected, Calyptahyla , represented several independent invasions from the mainland.

Maxson (1992) and Hass et al. (2001), using albumin immunological distances, suggested that Osteopilus is paraphyletic with respect to most other West Indian hylids (with the exception of Hyla heilprini , a Gladiator Frog). Hedges (1996) mentioned that unpublished DNA sequence data confirmed these findings. Anderson (1996) presented a karyological study of the three species of Osteopilus , indicating that her data were compatible with a monophyletic Osteopilus . Based on the comments by Hedges (1996), and immunological results of Hass et al. (2001), Franz (2003), Powell and Henderson (2003a, 2003b), and Stewart (2003) transferred Calyptahyla crucialis , H. marianae , H. pulchrilineata , H. vasta , and H. wilderi to Osteopilus that now includes eight species. Osteopilus is grouped together only on the basis of the immunological distance results, as no discrete character data set supporting its monophyly has yet been published. The species of Osteopilus available for our study were O. crucialis , O. dominicensis , O. septentrionalis , and O. vastus .

Phrynohyas: This genus was reviewed by Duellman (1971b). Although very distinctive externally, the only seeming synapomorphy in the diagnostic definition of Phrynohyas provided by Duellman (1971b) is the extensively developed parotoid glands in the occipital and scapular regions. Likely related to this character state, the viscous, milky secretions of the species of this genus could also be considered synapomorphic. Lescure and Marty (2000) transferred Hyla hadroceps to this genus; this was confirmed in a phylogenetic analysis using the mitochondrial ribosomal gene 12S by Guillaume et al. (2001). Pombal et al. (2003) described P. lepida . See Osteocephalus for further comments. Phrynohyas currently contains seven species: P. coriacea , P. hadroceps , P. imitatrix , P. lepida , P. mesophaea , P. resinifictrix , and P. venulosa . In our analysis we include P. hadroceps , P. mesophaea , P. resinifictrix , and P. venulosa .

Tepuihyla: This genus was defined by Ayarzagüena et al. (‘‘1992’’ [1993b]) for five species of Osteocephalus previously consid­ ered to constitute the O. rodriguezi species group (Duellman and Hoogmoed, 1992; Ayarzagüena et al., ‘‘1992’’ [1993a]). Ayarzagüena et al. (‘‘1992’’ [1993b]) differentiated Tepuihyla from Osteocephalus using the following character states present in Tepuihyla : subgular vocal sac, absence or extreme reduction of hand webbing, more reduced toe webbing, smaller size, absence of cranial coossification, large frontoparietal fontanelle, shorter nasals, and shorter frontoparietals. It is unclear which, if any, of these character states are apparent synapomorphies of Tepuihyla . There are eight species currently included in this genus: T. aecii , T. celsae , T. edelcae , T. galani , T. luteolabris , T. rimarum , T. talbergae , and T. rodriguezi . In this analysis we include only T. edelcae .

Trachycephalus: The relationships of this casque­headed taxon were discussed by Trueb (1970a) and Trueb and Duellman (1971). They diagnosed Trachycephalus from Argenteohyla , Osteocephalus , and Phrynohyas for having heavily casqued and co­ossified skulls, a medial ramus of pterygoid that does not articulate with the prootic, and a parasphenoid having odontoids. A likely synapomorphy of Trachycephalus is the presence of exostosis on the alary process of the premaxillae (Trueb, 1970a). Trachycephalus contains three species: T. atlas , T. jordani , and T. nigromaculatus . In this analysis we include T. jordani and T. nigromaculatus .

Species and Species Groups of Hyla Not Associated with Any Major Clade

Hyla aromatica Group: This group was proposed by Ayarzagüena and Señaris (‘‘1993’’ [1994]) for two species from the Venezuelan Tepuis, H. aromatica and H. inparquesi , which they could not associate with any of the species groups known from the Guayanas. Ayarzagüena and Señaris (‘‘1993’’ [1994]) noticed that the H. aromatica group shares some characters with the H. larinopygion group; however, they preferred to retain it as a separate group. They justified this decision based on the smaller size of members of the H. aromatica group, different coloration pattern, supraorbital cartilaginous process, vomerine odontophores smoothly S­shaped and with more odonto­ phores, small nasals, and large prepollex. They included as well other character states, that actually, like some of these just mentioned, are either shared with several neotropical groups (supraorbital cartilaginous process; Faivovich, personal obs.), or some species of the H. larinopygion group (vomerine odontophores smoothly S­shaped; Duellman and Hillis, 1990: 5), or with the H. armata group (labial tooth row formula; see Cadle and Altig, 1991), or they support the monophyly of the H. aromatica group (adults with strong odor). Considering the lack of evidence of monophyly for the H. larinopygion group, Ayarzagüena and Señaris (‘‘1993’’ [1994]) cannot be questioned for recognizing a separate species group.

Ongoing research by Faivovich and McDiarmid suggests that Hyla loveridgei should also be considered part of the H. aromatica group. For this analysis, we include H. inparquesi .

Hyla uruguaya Group: This group has never been mentioned as such in the literature. However, clear similarities had been shown by Langone (1990) between H. uruguaya and H. pinima (these species being almost undistinguishable). Possible synapomorphies of the H. uruguaya group are the bicolored iris (also shared with Aplastodiscus ; see Garcia et al., 2001a), the presence in tadpoles of two small, keratinized plates below the lower jaw sheath, and a reduction in the size of the marginal papillae of the posterior margin of the oral disc relative to the other papillae (Kolenc et al., ‘‘2003’’ [2004]). From this apparent clade we include H. uruguaya in our analysis.

Hyla chlorostea: Duellman at al. (1997) proposed the recognition of a species group to include the enigmatic Hyla chlorostea , a species known only from its holotype (a subadult male), which could not be associated with any known group of Hyla after its description (Reynolds and Foster, 1992). Unfortunately, we were unable to include this taxon in our analysis.

Hyla vigilans: Different perspectives concerning this enigmatic species were summarized by Suarez­Mayorga and Lynch (2001a). These authors rejected the possibility of a relationship with Scinax (as suggest­ ed by La Marca in Frost, 1985). Instead, they asserted that they suspected a possible relationship with Sphaenorhynchus or with H. picta (from the H. godmani species group) based on ‘‘oral disc and mouth position’’ of the tadpoles. We could not obtain samples of this species for our analysis.

Hyla warreni: This species, known only from two adult females, was described by Duellman and Hoogmoed (1992), who did not associate it with any other species or species group. Unfortunately, we could not obtain samples of this species for our analysis.

Other Genera

Aplastodiscus: The taxonomy and history of this genus was recently reviewed thoroughly by Garcia et al. (2001a). According to these authors the monophyly of the genus is supported by four putative synapomorphies: (1) the absence of webbing between toes I and II and basal webbing between the other toes; (2) bicolored iris; (3) females with unpigmented eggs; and (4) great development of internal metacarpal and metatarsal tubercles. Based on overall morphological and advertisement call similarities B. Lutz (1950) suggested a close relationship of this genus with Hyla albosignata . Garcia et al. (2001a) suggest that Aplastodiscus could be related with the H. albofrenata and H. albosignata complexes of the H. albomarginata group, as defined by Cruz and Peixoto (1984), based on the presence of enlarged internal metacarpal and metatarsal tubercles, and unpigmented eggs. Haddad et al. (2005) described the reproductive mode of A. perviridis and noticed that it was the same as that described by Haddad and Sawaya (2000) and Hartmann et al. (2004) in species included in H. albofrenata and H. albosignata complexes. Based on this, Haddad et al. (2005) suggested a possible relationship between these two species complexes and Aplastodiscus . Aplastodiscus is composed of two species, Aplastodiscus cochranae and A. perviridis ; we include both in our analysis.

Nyctimantis: This monotypic Neotropical genus was considered a member of the Hemiphractinae by Duellman (1970) and Trueb (1974). Duellman and Trueb (1976) reviewed the taxon and placed it in Hylinae . Duellman and Trueb (1976) considered Nyctimantis to be related with Anotheca spinosa because both share the medial ramus of the pterygoid that is juxtaposed squarely against the anterolateral corner of the ventral ledge of the otic capsule. Also, frogs of both genera are known ( Anotheca ; Taylor, 1954; Jungfer, 1996) or suspected ( Nyctimantis ; Duellman and Trueb, 1976) to deposit their eggs in water­filled tree cavities. However, Duellman (2001) latter placed Anotheca in the Middle American/Holarctic clade, implicitly suggesting no relationship with Nyctimantis . Considering the uncertainty of the position of Nyctimantis within hylines, at this stage it is difficult to interpret which character states are autapomorphic. We include the single species Nyctimantis rugiceps in this analysis.

Phyllodytes: The history of this genus was reviewed by Bokermann (1966b). Possible synapomorphies of the taxon are the presence of odontoids on the mandible and on the cultriform process of the parasphenoid (Peters, ‘‘1872’’ [1873]), something unique within the Hylinae . Peixoto and Cruz (1988) noticed that among the six species recognized at that time, four species ( P. acuminatus , P. brevirostris , P. luteolus , and P. tuberculosus ) share the presence of series of enlarged tubercles on the venter and an enlarged tubercle on each side at the origin of the thigh (Bokermann, 1966b: fig. 6 ). The other two species, P. auratus and P. kautskyi , have uniform granulation on the venter and lack enlarged tubercles on the thighs, as also seems to be the case in P. melanomystax , a species described later (see Caramaschi et al., 1992). Peixoto et al. (2003) described two additional species, P. edelmoi and P. gyrinaethes ; both have a tubercle on each side at the origin of the thigh. Phyllodytes edelmoi has a series of indistinct tubercles on the venter; in P. gyrinaethes they do not form series. Caramaschi and Peixoto (2004) added P. punctatus , which has two medial, poorly distinct rows of tubercles. Caramaschi et al. (2004a) resurrected P. wuchereri . Peixoto et al. (2003) suggested three different species groups based on color pattern, and Caramaschi et al. (2004) further expanded the definitions. The P. luteolus group is characterized by a plain pattern with a variably defined dorsolateral dark brown to black line on canthus rostralis and/or behind the corner of eye. This group includes P. acuminatus , P. brevirostris , P. edelmoi , P. kautskyi , P. luteolus , and P. melanomystax . The P. tuberculosus group has a pale brown dorsum with scattered dark brown dots and includes P. punctatus and P. tuberculosus . The P. auratus group has a dorsal pattern of two dorsolateral, longitudinal white or yellowish stripes, with each stripe being bordered by a dark brown or black line from posterior corner of eye to groin. This group includes P. auratus and P. wuchereri . Finally, P. gyrinaethes is placed in its own group for having red color on hidden surfaces of thighs and a highly modified tadpole. It is unclear if any of these groups is monophyletic. Tissues were available for P. luteolus and an unidentified species, Phyllodytes sp. , from Bahia, Brazil.

Lysapsus and Pseudis: The monophyly of the former subfamily Pseudinae has not been historically controversial; it is supported by the presence of a long, ossified intercalary element between the ultimate and penultimate phalanges. Haas (2003) added several synapomorphies from larval morphology, based on the study of larvae of two species of Pseudis . The limits and definitions of Pseudis and Lysapsus were reviewed by Savage and Carvalho (1953) and by Klappenbach (1985). From their observations it is unclear which character states support the monophyly of either genus. Savage and Carvalho (1953: 199) implicitly proposed the paraphyly of Pseudis , when they suggested that Lysapsus ‘‘seems to have arisen from Pseudis .’’ Garda et al. (2004) recently distinguished both genera on the basis of sperm morphology. In Lysapsus laevis (the only species of Lysapsus available to them) the subacrosomal cone is nearly absent, but it is clearly present in the four species of Pseudis they studied. Regardless, the monophyly of either genus has not been satisfactorily documented.

Morphological diversity within Pseudis includes large species, several of which were included in the past in the synonymy of P. paradoxa and were recently resurrected (Caramaschi and Cruz, 1998), and smaller species with a double vocal sac, P. cardosoi and P. minuta (Klappenbach, 1985; Kwet, 2000). Lysapsus includes three species, L. caraya , L. laevis and L. limellum ; we include in our analysis the last two. Pseudis is composed of six species: P. bolbodactyla , P. cardosoi , P. fusca , P. minuta , P. paradoxa , and P. tocantins , of which we include in our analysis P. minuta and P. paradoxa .

Scarthyla: Duellman and de Sá (1988) and Duellman and Wiens (1992) suggested that this monotypic genus was sister to Scinax , but more recently, Darst and Cannatella (2003) presented evidence supporting a sister group relationship between Scarthyla and ‘‘pseudids’’. The single species, Scarthyla goinorum , is included in our analysis.

Scinax: With roughly 86 recognized species, Scinax is the second largest genus within Hylinae . This genus includes the species formerly placed in the Hyla catharinae and H. rubra groups; a taxonomic history was presented by Faivovich (2002). The relationships among the species of Scinax were recently addressed by Faivovich (2002), who performed a phylogenetic analysis using 38 species representing the five species groups then recognized. Although he employed eight outgroups, the analysis is not a strong test of the monophyly of Scinax nor of the relationships of Scinax with other hylines. Duellman and Wiens (1992) suggested that Scinax is the sister group of Scarthyla and that this clade is sister to Sphaenorhynchus . Faivovich (2002) did not test this assertion because his selection of outgroups was heavily influenced by da Silva’s results (1998), which did not suggest a close relationship between Scinax and these two genera. Taxon choice in the present study will test more appropriately the hypothesis of ( Scinax 1 Scarthyla ) 1 Sphaenorhynchus .

Faivovich’s (2002) results suggested that Scinax contains two major clades: (1) a S. ruber clade composed of species that had been previously grouped into the S. rostratus , S. ruber , and S. staufferi groups; and (2) a S. catharinae clade composed of the species that were included in the S. catharinae and S. perpusillus groups. Faivovich (2002) continued recognition of these two species groups within the S. catharinae clade, as well as the S. rostratus group within the S. ruber clade, as the individual monophyly of the S. catharinae and S. rostratus groups were corroborated by his analysis. The S. perpusillus group is recognized because its monophyly could not be tested, and it still awaits a rigorous test. All species previously included in the nonmonophyletic groups of S. ruber and S. staufferi are included in the larger S. ruber clade, without being assigned to any group. For a list of the species currently included in Scinax , see page 95.

In anticipation of a forthcoming study of the phylogeny of Scinax by Faivovich and associates, we include only S. berthae and S. catharinae as exemplars of the S. catharinae clade, and S. acuminatus , S. boulengeri , S. elaeochrous , S. staufferi , S. fuscovarius , S. ruber , S. squalirostris , and S. nasicus as exemplars of the S. ruber clade.

Sphaenorhynchus: More has been written about nomenclatural confusion surrounding Sphaenorhynchus than about its systematics (see Frost, 2004). This genus has been reviewed by Caramaschi (1989). Duellman and Wiens (1992) proposed the following synapomorphies for Sphaenorhynchus : posterior ramus of pterygoid absent; zygomatic ramus of squamosal absent or reduced to a small knob; pars facialis of maxilla and alary process of premaxilla reduced; postorbital process of maxilla reduced, not in contact with quadratojugal; neopalatine reduced to a sliver or absent; pars externa plectri entering tympanic ring posteriorly (rather than dorsally); pars externa plectri round; hyale curved medially; coracoids and clavicle elongated; transverse process of presacral vertebra IV elongate, oriented posteriorly; and prepollex ossified, bladelike. The genus is composed of 11 species: S. bromelicola , S. carneus , S. dorisae , S. lacteus , S. orophilus , S. palustris , S. pauloalvini , S. planicola , S. prasinus , S. platycephalus , and S. surdus . In our analysis we include S. dorisae and S. lacteus .

Xenohyla: This genus was named by Izecksohn (1996) for the bizarre frog Hyla truncata , which had previously been suggested to be related to Sphaenorhynchus by Izecksohn (1959, 1996) and Lutz (1973). According to Izecksohn (1996), Xenohyla shares with Sphaenorhynchus the reduced number of maxillary teeth, a relatively short urostyle, and the development of the transverse processes of presacral vertebra IV; furthermore, Xenohyla shares with Sphaenorhynchus the quadratojugal not in contact with the maxilla. Izecksohn (1996) suggested also a close relationship with Scinax based on the presence in Xenohyla of a coracoid ridge and an internal, subgular vocal sac. While the coracoid ridge is present in Scinax , it is also present in several other hylines (e.g., see Faivovich, 2002). The internal, subgular vocal sac is not a synapomorphy of all Scinax , but only of the S. catharinae clade. Caramaschi (1998) added X. eugenioi , a second species for the genus. We include X. truncata in our study.

CHARACTER SAMPLING

GENE SELECTION

Because this study involves the simultaneous analysis of taxa of disparate levels of divergence, we assembled a large data set, including four mitochondrial and five nuclear genes, spanning a broad range of variation, from the fast­evolving cytochrome b (Graybeal, 1993) to the much conserved nuclear genes such as 28S (Hillis and Dixon, 1991).

Ribosomal mitochondrial genes and cytochrome b have been employed recently in several phylogenetic studies of various anuran groups at various levels of divergence (Read et al., 2001; Vences and Glaw, 2001; Cunningham, 2002; Salducci et al., 2002). Nuclear genes have been poorly explored for their use in anuran phylogenetics. The 28S ribosomal nuclear gene has been used in amphibians by Hillis et al. (1993). The proteincoding genes rhodopsin, tyrosinase, RAG­1, and RAG­2 were used to study problems at different levels by Bossuyt and Milinkovitch (2000), Biju and Bossuyt (2003), and Hoegg et al. (2004). In this study we include 12S, tRNA valine, 16S, and fragments of cytochrome b, rhodopsin, tyrosinase, 28S, RAG­ 1, and seventh in absentia. The last gene is used here for the first time in amphibians.

DNA ISOLATION AND SEQUENCING

Whole cellular DNA was extracted from frozen and ethanol­preserved tissues (usually liver or muscle) using either phenol­chloroform extraction methods or the DNeasy (QIAGEN) isolation kit. See table 2 for a list and sources of the primers employed.

Amplification was carried out in a 25­ml­ volume reaction using either puRe Taq Ready­To­Go PCR beads (Amersham Biosciences, Piscataway, NJ) or Invitrogene PCR SuperMix. For all the amplifications, the PCR program included an initial denaturing step of 30 seconds at 948C, followed by 35 or 38 cycles of amplification (948C for 30 seconds, 48–608C for 60 seconds, 728C for 60 seconds), with a final extension step at 728C for 6 min.

Polymerase chain reaction (PCR)­amplified products were cleaned either with a QIAquick PCR purification kit (QIAGEN, Valencia, CA) or with ARRAY­IT (Tele­ Chem International, Sunnyvale, CA) and labeled with fluorescent­dye labels terminators (ABI Prism Big Dye Terminators v. 3.0 cycle sequencing kits; Applied Biosystems, Foster City, CA). Depending on whether the cleaned product was purified with QIAquick or Array­It, the sequencing reaction was carried out in either 10 ml or 8 ml volume reaction following standard protocols. The la­ beled PCR products were isopropanol­precipitated following the manufacturer’s protocol. The products were sequenced either with an ABI 3700 or with an ABI Prism 377 sequencer. Most samples were sequenced in both directions.

Chromatograms obtained from the automated sequencer were read and contigs made using the sequence editing software Sequencher 3.0. (Gene Codes, Ann Arbor, MI). Complete sequences were edited with Bio­ Edit (Hall, 1999).

MORPHOLOGY

Because the present study is mostly based on molecular data, the failure to include a thorough morphological data set doubtless is its weakest point. As trained morphologists, most of the authors of this paper think that a phylogenetic hypothesis that explains all the available data is the best hypothesis that we can aspire to, and that no class of data is better than any other. Apart from da Silva’s unpublished dissertation, which is comment­ ed upon below, published comparative studies involving a diversity of hylid exemplars are rare. Major exceptions are the thorough osteological studies by Trueb (1970a) and those on hand muscles of Pelodryadinae (Burton, 1996), distal extensor muscles of anurans (Burton, 1998a), and foot muscles of Hylidae (Burton, 2004) . Explicit character descriptions in the context of phylogenetic comparisons include those by Duellman and Trueb (1983), Campbell and Smith (1992), Duellman and Campbell (1992), Duellman and Wiens (1992), Fabrezi and Lavilla (1992), Kaplan (1994, 1999), Cocroft (1994), Burton (1996, 1998a, 2004), Haas (1996, 2003), da Silva (1997), Duellman et al. (1997), Kaplan and Ruiz­Carranza (1997), Mendelson et al. (2000), Sheil et al. (2001), Faivovich (2002), and Alcalde and Rosset (‘‘2003’’ [2004]). Most of these studies were targeted in general to very specific apparent clades or to very large clades using very few terminals, which leaves particular sets of characters known for very few terminals. Unfortunately, for the inclusion of the characters employed in these studies to be informative, detailed anatomic work would be required on a very large number of terminals (besides the potentially serious need to redefine several characters), a task that we find impossible to pursue at this time. Much to our regret, we find that there are almost no published studies from which we could derive character scorings to enrich our data set without extensive work. The only data set that we thought could be included, due to its relatively dense taxon sampling, is the one resulting from the collection of observations presented by Burton (2004). Although its sampling of nonhylid taxa that match our sampled taxa is particularly sparse, we consider Burton’s study to be an important addition to this analysis. Characters are listed and discussed in appendix 3.

da Silva’s (1998) Dissertation

da Silva (1998) presented his Ph.D. dissertation on phylogeny of hylids with emphasis on Hylinae . Although da Silva’s dissertation has not been published, some of its results and conclusions were described and commented in detail by Duellman (2001). Because the present paper deals specifically with the phylogeny of Hylinae , we cannot avoid a few comments dealing with da Silva’s work. Considering the mostly coincident scope of both da Silva’s dissertation and this paper, it is evident that a thorough discussion and comparison of his results with ours would almost amount to the publication of his chapter on Hylinae relationships. This is a situation with which we feel most uncomfortable, because we think that this is a responsibility that rests on Helio R. da Silva.

From a purely practical perspective, at this point the integration of da Silva’s data set with ours is impracticable for two reasons: (1) The data matrix as printed in the dissertation distributed by the University of Michigan is incomplete, as it lacks the scorings for 10 characters (chars. 110–120) for all taxa. This is also the situation with the thesis that is deposited at the Department of Herpetology library of the University of Kansas, Natural History Museum (Faivovich, personal obs.). (2) A few scorings for groups that we are familiar with are not coincident with our observations on the same species, something suggestive either of polymorphism in those characters or mistaken scorings. 21 If this were the case, it would not be surprising, as scoring mistakes are to be expected in such an impressive data set. The problem with them is that once detected, they have to be corrected and the analysis has to be redone. It is evident that a revision of the data set is necessary before any integration can take place.

PHYLOGENETIC ANALYSIS

Our optimality criterion to choose among trees is parsimony. The logical basis of parsimony as an optimality criterion has been presented by Farris (1983). However, parsi­

21 For example, character 60 (anterior process of the hyale) is scored 0 (absent) in Aplastodiscus , where it is present in the material available to us (Faivovich, 2002; Garcia, personal obs.). Character 61 (anterolateral process of the hyoid plate) is scored 0 (absent) in Hyla albofrenata , H. albomarginata , H. albopunctata , H. albosignata , H. faber , and H. multifasciata , whereas it is present in the specimens available to us (Garcia and Faivovich, personal obs.)

mony has repeatedly been attacked from different perspectives, all of which tend to portray parsimony as inferior to such modelbased approaches as maximum likelihood. Criticisms of parsimony have centered on two main topics: statistical inconsistency and the notion that parsimony is an overparameterized likelihood model. As stated by Goloboff (2003), the emphasis on statistical consistency decreased following several studies showing that: (1) maximum likelihood can be inconsistent even with minor violations of the model when they were generated with a mix of models (Chang, 1996); (2) given some evolutionary models, maximum likelihood estimators could be inconsistent (Steel et al., 1994; Farris, 1999); (3) parsimony can be consistent (Steel et al., 1993); (4) assuming likelihood as a more accurate method, inferences based on trees suboptimal under the maximum likelihood could be less reliable than inferences made on trees optimal under otherwise inferior but faster criteria (Sanderson and Kim, 2000); and (5) at least under some conditions, parsimony may be more likely than maximum likelihood to find the correct tree, given finite amounts of data (Yang, 1997; Siddall; 1998; Pol and Siddall, 2001). Tuffley and Steel (1997) demonstrated that parsimony is a maximum likelihood estimator when each site has its own branch length. Farris (1999, 2000) and Siddall and Kluge (1999) suggested that the results of Tuffley and Steel (1997) were an indication that the model implied by parsimony (‘‘no special model of evolution’’ or ‘‘no common mechanism model’’) was indeed more realistic. However, likelihood advocates (Steel and Penny, 2000; Lewis, 2001; Steel, 2002) countered that models that assume constant probabilities of change across all sites are to be preferred on the grounds of simplicity (i.e., as having few­ er parameters to estimate). Goloboff (2003) demonstrated that parsimony could actually be derived from models that require even fewer parameters than the commonly used likelihood models.

The use of Bayesian Markov chain Monte Carlo (BMCMC) techniques has become quite popular among evolutionary biologists. However, for reasons outlined by Simmons et al. (2004), the posterior probability values of the clades cannot be interpreted as values of truth or support. Furthermore, Kolaczkowski and Thornton (2004) demonstrat­ ed, by using simulations in the presence of heterogeneous data, that parsimony performs better than both maximum likelihood and BMCMC over a wide range of conditions.

We contend that all serious criticisms of parsimony have been rebutted. We consider that while the first point mentioned (inconsistency of likelihood when the data are generated with different models) could certainly occur in any analysis, it is particularly problematic in the present one, because we are combining morphology with both mitochondrial and nuclear coding and non­coding genes. Furthermore, for a data set of this size, maximum likelihood is quite impractical to apply for computational reasons.

For the phylogenetic analyses of the DNA sequence data, we used the method of Direct Optimization (Wheeler, 1996, 1998, 2002), as implemented in the program POY (Wheel­ er et al., 2002), a heuristic approximation to the optimal tree alignment methods of Sankoff (1975) and Sankoff and Cedergren (1983). Sequence alignment and tree searching have traditionally been treated as two independent steps in phylogenetic analyses: sequences are first aligned, and a fixed or static multiple alignment is then treated as a standard character matrix that is the basis for tree searching in the test of character congruence. However, there may be other equally defensible multiple sequence alignments that would require fewer hypothesized transformations to explain the observed sequence variation; an explanation that requires fewer transformations is more parsimonious and is therefore objectively preferred over explanations that require a greater number of transformations (see De Laet [2005] for a much more sophisticated approach to the problems of constructing multiple alignments prior to tree searching). Direct Optimization seeks the cladogram­alignment combination (i.e., the optimal tree alignment) that minimizes the total number of hypothesized transformation events required to explain the observations. Within this framework, insertion/deletion events (indels, gaps) are historical evidence that is taken into account when hypothesizing common ancestry.

The simplest minimization of transformations is obtained when tree searches are conducted under equal weights for indels and all substitutions (1:1:1, this is the ratio of the cost of opening gap:extension gap:substitutions) (Frost et al., 2001). This weighting scheme implies that indels are as costly as the number of nucleotides they span. This is not a situation with which we are comfortable, inasmuch as a single deletion event could entail more than a single nucleotide and hence necessarily require a lower cost than if all the nucleotides it includes were lost independently of each other. However, theoretical justifications for the selection of differential costs for gap opening and gap extension are not evident.

De Laet and Smets (1998) suggested that parsimony analysis searches for the trees on which the highest number of compatible independent pairwise similarities can be accommodated; that is, they described parsimony as a two­taxon analysis. When dealing with static data sets, this approach and the minimization of transformations give the same rank of tress. However, De Laet (2005) showed that when considering parsimony as a two­taxon analysis in the presence of inapplicable character states (e.g., unequallength sequences), the minimization of transformations (as obtained under 1:1:1) does not maximize the number of accommodated compatible independent pairwise a priori similarities. De Laet (2005) suggested that sequence homology has two components, homology of subsequences (the fragments of sequences that are comparable across a branch) and base­to­base homology within homologous subsequences. When maximization of homology is transformed into a problem of minimization of changes, the optimization of the two components that maximizes the accommodated independent pairwise similarities is obtained by summing up the cost regimes that are involved for each component. The number of subsequences is quantified by counting the number of insertion/deletion events (independent of their length, and therefore represented each as a whole by a unit opening gap). Base­to­base homology within homologous subsequences is maximized when substitutions are weight­ ed twice as much as unit gaps (Smith et al., 1981). These result in a substitution cost of 2, a gap opening cost of 2 1 1 (the same cost of a substitution plus the cost of the first unit gap), and a gap extension cost of 1. All this development rests on the perspective of parsimony as a two­taxon analysis (De Laet and Smets, 1998). The most immediately appealing aspect of De Laet’s perspective is that it offers a rationale for the use of gapextension costs different from substitution costs, thus avoiding giving an insertion/deletion event of n nucleotides the same weight of n substitutions.

We conducted our searches using equal weights for minimizing transformations. In order to examine the effect of the gap treatment in our results, and following De Laet’s development (2005), we also submitted our final tree to a round of tree­bisection and reconnection branch swapping (TBR) by using a weighting scheme of 2 for substitutions and morphological transformations, 3 for a gap opening, and 1 for a gap extension.

This study is guided by the idea that a simultaneous analysis of all available evidence maximizes explanatory power (Kluge, 1989; Nixon and Carpenter, 1996). Consequently, we analyzed all molecular and available morphological evidence simultaneously. The analysis was performed using subclusters of 60–100 processors of the American Museum of Natural History parallel computer cluster.

Heuristic algorithms applied to both tree searching and length calculation (i.e., alignment cost) were employed throughout the analysis. As with any heuristic solution, the optimal solution from these analyses under Direct Optimization represents the upper bound, and more exhaustive searching could result in an improved solution. Considering the large size of our data set, we tried two different approaches. The first strategy tries to collect many locally optimal trees from many replications to input them into a final round of tree fusing (Goloboff, 1999). For the second strategy, quick concensus estimates (Goloboff and Farris, 2001) are used as constraints for additional tree searching, following the suggestion of Pablo Goloboff (personal commun.).

For maximizing the number of trees for tree fusing, we employed two different routines:

1. Three hundred fifty random addition se­

quences were done in groups of 5 or 10,

followed by a round of tree fusing, sending

the best tree to 10–25 parsimony ratchet cy­

cles (Nixon, 1999a) using TBR, reweighting

between 15 and 35% of the fragments, keep­

ing one tree per cycle, and by setting the

character weight multiplier between two and

five in different replicates, with a final round

of TBR branch swapping. Tree fusing was

always done fusing sectors of at least three

taxa with two successive rounds of fusing. 2. One hundred fifty random addition sequenc­

es were built in groups of 5, 7, or 10 by

submitting the best of each group to 10–25

ratchet cycles using TBR. The 40 best trees resulting from these analyses where submitted to tree fusing in groups of five, and the resulting eight trees were subsequently fused. This final tree was submitted to 30 replicates of Ratchet using the same settings as above, with the resulting trees being submitted to a final round of TBR branch swapping. Alternatively, we did 50 random addition sequences followed by a round of TBR and made an 85% majority rule consensus, as suggested by Goloboff and Farris (2001) to quickly estimate the groups actually present in the consensus of large data sets without having to do intensive searches. The approach of Goloboff and Farris (2001) assumes that groups that are present in all or most independent searches are more likely to be actually supported by the data. To speed up the searches for the estimation of the quick consensus, we treated the partial sequences of the RAG­1, rhodopsin, SIA, and tyrosinase genes as prealigned. Once the quick consensus was estimated, it was inputted in POY as a constraint file, with which we built 100 Wagner trees, each followed by 10 ratchet replicates. All trees resulting from these constrained searches were fused in groups of different size, and the final trees were submitted to a round of TBR. The original constraint file was not used during the fusing and final TBR steps. While all searches were done using standard direct optimization, all were submitted to final rounds of TBR under the command ‘‘iterative pass’’ (Wheeler, 2003a). This routine does a three­dimensional optimization, taking into account the states of the three ad­ jacent nodes of the internal node of interest. Because any change in the reconstructed sequence could potentially affect adjacent nodes, the procedure is done iteratively until stabilization is achieved.

The large size of the data set imposes a heavy burden in computer times to estimate support measures. Bremer supports (Bremer, 1988) were calculated using POY, without using ‘‘iterative pass’’. Parsimony Jackknife values (Farris et al., 1996) were calculated using the implied alignment (Wheeler, 2003b) of the best topology. In turn, this implies that the parsimony jackknife values could be overestimated. Parsimony Jackknife was calculated in TNT (Goloboff et al., 2000); 1000 pseudoreplicates were performed. For each pseudoreplicate the best topology was searched for by using sectorial searches and tree fusing, starting with two Wagner trees generated through random addition sequences.

Final tree lengths under the 1:1:1 weighting scheme were checked with TNT. Lists of synapomorphies were generated with TNT; only unambiguous transformations common to all most parsimonious trees were considered.

For the analysis, the complete 12S­tRNA valine­16S sequence was cut into 14 fragments and the partial 28S sequence was cut into 4 fragments coincident with conserved regions (Giribet, 2001). Although this constrains homology assessment, the universe of alternative ancestral sequences that has to be explored is a more tractable problem than using long single fragments. The sequence files as they were input into POY are available from http://research.amnh.org/users/julian. Tree editing was done using WinClada (Nixon, 1999b).

Figures

Fig. 6. Current state of phylogenetic knowledge of the 30­chromosome Hyla. Redrawn from Duellman (2001: 857) with the addition of Kaplan’s (2000) suggestion regarding the relationships of Hyla praestans with the H. garagoensis group. Numbered synapomorphies, as textually described by these authors, are: 1, 30 chromosomes; 2, reduced quadratojugal; 3, 1/2 labial tooth rows; 4, nuptial excrescences absent; 5, tadpole tail xiphicercal; 6, tadpole mouth terminal; 7, 0/1 labial tooth rows; 8, 0/0 labial tooth rows; 9, one ventral row of small labial papillae in tadpoles; 10, extensive axillary membrane; 11, longitudinal stripes on hindlimbs of tadpoles; 12, one ventral row of large labial papillae in tadpoles; 13, tadpole body violin­shaped in dorsal view; 14, body of tadpole depressed; 15, labial papillae absent in tadpoles; 16, internal surface of the arytenoids with a small medial depression.